Reducing export tax rebates for certain products would make the tax treatment of those exports less neutral. Nevertheless, given the current fiscal constraints, trade frictions and excessive competition in some industries, this may be a second-best option worth considering.

This brief estimates which products could be considered for reductions in export tax rebates, and the amounts involved, under three policy objectives: raising fiscal revenue, easing trade frictions and curbing “involutionary” competition (excessive competition that erodes profitability).

This Brief draws three main conclusions:

• Amid tight fiscal constraints, reducing export tax rebates for certain products could help improve the allocation of fiscal resources. The 14 products identified under all three criteria account for an estimated RMB 162.8 billion in rebates and could be prioritized for further assessment.

• Although reducing export tax rebates would compromise tax neutrality, it could also help address some trading partners’ concerns and ease trade tensions.

• If such reductions are pursued, policymakers should carefully consider which products to target, when to implement the changes and how to phase them in. A reasonable transition period would give firms time to adjust their pricing and business arrangements.

Read the full report here.

Panelists include:

• Xiaochuan ZHOU, Former Governor, People’s Bank of China

• Jason FURMAN, Aetna Professor of the Practice of Economic Policy, Harvard University; Former Chairman, Council of Economic Advisers of the White House

• Kaushik BASU, Professor, Cornell University, United States; Former Senior Vice President and Chief Economist, World Bank

• Xiaojuan JIANG, Professor, University of Chinese Academy of Social Sciences; Former Deputy Secretary-General, State Council of the People’s Republic of China

• Bin YANG, Vice Chancellor of Tsinghua University Council

• Debora REVOLTELLA, Senior Fellow, Bruegel; Former Director of the Economics Department and Chief Economist, European Investment Bank

• The moderator is Chang Yong RHEE, Nonresident Senior Fellow at the Peterson Institute for International Economics and former Governor of the Bank of Korea.

Below is a transcript of the session, which is produced with the support of AI and has not been subject to the review of the panelists themselves. Among the panelists, Jiang Xiaojuan spoke in Chinese and the transcript is the translated version.

Rhee Chang-yong:

Let’s begin our discussion. In the first round, we’ll talk about AI and productivity growth in general. Let me start with Professor Furman, because the United States is leading at the frontier in AI modeling, computing capacity and AI adoption.

Professor Furman, do you expect AI to generate productivity gains faster or slower than previous technological revolutions, such as the Industrial Revolution or the Internet Revolution? In fact, in the US, we see nonfarm labor productivity growth has accelerated since 2022, roughly twice its pre-pandemic pace. So is AI already contributing to the US growth acceleration? And could these productivity gains help ease inflationary pressures, as some economists suggest?

Jason Furman:

Thanks for having me. The most important fact about AI is that it’s incredibly advanced in terms of its capabilities, but we’re still debating whether we’re seeing it in the macro data. I believe that there is a credible case that it is showing up. You cited some of the speed-up of productivity growth. When you look at both micro, bottom-up surveys of how businesses are using it, the industrial pattern of which industries have seen the growth, or just how much productivity growth has risen relative to trend, it’s plausible that maybe half a point, maybe even as much as an entire point of the cumulative productivity growth over the last five years was fueled by AI, but I wouldn’t rule out zero.

Same thing in the labor market. The unemployment rate is 4.2%. That is basically where it was prior to ChatGPT. You can maybe see it in certain pockets of jobs, certain age groups, but for the most part, you don’t see it in the aggregate numbers. That raises an important question, which is: why is that? One possibility is that good ideas are hard to find, and so it takes more technological inputs to get a given output. Another possibility is that we’re going down a J-curve, where you hire people upfront to figure out how to use AI. That hurts productivity in the short run, but you get it in the long run.

Then a final one, which is related to that, is you need a set of complementary inventions and redesigning of workplaces in order to get it to work. Well, I’m pretty firmly in the camp of those second groups of things. I expect we are going to get more productivity growth. That being said, if I had to put down a number, it would be numbers more like 0.75 percentage points than 15, 20, 50 or 100 percentage points, which many in Silicon Valley are talking about. But if a country like the United States could grow at 2.5% a year, that would be quite impressive; 3% a year, I would find outlandish. And numbers above that, I think, are from people in Silicon Valley who haven’t made contact with the reality of just how slow and complicated diffusion is in the economy.

Then, very last sentence, I think it’s more likely to be inflationary than deflationary, because you get the demand associated with the AI boom before you get the supply associated with it. So right now, I believe it is pushing up inflation, pushing up interest rates, and I don’t think that will reverse in a big way.

Rhee Chang-yong:

Thank you. I’ll move to Governor Zhou. China appears to be the only country currently competing with the United States in developing frontier AI models. And we’ve recently seen AI adoption in China increasing really rapidly. How do you expect the speed and magnitude of AI-driven productivity gains in China to compare with those in the United States?

Zhou Xiaochuan:

It’s a good and difficult question, because AI is very dynamic and there is a lot of uncertainty.

The financial sector tries to organize investment into AI-related sectors, which may have three categories. One is large models, and the computing power and data centers for training those models. Secondly, application areas which China may pay more attention to. The third one is supporting facilities, including semiconductors, power supplies, and water for cooling. Investment in these supporting sectors, I think, is relatively safe, because no matter whether AI is going to use a lot of electricity and water, other sectors — especially those related to climate change, will need more energy supply and water.

As for productivity, I think it mainly comes from application areas. I may talk about the potential technologies which may benefit productivity and are also related to the risks.

Number one is the so-called model lightening, that is, downsizing the large model into a small-size application which could be put into cars, laptops, or mobile phones. So there are several technologies for downsizing to have a relatively smaller size of model to be used. I think it’s very much related to productivity.

Another is distillation. For many people, distillation may raise concerns about infringement of intellectual property rights. But actually, distillation means that your knowledge and intelligence are actually upgrading from the intelligence in the past generation. Then you get simplified and precise concepts that are established on the existing knowledge and intelligence. So actually, distillation helps a lot for further application of AI, which may also be related to productivity in a variety of applications which may bring returns to the investors.

There are also some other related developments of technology. I will only identify a few of them. One is the large model evolution. Models are upgraded very quickly, some models fade out while other new models fade in. Another is the effort to develop AI-specific semiconductors—I mean the IC chips. Recently, Nvidia and Microsoft are trying to have a new design of IC chips which may combine graphics computing, traditional CPU computing and memory together. The third one is the large simulation models, which later on may generate the data, but do not depend very much on the existing large data storage in the data center.

There are a few very important directions which may assist in AI applications and bring productivity gains. But they also involve risks. We see a lot of investment put into the data center establishment, however, in the future, how much the AI large models and applications are going to rely on the data centers and the computing power is questionable.

Rhee Chang-yong:

Thank you, Governor Zhou. You really gave us a lot of food for thought. Especially, you mentioned sensitive issues with distillation, so probably I have to ask Professor Jason Furman what you think about his remarks on distillation. But having said that, I think you mentioned several application issues, not just for the large AI models.

And I will ask Professor Yang, at Tsinghua University you must have seen many students and young entrepreneurs adapting to AI technologies. So in China, can you address the same issue, the speed and adoption of AI, especially from the perspective of young students, researchers and entrepreneurs in China?

Yang Bin:

Thank you. Following up on Governor Zhou’s bigger picture, I just want to add some views from a university perspective.

I see every day how young students and researchers are adapting to AI. Tsinghua set up its own new School of AI to train PhD students, and also an AI+ College for undergraduate education, and empowering various disciplines. Both are working closely with industry players, which have much stronger compute, data and even talent bases. So I’m so proud that at Tsinghua University, some of my colleagues and also alumni are now leading Chinese AI initiatives.

How do I compare the pace of productivity gains and innovation in China and the US? Let me first be honest. I think both countries are still in a very experimental stage. The rules of AI innovation, I think, are not the rules of the Internet era. They are quite different. The Internet playbook no longer works. For example, since the Spring Festival of this year, many startups have tried to encourage people to burn more tokens. But after the second half of the year, they (started to) evaluate the real economic return from those tokens. They become more mature or rationalized. That’s my observation.

With that in mind, I have three observations regarding your question. First, on innovation, I think the frontier is moving fast on both sides of the Pacific and closely connected. Some of the best people at American frontier labs actually graduated from Tsinghua. What impressed me most in China is how AI diffuses, how it is applied to different industries and companies. No one is debating whether we should use AI or not, and actually, I have seen the early economic returns coming from AI-native companies. They have no obstacles or barriers to transform themselves into real AI companies.

Second, on productivity gains, and this is a harder part, which is to actually achieve economic returns. I think it’s not a technological problem. It’s an organizational problem. How we start from AI and, based on AI, redesign the whole process, the whole work and the whole organization.

For well-established organizations, it’s even harder for them to transform themselves compared with those startups and AI-native companies. So it takes time. I think people should be patient, and finance should be patient. But if there is not enough financial capital— I mean patient capital— there should be some ups and downs for the market.

Third, on research, that’s quite a new trend. Before May or June, many of my colleagues used to talk more about how AI empowers our teaching and learning. But now they talk about how AI upgrades and speeds up our research in different areas, especially mathematics and materials science. For example, some of my colleagues want to apply to spend half a year or one year on sabbatical or on leave to work in frontier labs. Why? Because s ome frontier labs have models that are not even released, but they can leverage them to do their basic research. Don’t think basic research is far from economic returns. In the long run, they will contribute to economic development.

Rhee Chang-yong:

Thank you. You remind me that the new generation who keep up with AI are currently playing with it very well. I think many Korean mothers are now in trouble, because when their six- or seven-year-old children ask questions, the parents give an answer, and the children check and say that the mother is wrong. So I think we don’t know how to educate them at this moment.

Deborah, you just heard about China and the United States. When you were Chief Economist at the European Investment Bank, your team published a paper on the productivity impact on individual firms. But from a broader perspective, how is the European situation? Is it similar to the US and China?

Debora Revoltella:

I think what we can say is that the European situation is similar in terms of what we see today in the global economy, but a little bit less strong. If we look at the impact of AI on the global economy at this moment, we mostly see the effect on capital accumulation and trade, somehow a wealth effect. On capital accumulation, we see much stronger effects in the US. But we also see that in Europe, there is some acceleration of investment related to AI, particularly in the corporate sector. The AI part of corporate investment is accelerating more than the traditional, non-ICT-related part of corporate investment.

On the trade side, I tend to think that a big part of maintaining globalization is still related to the very strong push of AI-related trading goods. That’s very, very visible in Asia— I think that’s the part benefiting the most. But also in Europe, we see that there is a bottom on the import and export side, and there is an increase in the share of AI-related trade in goods in overall trade. That’s quite important, because it tells you that one of the drivers of the economy is really the new AI economy that is booming. We also see another effect, and that’s much more complex. It’s something I’m studying.

If we look at the balance of payments on the services side, I think the whole part of AI, the dependency on models and the cloud effect, is starting to be visible. In Europe, we see it in a very blurred way, also because of the strange arrangements of AI companies in Ireland. But actually, you start seeing effects, and they are also symptoms of the kinds of dependencies that are going to develop over time in different countries. That’s what we see today. So actually, it’s just an acceleration of some parts of the economy. But then, on the productivity effect, you were mentioning the paper I was writing with former colleagues and colleagues from the BIS.

We actually looked at the adoption of firms, so we had a special database on firm adoption. We see a couple of things. We see that in Europe, firms tend to adopt AI like US firms in terms of general adoption. But then if you look at the depth of adoption, actually there is a much lower share of firms that systematically use it in different ways, and also a lower share of firms that go from using general models to building in-house applications. So that’s the part where the effect in terms of productivity will be stronger. We estimated the effect on productivity. Nevertheless, with the firm-level data, we see an effect, which is very nice, exactly in the middle of what the literature is saying, 0.4-1 percentage point, exactly what the literature says.

Also we see some kind of aggregate effect. So the way in which we put it is that 10% of the TFP growth that we have seen in Europe in the last five years is generated by AI. But I think the most important point is the enablers. So this estimate is just the first part. But what we also find is that to really have the productivity-enhancing effect of AI, you need firms to also invest in intangibles, invest in skills, and you need the capacity of the system to reallocate resources. And that’s very interesting, because at the European level, these are the things that for 20 years we have been saying we need to do more on: flexibility of the system and reallocation of resources, etc.

So that brings us back to the point that maybe in the final take-up of productivity related to AI, Europe may have to push more There is also another important enabler related to AI, and that is scale. There are a couple of points. On the one side, on the data side, Europe has some kind of excellence in the health sector, but in other sectors, we need to do more in terms of pooling data. There are very interesting experiences coming from the industry side in trying to put data together in a comfortable way for companies. There is a very nice example on the automotive sector.

But then Europe is also waking up on another point: that is using demand more actively to create scale, And it’s very badly perceived, I think, internationally. But there is a need for doing that, and many other countries are doing it. I think it’s coming from procurement side, trying to create a base for demand that is European-based, but also the de-risking instruments on the demand side.

In the last point, I have happened to, as an external, look at what China is doing in the new economy sector, and there are a couple of examples that I find extremely interesting. I don’t think they are generalized, but a very nice example of what could be done in the maximizing the benefit of AI adoption.

One example I talked with one of the business angels that is supporting innovative companies on AI. They seem to have a very strict target in passing from research to application on the market. Very strict, but actually it’s successful because it helps bringing to the market the very early technology, Another point, I know that there was an idea of providing an instrument that insures the risk of first technologies going in the market. I understand that the banks didn’t want to provide the service, but I think the idea is extremely interesting: de-risking when you have new technologies that have to go in the market.

And the last one, moving as a system. I was fascinated by one example. I think it’s a humanoid center that is trying to invite other companies to bring a robot to be trained together, and that creates the big pool of data that actually improve the training, etc. So I think there are very interesting examples to look at.

Rhee Chang-yong:

Thank you, Deborah. I think by listening to our previous all speakers, at least for United States, China, and advanced economies, there are many things are changing, and even though we are saying that you know aggregate level productivity gains are still at this moment not clear, but by listening to you, my impression is that you are expecting that even productivity gains in aggregate level will coming soon. So it’s quite maybe the magnitude may be different, but that’s my impression that you are actually quite positive about the future that this AI will bring to us our economy.

Kaushik, you were the chief economist at the World Bank. So, from emerging market and the low income country perspective, how do you think this picture will change or differ in low-income countries, and how whether AI will help narrow the gap between the rich and poor countries or the other way?

Kaushik Basu:

Thank you very much. This is speculating about a world that we really understand very inadequately, because the changes have been so quick that people are fumbling to get a handle on what is happening to the world that we inhabit.

Keeping that in mind, that as the bottom line, if we look at the developing world, emerging economies, low-income countries, where do they stand and where will they end up? Let me back up once and make a comment about the world economy, and then I’m going to move to that.

There was reference to the industrial revolution right at the start, and we have to remember that if things go well during this AI revolution, our productivity could rise, growth could rise in dramatic ways, and dramatic in a way that people find difficult to imagine.

Before the industrial revolution, world growth used to be 0.3 percent per annum. The whole world’s GDP at 0.3 percent per annum. After the industrial revolution, after about 100 years, that became 2.1 percent per annum, so seven times faster the world is growing. If something similar is going to happen, our first sight is it’s a bit of a dizzying thought: the world growing at that phenomenal rate. But one thing we have to keep in mind: growth of GDP. And I don’t have to explain this to economists. Does not mean more cars, more homes, more mansions, more airplanes, better health, better education.

These are all things we value. GDP is anything that we value. So a very differently constituted GDP could be going through the roof. Where does the developing country stand over here? My own expectation is we talk about K-shaped recoveries, path ahead, the world, and if you think of the nations as constituents, it’s a K-shaped future that we are looking at. I do believe that some paths are going to go through the roof, but other paths, unless we take real serious policy measures, could do very poorly in this new world. You could really see a K-shaped recovery, which is going to lead to social and political troubles of a kind that we can’t imagine.

So that’s the possibility. That’s the prospect. Let me back up a little bit and touch on India specifically, since before going to the World Bank, I was working very closely with the Indian economy. India stands at a very unusual juncture, because at one level, when the information technology changes took place, India was one of the biggest beneficiaries, but it was a very strange relationship. A large number of well-educated people with engineering and other technical skills helped link India with Silicon Valley. The Bangalore–Silicon Valley connection played a major role in India’s takeoff.

Politics leaves that path unclear. For India, a lot of how India does out of this, there’s enough expertise, enough talent in the country. The prospects are, I think, very high. But global relations will matter because we live today in a very, very supply chain linked economy, so you can’t just as an isolated country do well. Even a big country like India - population wise, it’s now the biggest country, I don’t know if that’s good or bad - but even for that, the linkages are critically important.

Even for an economy as strong as China’s, the linkages are extremely important. China is making investment in high-tech education, patent building, if you look at globally today, China’s intellectual patents being taken out is almost double the United States, which was unimaginable 10 years ago. Intellectual patterns being taken out, China highest, U.S. Japan, Korea, India is the fifth, but with a bit of a large gap over there.

So there is this mixture of countries. One just keeps one’s fingers crossed, and hopefully we’ll get into that. There is the political and social side. If we can’t manage that, then this huge prospect can turn into a very chaotic political situation. I hope gatherings like this, best minds being brought together, will think in terms of political social moves, which can convert the huge opportunities into actually higher standards of living, especially for the developing world that needs it.

Rhee Chang-yong:

Very well noted.

Now let’s move beyond productivity gains, and now turn to the social implication of AI. I want to ask Professor Jiang. She has worked for the government quite a long time. In August 2025, China’s State Council announced its AI Plus initiative to accelerate AI adoption across in China. But when you actually launch these new initiatives, were there any concern that rapid automation could reduce labor demand? Please.

Jiang Xiaojuan:

The Chinese government attaches great importance to the “AI Plus” initiative. We hope that AI will not only play a role within the AI sector itself, but also empower a much wider range of industries.

As for the impact of AI on employment, this is a definite effect, and it is already happening. But so far, it has not become particularly serious. In terms of the Chinese government’s response, we are taking an active approach. First, we are encouraging AI to create new jobs. This is already quite evident in some areas. Take the film and television industry, for example. With the emergence of AI, employment in the traditional film and television production industry, including films and TV dramas, has fallen from 1.3 million to 700,000, because many stages of production can now be handled by AI. At the same time, however, a completely new industry has emerged in China, known as “short dramas,” and it is growing very rapidly. The short-drama industry has created more than one million new jobs. As a result, total employment in the sector has actually increased. And this increase has been more broadly shared: jobs and income are no longer concentrated in the leading film and television companies. Instead, many young people can now produce short dramas themselves. Even a short drama watched by only 10,000 or a few thousand people can generate a modest income. So AI has also created a great many new jobs.

Second, we need more flexible assistance policies for people who have already lost their jobs. The government is now paying close attention to social protection for people in flexible employment, and is stepping up its efforts in this area.

Looking ahead, my personal feeling is that AI will continue to have an impact on employment. So the overall design of the social security system is not simply a matter of social security policy; it is also a matter of how the entire economic system is structured. Take taxation, for example. We have tax policies designed to encourage businesses to invest, but the incentives are mainly directed toward investment in equipment. Businesses can receive certain tax benefits, such as VAT refunds on equipment investment. But there are not many comparable institutional arrangements for companies that hire more people. So the overall design of the national tax system should be more oriented toward mechanisms that encourage greater employment. The government should make greater use of the tax system to encourage businesses to hire people. This is also a very important institutional arrangement.

Last night, I saw a news report that Elon Musk plans to give every child in the United States one share after SpaceX goes public. In other words, he wants to share the returns generated by technology companies with the entire population. As these issues emerge and society continues to progress, I believe new ideas and ways of thinking will also develop.

Rhee Chang-yong:

Thank you. One thing I echo with you is that probably need some kind of tax reform, which in favor of employment rather than investment. That seems to be a very important issue.

Let me move back to Professor Furman online. Beyond employment and income inequality, you recently made an interesting point that some major U.S. AI companies appear to seeking market dominance, potentially creating monopolistic structures. So, can you elaborate more what you have in mind and some other risk that AI will bring to our society, Professor Furman?

Jason Furman:

Yeah, and I’ll try to be very brief because so much we want to hear from the whole panel. I think paradoxically, people think this will be like Google or Facebook, where whoever goes first will capture the whole market. But it’s not like those technologies. I can use ChatGPT, and you can use DeepSeek. Moreover, I can use ChatGPT one day and switch to DeepSeek the next day. Moreover, we’re already seeing the more advanced the technological capabilities get, companies don’t want that. They don’t need that. It is much, much more than they need. They’d much rather have the cheaper and faster models.

And for all of those reasons, I think the better the technology gets, it will tend not towards winner take all like we’re used to from previous technologies, but instead very vicious competition, the Bertran type competition we’re seeing now, where price almost falls down to marginal cost, sometimes maybe even below marginal cost, and so that’s in some ways a good thing if what you’re worried about is concentration.

In another way, it should exacerbate your fears about a bubble because even if we get the productivity growth, this vision of the future says you won’t be able to monetize it. So I expect, if anything, this to be one of our most competitive technologies.

Rhee Chang-yong:

Okay, thank you. Also, Debora, you mentioned also one time that you have a serious concern about AI dependence and market competition, and you mentioned that the few dominant companies, mostly in the United States, could capture most of the economic return from AI investment. So, in that case, how can European countries avoid becoming locked in in their technologies and ensure fair share of the benefit?

Debora Revoltella:

I think I have a very, very similar view of what Jason was saying, in the sense that the concern of having a more monopolistic competition coming up is actually kind of fading, and there is a bigger component that is associated to actually open model entering in the market and competing and offering a very valuable solution, particularly for models that are more on the application side, so the converting for the application side at the industrial level, etc. That’s very good in the development. And if we look at the European companies, they are moving toward that in the sense we were asking firms how they approach AI, they use generalized models, but they are also trying to build new internal application based on open model.

And the trick is that they very like the fact of downloading the weight and keeping it internally, so they don’t have to constantly share the data. And that’s a very strong selling point. It helps global competitions. It helps the concern on sharing information. The concern I have, but it’s very new, and we have all to look into it. It’s a draft proposal in terms of export control for a frontier model.

I understand that this may actually carve out exactly what I’m talking about: the downloading of models for companies to use on their local servers. But this is very important, because if that is not carved out, it will create a lot of uncertainty and, if you will, a lot of delays in the broader use of these models across Europe.

So I think the competition that’s coming is very good. And I fully agree with Jason on what he was saying about Europe. This is particularly important because we are building these dependencies on models from outside.

Rhee Chang-yong:

Okay, thank you. Kaushik, for this issue of possible monopolistic structure and uneven share of the benefit and the profit from AI investment, from a low-income country perspective, how the government has to react to this kind of risk? Do you think some kind of voluntary agreement among AI developers is going to be enough, or do we need something more? What do you think?

Kaushik Basu:

There is a lot that individual governments can do, but I also do believe that this is such a global change that for individual countries there are also very severe limits to what each one can do without causing a flight of capital, flight of critical resources going elsewhere. So you will also need coordination.

See, one of the things that is happening as a result of this change is the loss of voice. This not enough is being written about. This is that if inequality continues to grow the way it’s growing, and I’m not referring to general inequality across the population of Gini coefficient type inequality, but inequality of a few tech super rich people and the rest, there is a lot of writing now that that’s going sky high, that’s having one social impact, which is affecting countries.

You know there are anthropological studies in village councils when people sit around and talking. When the super rich landlord walks in, people lose their voice. Suddenly they can’t speak anymore. The villagers, something like that is happening a bit in the world, where the voice is being captured by very few, and for developing countries that means you’re losing voice. One is seeing some initiatives, including in China, prominent initiatives, to pull the South more into this global space of conversation, and I think that is going to be critically important.

If I may touch on something that you mentioned about the system of taxation and job creation, I do have some hesitation about whether that will work. If jobs begin to vanish because whatever is onerous, an AI or AGI-powered robot can do it. Jobs will have to be created, which is the Keynesian job of digging holes and filling them up. That’s not going to be productive. So I do believe that we are reaching a stage of the world where some universal basic share, and you were referring to that with individuals doing, but it is needed because if regular work begins to vanish, the world is still very well off. Lots of goods and services available, but regular work is vanishing. The working class gets affected. Just to employ them with no productivity will mean a fiscal strain of an enormous kind.

What we do need to think of is very innovative form of profit sharing. A segment you want the profit motive, but a segment of the profit has to be shared. Old ideas that go back to economists like Marty Weitzman, who was writing about it once upon a time. We need to explore those now in today’s world if we don’t want political and social chaos to destroy the opportunities.

Rhee Chang-yong:

Good, well said. Professor Yang.

Yang Bin:

Thank you. Chinese government always appreciate a kind of, just as Professor Jiang Xiaojuan mentioned, pro-worker policy and put people in the center of the whole policy consideration and protect the people rather than protect the jobs. Jobs may be created or jobs may disappear, but always people will be…get some solutions in the future. Not just the government. Now the biggest challenge, in my view, is the first job for young people, or the entry job issue. So I do see the NGOs or social organizations can place some kind of philanthropies to set up some in between institutions to connect the university and the companies.

The companies they don’t have the motivation to keep the entry-level jobs for the younger because agentic AI could play a better or more effective contribution. But also another dimension for the entry-level job is training. So how we can develop the future younger generation? Maybe government and NGO and part of the university could contribute to this very urgent problem.

Rhee Chang-yong:

Okay, thank you.

Professor Jiang, I think you mentioned several policies that China government is taking to address this issue, but I have one specific question. Many Chinese AI companies are releasing their models as open source, and which make harder to recover their investment. So, how can China balance the benefit of open source development with the need to ensure adequate return from AI investment?

Jiang Xiaojuan:

Most of the people here have an economics background. The emergence of open-source AI models is, to a considerable extent, similar to what happened with open-source software back then. The people behind these models are idealists, I would say. Take DeepSeek, China’s first open-source model. At the time, I would argue that its main source of business revenue did not come from the model itself. Yet the team was very eager to make its excellent model open source so that everyone could use it.

There was a phrase they used at the time, which was the hope to provide a “public good for global technological development.” So open-source models did not emerge in the first place out of a consideration of business models. But once open-source models became available, closed-source models faced a very significant challenge, because people were no longer using them. As a result, open source and closed source became different business options for different companies. In practice, however, the emergence of so many models has intensified competition and brought about many new changes. Still, I want to give open-source models a thumbs-up first, because they have brought greater equality of access in the AI era.

Many countries, apart from China and the United States, do not have leading AI models of their own. Thanks to open-source models, developers in these countries can build new models on top of ours, tailored to their domestic industries and national circumstances. ChatGPT has only 300 million users in the United States, while India has 700 million users. So open-source models have helped make AI more accessible.

There is now a wide range of business models. Some companies integrate cloud services with models: the model itself is free, but the cloud infrastructure belongs to the company, and users pay when they use its cloud services. With the emergence of agents, there are also models accessed through APIs, MCTs and other interfaces, where commercial users pay according to the number of tokens they use. We also have many cases of private deployment. In sectors such as banking and finance, where errors are simply unacceptable, companies cannot take an open-source model and use it directly. They need to deploy it privately and fine-tune it, and they pay the companies that provide these deployment services. For consumer-facing services, users may get three free uses a day and then have to pay for additional use. Across the entire chain—from cloud services and computing power to open-source models and tokens—a variety of business models are taking shape. There are many ways to make this work, and I don’t think there is any major problem here.

Rhee Chang-yong:

Thank you very much.

Governor Zhou, I left the last hard discussion to you because you have been the voice of the wisdom in our international society. In your opinion, what kind of international cooperation is most needed at this moment?

Zhou Xiaochuan:

Because of the very great impact of AI on the global economy and the social structure, people call for international coordination. Recently, Chinese President Xi Jinping and U.S. President Trump reached a kind of consensus to consider how to set up a governance mechanism for AI. However, because AI is so dynamic, it’s hard for politicians and some ministers to really know how to set up governance mechanism for AI.

I think there are three types of things that they can surely do. One is to control military use. Another is to control hackers and those involved in fraud. Third is to try to protect individual privacy, which relates to how to use the data. The other things, I think, are still in the observation and research stage, as several participants have already talked about, such as the impact on the job structure, the distribution issues, and maybe the other impact on the social structure. But I think up to now it’s not yet the stage to formulate a consensus. But we may keep watching over that.

Personally, I would like to emphasize two things.

One is that along with AI development, global equilibrium, the supply-demand relationship and industrial structures may undergo substantial changes. So, as financial sector people or economists, we may try to say that we need to let the price mechanism play an important role in responding to this kind of change in the global equilibrium. So maybe we do not need to rely very much on administrative measures to prevent those impacts, but rather need to let price adjustments reflect what the structural changes are going on and what kind of response we have to adapt to the new structure.

Another issue, I think, for the technological evolution is: we may surely face the fact that some of the investment may go into failure, while some investment may be successful. But considering that so large amount of investment now is going into the AI facilities, if a part of them goes to failure, it can cause financial turbulence and may create a crisis. It needs a very large action to bail out our economy and some of the financial institutions. So we may need to conduct research and make preparation to deal with that kind of turbulence and prepare some instruments in our toolkits to deal with that kind of impact.

That’s my suggestion.

Rhee Chang-yong:

Thank you. As a former central bank governor, I cannot agree more than your last point. That’s very important immediate homework. Thanks very much, and we really enjoy the excellent discussion from the distinguished speakers. And also, I’m especially grateful for Professor Furman, who stayed through the midnight, and thanks very much for that. So let’s give a panelist a warm round of applause.

-END OF PANEL-

Manufacturing PMI fell markedly, with both the output and demand indices declining, partly because some demand had been brought forward ahead of the National Day holiday. The input and ex-factory price indices both declined from high levels, mainly reflecting a correction in nonferrous metal prices, while oil prices continued to rise.

Services PMI was weighed down by weaker activity in the financial sector amid a stock market correction, as well as the timing of this year’s Mid-Autumn Festival, which fell in September. Extreme weather, including typhoons, and the rollout of policies promoting the sale of completed homes rather than pre-sales weighed on construction PMI.

Based on these data, we have three observations:

1) Weakening manufacturing supply and demand point to renewed downward pressure on the economy.

2) Although manufacturing price indices eased, rising oil prices may further squeeze the profits of midstream and downstream firms.

3) External demand is expected to continue growing at a relatively rapid pace in the near term, while domestic demand will depend largely on how effectively the growth-support measures introduced since September are implemented.

The full analysis is available here.

Below is a transcript of Governor Zhou’s speech, which has not been subject to the review of himself.

Moderator:

China appears to be the only country competing with the United States in developing frontier AI models, and we’ve recently seen AI adoption in China increasing rapidly. How do you expect the speed and magnitude of AI-driven productivity gains in China to compare with those in the United States?

Governor Zhou:

It’s a good and difficult question, because AI is very dynamic and there is a lot of uncertainty.

The financial sector tries to organize investment into AI-related sectors, which may have three categories. One is large models, and the computing power and data centers for training those models. Secondly, application areas which China may pay more attention to. The third one is supporting facilities, including semiconductors, power supplies, and water for cooling. Investment in these supporting sectors, I think, is relatively safe, because no matter whether AI is going to use a lot of electricity and water, other sectors — especially those related to climate change, will need more energy supply and water.

As for productivity, I think it mainly comes from application areas. I may talk about the potential technologies which may benefit productivity and are also related to the risks.

Number one is the so-called model lightening, that is, downsizing the large model into a small-size application which could be put into cars, laptops, or mobile phones. So there are several technologies for downsizing to have a relatively smaller size of model to be used. I think it’s very much related to productivity.

Another is distillation. For many people, distillation may raise concerns about infringement of intellectual property rights. But actually, distillation means that your knowledge and intelligence are actually upgrading from the intelligence in the past generation. Then you get simplified and precise concepts that are established on the existing knowledge and intelligence. So actually, distillation helps a lot for further application of AI, which may also be related to productivity in a variety of applications which may bring returns to the investors.

There are also some other related developments of technology. I will only identify a few of them. One is the large model evolution. Models are upgraded very quickly; some models fade out while other new models fade in. Another is the effort to develop AI-specific semiconductors—I mean the IC chips. Recently, Nvidia and Microsoft are trying to have a new design of IC chips which may combine graphics computing, traditional CPU computing and memory together. The third one is the large simulation models, which later on may generate the data, but do not depend very much on the existing large data storage in the data center.

There are a few very important directions which may assist in AI applications and bring productivity gains. But they also involve risks. We see a lot of investment put into the data center establishment; however, in the future, how much the AI large models and applications are going to rely on the data centers and the computing power is questionable.

Moderator:

You have been a voice of wisdom in the international society. In your opinion, what kind of international cooperation is most needed at this moment?

Governor Zhou:

Because of the very great impact of AI on the global economy and the social structure, people call for international coordination. Recently, Chinese President Xi Jinping and U.S. President Trump reached a kind of consensus to consider how to set up a governance mechanism for AI. However, because AI is so dynamic, it’s hard for politicians and some ministers to really know how to set up governance mechanism for AI.

I think there are three types of things that they can surely do. One is to control military use. Another is to control hackers and those involved in fraud. Third is to try to protect individual privacy, which relates to how to use the data. The other things, I think, are still in the observation and research stage, as several participants have already talked about, such as the impact on the job structure, the distribution issues, and maybe the other impact on the social structure. But I think up to now it’s not yet the stage to formulate a consensus. But we may keep watching over that.

Personally, I would like to emphasize two things.

One is that along with AI development, global equilibrium, the supply-demand relationship and industrial structures may undergo substantial changes. So, as financial sector people or economists, we may try to say that we need to let the price mechanism play an important role in responding to this kind of change in the global equilibrium. So maybe we do not need to rely very much on administrative measures to prevent those impacts, but rather need to let price adjustments reflect what the structural changes are going on and what kind of response we have to adapt to the new structure.

Another issue, I think, for the technological evolution is: we may surely face the fact that some of the investment may go into failure, while some investment may be successful. But considering that so large amount of investment now is going into the AI facilities, if a part of them goes to failure, it can cause financial turbulence and may create a crisis. It needs a very large action to bail out our economy and some of the financial institutions. So we may need to conduct research and make preparation to deal with that kind of turbulence and prepare some instruments in our toolkits to deal with that kind of impact.

That’s my suggestion.

-END-

Our estimates show that both trends weigh on the consumption rate, but declining fertility has the larger and earlier effect. Under the UN medium variant projection, changes in the age structure alone (aging) lower the consumption rate by about 0.30 percentage points by 2035 and by a cumulative 1.35 percentage points by 2075. Taking declining fertility into account widens the 2035 decline to about 1.27 percentage points, an additional drag of nearly 1 percentage point that peaks at about 1.14 percentage points around 2040.

Based on these findings, we draw four policy implications:

• The drag from declining fertility on the consumption rate is larger and comes sooner than that from aging, and is largely realized before 2040.

• Demographic effects matter far less than household expectations and the macroeconomic environment, and improving these may be enough to offset them.

• Raising the household share of national income would lift consumption even with an unchanged consumption rate, thereby offsetting the decline in the consumption rate.

• Demographic trends are hard to reverse in the short term, and even a rebound in fertility would take more than two decades to produce noticeable effects. A more realistic approach is to accept these realities and adopt proactive policy responses.

Read the full-length analysis here.

First, the Fed’s decision to pause rate hikes largely reflects a desire to assess how the AI-related market correction, Treasury yield movements, developments in Iran, and energy price shocks will unfold. With inflation showing little sign of easing, we expect a 25-basis-point rate hike in December, potentially followed by another hike next year.

Second, Warsh described economic conditions that would justify a rate hike but chose to hold rates steady, leaving an inconsistency in the Fed’s policy communication. However, markets had already largely priced in an October pause and a December hike. The relatively muted market reaction suggests that this inconsistency has not significantly undermined the Fed’s credibility.

Third, Warsh’s reaction function is becoming clearer, and Fed policy is increasingly sensitive to the AI cycle. Beyond employment and inflation, Warsh pays close attention to overall financial conditions and the economic impact of AI. His policy framework for assessing AI is also evolving: beyond its long-term supply-side effects and short-term demand shocks, he is increasingly considering how AI investment itself responds to interest rates and financing conditions.

On October 28, local time, the Federal Open Market Committee (FOMC) held the federal funds rate target range at 3.75%–4.00%, in line with market expectations. The decision passed by a 10–2 vote, with Fed Governor Lisa Cook and Cleveland Fed President Beth Hammack favoring a 25-basis-point rate hike. The policy statement upgraded its assessment of economic activity from “expanding at a solid pace” to “expanding at a strong pace.” It noted that job growth was broadly in line with labor supply growth and that the unemployment rate had changed little. At the same time, the Committee emphasized that inflation remained elevated and that more information was needed to determine the appropriate policy stance.

Fed Chair Kevin Warsh also announced changes to the FOMC meeting structure. Beginning in 2027, one of the eight annual meetings will be devoted to monetary policy strategy, with no interest rate vote and likely no policy statement. By 2028, the number of strategy meetings will increase to two, reducing regular policy-setting meetings to six per year.

The press conference provided three additional insights. First, Warsh explicitly acknowledged for the first time that rising interest rates had begun to affect AI capital expenditure through credit spreads and financing conditions. Second, he announced that he would participate in a public hearing of the Fed’s Productivity and Employment Working Group on November 16 to discuss AI’s implications for productivity, inflation, and employment. Third, while maintaining his opposition to routine forward guidance in normal times, he acknowledged that it could be useful during crises such as the one in 2008.

We draw three main conclusions from this FOMC meeting.

First, the Fed’s decision to pause rate hikes largely reflects a desire to assess how the AI-related market correction, Treasury yield movements, developments in Iran, and energy price shocks will unfold. With inflation showing little sign of easing, we expect a 25-basis-point rate hike in December, potentially followed by another hike next year.

Economic fundamentals alone would have justified another rate hike. Economic activity remains strong, employment is close to full employment, capital expenditure is resilient, and inflation is still well above the 2% target. The pause therefore does not reflect weakening fundamentals. Rather, the Fed wants to assess the broader effects of recent market adjustments and geopolitical shocks before proceeding with further tightening.

At the press conference, Warsh noted that financial and credit conditions had tightened for some businesses and households since September, while market participants had become more cautious in allocating capital. Although he did not consider overall financial conditions restrictive yet, he suggested that they were moving in a direction more consistent with sustainable long-term growth and price stability. He also pointed to episodes of financial market volatility and shifts in asset price dynamics, alongside renewed increases in energy and related prices in recent weeks.

Both the AI market correction and the energy shock could affect the policy outlook in opposite directions. If the AI correction remains confined to valuations without spilling over into financing or investment, its dampening effect on demand and inflation will be limited. But if credit spreads continue to widen and weigh on capital expenditure, the urgency of further rate hikes will diminish. Similarly, higher oil prices could fuel inflation while eroding household purchasing power, weakening consumption, and tightening financial conditions. The October pause is therefore best understood as a risk-management decision to wait for greater clarity.

Second, the Fed’s policy communication remains internally inconsistent, but has not triggered a significant credibility shock, as markets had already priced in an October pause and a December hike.

This meeting resembles the July meeting in one important respect: Warsh described an economic environment that would warrant a rate hike, yet chose to leave rates unchanged. At the press conference, he justified the pause by emphasizing the need for more information, characterized inflation control as a gradual process, and again declined to provide forward guidance on the December decision.

During the press conference, the two-year Treasury yield fell by approximately 4 basis points, while the 10-year and 30-year yields rose by around 5 and 6 basis points, respectively, resulting in a marked steepening of the yield curve. This pattern suggests that markets still see the Fed as inclined toward further tightening, but perceive less urgency to act.

Nevertheless, the market reaction was far more subdued than in July. Equity prices edged higher, the dollar weakened slightly, and other major asset classes showed no significant signs of either a flight to safety or fears of runaway inflation. Markets had already widely anticipated an October pause and viewed December as the next likely window for a rate hike. Thus, although Warsh did not fully explain why the Fed was holding off on tightening now, the decision did not materially deviate from market expectations. The inconsistency in policy communication has therefore not translated into a significant loss of credibility.

Third, Warsh’s reaction function is becoming clearer, and Fed policy is increasingly sensitive to the AI cycle.

Warsh’s remarks suggest that his policy decisions are guided not only by employment and inflation data, but also by how financial markets respond to monetary policy. When higher long-term interest rates, wider credit spreads, and market volatility are already tightening financial conditions, he is inclined to slow the pace of policy adjustment. Conversely, when financial conditions ease, asset prices rise, and demand strengthens, the case for further rate hikes becomes stronger.

Warsh’s framework for assessing AI’s economic impact is also evolving. Previously, he emphasized two main channels. The first is whether AI can raise productivity, reduce costs, and expand potential supply. The second is whether AI-related capital expenditure boosts aggregate demand, inflation, and long-term interest rates, while rising equity prices temporarily ease financial conditions. Productivity gains would support a more dovish policy stance, while stronger AI-driven demand would call for a more hawkish one.

This meeting introduced a third channel: AI investment itself is sensitive to interest rates and financial conditions. In practice, AI investment increasingly relies on corporate bonds, high-yield debt, and private credit. As long-term yields rise and credit spreads widen, financing costs have increased significantly for some data center operators and second-tier AI companies. Their capital expenditure is therefore not immune to tighter monetary conditions.

This suggests that the AI cycle could become a key constraint on how far the current rate-hiking cycle can go. With financial conditions already tightening and AI financing entering a more vulnerable phase, the remaining scope for rate hikes may be limited. Our baseline remains one rate hike in December, with another possible hike next year depending on economic and financial conditions. However, a sharp market selloff or sudden halt in AI capital expenditure could bring the tightening cycle to an earlier end.

1. Industrial profit growth slowed after adjustment for one-off (accounting) factors, revenue growth also softened further. (Pre-adjustment) headline industrial profit growth picked up to 14.4% YoY in September from 4.2% YoY, while revenue growth slowed to 6.6% YoY from 7.3% YoY (Figure 1). However, upon closer examination, the pick-up in September industrial profit growth was entirely driven by an abnormal surge (+224% YoY) of profit by the Liquor, tea, and beverage (LTB) industry, while profits of the same sectors “tanked” in August 2026. It was likely the case that major companies in the sector booked their chunky “residual” profit in August last year but in September this year (Figure 2). After adjustment, industrial profit growth (ex. LTB) decelerated to 8.2 % YoY in September, from 13.2% YoY in August and 22.4% YoY in 2Q 2026 (Figure 3).

2. Divergence between AI vs. non-AI, upstream vs. downstream profit growth intensified further (, after adjustment), while overall profit margin narrowed. As illustrated by Figure 4, AI related sectors and upstream (energy, mining & metals) industries has contributed to the lion’s share of industrial profit growth this year. However, the rest of the mfg. sector still managed to book positive profit growth before July -- the divergence widened after July, with the collective profit growth from non-AI & non-energy related sectors turned negative. Figure 5 & 6 illustrated the diverging path between the high-flying AI-related and upstream sectors vs. the rest, where the “crowding in” effect from the high-flying sectors reversed (implying margin squeeze) – the “inflection point” coincided with heightened pressure of fiscal tightening.

3. We are not at the “9.24 moment” for policy, yet. The financial regulators of China hosted a high-level press conference on September 24 2024, vowing to backstop the deterioration of growth momentum. The politburo followed up with a statement with similar message 2 days later. The decisive policy communication, followed by forceful fiscal loosening for the next 2-3 quarters, sparked a surge in the domestic stock market, aka. the “924 rally”. Although the downstream industries are experiencing profit contraction, we are still some times away from a “924 moment” now, since ① overall profit growth is still much higher than in Sep. 2026 (-27.1% YoY) (Figure 1), ② overall cash flow of the industrial enterprise improved on higher nominal growth and (Figure 7), ③ by the same token, central government revenue growth improved this year, as value-added tax revenue, the most sizable taxation income in China, is highly correlated with PPI.

4. Fiscal conditions hold the key to both the headline and structure of industrial profit growth. Overall profit growth, esp. that of the non-AI mid-to-downstream sectors, coincides with domestic demand growth. As we have analyzed in September Economic data: Marginal Stabilization, with a few Dark Clouds, published on October 19, 2026, fiscal tightening since 2Q has been the most important driver of slower domestic demand growth (granted, the squeeze from rising energy price also dampened demand on the margin). Meanwhile, stronger fiscal impulse with help mitigate the pressure from higher raw material and hardware cost by lifting end demand.

• Pierre-Olivier GOURINCHAS, Nonresident Senior Fellow, Peterson Institute for International Economics (PIIE); Former Economic Counsellor and Director of the Research Department, IMF

• Claudio SOTO, Board Member, Central Bank of Chile

• Claudio IRIGOYEN, Chief Economist, BofA; Managing Director, BofA Global Research

• Prakash KANNAN, GIC Chief Economist; Director of GIC Total Portfolio Strategy Group

• Tao WANG, CF40 Non-Resident Senior Fellow

• Gene MA (Moderator), Chief Representative, Asia/Pacific and Head of China Research, Institute of International Finance

Below is a transcript of the session, which is produced with the support of AI and has not been subject to the review of the panelists themselves.

Gene MA:

Welcome to the afternoon session. We have a line of superstar speakers. On my left is Pierre-Olivier Gourinchas. He right now is a senior fellow at the Peterson Institute for International Economics (PIIE). Formerly, he was the Director of Research at the IMF.

Mr. Claudio Soto is on the Board of the Central Bank of Chile. I think you are probably the one who traveled the greatest distance to come to Shanghai.

We have another Claudio in the middle, Claudio Irigoyen. And he’s the Head of Research and Chief Economist at Bank of America.

Prakash Kannan—both of us are based in Singapore——he is the Chief Economist and also the Director of Portfolio Strategy of GIC, one of the world’s largest sovereign wealth funds.

Finally, Ms. Wang Tao does not need much introduction. For a long time, she is, I think, one of the most experienced and well-respected China economists at UBS. Right now, she is a senior advisor.

Maybe I can start with Pierre. In the past ten months, we have experienced wave after wave of shocks -shocks from oil supply, AI build-out, rising yields, and fragmentation. We don’t even talk about tariffs anymore because we’ve gotten numb. But nonetheless, the global economy still seems to be just chugging along, not doing great, but also not poorly.

Why is that the case? Can you give us a survey of the landscape and connect the dots?

Pierre-Olivier GOURINCHAS:

I’d be happy to do that. I will make five points.

First, when we’re looking at the global economy, the IMF World Economic Outlook was just released a week ago. I’m not working with them anymore, but I read it carefully, and I agree with the broad analysis. What we’re seeing is a global economy that is facing a very significant negative supply shock. That started with the US-Iran war back in February, was expected to fade away, but is not fading away, and so energy prices remain very elevated. That’s a negative supply shock. It’s less activity and more inflation.

Then we have the AI build-up. The AI part, we’re not seeing it in productivity yet. It’s not a positive productivity shock or a positive supply shock, it’s a big demand shock. So we have a negative supply shock, we have a positive demand shock. They have contrasting effects on activity, but they have compounding effects when it comes to inflation and prices.

That’s exactly what you see when you look at the growth revisions in the latest World Economic Outlook. There are a lot of countries where growth is revised upwards significantly, and these are countries that tend to be positively affected by the AI build-up. And there are countries where growth is revised negatively and significantly, and those tend to be the countries that are hit by the war and the high energy prices. But when you look at inflation revisions in 2026, for most countries, it went up. There were positive inflation surprises on a Q4 basis, which is the way I like to look at the data. What that implies is that we have divergence when we look at activity, but we have convergence when it comes to inflation.

That brings me to my second point, which is that as a result of these positive surprises on inflation, pretty much for everyone, some of them because energy prices are up, some of them because the demand component coming from AI is also playing a role, monetary policy is resynchronizing. What we’re seeing in the second quarter and the third quarter is, central banks around the world are moving into hiking mode. 14 out of 33 major central banks have raised rates in Q2. Only two of them did that in Q1 of 2026.

So most central banks are tightening regardless of what’s happening to activity. That’s because, first, this negative supply shock. Usually we say, well, central banks should see through it. If it turns out to be more persistent than expected, then you can’t afford to see through it. And many countries are finding themselves there. For others, the demand component is also something that they need to deal with in order to hit their inflation targets. This is the broad picture.

My third point is, as a result of this new hiking cycle that has started - but not only that, there are a number of other things behind it and I’ll mention some of them - we’re also seeing real rates coming back up around the world. We’ve seen government yields going up, but more generally, we’re seeing interest rates going up. So, it’s a combination of monetary policy in response to higher inflation. It’s a combination of concerns about fiscal risk. It’s a combination of the impact of AI, which is absorbing a lot of loanable funds and therefore is driving up equilibrium interest rates, but it’s pushing up interest rates now.

If you look at where interest rates are now compared to, let’s say, 2000, they’re at about the same level. The ten-year rates are about the same level as they were in 2000, or 25 years ago. So you might say the level is not something that is out of bounds with what we’ve seen in the past. Even the rate of change, the increase in rates, is also not something that is out of the ordinary.

So why is that a problem? Well, it’s a problem because the base to which these higher rates apply. The level of debt, in particular public debt, is much higher now than it was 25 years ago. Public debt in the US was around 33% in 2000. It’s close to 100 now, so it’s been multiplied by three. Therefore, the impact of the same increase in interest rates is three times larger. That’s my third point, which is that this increase in real interest rates is having some real effects on eroding fiscal space much faster in many countries.

My fourth point is that a year ago, I’m sure a lot of the discussions here were about tariffs and were about the trade war. That has disappeared from the conversation. In a sense, we are in a new normal. There are tariffs that have been imposed by the US on the rest of the world. They might settle around 10% under new statutes - it’s not IEEPA anymore, it’s under Section 301. But by and large, this level of tariff is not going to break the world economy, and so the world economy has absorbed the tariffs. There’s been a lot of rerouting, there’s been a lot of adaptation, but that’s not where the main risks are right now.

Where are the risks?

One is in financial vulnerabilities that are building and continuing to build. Some of them are cross-border, related to global imbalances, some of them related to the AI build-up. These financial vulnerabilities are growing and could become a risk.

Second, fragmentation is and remains a major risk. We’ve seen it on the trade side. I just told you that it didn’t break the global economy so far. It’s unlikely to break it going forward. Financial fragmentation could have a much larger effect. It’s hard for me to imagine a world in which we fragment more and more on the trade side without fragmenting also on the financial side. I’m much more worried about the risks of financial fragmentation.

Both the financial vulnerabilities and the fragmentation more generally require fiscal space to build resilience and buffers, and that fiscal space has in part evaporated as interest rates have gone up and countries have not done the hard work of adjusting. So trade is not the main risk, but it doesn’t mean the risks are not there.

Fifth and final point, China. We’re in Shanghai. It faces the same shocks. It’s exposed on the energy side. It’s exposed on the AI side as well. There is an AI build-up and AI boom also in China. It’s more moderate than what we see in the US, but it is there. And so there is dependency there. Clearly, the right policy for China is to become more resilient in terms of domestic drivers of its economy.

Gene MA:

Excellent. This is really a very comprehensive survey from above. Now let’s zoom in a little bit. Maybe I can start with you, Mr. Soto.

Pierre-Olivier just mentioned that we’re seeing divergence in terms of productivity, and convergence in terms of inflation. Both the energy shock and AI build-out are inflationary. So how do all these forces affect Latin America these days? And how is Latin America and emerging markets broadly coping with those kinds of shocks?

Claudio SOTO:

Thank you to the organizers for the invitation.

As you say, Latin American countries have been affected by these two major shocks that we’re talking about. But I think there are some particularities within the region, particularities among countries, and also some particularities vis-à-vis what is going on in Europe, the US or other countries. So let me talk about these particularities.

First of all, the oil shock is a supply shock with inflationary impact all over the countries. There are some countries that are net exporters of oil and fuel, so those countries have benefited from an increase in their terms of trade. And there are other countries that are net importers, so they’ve been suffering the impact of the shock. This is the case of my country, Chile. In terms of inflation, we’re both facing the pressure of higher energy prices, but on the impact on fiscal space or the boost to demand, there are different impacts here depending on the country.

Second, in terms of the AI demand boost, Latin America in particular is one of the regions that is less integrated into the value chains that are pushing up activity globally speaking. There are some indirect impacts, for example, minerals that are critical for the development of AI are important in countries such as Chile and Peru, which are important producers of copper. But in general, we are far away from the value chains.

Therefore, all the positive impact in terms of demand, we’re not benefiting from that. So we have the supply shock coming from the high energy prices; and in the case of the AI boom, we’re also suffering an externality, which is related to the tightening of financial conditions that is produced by the increasing demand from AI. So in that sense, we’re not benefiting fully from the global development; rather we’re suffering from the impact of those two shocks.

Then I would add something more, which is the impact of political uncertainty specifically in the region. We have been going through a period of changes in government in several countries, where we have been moving from governments from one political side of the spectrum to another. One very drastic shift in political position that created uncertainty, and the global shock, particularly the supply shock, has created uncertainty, and uncertainty is a drag to growth. So, if you look at the region as a whole, growth this year will be very modest, and the outlook for the future is that growth will remain relatively modest. We are not benefiting from the tailwinds and we’re suffering from the headwinds.

Now, in terms of the policy response, countries in the region are starting from different starting points. There are countries that implemented very expansionary fiscal policy in recent years that are facing, on top of the shock, inflationary pressure coming from domestic policy in the past, so they need to be more aggressive in terms of taking steps towards addressing inflation. Fiscal consolidation is a requirement, but also tighter monetary policy. We’re seeing that in countries such as Brazil, Colombia.

But there are other countries that are in a much better position to confront this supply shock. This has been our position at the central bank in Chile. Being a transitory shock, although it has been more persistent than initially thought, but being a transitory shock in nature, eventually as long as inflation expectations are well anchored, you can look through. There is a big question here. How much can you sustain inflation that is above target and look through and not move your policy stance? This is a question that we will address in time, but at least from our point of view. We still have some space to not immediately react to this.

In sum, there is heterogeneity within the region, but the main impact of the global forces that we see around the world has been rather negative in terms of activity and inflation in the region.

Gene MA:

In Latin America, the whole political positioning has turned towards the right. The voters voted in either right-wing or right-leaning politicians in Argentina, Colombia, Peru, and probably very likely very soon in Brazil. Then I have two follow-up questions.

How would such a rightward change in Latin America affect the economy? It seems the stock market reacts positively, but I don’t know about the impact on the whole economy. Secondly, how would that affect relationships with Asia?

Claudio SOTO:

First of all, there is a big question why this is happening. One answer is people now realize that employment growth is more important, so they’re voting for governments that are more favorable to growth. An alternative explanation is that the opposition is winning the elections, because all the elections they’re having on the continent are being won by the opposition. So it’s not clear why that is the case.

Now in terms of the challenges, it’s true that the new governments are more pro-market, if you want, that probably they are more in favor of growth and investment. But delivering outcomes is not easy. I think the patience of the population is very limited, and it’s very easy to lose popularity and to lose confidence. That is one of the main challenges for these new governments. They promise growth, investment, but if they don’t deliver soon, then that would be complicated.

Now in terms of the relationship with the main powers, we’re a continent that produces commodities. I think commodities are needed, in particular with the green transition and the AI boom, the minerals are critical. We are still playing an important role in the world economy, and we’re going to keep playing that role in the future.

Gene MA:

Yes, just like former US President Bill Clinton said during his campaign, “It’s the economy, stupid.” Deng Xiaoping said a similar line, “Development is the hard truth.” You need to grow the economy, after all.

We have another Claudio. I want to turn to you. You cannot have a conference these days without talking about AI. Can you share with us your understanding on how AI would affect the economy, jobs, growth, productivity, demand for energy, and the impact on financial sectors? How is the Bank of America utilizing AI?

Claudio IRIGOYEN:

I think the discussion on AI cannot be separated from the very complex world that we are living in. I always joke that the world became so complicated that the Bank of America had to appoint an Argentine to run global economic research. We come from the future in many dimensions. When we think about AI, I think I’ll separate my remarks on growth, inflation, productivity, labor market and financial implications.

On growth, most of the AI story is driven by demand. It’s basically an investment boom that is affecting positively countries that are very much ingrained in the supply chain of the AI ecosystem. And the AI boom, or the investment boom, is impacting also the valuation of financial assets that are creating significant wealth effects that can be seen in the data. In particular, in the U.S. we have very good credit card data in the U.S. and we divide consumption dynamics by high, medium, low-income brackets, and we can see the wealth effects very clearly in the high- and middle-income brackets. That sort of created the what it was called the K-shaped type of economy.

Now we are seeing a closing of that K shape because now lower-income brackets are also spending more. They are also having wealth effects. This is ironically associated with the affordability issue that you have in the U.S. Because lower income brackets, especially Gen Z, cannot buy 5% of an apartment, but they can buy 50k of stocks, and they think that they can only go up, which we know is not true. Those are the two things that they are spending the money on: stocks and sport betting. What can go wrong? So this is one very important dimension.

The other important dimension about investment in AI is that all the stages of the AI ecosystem, from the producers of energy, producers of chips, cloud computing, the models are a winner-take-all type of markets, and therefore companies are investing to win market share. That means that they are rationally over investing, because not all of them will win. Three years from now, we are not going to have all the companies that we have now.

Therefore, at some point, the market will have to figure out whether the slowdown in investment is because some companies are dropping out of the race, or because the final demand will not be as big as initially thought. So, that will be an important signal extraction problem for the market, and if the market corrects, that will immediately feed through the wealth effect on consumption in the U.S. That’s one of the probably biggest financial risks that I’m going to talk about later.

So that’s the growth story, and as was mentioned before, the global economy has been pretty resilient. 3.5%, 3.4% global growth is an amazing number, given that everything that happened, and AI is sort of a part of that story, and the US will probably continue outperforming Europe or the rest of developed economies, growing probably twice as fast as the rest of the developed economies, because everything is basically happening there.

On prices, the question is always: Is AI inflationary or deflationary? The short answer would be: Well, it’s going to be inflationary first, inflationary later. When you have the productivity gains, but to be honest, I don’t like that approach. From an accounting perspective, probably that’s true. But to be honest, what AI is going to do is it’s going to change and affect significantly relative prices. If it’s going to be deflationary or inflationary will depend on whether central banks accommodate the shock or how they accommodate the shock.

But long story short, I would never use the story of AI to avoid hiking interest rates because AI will save us and it’s going to be deflationary. If that happens, it’s going to be a few years away, which leads me to the productivity story.

Are we seeing improvements in productivity? I think we are seeing a lot of improvements at the micro level. You can identify significant improvements at the micro level. We are not seeing that those improvements at the macro level yet, and it’s going to take a while, and it’s not even clear what we are talking about when we’re talking about productivity gains. Are we talking about labor productivity? Are we talking about total factor productivity?

We estimate that today AI is probably adding to global growth 10 basis points of global growth. So instead of 3.4%, it would be 3.3% without the AI productivity gains, and in the best case scenario, 10 years from now, it can add to a full percentage point of growth for global growth. As you can see, the range is huge. There is a lot of uncertainty about that. But long story short, we are not seeing the productivity gains yet at the aggregate level, which leads me to the labor market.

In the labor market, we are not seeing net destruction due to AI. Obviously, it’s much cooler for a company to say that they are laying off people because they are investing in AI and not because they are inefficient But the truth of the matter is that we are not seeing a net impact yet.

And I am on the camp that AI will destroy a lot of jobs, but will create also a lot of jobs. 60% of the jobs that we have today did not exist in 1940, and I think people have a symmetric imagination. It’s easy to imagine the jobs that will be destroyed, but it’s more difficult to imagine the jobs that will be created. But at Bank of America, we are hiring different people on jobs that didn’t exist just a couple of years ago, for all the AI deployment, etc.

So, I think the most important impact on the labor market will be also changing relative prices, because think about a human being as a portfolio of factors, like a financial portfolio that depends on underlying factors. Think about a human being as having two factors: IQ and EQ. Let’s assume that the price of IQ goes to zero, then the relative price of EQ, which is dealing with people, managing people, has to go up.

So you’re going to look for different skills—I’m saying this because it’s important—if you just assume straightforwardly that you’re going to replace a lot of juniors, you’re going to have an apprenticeship problem down the road, because who’s going to be the senior of tomorrow if the juniors don’t develop the critical thinking to be the seniors of tomorrow. This idea that you just can’t ignore the juniors, replace them all, and then say “okay, now it’s not an issue”, because we develop critical thinking. But 5 or 10 years from now, it’s going to be a huge gap. So, I’m not that pessimistic on that front.

Finally, on financial implications, I think there is an obvious one. Nobody knows whether the valuation of companies is the right one. And you are subject to multiple equilibria at this point, and if there is a correction in the stock market – whether it makes sense or not but the correction happens anyway - it’s going to affect directly consumption through the wealth effect.

The second one is I am not that optimistic about the pace of adoption. One of the things that caught my attention as one of the main takeaways in Bangkok was the concerns on cybersecurity, because Bank of America, for instance, is not going to take stupid cybersecurity risk for me to do a PowerPoint presentation in three hours, rather than six, right? So we are actually the most attacked bank in the world, and the reason why a zero success ratio. You might wonder why cyber-attacks continue in a bank that is basically so safe. It’s because cyber attackers try to learn from the defense system in order to attack weaker corporations.

So, this is something the defensive investment in order to avoid cyber-attacks will delay the pace of adoption. The dot-com bubble was a bubble not because those productivity gains never happened, but because it took longer than priced in by the stock market to materialize. So my bearish side of the story is not so much on destroying a lot of jobs or stuff like that. It’s more the pace of adoption can be more slow because companies would be very defensive on that front.

Gene MA:

That affects so many aspects of our life and economy. I give you one example. I know a bank. The CTO, Chief Technology Officer, told me they have a plan to lay off about 1/3 of the IT staff in about five years. This we’re talking about tens of thousands of jobs in a large bank. At the same time, they need to double the spending on the IT system, build two parallel systems to guard against the all kinds of cyber risks.

If I understand you correctly, earlier you also mentioned that the impact of AI means, we cannot afford to be dovish on rate, right? Because competing for capital, we do not see any deflationary impact anytime soon, right?

Then I want to turn to Prakash. I think Asia is right at center of all those shocks, they have shock on energy side and they are the provider of IT products, so the taxi driver of Bangkok was complaining that the gasoline price doubled from beginning of the year and it’s hard to make a living.

Pierre-Olivier just mentioned that the global trade system has refigured and rewired, so that’s why the trade, especially in the Asia Pacific region, is going to the roof. While the taxi driver in Bangkok was complaining, the young engineers at SK Hynix finds it a lot easier to find a date.

So, how is Asia dealing with other shocks? Then, I have another question: how should the global investors position themselves in views of the things we just mentioned?

Prakash KANNAN:

I’ll make two points.

The first one I would make is, to put your question in the right context, I think the single defining feature of the global investment environment right now is a competition for capital, and you are seeing this not only from the fiscal authorities or governments around the world, but you’re also seeing it from corporations, not just the tech corporations, but actually also the broader kind of businesses.

And if we tie that back to Asia, the one interesting statistic is that capital goods imports in Asia is growing at 43% year on year, and this is not just an AI story. When we usually think about this investment boom, the focus is about data center investment in the U.S. But here, the reason why capital goods imports is so strong, yes, there is an AI story, but there’s also a defense investment story post Middle East conflict. There’s a big energy security investment story, and then there still is a supply chain resilience build-out that’s happening as well. So all of this is that this huge kind of competition for capital. It’s a global phenomenon which ties into these points made here that understandably the equilibrium interest rate is moving up.

Now the problem here…Claudio mentioned it a little bit in terms of the K shape. I think there is also starting to be a differentiation. In Asia, the one interesting dimension in this competition for capital is actually also competition across different governments, and here the benefits of the surplus countries relative to the deficit countries is starting to be more evident. Again, it’s not just an Asia thing; you see it in Europe as well. Europe, in aggregate, is a surplus economy, and they are also trying to find ways to mobilize pension savings or different forms of savings towards all these big investment needs. Similarly, in Asia as well, you see the surplus economies finding ways to make capital more productive, even domestically.

The challenge then comes on who’s not on the other side. So, if you are deficit economy, if you are not plugged into the AI supply chain, if you are a net energy importer, I think this is a very challenging environment. There was a mention of this actually at the IMF’s regional economic outlook for the Asia Pacific region, which is that in a way a lot of the positive stories coming out of Korea and Japan mask some of the strains, particularly in the lower income economies, especially if you’re also a net importer of energy. So that’s the Asia side.

On your second question of well, what does an investor do in this environment? I think this kind of age-old principle of diversification, this thing of what’s typically called the only free lunch in investing, remains relevant today. Now I know it sounds a little bit like motherhood and apple pie, but actually diversification as an investment strategy has not paid off in the last 10 years. Capital returns have been extremely concentrated in one country, in one sector, and to a certain extent in a few companies. But I do think that now, there is a time for the benefit again of diversification to return for a bunch of reasons I can go to.

The second thing, which I think is going to be very important, is the pricing of liquidity. The one thing that a lot of investors did, even from post global financial crisis, was that when interest rates were very low, in order for you to hit any absolute return target for a portfolio, you had to move up the risk curve, and part of moving up the risk curve was to take on a lot more illiquidity risk. Now it turned out that was fine. Certain segments ended up rewarding investors, but I think you’re getting to an environment where the optionality of having liquidity in your portfolio, for all the reasons that have been mentioned here, because the investment and environment is changing really rapidly, I think that’s going to force a rethink. Like, how do you bring more optionality in your portfolio? How do you bring more liquidity? How do you get more agility? These things are going to be extremely important.

But having said that, if you think about solving the retirement problem, which is a feature that faces so many countries in the world, the move up in interest rates actually has a silver lining to it. If you think about the actuarial rate of return for a lot of pension funds in the U.S., it’s usually about 7%. You can build a portfolio now that gets you a 7% yield with not taking a lot of credit risk. When I look at a U.S. real interest rate on the tip security, that’s close to 3% real. Now the path to get there has been painful for fixed income investors, but it’s a great starting point to build a portfolio.

Again, part of the problem in the past was that when you had low interest rate, you forced a lot of people to take risk in places that they were not comfortable with. Right now, when you are getting to perhaps a more correct, if you will, kind of cost of capital, I do think if you are a saver, this is actually a good time. Going back to my point, which is that if you are in this position of a competition for capital, you are actually in a very beneficial position if you are a saver and you think you can actually build a very decent portfolio that puts you in a good position.

Gene MA:

Yes, so the current market condition makes diversification and the liquidity more important than before, and also it’s a good time to be a saver than borrower right now. I turn to Dr. Wang. What’s your advice to boost demand? That’s a question people have discussed for ages. Is there any new idea that can boost the demand domestically to balance trade?

Wang Tao:

Thank you. Let me just maybe take a step back before I dive in right into China’s domestic demand. As the previous speakers said, in the last year and a half or so, we had tariffs, we had energy shocks, we had AI boom, and China is in the middle of it. So we cannot really just talk about the domestic demand without looking at the overall picture.

Compared to the beginning of last year, China has done remarkably well, compared to we at that time thought not only the tariff but two other shocks were going to come and how was the AI shock was more about tech export restrictions from advanced economy.

That is partly because, as Prakash mentioned, Chinese exporters among all these tariff uncertainties created by the U.S. were able to remain very resilient. Part of the way is diversify into other markets, exporting capital goods, and of course, another big part is the AI boom. AI boom led to investment in China, especially with the coming out of DeepSeek, a lot of people, including in China, realize there are vibrant companies and innovation and so on.

But of course, recently we have seen the strength of export. A big part has been AI-related, so that AI had a big impact on the aggregate demand in China and on the energy side as well. So, China being the largest importer globally has held up remarkably well.

I’m no in no position to give anybody advice, but if we look at the overall economy, then one probably has less urgency to feel there is a need to do big stimulus now that China seems to be doing quite well in many of the sectors.

Of course there are also vulnerabilities there. The external side things could still turn. On the domestic side, two key things are underlying the current weakness. One is the continued property market adjustment that has not yet bottomed, and we are now in the sixth year. And Chinese households hold most of their assets in property, so that’s a big hit on their wealth. And the second thing is local government finance, which is partly related to the property market. These two things at the moment are negative factors.

To boost consumption in the short term, if the government can facilitate the adjustment in the property market, and the central government can provide more funding or debt restructuring to local government, that can release some of the negative effect coming from that.

But fundamentally, over the longer run, to boost domestic consumption, you need higher income and higher confidence in the household. The confidence can be helped by property sector stabilizing, but the most important thing is also the labor market and labor income.

At the moment, China has seen good growth in many sectors, AI related, export related, renewables, and so on. Manufacturing sector is in general doing quite well, but we have not seen translation into labor market improvement. So on that front, what are the kind of policies that can help boost labor employment? That I fear that is a challenge facing many governments at this moment, especially when you feel you have to let AI develop.

On AI’s application, it’s very rapid in China and creating demand in terms of investment and export demand, but it’s not having as much of a wealth effect as in the U.S. through the financial markets on the household. That’s an area that is worth noting.

Gene MA:

I think two of you mentioned that you’re worried about the financial vulnerability, and in different times we have different kinds of vulnerability. Back in 2008 or 2012, we had debt, U.S. household debt and European debt. These days, we have a rise in the yield and the cyber risk created by AI. I’m sure the crypto community is very concerned about advancing of quantum technology. What are the other vulnerabilities we should be concerned with for the financial system?

Claudio IRIGOYEN:

All the opacity related to private credit (are worth noting).

I mean, Lehman happened because we didn’t understand the cross exposure that significant players had, or when we learned it was too late, right? So I think the next crisis will come from something we don’t understand, not for all the usual suspects that we are thinking about, and that’s the part I think we have less visibility, and the market is complacent because private credit is small until it’s not.

Now, on the financial risks, I was interested in hearing Prakash say, “Okay, well, now if you’re a life insurance company, maybe you don’t have to take too much risk, you can build a portfolio that will deliver something that is closer to 4% or 5% without taking too much risk.” But that’s not what we are necessarily seeing.

For instance, we’re seeing a lot of life insurance companies going into private credit and private equity, and actually being bought by private equity firms to provide the funding to their asset side. And so that’s where I think there are a lot of risks that potentially can be building that we have to watch very, very carefully.

Prakash KANNAN:

No, I think it’s a path versus destination problem, and part of the reason why a lot of insurance companies, pension funds, were forced to take risk at the time was because if you… I mean, rates have moved up only very recently. So when you look back, especially at the 2021-22 vintages, I think that was a period where you really had a lot of push into the illiquid space and areas of opacity. If you were starting to build a portfolio today, I think you’re starting from a very different risk-free rate than you were in that period. But the general point, I think, I agree with Claudio, is that it’s the areas of opacity, which is where I think the financial vulnerabilities are going to be. Private credit is one.

I’m actually also concerned about market making. I think there’s so much in the market making formation within capital markets right now that we just don’t understand. Long-term investors such as us, we’re not the marginal price maker anymore. There’s high-frequency trading. There’s algorithmic trading. AI is also starting to play an increasingly important role, and so those kinds of dynamics, I think, we just don’t understand the amplification mechanisms until you really get into a serious downturn. So that part again is to the general point of opacity. I think that’s what I worry about.

Gene MA:

Private credit and the marketing agent. So we’re running out of time. I surveyed the audience. I didn’t see anyone doze off, and not even many people looking down on their phones, all thanking to our wonderful speakers. Thank you very much. Let’s give a round of applause.

-END OF PANEL-

Year-on-year growth broad fiscal revenue turned negative in September, mainly because a decline in corporate income tax and slower growth in other items pushed general public budget revenue into a year-on-year decline, while a steeper fall in land-sale revenue markedly widened the year-on-year decline in government fund revenue.

The year-on-year decline in broad fiscal expenditure narrowed in September, although the contraction narrowed: general public budget expenditure was flat, while government fund expenditure fell less sharply, supported by accelerated issuance of new local government special-purpose bonds.

Together with a broad fiscal deficit that was smaller than a year earlier, these developments suggest that the broad fiscal stance remains relatively tight.

Based on these data, we offer three observations:

• Cost shocks are squeezing the profits of midstream and downstream firms, weighing on tax revenue growth.

• The new policy on selling completed homes has widened the year-on-year decline in land-sale revenue, further dragging on government fund revenue growth.

• With revenue under pressure, further fiscal support is urgently needed to lift broad fiscal expenditure growth.

Read the full analysis

• Mário CENTENO, Former Governor of the Bank of Portugal; Former Minister of Finance of Portugal

• Fang CAI, Chairman, Academic Committee, China Finance 40 Forum; Member, Chinese Academy of Social Sciences

• Mark PEARSON, Acting Director, Directorate for Employment, Labour and Social Affairs, OECD

• Ekkehard ERNST, Chief Macroeconomist, International Labour Organization

• Yang DU (Moderator), Director, Institute of Population and Labor Economics, Chinese Academy of Social Sciences

Below is a full transcript of the session, which is produced and translated (Prof Cai spoke in Chinese) with the support of AI and has not been subject to the review of the panelists themselves.

Left to right: DU Yang, Mário CENTENO, CAI Fang, Mark PEARSON, Ekkehard ERNST

DU Yang:

Thank you. Welcome to this interesting session, themed “Reshaping Labor in the Age of AI: Where Do Humans Fit In?”

In this panel, we have four distinguished speakers. First is Mr. Mário Centeno, former Governor of the Bank of Portugal and former Minister of Finance of Portugal. Second is Prof CAI Fang, Chairman of CF40 Academic Committee. Third is Mr. Mark Pearson, Acting Director of the Directorate for Employment, Labor and Social Affairs at the OCED. Fourth is Mr. Ekkehard Ernst, Chief Macroeconomist of the International Labor Organization. Thank you for joining this session.

AI has a significant impact on micro-economy, productivity and employment. We will start with the first round of discussion, where each speaker spends about five minutes to talk about the impact of AI on productivity and employment according to your area of expertise. Let me start with Mário.

Mário CENTENO:

Thank you for having me in this panel. I’ll take the following view.

AI, at a very micro level, in terms of employment and its function in the economy, will have a positive impact on productivity. This looks pretty much a no-brainer. That’s why we are adopting it. That’s why we are putting these many resources into AI. As a consequence (of AI) now entering into the labor market, there will be job losses, and there will be new jobs, jobs that we did not even dreamed of, that will be created, given the possibilities that AI brings to the labor market and to the production function of firms.

As a consequence of these job losses and gains in new jobs, there will be displacement in the labor market. And displacement always comes with some form of churning, and churning is the thing that labor economists love the most. We love to see the labor market churn, because that’s where productivity progress occurs at the firm level and also at the aggregate level. So in my view, what is of concern is not as much this process itself, but it is the dimension of the process. So maybe we face too much of a good thing, because we need to think on public policy to respond to these.

AI will change some fundamental things in the labor market: with the gains in information we have, there may be some concern on economic rents that are the result of this process. Why can there be economic rents? Because firms may keep and appropriate a very important fraction of the surplus in the market. And if this occurs, we may see welfare losses in the labor market. The other thing that can be of concern is how much are we asking to the matching function, the function that joined workers and firms, because when we ask too much to the matching function, the final result may be higher unemployment, more inequality, and not necessarily more employment in the market.

So I think it’s important to understand in each market: what is the initial point and what is the process behind it? As of the initial point, most labor markets today work in full employment. Labor market has been the star of this business cycle, all over the place we have more employment, lower unemployment, and labor force participation has increased.

So if we shock the labor market from a situation of full employment, we must be ready for some job losses on net terms. Maybe we are not able to redirect immediately every worker that loses a job to a new position. The same thing is true as of aging. We do know that when there is a productivity shock, the workers that suffered the most with this productivity shock are older workers, those that indeed are capturing a rent in the labor market and firms have the greatest incentives to get rid of If this occurs, and given that older workers are the fraction of the labor market that increased the most employment in recent years (at least in Europe and in the US this has been the case), we again may be facing a problem of what to do with these workers.

DU Yang:

Thank you. Now let me move on to Professor Cai.

CAI Fang:

Let me begin with an overall assessment. We are all currently discussing the impact of AI on employment. Yet, whether in China or in other countries, if my understanding is correct, we have so far seen no clear evidence of employment shocks caused by AI in the aggregate macroeconomic data. At the same time, we can already detect certain early signals and effects from recruitment data and at the micro level. Some of these effects are real—for example, a reduction in the number of new hires in programming. Others, however, stem from employers’ expectations: they have adopted a wait-and-see attitude, neither laying off workers nor recruiting new ones, resulting in a state of stalemate. I believe this is a temporary period of calm. After this, we are likely to see more pronounced manifestations in the labor market. This therefore presents us with a valuable opportunity to analyze the characteristics of AI’s impact on employment and to make the necessary theoretical and policy preparations.

In this first round, I would like to focus on three key words. The first is productivity. From a highly macroeconomic and purely theoretical perspective, if AI can clearly raise labor productivity and if the productivity dividend materializes instantly, then we can use that productivity gain to generate new demand and new jobs. In other words, there will be both job destruction and job creation. We know that AI will certainly raise productivity—whether by accelerating scientific and technological progress and delivering technical efficiency; by enabling us to reallocate resources with less waste and fewer trial-and-error costs, thereby improving allocative efficiency; or by allowing high-productivity firms to survive and expand while low-productivity firms exit. All of these channels for raising productivity can be accelerated by AI.

There is, however, a problem: like any previous technological progress, the application of AI is uneven across industries, sectors, firms, over time, and even across different production processes, labor tasks, and value chains. As a result, when labor input is reduced in one particular chain or link in order to cut costs, job destruction begins immediately. Yet productivity is constrained by overall equilibrium and does not necessarily rise at once. Consequently, we are likely to observe inconsistencies between productivity gains and employment shocks, as well as inconsistencies across sectors. This is precisely what we must address. Moreover, we already see that AI is displacing many relatively high-quality, high-human-capital jobs, forcing the displaced workers to move elsewhere. This type of mobility is not what labor economists would hope to see.

The second key word is the J-shaped trajectory of productivity—this is not my own invention, but a finding from research teams at Stanford and elsewhere. The labor-productivity gains from AI will almost certainly follow a J-shaped path. What does this mean? There will be a period during which productivity does not rise; yet, as I just noted, labor substitution may already have occurred before productivity improves. The labor market cannot wait for the full J-curve to unfold—that is, we cannot wait until productivity has genuinely risen before putting institutional arrangements in place to address employment. We must respond in a timely manner. In a particularly adverse scenario, during the lower part of the J-curve we could see increased job destruction, slower income growth, worsening income distribution, and consequent damage to the circular flow of the national economy, which would in turn hinder the very rise in labor productivity.

My final point is a conclusion: technology itself cannot resolve the socio-economic impacts of AI, especially its effects on employment. Therefore, we must rely on government intervention, together with the participation of finance and other domains, in order to coordinate a solution that steers AI toward beneficial outcomes.

DU Yang:

Mr. Pearson, I have read some fellow colleagues from OECD who talk about the productivity improvement by using AI in OECD countries. Also, we see some OECD studies about the empirical evidence, which support the argument Professor Cai just mentioned. Can you elaborate more to us?

Mark PEARSON:

Absolutely. Good afternoon, everybody. Great to be here.

Usually there’s somebody on a panel like this who will at some point start talking about a jobs Armageddon, and somehow nearly everything that we do in the labor market will be replaced in the future by AI. Fortunately, my colleagues were not, they were a bit more balanced than that, although we did hear some of that argument this morning. And who knows, in the long run, maybe that’s what will happen. But in the short term, when we actually look at the numbers, we’re certainly not seeing it yet.

Mario’s already mentioned the fact that we have record levels of employment and very low unemployment. Looking in more detail, we did a survey this summer between June and August in 14 countries, including the US, Japan, and European countries, to look at what’s happening both of employers—it looks at 8,000 employers—and tens of thousands of employees. What we found is that when you look at the employers who have adopted AI and their employment practices, only 11% of those firms that have adopted AI have reported that they’ve cut employment. When you dig a bit further and ask how many of those reduced employment because of AI, it’s only 9%. In other words, we’re talking about somewhere between 1-2% of all employers that are adopting AI say that they are reducing employment due to AI. Now it may come in the future. We all know examples of jobs that may be entirely replaced by AI, but overall at the moment, the idea of a jobs Armageddon is looking way into the future, if at all.

Now, but as both of my colleagues have already said, saying that there’s no change in the overall number of jobs is not the same as saying there’s going to be no change. There will be lots of change. We do have to be careful, though, about who will be affected. An example I always like to give is if you think in the end of 1970s, office administrators, secretaries, as we used to call them—think about the tasks that they used to do, in terms of taking dictation, typing, making copies of things, putting things in the post…Virtually everything they did in the 1970s has been automated. Yet, if you look in labor markets like the United States, there are now actually more office administrators, secretaries now than they were in the 1970s. Tasks change, or rather jobs change. So the tasks that people do within those jobs will change, and that’s what we’re going to see an awful lot of in the future.

I do want to mention that the distribution issue does matter a lot. We’ve already heard and talked about older workers. I’d like to say something just about younger workers. There’s a lot of talk about entry-level jobs be vanishing because of AI. I think it’s not just AI. If you look at the numbers, that decline in the position of young graduates goes back to about 2014, and it’s probably in those days more to do with things like offshoring, changes in employer practices wanting to look at actual skills, work experience, a kind of gradual fading of the belief in the signaling that we get from formal qualifications. All those things were going on. It may well be that AI is an additional reason why young people are going to find it difficult to enter into employment, and I do think that’s going to be a key thing for our policy response, to make sure that we do have a way into the labor market for young people.

DU Yang:

Thank you. Mr. Ernst, I know the ILO has already done some research on the impact of AI on global labor market, in particular on developing countries, which is quite weak in terms of model development by themselves. What’s the new evidence you see in terms of productivity and impact on employment in South countries?

Ekkehard ERNST:

Thank you very much. I think my co-panelists already had laid the ground quite interestingly, and what we see in terms of employment generation in low and middle income countries, it’s a similar story. Employment has never been so high, unemployment has never been so low across the globe than it is now. The flip side of this is that productivity has not increased, so we don’t actually see all these promises that are being made from AI. We don’t see that in the numbers yet.

What is interesting, you mentioned our studies, then in different countries, what you actually see is that certain companies report specifically, for particular tasks, enormous productivity gains. Your customer service throughput, CV scanning and reviewing, some of these tasks have seen significant increases in productivity, but at the same time, it doesn’t aggregate up.

Now some of it has to do with the fact that companies internally need to still restructure their work. They need to figure out what they can do more and what they can do less with. It takes some time for this adjustment, but a lot of it has to do with an interesting phenomenon, which is that the gross time you save on certain tasks is not the net time, partly because of hallucination, partly because you still need a human in the loop to verify certain accounts. And so this “verification tax” is actually quite important.

In addition, a lot of these tasks that are automatable (take) only a short amount of time in your overall day. When you take a software engineer for instance, software engineering is a typical occupation that seems to be targeted particularly by AI. The actual coding is maybe a third of the time as a software engineer does. A lot of time is used in meetings, in planning your tasks, et cetera. A lot of the time is not spent specifically on coding, and so AI doesn’t really help with all of these other tasks. That’s called the Jagged Frontier (issue), and it really explains to a large extent what we call the AI aggregation paradox.

Now here’s the trick, and you mentioned the developmental model, which I think is an important element to take into consideration. What we know from the exposure measures of AI is that it targets mostly high-skilled employment, which is why the concern was mostly in high income countries where there’s a larger share of this. But it also targets specific customer service exports, for instance in India or in the Philippines, where a lot of the client services are being exported. They’re not consumed domestically.

If these high skilled service tasks can be automated, and parts of them actually can, that means that the countries have less potential to have this traditional path of structural change, where you would move from low value added, low skill jobs to high value added, high skill jobs. I call it premature de-sophistication. It’s a similar story that we had previously with industrialization paths in certain emerging countries. You see the same phenomena emerging now in services.

In the move towards high-skilled services, there are roadblocks increasingly from AI. For the moment it doesn’t have an aggregate effect, but that’s actually a problem, because it does prevent a country from growing faster, creates severe conditions for certain groups on the labor market to find gainful employment, and will slow down the path of structural adjustment, maybe not so much in China, but certainly in countries like the Philippines, India, and certainly low income countries in Africa as well.

DU Yang:

Thank you, Ekkehard. Now let’s start our second round discussion about policy.

Mário CENTENO:

Let me start with a parallel with the previous technological evolutions we experienced. Labor economists spent decades discussing the impact of skill-biased technological change in the labor market. It is rather funny that we can keep the same skill-biased technological change, but we have to reverse everything inside the analysis.

The skill-biased technological change of the past was against low skilled workers, because demand for skills increase, and now we have a revolution that eat particularly the more skilled people in our society. Eka was giving the example of computer programmers. We sent dozens and dozens, thousands and millions of people to our universities to be trained as computer programmers, and now suddenly there’s a shock in the demand of these people.

And Mark mentioned an event of the 1970s. I have another one that I think is useful to figure out what is going to happen and where policy should be directed, which is what happened with the demand of physics in the 1970s. During the [international] race for the space, the better minds of the US were sent to study physics and to have PhDs in physics, and suddenly this race stops because of the lack of money. There was a crisis: at the same time that we were increasing the supply, the demand shrank.

And what happened? Well, unemployment, lower wages. These people had to be directed to other occupations. The good thing, if there is anything good in this story, is that these people were the more skilled and the better students in American universities, so they did not find it very difficult to get another job. But that is where policy should be directed. There are three things.

One is inequality. Massive displacement and churning may imply that if inequality for those with a job increases, and it is very likely to be increased because these top productivity jobs will pay very high wages, whereas those jobs that are not affected by AI will remain with the same wages, inequality for those with a job will increase. But it will increase even more if we take the population as a whole, because there will be people that will need to find different jobs or no job at all. So inequality is something that we need to consider, and we know how to do that, because our economies and societies are used to implement redistributive policies. Maybe here again it’s the scale that may [scare] at least myself.

The second is education, [which is] the signal that we sent on what is needed by the labor market. What is the impact of digitalization on the quality of education, because this is the more pure input to the labor market? We have seen with some concern: at least for the OECD average, a reduction in the quality of the exams, the PISA, is very clear on that, and that is probably connected with digitalization. Will it exclude people in terms of quality from the labor market, because that’s the essence of the so-called matching function? So we need to also put some attention on the education part in terms of policies.

The final one is intergenerational issues. We suffered a big social crisis. We are still seeing the result of it, because at some point the young generation concluded that it will not be better off than the previous generation. If this is true, this intergenerational problem will become even harder in social terms, [potentially leading to] social unrest. So we need to, in terms of policies, think very carefully of what are we offering to these young generations. As Mark was pointing out, I think it’s very important to think of them.

So I will put the emphasis on redistribution, education and the quality of education, and for what are we educating our children, and then on this smooth transition in the labor market.

DU Yang:

Thank you. Prof Cai.

CAI Fang:

We have discussed at length the current degree of impact of artificial intelligence on the labor market, yet the supporting data do not appear particularly strong at present. We must nevertheless recognize that the development of AI is advancing by leaps and bounds. What we observe happening today is by no means what will occur tomorrow; the high-exposure occupations calculated today reflect only today’s situation. High exposure does not mean that the impact has already materialized. Therefore, the high-exposure estimates made today may no longer hold tomorrow. In this sense, I believe we cannot accurately foresee what will happen in the future, nor can we precisely estimate the scale of the shock.

From this perspective, it is of great importance to make policy preparations now and to build an institutional platform for response. Whether through government intervention, policy measures, or the participation of various macroeconomic policies, all efforts should be directed toward common goals and form coordinated instruments. On this basis, I put forward four recommendations.

First, we should focus on promoting equity in access to artificial intelligence. In every era, the emergence of new technologies creates divides and produces so-called K-shaped polarization; the divide brought by AI may be unprecedented. In China’s labor market, a large part of the structural employment contradictions is concentrated at the two ends—“the old and the young”—that is, older workers and young job-seekers. For older workers, the technological gap they have never previously faced is likely to widen further; for young workers, the entry-level skills they previously possessed are being devalued by AI. Therefore, addressing structural employment contradictions by starting with these two groups has become an urgent priority.

At the same time, we must recognize that if the impact of AI today already appears as a technological divide, then we can expect that tomorrow’s impact will be different and may affect an even broader range of people. Employment-support policies must therefore be continuously upgraded; this is a key point. Moreover, promoting equity in AI is also a key to ensuring that labor productivity actually rises, and it helps avoid the so-called Solow paradox.

Second, we must guide artificial intelligence properly. AI researchers raised the so-called alignment problem from the very beginning. At present, this requirement for alignment should be expanded—from the technical domain to the economic and social spheres, macroeconomic policy, industrial policy, social policy, and every other relevant field—so that AI is required to achieve alignment across the board.

It is especially important to recognize that the AI we discuss today consists only of foundation models. What will truly take effect and influence our lives in the future are the practical applications of these models. When they are transformed into embodied intelligence, and when humanoid robots truly acquire intelligence, their impact on workers is likely to be the greatest. Therefore, throughout this process we must fully uphold guidance, incentives, and regulation, and firmly hold a bottom line: from the employment perspective, AI must become an enhancer of workers’ capabilities rather than a destroyer of workers’ jobs. In this sense, there is much we can do.

Third, how should human beings be placed? Although this may appear somewhat forward-looking, it is already a topic of urgency today.

We should focus on proactively cultivating those industries that have high demand elasticity but currently low labor productivity—many service sectors, for example. Since the purpose of using AI is to raise labor productivity, why encourage low-productivity industries? Because raising overall labor productivity is precisely what creates room for high-demand-elasticity industries to survive; we cannot allow the development of these industries, the creation of jobs, employment, and incomes to become decoupled from individual labor productivity. “How to place human beings” is both a direction for future efforts and a firewall for today—an important mechanism for buffering employment shocks. To this end, we can create new demand, open new markets, and generate new jobs, all of which stem from the comprehensive development of human beings.

Fourth, judging from China’s current situation, my personal assessment is that the impact of AI on employment has already begun to appear. Although it has not yet manifested on a large scale, it has already become intertwined with structural employment contradictions and labor-market mismatches. These structural contradictions will intensify further, and they also reflect the negative effects of AI on employment.

Accordingly, we should start from structural employment contradictions to address the future shocks brought by AI. The most important point is how labor-market institutions—such as the minimum-wage system, the minimum living-security system, and the labor-contract system—can be linked to new forms of employment. In the past, both labor-market institutions and social-security systems in China were closely tied to unit-based employment. As employment forms become increasingly flexible and the share of non-unit employment gradually rises, we must explore new institutional arrangements so that labor-market institutions, systems for protecting workers’ rights and interests, and social-security systems can adapt to the new forms of employment. This involves many urgent goals of institutional construction.

At the same time, our systems of social protection, social security, and social welfare should become more inclusive. In the process of AI’s impact on the labor market, it is becoming increasingly difficult to determine whether the shock experienced by an individual is due to personal factors or macroeconomic ones. Therefore, to ensure that everyone receives adequate protection, the inclusiveness of the system must be continuously strengthened. In this sense, since the 15th Five-Year Plan emphasizes investment in people, the resolution of structural employment contradictions and the response to future AI shocks should also be incorporated into the overall strategy of investing in people.

DU Yang:

Thank you, Professor Cai. I believe we have just heard a very important point: in the age of artificial intelligence, the Baumol effect may no longer be a “Baumol’s cost disease,” but rather a mechanism through which people share in the productivity gains brought by AI. I now invite Mr. Pearson to speak.

Mark PEARSON:

There’s been some super important suggestions already, and I agree with all of them. I can’t go into huge detail on all the policies that I’d like to mention, so let me quickly mention three very important ones.

First is giving guidance to employers on how they introduce AI, because evidence already is showing that when it’s done well, for example with consultation with workers and training, you’ll get much better effects than if you do it without those things.

Second, support displaced workers. We know there’s going to be churn in the labor market. We know people will lose jobs and new jobs will be created. Some countries do much better than others in enabling that churn to be positive.

Third, early career pathways. We’re going to need things like new apprenticeship systems to be developed on a much more widespread scale than they currently are.

But the main thing I do want to mention is skills policies and what sorts of skills we actually need, because it’s not really just the technical skills in AI that we need. When you look at the professions most exposed to AI and scrape from the web (to see) what skills are being asked for by employers, you find things like data analysis and interpretation skills, digital skills—I don’t think either of them is hugely surprising—but also creativity and problem solving. So we’re talking about skills that you might not expect AI to have: a mixture of skills that help people use AI better, but also skills that enable them to do things AI cannot to do. And that’s really the direction our skills development needs to go in.

When you again look at how the general AI literacy of the workforce, it matters enormously. This should surely be the main aim of our skills policy—not developing technical skills in AI, but simply making sure that a large number of workers can use AI effectively.

Our surveys show that 89% of workers say AI has improved their productivity in companies that provided training, compared with just 69% in companies that did not provide training when AI was introduced. You can already see the effects of training, yet only half of employers introducing AI actually provide any training. It boggles the mind. You can’t just introduce a big new technology like AI and expect people to get on with it. It won’t work as well as if you provide some training.

There is a big move into developing more AI-relevant training, and that’s great. However, the way it’s being done will embed the existing inequalities in training: the people least likely to receive training at the moment are low-skilled people. In OECD countries, about 20% of low-skilled people take part in training each year, compared with around 50% of high-skilled people. If we’re going to introduce AI without increasing inequality, we need to make sure that people with lower skill levels, who are less likely to take advantage of existing training possibilities, also can take advantage of AI training. Otherwise, we’re going to get the inequalities embedded.

And we know what we should be doing. We need short courses, modular certificates awarded at the end of each course, and certification of the quality of training providers. We already know a lot about how to make training more effective. AI means that we need to do that much, much more urgently than we have done in the past.

Ekkehard ERNST:

I wanted to offer three complementary points to what we just heard in terms of policies.

First, there seems to be some agreement among all of us that transitions will be important and that we need to support them. But what’s surprising is that we’re actually seeing a decline in job-to-job changes, job hopping, or job-to-job transitions. By the way, this is not just an issue in advanced economies; it’s also true in many emerging economies.

The question is: where does it come from? There is an element that is really policy relevant, which has to do with an increasing labor market concentration that we see across countries and across income levels. This is not only related to AI, because it started long before the generative AI models we have today.

It has a lot to do with company policies. We see non-compete clauses. Professor Cai mentioned platform work, where people are falsely classified into non-employment status. So there is a policy-relevant issue here that needs to be tackled in order to help people to be able to move from one job to another.

I’m proud to say that the International Labor Organization has adopted this summer a new platform work directive that will help countries implement legislation that will clearly define the employment status of platform workers. So there is already a small element that will help clarify how employees should be classified in terms of social security contributions, occupational safety and health, and other labor-market-relevant regulations.

That’s my first point. We need to identify where it comes from that we don’t see these labor-market transitions, and that there are policy-relevant issues that need to be tackled.

The second point is on skills. We should take advantage of AI as an infrastructure to help us promote a more integrated, forward-looking type of skills policy. Reskilling policies that have been implemented have actually been surprisingly ineffective, and part of it has to do with the fact that we tend to look only at the next job, or the next [incremental] type of activity that one can do.

Singapore has an interesting approach to that. It identifies occupational pathways through the labor market that will help workers and employees to identify not only where the next job is, but where the next job in five or ten years is. That helps prepare the person in terms of the skills and competencies that need to be acquired. Ideally, these efforts would then be integrated into an institutional environment that provides not only technical skills but also broader, fundamental competencies.

Resilience is the buzzword of the day, but it’s more than a buzzword. It actually helps people adjust, especially in combination with an activating social welfare state. We have very good examples in Europe. Denmark is one, Switzerland is another. Both have an activating social security system. You not only receive replacement income if you lose a job, but also the full battery of incentives for you to look for the next job.

The last point I want to make relates to my initial remarks about what countries actually need to be concerned about, especially in the emerging world, where they are transitioning to higher-value activities. We need to identify how to increase productivity in those sectors and segments of the economy that have traditionally been classified as low-value-added and low-productive sectors.

I’m thinking in particular of one sector: healthcare. Healthcare is a highly labor-intensive activity, and it is the only sector that has consistently seen employment growth over the last 20 to 25 years. Demand for healthcare services is increasing, and it will continue to grow as people get older and older. We need to figure out how to provide healthcare services in this sector much more effectively and productively.

Related to it is one point that is particularly dear to my heart. People are actually busy. They might not be in employment, but they are doing a lot of things that are not remunerated by the labor market. I call that activity. So we need to shift our policy focus from full employment – we are currently close to full employment – to full activity where social and care activities are properly remunerated. Again, AI as an infrastructure tool can help us identify where these people are adding value to the society. These are socially valuable activities.

And as Mark pointed out, AI can certify and track these activities and help people get real social value from them. I believe there is a possibility for countries to focus on not only the top line sectors – finance, AI, industry – but really have a whole battery of sectors that are currently considered to be low-value-added and low-productive and make them more attractive and productive in the future. Thank you.

DU Yang:

Thank you. Since we still have some minutes, I have a question for all of you. It’s a big question. Since the first industrial revolution, we have seen that technological progress has always been labor saving. This time at the age of AI it’s the same. AI saves labor.

But we also see that productivity and employment have grown a lot since the first industrial revolution. So what can we learn from the history in the age of AI?

Mário CENTENO:

I think you already answered the question. It’s going to be more employment.

DU Yang:

What’s the difference and what will be the same?

Mário CENTENO:

Composition will be totally different. We need to create institutions that support that progress. Daron Acemoglu refers to pro-worker AI. I’m not going to discuss it because honestly, I did not understand completely the concept, but there are regulations that work towards pro-worker AI. There are two institutions very important and interlinked that we did not mention so far and that are important to also discuss, and I’m going to be very brief, because this will probably entail a big discussion, which are unions and collective bargaining.

The AI market is a completely atomistic market. When I mentioned in my first intervention the appropriation of rents by firms, it occurs in completely atomistic labor market in which the firm knows exactly, facing each worker with an agentic AI element, how many hours one works and which days of the week a specific worker prefers to work. We face a complete asymmetry of information from the side of firms and workers facing this very, very atomistic labor market. This is probably the biggest concern in terms of the equilibrium wages and the capacity of workers that, over history, as you were very rightly pointing out, were able, and we all altogether, to introduce technology and new institutions in the labor market. Two of them I see at great risk, which are unions and collective bargaining.

CAI Fang:

What is different about the potential employment shocks we face compared with those in history? I believe the greatest difference is that historical employment shocks were someone else’s history, whereas what we face today is our own history and our own reality. Historically, there was indeed a time lag between job destruction and job creation. In the history books and from the data, however, we may overlook this lag and conclude that new technologies eventually create more jobs. Today, when we encounter job destruction, if new jobs of comparable quality and quantity are not created in time to absorb the displaced, we will face structural employment contradictions and possible declines in income. Therefore, the most realistic imperative for us is that we must know what to do before job destruction has even occurred.

Turning to specific characteristics, we can consider three comparable situations. The first is the Luddite movement during the First Industrial Revolution. The background of the Luddite movement was mainly that the relatively high-end, reasonably well-paid jobs previously performed by handicraftsmen were destroyed. At the same time, many more jobs were created that anyone could enter and perform—jobs that required only cooperation with machines. This absorbed large numbers of cheap laborers, the industrial reserve army, as well as women and child workers. Under these circumstances, wages did not rise and the condition of the working class deteriorated. Until improvement finally came, society remained in a relatively poor state for nearly half a century—what has been called the “Engels Pause.”

The second comparable situation is the period when the United States and Europe experienced the impact of globalization. During this period, large numbers of skilled and semi-skilled manufacturing jobs at the intermediate level were lost, and the labor market tended toward polarization. Some workers moved into lower-tier jobs with falling incomes; others may have entered higher-tier service occupations. Employment was ultimately resolved, yet income disparities widened and generated popular discontent.

This round of technological change driven by artificial intelligence, in my view—at least for emerging market economies—also begins by substituting relatively high-end jobs and leaves workers with no retreat. That is, when workers are displaced from relatively high-end positions, there are no better jobs available for them to move into; they can only move downward into lower-tier jobs. The result of this downward movement is an overall decline in labor productivity, which runs counter to the structural path of labor-productivity evolution. At the same time, the condition of workers may deteriorate. Historically, even under the most ideal circumstances, the process of employment recovery and adjustment took a very long time, during which workers’ living standards suffered. If we rely solely on the self-adjustment of the labor market today, many workers will be harmed and the process will take far too long; society cannot afford to wait. Therefore, compared with any previous period in history, the present moment has an even greater need for government policy intervention and institutional reform.

Mark PEARSON:

I suppose we have a paradox. This is one of the big lessons from the history. The paradox is that if you look at the potential increase in real incomes over the next decade due to use of AI, our estimates are between 0.15% per year and 0.95% per year over the next decade - really big gains compared to what we’ve seen happening to real incomes in the recent past.

Similarly, I just mentioned earlier that only less than 2% of AI adopting firms are saying they’re reducing jobs. What I didn’t say is that there’s an awful lot more AI adopting firms say that they’re increasing employment due to AI, so there should be a lot of positivity about the labor market impacts of AI. In fact, when you look at what workers think, 16% of workers in OECD countries are very scared that they will lose jobs due to AI in the next year. 40% are moderately scared that they will lose jobs in the next year. So there’s this paradox of potential being in there for real benefits to workers, and yet they’re very scared. It comes back precisely to the fact that there will be churn in the labor market, and workers don’t trust the institutions and policies that we currently have to help them navigate those changes.

To be honest, I think they’re right. I think you’ve heard the reasons why they’re right during this discussion. We don’t currently have the policies in place that will help navigate that impact that many of them will be facing, due to churn in the labor market. I think the lesson from history, because this is a very similar pattern, I think you could say, as Professor Cai has already talked about. There is this gap between when the initial impact comes in place and then before you see the benefits more generally. That’s common practice. We need the institutions. It took us 50 years to get the institutions during the first industrial revolution. We can’t afford 50 years to manage the AI transition more effectively.

Ekkehard ERNST:

There’s one big difference with respect to historical patterns with this AI transition, which is that for the first time, the benefits of these new technologies are being appropriated by a small number of companies and will not be eroded away, at least not under the current institutional framework. I’m very much in line with Mario on this. The distributional issue will be much more important and much more persistent to tackle. Even though I’m coming from the International Labor Organization, I’m not convinced that collective bargaining and trade unions can help precisely because of that high concentration of wealth that accumulated in a few number of companies.

What I believe needs to be done is an overhaul of our legal framework that allows us to create completely new markets. Daron Acemoglu recently proposed, for instance, creating data markets so that other companies can use the data that is already out there to develop their own models and to help these rents that are being appropriated by a few number of companies being distributed across the labor market.

The other proposal that I always make is, and interestingly Anthropic has earlier this week offered a kind of a first step to that solution, is that we need to have sovereign digital wealth funds, where governments can actually invest in these companies and benefit from this massive amount of wealth accumulation in order to then help distribute it, either through direct payments to citizens or through indirect payments to infrastructure development, etc.

There are possibilities for doing that, but we need to reinvent our institutions to tackle that particular problem, which is the high concentration of wealth that comes with a data infrastructure that is very much geared towards only a few companies that are currently benefiting from this technology.

DU Yang:

Thank you. Since we are running out of time, please join me in thanking all the speakers.

-END OF PANEL-

In a recent Bloomberg TV interview during the 2026 Bund Summit, Guo Kai, President of CF40 Research, discussed the findings of “China’s Economy: Toward a New Paradigm”, our annual flagship report examining China’s changing growth drivers and the evolving role of the property sector, manufacturing, and the new economy.

Watch the Full Interview

Since 2021, CF40 has launched a major research project on green finance each year. The theme of this year’s Report is “Green and Low-Carbon Synergy and Sustainable Finance.” Wang Xin, Director-General of the Research Bureau of the People’s Bank of China and a CF40 member, served as project leader. All project team members participated in a personal capacity.

The Report takes the deep interlinkages between the green and low-carbon transition and China’s domestic and international macroeconomic and financial developments as its central thread. Moving beyond the traditional “green for green’s sake” perspective, it advances a path of “synergistic development” to address the complex landscape of multiple objectives and overlapping risks, providing a systematic framework for financial support for green and low-carbon development.

Following the release, Wang Xin held a thematic discussion on “Looking Ahead to COP31: How to Advance Climate Action?” with Agnès Bénassy-Quéré, Deputy Governor of the Banque de France; Pan Jiahua, Professor of Economics at the Hong Kong University of Science and Technology (Guangzhou) and Vice Chair of China’s National Expert Panel on Climate Change; and Jean Pisani-Ferry, Senior Fellow and Founding President of Bruegel and Professor at Sciences-Po Paris.

The Significance of Advancing Green and Low-Carbon Synergy in the New Context

Green and low-carbon development has become deeply embedded in macroeconomic performance, financial risk prevention and global governance, reflecting its inherent features of cross-sector relevance, multi-stakeholder interaction, and coherence between domestic and international efforts. At present, intensifying geopolitical tensions, faltering multilateral governance mechanisms, sluggish world economic growth and frequent climate-related disasters are exerting more severe shocks on economic and financial stability. Facing multiple objectives and overlapping risks, a coordinated overall solution is required.

The Report points out that promoting green and low-carbon synergy serves several strategic purposes simultaneously: facilitating growth, mitigating risks, conserving ecosystems, improving livelihoods and strengthening international cooperation. Its significance can be understood at three levels:

• At the level of international cooperation and governance, the low-carbon transition of energy systems is critical to energy security; international green trade and green investment and financing open up new space for international economic and trade cooperation; and international cooperation on sustainable finance contributes to advancing reform of international economic and financial governance.

• At the level of domestic economic and financial stability and development, amid the structural contradiction of strong supply and weak demand, the transition of high-carbon industries, climate adaptation, and the conservation and sustainable use of natural ecosystems hold substantial potential for investment, employment and consumption, offering a new path to resolve current economic difficulties and safeguard financial stability.

• At the level of natural ecosystems, climate mitigation and climate adaptation, as well as climate action and natural ecosystem conservation, reinforce one another; financial support can work in the same direction to deliver synergistic gains.

Centering on the links between the green and low-carbon transition and macro-financial development, the Report constructs a progressive, multi-layered analytical framework.

The first layer concerns internal coordination within green and low-carbon development itself, encompassing synergies between climate mitigation and climate adaptation, as well as synergies between climate action and natural ecosystem conservation. It emphasizes the application of nature-based solutions (NbS) and nature as infrastructure (NAI), and innovatively brings financial support for the circular economy into the research scope. The second layer concerns coordination between the green and low-carbon transition and domestic macro-financial development, covering the scientific decomposition of carbon-reduction targets, the use of market-based mechanisms to promote carbon abatement and backward production capacity cutbacks, the expansion of climate adaptation investment and green consumption, and the assessment and prevention of climate- and nature-related financial risks. The third layer addresses coordination from an international perspective, focusing on green trade, cross-border green investment and financing, international financial governance and international just transition, in particular leveraging green finance to advance reform of international economic and financial governance. The Report specifically notes that climate adaptation research carries substantial weight at each layer, while financial support and risk prevention run throughout the entire framework.

Climate-Nature Synergies in Green and Low-Carbon Development

The Report holds that climate mitigation and climate adaptation are intrinsically linked and must be advanced in tandem. For a long time, global climate efforts have centred predominantly on mitigation focused on carbon reduction. Yet as extreme heat, heavy precipitation, droughts, wildfires and other hazards grow more frequent, the importance and urgency of climate adaptation have become increasingly evident. At the same time, climate action and natural ecosystem conservation are inherently aligned: climate action helps safeguard natural ecosystems, while healthy ecosystems sequester carbon, expand carbon sinks and strengthen climate resilience. The Report emphasizes that nature-based solutions (NbS) and nature as infrastructure (NAI) are important levers for realizing synergistic climate and nature benefits.

At present, climate-nature synergies still face prominent shortcomings. Investment in climate adaptation remains severely insufficient: adaptation finance accounts for less than 30 percent of international public climate finance, while funds directed to combined mitigation and adaptation represent only 11 percent. Relevant policy and governance arrangements remain fragmented, with fiscal funds managed along segregated budget lines with earmarked uses and insufficient synergy orientation. Supporting incentive mechanisms are inadequate, with a stark mismatch between long-cycle, weak-cash-flow projects and the short tenors of financial products.

The Report recommends strengthening top-level institutional design and accelerating the establishment and application of climate adaptation finance standards and biodiversity finance standards; establishing dedicated libraries for climate adaptation and ecological projects alongside project-benefit assessment frameworks; further aligning climate-financing instruments and developing carbon-sink and ecological-value trading; and strengthening international cooperation and capacity building.

Synergies between Green and Low-Carbon Development and the Macroeconomy and Finance

On synergies between green and low-carbon development and the macroeconomy and finance, the Report focuses on the decomposition of carbon-reduction tasks, market-based backward production capacity cutbacks and carbon abatement, climate adaptation investment, green consumption, and financial risk prevention.

1. Scientific decomposition of carbon-reduction tasks

Based on simulations drawing on historical responsibilities, development stages and abatement potential, the Report proposes a potential pathway for China’s aggregate carbon emissions over 2026–2035 and builds emission-reduction trajectories along both regional and sectoral dimensions, providing a reference for localities in formulating their emission-reduction action plans. Under the regional decomposition, once the national carbon-allowance cap enters a downward trajectory in 2030, allowances for eastern provinces would decline first, those for central regions would register moderate falls and those for western provinces would remain broadly stable. Going forward, abatement priorities would focus on regions with large economic output or concentrated fossil-energy production. Under the sectoral decomposition, the power, steel and cement sectors represent the core abatement priorities and are projected to account for 65–70 percent of total national emission reductions over 2026–2035.

2. Using market-based mechanisms to promote synergy between backward production capacity cutbacks and carbon abatement

The Report notes that sectors with backward production capacity often face considerable carbon-reduction pressure, and that backward production capacity cutbacks can be paired with carbon abatement. Operationally, it recommends that central authorities establish a special purpose vehicle (SPV). Emission-reducing enterprises would apply to competent industry authorities for certified emission reductions; the SPV would purchase the emission-reduction certificates and disburse compensation payments to enterprises promptly, thereby jointly advancing backward production capacity cutbacks and carbon abatement. Funding sources for the SPV could include transition bonds, loans, and local or sectoral capacity-exit funds. As carbon prices rise and the certificates appreciate in value, the SPV could monetize them on the carbon market, making the mechanism self-sustaining. To illustrate, a hypothetical 10 percent backward production capacity cut in Hebei Province’s steel and cement sectors in 2025 could deliver total emission reductions of 22.259 million tonnes, with the SPV disbursing RMB 1.558 billion in compensation payments.

3. Scaling up climate adaptation investment and unlocking green consumption potential

The Report estimates climate adaptation investment needs for the 15th Five-Year Plan period (2026–2030) in three representative sectors—agriculture, urban construction and health care. Total agricultural climate adaptation investment needs are estimated at approximately RMB 1.05 trillion over the period, or roughly RMB 210 billion per annum. For the 39 deepened pilot climate-resilient cities (districts), total climate adaptation investment needs are estimated at roughly RMB 300–400 billion, or approximately RMB 60–80 billion per annum; extrapolated nationally, annual investment requirements are estimated at around RMB 400–540 billion. Health-care sector climate adaptation investment needs are estimated at approximately RMB 1.88 trillion, or around RMB 380 billion per annum. The Report estimates that improvements to China’s basic medical-insurance system can lift national consumption by roughly 7 percent.

On green consumption, China’s green-consumption market reached RMB 12.18 trillion in 2025, accounting for 24.3 percent of total retail sales of consumer goods. The Report projects that the green-consumption market will exceed RMB 18 trillion by 2030, with its share rising to around 30 percent. It recommends accelerating the development of green consumption finance standards, exploring e-CNY and smart-contract applications to incentivize green consumption, and studying the inclusion of green consumption finance within the scope of the existing relending facility supporting small businesses.

4. Aligning green and low-carbon development with financial risk mitigation

The Report systematically reviews the latest research findings and domestic practices on climate physical risks, transition risks, and nature- and biodiversity-related risks, with particular attention to physical risks and emerging risks, such as the energy and water consumption associated with large-scale AI deployment.

The authors note that many of China’s computing-power centers are deployed in western regions, some of which face water scarcity; the related risks and challenges warrant close attention. Further in-depth research is also needed on the climate- and nature-related risks facing China’s cross-border investment. A significant share of China’s outward direct investment is concentrated in climate- and ecologically vulnerable industries and regions where local climate adaptation capacity is relatively weak; climate and ecological issues may also trigger international litigation and other consequences, transmitting risks to “going-global” enterprises and financial institutions. The Report recommends deploying catastrophe insurance, catastrophe bonds and other instruments to build multi-tiered risk-diversification and risk-sharing mechanisms; integrating climate- and nature-related risks into macro-prudential management; strengthening cross-departmental coordination; and advancing international cooperation on climate adaptation.

Green and Low-Carbon Synergy from an International Perspective

From an international perspective, green and low-carbon synergy encompasses several dimensions, including green trade, cross-border green investment and financing, sustainable finance and international cooperation, and international financial governance.

The Report notes that green trade and cross-border green investment and financing serve the fundamental interests of all countries and open new space for international trade and investment cooperation. Expanding the use of the RMB in developing green trade and cross-border green investment and financing can lower settlement costs for trading partners and financing costs for offshore entities, expand the supply of global safe assets, and reduce economies’ reliance on any single sovereign currency. Substantial opportunities exist in the energy transition, particularly in grid construction in developing countries. The Report recommends building a diversified, multi-tiered green trade financing system, improving the matching of financial services, deepening international cooperation on green investment and financing, actively participating in the formulation of global green trade and economic rules, and opposing green protectionism.

On international cooperation and international financial governance, the Report holds that sustainable development has become a global consensus and trend, and that financial support for sustainable development is both necessary and urgent. Sustainable finance is closely linked to the four major pillars of international financial governance—the international monetary system, the global financial stability architecture, the governance of international financial organizations, and cross-border payment systems—and is expected to become an important driver of reform in international financial governance.

Specifically, in the international monetary system, the green and low-carbon transition has spawned carbon trading, the pricing and settlement of critical resources, trade in green goods and services, and investment in high-grade sustainable finance instruments, tying the core functions of currency more closely to “green attributes”. In the international financial stability architecture, climate- and nature-related risks should be more comprehensively and systematically integrated into financial stability monitoring and assessment frameworks. In the governance of international financial organizations, economies committed to sustainable finance development and international cooperation could strengthen their voice and influence. In cross-border payment systems, the demand driven by cross-border green investment and financing is pushing payment infrastructure towards higher efficiency and greater diversity; promoting interoperability among central-bank payment systems and exploring applications for new instruments, including central bank digital currencies (CBDC) and smart contracts, could better drive global sustainable development.

The Report emphasizes that green trade and cross-border green investment and financing are also important channels through which sustainable finance can support international just transition. Traditional climate negotiations have tended to emphasize financial support from developed to developing countries and the scientific allocation of emission-reduction targets. The Report argues that, by developing green trade and cross-border green investment and financing, China can further strengthen coordination and cooperation with developing countries under frameworks such as South-South cooperation and the Belt and Road Initiative, support international just transition through green investment and financing, provide developing countries with more technology, funding and capacity-building support, optimize global arrangements for abatement cost sharing, and build inclusive, equitable global green transition partnerships.

He also proposed moving loan pricing from relying predominantly on the loan prime rate (LPR) toward a more diverse set of benchmarks, and gradually phasing out benchmark deposit and lending rates for financial institutions.

We offer 4 observations on these proposals.

First, shifting the tenor of the central bank’s open market operations from seven days to overnight is an important step in the market-oriented reform of the interest rate management system.

Second, lending rates will become further differentiated: the LPR will remain the benchmark for residential mortgages, while corporate loans may shift to pricing at a spread over DR001.

Third, establishing a liquidity support facility for nonbank financial institutions is an important addition to the central bank’s liquidity support framework.

Fourth, changing the interest rate anchor for open market operations follows the logic of earlier reforms. It should not be interpreted as a disguised rate cut, nor does it imply a change in the monetary policy stance.

Please see here for more in-depth analysis.

The discussion covered the economic impact of AI, potential circular risks in AI investment, U.S.–China technology competition and AI safety cooperation, global imbalance, tariffs, and monetary policy independence.

Below is the full English transcript of the dialogue.

Transcript

Huang Yiping:

Good evening, Bob. Welcome back again. I’m delighted to have the opportunity to talk to you.

At the 2025 Bund Summit, you mentioned you were taking AI tutorial sessions twice a week. I’m not sure if you’re still continuing, but I’m reading a lot of the articles and speeches by you thinking about AI. So I would like to start with questions about AI.

You have expressed positive views on AI as a general-purpose technology in reshaping the economy, especially in lifting productivity, improving fiscal conditions, and so on. You also said that the Solow paradox is not applicable to AI. I wonder if you’d like to elaborate a little bit on the impact of AI on the economy, especially on growth outlook and fiscal conditions.

Robert Rubin:

I think my view is a little bit more nuanced and complicated, Yiping. I do think AI could create a lot of productivity effects. But the other side of that coin is the potential for large-scale job loss. Or even if you don’t have job losses in terms of the total economy, there is a serious – actually high possibility that a lot of knowledge workers will lose their positions and have to transfer into something else, and the transition complexities could be enormous, including the fact that the knowledge workers don’t have the skills for other kinds of jobs. It’s very uncertain as to what effect it will have on growth, taking everything into account that I just said, and what effect it will have on our fiscal conditions.

Huang Yiping:

This reminds me of some literature after the First Industrial Revolution. An economic historian at Oxford called Robert Allen did a study looking at income distribution after the First Industrial Revolution. He used the term “Engels’ pause” to describe the phenomenon that the real output per worker increased significantly while real wages stagnated for some time. I guess that’s one of the issues we might also be concerned with.

Robert Rubin:

Absolutely, Yiping. It could well be that AI will increase the income to capital but reduce it to labor. That not only has an income distribution effect, but also has an effect on tax revenues, because taxes are much higher on labor than they are on capital.

Huang Yiping:

That’s true. One implication is that it might affect income distribution and therefore the welfare of the people, especially the working class. The other thing that I also worry about as a macroeconomist is aggregate demand if, for instance, aggregate supply really moves up as a result of AI’s positive impact on productivity, but income distribution, jobs, and so on may actually cap the growth of the aggregate demand, and that may require some public policy action. I don’t know if you have any recommendations. You’ve heard about the universal basic income (UBI) proposal.

Robert Rubin:

I wrote an op-ed at the New York Times years ago, and my argument was that’s a terrible idea. First place, people want to work. They really do. Part of their identity is caught up in their work. They want opportunity. And I don’t think that Universal Basic Income, UBI, is the answer to this, but I think you’re right in saying that these are very big issues, and there’s enormous uncertainty about how all this is going to work out.

Huang Yiping:

That’s right. So it does require more proactive public policy action. Certainly we should think about it more seriously. The other risk you warned in your early discussion is the so-called circular risks. I.e., the large tech companies making huge procurement commitments, and then you also find the upstream suppliers borrow heavily on expansion. Obviously the potential risk is what if the procurement commitment could not materialize at some point for some reason, then you might actually see a crash. I’ve heard the view that the AI investment is much, much larger than the investment during the Internet bubble. So how far away do you think the risk is from a systemic outbreak?

Robert Rubin:

My impression is that there is a lot of exposure in terms of lending to AI-related activities, but I think the chance of a crash is pretty low. The reason is that the financial system is really in very good shape in the United States. The big banks are very well capitalized, and that’s where the lending to the AI-related businesses is taking place. The regional banks, which are not as well capitalized, have not been in the AI lending business. Now, there are non-bank lenders, so there are issues that may have to be dealt with. We could have individual companies go bad, but the risk of a systemic problem is probably very low. I could be wrong.

Huang Yiping:

I see. You think it’s not a systemic financial risk – but could it have a major impact on growth or real activities?

Robert Rubin:

I think everything related to AI involves massive uncertainties, but the chance of a systemic financial problem is very, very low. If you have enough companies have difficulty, you could in fact have some effect on growth, and there may be some individual companies that do badly and maybe even go bankrupt. They could happen, but I think the overall effects of all that still involve massive uncertainty. My own impression would be that there is some risk to growth if there are enough companies that run into trouble, but the risk of a financial crisis I think is extremely low.

Huang Yiping:

Okay. Professor Jason Furman last year made some estimates on the contribution from AI investment to GDP growth. I think he was saying at some point data center-related investment contributed at least 90% of GDP growth. If the circular risk you mentioned earlier actually materialized...

Robert Rubin:

I don’t remember the 90%, but I think he did say something like that. What he was saying is that a lot of investment demand now is related to AI. So it’s true that if you have a contraction in that area for the reasons you say, that could have a growth effect. But what I was saying is we have ups and downs in our economy. That’s the nature of a market-based economy. So yes, that could happen, but I think the probability of a financial crisis is very slight. If you have some sort of reduction in demand for the reasons Jason said, yes, you could, but the other side of that coin is that all of this investment could create a great deal of productivity. On the other hand, as I said a moment ago, the offset there may be job losses. And then the thing you haven’t mentioned, which is the thing most on my mind, is the safety risks.

Huang Yiping:

Agree. I remember also having conversation at some stage with Jason on this particular topic. His view is probably similar to yours. He said even if there is significant slowdown in data center-related activities, that doesn’t mean 90% of the growth will be gone because resources will be redirected to some other activities.

Robert Rubin:

A, I agree with that. B, there’s still going to be, at least in my opinion, even with all the political problems around data centers that now exist in the United States, I still think there’ll be a lot of AI investment. What I think is a much bigger issue than any of this and that is the question of AI safety. Hank Paulson, who many in China know well, and I wrote an op-ed about maybe a month ago or thereabouts. I think that is a risk and there are vast uncertainties around that, and there is certainly the possibility—probably a pretty high probability that there could be serious and even severe adverse effects, including cyberattacks of all kinds. Especially now with open source, that becomes available to much larger numbers of people, and with recursive self-improvement and AGI and ASI, it’s moving ahead very quickly. There’s no very good answer to that, but one thing I am absolutely convinced of, and that is that our political systems in both our countries should be focused on this very carefully, very intensely, and we should be working together, because if one country focuses and the other doesn’t, that isn’t going to work either.

Huang Yiping:

That’s right. I will follow up on your and Secretary Paulson’s proposal later on. But before that, I want also to ask you about the technological competition between China and the US. In our early dialogues, you said China has very strong technological innovation capability, and that even if the US deliberately restricts it, China will eventually achieve its technological development goals, and cooperation benefits both China and the US. I guess the question to you here is, how deep do you think is the political consensus in the US about restricting China in terms of technological development, particularly AI development? And will things change or loosen after the midterm elections in the US?

Robert Rubin:

On the first part of your question, I don’t know the answer to that. Right now, unfortunately, I think there’s a lot of animosity and distrust with respect to China, and I think we’ve got to get past that somehow or other to deal with these AI safety issues, or I think both of our countries are going to pay an enormous price. That at least is my view. What was the second part of the question? I’m sorry, Yiping.

Huang Yiping:

Oh, the question is, will the political consensus change?

Robert Rubin:

Oh, the midterms! No, the midterms... The midterms are really, really, really important, but they’re not going to affect policy very much because Trump is still going to be president. First place, Congress itself is going to be complicated. I think the midterms are going to be complicated. I think the outcomes could be very significant, though, in terms of the political environment in the United States. But whatever happens there, I don’t think Congress is going to be able to effectively do anything in this area. I think that has to come from the administration. And, you know, they have taken some actions, but they seem to me to be not even remotely adequate relative to the issues we face.

Huang Yiping:

OK. You also listed AI and climate change as the two existential risks facing humanity. And you just mentioned about the proposal you and Secretary Paulson made on possibly a US-China AI cooperation treaty modeled on nuclear arms control. What specific mechanisms would such a treaty entail?

Robert Rubin:

I think our point was slightly different. Our point was that the Soviet Union and the United States had a massive distrust of each other, but they recognized the dangers of nuclear development, and so they were willing to enter into a treaty. Unfortunately, the United States and China have this distrust and mutual antagonism. I think it’s enormously against the self-interest of both of our countries, but I think it’s where we are right now. So what we said, just as happened with SALT and the nuclear matters, we should put aside distrust, put aside the antagonism, and work together on an issue that can so massively and powerfully and adversely, severely affect both our countries. We didn’t have specifics, but if the two countries were to do that, then they would intensely engage right now, both each in their own countries with respect to their own AI and in working with each other.

Huang Yiping:

Okay. So if I understand what you’re saying, your purpose is to contain downside risk, not necessarily to create opportunities for cooperation in the AI space.

Robert Rubin:

No, I think... Well, I think the cooperation would be to contain the risk, just as you say. But there might also be cooperation in all kinds of ways that would also enable both countries to better realize the upside.

Huang Yiping:

Okay. Thank you very much. Bob, the other issue which is getting attention quite a lot lately is the new global imbalances. And obviously, this was an issue that was heatedly debated 20 years ago, and it came back again this year. At obviously, the G7 Summit, G20 Finance Ministers and Central Bank Governors meetings, and so on. On this particular issue, if I understand correctly, there is a kind of consensus that the global economy may need to be rebalanced. But who should take the main responsibility? What do you think we should look at this issue, and what are the useful policy actions we need to take?

Robert Rubin:

I think there are two issues around imbalances. One of them is, for example, the United States, we have these huge fiscal deficits that contributes to our trade deficits, and that’s something we need to address. We’re still in a very strong position in terms of our economy, despite what I think are the counterproductive actions of this administration, because we have tremendous underlying strengths. So I think we can still do this with relatively little difficulty. There are other countries where I think the difficulties are greater. Another issue, I think, is consumption in China. I remember when I were with Zhu Rongji, and we were saying China should increase its consumption in its own interest and the interest of the rest of the world. That’s still the case, I think.

Huang Yiping:

Yes, I agree. Probably the way for China to go ahead is to rebalance first the domestic economy, and as we know, the external account surplus is determined by savings-investment gap. And that’s probably applicable for...

Robert Rubin:

Well, we have to do the same thing, exactly to your point, by dealing with our fiscal deficit. And I believe that... we’re still in a very strong position with respect to doing that, but it does take a certain political will, and my guess is we’ll have to wait for the next administration for that to happen.

Huang Yiping:

Right. One of the issues we look at is (tariffs). President Trump introduced these tariffs, and we thought it might be removed at some point, but it’s stuck or it looks like it’s a very sticky... Tariffs have not changed the deep structure of US-China trade imbalances. Some analysts argue that the truly promising rebalancing is not more thorough decoupling, but converting part of China’s exports to the United States into Chinese production capacity formed inside the United States. How do you view the feasibility of this rebalancing path of capital rebalancing trade? What clear rules would we need?

Robert Rubin:

Two quick comments. One, I think tariffs have been a terrible idea. I think that they’re regressive, I think they adversely affect growth. Trump has this deeply embedded, non-analytic belief in tariffs, so we’re not going to get rid of those until Trump is no longer president. Beyond that, I think it would be wonderful to have Chinese investment in the United States. It may have some effect, it’s not clear to me why it would have any effect, actually, on the imbalances, maybe it would. But I don’t think it addresses the essential problem, which the one you mentioned, which is... and I also mentioned the need for consumption in China. And the need for us to take care of our fiscal deficit.

Huang Yiping:

That’s right. So maybe my last question will be tariff related. You just mentioned this is probably seriously misguided policy actions.

Robert Rubin:

And it’s wildly and tremendously unpopular politically.

Huang Yiping:

Politically not popular. Because you also pointed out that they are inherently regressive because they hurt the low-income households more than the wealthy. Early on we had some kind of a little bit puzzling that tariffs did not really push up the inflation rate, but now it looks like start happening. Do you think this is causing some issues in the economy and in the society, and whether this tariff issue might be becoming a driving force for the future direction of economic policy in the U.S.?

Robert Rubin:

Well, I think tariffs are increasing costs for both the consumers and producers, and secondly, I think that those increases in costs can create inflation expectations to build on top of the costs that the tariffs increase. So I think for both reasons, and then I think all of that affects us adversely economically. It affects middle-income people, low-income people most, and that is why such widespread political reaction to it. And I think we would be well served to get rid of the tariffs, except for our legitimate supply chain concerns, but I think that’s not going to happen until we get a new administration.

Huang Yiping:

I see. So at least, bottom line, it looks like in the near term we are not going to see easing of monetary policy as some politicians anticipate.

Robert Rubin:

Monetary policy? Wait, you said monetary policy?

Huang Yiping:

Yeah, monetary policy.

Robert Rubin:

Oh, monetary policy.

Huang Yiping:

The president wanted to cut the interest rate.

Robert Rubin:

Yeah, I know. He does. I tell you, Kevin Warsh is very well equipped for this job. Warsh has repeatedly expressed his support and belief in and commitment to the independence of the Fed, and I believe that’s what he’ll do. We face an inflationary, at least I think, a serious inflationary problem. And my guess is he will continue, and the Fed and the FOMC will continue to focus on that. They certainly should, and I believe they will.

Huang Yiping:

Right. Well, thank you very much, Bob. Unfortunately, the time is up again. But really appreciate your time, and we learned a lot from you. We hope to be able to continue to learn from you in the future Summit.

Robert Rubin:

I am honored to be here, Yiping, and I wish you a good day.

Huang Yiping:

Yeah. Thank you. Bye bye.

Robert Rubin:

Thank you.

Since 2022, the deep real estate adjustment has exerted a sustained and far-reaching drag on China’s domestic demand. Over the same period, emerging industries, particularly the “new three,” expanded rapidly. In 2025, the new economy accounted for a larger share of GDP than real estate. In 2026, the real estate adjustment began to show signs of structural stabilization, and China formally entered the “post-real-estate era.”

As the engine switches, how will the macroeconomic picture be redrawn? And what policy choices should be made? At the 2026 Bund Summit held on October 19–22, the CF40 Research Annual Report (hereinafter “the Report”) was released. Titled “China’s Economy: Toward a New Paradigm—Drivers, Trends, and Policy Implications,” the Report provides an overall framework for understanding China’s current economic transformation.

Guo Kai, President of CF40 Research, presented the Report and then joined a roundtable discussion on the Report and topics related to economic growth with Bai Chong-en, Dean of the School of Economics and Management at Tsinghua University; Sonali Jain-Chandra, Deputy Director of the Asia and Pacific Department at the International Monetary Fund; and Zhang Bin, CF40 Non-Resident Senior Fellow. Zhu He, CF40 Senior Fellow, moderated the roundtable.

The Report uses the term “new paradigm for the Chinese economy” to describe the transformation of China’s economy taking place as complex domestic and international variables interact, and identifies four key drivers moving China into the new paradigm: (1) the real estate and manufacturing capacity cycles approaching their troughs; (2) the rise of the new economy and the AI and energy revolutions; (3) population aging and continued fiscal consolidation; and (4) rising global geopolitical risks.

On this basis, the Report sets out five key trends of the new paradigm that warrant particular attention, involving the most important core macroeconomic issues—investment, saving, consumption, external surpluses, interest rates, inflation, the business cycle, and Chinese firms’ overseas expansion. Finally, the Report offers a series of policy reflections covering macroeconomic policy as well as reforms and social policy.

Explore the Full Report

I. Four Key Drivers of the New Paradigm

First, the real estate and manufacturing capacity cycles are nearing their troughs, easing the drag on growth. On the one hand, the Report systematically reviewed 22 episodes of major real estate adjustment worldwide since 1970 and found, through comparison, that China’s current real estate adjustment has already been considerable in both duration and magnitude. Since the start of 2026, China’s real estate market has begun to show signs of “structural stabilization,” and going forward, changes in housing inventories are likely to reinforce this emerging pattern of “structural stabilization.” In addition, a growing number of property developers are entering substantive debt restructuring, and the scale and pace of restructuring will be the key points to watch. On the other hand, China’s manufacturing capacity cycle is also gradually bottoming out, driven mainly by firms’ own supply-side adjustments. The gradual bottoming out of both cycles is jointly moving China’s economy into a new phase of rebalancing.

Second, the new economy is rising at an accelerating pace, while the AI and energy revolutions create new growth opportunities. Over the past five years, the new economy—represented by new energy vehicles, clean energy, high-end equipment, and digital technologies—has expanded rapidly and has gradually surpassed the traditional economy represented by real estate in terms of incremental growth. According to CF40 research, by 2025 the new economy’s total value-added share had surpassed that of the real estate sector, and the new economy had become an important component of GDP. At the same time, technological progress in AI, including large language models and robotics, has accelerated, creating new potential for sustained productivity gains, and related investment demand has begun to surge. The energy revolution, in turn, complements the development of AI and advanced manufacturing by lowering electricity costs and expanding the supply of clean energy.

Third, population aging combined with ongoing fiscal consolidation imposes long-term constraints on economic growth. China’s demographic structure has entered a clear turning point. Over the next three decades, the working-age population will remain on a declining path, and the pace of decline will accelerate after 2030. As the population declines, the accumulation of human capital may partly offset the decline in effective labor, although this effect remains to be seen. The rigid pressure that aging places on fiscal expenditure is becoming increasingly evident, and, according to CF40 estimates, aging may not lower the saving rate. Meanwhile, against the backdrop of fiscal consolidation, the room for maneuver and the actual stabilizing capacity of fiscal policy will be affected to some extent.

Fourth, populism and geopolitical risks are rising worldwide. This structural economic transformation will take place in a geopolitical environment that differs markedly from that of the past two decades: first, restrictions and barriers affecting global trade and cross-border investment are rising rapidly, with tariff changes providing a direct illustration of this shift; second, the influence of right-wing populism keeps growing; and third, regional security risks around the world are likely to remain elevated and could rise further. Foreseeably, concerns about national security, economic autonomy, and supply-chain stability will become key determinants of economic policy across countries. Over the longer term, rising populism and growing geopolitical fragmentation pose challenges to both China’s merchandise exports and firms’ overseas expansion, although they need not lead to weaker Chinese exports in the short term; Chinese exports could instead continue to strengthen, benefiting from fiscal expansion in other countries.

II. Five Key Trends of the New Paradigm

First, the investment rate is likely to stop declining and may even edge up, helping the neutral interest rate stabilize at a low level.

Based on three considerations, the Report expects China’s investment rate to stop declining and possibly edge up: first, real estate investment has already adjusted deeply, and its drag on the investment rate is expected to diminish further; second, substantial investment demand remains in the new economy and AI-related sectors, helping stabilize manufacturing investment; third, infrastructure investment is unlikely to expand significantly, but central government funding and structural investment needs will still provide a floor for infrastructure, keeping overall growth relatively stable.

The Report notes that replacement investment is another important factor supporting the overall investment rate. In China, depreciation as a share of gross capital formation rose from 45% in 2010 to about 63% in 2023, and an important feature of China’s future investment may be “less expansionary investment and more renewal investment.” Against this backdrop, the neutral interest rate is likely to move from its previous phase of persistent decline into a phase of stabilization at a low level, with conditions for a gradual rise over time. However, because its absolute level remains close to zero and inflation will remain low, the real interest rate could remain significantly above the neutral rate; the effectiveness of monetary policy may be affected, and fiscal policy becomes even more important.

Second, real estate stabilization will help household consumption recover, but whether the consumption rate can rise sustainably over the long term remains uncertain.

Over the past several years, the continuing real estate adjustment has been the most important factor behind the decline in the household propensity to consume. As the real estate market gradually stabilizes, the factors that have persistently weighed on household consumption will weaken. However, a sustained increase in the consumption rate will also require households to form more stable expectations about future income and employment. The household consumption rate depends on two factors: households’ share of national income and their propensity to consume. Both could improve in the coming years, but the outcome will depend heavily on policy choices.

The Report points out that the negative wealth effect of the real estate adjustment is fading, but households’ propensity to consume has yet to fully recover to its pre-pandemic level, and further improvements in social security and public services must also take fiscal sustainability into account. Rather than rising sharply, therefore, the consumption rate is more likely to stabilize as the drag from real estate fades. If supportive policies are fully implemented, the consumption rate could rise.

Third, a peak in the current account surplus-to-GDP ratio, gradually firming inflation, and renminbi appreciation.

Over the past few years, the deep adjustment in the real estate market has been the main driver of the widening current account surplus: the real estate downturn led to a sharp decline in the investment rate, while the saving rate adjusted with a lag, and the resulting widening of the saving–investment gap translated directly into a larger current account surplus. This was verified by CF40 research examining changes in current account balances in sample countries that had experienced major real estate adjustments. As the real estate market gradually stops falling and stabilizes, the drag from real estate investment will diminish, while household precautionary saving will stop rising. The saving–investment gap will therefore narrow, and the current account surplus will stabilize at a high level rather than continue to widen.

On inflation, the Report finds that three forces have jointly driven the year-on-year PPI inflation rate from negative to positive territory: the bottoming out of the manufacturing capacity cycle, strong demand from the AI industry, and the sharp rise in crude oil prices. Of these, the first two forces have some persistence in their contribution to PPI. The recent improvement in China’s year-on-year PPI growth should therefore not be viewed simply as so-called “imported inflation”; the underlying economic fundamentals are more important and warrant greater attention. Compared with the PPI, the recovery in headline and core CPI inflation has been more gradual, consistent with the broadly stable household consumption rate over the short to medium term and the gradual improvement in households’ propensity to consume. As consumption of services expands and the negative wealth effect from the real estate market weakens, headline and core CPI inflation are expected to continue recovering moderately.

On the exchange rate, improvements across multiple factors are moving the renminbi into an appreciation phase, and the trend could continue. Domestically, the real estate market has shown signs of structural stabilization, and the Chinese economy has begun to emerge from a period of low inflation; internationally, a new shift has emerged in the dynamics of China–U.S. competition, with China displaying strong resilience in trade, technology, and critical resources. The Report notes that, given that inflation in China is likely to remain about 3–4 percentage points below that in major advanced economies, that China’s manufacturing productivity is likely to grow about 1–2 percentage points faster than in other major economies, and that renminbi-denominated assets have gained stronger diversification and relative safe-haven properties, even if the renminbi were to appreciate by 5% annually, the impact on China’s export competitiveness would be close to neutral.

Fourth, the scope for fiscal expansion remains relatively constrained, and the pace of fiscal spending is becoming more important.

The Report notes that, in an environment of weak real estate activity and subdued private credit demand, both the multiplier and signaling effects of fiscal policy may be stronger than in the past. The scale and timing of fiscal expenditure could become the most important determinant of cyclical fluctuations in China’s economy in the future. Compared with fiscal policy, monetary policy may face more constraints.

It should be noted that, under fiscal consolidation, the scope for fiscal expansion is relatively limited, which means the economy may enter a period of low volatility over the next few years. By comparison, the pace of fiscal spending within the year will become more important, because the speed of intra-year fiscal spending—the timing of funding releases and project starts—will directly affect short-term investment, consumption, and credit demand. This may offer market participants a new way to monitor and interpret fiscal policy.

Fifth, Chinese firms have stronger incentives to expand abroad but face greater uncertainties.

Since 2023, China’s outward investment has returned to an upward trajectory, while its composition has also shifted structurally, with greenfield investment replacing cross-border M&A as the dominant form. The increase in greenfield investment means that Chinese firms are directly constructing overseas factories and production lines, and the relationship between Chinese firms and overseas economies is extending beyond trade into production. By industry, China’s outward investment has become increasingly concentrated in a small number of sectors in recent years, with automobiles, basic materials, and energy forming the main industry chains driving current overseas expansion—a pattern that closely reflects China’s accumulated advantages in the “new three” sectors.

For some time to come, overseas expansion by Chinese firms will remain a major trend, and its form will further evolve toward overseas investment, localized production, and the establishment of supply chains abroad, potentially laying a new foundation for renminbi internationalization. Behind this lies a dual logic: externally, rising trade barriers mean firms need to locate part of production closer to end-consumer markets; internally, the domestic market has shifted from demand-driven expansion to competition for market share in mature markets, and firms with technological, cost, and supply-chain advantages look to overseas markets as new sources of growth. It should be noted, however, that overseas direct investment places much greater demands on firms’ organizational capabilities than exports, and a more fragmented geopolitical environment means firms will need to reassess the trade-offs among efficiency, resilience, and localization—their risk management capabilities will face new tests.

III. Policy Choices under the New Paradigm

The Report notes that, as the economy enters the new paradigm, the challenges facing policymakers are also changing. The most important short-term policy priority is to promote reflation, help the economy emerge from a period of low nominal growth as quickly as possible, and remove bottlenecks in domestic economic circulation. Over the medium to long term, the central policy objective is to gradually establish, through a series of structural reforms, a new growth mechanism suited to the post-real-estate era.

In the short term, the Report recommends that nominal GDP growth be considered an important reference for macroeconomic policy and stresses that it should be viewed as a macroeconomic objective to which both fiscal and monetary policy contribute: fiscal policy needs to assume the primary role in stabilizing aggregate demand as soon as possible, while monetary policy should move away from pursuing credit-volume expansion toward lowering interest rates, improving inflation expectations, and restoring nominal demand. One possible aggregate benchmark is that “nominal growth in broad fiscal expenditure should exceed nominal GDP growth.”

The Report notes that, given that local governments face multiple constraints, including declining land-related revenues, debt resolution, and rising mandatory spending commitments, near-term fiscal expansion will need to rely more heavily on the central government. On the one hand, the central government could increase support for subsidies to low-income groups, public services, and social security. On the other hand, central-government infrastructure and public investment should be directed more toward projects with clear social returns, such as the upgrading of existing facilities and incremental supporting infrastructure related to metropolitan-area development.

Short-term reflation can help the economy overcome weak demand, but it cannot substitute for structural reforms. Over the medium to long term, the key policy tasks are to gradually resolve the fiscal and debt problems inherited from the real estate era, establish new mechanisms that translate growth in the new economy into household income and consumption, and adapt to changes in the global economic environment. The Report proposes four priorities:

First, advance fiscal and tax system reform and rebuild the fiscal revenue and expenditure framework for the post-real-estate era. Gradually develop sources of fiscal revenue that are less dependent on real estate and land cycles, further adjust fiscal relations between the central and local governments, and shift the composition of fiscal spending from its emphasis on investment in new construction toward the maintenance of existing facilities, human capital investment, and the provision of public services.

Second, accelerate real estate and local government debt restructuring and complete the balance-sheet cleanup of the old growth model. Developers’ debt should be resolved more quickly through market-based restructuring, bringing assets back into productive use, and mergers and consolidation, while local governments need to shift gradually from emergency debt resolution toward systematic debt restructuring.

Third, develop the service sector, raise household disposable income, and translate new-economy growth into consumption demand. China could further reduce market-entry and cross-regional operating barriers in services, expand participation by private capital and foreign investment, and increase the supply of high-quality services, allowing the service sector to serve as an important bridge between productivity gains and household income growth. At the same time, households need greater disposable resources within the distribution of national income, and the policy focus could shift further from short-term, one-off consumption subsidies toward improving the stability and predictability of household income.

Fourth, expand high-standard opening up and support firms’ shift from product exports to global operations. On the one hand, China can continue to expand domestic market opening, particularly institutional opening in services and investment, while increasing the stability and predictability of domestic market rules; on the other hand, a more comprehensive service system is needed to support firms’ overseas expansion.

On October 19, as part of the 2026 Bund Summit co-organized by China Finance 40 Forum (CF40) and Tsinghua University:

• Jacob A. FRENKEL, former Governor of the Bank of Israel, Chairman Emeritus of the G30 and President of the Frenkel-Zuckerman Institute for Global Economics;

• Donald KOHN, Economic Studies Fellow at Brookings and former Vice Chairman of the Federal Reserve;

• Klaas KNOT, former President of the Central Bank of the Netherlands and Distinguished Visiting Fellow at the Peterson Institute for International Economics;

• Philip R. LANE, member of the Executive Board and Chief Economist of the European Central Bank; and

• Masaaki SHIRAKAWA, former Governor of the Bank of Japan and Distinguished Professor at Aoyama Gakuin University,

focused on “Central Bank in Transition: Responsibilities, Markets and Macroprudential Risks” in a roundtable discussion.

The discussion was moderated by Ceyla PAZARBASIOGLU, Chief Operating Officer of the G30 and former Director of the IMF’s Strategy, Policy and Review Department.

Left to right: Ceyla PAZARBASIOGLU, Donald KOHN, Klaas KNOT, Philip LANE, Masaaki SHIRAKAWA, Jacob FRENKEL.

On central bank independence, the discussion converged on a key principle: independence does not mean being detached from or above the government, but rather fulfilling the mandate entrusted to the central bank. Achieving price stability is the central bank’s primary responsibility and also the best safeguard of its independence. Klaas Knot emphasized that the value of independence lies in lowering the social cost of disinflation. He contrasted the deep recession needed to bring inflation under control in the 1970s with the euro area’s experience in the 2020s, when inflation was brought down without a recession.

But independence does not guarantee that central banks will always get policy right. Masaaki Shirakawa cautioned against equating price stability with macroeconomic stability, arguing that this narrow focus had led to insufficient attention to financial imbalances. He also reflected on the risks of large-scale quantitative easing deployed to combat deflation, stressing that policy instruments should be matched to the objectives they are best suited to achieve and that the principle of proportionality should be respected.

The panelists held that central bank independence is indeed being squeezed by structural changes, fiscal pressure and political pressure, and that the risk of fiscal dominance does exist, but that financial markets can play a certain corrective role. Donald Kohn said he did not expect the Federal Reserve to abandon its price stability mandate and did not see fiscal dominance taking hold in the United States, despite mounting pressures on monetary policy.

Turning to financial risk, Donald Kohn, Klaas Knot and other panelists highlighted the difficulty of monitoring risks in non-bank financial intermediation, including hidden leverage, inadequate liquidity buffers for margin calls and fragmented supervisory responsibilities. Central banks may be held accountable when problems emerge, even when they lack the authority or tools to prevent or manage them. Ceyla Pazarbasioglu stressed in this connection that central banks and other supervisory authorities need to fill data gaps and prepare in advance—calls that have persisted for years but still lack implementation.

AI emerged as another major source of uncertainty. Donald Kohn argued that AI affects demand before supply, highlighting the need to study its implications for labor markets. Jacob A. Frenkel cautioned that major supply-side changes would take time to materialize, while Masaaki Shirakawa focused on AI’s potential impact on democracy and the implications for monetary and financial stability policy.

Below is a transcript of the roundtable produced by CF40 with support of AI.

Ceyla PAZARBASIOGLU:

Pleasure to be here, and we really look forward to even more collaboration with the CF40. This Bund Summit could not have come at a more important and timely point in time, so it’s really great to be here.

I will say a few words about the Group of 30, because not all of you may know it. The Group of 30 was actually established in 1978, so in a couple of years, it’ll be 50 years. The idea is to foster a deep, nonpartisan dialogue between economic and financial leaders at a time of important structural changes with financial stability implications. So the role, or the mission, of the Group of 30 has been to identify blind spots and talk about financial stability risks, monetary policy issues, global governance before they manifest themselves as a systemic crisis.

I’m coming from Bangkok, where we had the IMF--World Bank Annual Meetings. I’ve been at the Fund since ‘92. I retired last month---in May, rather---and it was the first time I attended as a non-IMF official. And it was very interesting to see how many challenges there are, and how the global architecture, financial architecture, is again at crossroads.

Things can go really well, but there are also a lot of risks. And there is the relative calm financial markets but with many fragilities and vulnerabilities, especially for central bankers. So this is why this panel is very timely. And we have a very impressive panel. I’m going to introduce them very briefly. They’re all known, very well known. Their bios will take us the whole hour. So I’m gonna say just a few words.

First of all, we have Jacob Frenkel on the line. He just arrived in Tel Aviv and kindly agreed to participate at little hours in the morning. So welcome, Jacob. He’s the former Governor of the Bank of Israel and the Chairman Emeritus of the Group of 30, has incredible experience in private sector, public sector and so on, and brings a lot of expertise.

Klaas Knot, who is the former President of the Central Bank of the Netherlands and also a Group of 30 member, and again, with incredible experience.

Don Kohn, former Vice Chairman, Board of Governors of the Fed. Again, very impressive CV in terms of private sector and public sector work.

Same with Philip Lane, who is the Chief Economist of the ECB.

And last but not least, Masaaki Shirakawa-san, who is the former Governor of the Bank of Japan and a G30 member. So thank you very much for participating in this panel.

I’m going to start with you, Jacob. The question is central bank independence. This is under a lot of pressure with many structural changes and fiscal pressure, political pressures, macroeconomic vulnerabilities and so on. What are your views in terms of this important issue, which is under threat? Let’s face it. Central bank independence is under threat. There is fiscal dominance. So what needs to be done and by whom?

And then I’ll ask each of the panelists to come in from their perspectives.

Jacob FRENKEL:

I am privileged to be here again at the Bund Summit, one of the most important international gatherings, and especially as we have now a collaboration between the CF40 and the Group of 30.

My dear friends, the question that was asked about central bank independence, I can tell you that when I participated in similar gatherings 30 years ago or even 40 years ago, the same questions were there, but the answers are different.

And the answers are different because the world has changed dramatically. We are in an interdependent world. And what is extraordinary is even if you look now around the table, you see that whether you are coming from Asia, from Europe, from the United States, there is a corpus of knowledge that connects the profession and the issues of central banking. And that’s what makes it so fascinating. So I’m privileged to be here again. And the role of China in this regard is extremely important, and should be even more so.

Coming to the question of central bank independence - all of us, when we practiced our central banking, we faced many of the political leaders who are asking: why should you, who was elected by the people, and now you suddenly want to be independent? And I think that there is an interesting question here, and one needs to provide a clear answer.

My clear answer is when we talk about independence, we do not mean unrelated to the government. We do not mean it’s more important than the government. Independence is about implementing the mandate that was granted to the central bank by the government.

Like in every range of countries, there are multiple objectives, and that’s a challenge. How do you bring about harmony of the various policymaking apparatus while at the same time sticking to the mandate? What is the main mandate for most of the central banks today?

It’s the role of price stability, and I want again to emphasize. In many public debates, there is the question of why do you think that price stability is the most important objective? After all, we have employment, we have so many other...

I’m not trying to say which one is the most important objective for society. We’re talking about the central bank. For the central bank, the most important objective to achieve is price stability. And the reason is the tools that we are in the position of the central banks are designed to bring price stability, and you cannot have or cannot achieve the other objectives unless you have price stability. And the best agency to bring it about is the central bank.

One of the issues that came up all over the world over the past decade has been the role of data. How do you use the data in the best way to bring about good decision-making? And what we saw are two elements.

Number one, in many cases, central banks will, as we say, they fell behind the curve. People say in retrospect that in some cases, central banks have allowed inflation to run faster than was desirable, and without stepping in sufficiently early. And I think that’s a very important practical and analytical issue, which has to do with data dependence.

If you say that your decisions are going to be exclusively depending on the data that you receive from the market, then by definition, you will lag. Because until you see the data, you will not act. And that’s a very important lesson that we have learned. You must act in an anticipatory way, even before the data come in front of you. And that’s what makes it a professional challenge. The role of models is exactly to help the central banks to see how they can anticipate what is happening, potentially even prevent the bad behavior.

One of the dangers, and one of the reasons why it is more important now than ever, has to do with the fact of speed. The role of capital markets has increased dramatically over the past few decades. And what capital markets are doing, they’re bringing the future into the present through expectations, and they’re operating very quickly. And that’s why when the role of speed comes in, there are also greater risks. And when there are greater risks, they are losing the language of finance and financial markets, risk management, expectations, and the last most important asset. So this is credibility. Because the markets must believe that the central bank knows what it is doing, and also brings about the outcome.

And I want to make one last point, which is there are always three elements in communication. Element number one is the transmitter. Element number two is the receiver. They must be fully aligned. If the message is not clear, if the transmitter is not well defined, and if the receiver is not fully aligned, the message gets out garbled. Therefore, what the central bank is doing is for both the transmitter providing the message, and the receiver translating it into the financial markets.

Finally, there is an issue of mandate, and here there has been also some confusion. Many of the central banks, at least in the more recent years, have had a very sharp mandate: price stability. Some central banks, for example, the European Central Bank, and we have on the panel a representative from the ECB, they will tell you that they do look at other objectives, as long as they are not compromising the primary objective, which is price stability.

Now, if you look at the United States, the mandate of the United States Federal Reserve Act, and we have Don Kohn who will tell you about it, is related to price stability, is related to maximum employment. We are calling it the dual mandate.

But as the years pass by, you realize that this dual mandate, or multi-mandate, if you want to call it, is really not so operational. What is operational? And that’s what Paul Volcker used always to say. And Paul Volcker, for our friends, has been the legendary Chairman of the Federal Reserve. He always said, I think that all objectives are important, but the only way, as the central bank, can help to bring about the attainment of all the wide range of objectives, is by securing price stability. That’s why [I’m focusing so much] on price stability.

Ceyla PAZARBASIOGLU:

Yeah, that was an excellent segue to Don Kohn, who can tell us about the US experience, and your views on the role of tools, data dependence, communication. So, Don. Thank you.

Donald KOHN:

Thank you, Ceyla, and thank you for having me at this wonderful conference.

So I’m afraid that the pressures on central bank independence have been very intense in the United States. But I think the United States is not that different from a lot of other countries. There’s been a lot of populist uprising, or people unhappy with the technocratic institutions. That has been embodied, to some extent, in the rise of autocrats or would-be autocrats like Donald Trump, who doesn’t want any constraints on his power. So, and that’s happening at the same time we have the fiscal pressures that you mentioned, which will pressure monetary policy, and you have all these supply shocks.

So I think one of the best safeguards of independence is hitting that mandate that Jacob was talking about: price stability. And with these repeated supply shocks, though, the petroleum prices, the tariffs, etc., then I think it makes it much harder to hit that mandate. So all these things, I’m afraid, aren’t going to let up anytime soon.

You ask whether we saw fiscal dominance coming. I don’t see fiscal dominance coming in the United States. I don’t see the Fed giving up on the price stability mandate, but I do see that pressure adding to what’s going to happen. And I don’t, and one reason mandates won’t change—this is carrying forward what Jacob said, is people really don’t like inflation.

So I think we’ve learned over the past decade that inflation is very unpopular, and it will be necessary to take steps, and the central bank, as Jacob said, is the instrument to take steps against it.

What can central banks do to protect their independence? I think one: they can be nonpartisan. Make sure you’re not lining up with one side of the issue or another. You keep your focus on your core objectives, price stability, financial stability.

Yes, in the U.S. it’s also the dual mandate labor market, the full employment labor market. But that really comes from achieving price stability. So we need to keep that focus. As Bob Rubin said, the Fed seems to have that focus front of mind now, and I think that’s very, very helpful. Be successful as possible in achieving those mandates. I think that requires flexibility and humility about what we know and we don’t know, and adaptability to changing circumstances. You have to acknowledge the uncertainty that we’re working under, but react to it in a flexible way, provided you have that goal, that price stability goal in mind.

And the last thing I would do is pick up on Jacob’s point about communication. I think you have to be very clear of your analysis of the situation, where you think it’s going, what the forces are at work on monetary policy, and how monetary policy is going to achieve its mandate. So I think communication is critical to all of these things.

Our challenge as central bankers and former central bankers is to communicate to the general public and the legislature. I think we do a very good job communicating with the markets and academia, both of which are well represented in this room, what we don’t do a very good job is communicating to the general public, and it’s very hard. We tend to lapse into technocratic discussions, but I think that’s something we need to work on hard and work on the legislature, who are, after all, give us our mandates.

Ceyla PAZARBASIOGLU:

That’s great. What do you think?

Klaas KNOT:

Well, thank you very much, Ceyla. And I think a lot has already been said by Jacob and Don.

But I want to begin with saying that if you put together a sort of panel of central bankers arguing for central bank independence, there’s always a risk that the sort of perception easily arises of a self-serving line of argumentation. Central banks don’t like to be pressured by politicians or by markets or whatever, etc. What I want to emphatically say here, that is not the essence of central bank independence.

Central bank independence is about significantly lowering the cost of disinflation, and a very simple comparison of the high inflation of the 1970s with the high inflation of the 2020s, I think, illustrates the point quite carefully.

In the 1970s, we had totally unanchored inflation expectation, and Paul Volcker needed to organize a deep, deep recession to break the back of inflation expectations.

In the 2020s, I was in the Governing Council. We had a look at inflation expectations every meeting, and even while actual inflation was in the double digits at some point, the highest number on inflation expectations that I saw was 2.6%.

And so we managed to engineer a disinflation without a recession, yeah, growth slowed, but no recession, and unemployment did not go up even by a notch.

So, first point: central bank independence is not there for us; it’s there for you. It’s there for the society to significantly lower the cost of disinflation.

Now, in the euro area, I think we are pleased by the fact that the sort of legal enshrinement of central bank independence in the **Maastricht Treaty** is probably stronger than in any place else in the world.

So literal attacks on central bank independence, I’m relatively relaxed about it. There is, of course, the issue of fiscal dominance, that is also looming in the background in the euro area.

The massivity of the fiscal adjustments in the primary structural needed to only bring about a stabilization of debt levels, let alone having to bring them down. We’re talking here about three to five percentage points of GDP adjustment needed, and we all know historically that’s pretty unprecedented.

But I would also say that, and I agree here with Don, that there are also natural sort of countervailing powers here. Suppose that a central bank was to give in to fiscal dominance and not raise interest rates, policy rates, at a time in which everybody could blatantly see that there was an inflation problem and policy rates needed to be raised. What then, of course, will happen is that inflation risk will go up, and the long end of the curve will go up. And I would argue that that is much more painful for governments to see the long end of the curve getting out of control, rather than having to survive a few 25 or 50 basis points more hikes in the overnight part of the curve.

So, in that sense, yes, there is the risk of fiscal dominance. But I do also think that financial markets have a sort of corrective role to play here, which I think, to some extent, mitigates difference.

Ceyla PAZARBASIOGLU:

Great, you see this in ECB, of course.

Philip LANE:

So the phrase central bank independence. So far, we’ve talked about it in relation to monetary policy, and I agree with what Klaas said. No one is going to be helped if the central bank is not trusted to bring inflation back to target, so there may be shocks that can lead inflation away from 2%. But if there’s a belief that we’re not committed to bring it back to 2% then there’s going to be an inflation risk premium in the long-term yield, making financing more expensive for governments and being very counterproductive.

But let me just, for the sake of differentiation or expansion, mention that central banks, of course, very involved also in financial regulation and supervision, and also in the payment system. So, in these areas, it’s important to emphasize that central banks play their role, but in the context of a wider legislative framework.

And so, let me mention in the European context maybe two initiatives. One is the digital euro, our central bank digital currency. This is something that we identified as important for the monetary system, but very appropriately, in the end, right now it’s going through the legislative process.

So, in other words, we are not independent in kind of creating something which has much wider implications, equity for supervision of the financial system, the ECB is very involved in banking supervision, but we don’t write the law. The law is written by Parliament, and we implement supervision inside that law. So independence everywhere is inside the mandate, so maybe that’s important.

But maybe I have to respond to Jacob’s point about data dependence, and of course, I think what’s important is to take the middle ground. It’s possible if you are only forward-looking or forecast-looking, that as Don said, you have to be humble. We can get the forecast wrong, and so if you overreact on monetary policy to something in the forecast, then you can make mistakes.

So I’m against under-reacting. I’m against overreacting, and so that middle ground will have some data dependence in there, but not backward looking. The data you want to see is the data that helps you project forward where inflation is going to be a year or two from now. And this year is important for that.

We have inflation going up, but we actually have inflation going up less quickly than we expected back in March, because food inflation was less than we expected. The energy shock was less than expected. So when you when you don’t have perfect forecasting power, the data dependence does help to some extent.

Ceyla PAZARBASIOGLU:

That’s true. It’s hopefully also with technology and more real time data, more tools in terms of making decisions. But yeah, it’s been a challenge for central banks with all these structural changes happening at the same time. You have been through a lot of these issues in Japan, so very interested to hear your views, Masaaki.

Masaaki SHIRAKAWA:

Central bank independence are now being challenged, and central banks are demanded too many things, and this is primarily because the problem our society and economy are faced are so difficult and complex. Therefore, central bank tend to become easy target of the criticism by others, and this is rather obvious among central bank community, I would say, but just to say this is not enough.

And I’m a former central bank governor. I’m not the current central government. But still, we have to think about why this kind of situation was created, and to what extent central bank itself is responsible, and in this sense, we have to look back on the long history of why central bank gained independence.

And as many people argue, the current state of independence derived from the bitter experience of high inflation in 1970s and early 1980s. Therefore, idea was created that the objective of monetary policy was price stability, and this task was delegated to independent central bank, and inflation targeting framework embodied this idea, and this framework worked reasonably well up until 20 years ago.

And then global financial crisis did happen exactly when the inflation targeting was hailed as a great success, and in retrospect, the typical notion that price of inflation targeting, they created the force that price stability is enough in achieving macroeconomic stability, and we tended to be less attentive to various form of financial imbalances, and then central bank deployed Quantitative easing on a massive scale for fear of deflation or for fear of zero low interest rate, and then high inflation did happen.

So we have to wonder why this happened, and I draw two lessons. One is we have to be attentive to the imbalances, which may not be captured by stability alone, and also we have to take the right balance.

And as I said, central bank deploy aggressive Quantitative easing for fear of deflation. And the idea supporting this is so-called the assignment principle. The policy should be assigned to the objective which each policy instrument is most effective.

So monetary policy is trying to achieve price stability. That is good, but going too far is sometimes perilous, and if side effect is negligible, it is okay. But if side effect is non-negligible, going too far is somewhat perilous, and it is like that, I respect the decision by the ECB in 2021 when they announced monetary policy framework, and one passage was quite impressive. That was the importance of proportionality principle, and I think a proportional principle is very important. So on our part of central bankers, we have to be attentive to these kind of things.

Ceyla PAZARBASIOGLU:

Yeah, that’s brilliant, and you’re right. I remember there was a lot of discussion about understanding how asset prices move and what that means for price stability mandate of the central bank and thinking through macro-prudential policies before the great financial crisis, but especially then, which is a great segway to talk about all the changes in terms of financial structure after the GFC.

We had regulations. There was a lot of concern about unregulated non-banks, shadow banks, shadow intermediaries, however we call them, that these may become much more important in terms of financial intermediation and potential implications for financial stability.

So, Klaas, you and Agustín Carstens put together an excellent Group of 30 report on non-bank financial intermediation and potential risks, and in terms of financial sector stability. So, can you tell us a little bit what that means for central banks?

Klaas KNOT:

Yeah, thank you, Ceyla. I mean. Before delving into vulnerabilities and risk, of course, it should be observed that in and of itself, non-bank financial intermediation has a lot of benefits. Right? I mean, it broadens access to finance, it diversifies funding sources, price discovery, innovation. And importantly I also think that in principle it has a capacity to absorb shocks than sometimes is the case in the traditional banking system. But it is also true that what we’ve seen is that NBFI has taken on a larger role in what we call key funding markets, like for instance sovereign bond markets, and there are some emerging concerns around an area like, for instance, private credit.

If you think about the risks in NBFI, they are clearly different from the ones in banking. That has everything to do with the different liability structure. This is an area where we worry about liquidity mismatches, spikes in liquidity demand that lead to sort of herd behavior, where we also have to worry sometimes about pockets of hidden leverage.

In principle, you expect to see less leverage in NBFI than in the banking system. But unfortunately, in a couple of these failures that characterized very much the NBFI risks in the 2020s, there were pockets of hidden leverage coming to the fore in places where we had not expected them, and a third element that I would mention as the main risk is the lack of liquidity preparedness for margin calls.

That, to me, is a very, very serious issue. If I can just refer to systemic liquidity stress tests by the Bank of England, where like 90% or so of the simulated spikes in liquidity demand was actually could be traced back to liquidity needed for margin calls, and that was often not available.

If you take, for instance, the increased role of hedge funds in the government bond markets, that is, I mean, to some extent, the byproduct of our regulation of banks, where we pushed market making out of the banks. Hedge funds are filling that void, but 50 to 70 percent of that is against zero margining. So not even low margining; it’s against zero margining. So that is, I think, an area of risks in and of itself.

And then, secondly, there is also risks emanating from the relationship between NBFIs and banks, because it’s not a clean break. It’s not a simple all risks have migrated out of the banks into NBFI, and that that offers a different set of risks. No, there is a much more symbiotic relationship between banks and NBFI, where, for instance, credit risk may have migrated from banks to NBFI, but since NBFI don’t have access to deposits, don’t have access to central bank facilities, they often have turned to the banks for their funding with contingent credit lines as such.

Private credit, it has already been mentioned, around 30% of the funding of private credit is produced by the banks, and there are concerns in private credit about opacity, about of pro-cyclicality, about overreach into the let’s say retail investor space, etc. So that I think are risks that might sort of come back through the back door here as well.

Now, what do we need to do about it? First and foremost, try to fill data gaps, because NBFI is a very heterogeneous space. There are numerous interactions, etc. But our visibility of that area is clearly less than the visibility that we have on the banking system, and in order to charge financial stability risks coming from NBFI, we really need a comprehensive view of the area. We also need to think a little bit better about our crisis management tools if something goes wrong.

So far, in a sense, we’ve been lucky that the NBFI accident took place in a world where monetary policy was highly accommodative, and therefore asset purchases could sort of form an effective answer. But it has already been mentioned: we are now in a world of repetitive negative supply shocks where central banks are probably more often approaching their inflation targets from above than from below. And in that world, of course, central banks have to be much more careful with sort of liquidity provisions, even if there is an accident somewhere in the system. So that I think also requires more ex ante thinking, and part of it is what we try to offer in the report.

Ceyla PAZARBASIOGLU:

Yeah, that’s great, and I guess I’m going to ask the other panelists to come in. I guess the problem is for the last what five years, if not more, 10 years, FSB, BIS, all central banks, IMF have been saying that we need to close data gaps, that we need to be prepared, and so on. But there seems to be a lack of coordination or lack of- I don’t know what it is-willingness, perhaps, to actually do something about this because the recommendations are excellent, but no takers.

So I guess maybe there should be more emphasis in preparedness and do some what do we do if something something will go wrong? How bad, we don’t know, but perhaps be more prepared because I am amazed how little some progress has been made, like Bank of England, yourselves and so on. But still, U.S. is not doing as much on this issue, so it would be good to hear. Maybe I start with you, Don.

Donald KOHN:

I think financial stability for central banks, in part because of what you just said, Ceyla, presents particularly difficult challenges, and the challenge, at least at the Federal Reserve was we get blame whenever anything goes wrong, and we don’t have the tools to prepare or deal with it.

And one reason I think, Ceyla, that the preparation hasn’t been as thorough and as advanced as we all would like, is that the authority for non-bank financial institutions is widely dispersed, and it’s certainly dispersed in the United States, the SEC, CFTC, and then a bunch of ones that are regulated by states and aren’t regulated at all, really. And it’s seen as a much more politically sensitive thing. So it’s an area in which the politicians feel they have a bigger say than they do for monetary policy, central bank independence. So there are these constraints, and I think it becomes a problem.

Central banks have a unique perspective, I think, on financial markets because monetary policy is carried out through the financial markets. I think central banks have a real knowledge and feel for how financial markets work, that’s not always dispersed in these other authorities. Then the central banks have the macro perspective, so they can bring a broader perspective on spillovers and externalities than the agencies that are dealing with their individual markets, but it’s not always that easy to get that perspective done.

And I was on the financial stability board or the financial policy committee at the Bank of England. The bank has a really nice setup. It has a macro prudential, micro prudential, and monetary policy, and there was a lot of coordination between them. But that’s a very unusual situation. So you’ve got a lot of difficult edges rubbing against each other when it gets to non-bank financial.

Ceyla PAZARBASIOGLU:

Yeah, FSB tried to bring the securities regulators, banking regulators, all together, right?

Klaas KNOT:

No, just one thought. I mean, the point you raised, Ceyla, was about the implementation record. I mean, we have that in the report as well with the coloring table, etc. When I wrote my letters as FSB chair to the G7 and the G20, I always sort of wrote that the implementation record is uneven. That’s a very nice euphemistic word to say that it is quite poor, and that is a problem.

It’s of course a broader problem. It’s not only related to NBFI recommendations that the implementation record of global standards is on a declining path and has been on a declining path, and this is really a problem if we want to contain the risks coming out of these. The recommendations are there, but they need to be implemented.

Donald KOHN:

Why do you think it’s on a declining path? People are forgetting 2008 as time goes by.

Klaas KNOT:

Yeah, and I guess that NBFI is even more sort of... the banks at least have some appeal. People understand what happened there, what went wrong, and there was urgency, etc. But NBFI is even more technocratic, and yeah, legislators have different priorities. It’s not so much that they are against it, I think, but it’s just that they don’t sense enough sense of urgency.

Ceyla PAZARBASIOGLU:

Yeah, there’s also a bit of a chicken and egg, right? There’s not enough data. It’s very opaque, so the linkages and potential implications for financial stability are not very well understood.

Philip LANE:

So this is an area which is probably even more global than banking. So when we talk about private credit. So a lot of private credit provision in the EU area is coming from non-European firms, especially American firms. And equally, the private credit funds which originate in Europe, a lot of their provision is in the US and elsewhere. So essentially, we have this inherently global activity, which requires that global coordination, which is very difficult to do.

Let me mention, just in terms of an institutional framework, in Europe there’s also an additional infrastructure called the European Systemic Risk Board, which indeed brings together all the different supervisors, regulators, the ECB, the European Commission, and I think that does a very good work to really make sure everyone has the same level of understanding. But it does hit the limit, it helps with understanding of what’s going on in Europe, but in terms of understanding the exposures, which are global, not so much.

Let me maybe just come back to the basic point here: you know, in a European context, one of the big projects at the moment is savings investment in European union. It’s identified that we do want European households to develop a wider portfolio, not just holding bank deposits. We do want more intermediation, not just banks intermediating. So, what’s quite advanced already in the US in terms of private credit and so on, it is in Europe, but much less so. So, I mean, trying to make sure this develops in the way that that is, we get the benefits in terms of risk management and so on, without some of the downsides that Klaas mentioned. I think is important.

But let me also say one of the spillovers or connected debates now is also with essentially cyber assets, because it’s fundamentally a lot similar. One of the big questions we have to ask is: if the amount of deposits the system goes down, what does that do to credit provision, even when there is a spill back to the banking system?

So for example, with stable coins, through different mechanisms, a lot of the assets of stable coins intermediaries will end up back in the banking system, but these are now essentially corporate deposits, not sticky household deposits, and so it does change the nature of lending. So there’s so many different pathways here. It is a big project to understand this, but again, all in the context of a more diversified financial system is will contribute to stability. We just have to make sure where we can really mitigate risk.

Ceyla PAZARBASIOGLU:

That’s a great point, actually.

Masaaki SHIRAKAWA:

Okay, I’d like to talk about the role of central banks in financial stability. And financial stability is not result of one single good policy; it’s a result of multiple good policies. And also, institutional setting of supervision and regulation and bankruptcy is quite different across countries, but even so, the role of central bank in the area of financial ability is quite important for two reasons. One is central bank is one resource of liquidity, and the other is central bank is most effective and powerful think tank in each jurisdiction. In that sense, central bank is a role to pose the question about financial stability always.

And I’m a fan of financial history, and when we look back on the history of financial crisis, we encounter two lessons with the same commonality. One is always non-bank financial intermediaries play some role, in the case of Japan. It was Jusen, which was the subsidiary of banks. This non-bank financial institution extended credit to real-estate exchange so aggressively. The same is true for global financial crisis. In this case, the investment bank and SIV played a greater role, but eventually the problem service is a form of liquidity shortage.

So, for instance, we have to pose the question: What is the source of potential liquidity shortage, and how is it materialized and how is it dealt with and somebody have to pose this question. So even though central bank is not equipped with regulatory policy and monetary policy, central bank could contribute to financial stability greatly.

The same is true for digital currency, and the tokenization and DLT are very attractive idea, and we can understand why DLT reduces payment costs by eliminating an entity who manage account, but this entity is not a lazy middleman who just hold account. This entity has a great role, and as Tobias Adrian of IMF argued recently, tokenization constitutes a structural reallocation of trust within the financial system. I quite agree with what he said.

So we have to ask a question: How the previous role say on the monitoring the party who are compliance with financial transaction are performing, so in that sense, central bank has a greater role, even though it may not be a single supervisor in the matter.

Ceyla PAZARBASIOGLU:

Excellent point, and I think liquidity is clearly something that needs to be managed, and this very global environment we are in, that everyone is talking about, it’s also cross-border and dollar liquidity provision. So it makes it even more important that the authorities coordinate, the central banks coordinate across, which hopefully will happen. Jacob, any thoughts from you on this issue?

Jacob FRENKEL:

Yes. First of all, there was an incredible range of topics and insights that were now presented by the panel, which is very impressive. I’d like to only to make a few comments.

Number one, the fact today is that most of the banking activities in most countries do not take place in what we normally called banks. So it means, therefore, that when you have a banking supervisor, at the end of the day, the conventional responsibilities cover only a subset of the banking activities, and therefore, when we speak about risks and things of that type, we need to make sure that there is some alignment there, and it is a dynamic issue.

The second point I want to make is that if we spoke about shadow banking, that whatever we do, it is still the case that the best way to address the challenge of shadow banking is to bring the sunshine, to take them out of the shadow, openness and transparency is the key. We are only surprised from things that we do not see, and you know there is a saying that what you see is what you get, and what you don’t see gets you.

The third point I want to make is that, and I like very much the perspective that Klaas made, that independence of central bank is designed, one of the achievements is that it lowers the cost of disinflation, which means that while we want very much to ensure that we have price stability, we must recognize that there will be deviations, and we will need to address stabilizing and reducing the inflation that we missed. And that’s where the independence is so critical.

My final remark about this has to do that, this brings me to the point of memory. Markets have memory, but policymakers also need to have memory. And the fact is that we have learned in the hard way that the most important three elements to bring about resilience is. Capital, watch for excessive leverage, and make sure there is enough liquidity. If you have those three, then you are in a way more ready to address whatever structures that we have.

The world is very complex, and there is a tendency to find very complex solution to a complex world. And here I will only remind us about the dictum of Albert Einstein, saying you should always look for the simplest solution, but not simpler than that. Which is the world is complex. Don’t be surprised that it is a matter for professionals when political leaders or politicians come into this world, they are not always well equipped, and that’s where the independents come in to make sure that we protect ourselves from ourselves and from our leaders that are very important. But in a professional world, this is not always the thing that we have, and also my final remark was as Ceyla mentioned about the importance of communication and coordination, and the fact is that we are normally operating best during a crisis, and that’s one of the problems.

We know about the syndromes. There is a big story about a firefighters brigade that were sitting in the firefighters station and playing cards. And when they played cards, their boss said, “Well, you are doing something, nothing, and I will just dismiss you.” And that’s when the fire came, and there were no firefighters. It’s very important to understand that the role is to deal with the challenge that is not yet with us, but the challenge that may come, and this is where the preparedness that Ceyla spoke about is so important.

Ceyla PAZARBASIOGLU:

Yeah, very, very much agreed. I think opacity is a key concern. If you look at crises, and I’ve worked on quite a few, it’s usually leverage and opacity that creates the loop, which I think is critical in the context of AI. This was discussed a lot during the annual meetings. AI is reshaping productivity, labor markets, growth, also financial markets. So I’d love to hear from the panelists, we have seven and a half minutes, quickly about what are the implications for monetary policy potentially transmission channels to financial stability, and what would be your recommendation for either research or policymakers to focus on. I’ll start with Don and then go around the table.

Donald KOHN:

Well, I’ll try to be brief, Ceyla. It’s a huge subject. I mean, everybody has talked about the balance between aggregate demand and aggregate supply. The AI build out really is affecting demand before it affects supply. So it’s really adding to the inflation pressures, and it’s one of the reasons the Fed has cited for tightening monetary policy and being on that track.

I think you asked about what to research here, and one of the things I’ve been interested in trying to think a little bit about is the labor markets. So one of the things that’s happened in the U.S. already is we’ve had some pressure on demand and inflation. The labor markets have been very well behaved. Wage increases are low, consistent with the 2% price stability, if not a little lower, unit labor costs are below 2%. So we are starting from a different position than they usually think of with demand, and I think the central banks need to think about what this AI effect might be on labor markets. I can see an argument for a higher U*, the unemployment rate higher, because people have to find the job that suits them. There’ll be a lot of shuffling, but I do think what we’re seeing perhaps now is an argument that maybe U* is a little lower, the neighborhood is a little lower because workers are worried. They’re concerned. They’re not asking for wage increases. So, my response to your question, what should we be focusing our research on? I think the interaction of AI and labor markets and monetary policy seems a bit fitting.

Ceyla PAZARBASIOGLU:

Yes, which is very complex. Do you want to…?

Klaas KNOT:

Yeah, very briefly, I would want to keep a close eye on what part of the AI investment boom is actually debt finance, because this seems to be an area that forms equity finance and risk absorption capacity, but the numbers I’ve seen is are already on the order of magnitude of around 1/3, and history teaches me that these numbers are usually the floor because, on top of that, often there are off-balance sheet commitments on which we don’t have much visibility. But that would worry me a little bit because I would hope that we all agree that at least there is some potential for a correction to take place at some point, and given also the sensitivity of U.S. consumption for wealth effects nowadays, that might very quickly have wider ramifications.

Philip LANE:

So I think Europe is a very interesting location for this debate because we’re not producing the frontier models that the US are producing and China is producing. But what we are seeing is even the global supply chain for AI includes European firms. So, trade this year is stronger than we expected, including due to the trade impetus from the AI global supply chain, we are seeing firms investing. We think investment is about 10% stronger because of AI, and about a quarter of credit growth is being driven by AI, even though that sector is relatively small in overall scale, so in other words, there is a AI stimulus to the European economy, but it’s not at the same level as what we see in the US, for example.

But the financing is global, so we are seeing the increase in the long-term interest rate, and a lot of that, from our point of view, is a spillover. But let me agree with Don. We see it in our wage data. Wages this year are relatively moderate, despite the inflation increase. And I think it does connect to two concerns in Europe. One is the fear of jobs being lost through AI, but also the fear in a European context of firms, if you like, seeing a decline in activity because of increased competition from China. So there are two squeezes on European firms, which help to explain why wages are contained. So this year, maybe unusually, the area of inflation which is not really going up is services inflation. So we have energy, and to some extent, we have goods inflation because of the AI spillover, but services are pretty quiet, and so that is, if you like, maybe not what was originally expected, and I do think AI has a lot to do with it. But in the end, I mean, this is, I think, task number one for central banks is to make sure we are keeping up. It’s not international account data. It’s not hard data. So running surveys, really understanding the very fast-moving dynamic here, so it’s a super important area for central banks.

Ceyla PAZARBASIOGLU:

Saturday, we had the international banking seminar of the Group of 30, and Pablo was moderating the current head of BIS, and he talked about the BIS survey, which with households about what their expectations are, and it’s really very interesting results. If you haven’t seen it, not expected comments. So, yes.

Masaaki SHIRAKAWA:

In the interest of time, I don’t talk about even impact of AI. Rather, I’m interested in the impact of AI on democracy, which affect conduct of monetary policy and financial security policy. In this respect, I recall what happened when internet was starting to be used 30 years ago. There was an optimism that internet improved democracy by democratizing information flow and access. And of course, we had some positive events, but at the same time, we had various forms of adverse impact, and I wonder how AI improves the democracy by which monetary and financial stability is affected.

Ceyla PAZARBASIOGLU:

That’s very profound. Yes, agreed. Jacob, last word.

Jacob FRENKEL:

Well, since we are getting to the time limit, I will only make very brief point. Number one, the AI over the past two years, when we were looking at the world, that we saw so many signs of fragility and so many risks and so many dangers. And then the performance of the world ended up being better than what we feared. And normally, when we don’t know the answer, most people said yes. But it’s AI. AI became the answer to all missing forecasts in this regard. And I think that we should be careful on relying too much before we understand how it does it, and I think that Kristalina, the managing director of the IMF, when she made the presentations, she drew always two types of distinctions: the effect on demand versus the effect on supply. And when we speak about the effect on supply, we need to recognize that the AI is something which is called GPT, general purpose technology. So its impact on the supply has to be understood in the context of the general purpose technology that may be very very significant, but it is not coming overnight. On the demand, the valuation of assets that are still based on expectations and hopes for the future is already stimulating the demand. So the question is: Will it be coming to the central banks? Will it be more inflationary or less inflationary? This is the way in which the taxonomy works.

So let me just conclude by saying this remark. I think that the perspective that that our good friend Masaaki Shirakawa-san gave, the effect of it not just on the usual channels of what things that we are measuring, but rather on the functioning of the labor system, democracy, and what does it do in this regard? That’s an angle that is very important to highlight. Thank you.

Ceyla PAZARBASIOGLU:

Very much agreed. Well, we are out of time. This was an amazing panel. Thank you so much for your expertise, experience, wisdom, and in these very difficult times, I think it’s important for us to be, of course, optimistic about what technology can bring. But given all the geo-economic fragmentation, given all the opacity, and so on, I think it’s really important that we think of what bad things can happen and be prepared for them, and then hopefully by that we will minimize the cost. So, please, round of applause for the excellent panelists. Thank you. Thank you, Jacob.

2026 Bund Summit, jointly hosted by the China Finance 40 Forum (CF40) and Tsinghua University (THU), will take place on October 19 to 22, 2026, in Huangpu District, Shanghai. The four-day event will focus on “Technology, Geopolitics and the Global Economy.”

Committed to international engagement and professional depth, the event features 19 Bund Roundtables and 12 closed-door meetings. Discussions will address key issues in global monetary and fiscal policy, trade and investment, artificial intelligence, Fintech, technology finance, asset management, green finance, and Renminbi internationalization, among others.

A series of Track II exchanges involving leading global think tanks will also take place from October 18 to 24. That means the Bund Summit will anchor a seven-day program of international exchanges.

See below for a glance of the sessions we will have in the coming 4 days, and stay tuned at Bund Summit - Home.

On August 28, China’s Ministry of Housing and Urban-Rural Development (MOHURD) and other authorities issued the Notice on Improving the Commodity Housing Sales System and the Opinions on Reforming and Improving Real Estate Credit Management to Accelerate the Building of a New Model for Real Estate Development (hereinafter “the new policies”), and implementation rules have since been issued in multiple localities. Drawing on data for A-share listed developers and national macro data, this Brief estimates the impact of the new policies on construction, land acquisition, and capital turnover. Our main assessments are as follows:

First, the new policies will reshape developers’ turnover and credit models. While creating contractionary pressure in the short term, they will also accelerate market clearing, supporting a more rational development of the real estate market in the long run. Under the new model, developer financing shifts from interest-free presale proceeds collected in advance to interest-bearing development loans covering the entire construction cycle, and high-turnover developers will face pressure to contract. The shift from lending based on developers’ creditworthiness to lending based on individual projects’ repayment capacity may widen regional divergence in new construction starts. In the long run, however, the new policies will bring about a survival-of-the-fittest process among developers and real estate projects.

Second, whether development loans can fully replace presale proceeds is the key variable determining the short-term impact of the new policies. For new development loans to fully replace presale proceeds under the old model, demand for development loans in 2027 would need to roughly double relative to 2026. If a large development loan shortfall emerges in the short term, funding pressure will spill over to land acquisition and, in turn, translate into fiscal pressure on local governments.

Third, under the new model, developers’ land acquisition capacity will be constrained in both the short and long term; the long-term decline in land acquisition is an inevitable consequence of stripping developers of their shadow banking function. In the short term, the land acquisition funding gap for high-turnover developers may exceed 30%; in the long term, developers’ equity-funded land acquisition capacity per unit of time will fall by nearly 40%.

Fourth, under the new model, capital is tied up for longer and developers’ returns on capital will decline. With little room for significant house price increases, pressure on returns may be passed upstream along the industry chain, pushing down land and material prices and, in some cases, increasing construction payment arrears. However, as the gains from high turnover disappear, developers may gradually shift their focus from expanding development scale to improving project quality in order to sustain operating profits, which is conducive to the long-term healthy development of the real estate market.

Fifth, extending mortgage terms can modestly improve homebuyers’ affordability, but its boost to demand is limited, and it has no material effect on developers’ use of funds under the new model.

I. The August 28 Policies: Accelerating Completed-Home Sales and Reducing Developer Leverage Risks

On August 28, 2026, MOHURD, the Ministry of Natural Resources, and the National Financial Regulatory Administration (NFRA) jointly issued the Notice on Improving the Commodity Housing Sales System. On the same day, the People’s Bank of China (PBOC) and the NFRA jointly issued the Opinions on Reforming and Improving Real Estate Credit Management to Accelerate the Building of a New Model for Real Estate Development (collectively referred to below as “the new policies”).

(i) Key Policy Adjustments: Higher Presale Thresholds, Full Supervision of Sales Proceeds, and Development Loans Covering the Entire Construction Period

On the supply side, the new policies make two main adjustments. First, they tighten the use of presale proceeds: presales are deferred until topping out, sales proceeds are subject to full supervision throughout the entire process, and mortgage disbursement is deferred until after completion. Second, development loans replace presale proceeds in financing construction: loan terms cover the entire construction process, a lead bank system is introduced, and financing relies more on the debt-servicing capacity of the project itself.

There are also supporting measures on the demand side, including extending the maximum term of individual housing loans from 30 to 40 years, the impact of which is estimated below.

Overall, the new policies will make the use of real estate development funds more transparent and reduce mismatches between funds and the projects and periods in which they are used; help convert corporate credit into project credit, further stripping developers of their shadow banking function; and advance completed-home sales, safeguarding homebuyers’ rights and reducing delivery risks.

(ii) Local Implementation Rules: Consistent Core Provisions, with Transitional Arrangements Providing Some Buffer

In late September, Beijing, Shanghai, and other localities successively issued implementation rules clarifying how the new policies will be applied. The buffers that localities have added on top of the new policies mainly concern the dividing line between the old and new rules for existing projects, installment payments for land, and whether completion filing can be handled on a building-by-building basis.

First, a transition period for projects on land already acquired but without a construction project planning permit. Some localities allow projects that had not obtained the permit before the new policies but meet certain conditions to continue presales under the old rules. This allows some existing projects to temporarily retain their original financing arrangements, buying time for developers to gradually increase development loans and adjust their funding arrangements.

Second, longer payment periods for land. Some localities allow land payments to be made over two years interest-free, reducing the equity developers need to commit at the initial stage of land acquisition and freeing up more funds for turnover.

Third, permits and completion filing on a building-by-building basis. In some localities, this arrangement allows developers to construct, complete, and collect sales proceeds building by building, reducing the construction funds they need to advance and helping sustain capital turnover.

These arrangements operate over different time horizons. The grandfathering rules determine whether projects shift to the new rules en masse in 2027 or in phases over 2027–2028, whereas installment land payments and building-by-building filing can continuously reduce upfront investment and accelerate cash collection, easing funding pressure over the long term.

II. Potential Impact of the August 28 Policies

……

Please visit CF40 Research’s website to read this article in full.

New total social financing (TSF) was broadly flat year-on-year, mainly supported by government bonds and bill financing, while household and corporate financing demand remained weak.

M1 and M2 growth slowed, likely weighed down mainly by sluggish domestic credit expansion and slow fiscal spending, while foreign exchange may have continued to see net inflows.

Based on these data, we offer three observations, available here.

CPI year-on-year growth edged up, mainly lifted by energy and communication equipment. CPI month-on-month growth slowed, mainly because of a seasonal fall in travel prices and a smaller rise in oil prices. Adjusted core CPI rose slightly year-on-year, showing that consumption is stabilizing at the margin.

PPI rose on both a year-on-year and a month-on-month basis. This was mainly driven by raw materials such as energy and metals, and by AI-related sectors. Price increases in midstream and downstream industries were limited, which suggests their profits may come under further pressure.

We’d like to share three observations after combing through the data. Please see here for the analysis.