In my recent interview with Steven Budgen on the Achieving Alpha Podcast (see the podcast below), we delved into some pressing economic issues, particularly focusing on China’s economic trajectory and its implications for the global economy.

Here are my key insights and reflections from that discussion, as well as references to my 2014 blog post, “China might never be the largest economy in the world.”

Demographic Challenges in China

China faces significant demographic headwinds. The country’s labour force has been shrinking for more than a decade, and this trend is expected to continue in the coming decades. This demographic shift mirrors what Japan experienced, leading to a continuous slowdown in growth. As the labour force declines, so does the potential for sustained economic growth. Without a reversal or mitigation of these demographic trends, China’s growth will continue to slow.

Investment Imbalances

For years, China has maintained extraordinarily high investment ratios, often exceeding 40% of GDP. While investment is typically seen as a positive driver of growth, in China’s case, it has led to overinvestment and malinvestment, particularly in the real estate sector. A significant portion of this investment has been driven by government subsidies and easy credit, resulting in unsustainable growth patterns. Moving forward, China needs to shift from an investment-led growth model to one that is consumption-led to ensure economic stability and health.

Policy and Governance Issues

Since President Xi Jinping took power, China’s economic policies have become increasingly less market-friendly. The shift towards more totalitarian governance and aggressive foreign policy has scared off international investors. This change, coupled with a lack of substantial economic reforms, has stifled productivity growth and innovation. Total factor productivity, which is crucial for sustainable long-term growth, has been negatively impacted by these policy shifts.

Future Projections for China’s Economy

Given the demographic pressures and the need to normalise investment levels, I predict that China’s real GDP growth trend will continue to decline, potentially nearing zero within the next decade or so. The government’s attempts to stimulate the economy through demand-side measures, such as fiscal stimulus and credit policies, are unlikely to be sufficient without significant supply-side reforms.

In my 2014 blog post, “China might never be the largest economy in the world,” I argued that China’s growth would slow dramatically due to demographic challenges, overinvestment, and the diminishing returns of the catch-up process. These factors remain relevant today and continue to support the view that China’s growth will decelerate significantly.

Political Constraints

The Chinese Communist Party’s reluctance to liberalise politically limits economic freedom and innovation, further constraining economic growth. The lack of supply-side reforms means that the slowdown in trend growth is likely to continue. China could do a lot to open up its economy, develop financial markets, and liberalise its political and economic systems, but these steps seem unlikely under the current political regime.

Global Impact and Comparisons

There are notable parallels between China’s current situation and Japan’s economic stagnation. However, Japan managed to navigate its crisis without major political upheaval due to its democratic governance structure. In contrast, China faces greater challenges due to its lower per capita income and political structure.

Impact of AI and Technology

I’m optimistic about the potential of AI to drive productivity growth globally, particularly in high-income countries. However, China, being a manufacturing-focused economy, may not benefit as much from AI-driven productivity gains as high-income countries. AI is likely to introduce significant productivity gains in sectors like finance, healthcare, and retail, which are more prevalent in developed economies.

US Economic Prospects

Despite the challenges faced by China, I’m very optimistic about the US economy. Several factors contribute to this optimism: the potential of AI, the impact of anti-obesity drugs, and strong immigration trends. These elements could lead to significant productivity gains and robust economic growth in the US, contrasting sharply with the challenges faced by China.

Policy Recommendations

For China to regain robust economic growth, it needs to implement significant supply-side reforms, including opening up its economy, developing financial markets, and liberalising political and economic systems. Additionally, central banks, particularly the Federal Reserve, should focus on nominal GDP growth targets rather than strict inflation targets to better manage the economy in the face of productivity shocks from technological advancements.

Concluding Thoughts

Economic freedom and innovation are essential for sustained economic growth. Historical perspectives and current global economic trends underscore the importance of policies that enhance productivity and economic flexibility. As we look to the future, it is crucial for policymakers to focus on these areas to navigate the challenges ahead.

These are some of the critical points Steven Budgen and I discussed on the Achieving Alpha Podcast. I believe understanding these dynamics is essential for anyone interested in the future of the global economy. You can read more about my earlier predictions in my 2014 blog post, “China might never be the largest economy in the world.”

I have had a look at a new interesting paper – “Financial Statement Analysis with Large Language Models” – where the authors investigates whether an Large Language Model (LLM) can predict earnings as effectively as human analysts.

Summary of the Study

In the study, researchers Alex G. Kim, Maximilian Muhn, and Valeri V. Nikolaev from the University of Chicago examine whether LLMs can perform financial statement analysis similarly to professional analysts. They provide standardized and anonymous financial statements to GPT-4 and ask it to predict the direction of future earnings.

Methodology:

1. Data Preparation: The financial statements are anonymized and standardized, removing any identifying information to ensure the model can’t rely on prior knowledge of specific companies.
2. Prompts: Two types of prompts are used – a simple prompt and a Chain-of-Thought (CoT) prompt. The CoT prompt guides the model through a detailed analytical process, similar to a human analyst’s workflow.
3. Comparative Analysis: The LLM’s predictions are compared to those of human analysts and other machine learning models, such as logistic regression and artificial neural networks (ANNs).
4. Evaluation Metrics: The accuracy and F1-score are used to measure prediction performance. The study also examines the economic usefulness of trading strategies based on LLM predictions.

Main Takeaways

1. Prediction Accuracy: The study finds that GPT-4, especially when using the CoT prompt, can predict earnings changes with an accuracy that surpasses human analysts. This performance is on par with, and sometimes exceeds, state-of-the-art specialized machine learning models.
2. Economic Insights: GPT-4 doesn’t just rely on training memory but generates useful narrative insights about a company’s future performance. This capability allows it to form predictions that are often more accurate than those made by human analysts.
3. Market Performance: Trading strategies based on GPT-4’s predictions yield higher Sharpe ratios and alphas compared to those based on other models, suggesting potential financial benefits.

My Perspective: LLMs and Market Predictions

While the study convincingly demonstrates that LLMs like GPT-4 can predict earnings changes effectively, it’s important to distinguish between predicting earnings and predicting market prices.

Earnings predictions do not directly translate to market predictions, especially in highly liquid markets like the US stock market.

The reality is that in liquid markets, where numerous participants are using advanced machine learning techniques, beating the market consistently is incredibly challenging. These methods have become widespread, and any potential edge from using LLMs would likely be arbitraged away quickly. Thirty years ago, there might have been more opportunities to exploit such models, but in today’s highly efficient markets, the competition is fierce.

Where LLMs Can Make a Difference

If one aims to leverage LLMs and machine learning to “beat the market,” the focus should shift to less liquid markets. Illiquid markets, such as certain niche commodities or retail prices (or the price of for example soccer players), may offer more opportunities for these models to provide a competitive edge. In these markets, the relative lack of sophisticated competition and slower dissemination of information could allow LLMs to identify and exploit inefficiencies more effectively. But with the explosion in the use of LLMs and Machine Learning models we will use see these markets becoming “as financial markets” in terms of efficiency and market pricing.

Conclusion

In conclusion, while LLMs show great promise in performing financial analysis and predicting earnings, their ability to outperform in highly liquid markets remains questionable. For those looking to leverage these advanced models, the key may lie in targeting illiquid markets where their predictive power can provide a true advantage.

Contact:

Contact:

Lars Christensen

+45 52 50 25 06

LC@paice.io

I have been thinking about the impact on the economy of an increase in productivity y*, for example, due to the AI revolution, and I assume that it takes time for the central bank to realize that the increase in y* has also increased the natural real interest rate r*.

I have tried to model this in a simple (New?) Keynesian style model.

I should stress this is preliminary work and is basically just an exercise in seeing how much modelling I can do using ChatGPT 4o (and Python), while at the same time gaining real insights into the macroeconomic process and obtaining new nuances in the relationship between productivity shocks, imperfect central banking, and inflation.

In the modelling I have done so far, I assume that the central bank forms expectations of r* through an adaptive process over four years.

The graphs below illustrate the dynamic response of the macroeconomic variables to a shock in potential output growth.

Initial Shock and Aggregate Demand

In the first year, the potential output growth rate increases from 2% to 2.5%. This represents an improvement in the economy’s productive capacity.

This surge occurs because the economy reacts positively to the higher productive capacity, leading to increased economic activity and higher aggregate demand.

However, notice in the graph that aggregate demand initially ‘overshoots’ the increase in potential output. This is a result of the actions of the central bank, as we will see below.

Central Bank’s Initial Response

Despite the immediate increase in potential output grow, the central bank does not immediately adjust its perception of the natural rate of interest. The central bank is, so to speak, backward-looking. And for good reason – the natural interest rate cannot be observed in real-time.

As a result, the central bank initially lags behind the curve, failing to recognize the new higher natural rate. Hence, we see in the graph that the natural rate (the orange line) increases in year 1, but the perceived natural rate (the green line) only moves up with a lag.

As the perceived rate increases, the central bank also starts to push up its policy rate (the blue line).

It is notable that while the central bank initially lags behind the increase in the natural rate, once it starts hiking interest rates, it will in fact increase rates more than the rise in the natural rate. This is due to the fact that the central bank is not only reacting to the perceived increase in the natural rate but also to aggregate demand growth outpacing potential GDP growth.

Furthermore, it is also notable that the central bank will have to keep its policy rate somewhat above the natural rate (both the perceived and the actual) for some time to push down aggregate demand and tame inflationary pressures.

Consequently, because the central bank only slowly recognizes the increase in the natural rate, it will later have to overshoot its policy rate as well. More on that below, but first, let’s have a look at the inflation dynamics.

Inflation overshooting despite disinflationary effects of higher productivity growth

Due to the output gap created by the increase in aggregate demand and the lag in the central bank’s response, inflation begins to rise. The Phillips curve relationship in the model explains this rise, as the economy’s aggregate demand temporarily exceeds the new potential output, putting upward pressure on prices.

This could be paradoxical as we would normally expect a positive supply shock to put downward pressure on prices (and inflation – at least for a period). This effect is also in play, as a higher growth rate of y* tends to push down the output gap, but this effect is smaller (in this parameterization) than the demand effect coming from an overly easy monetary policy.

Catching Up: Central Bank’s Aggressive Response

Realizing the inflationary pressures, the central bank responds aggressively. The monetary policy rule dictates that the nominal interest rate be adjusted based on the deviation of actual inflation from the target inflation. Given the parameters in the model, the central bank increases the interest rate significantly. This adjustment is necessary to combat the inflation spike and bring inflation back to the target as discussed above.

Interest Rate and Natural Rate Convergence

As the central bank reacts, it also begins to update its perception of the natural rate of interest. In the interest rate graph above, we see the green line starting to approach the orange line. The learning rate parameter in the model ensures that the central bank updates its perception relatively quickly, though initially, it had been behind the curve. Over time, the central bank’s perceived natural rate converges to the actual new natural rate, which aligns with the higher potential output.

Aggregate Demand Slowdown and Inflation Stabilization

To force aggregate demand and inflation down, the central bank pushes the interest rate above the new natural rate of interest. This aggressive stance is crucial to moderating economic activity and addressing the inflation overshoot.

As the higher interest rates take effect, aggregate demand gradually declines as we see in the first graph above. In fact, we see that for a period, actual GDP growth y drops below the growth of potential GDP y*. This is necessary to bring inflation back to the assumed 2% inflation target in the model.

While the initial ‘boom’ that pushes actual GDP growth above potential growth only lasts 1-2 years and is followed by a fairly long period of growth below potential growth (but still above the 2% potential prior to the productivity boost), inflation remains elevated for a much longer period. In this parameterization of the model, it takes around seven years to get back to 2%, even though inflation is only slightly above the 2% target for most of this period.

I should, of course, stress that the actual parameterization of the model affects the size and length of the movements in GDP growth, inflation, and interest rates. I would not place too much emphasis on the scale of the shocks, but I fundamentally believe the model provides some important insights.

I would particularly highlight the following points:

Aggregate Demand and Potential Output

The model shows an initial spike in aggregate demand above the new potential output level, followed by a gradual convergence as the central bank’s policies take effect.

Interest Rates

The nominal interest rate initially lags behind the new natural rate but then adjusts significantly, overshooting the natural rate to combat inflation before eventually converging to the new natural rate.

Inflation

Inflation initially overshoots the target due to the central bank’s initial lag in response but gradually returns to the target as the central bank’s aggressive policies bring aggregate demand and inflation under control.

And again, even though I would caution against taking the numbers in this simulation too literally, I think there is an important insight that comes from the simulations. That is, we probably should not expect the AI revolution to be hugely deflationary if the central bank is slow to recognize that the natural interest rate has increased. This is due to the fact that demand-side inflation will likely more than offset the disinflationary effects of high productivity growth.

Is this what we are seeing now?

I do, in fact, think that we are seeing some of that right now in the US economy. At least, that could help explain why the US economy continues to grow robustly and why global stock markets and commodity markets continue to inch up while market interest rates have also increased.

The reason that market expectations of long-term inflation haven’t risen may reflect the fact that markets are indeed more forward-looking than central bankers. Markets realise that sooner or later r will have to move towards r*. That does not necessarily imply rate hikes but at least suggests that r might not be cut anytime soon.

The policy conclusion from all this is that central bankers should be forward-looking (we knew this), but it might be difficult given the uncertainty regarding the size of the shock to productivity and therefore to the natural interest rate. This underscores that the focus should be on nominal demand growth rather than short-term developments in inflation (which might either increase or decrease).

This means that central banks ideally should introduce nominal GDP targets rather than inflation targets. This is particularly important in periods of high uncertainty regarding the supply side of the economy.

PS: I certainly encourage any economist (or aspiring economist) to use ChatGPT 4o or other large language models to help with writing code for macroeconomic modelling. It is a great tool, but never forget to think. After all, you are the economist – not the LLM. If you want to play around with the model – see the appendix for more on the theoretical model as well as the entire Python code.

Contact:

Lars Christensen

+45 52 50 25 06

LC@paice.io

Appendix

Python code:

import numpy as np
import pandas as pd
import matplotlib.pyplot as plt

Initial conditions

y_0 = 0.02
y_star_0 = 0.02
r_0 = 0.02
r_star_0 = 0.02
pi_0 = 0.02
tilde_r_star_0 = 0.02
pi_star = 0.02

Parameters

alpha = 4.5 # Increased responsiveness to inflation overshooting
beta = 0.5
lambda_ = 0.7 # Increased learning rate for faster convergence
gamma = 1.0
kappa = 0.3 # Increased sensitivity of inflation to the output gap

Simulation parameters

years = 40
y_star_path = [0.02]1 + [0.025](years – 1) # y* changes to 2.5% at year 1

Initialize arrays to store results

y = np.zeros(years)
y_star = np.zeros(years)
r = np.zeros(years)
r_star = np.zeros(years)
pi = np.zeros(years)
tilde_r_star = np.zeros(years)

Set initial values

y[0] = y_0
y_star[0] = y_star_0
r[0] = r_0
r_star[0] = r_star_0
pi[0] = pi_0
tilde_r_star[0] = tilde_r_star_0

Run simulation

for t in range(1, years):
y_star[t] = y_star_path[t]
r_star[t] = gamma * y_star[t]
tilde_r_star[t] = tilde_r_star[t-1] + lambda_ * (r_star[t-1] – tilde_r_star[t-1])
r[t] = tilde_r_star[t] + alpha * (pi[t-1] – pi_star)
y[t] = y_star[t] – beta * (r[t] – r_star[t])
pi[t] = pi[t-1] + kappa * (y[t] – y_star[t])

Convert to DataFrame and multiply values by 100 to display percentages

df = pd.DataFrame({
‘Year’: np.arange(years),
‘y’: y * 100,
‘y_star’: y_star * 100,
‘r’: r * 100,
‘r_star’: r_star * 100,
’tilde_r_star’: tilde_r_star * 100,
‘pi’: pi * 100
})

Plot the results up to year 10

plot_years = 10
df_short = df[df[‘Year’] < plot_years]

plt.figure(figsize=(12, 8))

plt.subplot(2, 2, 1)
plt.plot(df_short[‘Year’], df_short[‘y’], label=’y (Aggregate Demand)’)
plt.plot(df_short[‘Year’], df_short[‘y_star’], label=’y* (Potential Output)’, linestyle=’–‘)
plt.title(‘Aggregate Demand and Potential Output’)
plt.xlabel(‘Year’)
plt.ylabel(‘Growth Rate (%)’)
plt.legend()

plt.subplot(2, 2, 2)
plt.plot(df_short[‘Year’], df_short[‘r’], label=’r (Interest Rate)’)
plt.plot(df_short[‘Year’], df_short[‘r_star’], label=’r* (Natural Rate)’, linestyle=’–‘)
plt.plot(df_short[‘Year’], df_short[’tilde_r_star’], label=’~r* (Perceived Natural Rate)’, linestyle=’-.’)
plt.title(‘Interest Rates’)
plt.xlabel(‘Year’)
plt.ylabel(‘Rate (%)’)
plt.legend()

plt.subplot(2, 2, 3)
plt.plot(df_short[‘Year’], df_short[‘pi’], label=’π (Inflation)’)
plt.axhline(pi_star * 100, color=’r’, linestyle=’–‘, label=’π* (Inflation Target)’)
plt.title(‘Inflation’)
plt.xlabel(‘Year’)
plt.ylabel(‘Rate (%)’)
plt.legend()

plt.tight_layout()
plt.show()

What should you spend the next 37 minutes on?

You should listen to me talk about why I remain so optimistic about the West, and especially why I am convinced that the USA will continue to be the world’s largest economy – and yes, why the status of the dollar as the global reserve currency is in no way in danger.

And you will hear me explain why Chinese growth is likely to continue stagnating.

That and much more – including the role “good governance” and AI play in future growth – in the latest edition of “The David Lin Report,” which I have just participated in.

Remember to follow me on Linkedin and X.

Tonight, I had a discussion with Nik Bhatia, who had invited me to participate in his podcast ‘The Bitcoin Layer’ to share my views on the continued weakening of the Japanese yen, and why I believe that the Chinese renminbi will follow the same weakening trend that the Japanese yen has experienced over the past 30 years (in real terms).

It’s about negative demographics and a lack of economic reforms. Nik and I also talked about the significance of AI for global growth – and interestingly, why this specifically impacts the yen negatively, given the Japanese monetary policy.