The Sixth Session of the Dialogue with a Nobel Laureate: Student Tea Talk Series at PHBS

On June 23, the sixth session of the “Dialogue with a Nobel Laureate: Student Tea Talk Series” was held at Peking University HSBC Business School (PHBS). The event featured Thomas J. Sargent, the 2011 Nobel laureate in economics and honorary director of the Sargent Institute of Quantitative Economics and Finance at PHBS (SIQEF), in a roundtable discussion with faculty and students. The session was moderated by Shi Jiao, associate professor at PHBS and deputy director of SIQEF.

Sargent delivers a lecture

The Logic of Rational Expectations

Responding to many students' questions regarding the concept of rational expectations, Professor Sargent delivered an in-depth lecture on the core logic, mathematical foundations, and applications of rational expectations models in modern economics. He emphasized that rational expectations is not a psychological model of how people actually think, but rather an assumption about how agents within a model behave. Within the model, agents make decisions based on their subjective models, and these decisions collectively shape the objective model. In equilibrium, agents only need to form joint histograms; the law of large numbers then recovers the true distributions. The statistical foundations of this process lie in the stationarity and ergodicity of time series.

In an equilibrium Markov process, each agent's subjective probability distribution — encompassing expectations about both exogenous shocks and endogenous variables — implicitly determines an objective distribution. When the Kullback-Leibler (KL) divergence between the subjective and objective distributions is set to zero, the rational expectations equilibrium proposed by John F. Muth (1961) is achieved. This is essentially a fixed point of the mapping from subjective to objective distributions — a “communism” of statistical models.

He stressed that rational expectations is not a model of how people think about forecasting future outcomes. Its main advantage lies in its simplicity and coherence: it eliminates free parameters describing subjective distributions and sets the KL divergence to zero, thereby making the model builder immune to the question, “If you are so smart, why aren't you rich?” This coherent system of conditional distributions laid the micro-foundations for modern macroeconomics, opening doors to framing problems such as time inconsistency of optimal plans, credibility and reputations, and “expectations management” in structured models.

Sargent answers questions from students

On AI, Learning, and the Limits of Models

In the interactive session, students raised questions on the relevance of rational expectations in the age of AI, the boundaries of human and machine capabilities in handling high-dimensional problems, and the predictive power of rational expectations models.

Q: Some people think in the age of AI, every person will have an AI agent to make all the decisions. Isn't the concept of rational expectations going to come into its own?

Sargent: Not going to happen soon. In very simple theoretical economics even the best AI algorithms cannot perform very well because there is too much to learn about.

Q: Since AI can handle low-dimensional problems, will people still learn to deal with higher-dimensional problems in the future, or can it be done by AI?

Sargent: People are not good at solving high dimensional questions, so we always compress it, so does AI. AI can be useful for some things and not at others. When I was teaching in the US, AI was very harmful. You have to almost give oral exams to avoid cheating. More and more people are now saying I don't have to learn things because AI does a good job. A small fraction of people do learn the basics and then use AI. They become fairly powerful. AI is a complement.

Q: Can rational expectations really predict people's behavior?

Sargent: Sometimes they work pretty well. It takes a great deal of effort to build a model. Some models are not even coherent.

Q: How can business students better equip themselves in the face of AI anxiety?

Sargent: The question is whether you think it will help you. If you learn the basics and use AI as a complement, I think AI is going to help you.

The tea talk provided students with a rare opportunity to engage directly with a Nobel laureate on the foundational ideas that continue to shape economic theory — and to reflect on how those ideas intersect with the transformative challenges of artificial intelligence.

By Annie Jin, Shi Ningjing 

Source: SIQEF and PR & Media Office