Financial Market Microstructure, Big Data, and AI

Well-functioning financial markets provide liquidity to efficiently allocate resources and are fundamental to economic growth. A crucial aspect of designing, regulating, and monitoring financial markets is the ability to learn from the past and make informed decisions for the future. However, the limited availability of historical data restricts the ability to study the past. In contrast, the recent explosion of data and the rise of Artificial Intelligence (AI) introduce new risks for the foreseeable future.

To date, little is known about how to (i) estimate liquidity in the past when high-frequency data are not available and (ii) simulate the impact of AI on financial markets in the future. At the core of both issues is the lack of a practical understanding of how the price of an asset is determined in financial markets, which prevents connecting high-frequency to low-frequency data and hinders realistic simulations. In the last two years, I have been working on a model of price formation that predicts and explains novel testable relationships among key variables in financial markets. Here I conjecture that such a model may be used to connect the microstructure of financial markets to macroscopic effects. The goal of this proposal is thus to finalize the model and use it to improve estimation and simulation in critical applications. To do so, this proposal is articulated around three parts. Each part involves empirical analyses using the NYSE Trade and Quote (TAQ) database.

(Part I): The main objective is to explain relationships among variables. I propose to do so by finalizing and validating a price formation model based on the last chapter of my PhD thesis that I have further developed during my postdoctoral research.

(Part II): The main objective is to estimate liquidity from low-frequency data. Specifically, I propose to (i) use the relationships in Part I to construct estimators of liquidity from low-frequency data, (ii) estimate liquidity in the historical sample when high-frequency data are unavailable, and (iii) study liquidity in the cross-section and the time series since 1926.

(Part III): The main objective is to simulate the impact of AI on financial markets. Specifically, I propose to (i) simulate financial markets as trading games where AI agents learn to trade by maximizing their utility in an environment based on the model in Part I, (ii) perform theoretical analyses of the equilibrium and simulation studies of the dynamics, and (iii) identify emerging trends by comparing the theoretical and numerical results to the empirical data.

The project foresees collaborations with world-leading experts in market microstructure and artificial intelligence: Prof. Albert S. (Pete) Kyle from the University of Maryland’s Robert H. Smith School of Business, Prof. Albert Menkveld from Vrije Universiteit Amsterdam, Prof. Björn Hagströmer from Stockholm Business School, Prof. Stefano M. Iacus from Harvard University, Prof. Nicolò Cesa-Bianchi from the University of Milan, Prof. Semyon Malamud from the École Polytechnique Fédérale de Lausanne (EPFL).

The expected outputs are academic publications, open data containing liquidity estimates, and open-source software implementing the estimators and simulators developed in this project. Overall, this project contributes to a deeper understanding of how financial markets functioned in the past, function in the present, and will function in the future.