Keynote Speeches
Reinforcement Learning Algorithm for Mixed Mean Field Control Games
Speaker: Professor Jean-Pierre Fouque (Distinguished Professor, University of California Santa Barbara)
Abstract
We present a new combined Mean Field Control Game (MFCG) problem which can be interpreted as a competitive game between collaborating groups and its solution as a Nash equilibrium between the groups. Within each group the players coordinate their strategies. An example of such a situation is a modification of the classical trader's problem. Groups of traders maximize their wealth. They are faced with transaction cost for their own trades and a cost for their own terminal position. In addition they face a cost for the average holding within their group. The asset price is impacted by the trades of all agents. We propose a reinforcement learning algorithm to approximate the solution of such mixed Mean Field Control Game problems. We test the algorithm on benchmark linear-quadratic specifications for which we have analytic solutions. Joint work with A. Angiuli, N. Detering, Mathieu Laurière, and J. Lin
The Use of Machine Learning in Treatment Effect in Treatment Effect Estimation
Speaker: Yu-Chin Hsu (Acting Dean, Institute of Economics, Academia Sinica)
Abstract
We present recent developments in double machine learning (DML) approach. The DML approach is concerned primarily with selecting the relevant control variables and functional forms necessary for the consistent estimation of an average treatment effect. We explain why the use of orthogonal moment conditions is crucial in this setting. We also discuss how DML approach can be applied to estimate the conditional average treatment effect (CATE) function conditional on a pre-specified coordinate.(This talk is based on “The Use of Machine Learning in Treatment Effect Estimation: Hsu, Lieli and Reguly (2022).” In F. Chan and L. Matyas (Eds.) Econometrics with Machine Learning, Springer.)
Personalized Loss Functions in Machine Learning for Risk Management
Speakers: Pierrick Piette (Speaker in person)（ISFA, Université Lyon 1）, Loisel Stephane (ISFA, Université Lyon 1), Chenghsien Jason Tsai (NCCU)
Abstract
Loss functions are at the core of machine learning algorithms: they mathematically determine the objectives that models are structured to optimize. However, insufficient attention is frequently paid to the choice of the loss function, and data scientists often retain the traditional ones, i.e. logistic for classification or mean squared error for regression. This can result in unoptimized algorithms regarding the economic reality of the company. This talk explores the concept of personalized loss functions in machine learning and how they can be used to better align the goals of statistical learning models with one’s specific risk management objectives. An example on lapse risk management in life insurance is discussed to quantitatively highlight the potential gain of personalized loss functions.