Machine Learning Methods for Portfolio Risk Measurement
DescriptionPortfolio risk measurement has been an important computational problem in financial engineering and risk management, especially when the portfolio involves complex financial derivatives and/or sophisticated mathematical models. Under these settings computer simulation is usually required in portfolio reevaluation, and a central research question is then how to improve the computational efficiency of the risk estimation via simulation. In this proposal, we propose new simulation methods for portfolio risk measurement, based on machine learning techniques. The main theme of these methods is to borrow the ideas behind the balance of the trade-off between exploration and exploitation in machine learning techniques to guide the allocation of simulation budgets for portfolio risk measurement, thus increasing the efficiency of the risk estimators. In particular, we consider several important classes of risk measures and plan to develop tailored-made methods by examine carefully their unique structures. In addition, we plan to study the statistical properties of these machine learning methods, including their consistency and central limit theorems, providing theoretical supports and better understanding of the insights of these methods.
|Effective start/end date||1/01/21 → …|