2022 CMS Winter Meeting

Toronto, December 2 - 5, 2022


Machine learning in finance
Org: Michael Chen (York), Hyejin Ku (York), George Lai (WLU) and Hongmei Zhu (York)

MICHAEL CHEN, York University

VICTOR HUANG, York University

YINGYAN JIA, Wilfrid Laurier University

ROY KWON, University of Toronto
Data-driven Integration of Norm Penalized Mean-variance Optimization  [PDF]

Mean-variance optimization (MVO) is known to be sensitive to estimation error in its inputs. Norm penalization of MVO programs is a regularization technique that can mitigate the adverse effects of estimation error. We augment the standard MVO program with a convex combination of parameterized L1 and L2 norm penalty functions. The resulting program is a parameterized quadratic program (QP) whose dual is a box-constrained QP. We make use of recent advances in neural network architecture for differentiable QPs and present a data-driven framework for optimizing parameterized norm-penalties to minimize the downstream MVO objective. Historical simulations using US stocks and global futures data demonstrate the benefit of the integrated data-driven approach.

YONGZENG LAI, Wilfrid Laurier University

CHIFENG SHEN, York Univeristy
Bayesian Online Changepoint Detection in Finance  [PDF]

Changepoints are utilized to split a data sequence into non-overlapping segments. The Online detection of changepoints method can be applied in many fields, especially in finance. In this talk, the case where the parameters between the segments are independent is examined, and an online algorithm for computing the probability of the most recent changepoint is derived.

YAODE SUI, Wilfrid Laurier University

WEI XU, Toronto Metropolitan University
Random Willow Tree with Application in Risk Management  [PDF]

Derivatives underlying a portfolio is popular on the market to diversify the market risk. However, existing method, the nested simulation, is quite time-consuming for pricing and managing the risk. In this article, we propose an efficient approach, randomized willow tree method. There are three main stages for our approach, portfolio distribution approximation, randomized willow tree construction and managing the risk of derivatives. We first generate some simulated paths to describe the evolution of dynamic portfolio values. Then, the minimal relative entropy (MRE) method is applied to approximate the distribution of portfolio values at each time based on the simulated data. After the approximated distributions are determined, a randomized willow tree can be constructed for pricing and managing the risk of derivatives underlying the portfolio. Finally, we apply the proposed approach to calculate annual dollar delta, 99\% VaR and CVaR of a particular derivative, i.e., a 19-year variable annuity with guarantee riders. This application demonstrates the efficiency and accuracy of the proposed approach compared with the common nested simulation technique, especially for a large pool of derivatives underlying the same portfolio.

© Canadian Mathematical Society : http://www.cms.math.ca/