勷勤数学•专家报告
题 目:Liquidity, exploration and inference-based learning to reduce transaction cost
报 告 人:王海婴 教授 (邀请人:杨舟)
香港中文大学
时 间:11月9日 15:30-16:30
地 点:数科院西楼二楼会议室
报告人简介:
Hoi Ying Wong is a Professor in Department of Statistics at The Chinese University of Hong Kong (CUHK). He has been CUHK Outstanding Fellow since 2019. With primary research interest in Mathematical Finance and Risk Management, he published over 105 research articles in academic journals such as MF, F&S, SIFIN, SICON, SINUM, MOR, EJOR, Automatica, JEDC, JBF, JEF, JRI, IME, SAJ, etc. He was an Associate Editor (AE) of SIFIN in 2016-2022 and has been AE of IJTAF since 2004. Prof. Wong also takes a concurrent role as Associate Dean of Science (Student Affairs) at CUHK. He has consulting experiences with Hong Kong Monetary Authority, Commercial Banks, Hedge Funds and FinTech firms.
摘 要:
Reinforcement learning (RL) for trading has been the fast-growing field in mathematical finance since the success of machine learning algorithms in many practical problems. Unlike classic stochastic control approach which outputs an optimal policy for a given stochastic model, RL enables the investor not only to optimize the reward with a model (exploitation) but also to interact with the market, or equivalently the environment, through generating alternative policies (exploration). The success of RL relies on striking the best balance between exploitation and exploration. In this research, we show that exploration is unavoidable due to the ever-existing liquidity issue in trading. The optimized policy cannot be executed perfectly because the random liquidity spread makes the realized policy be randomized, causing investor to encounter a natural and passive exploration. We propose to calibrate the temperature parameter, or the level of exploration, in RL through liquidity data. Then, we propose an inference-based learning approach to form investment policies which reduce transaction cost due to liquidity. Our proposal involves a novel relaxed stochastic control problem with a regularization that combines Shannon entropy and KL-divergence relative to the optimal exploratory policy. We derive explicit solution to the relaxed stochastic control problem. Simulation demonstrates how this approach effectively reduces transaction cost and outperforms classic and exploratory optimal policies. (This is a joint with with Sixian Zhuang)
欢迎老师、同学们参加、交流!