讲座时间：2022年5月23日 14：00-15：30

讲座地点: 腾讯会议 187 546 825

主讲人：Yonghwan Jo （辽宁大学）

讲座题目：Improving return forecast accuracy of machine learning by simple sample splitting

讲座简介：Predicting returns using machine learning is attracting attention in academia. However, there are no in-depth studies of the role of training and validation sets in estimating parameters and hyperparameters given the unique characteristics of financial data. In this paper, I show that the performance of return prediction differs significantly depending on how the sample set is splitted into training and validation sets. Specifically, I find that putting recent data into the validation set performed better than putting recent data into the training set. I show the recursive sample splitting on both training and validation sets outperforms conventional splitting methods when using nonlinear machine learning models. The results can be explained by the relationship between future excess returns and predictors, which changes over time because of structural changes or the market evolution from mispricing.

主讲人介绍：Yonghwan Jo is an assistant professor of the Advanced Institute of Finance and Economics at Liaoning University since 2018. Yonghwan completed his Ph.D. at Korea Advanced Institute of Science and Technology (KAIST) with a focus in empirical asset pricing. His research studies financial markets and investments regarding the relationship between risk and expected return. He is currently working on understanding the risk premiums in futures markets and also anomalies via machine learning techniques.