Research
Working papers
Empirical investigation on supervised machine learning models predicting equity risk premium
(Job market paper, Current manuscript, last update September 2023)
Abstract: We examine the predictive performance of supervised machine learning models in forecasting multi-horizon firm-level equity risk premium. We use an extensive collection of individual firms’ financial characteristics and US macroeconomic return predictors for data. We forecast excess returns for (1) all individual firms and (2) each group of firms belonging to the same industry sector in the US. We first show an out-of-sample fit for each forecast model. Second, we forecast and evaluate models pairwise to find ones with superior predictive ability. We also estimate model confidence sets collecting models with superior predictive ability. Finally, we test for a model’s conditional superior predictive ability, where a model’s predictive ability is determined conditionally on a priori chosen variable indicative of the state of the market.
Keywords: Big Data, Supervised Machine Learning, Return predictability, Forecast evaluation
Evaluation of supervised machine learning models predicting equity risk premium in South Korea (working)
Abstract: We examine the predictive performance of supervised machine learning models in forecasting monthly firm-level equity risk premium. We collect firms whose stocks are currently listed or were listed on Korea Exchange in the past. We collect firms from January 1990 to December 2019 through the Worldscope database and use their financial characteristics as predictors. We forecast and evaluate models pairwise to find ones with superior predictive ability. We also estimate model confidence sets collecting models with superior predictive ability. We also investigate the importance of predictors based on their mean decrease in l2 impurity using the random forest model. For our data, weekly and monthly price trends contribute the most.