Nonparametric predictive regression for stock return prediction
提出多步非参数预测回归模型，允许时变/非线性可预测性，用核估计方法证明大样本性质，实证显示在样本外预测和交易策略上优于历史均值、线性回归和多种机器学习方法。
.We propose a multi-step nonparametric predictive regression model, which allows for general locally stationary predictors and time-varying/nonlinear return predictability. We propose a kernel estimation method and establish the large sample properties in both short and long horizons. We apply our method to stock return prediction using a number of standard predictors such as dividend yield. The empirical results show that our proposed model can outperform the historical mean benchmark, linear predictive regression model, and several machine learning methods for some cases in terms of out-of-sample forecasting performance. We also compare our method with the historical mean benchmark using an economic metric. In particular, we show how our methods could be used to deliver a trading strategy that beats the buy-and-hold strategy over our sample period.