STOCK RETURN QUANTILE PREDICTION - USING IVX-QR METHOD

Abstract

This paper seeks to contribute to the empirical literature on forecasting US equity premium using predictive (quantile) regressions. Previously, Goyal and Welch (2008) found that popular predictors like the Dividend/Earnings (D/E) ratio fail to deliver significantly better forecast accuracy than the prevailing mean model. Cenesizoglu and Timmermann (2008), on the other hand, provided evidence of usefulness of these predictors in forecasting some of the quantiles of the equity premium distribution by implementing the CAViaR model (Engle and Manganelli (2004)). This paper further extends the application of quantile-based forecasting techniques along several directions. First, it investigates whether the quantile forecasting model selection based on the Akaike and the Minimum Description Length information criteria leads to improved forecasting performance. Second, it applies the IVX-QR method of Lee (2016) that filters persistent predictors in a way that preserves more statistical power than first differencing. When comparing the selected models with the corresponding unconditional benchmarks, our results show that some of them considered significantly outperform the benchmark forecasts and help investors attain higher utility even when considering the least favorable scenario of the US financial crisis of 2007-2009.