Quantile aggregation and combination for stock return prediction
研究了在股票收益预测中，通过聚合同一模型内不同条件分位数以及跨模型分位数平均的方法，能产生优于传统条件均值回归的预测，并在最优投资组合决策中带来更高的平均效用。
Model averaging for forecasting and mixed estimation is a recognized improved statistical approach. This paper is a first report on: (1). aggregate information from different conditional quantiles within a given model and, (2). model averaging with quantile averaging. Based on a subset of possible methods, we show that aggregating information over different quantiles, with and without combining information across different models, can produce superior forecasts, outperforming forecasts based on conditional mean regressions. We observe a variety of quantile aggregation schemes within a model can significantly improve over forecasts obtained from model combination alone. We provide simulation and empirical evidence. In addition economic value of our proposals is demonstrated within an optimal portfolio decision setting. Higher values of average utility are observed with no exception when an investor employs forecasts which aggregate both within and across model information.