Mosaics of Predictability: A New Framework for Asset Return Forecasting

This National Bureau of Economic Research working paper, titled 'Mosaics of Predictability,' introduces a novel approach to understanding stock return predictability. Authors Lin William Cong, Guanhao Feng, Jingyu He, and Yuanzhi Wang argue that predictability is a latent, asset-specific, and state-dependent characteristic rather than a uniform market feature. They develop an interpretable Panel Tree method that partitions the U.S. equity panel into persistent 'mosaic' patterns, allowing for cluster-specific forecasting models. The study finds that predictability is concentrated in stocks with large earnings surprises, high earnings-to-price ratios, and low trading volume. Furthermore, it exhibits countercyclical behavior, strengthening when market dividend yields are high and liquidity is low. By accounting for this heterogeneity, which conventional models often ignore, the authors demonstrate improved forecast accuracy and portfolio performance, achieving out-of-sample Sharpe ratios around 2. The research utilizes fifty years of data to map where significant signals arise versus where noise dominates, offering valuable insights for financial economics and asset pricing strategies.