Economic aggregation of return signals in global markets

Abstract

I provide novel evidence supporting the robust predictability of the “signal zoo” by clustering and aggregating 84 signals based on economic similarity. Economic clusters not only exhibit high (low) within-cluster (between-cluster) signal correlations — comparable to $k$-means clusters — but also produce composites that non-redundantly explain the cross-section of U.S. stock returns. All composites exhibit robust predictability in the U.S. and certain evidence in the global regions. Subsample and long-run return tests suggest that predictability primarily arises from risk, except for momentum, which is driven by mispricing. Composites generally outperform an average-signal strategy due to their superior ability to identify less noisy stocks.