Adjusting Return Horizon According to Trade Activity with Application to Return Autocorrelation

We adopt a new approach to study individual stock returns' predictability from prior returns, and show that over short-horizons (daily to weekly), individual stock returns exhibit continuation for moderate prior returns, that is, those in the -1% to 1% interval, and reversal for extreme prior returns, that is, those outside the interval. Moreover, The region of continuation encompasses the central 40 percent of the distribution. We explain our findings with the interaction of the two effects, information asymmetry and inventory costs, influencing price formation process. We study the nonlinear pattern in the cross-section of stocks with different characteristics, such as idiosyncratic volatility, size, and liquidity, and find that only part of the pattern is due to the cross-sectional heterogeneity. More importantly, we show that the nonlinear pattern is invariant in the cross-section of stock returns with different trade activity. The critical requirement for this invariance is to compute returns over horizons that incorporate the variations in (and is inversely related to) stocks' trade activity