Liquidity Premium and Investment Horizons

Battle Count: 102

Highly relevant for quantitative trading. The paper provides a theoretically grounded, empirically implementable signal (signed order flow and Kyle's lambda) for short-horizon return prediction using only publicly available CRSP data. The Fama-MacBeth results show statistically significant cross-sectional predictive power (NW t-stats of -2.27 and -2.25). The framework directly informs portfolio construction (favoring low lambda stocks for short-horizon strategies), market-making spread calibration, and liquidity-aware execution. However, the modest R-squared values and specification sensitivity of lambda-based predictions temper the practical signal strength. The adverse-selection mechanism provides a clear economic rationale for contrarian liquidity-provision strategies.

Moderate complexity. The core methodology involves: (1) constructing signed order flow from daily CRSP data (straightforward), (2) running within-month OLS regressions per firm (computationally intensive for 9,893 firms x 60 months but parallelizable), (3) computing Amihud ratios (simple), (4) running panel and Fama-MacBeth regressions (standard econometrics), and (5) expanding-window out-of-sample forecasting (requires careful implementation of rolling estimation). The main challenges are: handling CRSP data cleaning/filters, managing the large panel (448,393 observations), ensuring point-in-time index membership, and correctly implementing Newey-West standard errors. No machine learning or complex optimization is required.

The paper uses CRSP daily and monthly data (available via subscription) and Wharton Research Data Service for factor returns. Methodology is described in detail including variable construction, filtering criteria, and regression specifications. However, no code repository is provided, and the exact CRSP query/filtering pipeline would need to be reconstructed. The expanding-window out-of-sample procedure is described but implementation details (e.g., exact initialization) could benefit from more specificity.