Measuring Incomplete-Cross-Section Leakage: A Matched Placebo, a Susceptibility Screen, and Evidence from Taiwan and US As-Filed Data that the Channel Is Benign and Correctable

Many cross-sectional factors, such as industry-relative ratios, abnormal accruals, and the residual of a same-period regression, require the full cross-section of peer firms. Because firms file on staggered dates, a characteristic constructed shortly after a fiscal period closes is estimated from only those firms that have already reported. We term the resulting concern incomplete-cross-section leakage and assess whether it biases backtested performance. The conventional diagnostic, namely the disagreement between the early-filing signal and the complete-cross-section signal, does not identify the bias, because that disagreement combines three components: sampling variation from the smaller subsample, a difference in the covariate composition of the early filers, and any dependence of filing speed on the component of the signal under study. Only the third constitutes leakage. We isolate it by holding the covariate composition fixed, either through a matched control group or, equivalently, through the partial correlation between filing latency and the regression residual conditional on the regressors. This partial correlation is the selection-bias term of a Heckman model; because it is considerably less noisy than the disagreement statistic, we adopt it as the primary measure and provide it as an ex-ante screen, validated on a controlled experiment with a known planted bias. We apply the framework to two markets. For Taiwan, abnormal accruals and industry-adjusted inventory change exhibit no detectable leakage, and the associated long-short Sharpe ratio is bounded within 0.13. For the United States, where filing dates are observed directly through the SEC EDGAR system, profitability characteristics appear strongly distorted (filing latency correlates approximately -0.30 with industry-adjusted return on assets), yet the distortion is almost entirely attributable to firm size: larger firms file earlier and are more profitable, and conditioning on size reduces the partial correlation to approximately +0.09. The bias is therefore small in both markets, and the portion that remains operates through an observable characteristic that a size-aware release rule corrects. A companion paper develops that release rule; the present paper quantifies the magnitude of the bias it addresses.