Non-Random Sampling and Association Tests on Realized Returns and Risk Proxies

This paper investigates how data requirements often encountered in archival accounting research can produce a data-restricted sample that is a non-random selection of observations from the reference sample to which the researcher wishes to generalize results. We illustrate the effects of non-random sampling on results of association tests in a setting with data on one variable of interest for all observations and frequently-missing data on another variable of interest. We develop and validate a resampling approach that uses only observations from the data-restricted sample to construct distribution-matched samples that approximate randomly-drawn samples from the reference sample. Our simulation tests provide evidence that distribution-matched samples yield generalizable results. We demonstrate the effects of non-random sampling in tests of the association between realized returns and five implied cost of equity metrics. In this setting, the reference sample has full information on realized returns, while on average only 16% of reference sample observations have data on cost of equity metrics. Consistent with prior research (e.g., Easton and Monahan 2005), analysis using the unadjusted (non-random) cost of equity sample reveals weak or negative associations between realized returns and cost of equity metrics. In contrast, using distribution-matched samples, we find reliable evidence of the theoretically-predicted positive association. We also conceptually and empirically compare distribution-matching with multiple imputation and selection models, two other approaches to dealing with non-random samples