Using Multiple Imputation in the Analysis of Incomplete Observations in Finance

Incomplete observations are a common feature of financial applicationsthat use survey response, annual report, and proprietary bankingand security issue and pricing data. Finance researchers usea variety of procedures, including deleting offending observationsand imputing ad hoc values, that potentially fail to deliverefficient and unbiased parameter estimates. This article examinesthe application of a statistical framework, multiple imputationmethods, that minimizes incomplete data problems if the missingnesssatisfies certain criteria. When applied to two financial datasetsinvolving severe data incompleteness, the imputation methodsoutperform the ad hoc approaches commonly used in the financeliterature.