A Refined Fixed-Effects Estimator to Detect Fraudulent Action

Abstract

Active policy measures to resolve a problem must begin with the acknowledgement of the existence of the said problem. For instance, measures to curb tax evasion are only meaningful once the evading entities are systematically detected. I propose a method to econometrically detect units of observation that are potentially associated with malpractice. The method exhibits a remarkable detection rate and requires two stages. In the first stage, (biased) regression coefficients are systematically obtained through fixed-effects estimation (with varying lags). In the second stage, a structure is imposed on the (biased) estimates to single out the units of observation that are likely to involve malpractice. The method is explained in the context of sportspersons who may resort to drug intake (without reporting, for obvious reasons) for the qualifiers before an event of central importance. What makes the proposed exercise challenging is that unobserved drug intake may be correlated with observed physical attributes that may serve as regressors in a model to explain performance. Chemical processes of detection are often known to be costly and unhealthy and may not be free from errors.