The API and the Design of Experiments

Abstract

Residual analysis techniques, generically labeled Abnormal Performance Index (API) tests, have served as the primary experimental procedure in a wide variety of empirical financial accounting research studies. Through various types of API calculations, investigators have observed the security price behavior which preceded and accompanied such events as stock splits, secondary distributions, annual earnings announcements, accounting changes, and earnings forecasts.' Other API studies have drawn inferences about the content of quarterly earnings reports, tax accounting procedures, depreciation methods, and product-line reporting.2 The range of interpretations has included the demonstration of statistically unusual behavior, the imputation of information content or relevance to investors, and the expression of investor preference for a particular accounting technique. The discussion presented here attempts analytically to characterize the experimental design of API studies and to focus attention on the role of models of investor expectations. I address the questions of what constitutes an adequate model, what makes one model better than