SOME APPLICATIONS OF COMPUTER INTENSIVE STATISTICAL METHODS TO EMPIRICAL RESEARCH IN ACCOUNTING (BOOTSTRAP)

Market-based, empirical research in accounting contends with several inconvenient statistical properties of price data. These difficulties include departures from normality and the need to estimate and account for cross-sectional covariances in multivariate time series which are short relative to their cross-sectional dimensionality. Researchers usually do not consider the potential impact of non-normality on normal theory tests, and they often rely on asymptotic approximations even with small sample sizes. These pragmatic accommodations are difficult to avoid but they tend to confound the interpretation of the resulting statistical inferences. Computer intensive methods such as Efron's bootstrap method could help to mitigate this concern by partially substituting for exact sampling theory.;The three studies that were selected are generally regarded as authoritative and are not known to be materially flawed. They examine the security market responses respectively to voluntary disclosures of earnings forecasts by corporations (using normal theory tests), to announcements regarding the regulation of accounting procedures in the oil and gas exploration industry (using nonparametric tests on ranks), and to announcements regarding the regulation of mergers (using asymptotic standard errors for approximate generalized least squares estimators). The dissertation examines the published conclusions for vulnerability to the type of "confounding" described above and re-examines the data, using the bootstrap method to circumvent this issue. The analysis includes simulation studies that assess both the potential magnitude of the "confounding" and the performance of the bootstrap method in each particular application.;In all three cases the statistical "confounding" is at least potentially capable of altering the original qualitative inferences. In two cases this effect does not alter the original inferences, but in the third case--the merger regulation study--there are strong counterindications to the published conclusions. These results document both that the "pragmatic accommodations" identified above merit careful consideration in market-based research, and that the bootstrap method can mitigate them successfully in some cases.;Abstract argumentation alone cannot resolve (1) whether the "confounding" identified above is operationally significant to accounting research, or (2) whether bootstrap methods in practice succeed in mitigating this concern. This dissertation uses the data and statistical models of three published information content studies as vehicles for examining these two questions.