Multivariate Distributional Properties, Outliers, and Transformation of Financial Ratios

Accounting research that uses financial ratio data often assumes that sets of ratios have multivariate normal distributions. Multivariate distributional properties, multivariate outliers, and modified power transformations were examined to determine whether multivariate normality could be approximated for cross-sectional samples of financial ratios. The results were that the joint distribution of the financial ratios differed appreciably from multivariate normality and the financial ratio data contained multivariate outliers. Approximate multivariate normality was obtained by deleting multivariate outliers and applying modified power transformations to the ratios. Consequently, it would be possible to use multivariate outlier detection and transformation methods in accounting research to enhance statistical conclusion validity and to improve the effectiveness of decision models when multivariate methods that assume normality are used with financial ratios.