COMPUTER GENERATION OF HIGHLY SKEWED CORRELATED RANDOM VARIABLES

Abstract

ABSTRACT A multivariate approach was previously developed for generating non-normal correlated random variables (Taylor and Bender, 1988, 1989). The method exactly preserves each univariate probability distribution; however, there may be cases where the correlations between variables are not preserved. The objective of this study was to empirically determine how well the multivariate approach preserves the correlations for highly skewed probability distributions. Three bivariate combinations were examined: (1) both distributions positively skewed, (2) both distributions negatively skewed, and (3) one positively and one negatively skewed. The multivariate method preserved the correlation between variables with excellent accuracy except for the case involving a mixture of positively and negatively skewed distributions in conjunction with high correlation. KEYWORDS. Probability distributions. Computer simulation..