|
Abstract:
|
Audit sampling inference has been largely relied on central limit theory. However, due to the skewness of financial data, sample size required to allow a meaningful projection from sample to universe could be enormous and beyond audit capacity. Highly skewed and heavily tailed financial data often accompanied by a few extremely large values toward its tail. Thus, sample mean rarely follows normal distribution, unless sample size is extremely large. To minimize statistical inference bias resulted from small audit sample size for such skewed data, we developed a subsampling approach –– a nonparametric statistical method that applies the procedure of resampling without replacement from the universe to construct an empirical distribution for the statistical measures of interest (e.g. mean, median, variance, confidence intervals) between the universe and its corresponding samples.
|
