TESTING FOR MULTIPLE UPPER OUTLIERS IN DISTRIBUTION SAMPLES: A STUDY OF FOREIGN EXCHANGE DATA

Abstract

In this study, the existences of k-upper outliers are investigated in distribution samples of gamma, Normal and exponential by carrying out a simulation of ten thousand at different values of n using algorithm introduced by Tietjen-Moore, test statistics, and critical values were equally estimated from the algorithm. A Normal Q-Q plot was made which aims at distinguishing a data set that follows a normal distribution and one that deviates from normality. The algorithm was applied to Nigeria-US dollars foreign exchange rate, both on raw and logarithmic transformed data. The simulation study reveals the existence of upper outliers more in Gamma and exponential samples than the Normal sample. Empirical analysis shows that there are upper outliers in the raw data set but no upper outliers are found in the transformed data. The result in this paper would help the researcher in business and economics to take time to explore data before use and properly transform accordingly to avoid error in estimation.