• Communications for Statistical Applications and Methods
• /
• v.15 no.6
• /
• pp.1013-1026
• /
• 2008

Kolmogorov-Smirnov(K-S) statistic has been widely used for the model validation of credit rating models. Validation criteria for the K-S statistic is empirically used at the levels of 0.3 or 0.4 which are much larger than the critical values of K-S test statistic. We examine whether these criteria are reasonable and appropriate through the simulations according to various sample sizes, type II error rates, and the ratio of bads among data. The simulation results say that the currently used validation criteria are too lower than values of K-S statistics obtained from any credit rating models in Korea, so that any credit rating models have good discriminatory power. In this work, alternative criteria of K-S statistic are proposed as critical levels under realistic situations of credit rating models.

• Journal of the Korean Data and Information Science Society
• /
• v.24 no.3
• /
• pp.437-444
• /
• 2013

The object of study is to verify the distributions of response packets of 3 portal sites such as Naver, Daum, Nate. The period of experiments is from May 19th 2010 to November 7th 2012 and the number of experiments is 4,642. The distributions of Naver, Nate are biomodals. The distribution of Daum has long right tails. 3 distributions are different under 1% significance level using chi-square test and two sample Kolmogorov-Smirnov test. From proportions and percentiles, Naver has a distribution with the largest values. Nate is the second place, and Daum has a distribution with the smallest values. We must make portal pages light to increase response speed including other technologies. We expect our results to activate competition among portal sites.

• The Korean Journal of Applied Statistics
• /
• v.30 no.4
• /
• pp.513-528
• /
• 2017

Statistics are often used to test two samples if they have been drawn from the same underlying distribution. In this paper, we introduce several well-known distribution-free tests to compare distributions and conduct an extensive Monte-Carlo simulation to specify their behaviors. We consider various circumstances of when two distributions vary in (1) location, (2) scale, (3) symmetry, (4) kurtosis, (5) tail weight. A practical guideline for two-sample goodness-of-fit test is presented based on the simulation result.