Statistics Colloquium/PhD Proposal
March 31 (Fri) 11 a.m.
Alavi Commons Room, 6625 Everett Tower

Population Stability for Scoring Models

Bilal Yurdakul
Statistical Quant Analyst, Vice President
SunTrust Bank
Greater Atlanta Area

Abstract: Population stability index (PSI) is a metric to measure how much economic variables changes over time. It is widely used in statistical model monitoring. The PSI compares distributions by looking at the percentage of observations that fall within bins. The PSI is essentially a weighted average of the difference between bin percentages. If PSI is less than 0.10, then this is traditionally interpreted as ``little or no change occurred" in the economic conditions. If PSI exceeds 0.10 but is less than 0.25, this is interpreted as an intermediate shift. If PSI exceeds 0.25, then a signifcant shift has occurred and action is needed. A weakness of the PSI is its lack of statistical properties. Under the null hypothesis of no shift, how often does PSI declare an intermediate or significant shift? A standard statistical test applicable in this case is a chi-square goodness-of-fit test which has very well known properties. We will use the chi-square test as a standard for assessing statistical properties of PSI under the null situation and various alternative configurations.

A second problem of interest is detecting for possible shifts in probability of default over time. Comparing two probabilities over time is equivalent to conducting a 2x2 chi-square test of homogeneity. Treating the bins as strata, we can construct a Cochran-Mantel-Haentzel test for detecting a shift in probability of default. The CMH is most appropriate when a uniform shift in probability occurs throught the bins. When a shift occurs in only one or two bins, the CMH may not be very powerful. We will propose a test that looks at the maximum shift over the bins, by looking at the maximum odds ratio of default.

Bio:Bilal Yurdakul is a Quantitative Statistical Analyst of SunTrust Bank in the Greater Atlanta Area, and doctoral candidate in the Department of Statistics.
All statistics students are expected to attend.


Past colloquiums


Department of Statistics
3304 Everett Tower
Western Michigan University
Kalamazoo MI 49008-5152 USA
(269) 387-1420 | (269) 387-1419 Fax