FACTOR AND CLUSTER ANALYSIS IN CONTEXT OF BANK'S PROPENSITY SCORE MATCHING

Propensity score matching is a statistical matching technique, where the probability of an event occurrence is estimated by the score, which is mostly calculated using logistic regression (propensity model). It is classification task with a known number of groups. The aim of this contribution is to improve propensity model by using factor and cluster analyses. Due to cluster analysis and logistic regression reason, we apply a method of the significant variables selection based on the correlation comparison between explanatory variables and the target variable. We also use factor analysis with the Varimax rotation to create new variables to reduce the data set and include these factor variables in the process of matching. After the data set reduction, we apply the k-means and TwoStep clustering methods to add new information about the structure of selected objects for propensity score matching. The cluster quality (objects clustering) is compared by the silhouette coefficient. The effect of factor and cluster analyses application is measured by the total success rate of the created propensity model.