Modeling Mortgage Customer Retention with Binary Logistic Regression
A bank, looking to prevent the loss of mortgage customers at the end of the ‘lock-in’ period, asked for an analysis of customers’ personal circumstances to determine if any variables significantly influence the customer’s decision to stay or leave the bank. An analysis was performed using binary logistic regression to model the bank data and model selection was performed using forward stepwise selection with assessment criteria of Akaike Information Criterion, Bayesian Information Criterion, McFadden R-squared, and classification tables. The analysis resulted in a model that shows a significant relationship between a customer’s account status and their choice in being a branch user, a direct/telephone user and having an agent.