As rates drop, lenders are likely to see a flurry of requests to switch mortgages. Mortgage redemptions could be expensive from many aspects. For starters, a mortgage repayment can leave a lender with out-of-pocket charges, such as broker commissions and the opportunity cost relating to the loss in the interest income. Some estimates suggest that early redemptions costs could exceed 5% - 7% of projected portfolio income, increasing the income volatility and exacerbating an already fragile liquidity situation.
BBA’s Mortgage Retainer undertakes the customer lifetime value analysis to minimize the losses attributed due to prepayments (Y), by fine-tuning the pricing for the renewal pool and also by segmenting mortgages into different prepayment cohorts:
In order to address the problem of predicting a dependent variable which is binary, such as prepaying a mortgage, three principal methods can be applied:
BBA’s Mortgage Retainer uses logistic regression approach and applies discriminant analysis to rank-order the obligors based on the prepayment probability. Using Single Factor Analysis (SFA), we test the predictive power of independent variables Xi, Xj....Xn, on a standalone basis. For the residential mortgage prepayments, we identified a range of qualitative and quantitative factors, some of which are as follows:
The predictive power of a factor can be represented numerically by the Power Statistic. Only factors with a reasonable predictive power were considered for the next stage of analysis.
A perfect predictor of prepayment would rank all ‘prepaying’ obligors below the ‘non-prepaying’, hence if 20% of a sample were prepaying, we would expect these 20% to receive lower scores than the remaining 80%.
Once the list of the most attractive individual factors is finalized, the next step is to assemble the most powerful combination of factors that are predictive of prepayment. In principle, Multifactor Analysis (MFA) analysis estimates Y (Prepayment) from a transformation of a linear combination of independent variables.
Normally there will be several models that are equally good from a statistical standpoint, and thus expert review is important to ensure that no factors are overweighted, or that where two somewhat correlated factors give similar overall performance as part of the final model, which factor should be chosen.
In the final stage, we benchmark an acceptable prepayment threshold and calibrate scores from the Multifactor Analysis into a prepayment probability over a 1-year horizon. By calibrating scores of each obligor with prepayment probability, MR is able to generate portfolio distribution for the prepayment probabilities of obligors.
MR can be customized and implemented in 25 – 30 FTE days. Upon completion of the engagement, the clients can benefit by identifying upfront high risk mortgage pre-payers and apply tactics to reduce prepayments.
Author - Sohail Saad