Hi guys,
Most publications on applying machine learning to credit scoring focus on finding the probability of default. However, eventually, it seems that identifying profitable customers is what really matters.
I wonder if there are any reasons why we should apply machine learning to predict the probability of default instead of predicting profitability / predicting the probability of a customer being profitable.
Thanks!
Hi,
IMHO its done this way because of the following reasons:
1) its done from the point of view of the creditors, thus the most important question is whether they would get payments as due (e.g. as promised). The question of profitability, although important for the shareholders, does not exactly mean the same as being able to pay what is due on the borrowed funds.
2) Profitability is not as easy to define in an unambiguous way as defaulting on debt.
I agree that identifying profitable customers is what matters, but when analyzing a standardized product (or one fit fixed terms/payments/income) the degree of profitability would matter only as long as we are sure the payments will be made as promised. E.g. identifying profitability of the customers would be another important problem when we can target them separately and offer individual products/solutions (for which profitability matters).
Best regards,
Stan
I think both are correlated. High credit score -> Low Default and NPAs -> Higher Profitability. There is empirical evidence that suggest negative association between NPAs and Banks Profitability. The huge issues banks face today is default and every bank would like to minimize it. Higher default rates also increases the cost of the banks in terms of collection efforts and opportunity cost leading to lower profitability. Moreover, profitability calculations are infected with measurement errors, hence endogenity issues.
You can remember than ML is a tool to build a model from your data. Then you must decide what you want.
Hi, for me there is space for other than defualt risk factors while preparing credit rate. But it should be rather liquidity because profitability, that could be 'build' on paper (accounting) revenues, will not pay off the loan. We need to remember that also from bank's point of view, CASH IS KING!
I think there is enough room for building a model that is appropriate to the context. Profitability translates to a clients credit worthiness and does get captured in the Risk Score.
- Badges
- Science topic