Before I elaborate on the thoughts, I should give readers the typical risk management framework for existing customers used in consumer loans (In a mature market, the contribution of new users is relatively small). Users are segmented into different risk tiers corresponding to predicted default risks. It is similar to some credit rating agencies, like Moody which gives a credit rate to governments or companies. Based on the risk tiers, different account management actions may be applied. For good users, the limit can be increased so that more outstanding loans can be contributed by them, thus, improving the portfolio structure. Also, by increasing their limit, users tend to make more disbursements which benefits the business. For bad users, the limit can be reduced or even accounts can be frozen to reduce the loss. Readers may recognize that the language used is quite vague. What do we mean by good and bad users? What do we mean by better portfolio structure?
To summarize the concerns of different people I encountered regarding managing consumer loan portfolios, I try to list out the qualitative concerns:
- The predicted risk in the risk tiers may not be accurate if we observe a longer-term performance. Therefore, even for the best users, we should be careful about positive account management actions, like limit increase.
- The longer-term performance for users with higher predicted risk will be more volatile than those with lower predicted risk as they should be more fragile and more sensitive to macroeconomy changes. Therefore, we should be even more conservative.
- For the users with low predicted risk, there exists a limit of maximum outstanding as they are more rational. It is unlikely to boost their outstanding by granting more limits. Therefore, at a certain stage, we should stop or slow down the pace of limit increment for them.
- The above practice then raises concern about concentrating the loan outstanding on relatively high predicted risk users if we stop increasing the limit for best users, and continue to increase the limit for relatively worse users. In fact, it is observed that the group that makes the most money is not the best users but the more subprime users.
The concerns can be divided into three types. First, regarding the deviations from predicted/ expected risk. Second, the trade-off between profit and volatility of risk (volatility of risk can be directly translated to the volatility of return which I will elaborate on later). For example, the not-so-good users earn the most but they are more sensitive to macro changes. Third, they may exist a maximum allocation for a certain group of users, for example, the best users who will not borrow more than they can afford.
Sometimes, it is beneficial to have a more quantitative model if data is sufficient and available. Then, we can add more qualitative aspects based on the quantitative framework instead of purely relying on the qualitative judgement which can sometimes be inconsistent and vague.
When I try to think about the concerns, it seems that the modern portfolio theory framework may be applicable. Based on the mean-variance framework, it takes into account the risk (the variance of return. Risk does not mean default risk here), expected return, and correlations between asset classes in order to calculate a set of optimal portfolio compositions which maximize the return based on a given risk or minimize the risk based on a given return.
My initial thought is that it should be able to solve the concerns listed above. First, the deviations from expected risk (return) are considered. Second, the optimal trade-off between return, and volatility can be found. Lastly, a constraint of maximum allocation to a particular asset class can be added to the framework. Unlike investment portfolios, the nature of underlying assets is different. For a consumer loan portfolio, the underlying assets are effectively fixed-income security. The calculation of return is based on the DCF formula. The return is therefore affected by default risk, interest rate risk, and tenure composition. But since fixed-income security generates stable cash flow, the valuation is relatively easy compared with equity (Not sure if easy is an accurate word here. Probably I should use standard).
Using historical data, we can calculate the inputs of the mean-variance framework and output a set of optimal portfolios. Then, experts can further select the desired portfolio composition and further adjust the allocation based on the qualitative judgement which adjusts for other considerations like macro-economy trends and interest rate risk.