I came upon this interesting blog post by Iain Brown from SAS. He is discussing the key challenges in the development, refinement, calibration and validation of internal credit scoring and risk models within the banking sector. The topic is of on-going importance for the banks, in particular in view of the ever changing global and local risk regulations.
I think that Iain is absolutely right about the most important day to day challenges:
- Data quality: The quality of the final outputs of any model is critically dependent on the data quality and in particular on the consistency of the data on which the model is applied to. Hence, the phrase “garbage in, garbage out” is one of the most frequent I have heard among credit risk modelers and risk management software specialists. Banks struggle with that, despite huge investments and regulations requiring consistent and granular data going back at least 20 years.
- Scarce default history: Most current credit risk models extensively rely on past data to determine critical parameters like PD (Probability of Default). Luckily for the banks but somehow unluckily for the statisticians working in banks, defaults do not happen too often. Even for high-risk investments like microfinance the reported default rates are in many cases lower than 5% (source: mixmarket.org). This introduces significant uncertainty in the parameter estimation the statisticians have to come up with.
- Reject bias: In the banking industry, scorecards are commonly built to help determine the level of risk associated with approving a loan application. Scorecards are built using data from the past to help infer what would happen in the future. However any applicant declined by the bank cannot be used in the model, since the bank does not track the actual outcome (i.e. we do not know if we were right or wrong when declining the application).
- Forward-looking indicators: The idea is to look at macroeconomic trends like unemployment or ratio of domestic loans to GDP in order to be able to predict better the near future. It is an excellent idea but I have seen very few good implementations in practice.
- Model choice: This topic extends on point 1. There various models and model modifications which can be applied in credit risk. There is however no best-in-class and no one-size-fits-it-all model. The model must be carefully chosen, this choice is not trivial and must respect both the underlying data quality and the economic characteristics of the portfolio.
Going beyond the day to day business and taking a longer-term perspective, the following challenges can be added to that:
1. Dependence structure within the portfolio: At the end of the day for a bank it does not matter if some of its clients go burst. This is a normal part of the business. It matters if many default at the same time. To some extend this issue builds upon the forward looking indicators approach explained above. At the same time I think that it goes beyond it. Doing an analysis on macroeconomic factors like GDP ratios, etc. helps to get a more up-do-date forecast for the expected loss from defaults. Being able to model the dependence structure in the portfolio gives insights and measurement on the unexpected loss, that is the loss that could happen in case the economic parameters turn suddenly south.
2. Interaction credit risk and market risk: Suppose there are 2 banks engaged in some huge size derivative transaction, say a huge FX swap. Bank A has fully hedged its FX exposure, bank B has not hedged it at all. In case the market turns extremely against the bet of bank B, it will probably not be able to pay its obligation and will go burst. Suppose you are risk manager of bank A. Now is that really credit risk? Or is it a market risk? For credit risk, a risk manager would classically look at the credit rating, the balance sheet, the default probabilities of bank B. Using these parameters everything is fine. For market risk a risk manager would classically look at market FX rates and volatilities. Again everything is fine because bank A is fully hedged. It is actually the combination of a sudden market movement and credit risk that matters; and this requires techniques which go beyond the classical text-book approaches.
3. Sustainability: Suppose bank A is considering to give a huge loan to the mining company C. C has today a good P&L, excellent credit rating and stable cash flow. One would naturally conclude that this company has a very low credit risk. On the other hand, suppose the company C’s technology is known to be damaging for the water resources, and in this respect the company is far behind most of its competitors in the mining sector. What would happen to C if the government or the society react on the company’s externalities by e.g. removal of business permits? On the flip side, what is the real credit risk of (some) micro finance operations which empower poor people all over the world? In the long run such operations could create a huge new market of millions of consumers for the benefit of all, including and in particular the lender. A further example: those of you who have traveled to Nigeria have certainly smelled the huge public discontent about the omnipresent corruption and arrogance of the ruling elite (you can feel it even as you go to pick up your Nigerian visa at the consulate). I am personally surprised that it took he Nigerians so long to go on the streets to protest. And yes, the poor governance and corruption is not a patent of developing countries; I think it is rather a borrowed behavior from the West. Just recall for example the Enron accounting scandals in 2001; this is a textbook example for credit risk caused by poor governance. Every time I read a post on poor governance practice in United States Proxy Exchange community site I am thinking: shouldn’t capital lenders and investors incorporate this information in their analysis, tilt their exposures and risk-adjusted yield estimates accordingly and undertake active steps to minimize downside risk?