What are the impacts of macroeconomic factors on credit scoring?
In these turbulent times, many institutions have been considering how scorecards will be affected by the changes in the economic environment and how reliable their risk estimates are.
These questions have restarted a long standing debate; how best to incorporate economic factors into scorecards? Experian has developed an innovative approach that endeavours to answer this question and allow stress testing of the probability of default (PD) estimates to deliver more precise and robust risk measures.
Credit scoring techniques have been widely used for conducting an assessment of both consumers and commercial organisations' creditworthiness, with many techniques being developed using statistical methods such as logistic or linear regression. In the recent past, the focus of credit scoring in the underwriting process was primarily to create risk rankings, with little attention being given to the precision of the estimated probabilities of default values attributed to different economic conditions.
However, there are a number of applications where a precise estimation of PD in different economic conditions can make a difference, such as; loss calculation and provisioning, economic capital calculation, decision strategies based on loss or profitability, regulatory capital calculation, and so on. The Basel II Capital Accord reinforced the need not only for risk ranking measures but also the requirement for an accurate estimation of PD, explicitly stating that ‘economic conditions must be taken into consideration in the risk management process’. All of these influencing factors lead us to think about how to effectively include the effects of economic factors into our scoring models.
The first question that arises when we think about the impact of economic conditions on scoring models is how capable are the traditional credit scoring models that are already in place, and are they robust enough to endure different economic cycles. The verification that the PD prediction of a scoring model can remain precise enough over different economic conditions is key to answering the question of how capable current models are.
If the credit scoring models are producing precise estimates of PD, the average of the predicted PDs for a portfolio, in a specific period, will be very close to the actual default rate in that period. Considering the mean of the predicted individual PDs as an estimator of the default rate of the portfolio, we are able to assess the how precise the PD predictions are. Figure 1 shows time series of the mean predicted PD using an Experian’s business bureau score model in Brazil, called Credit Rating Serasa, and the actual default rate in companies in Brazil.
Figure 1: Comparison of Pd estimates and actual default rates.
Source: Experian analysis of a portfolio of Brazilian companies.
We can clearly see that the mean predicted PD doesn’t present a good fit of the actual default rate. Although graphical verification is not a formal procedure for the evaluation of credit risk robustness across different economic conditions, it certainly highlights a strong indication that differences in economic conditions can lead to bias in PD estimates produced by credit scoring models.
Usually credit scoring models are developed using samples extracted from historical data or using snapshots of a portfolio for a specific date. This implies that the model is developed capturing the risks of the applicants / clients that were current in the period the data was captured. Thus, the usual credit score or behavior score model is conditional to the economic conditions that were current in the period of the data that was used for its development. The use of the model under an economic scenario that is quite different from it can lead to biased risk estimates.
Changes in the economic scenarios can impact the probability of default estimates from credit scoring models. For example:
Effects over the parameters of the model – Changes in economic conditions can cause modifications in the risk relations that can be attributed to a predictive variable. For instance, in an economic downturn, liquidity and debt ratios may have greater importance in the estimation of PD of a company than in an economic expansion scenario. The same can occur with consumer credit, where variables indicating financial stability may be more important in stressed scenarios. Effects over the distribution of the predictive variables – Many variables can have their distribution clearly affected by the economic scenario. This is especially true with behavioral variables. For instance, in an economic downturn a greater number of registers of late payments is expected. Additionally the profile of new applicants can change as good or bad economic conditions make the need for credit or the consumption level vary over different segments of the applicants. Effects over a priori probabilities – The overall default rate increases in economic downturns and decreases if the economy is favorable. It causes variation of the a priori probabilities of default and produce bias in the estimates of PD generated by the Credit Scoring model.
In order to deal with those issues Experian developed a solution to estimate PDs of companies based in Latin America. The solution is hosted at Serasa Experian, the biggest credit bureau in Latin America, and is based on:
 |
Segmented approach – indentifying groups of companies that react in a similar way to macroeconomic factors and treat each segment differently |
|
Single correction factor – all bias effects due to changes in economic conditions were encompassed into one single correction factor that adjusts the good/bad odds or the PD produced by a scoring model to a forecasted economic scenario. Different segments with different sensitivities to economic factors have different correction factors. |
|
Econometric models that make possible to forecast the value of the correction factor that should be used for a given economic scenario. |
Besides more accurate prediction of the probability of default, the use of such an approach can have positive results, even if only applied for risk ranking and discrimination. This is due to the effects of different sensitivities to economic factors of the different segments. Figure 2 shows the mean predicted PD using the adjusted score model and the actual default rate in companies in Brazil. Clearly indicating the improvement in the estimation of PD.
Figure 2: Comparison of Pd adjusted estimates and actual default rates.
Source: Experian analysis of a portfolio of Brazilian companies.
As well as a more precise PD estimation, the new solution makes it possible to simulate forecasts of PDs in different economic scenarios, allowing Experian’s clients to perform stress testing of PD estimations. The same approach can be extended to consumer bureau solutions and custom projects for development of internal models of financial institutions.
Besides capturing the influence of a forecasted economic scenario into scores and, by consequence, into the expected loss of the portfolio, this approach is a way to increase the prospective characteristic of scores and risk management, not only “driving looking out the rear view mirror” but merging historical risk patterns with expected future economic conditions.
Fabio Wendling
Head of Analytics and Data Intelligence - Latin America
Decision Analytics
Experian
| Read more about Scoring and Analytics |
| Visit the website to read more » |
| Related links |
Contact us for further discussions about this article