Читать книгу Intelligent Credit Scoring - Siddiqi Naeem - Страница 8

Credit risk scoring, as with other predictive models, is a tool used to evaluate the level of credit risk associated with applicants or customers. While it does not identify “good” (no negative behavior expected) or “bad” (negative behavior expected) applications on an individual basis, it provides statistical odds, or probability, that an applicant with any given score will be “good” or “bad.” These probabilities or scores, along with other business considerations such as expected approval rates, profit, churn, and losses, are then used as a basis for decision making.

In its simplest form, a scorecard consists of a group of characteristics, statistically determined to be predictive in separating good and bad accounts. For reference, Exhibit 1.1 shows a part of a scorecard.

Exhibit 1.1 Sample Scorecard (Partial)

Scorecard characteristics may be selected from any of the sources of data available to the lender at the time of the application. Examples of such characteristics are demographics (e.g., age, time at residence, time at job, postal code), existing relationship (e.g., time at bank, number and types of products, payment performance, previous claims), credit bureau (e.g., inquiries, trades, delinquency, public records), real estate data, and so forth. The selection of such variables and creation of scorecards will be covered in later chapters in much more detail.

Each attribute (“age” is a characteristic and “23–25” is an attribute) is assigned points based on statistical analyses, taking into consideration various factors such as the predictive strength of the characteristics, correlation between characteristics, and operational factors. The total score of an applicant is the sum of the scores for each attribute present in the scorecard for that applicant.

Exhibit 1.2 is an example of the gains chart, one of the management reports produced during scorecard development.

The gains chart, which will be covered in more detail in later chapters, tells us the expected performance of the scorecard. Several things can be observed from this exhibit:

● The score bands have been arranged so that there are approximately 10 percent of accounts in each bucket. Some analysts prefer to arrange them in equal score bands.

● The marginal bad rate, shown in the column “marginal event rate,” rank orders from a minimum of 0.2 percent to a maximum of about 15.7 percent. There is some variability between the bad rate based on counts and the predicted bad rate from the model (average predicted probability) due to low counts.

● For the score range 163 to 172, for example, the expected marginal bad rate is 5.31 percent. This means 5.31 percent of the accounts that score in that range are expected to be bad.

● For all accounts above 163, the cumulative bad rate, shown in the column “cumulative event rate,” is 2.45 percent. This would be the total expected bad rate of all applicants above 163.

● If we use 163 as a cutoff for an application scorecard, the acceptance will be about 70 percent, meaning 70 percent of all applicants score above 163.

Exhibit 1.2 Gains Chart

Based on factors outlined above, as well as other decision metrics to be discussed in the chapter on scorecard implementation, a company can then decide, for example, to decline all applicants who score below 163, or to charge them higher pricing in view of the greater risk they present. “Bad” is generally defined using negative performance indicators such as bankruptcy, fraud, delinquency, write-off/charge-off, and negative net present value (NPV).

Risk score information, combined with other factors such as expected approval rate and revenue/profit potential at each risk level, can be used to develop new application strategies that will maximize revenue and minimize bad debt. Some of the strategies for high-risk applicants are:

● Declining credit/services if the risk level is too high.

● Assigning a lower starting credit limit on a credit card or line of credit.

● Asking the applicant to provide a higher down payment or deposit for mortgages or car loans.

● Charging a higher interest rate on a loan.

● Charging a higher premium on insurance policies.

● Adjusting payment terms for business customers.

● Asking the applicant to provide a deposit for water/electricity utilities services, or for landline phones.

● Offering prepaid cellular services instead of postpaid, or offering a lower monthly plan.

● Denying international calling access from telecommunications companies.

● Asking high-risk applicants for further documentation on employment, assets, or income.

● Selecting applicants for further scrutiny for potential fraudulent activity.

Conversely, high-scoring applicants may be given preferential rates and higher credit limits, and be offered upgrades to better products, such as premium credit cards, or additional products offered by the company.

Application scores can also help in setting “due diligence” policies. For example, an applicant scoring very low can be declined outright, but those in middling score ranges can be approved but with additional documentation requirements for information on real estate, income verification, or valuation of underlying security.

The previous examples specifically dealt with credit risk scoring at the application stage. Risk scoring is similarly used with existing clients on an ongoing basis. In this context, the client’s behavioral data with the company, as well as bureau data, is used to predict the probability of ongoing negative behavior. Based on similar business considerations as previously mentioned (e.g., expected risk and profitability levels), different treatments can be tailored to existing accounts, such as:

● Offering product upgrades and additional products to better customers.

● Increasing or decreasing credit limits on credit cards and lines of credit.

● Allowing some revolving credit customers to go beyond their credit limits for purchases.

● Allowing better customers to use credit cards even in delinquency, while blocking the high-risk ones immediately.

● Flagging potentially fraudulent transactions.

● Offering better pricing on loan/insurance policy renewals.

● Setting premiums for mortgage insurance.

● Deciding whether or not to reissue an expired credit card.

● Prequalifying direct marketing lists for cross-selling.

● Directing delinquent accounts to more stringent collection methods or outsourcing to a collection agency.

● Suspending or revoking phone services or credit facilities.

● Putting an account on a “watch list” for potential fraudulent activity.

In addition to being developed for use with new applicants (application scoring) or existing accounts (behavior scoring), scorecards can also be defined based on the type of data used to develop them. “Custom” scorecards are those developed using data for customers of one organization exclusively, for example, if a bank uses the performance data of its own customers to build a scorecard to predict bankruptcy. It may use internal data or data obtained from a credit bureau for this purpose, but the data is only for its own customers.

“Generic” or “pooled data” scorecards are those built using data from multiple lenders. For example, four small banks, none of which has enough data to build its own custom scorecards, decide to pool their data for auto loans. They then build a scorecard with this data and share it, or customize the scorecards based on unique characteristics of their portfolios. Scorecards built using industry bureau data, and marketed by credit bureaus, are a type of generic scorecards. The use of such generic models (and other external vendor built models) creates some unique challenges as some of the know-how and processes can remain as black boxes. We will discuss how to validate and use such models in a guest chapter authored by experienced industry figures Clark Abrahams, Bradley Bender, and Charles Maner.

Risk scoring, in addition to being a tool to evaluate levels of risk, has also been effectively applied in other operational areas, such as:

● Streamlining the decision-making process, that is, higher-risk and borderline applications being given to more experienced staff for more scrutiny, while low-risk applications are assigned to junior staff. This can be done in branches, credit adjudication centers, and collections departments.

● Reducing turnaround time for processing applications through automated decision making, thereby reducing per-unit processing cost and increasing customer satisfaction.

● Evaluating quality of portfolios intended for acquisition through bureau-based generic scores.

● Setting economic and regulatory capital allocation.

● Forecasting.

● Setting pricing for securitization of receivables portfolios.

● Comparing the quality of business from different channels/regions/ suppliers.

● Help in complying with lending regulations that call for empirically proven methods for lending, without potentially discriminatory judgment.

Credit risk scoring, therefore, provides lenders with an opportunity for consistent and objective decision making, based on empirically derived information. Combined with business knowledge, predictive modeling technologies provide risk managers with added efficiency and control over the risk management process.

Credit scoring is now also being used increasingly in the insurance sector for determining auto6 and home insurance7 premiums. A unique study conducted by the Federal Reserve Board even suggests that couples with higher credit scores tend to stay together longer.8

The future of credit scoring, and those who practice it, is bright. There are several issues, discussed later, that will determine the shape of the industry in the coming 5- to 10-year span.

The rise of alternate data sources, including social media data, will affect the industry. In reality, the change has already begun, with many lenders now starting to use such data instead of the more traditional scores.9 This issue will be discussed in more detail in several chapters. In many countries, the creation of credit bureaus is having a positive impact on the credit industry. Having a centralized repository of credit information reduces losses as lenders can now be aware of bad credit behavior elsewhere. Conversely, it makes it easier for good customers to access credit as they now have strong, reliable evidence of their satisfactory payment behavior. In addition, the access to very large data sets and increasingly powerful machines has also enabled banks to use more data, and process analytics faster. We will cover this topic in more detail in its own chapter authored by Dr. Billie Anderson.

Regulatory challenges will continue, but banks are better prepared. Basel II has overall improved the level of analytics and credit scoring in banks. It has introduced and formalized repeatable, transparent, and auditable processes in banks for developing models. It has helped create truly independent arm’s-length risk functions, and model validation team that can mount effective challenges. Basel II, as well as Basel Committee on Banking Supervision (BCBS) regulation 239,10 has also made data creation, storage, and aggregation at banks far better than before. IFRS 9 and other current regulatory initiatives such as Comprehensive Capital Analysis and Review (CCAR), Current Expected Credit Loss (CECL), and stress testing, as well as their global equivalents, will continue to expand and challenge analytics and credit scoring.

One factor that users of credit scoring will need to be cautious about is the increasing knowledge of credit scoring in the general population. In particular, in the United States, knowledge of bureau scores such as the FICO score, is getting very common. This is evidenced by the number of articles, discussions, and questions on how to improve the score (I personally get such questions via e-mail and on social media at least every week or two weeks – questions such as “How do I maximize my score in the shortest time?”; “If I cancel my card, will it decrease my score”; etc.). This factor can work in two ways. On the positive side, it may drive people to improve their payment and other credit habits to get better scores. On the negative side, this may also lead to manipulation. The usage of robust bureau data will mitigate some of the risk, while the usage of unreliable social media or demographics data may not.

The ever-present discussion on newer, better algorithms will continue. Our quest to explain data better, and differentiate useful information from noise, has been going on for decades and will likely go on for decades more. The current hot topic is machine learning. Whether it or the other more complex algorithms replaces the simpler algorithms in use in credit scoring will depend on many factors (this topic will also be dealt with in the later chapter on vendor model validation). Banks overwhelmingly select logistic regression, scorecards, and other such methods for credit scoring based on their openness, simplicity, and ease of compliance. Complex algorithms will become more popular for nonlending and nonregulatory modeling, but there will need to be a change in regulatory and model validation mind-sets before they become widely acceptable for the regulatory models.

The credit crisis of 2008 has been widely discussed and dissected by many others. Let us firstly recognize that it was a complex event and its causes many. Access to cheap money, a housing bubble in many places, teaser rates to subprime borrowers, lack of transparency around models, distorted incentives for frontline staff, unrealistic ratings for mortgage-backed securities, greed, fraud, and the use of self-declared (i.e., unconfirmed) incomes have all been cited.11 Generally, I consider it a failure of both bankers in exercising the basic rules of banking, and risk management in failing to manage risks. Some have even suggested that models and scorecards are to blame. This is not quite accurate and reflects a failure to understand the nature of models. As we will cover in this book, models are built on many underlying assumptions, and their use involves just as many caveats. Models are not perfect, nor are they 100 percent accurate for all times. All models describe historical data – hence the critical need to adjust expectations based on future economic cycles. The amount of confidence in any model or scorecard must be based on both the quality and quantity of the underlying data, and decision-making strategies adjusted accordingly. Models are very useful when used judiciously, along with policy rules and judgment, recognizing both their strengths and weaknesses. The most accurate model in the world will not help if a bank chooses not to confirm any information from credit applicants or to verify identities. As such, one needs to be very realistic when it comes to using scorecards/models, and not have an unjustified level of trust in them.

“… too many financial institutions and investors simply outsourced their risk management. Rather than undertake their own analysis, they relied on the rating agencies to do the essential work of risk analysis for them.”

– Lloyd Blankfein, CEO Goldman Sachs (Financial Times, February 8, 2009)