Printer Friendly

Using credit screening to manage credit risk.

Until recently, there never seemed to be a credit scoring model for business credit decisions - at least one that was simple enough to understand and cost-effective enough to develop. Credit managers have considered using scoring primarily for large numbers of decisions, such as consumer credit-granting decisions. In addition, the models were too generalized to be applicable to an industry or line of business or to incorporate specific company information and approaches. They also required a great deal of data, which could be expensive to obtain and maintain. So many credit managers probably believed that credit scoring was not applicable to their company's business. All of these may have been good reasons in the past, but they no longer hold.

Who Used Scoring Models?

The heavily statistical model represents only one form of scoring model - the oldest form. This traditional credit scoring technique has been around for decades. It is heavily used as a decision-making support tool by lenders or credit granters to individual consumers and small businesses.

Credit card, bank loan and mortgage lenders have long recognized the value of scoring and have put it to good use. Consumer credit scoring models offer significant savings in the time required to make credit decisions, and with many to make, this is invaluable. Also, by identifying the key variables, extraneous data can be eliminated, thereby making record-keeping easier. For banks or mortgage lenders, this means faster loan decisions and the ability to bid on loans of smaller amounts (because the cost of making the decision is so small).

However, credit managers are often faced with a much smaller number of business customers and credit decisions that represent significant dollar amounts of credit. They probably believe that this type of model did not apply to them, so credit scoring was not considered as an alternative. In fact, they were right. Heavy statistical models were not easily adaptable to business applications.

Why Limit Credit Scoring to Consumer Credit?

Credit scoring does not have to be limited. Many firms used a form of scoring probably without realizing it. For example, a credit department may establish a standard group of questions that each credit customer has to answer. Then a numeric value is assigned for the answers, and an aggregate point score is calculated. This rudimentary system is a type of credit score. Often it has been used to determine a customer's credit limit or whether to offer a customer any credit at all. This decision matrix might be enhanced by setting it up on a computer spreadsheet program. The credit manager might even have considered including some trade credit or bank reference data. However, the time required to gather the data and enter it into the spreadsheet program usually limited the amount of external data used. As John Sargent, president of Credit & Management Systems, Inc. suggests, "Scoring by itself is not helpful; you need to do something with the score." But, as Sargent and others point out, many credit managers use the credit scoring system in their heads.

Today's Business Environment Is Changing

Other influences in today's corporate environment have changed the way credit managers look at scoring. For instance, the corporate credit function has not been immune from the effects of downsizing. Today's managers must do more with a smaller staff, so it is only natural to look for help in other ways. Credit scoring can offer assistance without requiring additional people.

In addition, with so many decisions to make, it is important to be able to establish and administer a consistent, objective credit function. With companies striving for increased globalization, the need for consistency across borders is important. Adapting a credit scoring approach to each country can provide consistent, fair credit decisions.

Finally, credit scoring can help credit managers do more than make reactive credit decisions based on historic data and customer performance. It can help identify potential high-volume customers. It can also lead credit managers to increase credit limits to low-risk customers, thereby helping the marketing function bring in more business. Team play of this sort is the expected norm of today's environment. Credit managers should be ready to take advantage of any tool that helps them contribute to increasing corporate sales or business.
Figure 1

Delinquency Rate by Score Category

Class       Percentile       Scores       Rate of Severe Delinquency

5              1-10         101-376                   69%
4             11-30         377-449                   21%
3             31-70         450-492                   9%
2             71-90         493-537                   4%
1             91-100        538-660                   3%

Source: Dun & Bradstreet.

What Are the Major Types of Credit Scoring Tools?

There are several types of credit scoring:

Predictive scoring. These models are used to predict whether a customer will pay its bills on time, late or at all. They usually are used before credit is offered. They may also be used when a customer applies for an expansion to an existing credit limit. Predictive models have been derived from the consumer type and are heavily dependent on statistical input. Only in the last 10 years or so have these models been used by corporate credit managers to make credit decisions for business customers. These models can also be used to review whether a customer can be granted additional credit and still be expected to be a prompt payer.

Risk scoring. These models are used to predict whether existing customers are likely to pay or to extend their possible delinquency in paying. While similar to predictive models, risk scoring models evaluate existing customers and their collection potential. This allows credit managers to value their portfolio of credit customers. By determining which parts of the portfolio need attention first, a credit manager can make better use of his or her time and improve collection performance.

Default scoring. These models predict whether a customer is a candidate for bankruptcy. Obviously, early warning signals are important and are valuable items for any credit manager who may be uncertain about specific customers or specific types of customers.

Corporate credit managers can use any, all, or combinations of these models. Predictive scoring models are valuable aids for prescreening credit customers or targeting market segments. Likewise, credit managers look for exceptional reporting techniques that can help them use credit management resources most effectively. Risk scoring models do this by identifying key customers to follow without losing valuable follow-up time. Default models may be used for special cases or when economic circumstances suggest it.

Credit managers also are expanding their horizons for credit scoring to include foreign locations. As David T. Kresge, senior vice president and chief economist at D&B says, "Global scoring is happening. We're developing separate scores in each country, reflecting individual predictive factors and local information. Scores are then translated for comparison purposes. D&B reports scores for more than a dozen countries and provides a translator that equates the scores across countries."

What Do Business Models or Systems Look Like?

Many business models resemble their consumer model counterparts. For example, most have some standard required data from all applicants, just like the statistical consumer models. However, unlike consumer models, business models have additional parameters that differ by industry, location or type of business organization.

Business models also differ from consumer models by how information. used for the scoring process is gathered. Consumer models often use a single source or similar types of third-party sources for most, if not all, of the scoring data. Business models use third- party data, such as credit bureau sources. They enhance the models by integrating in-house data with external data. For example, one credit manager interviewed described his credit granting system, which is currently being developed, as a "weighted average scoring approach." His system for granting credit to retailing customers uses automatic data feeds from several trade credit bureaus, an in-house database, and a third-party PC-based system to consolidate data and provide necessary reports.

Delinquency Models

"Credit managers need a tool to go after potential delinquent accounts. Service providers, like D&B, have developed tools to help these firms," notes Ron Klausner, senior vice president for D&B Receivable Management Services.

Delinquency models usually show a distribution of results for a firm's customers. Figure 1 shows an example of the distribution of scores for a typical corporate credit portfolio using a D&B model. The result for the lowest class (5) indicates that 69 percent of the accounts with scores of 376 and lower, which represent the lowest 10 percent of companies, are likely to go more than 90 days past due over the next year.

A further example of how one firm uses delinquency models is useful. An automobile manufacturer used risk credit scoring to monitor customer payment performance. One customer, who was running into financial problems, had maintained payments for 20 months but had decided to skip a payment, figuring that the manufacturer would not notice a missed payment after so many on-time payments. The customer was very surprised to get a call after his payment was five days' overdue. The manufacturer had earmarked this customer as a problem delinquency (if he went delinquent) and started prompt collection follow-up, with the result that the customer re-evaluated his intentions of delaying payment.

Another example of a delinquency model specific to an industry is the score jointly developed by Dun & Bradstreet and Lightbridge, Inc. for the global telecommunications industry. The model, which will be available in the first quarter of 1998, provides a risk assessment of new or current customers in the United States. The statistical model provides estimates of the likelihood a customer (company) will be severely delinquent (i.e., more than 90 days past due) in paying its bills within a one-year period. Both firms used their experience and data sources to develop the model. Scores range from 101 to 660, with higher scores reflecting less risk.

Fair, Isaac's revenue score is a type of model that credit managers can use to identify customers who are more likely to provide increased revenue. It is intended for use in situations such as consumer credit cards.

Bankruptcy Models

Bankruptcy predictors have been popular credit tools for many years. One of the earliest forms of a bankruptcy model was the Altman Z-score. Today, bankruptcy scoring models have been refined to provide ongoing customer analysis.

A good example of recent developments is a bankruptcy scoring model for consumer credit customers, HorizonTM. This model has been jointly developed by Fair, Isaac and Trans Union. The model draws differences between consumers who are likely to go bankrupt and those that are likely to be high-profit customers. The model is intended to provide bankruptcy predictions with sufficient lead time for credit grantors to take appropriate actions to reduce their potential losses.

Something New Is Coming

A new type of predictive scoring model is emerging. A prime example is Dun & Bradstreet's Recovery Score. Traditionally, all customers were treated equally and received the same type of collection follow-up at regularly prescribed periods. Models like the recovery score provide estimates of each customer's likelihood to pay on time or to become seriously delinquent if past due. This information allows credit managers to focus on the high-risk (for continued delinquency) customers first. This applies the consumer approach to companies.

Third parties, like Dun & Bradstreet, who offer credit/collection outsourcing services, use D&B's tools to handle collection. Their results, which now can incorporate the recovery score to rank the credit portfolio, have shown that this approach works and can pay big dividends. By measuring performance in either the amounts collected over an average historical level or against an in-house control group, D&B reports that using a recovery score concept to prioritize customers is very effective. Further, if customer satisfaction (actually dissatisfaction) is measured in the form of complaints to the company that has outsourced its collections, the technique is successful. The customers who are less likely to pay (based on past track records) are most likely to complain - but as one D&B customer put it, they may not have liked it but they paid.

What Models Are Offered?

A sample list of firms that provide credit scoring models and services is shown in Figure 2. The names should be familiar to most credit managers. Most offer finished products (scoring models), consulting and advisory services and access to their data bases.

Some examples of the types of models offered by the major service providers include:

* Dun & Bradstreet. Predictors of Payment Delinquency - Commercial Credit Scoring, Industry-Specific Credit Scoring, Small Business Credit Scoring, Custom Credit Scoring and Business Failure Predictors - Financial Stress Scoring.

* Fair, Isaac Credit Bureau Scores - Risk scores, Bankruptcy scores, Revenue scores, attrition scores and collection scores.

* Credit & Management Systems, Inc. The Corporate Credit Manager[TM] - Risk/scoring models, Credit Limit Models.

Scaling the Scores

Credit scores are scaled in a similar range from one type of model to another or from one service provider's model to another. The scale for consumer models is 101-800, and this scheme has generally been used for business credit models by most of the service providers. In some cases, the range has been narrowed (usually with a higher starting value). Higher scores reflect lower risk. Also, as shown in Figure 1, many models show percentile rankings.

Global scores, range from one country to another. In these cases, the range of scores must be translated to a common denominator before they can be used. Because global scoring is so new, this translation is being done on a case-by-case basis.

How To Access Models

Credit managers have several choices for accessing credit scoring models. The method selected will depend on the overall nature of the credit manager's system and the number of decisions to be made. Also, if a credit manager develops an inhouse large-scale system, direct computer links with more than one service provider will be required.

For D&B models, scores can be obtained in an on-line fashion, or a firm can submit its entire portfolio for scoring on a batch basis. With other services, such as Fair, Isaac, the model resides on the client's computer and is operated internally. With smaller service providers, such as Credit Management Systems, the model is developed and installed on the firm's computer.

Advantages to Credit Managers

The advantages of using credit scoring include:

Consistency in applying credit reviews. If the models are developed properly with extensive input from all levels of experience and ability, the end product will represent the company's credit approach. It will also be administered objectively and fairly across all customers or credit applicants regardless of who is doing the review or how it is being conducted. "Credit scoring gives you the opportunity to clone the highest level of review," says Roger Mumper, vice president of sales with Credit & Management Systems, Inc.

Better organization to credit information. In addition to consistency, improved organization of fundamental credit information is a natural offshoot of this process. It is clear what pieces of information are required. This should make it easier to maintain the credit data base.

More efficient use of third-party data resources. Accessing third-party data bases is an important resource for any credit management function. Scoring takes it further by integrating this information into a decision-making system. Models also ensure that the information is actually being used, not just being filed.

Elimination of overly subjective approaches. As credit management becomes more developed, the need to eliminate subjective influences in-creases. Scoring models help greatly with this. For consumer credit, removing possible bias is important, especially in assuring that the credit granting firm is in legal compliance.

Better understanding of the process. Building a model, even if one's role is in supporting an outside expert, is an educational experience. For example, learning about the firm's actual credit process or about the differences in the process can be very valuable.

Improved performance. Incorporating credit scoring into the overall credit process should allow credit managers to practice more effective management techniques. For instance, they can concentrate their collection efforts on those customers who are likely to respond with payments if contacted earlier than the norm. This translates into better credit performance and better use of a credit manager's time.


Credit scoring also has some disadvantages, including:

Costs of development. Nothing this good comes cheaply. Although some corners can be cut, developing scoring models and incorporating them into an overall credit system can be a big project. That means that there will be additional costs, such as hardware, software, third-party data, consultants and more.

Over-reliance on models. Inexperienced users may see the model as faultless and may not challenge the results. This can lead to problems. Obviously, the answer is not to make the model the whole system, but just a part of the system.

Lack of quality or timely data. The model will only be as good as the data it uses. If the required data cannot be obtained, the model will not work as expected. Also, with heavily statistical models, there is the danger of reject inference. This is created when the data used to develop the model did not include full representation of acceptable and unacceptable customers. This can lead to bias in the model. Reject inference can be avoided by using public data as well as in-house data in developing the model. Most third-party service providers are aware of this problem and have built in safeguards to test for it.

Misunderstanding of what scores mean. Training and education about the model and its results are important aspects of the model-building process. If not done properly, problems will certainly occur. This can be more serious in certain cases, such as with consumer or small business credit decisions. The potential misuse of credit scoring models has drawn government notice as well. The Office of the Comptroller of the Currency (OCC) has issued advisory guidelines to commercial banks using credit scores as part of their lending decision process. The sidebar outlines some of the concerns the OCC has noted and lists some of its recommended tips. While the concern in this case is for consumer credit protection and fairness, these are appropriate considerations for commercial credit as well.

Trends and Developments

As use of credit scoring increases, there are several developments that will change the way this tool is used in business credit. Credit managers will be incorporating scoring as a general tool into their credit granting process and/or overall credit management systems. This will mean general acceptance of the approach and increased expectations from the tool. Scores will be used to establish a credit manager's priorities, such as indicating which customers need early attention or which ones offer the most potential for loss or gain. Credit managers will break the time-related model for collection follow-up by using products such as D&B's recovery score when it becomes available. This will make credit managers more effective users of their time. Further, credit scoring will be a common link in many new services, such as outsourcing. Finally, credit managers will extend credit scoring internationally, adapting the tool to country specifics and calibrating the scores across countries.

Figure 2

Examples of Providers of Credit Scoring Services and Models

Credit & Management Systems, Inc. Dun & Bradstreet Equifax Fair Isaac and Company Inc. SR Research, Experian (formerly TRW) Trans Union

HOW TO DEVELOP a Credit Scoring Model

Identify key factors you use or want to use to make credit decisions.

Identify sources of data for the preceding factors.

Set up a database of internal data and outside data to collect necessary information.

Link outside and inside databases.

Construct the model (in a preliminary form) and test with minimal data to be sure that it operates satisfactorily. (Note: This is still not an actual test of the model.)

Test the model on a sample of customers, including customers that you would accept and reject.

Evaluate the results and adapt the model.

Repeat the testing and evaluation cycle until you are satisfied.

Implement (real-time).

Plan for regular reviews of the model and allow for problem reporting or exception reports that can be filed by users of the model.



According to the American Bankers Association, the OCC is concerned about possible misuse of credit scoring models that can create discriminatory practices. The OCC, through its bank examinations, has found that some banks have had the following problems:

Inadequately trained staff to track how the model works

Insufficient information system at the bank to adequately support the model development, use and/or maintenance

Inappropriate use of scoring models, (e.g., applying the model to a different geographic market segment that was not included in the model's development).

Inclusion of data items that can be considered discriminatory under federal legislative standards. This may create something called unintentional disparate treatment, which is the adverse effect (i.e., credit rejection or restriction) on a protected group (e.g., minority groups) that is not justifiable as a business necessity.

Stephen Cross, the OCC's top compliance official, in an interview in the ABA Banking Journal in August 1997, offered the following suggestions to keep in mind when considering using scoring models, especially for consumer credit:

Revalidate [the model] often.

Don't tinker. . .without a complete statistical reworking.

Don't use preferential score cards for protected groups.

Avoid the disparate treatment trap.

Kenneth L. Parkinson and Joyce R. Ochs are the senior editors of Business Credit.
COPYRIGHT 1998 National Association of Credit Management
No portion of this article can be reproduced without the express written permission from the copyright holder.
Copyright 1998 Gale, Cengage Learning. All rights reserved.

Article Details
Printer friendly Cite/link Email Feedback
Author:Parkinson, Kenneth L.; Ochs, Joyce R.
Publication:Business Credit
Date:Mar 1, 1998
Previous Article:Put story power in your presentations.
Next Article:The Payment Undertaking based alternative payment security.

Terms of use | Privacy policy | Copyright © 2019 Farlex, Inc. | Feedback | For webmasters