Fraud Database

--by Kennis Wong

It's true that intent is difficult to prove. It's also true that financial situations change. That's why financial institutions have not, yet, successfully fought off first-party fraud. However, there are some tell-tale signs of intent when you look at the consumer's behavior as a whole, particularly across all his/her financial relationships.

For example, in a classic bust out case, you would see that the consumer, with pristine credit history, applies for more and more credit cards while maintaining a relatively low balance and utilization across all issuers. If you graph the number of credit cards and number of credit applications over time, you would see two hockey-stick lines. When the accounts go bad, they do so at almost the same time. This pattern is not always apparent at the time of origination, that's why it's important to monitor frequently for account review and fraud database alerts.

On the other hand, consumers with financial difficulties have different patterns. They might have more credit lines over time, but you would see that some credit lines may go delinquent while others don't. You might also see that consumers cure some lines after delinquencies…you can see their struggle of trying to pay.

Of course the intent "pattern" is not always clear. When dealing with fraudsters in fraud account management, even with the help of the fraud database, fraud trends and fraud alert, change their behaviors and use new techniques.

--by Kennis Wong

In Part 1 of Generic fraud score, we emphasized the importance of a risk-based approach when it comes to fraud detection. Here are some further questions you may want to consider.

What is the performance window?

When a model is built, it has a defined performance window. That means the score is predicting a certain outcome within that time period. For example, a traditional risk score may be predicting accounts that are decreasing in twenty-four months. That score may not perform well if your population typically worsens in two months. This question is particularly important when it relates to scoring your population. For example, if a bust-out score has a performance window of three months, and you score your accounts at the time of acquisition, it would only catch accounts that are busting-out within the next three months. As a result, you should score your accounts during periodic account reviews in addition to the time of acquisition to ensure you catch all bust-outs.  Therefore, bust out fraud is an important indicator. 

Which accounts should I score?

While it’s typical for creditors to use a fraud score on every applicant at the time of acquisition, they may not score all their accounts during review. For example, they may exclude inactive accounts or older accounts assuming those with a long history means less likelihood of fraud. This mistake may be expensive. For instance, the typical bust-out behavior is for fraudsters to apply for cards way before they intend to bust out. This may be forty-eight months or more. So when you think they are good and profitable customers, they can strike and leave you with seriously injury. Make sure that your fraud database is updated and accurate.  As a result, the recommended approach is to score your entire portfolio during account review. 

How often do I validate the score?

The answer is very often -- this may be monthly or quarterly. You want to understand whether the score is working for you – do your actual results match the volume and risk projections? Shifts of your score distribution will almost certainly occur over time. To meet your objectives over the long run, continue to monitor and adjust cutoffs.  Keep your fraud database updated at all times.

--- by Kennis Wong

In this blog entry, we have repeatedly emphasized the importance of a risk-based approach when it comes to fraud detection. Scoring and analytics are essentially the heart of this approach.

However, unlike the rule-based approach, where users can easily understand the results, (i.e. was the S.S.N. reported deceased? Yes/No; Is the application address the same as the best address on the credit bureau? Yes/No), scores are generated in a black box where the reason for the eventual score is not always apparent even in a fraud database.

Hence more homework needs to be done when selecting and using a generic fraud score to make sure they satisfy your needs. Here are some basic questions you may want to ask yourself:

What do I want the score to predict?
This may seem like a very basic question, but it does warrant your consideration. Are you trying to detect these areas in your fraud database? First-party fraud, third-party fraud, bust out fraud, first payment default, never pay, or a combination of these? These questions are particularly important when you are validating a fraud model. For example, if you only have third-party fraud tagged in your test file, a bust out fraud model would not perform well. It would just be a waste of your time.

What data was used for model development?
Other important questions you may want to ask yourself include:  Was the score based on sub-prime credit card data, auto loan data, retail card data or another fraud database? It’s not a definite deal breaker if it was built with credit card data, but, if you have a retail card portfolio, it may still perform well for you. If the scores are too far off, though, you may not have good result. Moreover, you also want to understand the number of different portfolios used for model development. For example, if only one creditor’s data is used, then it may not have the general applicability to other portfolios.