Scorecards, Buckets and Points, The Anatomy of a Credit Scoring Model

There are four primary components to any credit score; the scorecards, the characteristics, the variables and the weights.

Scorecards – The scorecards are actually scoring models but cannot stand alone as a freestanding credit scoring system. All properly designed scorecards are built to evaluate the risk of a homogenous population. Bankrupt consumers is one example. “Consumers with thin credit reports” is another example. There are many more examples of scorecards but FICO and other model developers don’t generally disclose the exact definitions.

The purpose of having multiple scorecards in a model is to optimize it’s performance for all different consumer credit file types. If your credit score just had one scorecard then it would likely do well for one group of consumers and perform substandard for all others. That’s not a good credit scoring system. The better your developer is at defining a unique population, one that support it’s own scorecard, the better results from your credit score. Currently the FICO scoring system has 10 scorecards (for older versions) and 12 (for FICO 08). The following three components all reside within the scorecards.

Characteristics – A characteristic is simply a question the models asks your credit report. So, for example, “how many inquiries do you have in the past 12 months?” or “what is your revolving utilization?” or “what is the oldest account on your file?” Each scorecard has a different set of characteristics, but many of the same characteristics reside across multiple scorecards.

No model developer discloses all of their characteristics but we do know some of them and we do know that there are thousands of possible characteristics to choose from when building a model. There’s actually software designed to think up characteristics.

Variables – If the characteristic is best described as a “question” then the variable is best described as “the answer.” So, if the model asked you “how many inquiries do you have in the past 12 months” then the variable could be “none” or “one” or “15.” That’s why it’s called a variable, because the answer to the question can vary.

Each of your answers is going to place you neatly into a bucket or bin or class, they’re all the same thing so don’t get confused by the term. For example, here’s how inquiries COULD be bucketed, binned, or classed…THIS IS AN EXAMPLE.

Variable Buckets for “Number of Inquires in the Past 12 Months Characteristic”

0 inquiries

1 inquiry

2-5 inquiries

6-10 inquiries

>10 inquiries

The decision on how to break up those buckets is made by the model developer. He or she is trying to come up with the best scenario, which yields the most predictive model. This is an important step because you can’t simply choose how to break up your buckets based on common sense or anecdotal evidence. It has to be based on science. Just because you “think” 5 inquires is worse than 2 inquiries it doesn’t mean that it’s actually true. In the example above, 2, 3, 4, and 5 inquires all mean the same thing, which is why they’re all in the same bucket.

This “bucketing” process is going to apply to almost every characteristic in your scoring model. NOTE: Just because your bucket looks one way in one of the scorecards it doesn’t mean it’s going to look the same way in the others. It could easily look like this in a different scorecard…

Variable Buckets for “Number of Inquires in the Past 12 Months Characteristic”

0 inquiries

1 inquiry

2-4 inquiries

5-8 inquiries

9-12 inquiries

>12 inquiries

Weights – Weights, or point values, is where your scoring model is most visible to lenders and consumers. This is where your final score is going to start coming together. The weight is the point value given to your variable. So, if I used the above example here’s what it could look like…

Variable Buckets for “Number of Inquires in the Past 12 Months Characteristic”

0 inquiries = 50 points

1 inquiry = 45 points

2-4 inquiries = 40 points

5-8 inquiries = 20 points

9-12 inquiries = 5 points

>12 inquiries = 0 points

Just as it is with characteristics and variables, the point values will be different in different scorecards. So, using my first example for inquiry bucketing your weights could look like this…(remember, this is the same characteristic just in a different scorecard)

Variable Buckets for “Number of Inquires in the Past 12 Months Characteristic”

0 inquiries = 60 points

1 inquiry = 55 points

2-5 inquiries = 50 points

6-10 inquiries = 20 points

>10 inquiries = 0 points

This is what confuses so many “credit expert pretenders’ because they generally want to assign a fixed point value to each item on a credit report. When you look at these inquiry examples you quickly realize there is not a fixed value per inquiry. The value or points you earn is based entirely on what bucket you fall into. You don’t lose 5 points per inquiry. That’s not how scoring works.

There ended the lesson!

From Credit CRM blog by Jamison Law