Credit score explanation

Abstract:

As the speed of business continues to accelerate, more and more companies are challenged to keep pace with the demands of their markets, their customers and competitors. The marketplace is shrinking, with distinctions among global, national and regional marketplaces blurring. To increase the efficiency of processing and to meet the demands of time-sensitive customers, businesses must make decisions in real time or near-real time. Larger organizations have been utilizing credit scoring to quickly and accurately assess the risk level of their prospects, applicants and existing customers. Increasingly, midsize and smaller organizations are appreciating the benefits of credit scoring as well.

The credit score is reflected in a number or letter(s) that summarizes the overall risk utilizing available information on the customer. The credit score empowers users to make quick decisions or even to automate decisions. Whereas a decision based on manual review may take hours, a decision based on a credit score will take seconds. However, not all credit score models are created equally. That's why organizations must understand and match their needs to an appropriate credit score model. To select the appropriate model, one must understand not only the similarities and differences of the credit score models available, but also understand and be comfortable with using credit scoring.

To feel confident about using risk scoring in your environment, you need to have a good conceptual and practical understanding of what a risk score is and how it is developed. A risk score is generated when information related to an entity (a business or a consumer) is fed into a risk model. The risk model examines the information and assigns the relative importance of each piece of information, aggregating the individual contributions of each piece of information into one risk score that summarizes the entity's risk level. That's it in a nutshell. Now let's explore in detail each step in the process of generating a risk score.

The Risk Model

You undoubtedly have heard the word "model" used many different ways. We have forecasting models, revenue models, analytical models, physics models and models of trains, planes and automobiles to name a few. Like a model plane, train or automobile, all models can be thought of as simplistic representations of the real world. As such, models are imperfect facsimiles, never 100 percent accurate. No one expects a model car to replicate the intricacies of its real-life counterpart, and no one should expect other models to be completely accurate either. Besides physical models, there are predictive models, which predict an event based on empirical data information observable in the real world.

A weather forecast is actually a prediction made by a system that takes empirical data, such as current temperature, humidity, wind conditions and other factors, to determine what the most likely weather conditions will be for the next several days. Weather forecast systems have certainly become more sophisticated over the years with advancements in technology. Whereas earlier weather forecast systems used a few observable factors to make predictions, current systems use hundreds of factors to make weather predictions almost an exact science. As a general rule, the more information that is available, the better the model will perform and the more accurate the prediction will be. However, no one should expect predictive models to be crystal balls. A weather forecast model would be 100 percent accurate only if it were to have perfect information of everything that affects the weather, which is impossible. How often have you dressed for a sunny day, as predicted, only to have it rain or vice versa? Certainly, there have been plenty of times when the forecast was inaccurate. Yet, we still trust in the weather forecast. Why? Well, for the most part, the weather forecast is accurate. It is valid in its prediction and has been consistently reliable over the years.

Credit bureaus offer models that predict credit risk. Similar to the weather forecast system, credit models generate a credit risk prediction based on readily available information. Credit risk models follow the general rule that more information leads to a better prediction. However, they also exhibit the weakness of all models: a small margin of error. Despite this limitation, credit risk models have been used for nearly two decades and repeatedly have proved to be both valid and reliable. Validity and reliability are two key measures that determine a predictive model's usefulness. Validity refers to whether a model does what it is made to do, and reliability refers to a model's validity over time.

Now that we understand what models are, their strengths and limitations, and how their usefulness is measured, let's explore how a model is actually developed.

Model Development

To facilitate understanding, we will provide a conceptual explanation of the process of model development, followed by a "real-life" example of each model building step.

1. Model development is driven by an objective, which answers the question, "What is the model supposed to predict?"

Let's say Martians do exist and they are of the scientific sort. One Martian scientist is interested in human beings - specifically, this Martian scientist has a burning desire to predict a human's weight!

2. Once the objective has been defined, the next step is to gather as much data as possible related to the objective. A common rule of thumb in model development is "The more data gathered, the better."

ince the Martian doesn 't know much about humans and only wants to predict weight from afar (Mars), he will need to collect all the data he can observe from a distance, including hair color, eye color, skin color, height, waist size, shoe size and so forth.

3. Next, the data is analyzed through a statistical process, usually called regression, to create a model. Common statistical packages used for regression include SAS, SPSS and even Microsoft?? Excel! The statistical process looks for any significant relationships between what the model wants to predict - called the dependent, or predicted variable - and all the possible variables that can influence the predicted variable. The variables that have a relationship with the predicted variable are called independent, or predictor variables. Once the regression process identifies all significant relationships between the predicted variable and the predictor variables, then a regression equation is generated that can be used to calculate the predicted variable when the predictor variable values are known. The model can be fine-tuned at this point to be more predictive by segmenting the population used to create the model into more homogeneous, or similar groups. Conceptually, this step is the most difficult to understand, so let's continue with the example, which will clarify this part of modeling.

The Martian now needs to look for relationships between the predicted variable, weight, and the predictor variables. Because the Martian is unable to observe the weight of people from afar, he actually will need to abduct a large number of people from Earth! This group of people who have been abducted is called a sample population. The size of the sample population has to be large enough to be representative of the entire population of people on Earth. Now the Martian can weigh all the people to capture the predicted variable values, capture other human features to get potential predictor variable values and then dump all the data, into the regression process. At this point, the regression process will look for any significant relationships between weight and all the other potential predictor variables, such as hair color, eye color, skin color, height, waist size, shoe size and so on. For simplicity, let's say that there are only significant relationships between weight and waist size and between weight and height. Then the regression process will generate an equation that best estimates a person's weight based on waist size and height. Let's say the equation is as follows: Weight (in Earth pounds) - 1.4 x waist size (in inches) + L 7 x height (in inches).