Scorecard Variable

    Characteristics about the company that are statistically relevant to predict the likelihood of a company defaulting.


    A set of characteristics about the company that will predict the likelihood of the company defaulting.

Scoring Models

Not all companies in a particular country behave the same, especially when looking at failure. The population is therefore separated based on characteristics of the company. This could be size, financials or legal form. Separate models are then built on each of the populations

Creditsafe Data 

Creditsafe data is data that we own or collect ourselves as a business. For example Creditsafe collects Payment Data from members of its Trade Payment Data partners. This data is unique to Creditsafe.

Third Party Data

Creditsafe collects data from a number of sources to ensure that we are providing a truly representative view of a company. For instance, we source business information from Companies House, the Land Registry, banks and     many others.

Trade Payment Data 

Trade Payment Data is a data set that is unique to Creditsafe. We collect information on how businesses are paying their invoices from our large collection of Trade Payment Partners. This provides unique insight into the payment habits of UK businesses, such as when an invoice is beyond terms, paid late or paid on time.

Probability of Default (PD)

The underlying risk that estimates the likelihood of a company failing. This is converted into and drives changes in the 1-100 score.

Definitions of Default

    The outcomes of companies which we have deemed as failing. This is what is predicted by the scorecard.

Business Population

The business population is the collection of businesses, for which we have information on, that fall into a specific economic region, e.g. the UK.


Migration Matrixes 

A measure of how a population moves from one place to another. This is usually used to measure how the population moves scores or Bands when launching a new scorecard.

Conversion Table

    Shows how the Probability of Defaults are converted into the 1-100 scores.

Gini (coefficient)

This is a measure of how good the scorecard is at predicting bad companies. It shows at every score many bads are caught in exchange for good companies. Ideally all the bad companies will have a score less than 30 and all the good companies have a score over 30. So companies can decline all customers with a score less than 30. However this is not possible and the Gini shows how close to this separation the model is.  

Out-of-time validation

When developing a scorecard the population is taken at one moment in time historically and then applied to the entire population continuously over time today. There are chances that the development population isn’t a true representation of the population the rest of the time. The scorecards are applied to a population that wasn’t used during the development. This can be after the development period and therefore called out of time.

 Alternatively, data can change over time and this test makes sure that the scorecards are stable in today’s economy and after any data / population changes

Logistic Regression Model

    The statistical method that is used to calculate a company’s risk of failure.

Expert model

Scorecard produced using a market expert using scores to define variables classings that are known associated with a company defaulting. The scores are then added to produce 1-100.


Statistical scorecard

Scorecard produced using statistical methods to produce the risk of a company failing. This is then converted into the 1-100 score.

Retrospective Analysis

A way of clients and customers comparing how well our models predict their customer failures. The scorecard is outputs the scores of their customers to match their records of customers in the past. This is then used to check how aligned our scorecard is to our customers’ records.