Bookshop

Author

Contents

Order


 

Practical Credit Scoring: Issues and Techniques

INTRODUCTION

 

In the 50 years that credit scoring has been around, it has placed itself at the heart of lending institutions decision-making. Slowly at first, with a few companies testing the concept, then more rapidly as the concepts became more widely accepted. With wider acceptance came greater depth of information and improved technology. Scorecards, look like they do today because of the limited technology of the past; scorecards had to be integers and were additive so that operators could calculate the score and compare the result to a pass mark in their heads. By the start of the 1980’s the US and UK were developing consumer credit bureaux that could capture lenders information and summarise it to aid future lending decisions. Account information sharing kick-started credit scoring as models became more complicated, sophisticated and predictive. Automation of credit bureau links enabled applicant information to be assessed rapidly and consistently. Gradually businesses overcame their reliance on subjective decision-making as the systematic, consistent approach was shown to out perform human judgement.  

To assist the transition, the decision engines also improved. They moved from rule-based, hard coded software to highly parameterized flexible solutions that empowered the risk management teams. Strategy software developed so that decisions could be given extra dimensions. Applications could be segmented for different treatment and decisions progressed from one-dimensional accept or decline, to include line assignment, product features and risk based pricing.

Initially the credit bureau was seen as an expense to be saved and many early scorecards were designed with two stages. Application score followed by a final score. Applications failing the application score would not progress to the credit bureau thereby saving the cost of a search. The power of the bureau data combined with the ability to tailor offers to risk segments, has resulted in an about face: the credit bureau search is now often the first stage after sufficient identity information has been captured. Some organisations now reject applicants who don’t meet credit bureau criteria, thus saving processing cost, or determine the second stage questions based upon the bureau outcome.

The 1990’s saw the rapid spread of scoring as companies recovered from the Recession, realising  that objective decision-making reduced operational risk, whilst statistical tools provided the control management needed to ensure the appropriate level of risk was being taken. Regulators also recognised that scoring provided an evaluation of the risk being taken by a lender and have now determined international rules for determining the level of capital that a bank should hold relative to the potential losses. As a result, what started as a trickle in the 1960’s has become a flood today with the mortgage lenders and others joining the approach to underwriting and portfolio management what was started mainly by the mail order and store credit companies.

The greater the discrimination of risk, the lower the capital requirement and so the pressure is on to find new techniques and data to improve the models. In response, credit bureaus have deepened their data and models and the incorporation of these in lenders’ scorecards is growing. Where a single bureau reference may be taken, some lenders now find benefit in taking multiple searches from multiple bureaus. 

This book covers the practical issues of building good application and behavioural scorecards. Many assumptions are made during a development and it is imperative that both the developer and the organisation appreciate what approaches have been used and what the implications are. The growing importance of and reliance on scorecards, means that the models must be robust and practical. A scorecard that appears statistically to be highly discriminatory must deliver those benefits. Time and time again you will read that operational practicality and the strategic use of the model is more important than the new technology used to build the scorecard.  

Peter Constance

Pancredit

May 2006