Tom’s Ten Data Tips – July 2007
Credit Scoring
Credit scoring has dramatically changed the face of the underwriting business. It is a “young” discipline. Only 30 years ago, the majority of credit acceptance decisions were taken intuitively by underwriters. As statistical evidence accumulated, and early adopters of automated techniques conquered the markets, a sea change took place. Nowadays, almost all credit decisions are processed automatically, using scorecards to determine default (and sometimes profitability) odds.
The benefits of objective risk assessment by credit scoring are many:
- Objective risk evaluation – an algorithm is consistent and never discriminates
- Efficient processing – automatic scoring of decision criteria generates faster decisions
- Statistical portfolio control – scorecard characteristics give expected default odds
- Controlled experimentation – on subsets of accounts with known scores, dedicated deviations from existing practices can be tried and tested
Source: Lawrence & Solomon
1. Default Levels Are Only Relevant In Relation To The Profit A Product
Is Generating
It
is typical for many organizations to display knee-jerk reactions to jumps in
write-off levels. However, because mature credit risk management is about
balancing risks and rewards, a default percentage should never be regarded without
reference to corresponding profit levels.
In times of economic downturn, it is quite expected to see a rise in defaults, and increased write-offs. This is not necessarily a bad thing, as an economic downturn will also cause people to tap into their credit reserves more readily, thus growing the revenue stream from this same credit product (you can safely assume average credit balances have raised). Only if the balance between risk and reward is shifting unfavorably, defaults versus interest income, should policy changes be considered (see also tip# 10).
2. Delays In Policy Change Cause Oscillation
When
a cutoff score is adjusted because there are too many write-offs, it takes a
while before the actual (aggregated!) percentage of defaults will go down.
Because if this, and impatience, management can be tempted to increase the
cutoff even further, before this is required. As a result, they may find
themselves with too few accepted applications.
When this happens, a “natural” response is to lower the cutoff score. Since this will not impact default levels until much later, sometimes another adjustment is made, before the new cohort of accounts have had sufficient time to “mature” (and get into arrears).The converse holds, too. This phenomenon can result in swings in default levels, which are caused by the very policies meant to “adjust” the percentage of bad accounts.
More on complex systems in next month’s edition of Tom’s Ten Data Tips titled “system dynamics”.
3. When Using Two Score Systems, They
Must Operate Simultaneously
There
can be very good reasons to use two separate scoring systems. For instance,
when using a bureau score in conjunction with a bespoke scorecard. There is one
pitfall that must be avoided, though: the two scoring systems must run in
parallel, not sequentially.
The two scoring systems must be integrated and overlaid, to effectively guide the decision process. When the systems are used in sequence, too few applications are accepted. This has a statistical reason. Low-side applicants from either system can be rejected, resulting in a net lower acceptance rate. At the same time, the number of bads from either system remains constant, but they must now be offset by less accepted goods. Very bad practice, resulting in a poorly performing portfolio.
4. IT Management Needs To Support
Operational Credit Management
When
automated application and maintenance scoring proved to be the key to
successfully competing in the last decades, the initial focus was on providing
portfolio and management summaries. At the moment, however, productivity gains
from successful IT deployment are focused mainly at supporting operational
processing at the “shop floor” level.
What we are seeing today with our clients, is that providing world-class support of the day-to-day operations provides biggest productivity gains. And these are immensely important to stay competitive in markets that are characterized by aggressive pricing schemes. Timely information, of immaculate quality, is crucial. If underwriters have any doubts about the validity of information “in the systems”, they will double check via alternative sources, immediately losing any productivity gains that automation was supposed to have. Interestingly, efficiency gains also improve throughput times, enhancing service and value to the customer.
5. Rapid Portfolio Growth Conceals
Problems
When
the rate of account acquisition grows, you are bound to see a decline in
defaults. The reason for this is that new accounts take time to mature. Only
after new accounts have been in use for a while, are beginning to carry
significant balances, and start getting into arrears, will this new cohort
begin to generate write-offs.
There is no ironclad rule against this conundrum except for controlled growth. For sure, there should be reports that break down write-offs by vintage, that is, the year/month in which accounts were acquired. In times of growth, the overall default percentage will decline, but the write-offs per vintage are still expected to remain constant. It is good practice to include accounts in overdraft and arrears, even though they haven’t “technically” defaulted, yet (a Dynamic Delinquency report). The general rule about risk portfolio management is to get bad news out as early as possible.
6. You Never Know What Are The Right Questions To Ask, But Don’t Stop Searching
Questions
on a credit application form have the role of a psychometric instrument: you
try to get the best possible “measurement” about the customer’s situation. You
need to balance the amount of information requested with the need to make the
best possible acceptance decision. More is not always better: only the poorest
credit risks are willing to fill out extensive application forms. So a longer
form will bring in relatively more bad credit risks, whilst the good risks will
refrain from applying (with you).
At the same time, you should be constantly updating your forms, on the lookout for newer and better items in terms of discriminatory power of the application process. You do this by adding new “test” items. Questions that are successful in sifting out bad risks today, may not hold up their statistical power in the future. For example, whether someone has a land line or cell phone plays a vastly different role in determining creditworthiness now, than it did five or ten years ago.
7. Group Product Lines By Terms And Conditions
When
reporting about the inherent risk in a multi-tier portfolio, it is easy to
loose track because of the proliferation of products that have been launched
over time. To manage this information overload, it is good practice to cluster
separate products, by product groups. What is a “good” criterion to cluster?
In terms of risk management, it is usually best to lump together products that are similar with regards to terms and conditions. This may be sensible for risk reporting, despite the fact that these products have been targeted at widely different audiences. Risk profiles are predominantly governed by the intrinsic product constraints rather than overt (e.g. marketing) characteristics. The needs for risk reporting can be demonstrably different form other business units.
8. Credit Risk Management Needs To
Permeate Departmental Boundaries
Credit
risk management should never be left to the credit risk department alone. World
class organizations like Citi Group, American
Express, MBNA and others, are renown for making risk
policies an integral part of all departments. It is still true,
though, that an integrated policy may nonetheless allow for diversity in
targets for separate departments, as long as there’s rhyme and reason.
Throughout the entire customer life cycle from acquisition, to maintenance, to collections, the credit risk policies need to work in unison. If marketing keeps an eye out for targeting the right kinds of credit risks, maintenance will be more successful through up- and cross sell. Conversely, taking on poor accounts will lead to an over burdened collections department. Business targets should be aligned with the corporate bottom-line, allowing all departments the most leeway to perform to the best of their abilities.
Historically, even through recessions, the profitability of a portfolio may well stay intact. This is mainly because utilization levels will likely go up, and therefore interest revenues for revolvers and interchange fees for transactors may go up. Businesses with strong risk management practices can use this turbulence to gain market share.
10. Cutoff Scores Should Be Set On The Basis Of Profitability Analysis
Given
the central importance of cutoff scores, they should be set (and understood!)
by senior management. The cutoff score determines the ratio of good to bad
accounts that flow in to the portfolio, and therefore will drive the write-off
levels in the future. For each band of scores, this ratio also has utilization
levels, which are the most important driver of profitability.
The key in setting a new cutoff score is to balance profits and losses. When the score is raised (tighter credit extension), the data for these considerations are (at least potentially) available, and so it should be calculated exactly. When the cutoff score is lowered, there is probably little if any quantitative basis for calculating profitability, so it should be simulated. This latter case requires close tracking of newly acquired accounts.
Further reading
Some excellent books on Credit Scoring:
Managing a Consumer Lending Business.
David
Lawrence & Arlene Solomon (2002)
ISBN# 0971753709
An Introduction to Credit Scoring.
Edward
Lewis (1994)
ASIN# B0006PHNTU
Credit Card Risk Management.
Richard Nelson (1996)
ISBN# 0965386503