The New Lending System - Putting common sense into compliance and risk assessment

SAS has been conducting research on analytic frameworks that combine expert judgment with best science. We have observed that effective integration of comprehensive views, especially customer-centric ones, with financial performance and the evolving economic and market realities, offers significant competitive advantages.

SAS has actually pioneered a new lending system that, compared with today’s typical loan underwriting systems, is:

1) simpler, yet more accurate,
2) thorough, yet concise,
3) becomes more, not less, predictive over time, and
4) is statistical in nature, yet is transparent and passes various tests of reason by simple inspection.

Moreover, these advances in loan underwriting approaches have the implications for the entire lending value chain.

CCAF (pronounced See-Caf) is short for Comprehensive Credit Assessment Framework. It is a new lending system that affords many benefits, in addition to those stated above, when compared with more traditional loan underwriting systems. Additional benefits include:

ease of understanding
stresses loan affordability
adaptibility and ease of updating
ability to be effectively monitored
is based on proven lending principles
provides adequate controls to limit risk
affords early warning in advance of performance declines

For those who want to learn more, just click on the links in this post to download a copy of the CCAF white paper or to obtain a copy of our book that describes the new lending system. Please share your comments on either, or on this or past blog content. I want to hear your opinions!

I call fact on that! I am in the process of preparing a keynote address entitled Credit Access And Risk Management: A Gap To Close, for an International Banking Conference sponsored by Asobancaria, Colombia's Banking and Financial Entities Association. The conference will take place in late November in Cartagena, Colombia.

During the course of my research on the high cost of being poor in Latin and South America, I learned that it is estimated that perhaps a billion dollars is maintained under mattresses and in cookie jars where it earns no interest and provides no indication of the thriftiness of the lower income tiers of society. Worse, as many as 85 percent of the population is out of the formal financial banking system, and as a result their in-person cash payment and deposit transactions require long trips, risk of robbery, and long waiting lines, in addition to carrying a high degree of cost (as much as five to ten percent of the actual payment amount for check-cashing, and as much as nineteen percent for US dollar international wire transfers to non-bank customers who receive income supplements from relatives and friends). Bank loans can carry as much as a 17 percent annual interest rate (35-40 percent for credit cards) and the common non-bank alternative sources for loans can charge as much as from 150 to 400 percent interest annually.

The cover story of the October 2009 RMA Journal explores the principle of risk homeostasis -- the notion that excessive faith in mechanisms aimed at risk-prevention actually encourages riskier behavior. Rick Nason, author of the article, asserts that:

"Risks in the financial services industry are constantly changing and evolving. Risk systems must change and evolve as well. Faithful adherence to what has worked in the past is a blueprint for extinction in the future."

When I was on Capitol Hill last year testifying as an expert witness before the House Financial Services Subcommittee on Investigations and Oversight, I asserted that we needed to move away from the "secret sauce" models of the past that relied on "substitutes for common sense." I came away with the impression that most observers felt that the FICO score and the heavily relied upon credit bureau data for consumer credit qualification have served the country well, and that there was no real need for moving to a new system. That was in July of 2008, two months prior to the financial system near meltdown in September. These days I am finding much greater interest in a new lending system (termed CCAF) and far greater acceptance of the assertion that the FICO score is an incomplete risk measure -- one that does not nearly take into account sufficient breadth of information to accurately reflect a consumer's creditworthiness.

In our latest book on the new lending system, my co-author Dr. Mingyuan (Sunny) Zhang and I began each chapter with an apropos quote. For our chapter on the borrower and loan affordability, we chose the following quote: "What the people cannot understand, they must accept on faith." We have sought to make the case that a new lending system is needed that will bring lending models closer to reality through the bonds of common sense and a sound and proven structural basis -- one that will afford consistent and comprehensive model visibility to all market participants, from borrowers to lenders to investors. Be assured that blind faith is not a requirement for the newly proposed lending system! We believe that "more disclosure is better," and that the vast majority of borrowers and investors would agree that we need to put the era of blind faith in FICO credit scores and reporting agency ratings of loan-backed securities behind us.

Adoption of CCAF (pronounced See-Caf) would significantly strengthen our risk management systems dealing with loan underwriting, loan portfolio management, loan collections and recovery, loan securitization, and investor reporting for loan-backed investments. While this would undoubtedly foster much greater faith in risk management systems, I do not believe it would encourage riskier behavior. Much to the contrary, it would effectively curb, possibly eliminate, ill-advised product choices and disproportionately large loan magnitudes based on such categories of factors as borrower income and capacity, borrower liquid capital reserves, historical collateral price levels, down payment percentage -- all of which act in concert with one another. As a result, regardless of whether the actors are well-intentioned, the types of behavior that led to the unaffordable combination depicted in the table below would no longer pose a threat, because they would result in a declined loan application!













Credit Risk Assessment: The New Lending System for Borrowers, Lenders & Investors, Clark Abrahams & Mingyuan Zhang, copyright 2009, SAS Institute, Inc. Reprinted with permission of John Wiley & Sons, Inc.

Yesterday, the conference coordinator for an external event where I soon will be speaking asked me if the new lending system I have been advocating would surpass the FICO Credit Score. In our latest book, published in April 2009 by John Wiley & Sons, Sunny Zhang and I make the case that we need a new singular measure of creditworthiness. In our view, the FICO Score is an incomplete, inaccurate and out of context risk measure that focuses only on how consumers have paid their credit obligations in the past without regard for prevailing, and anticipated, circumstances, their income, savings, and a host of other factors. More specifically:

FICO is a one-size fits all technology that is also a black box, i.e. consumers are not told what characteristics are included in their score and how many points they were assigned for each of those factors. Consumers do not even know how specifically to improve their FICO Credit Score! While it is true that credit scoring is more consistent that a purely judgmental system of credit granting, it is rooted in models that attempt to identify statistical correlations between “substitute factors” and poor loan performance. Many of those substitute factors have no direct logical connection to loan default or are viewed out of context, in stark contrast with such factors as capital and capacity. Consequently, when we experience significant change, such as the current economic downturn, FICO scores become irrelevant (i.e. what used to be predictive no longer is).

CCAF (pronounced See-Caf) is rooted in common sense and the basic guiding principles of credit granting, namely the 5 C’s of Credit – Character, Capacity, Capital, Collateral, and Conditions. CCAF first classifies borrowers holistically before assessing their risk. Hence, their CCAF score is in the proper context, and it considers all relevant factors. Furthermore, an importantly, CCAF is transparent in that borrowers can decipher how they rated on each of the factors considered during the evaluation of their loan application, and how they can boost their creditworthiness.

I was so thankful that the question was raised, comparing CCAF to FICO. Yes, the FICO score falls far short of meeting today’s credit market demands for greater relevancy, accuracy, and transparency throughout the lending value chain, i.e. borrowers-to-lenders-to-securitizers-to-investors! Replacing FICO with CCAF would enable borrowers to better balance product options and loan affordability, lenders to more effectively balance loan volume and quality, and investors to better balance risk and return. CCAF adoption will help prevent future financial crises and can also help deal with the fallout, namely loss mitigation and foreclosure avoidance.

In our research for our book, Sunny and I concluded that a couple of root causes of the current financial crisis were inadequate risk assessment and lack of transparency. The secret sauce, narrow, and historical-behavior based FICO Score is used pervasively throughout our credit system, and it was most certainly one of our critical risk measurement antennae that was fully extended when the financial time bombs and loan default missles hit our financial system. It appears that at least one other of our critical risk measurement antennae was, at the same time, severely retracted, namely common sense! CCAF adoption will provide a far more powerful and much needed risk antennae for borrowers, lenders, investors, and regulators and it will automatically go a long way to fully extend the common sense antennae!

Generation after generation has believed that homeownership is a core requirement to accumulating wealth, is integral to the fabric of the American way of life, and perhaps even our saving grace. In Thomas Sugruea’s August 14th piece for the Wall Street Journal entitled “The New American Dream: Renting,” we learn that “for millions of Americans at risk of foreclosure, the home has become something else altogether: the source of panic and despair.”

I recall the days I worked at Fair Isaac and Co., where I developed scorecards for installment and revolving credit products. Back then, and over the years, renters always got fewer points than those who owned their residence or were buying. Interesting. I often wondered why it was fair, and unquestioned, that credit scoring penalized credit applicants just because of their choice of lifestyle.

I grew up in rental housing – my parents always rented homes and apartments. My Dad always said “If something breaks or wears out (like the garbage disposal, furnace, air conditioner, a leak in the roof, etc.), then it is somebody else’s headache,” and he was compulsive about his financial affairs – very insurance-minded and he never paid a bill late in fifty years! According to the credit models, my Dad was higher risk because he rented!

That just doesn’t make sense, but that is not the only characteristic in credit models today that doesn’t make sense! I testified last year about “proxies for common sense” during a Congressional Hearing on what consumers should know about their credit score. I have been told by credit scorecard developers that the characteristic “Own/Rent” is a proxy for wealth and stability. Instead of scoring consumers on proxies for their stability, capital and capacity, why not categorize them based on the facts?

That is what the Comprehensive Credit Assessment Framework (CCAF) is about. Please be on the lookout for the next post, where Sunny Zhang will highlight the advantages of CCAF over today’s typical credit scoring-based underwriting systems.

In the September 2009 issue of the RMA Journal, the 5 C’s of Credit have again achieved center stage as “the banking world’s long-established tools for credit analysis,” to quote the author, Warren Stell. In the article, Mr. Stell makes the case for qualitative, in addition to quantitative, analysis that leverages the 5 C’s of Credit with a future-oriented focus. At the close of his article, he quotes Richard Fischer, president of the Federal Reserve Bank of Dallas, who commented to The Wall Street Journal on the causes of the credit bubble: “Finally … there was the ‘mathematization of risk:’ Institutions were ‘building risk models’ and relying heavily on ‘quant jocks’ when ‘in the end there can be no substitute for good judgment.’” Mr. Stell concludes that “Good judgment is the lender’s trump card, augmented with qualitative and quantitative analytical tools such as the Cs of credit …” which he advocates now ought to include Change.

We concur with his assessment and in our latest book we quoted Benjamin Franklin, who said “There is no substitute for Common Sense,” as a lead-in to the second chapter, which makes the case for a comprehensive credit assessment framework that is rooted in the 5 C’s of credit! It is not surprising that we are on the same wavelength with yet another RMA feature article in the space of 5 months.

In our book, we cover, in detail, how the new system we propose will greatly improve the vast majority of loan underwriting systems in operation for consumer and small business lending today. I have been receiving considerable positive feedback on the book from a wide variety of stakeholders since its publication in April.

I had a discussion with my son, who graduated from Med School on Sunday, about the complexity of lending these days. His reaction was “It’s not brain surgery! What’s the big deal?” To which I replied “Son, I have just co-authored an entire book on a new approach to lending and trying to cover the bases with a short white paper is like explaining quantum physics in 3 words or less!” My son replied “Fast moving particles, Dad. What’s your problem?!”

So, rising to my son’s challenge, I have put together some sound bites on CCAF! After all, it is not brain surgery!

Making a loan – sounds simple enough! If this country can put a man on the moon, you would think that we should be able to make loans without melting down the world financial system. Given recent experience, maybe not. The obvious upshot is that making a loan must be more complicated than we think!

OK, but if that is the case, are the complications inherent in the transaction of making a formal IOU between a lender and a borrower, or have we made lending a whole lot harder that it needs to be? I know I am treading on sacred ground here – challenging the status quo. Add to that the fact that when a lot of money changes hands, folks want to get their piece of the action. Well, who wants to grab that porcupine? The short answer is all of a sudden a whole lot of folks, including Congress. There is legitimate outrage surrounding the greed, lack of accountability, lack of transparency, and overly complicated transactions and processes that have come to be embraced by the Financial System Establishment.

I don’t know about you, but I yearn for simpler times – when we had a handful of guiding principles in lending called The 5C’s of Credit. Let me outline, simply, how an alternative system might work.

A credit score based on credit bureau data is incomplete, a flawed concept when it is used in isolation to measure a borrower's creditworthiness. How can you predict whether or not someone will pay a loan without considering all of the relevant facts?

In her recent BusinessWeek article, Jessica Silver-Greenberg discussed the inclusion of the type of merchants with whom consumers transact as a potential factor in calculating a consumer’s credit score. Examples of the types of business that the article in question cited that could lower the consumer’s credit standing were “tire and retreading shops, massage parlors, bars, billiard halls, and marriage counseling offices.”

My last posting highlighted gaps in credit bureau data. Following that line of thought, I have listed several benefits below offered by the use of credit-score-like tools in loan decisioning and the use of non-financial economic transaction data.

1.) It can speed and lower the cost of making a loan decision and pricing the loan, which can be crucial in the context of most consumer and some small business loans. It can also help micro-loans, where the cost of origination and servicing can easily make such loans infeasible at reasonable interest rates.
2.) Negative subjective factors are reduced in loan pricing and approval, reducing discrimination based on non-relevant factors, such as gender or ethnicity.
3.) The lending environment becomes more competitive as lenders no longer need to have a relationship with consumers or businesses in order to do business with them.
4.) With use of non-credit transaction data, vast numbers of consumers and small businesses can be brought into the financial mainstream and gain access to credit.
5.) With greater information, lending decisions become better, with lower rates of delinquencies, less overextension, and an increase in the number of performing loans.

The way credit scoring uses alternative data is basically same as it does for traditional data. The inherent problem for scoring as a technology is that it is unable to simultaneously consider all of the primary factors; rather it can only pick them one at a time. The order in which variables are selected affects their weight. Variables that get selected are not necessarily the most predictive individually because selection is based on their marginal contribution. So, even with expanded data, credit scoring still faces challenges to accurately qualifying thin-, or no-file borrowers.

On the other hand, CCAF should outperform credit scoring and underwriting systems composed of scoring with judgmental overlays applied after the fact, even when both methods use identical alternative data for input .This is due to the way in which CCAF is constructed and the previous discussion concerning the various gap components. One of the chief advantages of CCAF is that all primary givens are captured simultaneously via the handle structure for bad performance, thus eliminating path dependency in the choice of model factors. With CCAF, it is known which factors, in which combinations, present specific performance levels. Credit scoring does not have this capability, nor does it possess a feedback mechanism to adjust factor weightings over time as experience accumulates (e.g. credit scoring is not adaptive, rather its predictive strength diminishes over time). Even when credit scoring systems are re-developed the factors are again considered one at a time and selected in a particular sequence.

Defining the underwriting gap

The underwriting gap refers to the difference between the underwriting decisioning model and the borrower, business, and market realities. Narrowing the gap translates to more accurate loss predictions, fairer treatment of the customer, and less reliance on the assumption that the past determines future outcomes. Moreover, the introduction of the CCAF will have an impact on mainstream underwriting standards. Lenders will begin to recognize the benefits of putting borrowers, and their credit transactions, in the proper context before attempting to determine creditworthiness or how much to charge for a particular loan.

Many underwriting systems have evolved to be convoluted in the sense that they incrementally work off of partial information along the decisioning process. As a result, consumers may be over-charged, or they may be approved for loans that are not affordable. When working with incomplete information, it is not possible to properly construct factor weights, because their information value relative to predicting loan default can be significantly reduced.

There are several technical underwriting modeling components that can be leveraged to close the gap, namely a data component, a sampling/segmentation component, a model factor component, a model formulation component, and a model construction component. I will further explain those in my future postings. It is useful first to consider some examples of the outcomes associated with the gap, which impacts the full spectrum of borrowers.

First I can examine the case where a borrower is a well-established revolving credit user, primarily for the rewards benefit, possessing very strong capacity and ample liquid capital reserves. For such a consumer, there is less motivation to use credit to finance automobiles or consumer durable purchases. As a result, a credit bureau-based credit score may negatively assess a lack of, say, recent installment loan information, and thus be unable to calculate the ratio of revolving to installment debt, etc. Such a consumer may also be found to have too many accounts reported or too many accounts with balances if they use several credit cards. Furthermore, some delinquency in credit payments may simply signify a willingness to pay late fees for convenience and not any greater risk of default. In this instance, the added lowering of credit score in addition to late fee assessment may over-penalize the borrower. With credit score-driven risk-based pricing, this consumer will pay more for their financing, say a mortgage, than is necessary or reasonable.

On the other end of the spectrum, I may look at a thin-file borrower with a small amount of capital and a relatively small, but steady, income over several years, with very little existing debt. In this instance, his/her payment history of meeting obligations must play a greater role. Another important consideration would be the amount of non-credit obligations (for example rent, utilities, telecommunications, insurance, subscriptions) relative to income, and the impact of the proposed loan, or payment shock, relative to monthly cash outflows. Unfortunately, sparing, or lack of, use of credit in the past, and fast repayment of any debt will likely hurt, rather than help this borrower. In addition, in the absence of alternative data on rent, phone, utility, and other regular payments, it will be difficult for the borrower to obtain desired financing. Even if successful, these individuals will likely be required to pay more for their loans than the true default risk would dictate.

This conclusion is consistent with the Federal Reserve Board’s recent report to Congress on credit scoring, which noted that recent immigrants and young people were assigned lower scores by the models they developed than is appropriate, given the actual performance of these groups.

Judgmental components play an important role throughout the CCAF development process. In fact, CCAF can be considered as a “systematic” judgmental decision process that can be fully automated and updated after initial judgmental factors are integrated. Since it can be executed in an automated fashion like credit scoring, it can overcome some shortcomings inherent with traditional judgmental systems, and provide fast, consistent and efficient credit assessment. This can be crucial in the context of most consumer and some small business loans, and also for segmenting and monitoring micro-loans to reduce origination and servicing costs and better price the risk.

Credit scores make many assumptions and you know the old adage, “never assume.” Well, without these assumptions, credit scoring as we know it today, would not exist.

Credit scoring models are based on observed performance for a set period of time. Scoring system technology was originally sold chiefly on its ability to:

1.) speed the process - credit scoring was faster, even when performed in a manual environment. Scores at or above the cut-off are approved, and those below the score cut-off are declined. Choices around the “risk tolerance” threshold are always the most difficult ones.

2.) enforce consistency - credit systems enforce consistency because two consumers falling into the same categories for scored factors will receive identical scores. However two credit applicants falling into different categories for scored factors may still receive identical scores. End result is that two applicants may have identical scores, but not identical credit characteristics.

3.) precisely quantify bad performance risk - specific lenders and loan products with custom credit systems are capable of quantifying the odds of bad performance. They are developed based upon the compilation of large numbers of similar loans where the historical performance is known. The credit score is simply the odds of repayment, scaled to be a positive number that ranges within a few hundred points.

4.) outperform individual loan officers on bad loan prediction - when judgmental systems consisted of a collection of fairly autonomous loan officers, the scoring system always outperformed any individual loan officer. This was because the comparison was based upon all, and not just some, of the cases to be decided.

With these assumptions dissected, we can see how credit granting has become a data mining exercise of separating out who did and did not re-pay in the past, irrespective of causality, and not a reasoned response to who will likely re-pay. The real issue is which decision is actually most likely to come true. More on that next week!

As we continue to ponder how we got into the subprime lending crisis, it may be worthwhile to pause and look back at what existed before credit scoring.

Before the 1960s, loan officers lent money based on the 5 C’s of credit: Character, Capacity, Capital, Collateral and Conditions.