CR 7 – Credit Scoring and Retail Credit Risk Management

Retail credit risk is the chance of loss when individual consumers and small businesses fail to repay what they have borrowed from a bank. Once seen as the dull cousin of corporate lending and trading, retail banking was reshaped by new products, marketing methods and risk tools, and in the early 2000s it gave many banks growing, fairly stable profits. The same retail world also produced the poorly controlled subprime mortgage lending in the United States that fed the collapse of the securitization industry and helped trigger the 2007 to 2009 financial crisis. So retail credit is both a dependable earner and, when handled carelessly, a source of systemic damage.

Retail banking serves consumers and small firms. It takes consumer deposits and runs the main consumer lending businesses, and the lending side is where the credit risk sits. The principal retail credit products are worth knowing by name, because each carries its own risk drivers.

The Basel Committee definition of a retail exposure

The Basel Committee on Banking Supervision, which sets international standards for bank regulation, treats retail exposures as homogeneous portfolios with three defining features. They hold many loans, each small and of low value; the loans may have either a consumer or a business focus; and the extra risk added by any one exposure is small. Typical examples are credit cards, installment loans such as personal finance, educational loans, auto loans and leasing, revolving credits such as overdrafts and home equity lines of credit, and residential mortgages. A bank may manage small business loans as retail exposures as long as its total exposure to the small business borrower stays below 1 million euros.

Homogeneity is what makes the retail approach work: a pool of thousands of similar loans can be studied as a group, which is the starting point for credit scoring.

A retail credit portfolio is built from bite-sized exposures, so the default of any single customer can never cost enough to threaten the bank. A corporate or commercial portfolio is different: it often holds large loans to single names, and it frequently concentrates exposure in companies whose fortunes are tied together because they share a region or an industry. Granularity is the root of most of the contrasts that follow.

Expected loss behaves like a cost of doing business

In normal markets a retail book tends to behave like a well-diversified portfolio. Because so many small, roughly independent loans are pooled, the bank can estimate with reasonable confidence what share of the book will default and how much those defaults will cost. That expected loss can then be handled like any other running cost, in the same way as the expense of keeping branches open or processing payments, and it can be built directly into the price charged to customers. Commercial portfolios are dominated by a different problem. Their main danger is not the expected loss but the possibility that losses climb to some unexpected level, because a few large defaults or a correlated sector downturn can swamp the average.

Early warning signals and room to react

A more comforting feature of retail books is that trouble usually announces itself. Customers under strain change their behavior before they default, for instance by missing the minimum payment on a credit card, and well-run banks and their supervisors watch for this because it leaves time to act. The bank can tighten the rules on how much it lends to existing customers, shift its marketing and acceptance rules toward less risky applicants, or price the extra risk into higher interest rates for the customer groups where default has become more likely. A commercial loan book, by contrast, turns like a supertanker: by the time it is clear that something has gone wrong, it is often too late to change course.

Warning signals help only if banks act on them. Lenders are often tempted to ignore them to protect a fast-growing line, or even to chase volume by loosening standards, as US subprime mortgage lenders did before the 2007 to 2009 crisis.

Regulatory treatment and segmentation

Supervisors take the view that losses on retail books are fairly predictable, and they regard mortgage loans as safer because a specific real estate asset backs each loan. So under Basel II and III, the capital a retail bank must hold is modest next to what corporate lending demands. The relief comes with a data obligation. For clearly separated segments of each portfolio, banks must report three statistics to supervisors: exposure at default (EAD), loss given default (LGD) and probability of default (PD). Segments should be drawn using credit scores or an equivalent measure together with the vintage of the exposures, meaning the age of each loan since it was booked.

The dark side of retail credit risk is the danger that losses jump suddenly to unexpected levels because a systematic risk factor, one nobody foresaw, pushes many borrowers in the portfolio toward default together. Diversification across thousands of small loans protects a bank only while those loans default more or less independently. The 2007 to 2009 crisis showed how fickle that protection can be. After years of easy credit, house prices fell at roughly the same time across an economy as large as the United States, although the size of the fall varied a great deal between regions, and large mortgage portfolios turned out to be far less diversified than their owners believed. A systematic shift in lending practice can do similar harm: if an industry starts advancing money without checking borrowers’ incomes, it plants a hidden common risk in every portfolio that follows the practice. When the economy then weakens, default rates lurch upward and collateral values such as house prices fall unexpectedly.

The danger was visible before the crisis. Michel Crouhy, Dan Galai and Robert Mark described the dark side in the 2006 edition of their book and warned that subprime lending was risky for an unwary bank: if those customers defaulted far more than estimated, or their behavior shifted with a social trend, losses could cut through even wide margins and fees, and banks new to the sector lacked the history to forecast default reliably.

Four prime causes

• Too little loss history. Many innovative retail products have not existed long enough to build up the loss data needed for a dependable risk assessment.
• A sharp change in the economy. Even well-understood products can behave strangely when the environment changes suddenly, especially if several risk factors worsen together in a so-called perfect storm. The standing fear in mortgage lending is that a severe recession arrives together with rising interest rates, which would raise defaults just as house prices, and so collateral values, fall very sharply.
• Shifting social and legal norms. Whether consumers default depends on a complex social and legal system that keeps evolving. The growing social and legal acceptability of personal bankruptcy, particularly in the United States, appears to have contributed to rising personal default during the 1990s.
• Operational failures in the credit process. Consumer credit runs as a semiautomated decision process rather than a series of individual judgments. Any operational flaw in how customers are assessed is therefore repeated across the whole consumer portfolio.

These wild-card risks are hard to quantify almost by definition, so a bank should limit the number of its retail portfolios that are especially exposed to unfamiliar risks of the subprime kind. A small exposure to uncertainty can open a profitable line of business and give the bank time to collect enough data to measure the risk properly later; a large exposure leaves the bank at the mercy of events. Where big conventional books such as mortgage portfolios are vulnerable to several risk factors moving sharply at once, banks need stress tests to judge how damaging each plausible worst case could be.

How subprime standards slipped

From 2002 until the subprime crisis broke in 2007, the belief that real estate prices would keep rising, together with low interest rates, poor broker incentives and fierce competition, eroded underwriting standards, and loans went to borrowers who often could not afford them. Many subprime loans of the period combined several weaknesses at once: borrowers of poorer credit quality, combined loan-to-value ratios that were high, and income that was checked lightly or not at all.

Some of these loans were hybrid ARMs labelled 2/28 or 3/27: a low fixed teaser rate ran for an initial two or three years, after which the rate floated. The payment shock at reset meant the borrower would realistically need to refinance, which was feasible only if the house had risen in value; otherwise default was likely. Because so many of these loans were written around the same time, lenders had unintentionally set up a system-wide surge in which borrowers would all have to refinance or default at once. Consumer attitudes toward defaulting on mortgage debt also changed in ways that banks and rating agencies did not foresee. Commentators called the 2007 events a perfect storm, but it was a storm the banking industry had largely made for itself.

Post-crisis measures: qualified mortgages and ability to repay

After the crisis, reforms under the Dodd-Frank Act (DFA), including the creation of the Consumer Financial Protection Bureau (CFPB), set out to contain the dark side of retail lending. The CFPB rules require a mortgage originator to establish that the borrower can in fact afford to repay. If the loan meets the definition of a qualified mortgage (QM), the creditor may presume that this requirement has been satisfied.

The ability-to-repay rule asks the lender to weigh eight underwriting standards: the borrower’s current employment status, current income or assets, credit history, the monthly mortgage payment, monthly payments on any other loans tied to the property, monthly payments on obligations linked to the mortgage (property taxes, for instance), the borrower’s other debts, and finally the monthly DTI ratio, or residual income, that would result once the new mortgage is added.

Credit risk is the largest financial risk across most retail lines, yet retail banking also carries market, operational, business and reputation risks, several of which are moved out of the business line and handled centrally.

Interest-rate risk

Whenever the bank quotes specific rates to borrowers on one side and to depositors on the other, it creates interest-rate risk on both its assets and its liabilities. This exposure is usually passed from the retail line to the bank’s treasury, which manages it alongside the rest of the institution’s asset/liability and liquidity risk.

Asset valuation risk

Asset valuation risk is a special form of market risk in which the profit of a retail line depends on valuing an asset, a liability or a class of collateral correctly. The most important case is prepayment risk in mortgage banking. When interest rates fall, borrowers who want to remortgage repay their existing loans unexpectedly fast, and the mortgage portfolio loses the value it was expected to earn. Valuing and hedging assets exposed to prepayment is difficult because it depends on behavioral assumptions about customers that are difficult to validate. A second example is the residual value of cars at the end of leases in auto leasing. Where these risks are recognized explicitly, the retail bank’s treasury generally manages them centrally.

Operational risk

Operational risk in retail banking is generally managed inside the business where it arises. Customer fraud, for instance, is monitored closely, and new controls such as fraud detection systems are introduced when their cost is economically justified. Basel II and III require capital for operational risk across wholesale as well as retail activities, and a specialist branch of retail operational risk management has developed that borrows many concepts from firm-wide operational risk practice.

Business risk

Business risk is a central concern of senior management. It covers volume risk, such as the rise and fall of mortgage business as interest rates move; strategic risk, such as the spread of internet banking or new payment systems; and choices about acquiring or merging with other firms.

Reputation risk

Reputation risk carries particular weight in retail banking. The bank must keep its reputation with customers for delivering what it promises, and it must also keep its standing with regulators, who can withdraw its business franchise if they judge it to have acted unfairly or unlawfully.

A credit risk scoring model converts what is known about a borrower, whether a new applicant or a current customer, into numbers, and then combines those numbers, usually by simple addition, into a single score. The score serves as an indicator of the individual’s credit risk, interpreted as the probability of repayment: a higher score means lower risk. Such models run behind everyday decisions such as card applications, telephone accounts, medical claims and auto insurance.

Why banks score: avoidance, profitability, cost and consistency

Scoring lets a bank steer clear of the riskiest customers. It also shows whether a business line is likely to pay, by comparing the margin left after operating and estimated default costs. Cost and consistency matter too. Large banks have millions of customers and handle billions of transactions a year, and scoring lets them automate the decision process for small loans and credit cards as far as possible. Before scoring spread, a credit officer read each application and applied experience, industry knowledge and personal judgment to a large amount of information. A typical application held about 50 items of information, and some asked for as many as 150. The number of possible combinations is so vast that no human analyst could treat similar applications identically over time, whereas a scorecard weighs and treats every item in exactly the same way on every occasion.

Characteristics and attributes

The industry calls each item drawn from an application or a credit bureau report a characteristic; it matches a question on the form or an entry in the bureau file. The answer given is called an attribute. If the characteristic is time at address, then four years is an attribute; if the characteristic is residential status, then renting is an attribute. A scoring model judges not only whether an attribute is favorable or unfavorable but also by how much, and the weight assigned to each attribute comes from statistical analysis of the odds of repayment in past performance data.

Population odds

In retail banking the word odds is used loosely to mean probability, but the formal definition of population odds is the ratio of the probability of a good outcome to the probability of a bad outcome across the population. A bad customer here means one who becomes delinquent or whose debt is charged off.

An applicant drawn at random from a population with 15:1 odds therefore has a 1 in 16 chance of being bad, which is 1/16 = 6.25 percent. Odds of 24:1 work the same way and give 1/25 = 4%, while a bad rate of 2% corresponds to odds of 98 to 2, or 49:1.

Several statistical techniques are used to set the weights: linear regression, logistic regression, mathematical programming, classification trees, neural networks and genetic algorithms. Logistic regression is the most common.

An application scoring table

The table below lists the points awarded in an example application scoring table published by Edward Lewis in 1992. Each applicant collects exactly one attribute score for each of the seven characteristics, and the scores are added together. The last column is derived here and shows the spread between the best and worst attribute, which signals how strongly each characteristic separates good from bad applicants.

Source: Lewis, An Introduction to Credit Scoring, 1992. Attributes reordered from highest to lowest points; spread column added.

Points do not always move in a straight line with the characteristic: applicants aged 26 to 31 score less than those aged 18 to 25, because the weights are fitted to data rather than set by intuition. The worst credit reference is the only characteristic that can subtract points.

For scoring consumer credit applications there are essentially three kinds of model, and they trade cost against how closely they fit a particular lender’s customers.

• Credit bureau scores. These are widely called FICO scores, because Fair Isaac Corporation, the leading firm in retail credit analytics, developed the method. Firms such as Equifax and TransUnion also supply bureau scores in the US and Canadian markets. For a bank, a generic bureau score is cheap, quick to install and gives a broad view of an applicant’s overall creditworthiness whatever type of credit is being sought. Fair Isaac scores can be adapted to a lender’s preferences; they usually run from 300 to 850, and subprime lending typically targets customers scoring below 660.
• Pooled models. Outside vendors such as Fair Isaac build these from data collected across many lenders with similar portfolios; a revolving credit pooled model, for example, might be estimated from card data supplied by several banks. On price they sit between the two other types, dearer than a generic score but cheaper than a custom build, and they fit an industry without fitting any single company.
• Custom models. These are usually built in-house from the lender’s own applicant data and are tailored to screen a specific applicant profile for a specific product. Custom models have let some banks become specialists in segments such as credit cards and mortgages, and they can give a strong competitive edge in picking the best customers and offering the best risk-adjusted pricing.

What a credit bureau file contains

A credit bureau holds a credit file for each individual with a credit history, and each file carries five main kinds of information.

• Identifying information. Personal details that are not treated as credit information and are not used in scoring. Local law decides what may be collected; in the United States, equal opportunity legislation forbids scoring models from using a person’s gender, race or religion.
• Public records (legal items). Items from civil court records, including bankruptcies, judgments and tax liens.
• Collection information. Reports from debt collection agencies or from the credit granters themselves.
• Trade line or account information. Built from the monthly receivables data that lenders send to the bureaus, covering new accounts and updates to existing ones.
• Inquiries. Each time someone opens a file, an inquiry is logged. Credit granters see only the inquiries made in connection with an extension of new credit.

Some bureaus, Equifax among them, let individuals see their own score together with guidance on how to raise it and what-if analysis, such as the effect on the score of paying down card balances.

From a bureau score to a mortgage risk score

A bureau score is only one input into mortgage credit assessment. Lenders can combine it with other key variables, including the loan-to-value ratio and the quality of the loan documentation, to build a broader risk score.

One failure in the years before the 2007 to 2009 crisis was that some originators leaned too heavily on bureau scores and gave too little weight to this wider set of variables. After the crisis, mortgage lending returned to more traditional standards: full documentation loans, borrowers required to have credit scores above 680, and much larger down payments. The industry moved away from negative amortization, stated income, no income/no asset and no documentation loans, and from 100 percent financing.

Early scoring models were built only to rank applicants by relative risk, so the absolute probability of default attached to each applicant hardly mattered. Lenders used the ranking to choose a cutoff score, the point at or above which applicants were accepted, and the choice of that point rested on subjective criteria.

Goods, bads and the cutoff

Picture the score distributions of good and bad accounts drawn on the same axis. Bad accounts cluster at lower scores and good accounts at higher scores, but the two distributions overlap. Suppose the minimum acceptable score is 680. Rejecting every application below 680 keeps the lender away from most of the bad customers, who sit to the left of the line, but it also turns away the smaller group of good customers who happen to score below 680. Those rejected good accounts are false bads. On the right of the line, some bad accounts still score above 680 and are accepted; those are false goods.

Shifting the cutoff to the right removes an even larger share of bad accounts but forgoes a larger share of good ones, and shifting it left does the reverse. So the cutoff is a central business decision that fixes both the likely profitability of the product and the risk the bank takes on. With a cutoff in place, experience reveals the product’s loss rate and profitability, and the bank can move the cutoff to improve the margin and reduce false goods and false bads. Data build up quickly in retail, though only a longer series spans a full economic cycle. Loss rate and profitability statistics are usually refreshed every quarter.

From loss rates to implied default probabilities

The Basel Capital Accord requires banks to split each retail book into groups that share loss behavior, prepayment risk in particular, and to estimate both PD and LGD for each. A natural way to do this is to segment each portfolio by score band, with every band representing one risk level. For each band the bank measures the historical loss rate and, given an LGD estimate, backs out the PD implied by those losses.

A historical loss rate of 2 percent with an LGD of 50 percent gives an implied PD of 2% / 50% = 4 percent. The same arithmetic applied across a set of score bands produces a PD for each band, as the illustrative figures below show for a book with an assumed LGD of 60%.

The bank should then apply common risk management policies to all borrowers and transactions within a segment. Those policies cover underwriting and loan structure, economic capital allocation, pricing and other lending terms, monitoring and internal reporting.

A scorecard must give high scores to future good accounts and low scores to future bad ones, so its goal is to minimize the overlap between the two distributions. Risk managers then need to measure that performance and know when to adjust or rebuild the scorecard or change operating policy.

Building the CAP curve

The traditional validation tool is the cumulative accuracy profile (CAP) together with its summary statistic, the accuracy ratio (AR). To draw the CAP, rank every customer by score, starting with the riskiest and ending with the safest, and place that ordering along the horizontal axis as a cumulative percentage of customers. On the vertical axis plot the cumulative percentage of all actual defaults, taken from the bank’s records, captured up to each point.

Two benchmarks frame the picture. Suppose the model predicts that 10 percent of accounts will default over the next 12 months. A perfect model would place every one of those defaulters in the riskiest 10% of the ranking, so its CAP rises in a straight line to 100% of defaults at the first decile and stays flat thereafter. A random model with no power to tell good from bad captures defaults only in proportion to the population covered, so its CAP is the 45-degree line. A real model lies between the two, and the bank wants its curve as close to the perfect model line as possible.

The accuracy ratio

Let A_R be the area between the actual rating model’s CAP and the random model’s diagonal, and A_P the area between the perfect model’s CAP and the same diagonal. The accuracy ratio compares the two.

With a default rate d, the perfect model’s curve lies above the diagonal by an area of one half minus d over 2, which is a quick way to find A_P.

Monitoring and replacing a scorecard

A bank can track a scoring model, perhaps every quarter, by redrawing its CAP curve and replacing the model when performance slips. Scorecard performance rarely collapses overnight, but it can erode for two reasons: the characteristics of the underlying population can drift, and the population’s behavior can evolve so that different variables become associated with a high likelihood of default. A third reason for replacement is a change in the products the bank sells. If a lender that has offered auto loans sells that business and starts issuing credit cards, its target customers will very probably be different enough to justify a new custom scorecard.

Scoring began as a one-time check on an application, but banks now also run periodic behavior scoring on existing customers. A behavior score uses what the bank observes about a current customer, such as how the credit line is used, together with social and demographic information, to estimate the risk of default over a fixed period. The method resembles application scoring but draws on many more variables describing the customer’s past performance.

Beyond default: the family of scorecards

Risk modeling of this kind is no longer limited to predicting default. Lenders have moved toward judging how valuable a customer is to the bank, and scoring methods now cover response, usage and attrition as well as risk, so that one customer may carry several scores at once.

The trade-off between creditworthiness and profitability

The safest customers are not always the most profitable. A bank gains little by issuing expensive credit cards to highly creditworthy people who never use them. Customers who are a little more likely to default may still be worth more than higher-scoring customers if they borrow often or accept a higher interest rate. The key risk management question is whether that extra profit really covers the extra risk over the longer term, which comes down to asking whether the default rate of the marginally less creditworthy customer would shoot up if the economy turned sour.

From default scoring to profit scoring

To capture this interplay of risk and reward, leading banks are moving from traditional default scoring to product profit scoring, which estimates the profit the lender earns from a customer on a specific product, and on to customer profit scoring, which estimates what the whole relationship with that customer is worth to the lender. With that information, lenders can set limits, margins and product terms so that each relationship earns as much as possible, and they can keep adjusting risk, operating and marketing parameters during the relationship.

The customer relationship cycle

Best-practice banks organize these decisions around a four-stage cycle. Marketing initiatives decide whom to target with a fresh or tailored offer, what message to use, whether to mail and how often, guided by studies of how segments respond. Screening applicants means using scorecards to accept or reject applications, to set the initial credit line and to set pricing tiers suited to each client’s risk level. Managing the account is a continuing series of decisions based on observed behavior: raising or cutting a credit line, repricing, authorizing a temporary excess over the line, reissuing or renewing a line, choosing the service level, and collecting overdue interest or principal on delinquent accounts. Cross-selling closes the loop, as the bank uses its detailed knowledge of existing customers to offer them more products; customers who already hold checking and savings accounts might be offered a mortgage, a credit card or insurance. Risk management thus becomes part of the wider business decision.

Two post-crisis refinements

Since the 2007 to 2009 crisis, two trends have sharpened this approach. First, banks are working harder to understand how macroeconomic factors such as house prices and unemployment change the behavior of each score band, so that forecast default rates reflect where the economy currently stands in its cycle. This links to stress testing, so that decisions reflect both the consensus economic outlook and the capital costs and losses implied by the higher default rates of adverse scenarios, with further adjustment for social change such as new personal finance laws. Second, firms trial variations of an offer, see who responds, and track how the takers perform in the first months, for example on new credit cards. Lessons from this tasting exercise feed the wider campaign once their effect on capital costs and risk-adjusted profitability is understood. Both trends make retail decisions more forward-looking and granular than a static view built on history alone.

Risk-based pricing (RBP) means building risk-driven account economics explicitly into the annualized interest rate each customer is charged, account by account, so that customers with different risk profiles pay different prices for the same product. Both competition and regulation are pushing retail finance in this direction. The economic factors that enter the price are operating expenses; the probability of take-up, meaning the chance the customer accepts the offer; PD, LGD and EAD; the capital the transaction consumes; and the institution’s cost of equity.

Why one price for everyone fails: adverse selection

Banks increasingly see that charging a single price to all customers in a competitive market produces adverse selection. A uniform price looks cheap to high-risk customers, who are drawn in, and expensive to low-risk customers, who take their business to a lender that prices them more fairly. The bank ends up with a riskier book than its price assumed, and the full extent of the damage may become visible only when the economy deteriorates.

Tiered pricing

Tiered pricing means differentiating price by score band above the cutoff score, so that a higher score earns a lower price; put the other way, price rises as score bands become riskier. A tiered policy of this kind makes risk-based pricing easier to run and more effective. Many leading institutions use some RBP for new auto loan, card and mortgage accounts, with more longer-term considerations added since the 2007 to 2009 crisis, yet retail RBP is still in its infancy: pricing seldom reflects business objectives, low-balance accounts are poorly priced against high-balance revolvers, and cutoffs and tiers are often set by ad hoc rules of thumb.

What a well-designed RBP strategy delivers

A sound RBP strategy lets the bank map different pricing choices at the credit score level onto its key corporate metrics, such as revenue, profit, loss, risk-adjusted return, market share and portfolio value, and it is a critical part of best-practice retail management. It blends external market information, such as how likely the customer is to accept, a probability that itself moves with price and credit limit, with internal information such as the cost of capital. Its main benefits follow from this.

• It raises shareholder value by meeting management objectives while respecting several constraints at once, such as having to give up some profit for market share or some market share for lower risk. Mathematical programming methods, such as integer programming, have been developed to find the best pricing subject to those constraints.
• It gives retail bankers a pricing lever for balancing the goal of growing market share against the goal of cutting the rate of bad accounts. To grow share in a risk-adjusted way, a bank can trace how its bad rate climbs as it accepts a growing share of the population, a relationship known as the strategy curve.
• It recovers money that traditional pricing leaves on the table: improved pricing can raise the main corporate measures by 10 to 20 percent or more.
• It counters adverse selection by keeping low-risk customers and charging high-risk customers enough to cover their cost.

The same logic applies when nonbanks extend credit to consumers and small businesses, but RBP depends on systems and operations that many retailers lack. So retailers tend to rely on credit card payments and on financing provided by financial institutions.

Tactical and strategic uses of scoring

Tactically, scoring helps predict which customers will stay or leave, cut attrition among the right customers, match each customer with the most suitable product, and promote newer services such as retirement planning. Strategically, banks ask whether they extract enough lifetime value from each account, how much future value the portfolio holds and where it comes from, ideally benchmarking against peers on measures such as market share.

Because consumer credit grew up around application scoring, it traditionally judged every loan or borrower on its own. What lenders really care about, though, is how whole portfolios of retail loans behave, and the internal ratings-based modeling of Basel II and III has reinforced that portfolio view.

The Basel approach to retail capital

Basel III lets banks calculate regulatory capital under either a Standardized Approach or an Advanced Approach. Under the Advanced Approach the bank estimates its own PD and LGD and uses its consumer credit model to estimate the default loss distribution for each segment. The Accord splits retail into three subsectors (residential mortgages, revolving credit, and other retail such as installment lending) and gives each its own formula to capture the risk in its risk-weighted assets. This structure has pushed banks to estimate default probabilities precisely instead of relying only on relative credit scores, and to segment their portfolios well. Banks that convince supervisors their estimates are accurate can minimize the capital they hold against expected and unexpected retail default losses.

How retail securitization works

Securitization has been a defining feature of consumer lending. Before the subprime crisis began in 2007, around 50 percent of all US home mortgages were securitized. The crisis halted almost all private label mortgage securitization, meaning deals that carry no guarantee from a government-sponsored agency, but that market has been reformed since and is slowly reviving, while securitization of auto loans, credit card receivables and student loans stayed in relatively good health throughout.

The practice began in US home mortgages, where a substantial share was securitized by the late 1970s, with growth accelerating in the 1980s. Federal government sponsorship of three agencies was a catalyst: Fannie Mae (the Federal National Mortgage Association, FNMA), Freddie Mac (the Federal Home Loan Mortgage Corporation, FHLMC) and Ginnie Mae (the Government National Mortgage Association, GNMA). These agencies issue securities whose payments come from pooled home loans that banks and other lenders originated, and a mortgage must meet requirements on structure and amount to enter a pool. From the 1990s a private label market grew quickly, producing many kinds of mortgage-backed and other securitized products.

In a collateralized mortgage obligation (CMO), payments from the mortgage pool are divided into tranches, with the first tranche receiving the first payments and the others taking their turn. Asset-backed securities (ABSs) is the broader term for instruments based on other assets as well, such as card receivables, car loans, home equity lending and lease receivables. By selling the cash flows to investors, the bank receives principal up front instead of collecting it gradually over the life of the loans. The securities may be sold to third parties or issued publicly as tranched bonds, meaning classes of senior and subordinated bonds rated by a rating agency.

Risk transfer versus regulatory arbitrage

Securitizations differ in legal form, cash flow reliability and how much of the riskier tranches the bank keeps. Sometimes the bank genuinely shifts most of the portfolio’s risk to investors and so reduces the economic risk and economic capital tied to it; it gives up part of the borrowers’ income and keeps a margin that pays it for originating and servicing the loans. In other deals the structure is designed around the regulatory rules to cut the risk capital the bank must hold while transferring much less economic risk to investors. That practice is regulatory arbitrage, the exploitation of differences in the capital charges applied to different kinds of asset.

What went wrong and the reforms that followed

Three trends undermined mortgage and other securitization markets before the 2007 to 2009 crisis. Subprime and similar loans were originated specifically to be securitized, often by lightly capitalized and lightly regulated firms such as brokers with no long-term stake in loan quality. Subprime credit was packaged into complex securities whose high ratings rested on fragile assumptions. And banks failed to distribute the securitized risk, holding large amounts of it themselves, either directly or through investment vehicles.

A separate group of reforms aims to stop financial institutions from abusing customers, and these are likely to shape behavior in US retail markets for a long time.

Judgment still matters

Quantitative tools now support retail decisions from origination, where they assess creditworthiness, set lines and apply risk-based pricing, through servicing, where they anticipate payment behavior, detect fraud and prioritize collections. Yet every new product or marketing technique can slip a systematic risk into the portfolio, a common factor that pushes losses far above expectations when the economy or consumer habits shift. So scoring models must be used with a good dose of judgment, grounded in a deep understanding of each product and the customer segment it serves.