what's the point of credit scoring? loretta j. mester...*loretta mester is a vice president and...

Loretta J. Mester

3

What's the Point of Credit Scoring?

What’s the Point of Credit Scoring?Loretta J. Mester*

When one banker asks another “What’s thescore?” shareholders needn’t worry that thesebankers are wasting time discussing the ballgame. More likely they’re doing their jobs anddiscussing the credit score of one of their loanapplicants. Credit scoring is a statistical methodused to predict the probability that a loan ap-plicant or existing borrower will default or be-

come delinquent. The method, introduced inthe 1950s, is now widely used for consumerlending, especially credit cards, and is becom-ing more commonly used in mortgage lend-ing. It has not been widely applied in businesslending, but this, too, is changing. One reasonfor the delay is that business loans typicallydiffer substantially across borrowers, makingit harder to develop an accurate method of scor-ing. But the advent of new methodologies, en-hanced computer power, and increased dataavailability have helped to make such scoringpossible, and many banks are beginning to usescoring to evaluate small-business loan appli-cations.

*Loretta Mester is a vice president and economist in theResearch Department of the Philadelphia Fed. She is alsothe head of the department's Banking and Financial Mar-kets section.

4 FEDERAL RESERVE BANK OF PHILADELPHIA

BUSINESS REVIEW SEPTEMBER/OCTOBER 1997

Credit scoring is likely to change the natureof small-business lending. It will make it lessnecessary for a bank to have a presence, say,via a branch, in the local market in which itlends. This will change the relationship be-tween the small-business borrower and his orher lender. Credit scoring is already allowinglarge banks to expand into small-business lend-ing, a market in which they have tended to beless active. Scoring is also an important stepin making the securitization of small-businessloans more feasible. The likely result wouldbe increased availability of funding to smallbusinesses, and at better terms, to the extentthat securitization allows better diversificationof risk.

WHAT IS CREDIT SCORING?Credit scoring is a method of evaluating the

credit risk of loan applications. Using histori-cal data and statistical techniques, credit scor-ing tries to isolate the effects of various appli-cant characteristics on delinquencies and de-faults. The method produces a “score” that abank can use to rank its loan applicants or bor-rowers in terms of risk. To build a scoringmodel, or “scorecard,” developers analyze his-torical data on the performance of previouslymade loans to determine which borrower char-acteristics are useful in predicting whether theloan performed well. A well-designed modelshould give a higher percentage of high scoresto borrowers whose loans will perform welland a higher percentage of low scores to bor-rowers whose loans won’t perform well. Butno model is perfect, and some bad accountswill receive higher scores than some good ac-counts.

Information on borrowers is obtained fromtheir loan applications and from credit bureaus.Data such as the applicant’s monthly income,outstanding debt, financial assets, how long theapplicant has been in the same job, whetherthe applicant has defaulted or was ever delin-quent on a previous loan, whether the appli-

cant owns or rents a home, and the type of bankaccount the applicant has are all potential fac-tors that may relate to loan performance andmay end up being used in the scorecard.1 Re-gression analysis relating loan performance tothese variables is used to pick out which com-bination of factors best predicts delinquencyor default, and how much weight should begiven to each of the factors. (See Scoring Meth-ods for a brief overview of the statistical meth-ods being used.) Given the correlations be-tween the factors, it is quite possible some ofthe factors the model developer begins withwon’t make it into the final model, since theyhave little value added given the other vari-ables in the model. Indeed, according to Fair,Isaac and Company, Inc., a leading developerof scoring models, 50 or 60 variables might beconsidered when developing a typical model,but eight to 12 might end up in the finalscorecard as yielding the most predictive com-bination (Fair, Isaac). Anthony Saunders re-ports that First Data Resources, on the otherhand, uses 48 factors to evaluate the probabil-ity of credit card defaults.

In most (but not all) scoring systems, ahigher score indicates lower risk, and a lendersets a cutoff score based on the amount of riskit is willing to accept. Strictly adhering to themodel, the lender would approve applicantswith scores above the cutoff and deny appli-cants with scores below (although many lend-ers may take a closer look at applications nearthe cutoff before making the final credit deci-sion).

Even a good scoring system won’t predictwith certainty any individual loan’s perfor-mance, but it should give a fairly accurate pre-diction of the likelihood that a loan applicantwith certain characteristics will default. To

1Some of the models used for mortgage applications alsotake into account information about the property and theloans, for example, the loan-to-value ratio, the loan type,and real estate market conditions (DeZube).

Loretta J. Mester

5


Scoring Methods

Several statistical methods are used to develop credit scoring systems, including linear probabil-ity models, logit models, probit models, and discriminant analysis models. (Saunders discussesthese methods.) The first three are standard statistical techniques for estimating the probability ofdefault based on historical data on loan performance and characteristics of the borrower. Thesetechniques differ in that the linear probability model assumes there is a linear relationship betweenthe probability of default and the factors; the logit model assumes that the probability of default islogistically distributed; and the probit model assumes that the probability of default has a (cumula-tive) normal distribution. Discriminant analysis differs in that instead of estimating a borrower’sprobability of default, it divides borrowers into high and low default-risk classes.

Two newer methods beginning to be used in estimating default probabilities include options-pricing theory models and neural networks. These methods have the potential to be more useful indeveloping models for commercial loans, which tend to be more heterogeneous than consumer ormortgage loans, making the traditional statistical methods harder to apply. Options-pricing theorymodels start with the observation that a borrower’s limited liability is comparable to a put optionwritten on the borrower’s assets, with strike price equal to the value of the debt outstanding. If, insome future period, the value of the borrower’s assets falls below the value of its outstanding debt,the borrower may default. The models infer the probability a firm will default from an estimate ofthe firm’s asset-price volatility, which is usually based on the observed volatility of the firm’s equityprices (although, as McAllister and Mingo point out, it has not been empirically verified that short-run volatility of stock prices is related to volatility of asset values in a predictable way. Saundersdiscusses other assumptions of the options-pricing approach that are likely to be violated in certainapplications.) Saunders reports that KMV Corporation has developed a credit monitoring modelbased on options-pricing theory.

Neural networks are artificial intelligence algorithms that allow for some learning through expe-rience to discern the relationship between borrower characteristics and the probability of defaultand to determine which characteristics are most important in predicting default. (See the articles byD.K. Malhotra and coauthors and by Edward Altman and coauthors for further discussion.) Thismethod is more flexible than the standard statistical techniques, since no assumptions have to bemade about the functional form of the relationship between characteristics and default probabilityor about the distributions of the variables or errors of the model, and correlations among the char-acteristics are accounted for.

Some argue that neural networks show much promise in credit scoring for commercial loans,but others have argued that the approach is more ad hoc than that of standard statistical methods.(The article by Edward Altman and Anthony Saunders discusses the drawbacks.) A study by Ed-ward Altman, Giancarlo Marco, and Franco Varetto analyzed over 1000 healthy, vulnerable, andunsound Italian industrial firms from 1982-92 and found that performance models derived usingneural networks and those derived using the more standard statistical techniques yielded about thesame degree of accuracy. They concluded that neural networks were not clearly better than thestandard methods, but suggested using both types of methods in certain applications, especiallycomplex ones in which the flexibility of neural networks would be particularly valuable.

build a good scoring model, developers needsufficient historical data, which reflect loanperformance in periods of both good and badeconomic conditions.2

WHERE IS CREDIT SCORING USED?In the past, banks used credit reports, per-

2Patrick McAllister and John Mingo estimate that todevelop a predictive model for commercial loans, some20,000 to 30,000 applications would be needed.

sonal histories, and judgment to make creditdecisions. But over the past 25 years, creditscoring has become widely used in issuing



credit cards and in other types of consumerlending, such as auto loans and home equityloans. The Federal Reserve’s November 1996Senior Loan Officer Opinion Survey of BankLending Practices reported that 97 percent ofthe responding banks that use credit scoringin their credit card lending operations use itfor approving card applications and 82 percentuse it to determine from whom to solicit appli-cations. About 20 percent said they used scor-ing for either setting terms or adjusting termson their credit cards.

Scoring is also becoming more widely usedin mortgage origination. Both the FederalHome Loan Mortgage Corporation (FreddieMac) and the Federal National Mortgage Cor-poration (Fannie Mae) have encouraged mort-gage lenders to use credit scoring, whichshould encourage consistency across under-writers. Freddie Mac sent a letter to its lend-ers in July 1995 encouraging the use of creditscoring in loans submitted for sale to theagency. The agency suggested the scores couldbe used to determine which mortgage appli-cants should be given a closer look and thatthe score could be overridden if the under-writer determined the applicant was a goodcredit risk. In a letter to its lenders in October1995, Fannie Mae also reported it was depend-ing more on credit scoring for assessing risk.Both agencies have developed automatic un-derwriting systems that incorporate scoring sothat lenders can determine whether a loan isclearly eligible for sale to these agencies orwhether the lender has to certify that the loanis of low enough risk to qualify (Avery and co-authors).

Private mortgage insurance companies, suchas GE Capital Mortgage Corporation, are us-ing scoring to help screen mortgage insuranceapplications (Prakash, 1995). And it was re-cently reported that four mortgage compa-nies—Chase Manhattan Mortgage Corp., FirstNationwide, First Tennessee, and HomeSide—are involved in a test of the use of credit scor-

ing models for assessing mortgage perfor-mance, prepayments, collection, and foreclo-sure patterns (Talley). This test is being con-ducted by Mortgage Information Corp.

A growing number of banks are using creditscoring models in their small-business lendingoperations, most often for loans under$100,000, although scoring is by no means uni-versally used.3 It has taken longer for scoringto be adopted for business loans, since theseloans are less homogeneous than credit cardloans and other types of consumer loans andalso because the volume of this type of lendingis smaller, so there is less information withwhich to build a model.

The first banks to use scoring for small-busi-ness loans were larger banks that had enoughhistorical loan data to build a reliable model;these banks include Hibernia Corporation,Wells Fargo, BankAmerica, Citicorp,NationsBank, Fleet, and Bank One.BankAmerica’s model was developed based on15,000 good and 15,000 bad loans, with facevalues up to $50,000 (Oppenheim, 1996); FleetFinancial Group uses scoring for loans under$100,000 (Zuckerman). Bank One relies solelyon scores for loans up to $35,000 and approves30 percent of its loans up to $1 million byscorecard alone (Wantland). This spring, a re-gional bank in Pennsylvania began basing itslending decision for small-business loans up

3A survey reported in the American Banker in May 1995with responses from 150 U.S. banks indicated that only 8percent of banks with up to $5 billion in assets used scoringfor small-business loans, while 23 percent of larger banksdid (Racine). The smaller banks were less inclined to adoptscoring, citing small loan volumes. Fifty-five percent ofbanks with more than $5 billion in assets reported theyplanned to implement scoring in the next two years. In amore recent survey of larger banks—the Federal Reserve’sJanuary 1997 Senior Loan Officer Opinion Survey on BankLending Practices—70 percent of the respondents, that is,38 banks, indicated that they use credit scoring in theirsmall-business lending, and 22 of these banks said that theyusually or always do so.

Loretta J. Mester

7


to $35,000 exclusively on a credit score.4 Otherbanks have loan officers review the decisionsbased on credit scores: at First National Bankof Chicago it’s been reported that about a quar-ter of the small-business loan applications re-jected by credit scoring are approved after re-view, and an equal number that pass the scor-ing model are rejected. First Union looks atcredit scores as a supplement to more tradi-tional analyses of businesses’ financial state-ments (Hansell).

Credit scoring is now available to lenderswho do not have sufficient volumes to buildtheir own small-business loan scoring models.In March 1995, Fair, Isaac introduced its “SmallBusiness Scoring Service (SBSS),” a scoringmodel that was developed with RMA, a tradeassociation of commercial lenders. The modelwas built using five years’ worth of data onsmall-business loans from 17 banks in theUnited States, a sample of more than 5000 loanapplications from businesses with gross salesof less than $5 million and loan face values upto $250,000; banks provided data on good andbad accounts and on declined applications, aswell as credit reports on at least two of abusiness’s principals and on the business (Asch;Hansell; and Neill and Danforth).5 Separatescorecards were created for loans under $35,000and for loans between $35,000 and $250,000.The models found that the most important in-dicators of small-business loan performancewere characteristics of the business ownerrather than the business itself. For example,the owner’s credit history was more predic-

tive than the net worth or profitability of thebusiness. While this might seem surprising atfirst, it’s worth remembering that small busi-nesses’ financial statements are less sophisti-cated than those of larger businesses and thatthe owners’ and businesses’ finances are oftencommingled (Hansell). Other companies suchas CCN-MDS, Dun & Bradstreet, and Experian(formerly TRW) are developing or already havecompetitive products. These standardizedproducts make scoring available to lenderswith smaller loan volumes, but the models maynot be as predictive for these lenders to theextent that their applicant pool differs from thatused to create the scorecard.6

Despite its growing use for evaluating small-business lending, credit scoring is not beingused to evaluate larger commercial loans.While the loan performance of a small busi-ness is closely related to the credit history ofits owners, this is much less likely to be thecase for larger businesses. Although somemodels have been developed to estimate thedefault probabilities of large firms, they havebeen based on the performance of corporatebonds of publicly traded companies. It is notat all clear that these models would accuratelypredict the default performance of bank loansto these or other companies. (See McAllisterand Mingo for more discussion on this point.)To develop a more accurate loan scoring modelfor larger businesses, a necessary first stepwould be the collection of a vast array of dataon many different types of businesses alongwith the performance of loans made to thesebusinesses; the data would have to include alarge number of bad, as well as good, loans.

4For its small-business loans between $35,000 and$250,000, a lender makes the decision, but a credit score isalso calculated as a guideline. At this bank, a small-busi-ness borrower is one with annual sales of $2 million or less.

5A good account was defined as one that had not been30 days delinquent more than twice during the first fouryears of account history, while a bad account was one thatat least once had been 60 days or more delinquent (Asch).

6In personal conversation, the manager of the small-busi-ness lending department of a regional bank in Pennsylva-nia reported that it was because of this concern that the bankdoes not rely on the credit score from a standardized modelto make the approval decision for loans between $35,000 to$250,000.



Since the typical default rate on business loansis in the range of 1 percent to 3 percent annu-ally, banks would have to pool their data. Suchdata-collection efforts are currently under way.7

But the fact that loans to large businesses varyin so many dimensions will make the develop-ment of a credit scoring model for these typesof loans very difficult.

BENEFITS OF CREDIT SCORING:QUICKER, CHEAPER, MORE OBJECTIVE

Credit scoring has some obvious benefitsthat have led to its increasing use in loan evalu-ation. First, scoring greatly reduces the timeneeded in the loan approval process. A studyby the Business Banking Board found that thetraditional loan approval process averagesabout 12-1/2 hours per small-business loan,and in the past, lenders have taken up to twoweeks to process a loan (Allen). Credit scoringcan reduce this time to well under an hour, al-though the time savings will vary dependingon whether the bank adheres strictly to thecredit score cutoff or whether it reevaluatesapplications with scores near the cutoff. Forexample, Kevin Leonard’s study of a Canadianbank found that the approval time for con-sumer loan applications averaged nine daysbefore the bank started using scoring, but threedays after scoring had been in use for 18months. Barnett Bank reports a decrease fromthree or four weeks’ processing time for asmall-business loan application before scoringto a few hours with scoring (Lawson).

This time savings means cost savings to thebank and benefits the customer as well. Cus-tomers need to provide only the information

used in the scoring system, so applications canbe shorter. And the scoring systems themselvesare not prohibitively expensive: the price perloan of a commercially available credit scoringmodel averages about $1.50 to $10 per loan,depending on volume (Muolo). Even if a bankdoes not want to depend solely on credit scor-ing for making its credit decisions, scoring canincrease efficiency by allowing loan officers toconcentrate on the less clear-cut cases.8

Another benefit of credit scoring is im-proved objectivity in the loan approval process.This objectivity helps lenders ensure they areapplying the same underwriting criteria to allborrowers regardless of race, gender, or otherfactors prohibited by law from being used incredit decisions (see Credit Scoring and Regula-tion B). Bank regulators require that the fac-tors in a scoring model have some fundamen-tal relationship with creditworthiness. Even ifa factor is not explicitly banned, if it has a dis-parate impact on borrowers of a certain raceor gender or with respect to some other pro-hibited characteristic, the lender needs to showthere is a business reason for using the factorand there is no equally effective way of mak-ing the credit decision that has less of a dispar-ate impact. A credit scoring model makes iteasier for a lender to document the businessreason for using a factor that might have a dis-proportionately negative effect on certaingroups of applicants protected by law fromdiscrimination. The weights in the model givea measure of the relative strength of eachfactor’s correlation with credit performance

8Scoring is one part of an automated loan system, whichpermits banks to offer loans over the phone or via directmail, so that a costly branch network can be avoided. It’sworth mentioning, however, that the costs of a fully auto-mated lending operation at a large bank could be quite high,since reliability would be essential. As one lender haspointed out, when an automated loan system goes down,the bank’s lending operation is out of business. Hence,backup systems are necessary.

7Loan Pricing Corporation and several of its clients arepooling their data on commercial loans so that in severalyears there may be information on a sufficiently large num-ber of good and bad loans to begin building a scoring modelfor commercial loans to larger businesses (correspondencefrom John Mingo, Board of Governors of the Federal Re-serve System staff).

Loretta J. Mester

9


But not everyoneagrees that the objectiv-ity in scoring will benefitminorities or low-in-come individuals, whomay have had limitedaccess to credit in thepast. Some argue thatsince these potential bor-rowers are not well rep-resented in the loan dataon which the scoringmodels have been built,the models are less accu-rate predictors of theirloan performance. (See,for example, the discus-sion in “Mortgage CreditPartnership Project:1996-1997.”) This is a le-gitimate concern. But itneed not be the case thatthe models are less accu-rate, since the factors andtheir weights identifiedin the model could alsobe those that determinecreditworthiness of theu n d e r r e p r e s e n t e dgroups. One study byFair, Isaac indicated thattheir scoring model forinstallment loans was aspredictive for low- tomoderate-income loanapplicants as for the en-tire sample of applicants,although the low-income

subsample had lower scores. (With a cutoff

Credit Scoring and Regulation B

The Equal Credit Opportunity Act (implemented by the Fed-eral Reserve Board’s Regulation B) prohibits creditors from dis-criminating in any aspect of a credit transaction because of anapplicant’s race, color, religion, national origin, gender, maritalstatus, or age (provided the applicant has the capacity to contract),because all or part of an applicant’s income derives from publicassistance, or because the applicant has in good faith exercisedany right under the Consumer Credit Protection Act.

Scoring models cannot include information on race, gender, ormarital status. Recently, the Board amended its commentary onReg B to clarify the use of age in credit scoring models. Reg Bdefines an “empirically derived, demonstrably and statisticallysound, credit scoring system” as one that is: (i) based on data thatare derived from an empirical comparison of sample groups orthe population of creditworthy and noncreditworthy applicantswho applied for credit within a reasonable preceding period oftime; (ii) developed for the purpose of evaluating the creditwor-thiness of applicants with respect to the legitimate business inter-est of the creditor; (iii) developed and validated using acceptedstatistical principles and methodology; and (iv) periodically re-evaluated by the use of appropriate statistical principles and meth-odology and adjusted as necessary to maintain predictive ability.Reg B classifies any other system as a judgmental system, and suchsystems cannot use age directly as a predictive variable in themodel. However, if the model does qualify as an empirically de-rived, demonstrably and statistically sound system, the Board hasdetermined that it can use age directly in the model as long as theweight assigned to an applicant 62 years or older is not lower thanthat assigned to any other age category. And if a system assignspoints to some other variable based on the applicant’s age, appli-cants who are 62 years and older must receive at least the samenumber of points as the most favored class of nonelderly appli-cants. (Any system of evaluating creditworthiness may favor acredit applicant aged 62 years or older.)

(given the other factors contained in the model).Also, a well-built model will include all allow-able factors that produce the most accurate pre-diction of credit performance, so a lender us-ing such a model might be able to argue that asimilarly effective alternative was not avail-able.9

9But banks that override the model for certain borrow-ers need to be particularly careful in documenting the rea-sons for the override to avoid violating fair lending laws.Similarly, borrowers right at the margin of cutoff for ap-proval must be handled carefully.



score of 200, the acceptance rate for low- tomoderate-income applicants was 46 percent,while for higher income applicants it was 67percent. See Fair, Isaac.)10 Freddie Mac alsosays its system, called Loan Prospector, isequally predictive of loan performance, regard-less of borrower race or income (Prakash, 1997).

LIMITATIONS OF CREDIT SCORINGThe accuracy of the scoring systems for

underrepresented groups is still an open ques-tion. Accuracy is a very important consider-ation in using credit scoring. Even if the lendercan lower its costs of evaluating loan applica-tions by using scoring, if the models are notaccurate, these cost savings would be eatenaway by poorly performing loans.

The accuracy of a credit scoring system willdepend on the care with which it is developed.The data on which the system is based need tobe a rich sample of both well-performing andpoorly performing loans. The data should beup to date, and the models should be reesti-mated frequently to ensure that changes in therelationships between potential factors andloan performance are captured. If the bankusing scoring increases its applicant pool bymass marketing, it must ensure that the newpool of applicants behaves similarly to the poolon which the model was built; otherwise, themodel may not accurately predict the behav-ior of these new applicants. The use of creditscoring itself may change a bank’s applicantpool in unpredictable ways, since it changesthe cost of lending to certain types of borrow-ers. Again, this change in applicant pool mayhurt the accuracy of a model that was built

using information from the past pool of appli-cants.

Account should be taken not only of thecharacteristics of borrowers who were grantedcredit but also of those who were denied. Oth-erwise, a “selection bias” in the loan approvalprocess could lead to bias in the estimatedweights in the scoring model.11 A model’s ac-curacy should be tested. A good model needsto make accurate predictions in good economictimes and bad, so the data on which the modelis based should cover both expansions and re-cessions. And the testing should be done us-ing loan samples that were not used to developthe model in the first place.

It is probably too soon to determine the ac-curacy of small-business loan scoring modelsbecause they are fairly new and we have notbeen through an economic downturn sincetheir implementation. Studies of the mortgagescoring systems suggest that they are fairlyaccurate in predicting loan performance. In itsNovember 11, 1995, industry letter, FreddieMac reported some of its own research on thepredictive power of mortgage credit scores byFair, Isaac and CCN-MDS. The agency stud-ied hundreds of thousands of Freddie Macloans originated over several years and selectedfrom a wide distribution of lenders, productand loan types, and geographic areas; it founda high correlation between the scores and loanperformance. The agency also had its under-writers review thousands of loans and founda strong correlation between the underwriters’

10Fair, Isaac’s study used data on direct installment loanapplicants from July 1992 to December 1994. Low- andmoderate-income applicants were defined as those havinggross monthly incomes of less then $1750. By this defini-tion, one-third of their sample and one-fifth to one-half ofthe applicants of each of the individual lenders weredeemed low- to moderate-income applicants.

11For example, suppose owning a home means a personis less likely to default on a loan. Then if the majority ofapplicants that a bank approves are home owners, the dis-tribution of home ownership in the approved applicant poolwill differ from that in the total applicant pool. If this factis ignored in estimating the model, the model could not ac-curately uncover the relationship between home ownershipand loan default. The model would show that home own-ership is less predictive of good performance than it actu-ally is.

Loretta J. Mester

11


judgments and the Fair, Isaac credit scores.Avery and coauthors also found that creditscores based on the credit history of mortgageapplicants generally were predictive of mort-gage loan performance.12

Not all the news on accuracy is good, how-ever. In the November 1996 Senior Loan Of-ficer Opinion Survey, 56 percent of the 33 banksthat used credit scoring in their credit card op-erations reported that their models failed to ac-curately predict loan-quality problems by be-ing too optimistic. The bankers attributed partof the problem to a new willingness by con-sumers to declare bankruptcy. This is a reason-able supposition: this type of “regime shift” (toa world in which there’s less stigma attachedto declaring bankruptcy) would not be pickedup in a scoring model if it was not reflected inthe historical data on which the model wasbased. In response, 54 percent of the bankshave redefined or reestimated their models,and 80 percent have raised the cutoff score anapplicant needs to qualify for credit.

It’s important to remember, though, that acredit scoring model is not going to tell a lenderwith certainty what the future performance of

an individual loan will be. When loan approvaldecisions are based solely on credit scores,some borrowers will be granted credit but willultimately default, which visibly hurts thelender’s bottom line. Other borrowers won’tbe granted credit even though they would haverepaid, which, though less visible, also hurtsthe lender’s profitability. No scoring model canprevent these types of errors, but a good modelshould be able to accurately predict the aver-age performance of loans made to groups ofindividuals who share similar values of the fac-tors identified as being relevant to credit qual-ity.

Many considered the well-publicized denialof then Federal Reserve System GovernorLawrence Lindsey’s application for a Toys ‘R’Us credit card a failure of a credit scoringmodel. But the denial does not necessarilymean the model was faulty. The denial wasbased on the fact that his credit report showedtoo many voluntary credit bureau inquiries,and research by Fair, Isaac shows that as a group,applicants with seven to eight such inquiriesare three times riskier than the average appli-cant and six times riskier than applicants withno such inquiries (McCorkell).13

IMPLICATIONS FOR THEBANKING INDUSTRY

The spread of credit scoring, especially itsgrowing use in small-business lending, shouldlead to increased competition among provid-ers of this type of credit and increased avail-ability of credit for small businesses. Tradi-tionally, lenders to small businesses have beensmaller banks that have had a physical pres-ence, usually in the form of a branch, in the

12Avery and coauthors examined data from Equifax onall mortgages that were outstanding and whose paymentswere current as of September 1994 at three of the largestlenders in the United States. Each loan had a mortgagecredit history score and measures of performance over thesubsequent 12 months, to September 1995. For each loantype (conventional fixed rate, conventional variable rate,or government-backed fixed rate), regardless of seasoningstatus (newly originated or seasoned), borrowers with lowscores had substantially higher delinquency rates than thosewith medium or higher scores, although most borrowerswith low scores were not delinquent. They also examineddata from Freddie Mac on loans for single-family owner-occupied properties purchased by Freddie Mac in the firstsix months of 1994, which showed that borrowers with lowscores had higher foreclosure rates (by the end of 1995), andthat loans with both low credit scores and higher loan-to-value ratios had particularly high foreclosure rates. In ad-dition, credit scores were much stronger predictors of fore-closure than was income.

13A credit bureau inquiry refers to an inquiry intosomeone’s credit history at a credit bureau. A so-calledvoluntary inquiry is initiated when a person seeks credit.An involuntary inquiry can occur without a person’s knowl-edge as part of a routine review of existing accounts or aprescreening for a promotional mailing, for example.



borrower’s neighborhood. The local presencegives the banker a good knowledge of the area,which is thought to be useful in the credit de-cision. Small businesses are likely to have de-posit accounts at the small bank in town, andthe information the bank can gain by observ-ing the firm’s cash flows can give the bank aninformation advantage in lending to these busi-nesses. (Leonard Nakamura’s article discussesthe advantages small banks have had in small-business lending.)

However, credit scoring is changing the waybanks make small-business loans, and largebanks are entering the market using credit scor-ing and processing applications using auto-mated and centralized systems. These banksare able to generate large volumes of small-business loans even in areas where they do nothave extensive branch networks. Applicationsare being accepted over the phone, and somebanks are soliciting customers via direct mail,as credit card lenders do. For example, WellsFargo uses centralized processing for loansunder $100,000, soliciting these loans nation-wide, and uses credit scores not only in theapproval process but also for loan pricing. Forloans over $100,000, it still uses traditional un-derwriting, soliciting in areas where it hasbranches (Zuckerman).

Out west, in the 12th Federal Reserve Dis-trict, the largest banks have increased theircommercial loans of less than $100,000 andhave taken market share from smaller banks,while they have reduced their commercialloans in the $100,000 to $1 million size range,which are less easy to automate (Levonian).14

Many of the larger banks are finding that auto-

mated small-business lending allows them toprofitably make loans of a smaller size thanthey could using traditional methods. For ex-ample, at Hibernia Corp., the break-even loansize was about $200,000 before automation, butnow Hibernia has a large portfolio of loansunder $50,000 (Zuckerman). At Wells Fargo,the average size of a small-business loan is$18,242 (Oppenheim, January 1997).

This spring, a regional bank in Pennsylva-nia planned to solicit small-business loan ap-plications with a direct mail campaign to 50,000current and prospective customers who will beprescreened using the bank’s scoring model.The bank will exclude certain lines of businessand businesses less than three years old. Asimple application form will be used, with nofinancial statements required, and loans up to$35,000 will be approved based solely on thecredit score. Also this spring, PNC Bank Corp.opened an automated loan center in suburbanPhiladelphia through which it plans to process25,000 small-business loan applications fromacross the nation in the next year. While muchof the application process is automated andcredit scores are used, a lender makes the finalapproval decision on a loan application(Oppenheim, May 1997).

For many creditworthy small-business bor-rowers, the entry of the larger banks into themarket means more potential sources for credit.Some banks have found they’ve been able toextend more loans under credit scoring thanunder their judgmental credit approval sys-tems without increasing their default rates(Asch). Credit scoring may also encouragemore lending because it gives banks a tool formore accurately pricing risk.15 However, therelationship that a borrower has with its large

14Mark Levonian reports that between June 1995 andJune 1996, the largest banks in the 12th District increasedtheir holdings of loans under $100,000 by over 26 percent,while other banks increased their holdings of these loansby a little over 3 percent. (These figures are adjusted forbank mergers.)

15Banks that use scoring to develop risk-related loanpricing need to keep in mind fair lending rules and shouldavoid selectively overriding the model for some borrowersand not others.

Loretta J. Mester

13


creditor is likely to be very different from theone it has traditionally had with its small bank.The typical bank-borrower relationship, whichis built up over years of lending, allows forsubstantial flexibility in loan terms. A long-term relationship allows the bank to offer con-cessionary rates to a borrower facing tempo-rary credit problems, which the bank can latermake up for when the firm returns to health.(Mitchell Berlin’s article discusses relationshiplending.)

But automated small-business loans arelikely to be more like credit card loans than tra-ditional business loans, with the terms beingless flexible and set to maximize a bank’s prof-its period-by-period rather than over the lifeof a relationship. Monitoring these borrowerswould likely be more expensive for the bank,since the borrowers may come from outsidethe bank’s traditional lending markets. Thiswould tend to make the bank less flexible onits loan terms. Small businesses that value theflexibility of the traditional relationship loanwill have to seek banks that make loans on thisbasis, most likely smaller banks, as has beenthe case in the past. These smaller banks willmaintain their advantage over larger banks inmonitoring loans, since they have a goodknowledge of the local markets in which theyand their borrowers operate. Businesses thatfind it hard to qualify for loans based solely ontheir credit scores but that, nevertheless, arecreditworthy on closer inspection will need toseek funding from these relationship lendersas well.

Another way credit scoring may encouragelending to small businesses is by makingsecuritization of these loans more feasible.Securitization involves pooling together agroup of loans and then using the cash flowsof the loan pool to back publicly traded securi-ties; the loans in the pool serve as collateral forthe securities. The loan pool will typically havemore predictable cash flows than any indi-vidual loan, since the failure of one borrower

to make a payment can be offset by anotherborrower who does make a payment. The ex-pected cash flows from the loan pool determinethe prices of the securities, which are sold toinvestors. Securitization can reduce the costsof bank lending, since typically the loan poolis moved off the bank’s books to a third-partytrustee so that the bank need not hold capitalagainst the loans and the securities providewhat is often a cheaper source of funding thandeposits. (See Christine Pavel’s article for anoverview of securitization.)

Securitization has occurred with mortgageloans, credit card receivables, and auto loans,all of which tend to be homogeneous with re-gard to collateral, the loan terms, and the un-derwriting standards used. This homogeneityis important, since a crucial aspect ofsecuritization is being able to accurately pre-dict the cash flows from the pool of loans sothat the securities can be accurately priced.There have not been many securitizations ofsmall-business loans, partly because of theirheterogeneous nature.16 But credit scoring willtend to standardize these loans and make de-fault risk more predictable, steps that shouldmake securitization more feasible.17 As wastrue in the mortgage market, securitizationwould probably lead to an increase in small-

16In his 1995 article, Ron Feldman indicates that less than$900 million in small-business loans had been securitized,while $155 billion of these types of loans were outstandingat year-end 1994. He also provides descriptions of some ofthese securitizations.

17The difficulty that the borrower’s option to prepay amortgage poses for pricing mortgaged-backed securities isnot an issue for small-business loan securitizations, sincesmall-business loans have short maturities. For example,the November 1996 Survey of Terms of Bank Lending indi-cated that 85 percent of loans made in the survey periodeither had no stated maturity or a stated maturity of lessthan one year. The average maturity, weighted by loan size,of loans with stated maturities of longer than one day butless than a year was 64 days (Federal Reserve Bulletin).



business lending, with nonbank lenders play-ing a larger role. The market would becomemore liquid, since unlike loans, the securitiesare easily bought and sold; thus, diversifica-tion would be easier to achieve. Since diversi-fication lowers risk, loan rates could be lower.

CONCLUSIONWidespread securitizations of small-busi-

ness loans are still in the future. But credit scor-

ing is increasingly being used to evaluate small-business loan applications, something that wasnot widely anticipated a decade ago. Creditscoring will never be able to predict with cer-tainty the performance of an individual loan,but it does provide a method of quantifyingthe relative risks of different groups of borrow-ers. Scoring has the potential to be one of thefactors that change small-business banking aswe know it.

BIBLIOGRAPHY

Allen, James C. “A Promise of Approvals in Minutes, Not Hours,” American Banker (February 28,1995), p. 23.

Altman, Edward I., Giancarlo Marco, and Franco Varetto. “Corporate Distress Diagnosis: Compari-sons Using Linear Discriminant Analysis and Neural Networks (The Italian Experience),”Journal of Banking and Finance 18 (1994), pp. 505-29.

Altman, Edward I. and Anthony Saunders. “Credit Risk Measurement: Developments Over theLast 20 Years,” Journal of Banking and Finance, forthcoming 1997 (New York University SalomonCenter Working Paper S-96-40).

Asch, Latimer. “How the RMA/Fair, Isaac Credit-Scoring Model Was Built,” Journal of CommercialLending (June 1995), pp. 10-16.

Avery, Robert B., Raphael W. Bostic, Paul S. Calem, and Glenn B. Canner. “Credit Risk, Credit Scor-ing and the Performance of Home Mortgages,” Federal Reserve Bulletin 82 (July 1996), pp. 621-48.

Berlin, Mitchell. “For Better and For Worse: Three Lending Relationships,” Business Review, FederalReserve Bank of Philadelphia (November/December 1996), pp. 3-12.

DeZube, Dona. “Mortgage Scoring: Rules of Thumb,” Mortgage Banking (August 1996), pp. 51-57.

Fair, Isaac. “Low to Moderate Income and High Minority Area Case Studies,” Fair, Isaac and Com-pany, Inc. Discussion Paper (October 4, 1996).

Federal Reserve Bulletin, Table 4.23 Terms of Lending at Commercial Banks (February 1997), p. A68.

Feldman, Ron. “Will the Securitization Revolution Spread?” The Region, Federal Reserve Bank ofMinneapolis (September 1995), pp. 23-30.

Freddie Mac Industry Letter (July 11, 1995).

Hansell, Saul. “Need a Loan? Ask the Computer: ‘Credit Scoring’ Changes Small-Business Lend-ing,” New York Times (April 18, 1995), p. D1.

Loretta J. Mester

15


Lawson, James C. “Knowing the Score,” US Banker (September 1995), pp. 61-65.

Leonard, Kevin J. “The Development of Credit Scoring Quality Measures for Consumer CreditApplications,” International Journal of Quality and Reliability Management, 12 (1995), pp. 79-85.

Levonian, Mark E. “Changes in Small Business Lending in the West,” Economic Letter, Federal Re-serve Bank of San Francisco, Number 97-02 (January 24, 1997).

Malhotra, D.K., Rashmi Malhotra, and Robert W. McLeod. “Artificial Neural Systems in Commer-cial Lending,” The Bankers Magazine (November/December 1994), pp. 40-44.

McAllister, Patrick H., and John J. Mingo. “Commercial Loan Risk Management, Credit-Scoring,and Pricing: The Need for a New Shared Database,” Journal of Commercial Lending (May 1994),pp. 6-22.

McCorkell, Peter L. “Comment: Managing Credit Risk Means Getting the Right Mix,” AmericanBanker (March 20, 1996), p. 14.

“Mortgage Credit Partnership Project: 1996-1997,” Final Report, Federal Reserve Bank of San Fran-cisco, March 5, 1997.

Muolo, Paul. “Building a Credit Scoring Bridge,” US Banker (May 1995), pp. 71-73.

Nakamura, Leonard I. “Small Borrowers and the Survival of the Small Bank: Is Mouse Bank Mightyor Mickey?” Business Review, Federal Reserve Bank of Philadelphia (November/December1994), pp. 3-15.

Neill, David S., and John P. Danforth. “Bank Merger Impact on Small Business Services Is Chang-ing,” Banking Policy Report, The Secura Group, 18 (April 15, 1996), pp. 1 & 13-19.

Oppenheim, Sara. “Would Credit Scoring Backfire in a Recession?” American Banker (November 18,1996), p. 16.

Oppenheim, Sara. “Wider Rate Gap Between Small and ‘Small’,” American Banker (January 21, 1997),p. 10.

Oppenheim, Sara. “Gearing Up for Small-Business Push, PNC Building an Assembly Line,” Ameri-can Banker (May 27, 1997), p. 1.

Pavel, Christine. “Securitization,” Economic Perspectives, Federal Reserve Bank of Chicago (July/August 1986), pp. 16-31.

Prakash, Snigdha. “Mortgage Lenders See Credit Scoring as Key to Hacking Through Red Tape,”American Banker (August 22, 1995), p. 1.

Prakash, Snigdha. “Freddie Mac Exec Details Evolution of Credit Scoring,” American Banker (March6, 1997), p. 12A.

Racine, John. “Community Banks Reject Credit Scoring for the Human Touch,” American Banker(May 22, 1995), p. 12.



BIBLIOGRAPHY (continued)

Saunders, Anthony. Financial Institutions Management: A Modern Perspective, 2nd edition, Chapter10, Boston: Irwin: Boston, 1997.

Senior Loan Officer Opinion Survey on Bank Lending Practices, Federal Reserve System, November1996 and January 1997.

Talley, Karen. “Four-Lender Test Could Advance the Status of Credit Scoring,” American Banker(March 24, 1997), p. 12.

Wantland, Robin. “Best Practices in Small Business Lending for Any Delivery System,” Journal ofLending and Credit Risk Management (December 1996), p. 16-25.

Zuckerman, Sam. “Taking Small Business Competition Nationwide,” US Banker (August 1996), pp.24-28, 72.

Loretta J. Mester

17


What Determines the Exchange Rate:Economic Factors orMarket Sentiment?

Gregory P. Hopper*

Readers of the financial press are familiarwith the gyrations of the currency market. Nomatter which way currencies zig or zag, itseems there is always an analyst with a quot-able, ready explanation. Either interest rates arerising faster than expected in some country, orthe trade balance is up or down, or centralbanks are tightening or loosening their mon-etary policies. Whatever the explanations, the

underlying belief is that exchange rates are af-fected by fundamental economic forces, suchas money supplies, interest rates, real outputlevels, or the trade balance, which, if well fore-casted, give the forecaster an advantage in pre-dicting the exchange rate.

What is not so well known outside academiais that exchange rates don’t seem to be affectedby economic fundamentals in the short run.Being able to predict money supplies, centralbank policies, or other supposed influencesdoesn’t help forecast the exchange rate. Econo-mists have found instead that the best forecastof the exchange rate, at least in the short run,is whatever it happens to be today.

*When this article was written, Greg Hopper was a se-nior economist in the Research Department of the Philadel-phia Fed. He is now in the Credit Analytics Group at Mor-gan Stanley, Co., Inc., New York.



In this article, we’ll review exchange-rateeconomics, focusing on what is predictable andwhat isn’t. We’ll see that exchange rates seemto be influenced by market sentiment ratherthan by economic fundamentals, and we’ll ex-amine the practical implications of this fact.Sometimes, there are situations in which mar-ket participants may be able to forecast the di-rection but not the timing of the movement.We’ll also see that volatility of exchange ratesand correlations between exchange rates arepredictable, and we’ll examine the implicationsfor currency option pricing, risk management,and portfolio selection.

THE EXCHANGE RATE ANDECONOMIC FUNDAMENTALS

The earliest model of the exchange rate, themonetary model, assumes that the current ex-change rate is determined by current funda-mental economic variables: money suppliesand output levels of the countries. When thefundamentals are combined with market ex-pectations of future exchange rates, the modelyields the value of the current exchange rate.The monetary model might also be dubbed the“newspaper model.” When analyzing move-ments in the exchange rate, journalists oftenuse the results of the monetary model. Simi-larly, when Wall Street analysts are asked tojustify their exchange-rate predictions, they willtypically resort to some variant of the monetarymodel. This model is popular because it pro-vides intuitive relationships between the eco-nomic fundamentals and it’s based on standardmacroeconomic reasoning.

The reasoning behind the monetary modelis simple: the exchange rate is determined bythe relative price levels of the two countries. Ifgoods and services cost twice as much, on av-erage, in U.S. dollars as they do in a foreigncurrency, $2 will fetch one unit of the foreigncurrency. That way, the same goods and ser-vices will cost the same whether they arebought in the U.S. or in the foreign country.1

But what determines the relative price lev-els of the two countries? The monetary modelfocuses on the demand and supply of money.If the money supply in the United States rises,but nothing else changes, the average level ofprices in the United States will tend to rise.Since the price level in the foreign country re-mains fixed, more dollars will be needed to getone unit of foreign currency. Hence, the dollarprice of the foreign currency will rise: the dol-lar will depreciate--it’s worth less in terms ofthe foreign currency.

Money supplies are not the only economicfundamentals in the monetary model. The levelof real output in each country matters as wellbecause it affects the price level. For example,if the level of output in the United States rises,but other fundamental factors, such as the U.S.money supply, remain constant, the averagelevel of prices in the United States will tend tofall, producing an appreciation in the dollar.2

Future economic fundamentals also matterbecause they determine the market’s expecta-tions about the future exchange rate. Not sur-prisingly, market expectations of the futureexchange rate matter for the current exchangerate. If the market expects the dollar price ofthe yen to become higher in the future than itis today, the dollar price of the yen will tend tobe high today. But if the market expects thedollar price of the yen to be lower in the futurethan it is today, the dollar price of the yen willtend to be low today.

Here’s an example of how to use the mon-etary model: suppose we wanted to predict the

1When purchasing power parity holds, particular goodsand services cost the same amount in the domestic countryas they do in the foreign country. There is an extensive lit-erature that documents that purchasing power paritydoesn’t hold except perhaps in the very long run.

2In the monetary model, the price level must fall in thissituation to ensure that money demanded by consumers isthe same as money supplied by the central bank.

Loretta J. Mester

19


dollar-yen exchange rate. The first thing weneed to do is think about the relationships be-tween the fundamentals and the exchange rate.The monetary model implies that if the U.S.money supply is growing faster than the Japa-nese money supply, the dollar price of the yenwill rise: the dollar will depreciate and the yenwill appreciate. So, the analyst needs to assessmonetary policy in the two countries. The mon-etary model also implies that if output is grow-ing faster in the United States than it is in Ja-pan, the dollar price of the yen will tend to fall:the dollar will appreciate and the yen will de-preciate. Finally, the analyst must assess expec-tations about the future exchange rate. If themarket’s expectation of the future exchangerate were to change, the current exchange ratewould move in the same direction. When mak-ing an exchange-rate forecast based on themonetary model, the analyst must consider theeffect of all the fundamentals simultaneously.He can do this by using a statistical model orby combining judgment with the use of a sta-tistical model.

In practice, using the monetary model tomake exchange-rate forecasts is difficult be-cause the analyst never knows the true valueof the economic fundamentals. At any time,money supply and output levels are not knownwith certainty; they must be forecast based onthe available economic data. Of course, expec-tations about the future of the exchange rateare even harder to assess because these expec-tations are unobservable. The analyst can al-ways survey market participants about theirexpectations, but he can never be sure if thesurveys accurately reflect the market’s views.If we assume the monetary model is valid, thegoal of the successful exchange-rate forecasteris to predict the values of the fundamentalsbetter than the competition and then use themonetary model or some variant to derive fore-casts of the exchange rate.

The fatal flaw in this strategy is the assump-tion that the monetary model can be used to

successfully forecast the exchange rate once thevalues of the fundamentals are known. Al-though the monetary model had some earlysuccess, economists have established that themodel fails empirically except perhaps in un-usual periods such as hyperinflations.3 For onething, research did not establish a strong sta-tistical relationship between exchange rates andthe values of the fundamentals. Moreover, akey assumption of the model was found to befalse: the model assumes that the price levelcan move freely. Yet the price level seems tobe “sticky,” meaning that it moves very slowlycompared with the movement of the exchangerate.

What about other models? After the failureof the monetary model became apparent,economists went to work developing otherideas. Rudiger Dornbusch developed a vari-ant of the monetary model called the overshoot-ing model, in which the average level of pricesis assumed to be fixed in the short run to re-flect the real-world finding that many pricesdon’t change frequently. The effect of this as-sumption is to cause the exchange rate to over-shoot its long-run value as a result of a changein the fundamentals; eventually, however, theexchange rate returns to its long-run value.Ultimately, this model was shown to fail em-pirically: economists couldn’t find the strongstatistical relationships between the fundamen-tals and the exchange rate that should exist ifthe model were true.4

Another extension of the simple monetarymodel is called the portfolio balance model. Inthis approach, the supply of and demand forforeign and domestic bonds, along with the

3See the papers by Frenkel (1976, 1980), Bilson (1978),and Hodrick (1978) for empirical analysis of the monetarymodel.

4For an empirical treatment of the overshooting model,see the paper by Backus (1984).

What Determines the Exchange Rate: Economic Factors or Market Sentiment? Gregory P. Hopper



supply of and demand for foreign and domes-tic money, determine the exchange rate. Earlytests of the model were not very encouraging.5

Later, economists formulated a more sophisti-cated version of the portfolio balance model,in which investors were assumed to choose aportfolio of domestic and foreign bonds in anoptimal way. According to the more sophisti-cated portfolio balance theory, the degree towhich investors are willing to substitute do-mestic for foreign bonds depends on how muchinvestors dislike risk, how volatile the returnson the bonds are, and the extent to which thereturns on the different bonds in the portfoliomove together. Unfortunately, economists didnot find much empirical support for the moresophisticated version of the portfolio balancemodel.6

Economic News. Thus, the three major mod-els of the exchange rate—the monetary, theovershooting, and the portfolio balance mod-els—do not provide a satisfactory account ofthe exchange rate. Nonetheless, it is possiblethat news about the fundamentals affects the ex-change rate even if the fundamentals them-selves don’t influence the exchange rate in themanner suggested by the three major exchangerate models.

The news about the fundamentals can bedefined as the difference between what mar-ket participants expect the fundamentals to beand what the fundamentals actually are oncetheir values are announced. For example, mar-ket participants form expectations about thevalue of the money supply before the govern-ment announces the money supply figures, and

these expectations are translated into decisionsto buy or sell currency. These decisions ulti-mately help to determine the current level ofthe exchange rate. Once the government an-nounces the value of the money supply, mar-ket participants buy or sell currencies as longas the news is different from what they ex-pected. Thus, news about fundamentals, un-der this view, is an important determinant ofthe exchange rate.

The difficulty in testing this view is thateconomists don’t know how to measure thenews because they don’t know how to mea-sure the market’s expectations. One solutionis to assume that market participants form theirexpectations using a statistical device calledlinear regression. Using linear regression, aneconometrician could estimate the expectedlevel of a fundamental, such as the U.S. moneysupply, for each quarter during the past 20years. He could then subtract the value of theestimated expected money supply from its ac-tual value in each quarter to generate an esti-mate of the news about the quarterly U.S.money supply. The news for other fundamen-tals can be estimated in a similar way.

Once the econometrician has estimated eachfundamental’s news for each quarter duringthe last 20 years, he can check to see if it ex-plains the level of the exchange rate. Studiesby economists who have carried out this pro-cedure generally indicate that news about thefundamentals explains the exchange rate bet-ter than the three major exchange-rate mod-els.7 However, two factors make this resulthard to interpret. First, we have no direct evi-dence suggesting that market participants formtheir expectations using linear regression mod-els or that they form their expectations as ifthey were using these models. Second, these

5See, for example, the paper by Branson, Halttunen, andMasson (1977).

6See the papers by Frankel (1982) and Lewis (1988) forempirical analysis of the more sophisticated portfolio bal-ance model. The fundamental problem with the model isthat investors must have an implausibly high aversion torisk to explain the exchange rate.

7For empirical analysis of news models, see the papersby Branson (1983), Edwards (1982, 1983), and MacDonald(1983).

Loretta J. Mester

21


studies use the final values of the fundamen-tals, values released by governments months,if not years, after the forecasts were made. Yet,forecasters must use the government’s prelimi-nary estimates of the fundamentals when theymake their predictions. In other words, theeconometrician is assuming that market par-ticipants are making forecasts using informa-tion they don’t have. Hence, the result thatnews about the fundamentals seems to explainthe level of the exchange rate better than themodels is hard to interpret.

One way to avoid the problem of using fi-nal values of fundamentals is to collect the ini-tial estimates from newspapers, governmentannouncements, and wire services and exam-ine their ability to affect the level of the ex-change rate. Studies that have done this havefound that announcements about fundamentalsaffect the exchange rate only in the very shortrun: the effects of announcements generallydisappear after a day or two.

When we look at the evidence from the threemajor exchange-rate models, from the newsanalysis, and from the effects of announce-ments, it is hard not to be pessimistic aboutthe fundamentals’ ability to explain the ex-change rate. But the evidence we have exam-ined so far is backward-looking: the fundamen-tals don’t seem to explain exchange-rate behav-ior over the past couple of decades. However,we can also do a forward-looking analysis: dothe fundamentals help us forecast the level ofthe exchange rate?

The surprising answer to this question, givenby economists Richard Meese and KennethRogoff in the early 1980s, is no. Meese andRogoff examined the ability of the fundamen-tals to predict the level of the exchange rate forhorizons up to one year. They considered fun-damentals-based economic models as well asstatistical models of the relationship betweenthe fundamentals and the exchange rate thatdid not incorporate economic assumptions.They found that a naive strategy of using today’s

exchange rate as a forecast works at least aswell as any of the economic or statistical mod-els. Worse, they found that when they endowedthe economic or statistical models with finalvalues of the fundamentals—giving the mod-els an advantage that forecasters could notpossibly match—the naive strategy still wonthe forecasting contest. Despite many attemptssince the publication of Meese and Rogoff’sresults, economists have not convincingly over-turned their findings.

Thus, if we look backward or forward overperiods of up to a year, the fundamentals don’tseem to explain the exchange rate, contrary towhat standard models in international financetextbooks imply. But this result might be dis-missed by claiming that only the models testedhave failed to explain the exchange rate. Per-haps economists will discover a model thatworks in the future.

Although a fundamentals-based model thatworks is a possibility, evidence from othercountries suggests otherwise. In the EuropeanExchange Rate Mechanism (ERM), exchangerates between major European currencies arekept relatively stable by the countries’ centralbanks. If fundamentals are closely associatedwith the currencies, they should be stabilizedas well. However, when we examine Europeanfundamentals, we find that they fluctuate aboutas much as do the fundamentals ofnonstabilized currencies, such as the U.S. dol-lar. Hence, the evidence from the Europeanexperience does not suggest a close connectionbetween the fundamentals and the exchangerate, leading one to suspect that no fundamen-tals-based model will predict the short-runexchange rate.8

It’s possible that the fundamentals really doexplain the exchange rate, but we can’t see therelationship because we can’t observe the truefundamentals. Perhaps if economists discov-

8See Rose (1994) for a detailed discussion of this point.




ered different economic models that use fun-damentals other than money supplies and realoutput levels, the exchange rate could still beexplained in terms of basic economic quanti-ties. For example, some economic models im-ply that the true fundamentals are businesstechnologies and tastes and preferences of con-sumers. However, the evidence from Europeancountries renders this potential solution im-plausible. According to such a model, stabili-zation of European currencies in the ERM cor-responds to stabilization of the true fundamen-tals. But why should business technologies andtastes and preferences of consumers change lessin Europe than they do in the United States?At present, economists have found no evidenceto suggest they do and, indeed, have little rea-son to suppose that they will ever find suchevidence.

THE ALTERNATIVE VIEW:MARKET SENTIMENT MATTERS

The alternative view is that exchange ratesare determined, at least in the short run (i.e.,periods less than two years), by market senti-ment. Under this view, the level of the exchangerate is the result of a self-fulfilling prophecy:participants in the foreign exchange marketexpect a currency to be at a certain level in thefuture; when they act on their expectations andbuy or sell the currency, it ends up at the pre-dicted level, confirming their expectations.

Even if exchange rates are determined bymarket sentiment in the short run, the funda-mentals are still important, but not in the com-monly supposed way. From reading the news-papers, we know that market participants takethe fundamentals very seriously when form-ing exchange-rate expectations. Thus, if wewish to understand the level of the exchangerate, we need to know the values of the funda-mentals and, more important, how market par-ticipants interpret those levels. However, theevidence we reviewed shows no pattern ornecessary connection between the fundamen-

tals and the level of the exchange rate. Whenmarket participants use the fundamentals toform expectations about the exchange rate, theydon’t use them in any consistent way that couldbe picked up by an economic or statisticalmodel. As we have seen, we can do as wellforecasting the exchange rate by quotingtoday’s rate.

Although the naive forecast is at least as ac-curate as statistical or model-based forecasts,it’s still not very good. It’s just that statisticalor model-based forecasts are so bad that eventhe naive forecast can do at least as well. Howcan we improve our forecast? Unfortunately,economists are just starting to build models ofmarket sentiment, so we can’t get much guid-ance from economic theory just yet. Nonethe-less, we know that exchange rates are likelydetermined by market sentiment, so it seemsreasonable to try to understand the psychol-ogy of the foreign exchange market to improveforecasts of the short-run exchange rate.

To understand the psychology of the foreignexchange market, we need to know about thevarious economic theories. Even if they aren’tvery accurate, their implications may still in-fluence expectations in the market, althoughwe would not expect any particular model tohave any consistent influence. We also need tofind out what the market is thinking. Probablythe best way to do so is to be an active partici-pant in the foreign exchange market and to talkto other participants to learn which events theythink are important for a particular currency’soutlook. These events might be announce-ments of fundamentals, political events, orsome other factors. The analyst could thenconcentrate on forecasting those events. Ofcourse, there will probably be no pattern towhich events are important. For example, theU.S. budget deficit may well be important forthe dollar one year and unimportant the next.

Speculative Attacks. In some cases, theforecaster might be able to make a reasonableguess about the direction of the exchange rate’s

Loretta J. Mester

23


movement, even if he can’t be precise aboutthe timing. As an example, let’s review whathappened to the exchange rate between theSwedish krona and the German deutsche markin the early 1990s.

Sweden applied to enter the ERM in May1991 in a bid to stabilize its currency. To stabi-lize the krona-deutsche mark exchange rate,interest rates in Sweden and Germany had tobe the same. Therefore, the Swedish and Ger-man central banks couldn’t independently usemonetary policy—that is, change short-terminterest rates—if they wanted to keep the ex-change rate stable.9 If Sweden wanted to actindependently, it had to use fiscal policy (taxand government spending policies) to stimu-late the country’s growth rate.

However, a weak Swedish economy pro-voked speculators, who mounted an attack onthe krona in September 1992. Speculators knewthat the weak economy would tempt Swedento abandon its fixed exchange rate and usemonetary policy to cut short-term interest rates,especially since the new Swedish governmentwas adopting restrictive fiscal policy. Specula-tors believed that if the Swedish central bankcut the short-term interest rate, the kronawouldn’t be as attractive to investors. Thus,the speculators thought that after interest rateswere cut, the currency would depreciate withrespect to other ERM currencies. But sincespeculators expected the depreciation to hap-pen, they decided to sell the currency immedi-ately, i.e., mount a speculative attack on thecurrency.

This attack put the Swedish central bank inan uncomfortable position. To combat thecurrency’s depreciation, the central bank raisedshort-term interest rates temporarily to repel

the speculative attack—exactly the policy itdidn’t want in the face of sluggish economicgrowth. In fact, the Swedish central bank raisedthe short-term interest rate to an astonishing500 percent and held it there for four days.10

The speculators were deterred, but not forlong. The speculators understood that theSwedish central bank had to raise short-terminterest rates temporarily to support the cur-rency. But they were betting that the centralbank wouldn’t fight off the attack for long, es-pecially in the face of disquiet in the countryresulting from weak economic growth and thehigher interest rates needed to fight the specu-lative attack. The high short-term interest rateshad made the economic situation in Swedeneven more precarious, so, in November, thespeculators attacked again, selling the kronain favor of other ERM currencies. This time theSwedish central bank did not aggressively raiseinterest rates and the krona depreciated.

Profit opportunities such as this one cansometimes be exploited by speculators whorecognize that a country’s exchange-rate policyis inconsistent with the monetary policyneeded, given a country’s domestic situation.By paying careful attention to a country’s eco-nomic and political developments, a specula-tor can sometimes forecast the direction of a

9If a central bank can’t change the short-term interestrate independently, it can’t use monetary policy indepen-dently to stimulate the economy. Hence, countries with sta-bilized exchange rates must give up the independent use ofmonetary policy.

10If speculators expect the value of the currency to fall,and they are right, speculators can profit by selling the cur-rency short. As an example, suppose a speculator antici-pates that the value of the Swedish krona with respect tothe deutsche mark will fall in one week. The speculatorcould borrow krona and sell them for deutsche marks atthe current exchange rate. If the speculator is correct andthe krona does depreciate, at the end of the week the specu-lator can buy back the krona for fewer deutsche marks thanhe sold them for. Provided the krona fell enough over theweek, the speculator can repay the loan with interest andmake a profit in deutsche marks. However, if the centralbank makes short-term interest rates high enough, it canmake this transaction unprofitable. Thus, one defenseagainst a speculative attack is to dramatically raise short-term interest rates.




currency’s move when it breaks out of a stabi-lized exchange rate system. But the timing isnot easily forecast; it is probably determinedby market sentiment.11

WHAT ABOUT TECHNICAL RULES?Many market participants don’t rely on the

fundamentals. Instead, they use technical rules,which are procedures for identifying patternsin exchange rates. A simple technical rule in-volves looking at interest rates in two coun-tries. Suppose the first country is the UnitedStates and the second is Canada. If the one-month U.S. interest rate is higher than the one-month rate in Canada, the U.S. dollar will tendto appreciate with respect to the Canadian dol-lar. But if the one-month Canadian interest rateis higher, the U.S. dollar will tend to depreci-ate with respect to the Canadian dollar. Econo-mists and foreign exchange participants haveoften noted this fact.12

Indeed, it is possible to make money, on av-erage, by using this rule. The problem is thatimplementing this rule carries risk. There is anongoing debate about how big this risk is, andwhether the average profits are explained bythe level of risk. After all, it would not be sur-prising that the market pays a premium to thosewilling to assume substantial risk. Further-more, the profits may have occurred only bychance and may not recur. Sometimes, econo-mists report other technical rules that seem tomake money in the foreign exchange market.13

However, the considerations noted in the in-terest-rate differential rule apply to any tech-

nical rule. Even if the rule makes profits onaverage, the profits might be explained by thelevel of risk assumed in applying the rule.Moreover, the profits may well disappear whenwe account for technical statistical problems.Since economists are undecided at presentabout whether technical rules really do makemoney, it seems prudent to be cautious whenevaluating the merits of any such rule.

WHAT ABOUTLONG-RUN FORECASTING?

Even though economic models or the fun-damentals don’t help us understand the ex-change rate in the short run (except to the ex-tent that they influence market psychology),there is evidence that models do better in thelong run. For example, economists MartinEichenbaum and Charles Evans report thatcurrencies react as theory would suggest tounanticipated movements in the money sup-ply, but only in the long run, after a period ofabout two years. Standard monetary theorieswould imply that an unanticipated decline inthe U.S. money supply would lead to an ap-preciation of the dollar with respect to othercurrencies. Eichenbaum and Evans found thatthe dollar does, in fact, appreciate in responseto an unanticipated monetary contraction;however, the full effects on the dollar are notregistered until two years after the contraction,suggesting that models may well work in ex-plaining the exchange rate in the long run.14

IS ANY ASPECT OF THEEXCHANGE RATE PREDICTABLEIN THE SHORT RUN?

Although the level of the exchange rate inthe short run is not very predictable, volatili-ties and correlations of currencies are much

11For further discussion of the myriad problems that canarise when countries attempt to fix their exchange rates,see the article by Obstfeld and Rogoff (1995).

12See my 1994 Business Review article for a nontechnicaldiscussion.

13For an example, see Sweeney (1986).

14For further evidence on the effects of unanticipatedmonetary contractions on the exchange rate, seeSchlagenhauf and Wrase (1995).

Loretta J. Mester

25


more predictable. The daily volatility of a cur-rency measures the extent to which thecurrency’s value in terms of another currencyfluctuates each day. The value of high-volatil-ity currencies fluctuates more each day thanthat of low-volatility currencies. Correlationsmeasure the extent to which currencies movetogether. In general, volatilities and correlationsvary with time, rising or falling each day in asomewhat predictable way.

The time-varying nature of the daily vola-tility of the dollar in terms of the deutsche markcan be seen in the figure. Notice that, in 1991,days on which the volatility of the dollar is hightend to cluster together, and in 1990, days withlower volatility follow one another. Since dailyvolatility clusters together, it is predictable. Ifwe want to predict tomorrow’s volatility, weneed only look at the recent past. If daily vola-tility has been high over the recent past, wecan be reasonably sure that it will be high to-morrow.

This idea forms the basis for statistical mod-

els of a currency’s volatility. The GARCHmodel, developed by economist Tim Bollerslev,who built on work by economist Robert Engle,uses the volatility-clustering phenomenon topredict future volatility. In essence, a GARCHmodel measures the strength of the relation-ship between recent volatility and current vola-tility. Once this strength is known, it can be usedto forecast volatility. GARCH models havegood empirical support for exchange rates andare being used in practical applications in theforeign exchange market.15

GARCH models can be extended to handletwo or more currencies, and they can measurethe strength of recent correlations in predict-

Daily Percent Dollar Return on Deutsche Mark


15GARCH stands for Generalized Autoregressive Con-ditional Heteroskedasticity. For the technical details of howGARCH models work, see Bollerslev (1986). Examples oftechnical applications of GARCH models of exchange ratesinclude Bollerslev (1990) and Kroner and Sultan (1993).Heynen and Kat (1994) use GARCH to forecast volatility.

4

3

2

1

0

-1

-2

-3

-41990 1991

Percent



ing current ones. Once this strength is under-stood, it can be used to forecast correlations.

USES OF VOLATILITY ANDCORRELATION FORECASTS

Volatility and correlation forecasts have im-portant uses in finance. First, currency deriva-tives, securities whose value depends on thevalue of currencies, require measures of vola-tility and sometimes correlations to price them.GARCH models can supply estimates of thesevolatilities and correlations. Second, volatili-ties of individual currencies coupled with cor-relations between currencies can be combined

Using GARCH to Measure Portfolio Risk

Here, we illustrate the use of a GARCH model to manage risk in a simple portfolio of twocurrencies, the yen and the deutsche mark. Using daily data on the yen and the deutsche mark fromJanuary 2, 1981, to June 30, 1996, the time-varying volatilities and correlations were estimated usingEngle and Lee’s (1993a,b) GARCH model. Suppose we have a portfolio with $1 million invested inyen and $1 million invested in deutsche marks. Then we can calculate the value at risk (VaR) of theportfolio. The VaR is the maximum loss the portfolio will experience a certain fraction of the timeduring a specific period. For example, we can see from the table that daily VaR at the 95 percentconfidence level is $12,000. That means that 95 percent of the time, the largest daily loss on theportfolio will be $12,000. But 5 percent of the time, the loss will be bigger, sometimes by a substan-tial amount. The daily loss measures the difference between the value of the portfolio at the end ofone trading day and its value at the end of the next trading day.

As another example, consider weekly VaR at the 98 percent confidence interval. The numbersindicate that 98 percent of the time, the loss over five trading days will not exceed $35,000. But 2percent of the time, the losses will be bigger. See Hopper (1996) for more discussion.

Value at Risk of a Currency Portfoliowith $1 Million Invested in Both Yen and Deutsche marks

One-Day Horizon Five-Day Horizon95 percent $12,000 $27,00098 percent $15,000 $35,00099 percent $18,000 $41,000

These numbers for the value at risk apply to the risk in the portfolio on July 1, 1996, the day after theend of the data period. However, the reason for using a GARCH model is that volatility varies overtime. The value at risk would be higher in times of greater volatility and lower when the market isless volatile.

to determine the volatility of a portfolio of cur-rencies. Since the volatility of a portfolio mea-sures the extent to which the portfolio’s valuefluctuates, the volatility can be used to assessa portfolio’s risk. Portfolios with higher vola-tilities are riskier because they have a tendencyto lose more per day—or gain more per day—than do portfolios with lower volatilities (seeUsing GARCH to Measure Portfolio Risk). Finally,knowledge of volatilities and correlations canhelp an investor choose the proportions of eachcurrency to hold in a portfolio. For example,knowing a portfolio’s volatilities and correla-tions may show an investor how to rearrange

Loretta J. Mester

27

What's the Point of Credit Scoring?What Determines the Exchange Rate: Economic Factors or Market Sentiment? Gregory P. Hopper

the proportions of currencies in a portfolio sothat he has the same return, on average, but alower risk of loss.

CONCLUSIONThe evidence discussed in this article sug-

gests that economic models and indeed fun-damental economic quantities are not very use-ful in explaining the history of the exchangerate or in forecasting its value over the nextyear or so. This fact has important implicationsfor market participants. It is all too commonto encounter private-sector foreign exchangeeconomists who tell very cogent stories de-signed to buttress their short-term forecasts forthe values of currencies. These stories are of-ten based on plausible economic assumptionsor models. These economists hope that marketparticipants will act on their forecasts and tradecurrencies. However, if these forecasts are jus-tified by a belief that economic models or fun-damentals influence the exchange rate in theshort run, it’s likely they are not very good.Indeed, we have seen that these forecasts willprobably be outperformed by the naive fore-cast: tomorrow’s exchange rate will be what itis today.

On the other hand, to the extent that theseforecasts reflect market sentiment or a self-ful-filling prophecy, they may be useful. Unfortu-nately, it is difficult to judge when this is thecase. The difficulty is accentuated by theunobservability of market expectations. A fore-caster might be using a model he believes in,and his forecast might turn out to be correct ifthe market also temporarily believes the im-plications of the model. But it is hard, if notimpossible, to know what the market expects;hence, it is hard to judge the merits of a fore-cast.

Fortunately, the situation is better regard-ing volatilities and correlations, which followpredictable patterns. The GARCH model andits more sophisticated variants can be used toprice derivatives, assess currency portfolio risk,and set allocations of currencies in portfolios.Economists are continually discovering newempirical facts about volatility and correlations.No doubt the GARCH model will eventuallybe supplanted by an alternative, but for now,economists will use the GARCH model, orsome variation of it, to forecast volatilities andcorrelations of currencies.

REFERENCES

Backus, David. “Empirical Models of the Exchange Rate: Separating the Wheat from theChaff,” Canadian Journal of Economics (1984), pp. 824-46.

Bilson, John. “The Monetary Approach to the Exchange Rate—Some Empirical Evidence,”IMF Staff Papers, 25 (1978), pp. 48-75.

Bollerslev, Tim. “Generalized Autoregressive Conditional Heteroskedasticity,” Journal ofEconometrics, 31 (1986), pp. 307-27.

Bollerslev, Tim. “Modelling the Coherence in Short-Run Nominal Exchange Rates: A Mul-tivariate Generalized ARCH Model,” Review of Economics and Statistics, 72 (1990), pp.498-505.



REFERENCES (continued)

Branson, William. “Macroeconomic Determinants of Real Exchange Rate Risks,” in R.J.Herring, ed., Managing Foreign Exchange Risk. Cambridge, U.K.: Cambridge Univer-sity Press, 1983.

Branson, William, Hannu Halttunen, and Paul Masson. “Exchange Rates in the Short Run:The Dollar-Deutschemark Rate,” European Economic Review, 10 (1977), pp. 303-24.

Dornbusch, Rudiger. “Expectations and Exchange Rate Dynamics,” Journal of PoliticalEconomy, 84 (1976), pp. 1161-76.

Edwards, Sebastian. “Exchange Rates and News: A Multi-Currency Approach,” Journal ofInternational Money and Finance, 1 (1982), pp. 211-24.

Edwards, Sebastian. “Floating Exchange Rates, Expectations, and New Information,” Journalof Monetary Economics, 11, (1983), pp. 321-36.

Eichenbaum, Martin, and Charles Evans. “Some Empirical Evidence on the Effects of Mon-etary Policy Shocks on Exchange Rates,” NBER Working Paper 4271 (1993).

Engle, Robert F. “Autoregressive Conditional Heteroskedasticity with Estimates of theVariance of U.K. Inflation,” Econometrica, 50 (1982), pp. 987-1008.

Engle R., and G. Lee. “A Permanent and Transitory Component Model of Stock ReturnVolatility,” Discussion Paper 92-44R, Department of Economics, University of Cali-fornia, San Diego (1993a).

Engle R., and G. Lee. “Long Run Volatility Forecasting for Individual Stocks in a OneFactor Model,” Discussion Paper 93-30, Department of Economics, University ofCalifornia, San Diego (1993b).

Frankel, Jeffrey. “In Search of the Exchange Rate Risk Premium: A Six Currency Test As-suming Mean-Variance Optimization,” Journal of International Money and Finance, 1(1982), pp. 255-74.

Frenkel, Jacob. “A Monetary Approach to the Exchange Rate: Doctrinal Aspects and Em-pirical Evidence,” Scandinavian Journal of Economics, 78 (1976), pp. 200-24.

Frenkel, Jacob. “Exchange Rates, Prices, and Money: Lessons From the 1920s,” AmericanEconomic Review, 70 (1980), pp. 235-42.

Loretta J. Mester

29


Heynen, Ronald C., and Harry M. Kat. “Volatility Prediction: A Comparison of the Sto-chastic Volatility, GARCH (1,1), and EGARCH (1,1) Models,” The Journal of Deriva-tives, 2 (1994), pp. 50-65.

Hodrick, Robert. “An Empirical Analysis of the Monetary Approach to the Exchange Rate,”in J. Frenkel and H.G. Johnson, eds., The Economics of Exchange Rates. Reading, Mass.:Addison Wesley, 1978, pp. 97-116.

Hopper, Greg. “Is the Foreign Exchange Market Inefficient?” Federal Reserve Bank ofPhiladelphia Business Review (May/June 1994).

Hopper, Greg. “Value at Risk: A New Methodology For Measuring Portfolio Risk,” Fed-eral Reserve Bank of Philadelphia Business Review (July/August 1996).

Kroner, Kenneth F., and Jahangir Sultan, “Time-Varying Distributions and Dynamic Hedgingwith Foreign Currency Futures,” Journal of Financial and Quantitative Analysis, 28 (1993),pp. 535-51.

Lewis, Karen. “Testing the Portfolio Balance Model: A Multi-lateral Approach,” Journal ofInternational Economics, 7 (1988), pp. 273-88.

MacDonald, Ronald. “Some Tests of the Rational Expectations Hypothesis in the ForeignExchange Markets,” Scottish Journal of Political Economy, 30 (1983), pp. 235-50.

Meese, Richard, and Kenneth Rogoff. “Empirical Exchange Rate Models of the 1970s: DoThey Fit Out of Sample?” Journal of International Economics, 14 (1983), pp. 3-24.

Obstfeld, Maurice, and Kenneth Rogoff. “The Mirage of Fixed Exchange Rates,” Journal ofEconomic Perspectives, 9 (1995), pp. 73-96.

Rose, Andrew. “Are Exchange Rates Macroeconomic Phenomena?” Federal Reserve Bankof San Francisco Economic Review, 1 (1994), pp. 19-30.

Schlagenhauf, Don, and Jeffrey Wrase. “Liquidity and Real Activity in a Simple OpenEconomy Model,” Journal of Monetary Economics, 35 (1995), pp. 431-61.

Sweeney, Richard J. “Beating the Foreign Exchange Market,” Journal of Finance (1986),pp. 163-82.


what's the point of credit scoring? loretta j. mester...*loretta mester is a vice president and...

Documents