Consumer Credit Data not Supportive of Management Decisions in the U.S. Apartment Industry

Transcription

1 Consumer Credit Data not Supportive of Management Decisions in the U.S. Apartment Industry Michael Furick, Assistant Professor of Marketing, Georgia Gwinnett College, USA ABSTRACT Purpose: Credit scoring summarizes a consumer s credit history and credit report into a three-digit number. This score is widely used by banks and insurance companies to identify and sort consumer behavior into categories based on their financial history and then predict financial behavior. This research analyzed the results of using six different commercially available credit scores, applied in one apartment complex, to the task of selecting applicants for apartment rental based on financial data. Approach: Six credit scores were obtained for each applicant for apartment rentals in one apartment complex over three years. Each of the scores were correlated with actual consumer rent payments following move-in. Findings: This research determined that these six scores in this application are not predictive of consumer financial performance and possible explanations are given. These include consumer understanding of the legal nature of the apartment lease, the specific development of each of the commercial scores, profile of this apartment complex plus others. Implications/Value: Apartment managers do not have a predictive method to summarize credit history. Rental acceptance or denial decisions therefore are made locally and can vary widely between similar apartment complexes even though The U.S. Fair Housing Act of 1968 and 1988 prohibits setting differing terms and conditions. A wider study can begin to develop a predictive model. INTRODUCTION AND BACKGROUND The banking and financial services industry has historically used credit report data and specifically, credit scoring as a means of determining the credit worthiness of consumers applying for loans. The intent is to weed out, or at least identify those applicants that will become questionable accounts while, at the same time, offer lower interest rates and better products to those applicants that are most desirable. Credit evaluation decisions are important for the financial institution involved due to the high risk and potential financial cost associated with a wrong decision (Piramuthu, 1998; FDIC, 2007). Several studies indicative that scores tend to be predictive of risk in the underwriting of both credit and insurance. Some studies suggest that consumers are the beneficiaries of lower credit costs and insurance premiums due to the use of credit scores (or at least credit costs and insurance premiums properly applied to consumer risk). Credit scoring is now widely used in a number of countries including United Kingdom, Germany, Norway, South Africa, Canada, Australia, Sweden, India and others. The credit scoring process generates a credit score, which is a three-digit number that predicts the likelihood that an applicant will repay a loan and repay it on time. This score is based on the data in a consumer s credit report and is the result of a process of modeling the variables important in the extension of credit. This modeling process is a statistical analysis of historical data for both good consumers and bad consumers, using certain financial variables that have been determined to be important in the evaluation of a consumer s financial strength and stability. These variables and the weighting of these The Journal of International Management Studies, Volume 11 Number 1, February,

2 variables change for each model and for differing industries. Typical variables used by the banking industry include the following (Leonard, 1996) 1. Number of bankruptcies. 2. Number of credit cards/trade line. 3. Percentage usage of these trade lines (percentage of credit limit). 4. Credit history and payment performance. 5. Length of employment. 6. Income. 7. Occupation. 8. Residential status. 9. Length of time at current address. The analysis of these variables for most credit models produces coefficients that are translated into weight scores. For example, if length of employment is longer than 10 years then add 50 points, if longer than 5 years add 25 points, otherwise add no points. Adding together these weight scores for each variable for each new loan applicant produces an overall score (other modeling techniques are also used). The loan officer/decision maker relies on this overall score in making the loan decision. A number of different consumer credit scores are commercially available (Table 1) Table 1: Sample of Commercially available credit scoring models Reference Number Name of Score 1 Crossview FICO National Risk Score (used by the apartment complex in this research and 2 tested in this research)). 3 National Equivalency Score. 4 Old National Risk Score. 5 FICO Installment Score. (tested in this research) 6 FICO Installment II Score. 7 FICO Automobile Score. 8 FICO Automobile II Score. 9 FICO Finance Score. (tested in this research) 10 FICO Finance II Score. 11 FICO Bankcard Score. 12 FICO Bankcard II Score. 13 FICO Mortgage Risk Score (sold by Equifax as Beacon ; also sold by the third credit bureau Transunion with the brand name Empirica ). (tested in this research) 14 MDS Bankruptcy II Score. 15 Bankruptcy Watch. 16 Retail Risk Score. 17 TEC Risk Score. 18 Collection Score. 19 Collection Recovery Score (bankcard). 20 Collection Recovery Score (retail). 21 FICO Advanced Risk Score. (tested in this research) 22 Fraud Shield. 23 Sureview Non Prime Score. (tested in this research) 24 Automobile Risk Score. 25 Credit Union Risk Score. 26 Tella Risk Score. 22 The Journal of International Management Studies, Volume 11 Number 1, February, 2016

3 Fair, Isaac Company (FICO) is the leading provider of these models and scores and has sold over 10 billion scores over the past 20 years. FICO estimates that over 90% of the mortgage decisions in the United States are based on one or more of its FICO credit scores (FICO, 2014). The FICO scores, and those from other companies, are available from the three U.S. credit bureaus when the loan officer orders a credit report and each credit bureau offers a different set of scores as part of its product offering. These scores range in price from about $0.25 to about $3.00 for each one obtained with the credit report. Each of the scores listed in Table 1 has its own scale and direction of the scale. Some of the scores have a scale of 0 to 1000, while others have a scale from 300 to 850. Some of the scores are developed so that a higher number is better, but for other scores a lower number is better. For example, the FICO National Risk Score (number 2 in Table 1) uses a scale from 0 to 1000 and a lower score indicates a better applicant. This is opposite of the typical score where a higher score indicates a better applicant. Figure 1 shows the typical statistics for the FICO Mortgage Risk Score that is number 13 on Table 1. This is the most widely used score in mortgage loan banking and this score has a range of 300 to 850 with a higher score indicating a better applicant. Source: Fair Isaac Corporation Figure 1: Graph of delinquency rates for the FICO Mortgage Risk Score. FICO defines the delinquency rate as the percentage of borrowers in a score range, who reach 90 days past due or worse (including bankruptcy or account charge-off) on any account on their credit report over a two year period (Fair, Issac & Company, 2014). The response of a lending institution to these scores is to deny a loan or to adjust the interest rate on a loan to reflect their increased risk. Some of the advantages in the use of credit scoring in financial transactions, particularly real estate/mortgage transactions are as follows: 1. Eliminating subjectivity- numeric scoring eliminates much of the subjectivity associated with the credit approval process and eliminates the need for the loan officer s gut feel, thus promoting a more consistent method of quantifying risk (Graves, 2000). 2. Reduced discrimination risk- quantifiable and consistent guidelines may eliminate discrimination in lending (Graves, 2000). The Journal of International Management Studies, Volume 11 Number 1, February,

4 3. Faster response time to the consumer s demand for credit- the loan application process is significantly speeded up. 4. Accuracy- the use of credit scoring appears to have a high degree of accuracy in financial/mortgage transactions. The success of credit scoring in the banking industry has caused it to spread to other industries, most notably the auto insurance industry. A survey by Conning and Company determined that more than 90% of the insurance carriers surveyed use credit data and credit scoring, such as the FICO credit score, in their new business process for automobile coverage (Jones, 2001). This credit scoring is part of the process in determining who will get auto insurance and at what price the auto policy will be issued. A study (Monahgan, 2000) matched credit histories to 170,000 auto policies. Those with the best credit scores had a loss ratio of 74.1% while those with the worst credit scores had a loss ratio of 118.6%. (An auto insurance loss ratio is the amount paid out for claims on a policy divided by the premiums collected from the consumer on that same policy. So a loss ratio of 74.1% means the insurance company paid out 74.1 cents for every dollar in premiums collected, a very profitable account.) (Figure 2) Source: Use of Credit Information by Insurers in Texas, Texas Department of Insurance, December 30, Figure 2: Graph of personal automobile insurer claim history vs credit score using the FICO Mortgage Risk Score. STATEMENT OF THE PROBLEM, NEED FOR THE STUDY AND RESEARCH QUESTIONS The apartment complex that was studied is a 181 unit apartment complex in an older, slow growth southeastern U.S. city. This apartment complex has been using the FICO National Risk Score (number 2 on Table 1) as part of its new applicant process. The opinion of the property manager is that,: the credit score is not very helpful in choosing applicants. It does not seem to accurately predict which applicants will honor their lease to the end. We seem to have just as many lease termination problems with people with good scores as we do with people with bad scores. (personal interview) There does not appear to be a standard applicant selection process in the apartment rental market but credit scoring does not seem to be widely used. Seven other apartment complexes contacted have varying methods of choosing applicants with only two using any type of credit scoring (see Table 2). Possibly credit scoring is not used because the lack of success experienced by the subject complex has 24 The Journal of International Management Studies, Volume 11 Number 1, February, 2016

5 also been experienced by other complexes (it is not a goal of this research to investigate this). In these seven apartment complexes, consumer information is used as a barrier to entry, that is, credit and criminal information is used primarily to reject applicants. Table 2: Factors affecting Applicant Selection at Seven Apartment Complexes Apartment complex contacted Factors affecting applicant selection 181 unit complex that is the subject FICO National Risk Score used (number 2 on Table1) of this research Applicant rejected if previous landlord problems Applicant rejected if previous landlord problems 96 unit complex in Baltimore, Maryland Credit scores not used 68 unit complex in Washington D.C. 395 unit complex in Chicago, Illinois 264 unit complex in the same city as subject complex 210 unit complex in Athens, Georgia 190 unit complex in Nashville, Tennessee Applicant rejected if previous landlord problems Credit score not used FICO Advanced Risk Score used, number 13 on Table 1; minimum score must be 675 (see range on Figure 1) which is the best 15% delinquency rate of U.S. consumers Applicant rejected if previous landlord problems Applicant rejected if previous landlord problems Credit scores not used Applicant rejected if previous landlord problems or bankruptcy Credit score not used Income minimum three times rent At least 80% satisfactory accounts Applicant rejected if previous landlord problems or bankruptcy Credit score not used Income minimum three times rent At least 80% satisfactory accounts DATA ANALYSIS OF CURRENTLY USED CREDIT SCORE (NATIONAL RISK SCORE) A number of descriptors can be used to describe an ideal tenant for an apartment complex but primarily these are a) honoring the 12-month term of the lease, b) paying rent on time, and c) social and living habits (i.e. problem neighbor?). However, the property managers interviewed all viewed honoring the 12-month lease as the most important descriptor as this has direct financial consequences, and they can manage most other issues. It seems reasonable that non-payment of rent or late payment of rent would be an equally important consideration. However, late or non-payment would result in an eviction from the apartment complex and thus these payment issues would be recorded as the tenant having stayed for less than the 12 month term of their lease. Additionally, 12 month or longer lease terms are desirable as operating expenses are lower as the term of the lease increases ( apartments do not have to be repainted, and re-cleaned as often). Consequently, this research correlated credit scores versus the number of months that the tenant lived at the apartment complex (with 12 months or longer being considered a success ). The first part of this study selected 60 tenants at random from those that leased an apartment and analyzed the tenant rental history correlated with National Risk Score (#2 on Table 1 and currently used by the subject apartment complex). This particular model creates a number score that directly The Journal of International Management Studies, Volume 11 Number 1, February,

6 corresponds to risk. Specifically, a score of 100 indicates that this applicant has a 10.0% probability that they will NOT fulfill their financial obligations. A score of 525 would indicate that this applicant has a 52.5% probability that they will NOT fulfill their financial obligations (essentially the higher the number for the National Risk Score, the higher the risk). This is opposite to the typical credit score that has a range calibrated so that as the score number gets higher, the risk gets lower. Two analyses of the data were performed. The first examined the score using the data in its entirety and the linear regression results are displayed in Figure 3 and Table 3. There is no relationship between the score and the tenant s performance in honoring the 12-month term of their lease as the R Square approaches zero, which implies no correlation. Linear regression was used because as score changes, the risk changes, and this relationship and the score value operate in a linear fashion. (Consumers without enough credit history to run the scoring model by the credit bureau are given a score of zero.) Figure 3:Linear regression result for National Risk Score using all the data. Table 3:Linear regression result for National Risk Score using all the data Summary Regression Statistics Multiple R R Square 3.99E-06 Adjusted R Square Standard Error Observations 62 ANOVA Df SS MS F Signif. F Regression Residual Total Coefficients Std. Error T Stat P-value Lower 95% Intercept E X Variable The Journal of International Management Studies, Volume 11 Number 1, February, 2016

7 A second analysis of the data was performed that examined the data subjects (applicants) when divided into two groups based on their fulfillment of the 12-month term of the lease agreement. Group one contained those applicants that fulfilled the lease term and stayed for 12 months (or longer) and the linear regression results are presented in Figure 4 and Table 4. Group one represents the desirable applicants that the score should identify. Group two contained those applicants who stayed less than 12 months (i.e. not desirable tenants) and the linear regression results are presented in Figure 5 and Table 5. No correlation exists between the score and length of stay for either group one or group two. R Square for both groups is very low:.08 for group two and near zero for group one. This lack of correlation can also be seen when examining the percentage of applicants in the data set as shown in Table 6. Figure 4:National Risk Score results for tenants who satisfied the lease. Table 4:Linear Regression Results for National Risk Score for tenants who satisfied the lease Group One (rented 12 months or longer) Summary Regression Statistics Multiple R R Square Adjusted R Standard Error Observations 27 ANOVA Df SS MS F Signi. F Regression Residual Total Coefficients Std.Error t Stat P-value Lower 95% Intercept E X Variable The Journal of International Management Studies, Volume 11 Number 1, February,

8 Figure 5:National Risk Score results for tenants who did not satisfy lease. Table 5:Linear Regression Results for National Risk Score for tenants who did not satisfy the lease Group Two (rented less than 12 months) Summary Regression Statistics Multiple R R Square Adjusted R Square Standard Error Observations 17 ANOVA Df SS MS F Signif. F Regression Residual Total Coefficients Std.Error Error t Stat P-value Lower 95% Intercept E X Variable Score Number of Applicants Table 6: Distribution of tenants tested with National Risk Score. Perceived Risk Level Predicted Number who Rented 12 by Credit Score Months or more Number who Rented less than 12 Months None 1 to to 500 Over Unknown Low Medium High 7 (43%) 15 (63%) 9 (60%) 3 (60%) 9 (57%) 9 (37%) 6 (40%) 2 (40%) 28 The Journal of International Management Studies, Volume 11 Number 1, February, 2016

9 The number of applicants that rented for at least the 12-month term of their lease is about 60% whether the applicant score was in the preferred range of 1 to 200 (63%), the range of 210 to 500 (60%) or in the highest range over 500 (60%) as shown in Table 6. If this credit scoring model was accurately working as a predictor, the tenants in the preferred score range of 1 to 200 should have had significantly better results, which was not the case. One clear result from the data is that applicants without enough credit history to run a score represented the highest business risk since 57% of this group (9 of 16) stay for less than 12 months versus 38% (17 of 44) of those with a score that stay less than 12 months The National Risk Score, which has been used by this apartment complex for several years, is not helpful in choosing applicants for apartment rentals. There is no correlation between the credit score of an applicant and an applicant s honoring of their 12-month lease and no predictive value. ANALYSIS OF FIVE ADDITIONAL COMMERCIALLY AVAILABLE CREDIT RISK SCORES This second part of the study selected 111 tenants at random from those that leased an apartment and performed the same analysis as previously discussed for the National Risk Score. These 111 were selected from a pool of about 300. Of these 111 files, 83 were used in the analysis. The remaining 28 could not be used for various reasons: 16 had multiple persons on the lease so the financial obligation and score were unclear, five had no scores or limited scores, and seven had never actually moved in or no move-in data could be found. The management of the credit bureau involved in this research suggested these five scores as possibly the best choices for use in the apartment rental industry and the following five credit scores were tested. 1. Sureview Non Prime Score (number 23 on Table 1). This is a risk assessment tool developed by Experian specifically designed for non-prime bankcard issuers. It was developed to make predictions for five major classifications of consumers: 1) thin credit history and a limited number of derogatory trade accounts 2) young, full credit history and may have a limited number of derogatory trade accounts 3) mature, full credit history and may have a limited number of derogatory trade accounts 4) a high percentage of delinquencies or a bankruptcy on file 5) a high percentage of delinquencies or a bankruptcy on file and at least one of the delinquencies is recent. These classes of consumers tend to be apartment rental applicants. 2. FICO Mortgage Risk Score (number 13 on Table 1). This model uses an in- depth review of the information in a consumer s credit file and attempts to identify customers most likely to result in serious delinquency, charge-offs and bankruptcy. This model is also sold by Equifax under the brand name Beacon, and is one of the most widely used consumer credit score for mortgage loan applications. 3. Fair Issac Advanced Risk Score (number 21 on Table 1). This model helps determine which accounts are most likely to be profitable and which pose the greatest credit risk. It predicts the probability of serious derogatory credit behavior and indicates the likelihood that a customer will become seriously delinquent within the next 24 months (most apartment renters tend to a shorter term of 12 to 18 months). 4. FICO Installment Loan Score (number 5 on Table 1). This model predicts a consumer s performance on repaying short-term installment loans such as 36-month car loans or other leases. This type of financial transaction may be similar to the apartment rental decision. The Journal of International Management Studies, Volume 11 Number 1, February,

10 5. Fair Issac Finance Score (number 9 on Table 1). This model predicts a consumer s financial performance for loans originated at non-traditional finance companies, cash your paycheck here companies, or pawnshop-type lending businesses. These are short-term loans usually made to high-risk borrowers. All five additional credit scores were generated for each of the 83 tenants used in the analysis. As previously described, the data for each score was examined twice. First, each score was compared to all the data then, second, the data was separated into two groups of applicants: those who satisfied the lease and those who did not satisfy the lease.. The summary results are illustrated in Table 7. The low R Square results indicated that there is no relationship between the score and the length of time that the tenant honored the lease for any of the five additional scores, indicating no predictive value. Table 7 Summarized results of linear regression testing of all six commercial scores All Applicants Applicants who honored lease Applicants who did not honor Score Name R Square Mean Score R Square Mean Score R Square National Risk Score FICO Risk Score FICO Advanced Risk FICO Installment Loan FICO Finance Score Experian Sureview DISCUSSION Six commercially available credit scores were tested and it was determined that these scores were not predictive of tenant behavior in honoring of the lease. The problem with predictability of the six models may be based on the composition of the credit scoring model (statistical modeling issues). Specifically, these models were developed for other purposes such as home ownership and, when working accurately, tend to filter out the typical tenants for apartment rentals (i.e. younger in age, less time on the job, lower paying job, and so forth). The weighting of the variables used in the model creation is targeted to answer or predict a different longer-term consumer behavior question than is being asked in the apartment rental market. Second, these commercial scores may not be predictive of a tenant honoring the lease possibly because of a fundamental human behavior difference in the apartment rental decision that makes this decision different from other financial decisions (particularly the home buying decision). Possible causes could be the perceived short term nature of an apartment rental, or a misunderstanding on the part of the tenant of the strength of the lease as a legal document (for example, renting a car is a legal transaction that incurs no penalty when the car is returned early), or that the apartment decision is driven by lifestyle choices (children s school, loss of job, and so forth) rather than credit financial choices. Third, applicants of different socio-economic backgrounds likely rent apartments for a wide variety of reasons and these reasons may vary by type and location of apartment complex. Since only one apartment complex was tested, there could be a geographic bias or economic bias (this was a lower income complex) that was not apparent in this test. However, all the models tested were national models so this may not apply. Lastly, a lack of tenant understanding about the financial and legal implications of breaking an apartment lease may be one of the root causes. The decision to abandon an apartment before the lease is up may be a decision likely based not entirely on credit matters, but instead based heavily on lifestyle 30 The Journal of International Management Studies, Volume 11 Number 1, February, 2016

11 issues (loss of job, change of school for children and so forth). However, not honoring the lease for the full 12 month term can be noted on the tenant s national credit file as an unpaid debt and this can affect the tenant s future ability to obtain credit. This may not be widely understood by tenants. IMPLICATIONS Apartment managers do not have a predictive method to summarize a consumer s credit history into a meaningful and repeatable decision support system. As a result, the rental acceptance or denial decisions are made locally and decision criteria can vary widely among similar apartment complexes. In order to remain legal, the apartment complex must apply its decision criteria in a consistent, nondiscretionary manner without setting differing terms and conditions and be able to prove that it is doing so. Title VIII of the Civil Rights Act of 1968 (Fair Housing Act), as amended in 1988, prohibits discrimination in the rental of dwellings based on race, color, national origin, religion, sex, familial status (including children under the age of 18 living with parents or legal custodians, pregnant women, and people securing custody of children under the age of 18), and disability. Nearly forty years after Congress enacted the Fair Housing Act, millions of complaints are still filed each year. Housing discrimination doesn't always mean having a door slammed or a bigoted remark. Unsuspecting renters may be politely turned away from the housing of their choice, even though they are qualified. Without a meaningful model of credit data to direct local management acceptance/rejection of tenant applicants, apartment complexes may implement variable management decisions and thus be vulnerable to claims of discrimination. A wider study involving a number of apartment complexes can begin to develop a predictive model for the U.S. Apartment Complex industry. REFERENCES Avery, R.B., Calem, P.S., Canner, G.B., Bostic, R.W. (2003). An Overview of Consumer Data and Credit Scoring. Federal Reserve Bulletin, February, 2003, Board of Governors of the Federal Reserve System (2007). Report to the Congress on Credit Scoring and Its Effects on the Availability and Affordability of Credit. Submitted to the Congress pursuant to section 215 of the Fair and Accurate Credit Transactions Act of Fair, Issac & Company (2014). Understanding your credit score. Fair, Issac & Company consumer brochure. San Mateo, CA: Fair Issac & Company Documents. F.D.I.C. (2007). Scoring and Modeling. Risk Management Examination Manual for Credit Card Activities, VIII, March, Graves, D. (2000). The future of lending: Automated credit decisions in zero time. Business Credit, 102(9), Hartwig, R.P. (2005). No Evidence of Disparate Impact in Texas Due to Use of Credit Information by Personal Lines Insurers. Insurance Information Institute White Paper, January, Jones, W. (2001). Study: Insurance scoring becomes more favored. Best s Review, 102(5), 16 Leonard, K. (1996). Information systems and benchmarking in the credit scoring industry. Benchmarking for Quality Management & Technology, 3(1), Monahgan, J. (2000). The impact of personal credit history on loss performance in personal lines. Conference on ratemaking of the Casualty Actuarial Society, 3, Piramuthu, S. (1998). Feature selection for financial credit risk evaluation decisions. Operations and Information Management, 98(11), The Journal of International Management Studies, Volume 11 Number 1, February,