The full text on this page is automatically extracted from the file linked above and may contain errors and inconsistencies.
Federal Reserve Bank of Chicago Does It Pay to Read Your Junk Mail? Evidence of the Effect of Advertising on Home Equity Credit Choices Sumit Agarwal and Brent W. Ambrose REVISED November 21, 2008 WP 2008-09 Does It Pay to Read Your Junk Mail? Evidence of the Effect of Advertising on Home Equity Credit Choices∗ Sumit Agarwal Federal Reserve Bank of Chicago 230 South LaSalle St. Chicago, IL 60604 312-322-5973 sagarwal@frbchi.org and Brent W. Ambrose Jeffery L. and Cindy M. King Faculty Fellow and Professor of Real Estate Smeal College of Business The Pennsylvania State University University Park, PA 16802-3603 814-867-0066 bwa10@psu.edu November 21, 2008 ∗ We thank Jan Brueckner; Richard Buttimer; Souphala Chomsisengphet; Dimitris Christelis (European University Institute Discussant); Henrik Cronqvist (AEA Discussant); Jim Follain; Fu Yuming; Bert Higgins; David Laibson; Ken Lusht; Donna Nicholson; Devin Pope (NBER Discussant); John Quigley; Stuart Rosenthal; Nancy Wallace; and the seminar participants at 2008 American Economic Association meeting in New Orleans; the 2008 NBER Summer Institute; Baruch College; the Federal Reserve Bank of Chicago; the National University of Singapore; Syracuse University; the University of California-Berkeley; the University of North Carolina at Charlotte; and the 2007 Conference on Behavioral Approaches to Consumption, Credit, and Asset Allocation at the European University Institute in Florence, Italy, for helpful discussions and comments. Research assistance ably provided by Lauren Gaudino and Cosmin Lucaci. The views expressed in this research are those of the authors and not necessarily those of the Federal Reserve System or the Federal Reserve Bank of Chicago. Does It Pay to Read Your Junk Mail? Evidence of the Effect of Advertising on Home Equity Credit Choices ABSTRACT We examine the effect of direct mail (commonly referred to as junk mail) advertising on individual financial decisions by studying consumer choice of home equity debt contracts. Consistent with the theoretical predictions, we find that financial variables underlying the relative pricing of debt contracts are the leading factors explaining consumers’ home equity debt choice. Furthermore, we also find that the intended use of debt proceeds significantly impacts consumer choice. However, when we study a subset of consumers who received a direct mail solicitation for a particular debt contract (fixed- versus adjustable-rate), we find evidence that the relative pricing variables are less relevant in explaining consumer contract choice, even though they were presented with a full menu of debt contracts. Thus, our results are consistent with the view that advertising is persuasive. JEL Classification: D1; D8; G21; M3 Keywords: Persuasion, Advertising, Contract Choice, Home Equity Lending In 2005, the top 100 U.S. advertisers spent over $271 billion on marketing across all forms of media. Of this amount, over $55 billion was spent on direct mail advertisements, making “junk mail” second only to television (at $68 billion) in dollars expended on advertising.1 Financial institutions spent over $8.4 billion marketing a wide variety of investment and credit products (e.g., mutual funds, insurance contracts, bank accounts, credit cards, and mortgage loans to name just a few of the major categories), making the financial services the fourth highest industry by dollars spent on advertising.2 Obviously, one of the roles of advertising is to persuade the consumer to purchase a good or service. Thus, a natural question arises: To what extent does advertising or persuasion impact consumer financial decisions? We answer this question by examining the effect of direct advertising (often referred to as “junk” mail) on one of the most important financial decisions facing households – the choice of mortgage contract type. Financial economists now recognize that marketing and persuasion can affect consumer investment decisions. For example, studies of consumer investments in mutual funds indicate that marketing plays an active role in determining the money flow into funds.3 In addition to evidence from mutual fund trading, Grullon, Kanatas, and Weston (2004) find evidence linking firm product market advertising and investor interest. Furthermore, Barber and Odean (2005) document that exogenous factors calling attention to particular stocks can affect investor purchase decisions.4 Thus, these studies reinforce the idea that marketing can and does impact financial decisions. While previous research in economics and marketing indicates that advertising is effective, little is known about the impact that advertising has on altering consumer evaluation of financial decisions.5 That is, can advertising lead consumers to ignore important financial factors when faced 1 Source: Advertising Age, June 26, 2006. (http://adage.com/images/random/lna2006.pdf) To put this in perspective, automotive, retail and telecom were the top three industries in terms of advertising expenditures at $20.9 billion, $18.6 billion, and $9.9 billion, respectively (Source: Advertising Age, June 26, 2006). 3 In one of the first studies to explicitly examine mutual fund marketing, Sirri and Tufano (1998) suggest that mutual fund advertising lowers consumer search costs and that this can explain the link between advertising and fund flow. Similarly, Jain and Wu (2000) and Barber, Odean, and Zheng (2005) show that mutual fund advertising is related to money flow (investment). More recently, Cronqvist (2006) shows that mutual fund advertising impacts investors’ choices even though it provides little information. In addition, Reuter and Zitzewitz (2006) find that mutual fund flows are positively related to positive news articles in the financial press. 4 Although Barber and Odean (2008) do not explicitly examine the role of advertising, they do note that news events, excessive trading volume, and extraordinary returns can affect investment decisions. 5 The economics literature traditionally classifies advertising as being persuasive, informative, or complementary (Bagwell, 2007). The informative view assumes that advertising simply conveys information, (e.g., Stigler, 1961; and Nelson, 1974 and 1975.), while the persuasive view assumes that advertising can alter consumer preferences. 2 1 with an economic decision? In one of the few studies to examine this question, Bertrand et al. (2006) conduct a field experiment in South Africa, using personal loan contracts. Their experiment presents evidence showing that variations in the psychological features of the advertisement, as well as traditional economic variables, such as interest rates, impact consumer loan take-up rates. The results from this field experiment are consistent with the findings of Russo, Carlson, and Meloy (2006) that persuasive information can lead decision-makers to choose inferior alternatives.6 At the theoretical level, Mullainathan, Schwartzstein, and Shliefer (2006) build a simple model of persuasion that helps explain certain aspects of marketing – branding, advertisement, and product attributes (also see Mullainathan and Shliefer, 2005) – while Shapiro (2006) theoretically demonstrates that advertisement can be persuasive rather than informative. To the best of our knowledge, little research has examined the impact that advertising and persuasion can have on consumer choice in the mortgage market.7 Yet, for most households, their mortgage is their single largest financial liability and the choice between fixed-rate mortgage (FRM) versus adjustable-rate mortgage (ARM) contracts can have a substantial impact on the overall cost of home financing (Campbell, 2006). For example, the historic decline in interest rates between 2000 and 2003 resulted in a dramatic lowering of mortgage costs for ARM borrowers, which prompted the Chairman of the Federal Reserve Board to remark that U.S. homeowners “pay a lot of money for the right to refinance and for the insurance against increasing mortgage payments” through their preference for fixed-rate mortgages over adjustable-rate mortgages.8,9 We examine the consumer choice of fixed versus variable rate debt by focusing on the home equity lending market in order to determine whether consumers rationally price the interest rate 6 In a related area of the literature, numerous studies have focused on the role that information framing has on individual decision choices (see, for example, Kahneman and Tversky, 1979, and Tversky and Kahneman, 1981 and 1986, among others.) 7 Recently, Perry and Motley (2008) explore the differences in newspaper print advertising message content in the Washington, D.C. metropolitan area aimed at the prime and subprime mortgage markets. Their analysis reveals that newspaper print mortgage advertisements having persuasive elements were more common than print advertisements containing informative elements. 8 See remarks by Chairman Alan Greenspan, Understanding household debt obligations at the Credit Union National Association 2004 Governmental Affairs Conference, Washington, D.C. (February 23, 2004) [http://www.federalreserve.gov/boardDocs/speeches/2004/20040223/default.htm]. 9 Previous research on mortgage choice indicates that borrowers rationally price the interest rate risk inherent in adjustable-rate contracts. For example, Brueckner and Follain (1988) examined borrower choice for first mortgage contracts and found that the interest rate differential between fixed-rate and adjustable-rate mortgages and the interest rate level for fixed-rate mortgages are the primary factors explaining why borrowers choose fixed versus adjustable contracts. Their result suggests that borrowers react rationally to changes in the relative pricing of ARM and FRM contracts when originating mortgage debt. 2 insurance feature of fixed-rate mortgages in the presence of direct mail advertising. Our data come from a large financial institution (the data are proprietary in nature) that accepted home equity credit applications from a large number of branch offices. Furthermore, we utilize a natural experiment arising from the bank’s marketing campaign that allows us to determine whether an applicant was exposed to a direct mail solicitation prior to applying for a loan. During the marketing campaign, applicants arrived at the bank’s branch locations via one of two methods. First, applications were accepted at local branches from customers who were not targeted by the bank’s marketing campaign. We refer to these applicants as “walk-in” (WI) customers. Second, the lender received applications at the branch locations from customers who were targeted with a direct mail solicitation advertising a home equity product. We refer to these applicants as “direct mail” (DM) customers. We are able to test the persuasive view versus the informative or complementary view of advertising by examining the choices of the DM customers relative to the choices made by the WI customers. If the lender’s direct mail campaign is persuasive, then we should observe differences between the DM and WI customers’ mortgage choices. However, if the advertising is informative or complementary, then we should observe DM and WI customers responding similarly to changes in economic conditions. Previewing our results, we find that the WI customers reacted as expected and chose fixed-rate or variable-rate home equity credit products based on the prevailing interest rate and economic environment at the time of application. That is, WI customers selected the variable-rate product during periods with higher interest rates. The results from the analysis of the WI customers are broadly consistent with previous empirical work. However, in comparing product choice across DM and WI customers, we find that DM customers do not react as expected to changes in the economic environment. Using a variety of empirical methods, we control for possible differences between the WI and DM customers. As a result, we are able to isolate the effect of the direct mail solicitation on the customer’s mortgage decision. In particular, we show that consumers who receive a direct mail solicitation are more likely to ignore the important economic and interest rate environment factors that influenced the decisions of the WI customers. Examining the applicant choices reveals that 78 percent of the DM customers were influenced by the bank’s solicitation, while 22 percent responded to signals present in the economic environment. Further analysis of the subset of applicants who 3 were clearly influenced by the bank’s solicitation reveals that 74 percent were persuaded to originate a product that was opposite to the one selected by their counterparts who did not receive a direct mail solicitation. However, we also find that the direct mail solicitation could be classified as informative for 26 percent of the DM customers. In support of our finding that some borrowers were persuaded to make an incorrect product choice, we find that the three-month prepayment rate for the persuaded borrowers is almost four times higher than the corresponding prepayment rates on borrowers who were informed by the bank’s solicitation. Our results are consistent with evidence of consumer learning.10 Thus, while our study reveals that bank advertising has a persuasive effect on consumer financial decisions for a majority of the applicants who received a solicitation, we also find evidence that is consistent with the informative role of advertising for a smaller subset of consumers. Finally, we believe that our findings are not the result of sample selection issues arising from possible correlation between the customer’s decision to respond to the bank’s marketing campaign and the mortgage choice decision. First, we note that the choice of fixed-rate or adjustable-rate debt is a function of borrower expected tenure.11 As a result, we expect that if sample selection is present it should bias our analysis toward observing a higher probability of selecting the fixed-rate product, weakening the effect of a line solicitation and biasing our estimated coefficients away from statistical significance. Second, we find no differences in location patterns between WI and DM customers. Third, we confirm our assertion that sample selection is not affecting our results by estimating a bivariate probit model that jointly captures the customer’s decision to respond to the advertising campaign and the choice of mortgage product. Consistent with our assertion, we find that the estimate for the sample selection coefficient is insignificant. Thus, we do not believe that sample selection is biasing our findings of a persuasive advertising effect. The findings from this study support the growing literature revealing that consumers are highly influenced by the presentation of information that frames financial decisions. For example, Mandrain and Shea (2001) present evidence showing that the design of a firm’s retirement contribution plan has a meaningful impact on the participation choices of employees.12 In another 10 See Agarwal et al. (2005). See Rosenthal and Zorn (1993) for a discussion of borrower tenure and mortgage choice preferences. 12 Brown, Liang, and Weisbenner (2007) also show that the menu of investment options has a significant impact on participant portfolio choice. 11 4 example outside of economics, Kressel, Chapman, and Leventhal (2007) demonstrate that the format of survey questions has a direct impact on an individual’s end-of-life treatment choice. In addition, in an area more closely related to the mortgage choice decision examined in this paper, Shiller (2008) discusses anecdotal evidence that many subprime borrowers accepted mortgage terms that were probably not in their best interests, simply because the information about the contract appeared to come from an expert – such as a financial institution. Thus, our result indicating that advertising has a strong influence on consumer choice of mortgage product is consistent with this growing body of evidence showing that the way information is presented or that the way financial choices are framed can have a significant impact on the decision outcome. Our paper proceeds as follows. In the next section, we outline the previous theoretical and empirical studies of borrower mortgage choice and the role of advertising. In Section II, we discuss the differences between home equity lines-of-credit and home equity loans. We describe the data and empirical method in section III. In section IV, we present the results, and in section V, we compare the switchers and non-switchers. We provide robustness tests in section VI and we conclude in section VII. I Mortgage Choice and the Role of Advertising In this section we discuss the theoretical motivations for borrower mortgage choice and the role that lender advertising may play in influencing that choice. The theoretical literature on mortgage choice is well developed and offers a number of testable hypotheses. For example, Alm and Follain (1987) and Brueckner (1986) suggest that borrower risk aversion is a primary factor determining choice betweeen ARM and FRM. These models indicate that borrowers with low risk aversion and high discount rates should prefer the higher interest rate risk associated with adjustable contracts, while borrowers with relatively high risk aversion and/or lower discount rates should prefer fixedrate contracts. Thus, the trade-off between fixed-rate and adjustable-rate mortgages should depend upon the prevailing interest rate environment at the time of origination.13 Extending the earlier work on mortgage choice, Brueckner (1993) and Rosenthal and Zorn (1993) focus on the role that borrower expected mobility plays in determining the selection of 13 Empirically, Dhillion, Shilling, and Sirmans (1987) find that loan pricing factors are the primary determinants of borrowers choosing ARMs over FRMs in the first mortgage market. 5 adjustable versus fixed-rate debt. The comparative statics derived from these models indicate that the borrower’s propensity to choose an adjustable-rate contract is inversely related to her expected tenure. Furthermore, the comparative statics in Brueckner (1993) also support earlier theoretical models by indicating that the level of interest rates are inversely related to the attractiveness of the adjustable-rate debt. More recently, a number of researchers have recognized the complexity and importance of the optimal mortgage choice problem within the context of household life-cycle consumption models. For example, Campbell and Cocco (2003) solve a dynamic model of mortgage choice and consumption assuming that borrower income is risky. Their analysis implies that borrowers with high risk aversion will prefer fixed-rate mortgages and that mortgage choice may reveal unobserved heterogeneity in borrower risk profiles.14 In addition, Campbell and Cocco’s (2003) analysis suggests that when the yield spread between fixed- and adjustable-rate mortgages is large (small), homeowners should prefer ARMs (FRMs). Empirically, Campbell (2006) confirms this result by finding that the share of ARMs to total mortgage origination is directly proportional to the both the FRM– ARM interest rate differential and the level of the FRM interest rate. In a novel empirical test that takes advantage of the discontinuity in the U.S. mortgage market resulting from the distinction between “conforming” and “non-conforming” mortgages, Vickery (2006) finds that borrower mortgage choices are highly sensitive to changes in FRM interest rates.15 For example, Vickery (2006) estimates that a 10 basis point increase in fixed-rate mortgage rates corresponds to a 10.4 percentage point decline in the FRM market share.16 While the theoretical literature clearly shows that borrower choice of mortgage type should depend upon prevailing financial conditions at origination, the prior empirical research has relied on the use of originated loans, necessitating the use of econometric models to infer borrower sensitivity 14 See Sa-Aadu and Shilling (1994); Sa-Aadu and Sirmans (1995); and Chiang, Chow, and Liu (2002). In addition, borrower mobility (see Chan, 1996; and Gabriel and Rosenthal, 1993) and borrower perceptions of default risk (see Posey and Yavas, 2001) may play a role in contract choice. Recent theoretical work by van Hemert, de Jong, and Driessen (2005) and van Hemert (2006) also reinforces the link between borrower risk aversion and ARM preference. 15 “Conforming” mortgages are loans that are eligible for purchase by the housing government sponsored enterprises (GSEs), Fannie Mae and Freddie Mac. In contrast, “non-conforming” mortgages are ineligible for purchase by the GSEs. In general, “conforming” mortgages have loan balances below the conforming loan limit (updated annually) and meet other underwriting risk criteria set by the GSEs. 16 Recent work by Koijen, van Hemert, and van Nieuwerburgh (2006) using aggregate ARM/FRM market shares indicates that the inflation risk premium and prepayment option value are primary factors in determining ARM market shares. 6 to the interest rate environment.17 Yet, a recent analysis of the home equity lending market by Agarwal et al. (2008) reveals that lenders can and do alter loan contract terms during the underwriting process and thus effect the observed “choice” of fixed-rate versus adjustable-rate contracts. In this study, we focus on the borrower’s initial choice as revealed on the credit application. Thus, we are able to isolate the factors impacting borrower choice, free of any bias introduced through subsequent lender screening and underwriting. While the brief review above demonstrates that borrower mortgage choice has received considerable attention, no study has examined the impact of lender advertising on this choice. Economists have long considered the effect of advertising on consumer behavior. In a recent survey of the previous century of economic research on advertising, Bagwell (2007) notes that economist generally view advertising as falling into one of three categories: persuasive, informative, or complementary. Under the persuasive view, economists assume that “advertising alters consumers’ tastes.”18 According to the persuasive theory, firms advertise with the goal of altering consumers’ preferences so that they purchase the good or service being advertised.19 Thus, one outcome of persuasive advertising is that consumers purchase the “wrong” or “incorrect” amount of the product or service. In the context of our mortgage choice problem, the persuasive view of advertising suggests that a lender’s direct mail solicitation causes consumers to ignore their evaluation of the economic environment and thus select the advertised product. In contrast, the informative view, based on the work of Ozga (1960) and Stigler (1961), concludes that advertising provides consumers with information and lowers consumer search costs. In the context of the mortgage choice problem, the informative view of advertising suggests that the lender’s direct mail solicitation provides information to the consumers (for example, reminding them that attractive interest rates exist on home equity products). As a result, direct mail solicitations lower search costs but do not alter preferences for a particular product based on the prevailing economic environment. Under this view, the consumer’s choice should coincide with the type of product advertised in the solicitation. Finally, the complementary view assumes that consumers’ tastes and preferences are stable 17 See Brueckner and Follain (1988) for example. Bagwell (2007), p. 3. 19 Bagwell (2007) documents that the persuasive view developed from early research by Robinson (1933) and Braithwaite (1928). Bagwell comments that Braithwaite suggested that “advertising’s effect is to induce consumers to purchase the wrong quantities of goods that are not well adapted to their true needs...” (p. 10.) 18 7 and advertising complements them to encourage consumption.20 Under this view, the direct mail solicitation encourages prospective customers to want a home equity product from the lender, but the choice of product type still reflects their individual tastes and preferences. Our study of home equity product choice has the potential to differentiate these competing economic views of advertising. First, if the persuasive view of advertising is correct, then we should observe the direct mail customers ignoring economic and interest rate environment factors and selecting the mortgage product suggested in the solicitation. However, if the informative view is correct, then we should observe the consumer’s product choice coinciding with the product advertised in the solicitation, and this choice should be consistent with our theoretical expectations for product choice given the economic and interest rate environment prevailing at the time of application. Finally, if the complementary view is correct, then we should observe the direct mail customers selecting products in line with the theoretical predictions given the economic and interest rate environment, without regard to the type of product listed on the solicitation. Empirically differentiating between the three competing views of advertising is difficult. Thus, we utilize several independent methods to isolate the effect. First, we estimate a direct probabilistic choice model. Second, we analyze the DM customer who selected products other than the one on the solicitation. Third, we compare DM customers with a matched sample of WI customers in order to eliminate any spurious comparisons. Finally, we use information provided by the borrower about their intended use for the funds to identify borrowers most likely to be informed by the advertising from those most likely to complemented by it. Each method has advantages and disadvantages. However, by triangulating the results we can separate the persuasive effect of advertising from the informative and complementary effects. II Differences Between Home Equity Lines and Loans Home equity credit falls into two categories: home equity loans (i.e. “spot” loans) and home equity lines (i.e. credit lines or lines-of-credit). Agarwal et al. (2006) note that “a spot loan is a closedend loan extended for a specified length of time requiring repayment of interest and principal in equal monthly installments.” The interest rate on home equity loans is set at loan origination. In 20 See Bagwell (2007) for a discussion of the development of the complementary view. 8 contrast, they define a credit line as “an open-ended, variable rate, revolving credit facility that permits the consumer to borrow up to a predetermined amount (the line amount),” and note that borrowers usually are required to pay interest only on the used portion of the line during the first five years, after which the line becomes a fully amortizing loan. Significant differences exist between borrowers who choose lines versus loans, with line borrowers having greater wealth – as indicated by their having relatively more expensive homes, higher incomes, and greater home equity. For example, Canner et al. (1998) document that the median home equity for credit line borrowers is $41,000 greater than spot loan borrowers ($76,000 versus $35,000) and that the median household income for spot loan borrowers is $10,000 less than the median household income of credit line borrowers. Canner et al. (1998) also note a significant difference in the ages of line borrowers and loan borrowers; 23 percent of the loan borrowers versus 6 percent of line borrowers are below the age of 34. In addition, the 1997 Survey of Consumers shows that 49 percent of the households who prefer loans over lines are relatively more sensitive to interest rates and that “ease of use” is the primary motivation for credit line borrowers. Thus, it is imperative to control for borrower heterogeneity in analyzing contract choice. III Data As discussed in Agarwal et al. (2007), the data used in this study were provided by a large financial institution and consist of variable-rate home-equity lines-of-credit (HELOCs) and fixed-rate homeequity loans (HELs) issued to owner-occupants from March 2002 through December 2002.21 The credit lines are open for the first five years, and the borrower is only required to make interest payments on the utilized line balance during this period. After the fifth year, the line is closed and is converted to a fully amortizing, fixed-rate term loan with a remaining term of five to 15 years. The lender received applications from customers via two channels. First, the majority of applications were from customers walking into their local branch and requesting a home equity credit. At this point, the local loan officer provided the customer with a menu of various home equity products – with the primary choice being a variable rate (line-of-credit) or fixed-rate (loan).22 As 21 During the sample period, this institution had operations primarily in the New England, Mid-Atlantic, and Florida regions and was ranked among the top-five commercial banks and savings institutions by the FDIC. 22 Each product also contained a variety of pricing options based on the requested loan-to-value ratio. 9 previously mentioned, we refer to these customers as “walk-in” (WI) customers. Between March 2002 and December 2002, the lender received over 108,000 applications by WI customers. Second, the lender also received applications from customers who had received a direct mail solicitation advertising either a line-of-credit or home-equity loan. Between March 2002 and May 2002, the bank sent out over 3 million direct mail solicitations in 12 equally distributed waves (or campaigns) to potential customers (or households) with credit (FICO) scores above 640.23 Across these 12 campaigns, approximately 2.1 million customers were targeted with a line-of-credit solicitation while 981,000 received a home equity loan solicitation. Conditional on maintaining the approximately two to one ratio of line to loan mailings, the bank randomly selected customers to receive the line-of-credit or loan offer.24 That is, the bank did not specifically target individuals for a line or loan offer; rather, the bank randomly sent the line and loan mailings to customers with FICO scores greater than 640.25 Table I shows the mean FICO scores and geographic distribution of the customers sent the direct mail solicitations. Consistent with the bank’s practice of randomly selecting customers for the two product solicitations, we see that the average FICO scores of the line and loan groups do not differ economically.26 Table I also shows the average credit scores and geographic distribution for the customers that responded to the bank’s solicitation. As is typical in direct mail marketing campaigns, the response rate is low. For example, 20,500 customers responded to the bank’s line-of-credit solicitation for a 0.99 percent response rate and 11,249 customers responded to the loan solicitation for a response rate of 1.15 percent. We also see that the credit scores for responding customers are lower than the credit quality of the population receiving 23 In designing the marketing program, the bank requested that the credit bureau provide a random sample of households in the target area that had credit scores above 640 for the purpose of conducting a direct mail campaign. By law, the bank cannot request information from the credit bureau and then screen the households again prior to mailing the solicitation. That is, once the bank pulls a credit score for a household, it is obligated to send that household a solicitation. The solicitation does not indicate that the households are “pre-approved” for credit. Since the solicitation is based only on credit score, some households that ultimately respond to the offer may be denied credit by the bank based on other household characteristics that are not in line with the bank’s underwriting standards. 24 We confirmed the bank’s random assignment of the line-of-credit and loan offers through discussions with representatives at the bank. Furthermore, we note that the random assignment is consistent with industry practice in financial product marketing campaigns. For example, Ausubel (1999) reports a similar random assignment of direct mail solicitations in credit card offers. 25 The bank’s two-to-one targeting of lines-of-credit versus loans likely reflects the underlying profitability of these contracts and responses to competitive pressures. For example, Agarwal et al. (2006) show that home equity linesof-credit and home equity loans have significantly different default rates implying substantial differences in potential regulatory capital requirements under Basel II. 26 The precision underlying consumer credit scoring models is such that the bank would not distinguish between borrowers with FICO scores of 729 and 722. 10 the solicitation; this is consistent with the experience reported in other consumer loan research.27 Although the customers received a solicitation for a specific product (either a line-of-credit or a loan), at the time of application the local loan officer showed them the same product menu as the WI customers and they were free to choose either product. Additionally, the solicitation for a line offer provided the option for the customer to choose a loan offer (and vice versa). As previously mentioned, we refer to the customers who responded to this advertisement as “direct mail” or DM customers. Table II shows the descriptive statistics for the DM and WI customer groups. A comparison of the sample means between the WI and DM customers clearly suggests that the two groups are different. For example, on average the WI customers have higher estimated home values, greater income, more job seniority, and longer tenure at the present address. Furthermore, WI customers request greater loan amounts (consistent with having higher average house values) but lower loanto-value ratios. Combining these risk factors suggests that WI loans are lower risk.28 Although the DM and WI customer groups are distinct, the bank did not systematically target the line or loan solicitation to individuals who were more likely to respond to such solicitations. To confirm this, we report the summary statistics for location (neighborhood) demographic characteristics in Table III. As we know the zip code for all walk-in customers as well as all recipients of the bank’s direct mail solicitation, we aggregated census tract demographic information from the 2000 census to the zip code level. The columns under “Direct Mail (Mailings)” show the mean values of the zip codes for all recipients of the bank’s solicitation. The columns associated with “Direct Mail (Response)” show the mean values of the zip codes for the customers who actually responded to the bank’s solicitation. Finally, the columns “Walk-In” show the mean values for the zip codes corresponding to the WI customers. If the bank systematically targeted areas based on demographic 27 Ausubel (1999) reports a similar result for responses to direct mail credit card solicitations. Although it is possible that the WI customers were exposed to the bank’s direct mail marketing campaign through contact with the DM recipients (see Hong, Kubik, and Stein (2005) or Shiller and Pound (1989) for evidence of informal information transfer about financial products), we are unable to control or measure this possibility. However, if the WI customers did systematically respond to the bank’s direct mail campaign via “word-of-mouth” contact with the DM recipients, then this contamination should bias our analysis away from finding an effect for the advertisement. In our empirical analysis, we use the WI customers as the control group and thus any spill-over from contact with the DM customers would bias downward our estimates of the direct mail effect on the DM customers’ choice. This implies that any positive effect may actually be stronger than reported in our analysis. Furthermore, as part of our process for matching the direct mail customers with the home equity credit application database, we confirmed that the customers identified as direct mail applicants applied for the credit after the mailing date of the direct mail solicitation. 28 11 characteristics, then we should see differences in the demographic characteristics of the “Direct Mail (Mailings)” zip codes and the “Walk-In” zip codes. Similarly, if the customers who responded to the solicitation are concentrated in areas that are demographically distinct, then we should observe differences in the “Direct Mail (Response)” zip codes to both the “Direct Mail (Mailings)” and “Walk-In” zip codes. Comparing the mean values reported in Table III clearly shows that the WI and DM customers reside in demographically similar neighborhoods. For example, the median neighborhood income for persons who received a line-of-credit solicitation is $50,677; the median neighborhood income for walk-in customers who selected lines-of-credit is $50,597; and the median neighborhood income for DM customers who received a line-of-credit solicitation and responded is $50,595. As a result, it does not appear that meaningful systematic differences exists in the neighborhoods of WI and DM customers. IV Empirical Methods and Results Our empirical analysis focuses on identifying the effect that lender advertising has on financial decisions. As discussed previously, identifying advertising’s effect is challenging. In this section, we estimate a consumer choice model under the assumption that consumers choose the contract that maximizes their personal utility function and that this utility is maximized subject to a variety of economic and personal factors. Analogous to studies that examine the effect of various social programs or medical treatments, we include shift variables that identify consumers who received direct mail solicitations. IV.A A Model of Mortgage Choice We begin by letting Yi represent applicant i’s choice of mortgage type, where Y = 1 denotes the variable rate line-of-credit and Y = 0 is the fixed-rate home equity loan. As noted previously, we also observe that some applicants received a direct mail solicitation from the bank. Our goal is to determine whether this solicitation had an impact on the applicant’s mortgage choice. The classic approach for this type of problem is to estimate the following probit model of borrower home-equity 12 choice: P r(Yi = 1) = βXi + αIi + εi , (1) where Xi is a matrix of explanatory variables; Ii = [Iiv , Iif ]0 with Iiv equaling one if the applicant received a direct mail variable-rate line-of-credit solicitation and zero otherwise and Iif equaling one if the applicant received a direct mail fixed-rate home equity loan solicitation and zero otherwise; α = [αv , αf ] and β are a set of coefficients to be estimated; and εi is a standard error term. In (1), Ii represents a demand shift that isolates the effect of the direct mail solicitation on the borrower’s mortgage choice probability. Estimation of (1) provides an indication of whether receiving a direct mail solicitation affects the mortgage choice decision. However, we are also interested in whether the bank’s direct mail solicitation altered the borrower’s sensitivity to factors in the prevailing financial environment. For example, prior studies indicate that mortgage choice should depend upon relative borrower risk aversion and interest rates.29 Thus, we segment Xi into variables representing borrower riskiness (XiR ), interest rate environment (XiE ), and other demographic and geographic factors (XiD ). By interacting the direct mail solicitation variables with XiR and XiE , we are able to isolate whether the bank’s marketing efforts altered the applicant’s decision process. As a result, we obtain the following model P r(Yi = 1) = βXi + αIi + πR XiR Ii + πE XiE Ii + εi , (2) v , π f ]0 and π = [π v , π f ]0 . The coefficients π and π provide an indication of the where πR = [πR E R E E R E impact that the direct mail solicitation had on the applicants evaluation of the economic and risk factors. As discussed previously, the persuasive view implies that direct mail customers will ignore economic and interest rate factors. Thus, we should observe significant coefficients for the advertising shift parameters with αv > 0 and αf < 0 because the solicitation is supposed to alter consumer’s tastes or preferences making them desire the product being advertised, while the interaction terms 29 See Alm and Follain (1987) and Brueckner (1986). 13 should not be significant (πE = πR = 0) because the advertising effect is assumed to cause consumers to ignore other factors. In contrast, the informative and complementary views imply that the interaction terms (πE , πR ) should have the same sign as the base parameters (β), since informative advertising does not alter the consumer’s underlying preferences. IV.B Selection Issues A potential problem associated with the estimation of (2) is that the consumer’s response to the bank’s marketing effort may not be exogenous. That is, the household’s decision to respond to the offer may be correlated with the variables in X that impact the mortgage choice decision. The nature of the correlation arises from the fact that the choice of fixed versus adjustable-rate loans is a function of expected borrower tenure. For example, analysis by Rosenthal and Zorn (1993) suggests that borrowers with relatively higher expected mobility should prefer adjustablerate loans over fixed-rate loans. Since our study involves home equity credit (not first mortgages), it is reasonable to assume that applicants seeking home equity credit have low expected mobility – leading to a high tenure expectation. As a result, home equity credit applicants should have an unbiased preference for a fixed-rate loan, all else being equal. Thus, any bias introduced as a result of applicant self-selection should be skewed toward observing a higher probability of selecting the fixed-rate product, weakening the effect of a line solicitation and biasing our estimate of αv downward. As a result, any presence of applicant self-selection should bias the estimated coefficients away from our hypothesis that advertising will affect consumer choices. Thus, any finding of a persuasive effect from advertising would be stronger than indicated. In addition, we control for possible sample selection bias by estimating (2) as a bivariate probit model: ∆i = γZi + νi ; P r(Yi = 1) = βXi + αIi + πR XiR Ii + πE XiE Ii + εi , (3) (4) where (3) reflects the customer’s decision to respond to the bank’s direct mail solicitation with ∆i 14 equal to one if the customer responded to the direct mail solicitation and zero otherwise, Zi is a matrix representing the household’s credit (FICO) score at the time the bank began the marketing campaign along with location and campaign fixed effects, and νi is a standard error term.30 Equations (3) and (4) are estimated via full information maximum likelihood where ρ = Cov[νi , εi ]. By including all hard information captured by the bank’s application in Xi and aggregate information concerning the borrower’s specific location (neighborhood) in Zi , the bivariate probit framework allows us to effectively control for potential differences between WI and DM customers. However, as discussed previously, the descriptive statistics reported in Table III confirm that no meaningful differences exist at the neighborhood level between the recipients of the bank’s solicitation, the customers who responded to the solicitation, or the walk-in customers. IV.C Results Tables IV and V report the estimated coefficients for the bivariate probit model. First, Table IV reports the estimated coefficients for the first-stage sample selection model. We note that the estimated coefficient for the customer credit score at solicitation is negative and significant, indicating that offer recipients with higher credit scores are less likely to respond to the bank’s solicitation. We also include zip code location fixed effects and campaign fixed effects to control for potential differences in customer response to the solicitation arising from geographic location and mailing date. Table V reports the coefficients from the second-stage sample selection estimation. The estimated coefficients confirm prior research about the decision to choose between a variable-rate and fixed-rate contract. We also note that the sample selection correction parameter (ρ) is statistically insignificant. Given the large number of observations included in the analysis, it is not surprising that almost all of the estimated coefficients in Table V are statistically significant (at the 5 percent level), with the exception being the interaction of the direct mail dummy variables with the income and interest rate variables. Thus, we focus on the variables’ marginal impacts to provide guidance as 30 Because of the time lag between the start of the marketing campaign and the customer’s decision to apply for a home equity product, the customer’s credit score may change. We refer to the former as the “Solicitation FICO Score” and the later as the “Application FICO Score.” 15 to the relative importance of the factors impacting the borrower’s decision.31 Interestingly, only the estimated coefficients for interest rates (FRM/ARM rate differential and FRM rate level), the borrowers declared intended use of the funds (consumption or refinancing), and direct mail solicitation have marginal impacts greater than 10 percent.32 IV.C.1 Impact of Borrower’s Stated Use of Funds The borrower’s declared use of the debt proceeds clearly has a strong impact on choice of contract. The marginal impacts indicate that a borrower who intends to use the funds for consumption is 13.2 percent more likely to select the adjustable-rate line-of-credit than a borrower who intends to use the funds for home improvements. However, a borrower intending to refinance existing debt is 16.8 percent less likely to select the adjustable-rate line-of-credit than a borrower seeking funds for home improvements. Clearly, borrowers prefer the flexibility associated with the adjustablerate line-of-credit when using home-equity to smooth consumption while preferring the certainty of fixed-rate contracts when refinancing (or consolidating) existing debt. IV.C.2 Impact of Interest Rate Environment We find that the impact of market interest rates in influencing the borrower choice of contract is consistent with the theoretical predictions of Alm and Follain (1987) and Brueckner (1986) as well as the previous empirical evidence presented in Brueckner and Follain (1988). For example, the theoretical literature indicates that borrowers with high relative risk aversion and low discount rates are more likely to prefer fixed-rate contracts. Consistent with this theory regarding borrower risk aversion, we find that every one percentage point increase in the difference between the fixedrate and adjustable-rate interest rates results in a 12.4 percent increase in the probability that 31 Ai and Norton (2003) point out that the marginal effects typically supplied by conventional statistical software for interaction terms in non-linear models are incorrect. Thus, we follow Ai and Norton (2003) and calculate the marginal effects for the interaction variables as 0 00 ∂ 2 Φ(·) = πR Φ (·) + (β + πR Ii )(α + πR XiR )Φ (·) ∂Ii ∂XiR (5) 0 00 ∂ 2 Φ(·) = πE Φ (·) + (β + πE Ii )(α + πE XiE )Φ (·), ∂Ii ∂XiE (6) where Φ represents the standard normal cumulative distribution. 32 The “home improvement” intended use is the reference category for the use of funds dummy variables. 16 the borrower will select the adjustable-rate contract. Furthermore, every one percentage point increase in the home equity loan interest rate (the fixed-rate product) raises the probability that the borrower will select the adjustable-rate line-of-credit by 13.3 percent. To highlight the impact of interest rate movements on borrower contract choice, we estimate the probabilities of selecting a home equity line-of-credit for two sets of hypothetical borrowers based on the short term (three months) and long term (five years) daily Treasury rates observed between January 2003 and December 2004.33 We construct our hypothetical borrowers by assuming that one set of borrowers are “walk-in” customers (DM = 0) and the other set received a direct mail solicitation (DM = 1). We further assume that each borrower has values for the other variables in Xi equal to the sample means. Finally, in each set we assume that one borrower indicated that she would use the debt proceeds for consumption and that the other would use the debt proceeds to refinance. Based on these values and the estimated coefficients in Table V, we calculate the time-varying probabilities that each borrower would select an adjustable-rate line-of-credit.34 To highlight the interest rates used in the simulation, Figure 1 shows the FRM interest rate and interest rate differential between January 2003 and December 2004. Over this period, the fixed-rate reference interest rate varied from 3.1 percent (in June 2003) to over 4.8 percent (in June 2004). Furthermore, this period saw a decline in the yield curve from 3.1 percent (in May 2004) to 1.3 percent (in December 2004) indicating a substantial drop in short-term interest rates. Figures 2 and 3 show the probability of selecting an ARM for borrowers with consumption and refinancing motives, respectively. Looking first at the consumption-motivated borrowers in Figure 2, we see that the probability that the walk-in customer will select the adjustable-rate product ranges between 54 percent and 77 percent, and closely tracks the FRM interest rate. During periods when the FRM interest rate is relatively low, the probability of selecting the adjustable-rate product is correspondingly low – indicating that borrowers are correctly anticipating higher future rates and thus seek to lock-in the lower current rate. As interest rates increase, the probability of selecting an adjustable-rate product increases. Comparing the walk-in customers with the directmail customers, one striking observation is the relative interest rate insensitivity of the directmail customers. That is, the probability that a direct-mail customer who received a line-of-credit 33 See Brueckner and Follain (1988) for a similar simulation. Obviously, the probability of selecting the fixed-rate home equity loan is simply one minus the adjustable-rate probability. 34 17 offer only varies between 81 percent and 84 percent. Furthermore, for borrowers who received a solicitation suggesting a home equity loan, the probability of selecting the adjustable-rate product only varies between 8 percent and 10 percent. Figure 3 reveals a similar pattern for the borrowers indicating a refinancing motive. Again, we see that the walk-in customers are relatively sensitive to changes in the interest rate environment with the probability of selecting an adjustable-rate product declining when interest rates are low and rising after interest rates have moved up. Furthermore, the direct-mail customers who indicated a refinancing motive show a pattern of being relatively insensitive to interest rate movements. this again is similar to the pattern for consumption-motivated borrowers. When we compare the changes in ARM selection probability between walk-in customers based on their indicated use, we find that borrowers with refinancing motives are more sensitive to changes in the underlying yield curve than borrowers with consumption motives. For example, between March 9, 2004, and May 13, 2004, the FRM benchmark increased from 3.73 percent to 4.85 percent and the yield curve (interest rate differential) increased from 1.72 percent to 3.01 percent. As a result of this dramatic increase in longer-term interest rates, the probability that a walk-in refinancing borrower would select an adjustable-rate product almost doubled, from 16 percent to 29 percent. However, during the same period, the probability that a similar consumption-motivated borrower would select an adjustable-rate product only increased from 62 percent to 77 percent. IV.C.3 Impact of Bank’s Advertising Campaign We now examine the effect of the bank’s solicitation on the applicant’s mortgage choice. The marginal effects reported in Table V clearly indicate that the bank’s direct mail solicitations had a significant impact on borrower product choice. After controlling for all other factors, a borrower receiving a line-of-credit solicitation was 17.7 percent more likely to select a variable rate line-ofcredit than the fixed-rate product. Similarly, borrowers who received a direct mail loan offer were 14.7 percent less likely to select the variable rate product than the fixed-rate product. In other words, the results indicate that αv > 0 and αf < 0. Consistent with the persuasive view, Figures 2 and 3 highlight the impact of the bank’s direct mail solicitations and provide evidence that the interaction terms are not significant. Both figures indicate that WI customers are relatively sensitive to changes in the market interest rate 18 environment. In comparison, the estimated probabilities for the customers who responded to the direct mail solicitation are virtually constant over the entire period. As a result, it does not appear that the customers who received a direct mail solicitation respond to changes in the economic environment (as reflected in movements in interest rates) in a manner that is consistent with the WI customers. The lack of response for the DM customers to changes in the interest rate environment is consistent with persuasive view that the advertising caused the customers to ignore other factors. That is, πE = 0 and πR = 0. The results presented here make a compelling case against the informative view of advertising. Under the informative view, we should observe similar sensitivities to changes in the interest rate environment for the WI and DM customers. Thus, the relatively constant probabilities of ARM choice for the DM customers is inconsistent with the informative view and suggests that the lender’s solicitation was persuasive. The results are less clear for differentiating the persuasive effect from the complementary effect because of differences between refinancing-motivated and consumption-motivated borrowers. The complementary view suggests that we should observe similar probabilities of ARM selection for the DM customers regardless of the type of solicitation received, since the solicitation is supposed to remind the customers of the availability of credit but not alter their tastes or preferences for a particular type of credit. However, the results for the consumption-motivated borrowers do not support the complementary view. Figure 2 shows that the probability of selecting an ARM for the consumption-motivated customers who received a loan offer is significantly lower than the corresponding probability for the borrowers who received a line offer. For the complementary view to hold, we expect a relatively high probability of selecting an adjustable-rate line-of-credit for borrowers who intend to use the funds for consumption because interest is only charged on the amount of credit utilized. Thus, Figure 2 reveals that borrowers who received a home equity loan solicitation are not only insensitive to interest rate movements, but they also have a low probability of selecting the adjustable-rate line-of-credit. This finding is inconsistent with the complementary view of advertising. In order to refrain from “overselling” our results, however, we note that the results for the refinancing motivated borrowers may be consistent with the complementary view. We expect that refinancing-motivated borrowers should prefer the stability of the fixed-rate product when 19 consolidating their debts.35 Consistent with this expectation, we see that the probability of selecting an ARM is almost zero for the refinancing customers who received a loan solicitation. We also see that the probability of selecting an ARM is approximately 30 percent for customers receiving the line solicitation. However, we note that the WI customers also have an ARM probability averaging approximately 20 percent. Thus, we cannot state that the solicitation altered the line versus loan preferences for the refinancing customers, since the estimated probability of selecting an ARM for them is consistent with the estimated probability of selecting an ARM for customers who did not receive a solicitation. As a result, one could view the results presented in Figure 3 as also being consistent with the complementary view. However, again, we observe that the lack of sensitivity to changes in market interest rates for the DM customers is inconsistent with the informative view. We explore the issue of informative versus persuasive advertising in Section 7. IV.D Summary To summarize, our analysis reveals that borrower mortgage choice is sensitive to the economic environment. Yet, we also observe that a subset of borrowers who received solicitations or “cues” from the bank did not select a mortgage product in a manner consistent with theoretical expectations. Overall, the results suggest that the lender’s advertising campaign had a persuasive effect.36 In the next section, we explore the impact of the direct mail solicitation in greater detail by highlighting the differences between borrowers who ignored the lender’s direct mail cue and switched to the alternative product and those who selected the product advertised in the mailing. 35 See Canner et al. (1998). A natural question arises as to whether the bank’s advertising had an “economic” impact on the borrower. Unfortunately, we are unable to calculate a direct cost. However, we note that fixed-rate products have higher interest rates than variable-rate products, and thus, borrowers who were steered to fixed-rate loans but who should have selected the variable rate product did bear a higher interest cost. As noted above, Greenspan (2004) suggested that many borrowers may have incorrectly preferred fixed-rate products. However, it is also not the case that all borrowers persuaded to select the variable-rate contract benefited from such steering. Variable-rate products do expose borrowers to greater future interest rate risk, and as discussed in Section 2, theoretical models show that fixed-rate contracts should be preferred by some borrowers depending upon economic and demographic factors. As a result, borrowers who were persuaded to select the variable-rate line-of-credit were exposed to greater interest rate risk than appropriate since they should have selected the fixed-rate contract as suggested by the theoretical predictions based upon the prevailing economic conditions at the time of origination. 36 20 V Switchers versus Non-switchers As discussed above, during the application process, all customers are presented with the full loan contract menu. Thus, even though the DM customer may have received a solicitation advertising a line-of-credit, the customer also had the option of applying for a home equity loan. By matching the database that tracked the customers who received a direct mail solicitation (for either a loan or line-of-credit) with the database of applications, we can identify instances when the borrower switched products. For example, if the borrower received a line-of-credit (or loan) solicitation, but applied for a home equity loan (or line-of-credit), then we classify that borrower as a “switcher.” However, if the borrower received a line-of-credit (loan) solicitation and also applied for a line-ofcredit (loan), then we classify that borrower as a “non-switcher.” The presence of switchers raises an interesting question: Do observable differences exist between the switchers and non-switchers? In other words, can we identify the customers who are more likely to be persuaded by the bank’s solicitation? Table VI shows the descriptive statistics for the DM customers based on whether the borrower switched products. We note that out of the 31,749 customers who received a direct mail solicitation, 22 percent selected a product that was different from the one in the solicitation. Furthermore, we see that 2,375 (21 percent) of the applicants who selected the fixed-rate product received a direct mail offer for a variable-rate product, while 4,623 (23 percent) of the applicants who selected the variable-rate product received the fixed-rate solicitation. In order to focus on the explicit differences between customers who switched from the solicited product and those who did not, we estimate a simple logit model for switch versus no switch. Table VII reports the estimated coefficients. The results indicate the customers having the characteristics of being more financially sophisticated (higher incomes and higher credit scores) are more likely to switch away from the advertised product. For example, the marginal effects imply that a customer with a FICO score of 774 is 24 percent more likely to switch than a customer with a FICO score of 724.37 Interestingly, we also see that older customers are less likely to switch than younger customers. For example, the marginal effects indicate that a 56-year-old customer is 33 percent less likely to switch than a 46-year-old customer.38 This result is consistent with the findings of 37 38 The mean FICO score for the DM customers is 724. The mean customer age at date of application if 46.5 years. 21 Agarwal et al. (2007) that financial sophistication declines with age. We also see that the customer’s indicated intended use of the funds has a direct effect on the probability of switching. For example, borrowers using the loan proceeds for refinancing are 2.7 percent more likely to switch, while borrowers using the funds for consumption are 5.6 percent less likely to switch than borrowers using the funds for home improvements. These results are consistent with the idea that customers seeking to rate refinance are sufficiently sophisticated that they respond to incentives present in the economic environment and are not persuaded to accept the offer presented in the solicitation. VI Robustness Tests As a robustness check on the evidence presented above, we analyze the impact of the direct mail solicitation using a matched sample method. This method requires that we create a matched sample of WI customers that is statistically similar to the DM customer sample. Since the bank targets a subsample of the WI population to receive a direct mail solicitation (those with credit scores greater than 640), our analysis concentrates on the subset of WI customers identical to the DM customers with the exception that they did not receive a solicitation. In this context, the direct mail solicitation is the experimental “treatment,” and our goal is to assess whether it has any impact on mortgage choice. Under the null hypothesis that consumers are financially rational and choose debt contracts based on the prevailing economic environment, we should not observe a difference in the factors affecting the mortgage choice between the two groups. We begin by matching the 108,117 walk-in consumers to the 31,749 direct mail consumers using the nearest centroid sorting algorithm (see, Anderberg, 1973, and Hartigan, 1985) based on the Euclidean distances computed over all demographic and financial variables within a zip code.39 Once we obtain the probability of the distance to the centroid, we rank order the 108,117 WI observations and choose the closest 31,749 accounts. Thus, the clustering procedure produces a sample of WI consumers that matches the DM consumers along these financial, demographic, and geographical variables. 39 As a robustness check, we also computed the Euclidean distances over a set of five pre-determined ‘key’ variables. The sorting algorithm produced an approximately 99 percent overlap between the respective WI subsamples. As a result, the results reported below are not qualitatively different. 22 Table VIII reports the descriptive statistics for the DM customers and the matched WI sample. It is clear from examining the mean values in Table VIII that the matching algorithm produces a WI sample that closely resembles the DM customers in terms of credit quality, loan amount, house value, income, and borrower age. For example, the average FICO scores and loan-to-value ratios of the two groups are within approximately 1 point of each other, and the difference in the average borrower incomes is about 2 percent ($2,462). To further demonstrate that the nearest centroid sorting algorithm produced samples that are closely matched on all observable information, Table IX shows the descriptive statistics for the DM and WI customers based on the type of credit contract selected. Comparing the mean values for the WI and DM customers based on the selected product reveals little economic difference between the two groups. VI.A Evaluation of Customer Choice We estimate separate logit models of borrower choice for the matched walk-in and direct mail samples. We then compare the marginal effects to determine the sensitivity of borrowers to the independent variables based on whether or not they received a solicitation. Effectively, this method is equivalent to estimating a single model over both samples and interacting a dummy variable for direct mail with each variable. Table X reports the consumer choice logit model for the matched walk-in sample. Not surprisingly, the results are consistent with the results reported in Table V. The marginal effects indicate that WI borrowers are sensitive to changes in the interest rate environment. For example, a one point increase in the fixed-rate reference interest rate results in a 14.3 percent jump in the probability that the borrower will select the adjustable-rate line-of-credit. Furthermore, we also see that the borrower’s intended use of funds significantly impacts their product choice. Borrowers intending to use the home equity funds for consumption are 12.8 percent more likely to choose the variable rate line-of-credit, while borrowers indicating that they are refinancing existing debt are 20.3 percent more likely to select the fixed-rate loan. Table XI shows the estimated coefficients and marginal effects for the borrower choice model estimated on the direct mail sample. In contrast to the WI borrowers, we first notice that none of the independent financial and demographic variables have marginal effects above 10 percent. Furthermore, many of the key variables identified in the WI sample are no longer statistically 23 significant. For example, neither the term structure variable (rate difference) or the FRM reference interest rate are statistically significant. This result suggests that, in direct contrast to the WI borrowers, the direct mail customers are not basing their mortgage choice decision on the key factors identified by theory. Comparing the marginal effects in Tables X and XI indicates that the intended use of the loan funds affects both the WI and DM customers’ choice. However, we see that DM customers’ choice is less sensitive to their stated use of funds. For example, DM borrowers indicating that they are refinancing are 8.3 percent more likely to select the fixed-rate product, while the refinancing WI customers are 20.3 percent more likely to select the fixed-rate product. Consumption borrowers also display a similar, but less dramatic, difference. We included the dummy variable Line Solicitation in this model to isolate the impact of the type of direct mail offer sent to the customer.40 The marginal effect clearly indicates that this variable has the largest impact on the customer’s choice. Customers who receive a line-of-credit direct mail solicitation are 44.7 percent more likely to select the variable rate line-of-credit than the fixed-rate loan product. The impact of this variable far exceeds the effect of any of the other variables. Thus, it appears that the bank’s solicitation even significantly dampens the effect of the customers’ intended use of the funds. Finally, we examine the choice of borrowers who received a direct mail solicitation, but chose the product not advertised. Did these customers ignore the bank’s direct marketing cue and select the product consistent with prior theoretical predictions? Table XII reports the results from this model. Again, we compare the marginal effects to the baseline WI customers to identify any differences in sensitivity. The results in table XII clearly indicate that the DM customers who switched are similar to the WI customers in that they are sensitive to the interest rate environment. The marginal impact of a one point increase in the yield curve results in an 11.1 percent increase in the probability that they will select the variable rate product. This result compares favorably with the 10.3 percent effect observed for the WI customers (in Table X). Similarly, we see that a one point increase in the reference fixed-rate mortgage rate increases the probability of selecting the variable rate product by 8.2 percent (compared with 14.3 percent for the WI customers) in Table X. Finally, we also note that consumption and refinancing motivations have the same effects on 40 Line Solicitation equals one if the borrower received a variable rate line-of-credit offer and zero otherwise. 24 the DM switching customers as the WI customers. Thus, the results from Table XII are consistent with the complementary view of advertising. These customers responded to the bank’s offer letter, but still reacted to the economic environment in selecting the product. VI.B Persuasive versus Informative Advertising As noted in Section 5.2, our analysis makes a compelling case that the bank’s direct mail solicitation was largely persuasive and not informative. In this section we explore this distinction in greater detail. Our objective is to identify the applicants who were persuaded or informed by the bank’s solicitation. We focus our analysis on the 24,751 customers (78 percent of the direct mail customers) who selected the product that was advertised (i.e. those who did not switch products) as these were the individuals most likely to be persuaded or informed by the bank’s advertising campaign.41 In order to determine the product that should have been selected, we use the estimated coefficients from the matched walk-in sample model (Table X) to generate a prediction of whether the customer should select the adjustable-rate or fixed-rate product.42 We then compare the customer’s model prediction to their actual selection. Table XIII reports the frequency of persuaded versus informed consumers. Based on our classification scheme, we see that 74 percent of the borrowers were effectively “persuaded” by the bank’s direct mail solicitation. That is, these borrowers selected the product that was featured on the solicitation but was opposite the one predicted by the model. However, we also note that 26 percent of the customers were “informed” by the bank’s solicitation, since they selected the product predicted by the model and it was also featured on the solicitation. Although the analysis above suggests that 74 percent of the DM customers were persuaded to select a product that was counter to the one predicted by our model, it is possible that our model has a high predictive error rate resulting in a large Type II error. Thus, to gain a greater appreciation for whether model predictive error can explain these results, we examine the model predictive accuracy using a hold-out sample of customers that were not exposed to the bank’s direct mail solicitation. Recall that the above analysis is based on the borrower choice model for the matched 41 By definition, the 6,998 customers who selected the product opposite to the one that was advertised in their solicitation letter could not have been persuaded. For the advertisement to be persuasive, the customer would have to select the product that was featured on the solicitation. 42 We use a 50 percent cutoff criteria to determine whether the customers should select the adjustable-rate product. 25 sample of 31,749 walk-in customers, leaving 76,386 walk-in customers as a defacto hold-out sample. Thus, by estimating the predicted product choice for the hold-out walk-in customers, we are able to observe an unbiased estimate of the model’s predictive accuracy. Table XIV reports the results of this test. The results clearly indicate that the model’s predictive accuracy (using the 50 percent cutoff criteria) is very high. Table XIV shows that the model is able to predict the actual product choice for 85 percent of the customers implying a Type II error rate of 15 percent. In contrast, the predictive error rate for the direct mail customers is 74 percent. We feel that this is compelling evidence that the bank’s marketing campaign did have a persuasive effect. One concern with our conclusion is that we may be attributing a persuasive effect to the bank’s marketing campaign for borrowers who may not care that they selected the “wrong” product simply because the costs associated with making an “incorrect” decision are trivial. For example, our analysis could classify borrowers as being “persuaded” even if they originated a line-of-credit in order to have ready access to funds in the future. These borrowers would clearly not select a fixedrate product, even if the economic environment pointed to it as the optimal choice, since they would not be utilizing the funds immediately. In order to test whether this effect could be responsible for our results, table XV shows the average takedown (or utilization) rates at origination, month 12, and month 24 for the matched WI and DM customers who selected the line-of-credit and actually originated a loan. If the DM customers viewed the costs associated with the line-of-credit as trivial, then we would expect to find their utilization rates substantially lower than the WI customers. The results clearly reveal that the average utilization rates for WI and DM customers are comparable and thus do not support the hypothesis that the costs associated with selecting the line-of-credit are trivial. As a final test of whether the advertising was persuasive or informative, we examine the ex post origination performance of the applications that were actually booked. If the advertising was persuasive such that it caused borrowers to select the wrong product, then we would expect to observe these customers learning about their mistake and making adjustments accordingly. To test for this effect, we examine the loan prepayment rates over the three months after origination. The three month window is a sufficiently short period that exogenous factors (such as changes in interest rates or household mobility) should have a minimal impact on borrower prepayment decisions. If the persuaded borrowers learn that they selected the incorrect product, then we should observe a 26 higher prepayment rate for these borrowers than for borrowers that we identified as being informed or complemented.43 To examine the differences in prepayment, we identified all applications that ultimately resulted in loans or lines being booked. We note that approximately 89 percent of the DM and WI matched sample applications resulted in booked loans or lines (28,099 DM customers and 28,256 WI customers, respectively). For the customers identified by our model as being persuaded by the bank’s solicitation, we note that approximately 90 percent of the applications resulted in booked loans or lines. Similarly, approximately 86 percent and 87 percent of the complemented and informed customer applications, respectively, resulted in booked loans or lines. Turning first to the persuaded customers, we observed that 707 prepaid during the three months after origination, implying an unconditional prepayment rate of 4.3 percent. In contrast, we observed an unconditional three-month prepayment rate of 2.9 percent for the complemented (switchers) and informed customers.44 In contrast, the unconditional three-month prepayment rate for the walk-in customers is 1.7 percent.45 Table XVI reports the estimated coefficients for a simple logistic prepayment model.46 Using the empirical mortgage performance literature to provide guidance in specifying the independent variables in the prepayment model, we estimated the following model: P r(hi = 1) = βXi + αAi + ε, (7) where hi equals one if the mortgage prepays during the three-month period following origination, 0 k and zero otherwise; Xi is a matrix of explanatory variables, Ai = [APi , AIi , AC i ] with Ai (k = P, I, C) equaling one if borrower i was identified as being persuaded (P ), informed (I), or complemented (C).47 Following Agarwal et al. (2006), we include in the set of explanatory variables (X), a series of variables designed to capture the financial incentives to repay the loan. The variables include the value of the borrower’s prepayment option (OPTION ), an indicator of whether the prepayment 43 We note again that the home equity loans and lines were “no fee” products. Thus, the borrowers were able to repay their loans and lines without penalty. 44 We observed 172 prepayments out of 6,018 complemented borrowers (or switchers) and 163 prepayments out of 5,619 informed borrowers. 45 491 out of the 28,256 walk-in loans booked prepaid within the first three months after origination. 46 Given the short time horizon of our prepayment model, we estimated the prepayment model using a logistic framework rather than with a hazard rate model. 47 The walk-in customers are the reference category. 27 option is “in-the-money” (InMoney), and a variable (DSpread ) that captures the interaction of between InMoney and OPTION.48 Not surprising, given that we examine only the three-month period after origination, none of the financial variables are significant, indicating that changes in the economic environment over the three months after origination did not impact borrower prepayment behavior. However, consistent with the theory that persuaded borrowers may have recognized that they selected the “wrong” product, we see that the coefficient on the variable indicating that the borrower was persuaded is positive and significant. The marginal effects suggest that the prepayment rate for persuaded borrowers is almost four times as high as the prepayment rate experienced by the walk-in customers. Furthermore, the coefficients indicating whether the borrower was informed or complemented are insignificant, suggesting that the three-month prepayment rate for these borrowers is not statistically different from the prepayment rate for the walk-in customers. Thus, our analysis shows that the borrowers most likely to have made a mistake by following the bank’s advertising cue (the borrowers identified as being persuaded) are significantly more likely to quickly prepay out of this product than the typical walk-in customer not exposed to the bank’s solicitation. VII Conclusions Financial economists now recognize that marketing and persuasion can have important effects on consumer decisions. In this paper, we examine the effect of direct mail (or junk mail) advertising on individual financial decisions by studying consumer choice of debt contracts. The results from our analysis suggest that financial variables underlying the relative pricing of debt contracts are the leading factors explaining consumer debt choice. Furthermore, we also find that the intended use of the debt proceeds significantly affects consumer choice. In particular, we find that borrowers who intend to use the debt proceeds for consumption are 13 percent more likely to select the adjustable-rate line-of-credit and borrowers who are refinancing existing debt are approximately 17 percent less likely to choose the adjustable-rate line-of-credit than the borrowers 48 The OPTION variable follows Deng et al. (2000) and is calculated as OP T IONi = Vi − Vi∗ , Vi (8) where Vi is the present value of the remaining mortgage payments at the current market interest rate and Vi∗ is the present value of the remaining mortgage payments at the contract interest rate. 28 who intend to use the funds for home improvements. With respect to the impact of advertising on borrower choice, we find evidence that the lender’s advertising campaign had a persuasive effect on consumer contract choice. We arrive at this conclusion based on a variety of tests. First, coefficient estimates from a consumer choice model reveal that receipt of a line-of-credit direct mail solicitation increased the probability of selecting a variable rate line-of-credit contract by 17.7 percent while receiving the fixed-rate loan direct mail solicitation reduced the odds of selecting the variable rate product by 14.8 percent. Second, analysis based on a matched sample method reveals that none of the financial and demographic variables that are important for the control group’s product selection have an impact on the product choice decision for the direct mail sample. In fact, the customers who received a direct mail line-of-credit solicitation are 44.7 percent more likely to select the line-of-credit product than the fixed-rate product. Third, analysis of the product choice model coefficients for the group of borrowers who ignored the bank’s direct mail solicitation and selected the product not advertised reveals that the key financial and demographic variables have the same signs and magnitude as the control group. This finding suggests that the advertisement had a complementary effect for this set of borrowers. Fourth, for the group of borrowers we identified as likely being persuaded, we find that the odds of prepayment over the three-month period after loan origination is almost four times higher than the prepayment rate experienced by the control group. Fifth, the estimated coefficient for the sample selection correction parameter reveals that borrower self-selection to respond to the bank’s advertisement does not affect our analysis. Finally, we also note that a substantial (one-third) portion of the consumers who received a direct mail solicitation did not view the offer as persuasive because they remained sensitive to the economic environment as theory predicts. Thus, the evidence indicates that these consumers viewed the direct mail advertisement as complementary to their decision-making process. The results from this study suggest that further research is needed in order to understand the reactions of individuals to various information cues. For example, in the wake of the on-going financial crisis in the mortgage and housing markets, banking and consumer regulatory agencies are exploring the issue of information disclosure in the residential mortgage market.49 Thus, being able to identify individuals who are most susceptible to financial advertisements may aid in identifying 49 For example, the Federal Trade Commission Bureau of Economics recently held a conference (“Consumer Information and the Mortgage Market,” May 29, 2008) in order to explore issues associated with consumer mortgage knowledge and consumer understanding of mortgage disclosures. 29 the type of information that is critical to making informed financial decisions. Yet, the results from this study indicate that the responses to bank marketing campaigns vary across individuals, implying that any regulatory action should reflect the heterogeneous responses of individuals to financial information. 30 VIII References Agarwal, S., B. W. Ambrose, S. Chomsisengphet, and C. Liu. 2006. An empirical analysis of home equity loan and line performance. Journal of Financial Intermediation, 15: 444-469. Agarwal, S., B. W. Ambrose, S. Chomsisengphet, and C. Liu. 2008. The Role of Soft Information in a Dynamic Contract Setting: Evidence from the Home Equity Credit Market. Working Paper, Pennsylvania State University. Agarwal, Sumit, Souphala Chomsisengphet, Chunlin Liu and Nicholas S. Souleles 2005. Do Consumers Choose the Right Credit Contracts? Mimeo, The Wharton School, University of Pennsylvania. Agarwal, Sumit, John Driscoll, Xavier Gabaix, David Laibson, 2007. Age of Reason: Financial Decisions over the Life Cycle Mimeo, Harvard University. Ai, C. and E.C. Norton. 2003. Interaction Terms in Logit and Probit Models. Economic Letters, 80: 123-129. Alm, J.R. and J.R. Follain. 1987. Consumer Demand for Adjustable-Rate Mortgages. Housing Finance Review, 6 1-16. Anderberg, M.R. 1973. Cluster Analysis for Applications, New York: Academic Press, Inc. Ausubel, L. 1999. Adverse Selection in the Credit Card Market. Working Paper, University of Maryland. Bagwell, K. 2007. The Economic Analysis of Advertising. in Handbook of Industrial Organization, Vol. 3 edited by M. Armstrong and B. Porter, North-Holland: Amsterdam. Barber, B.M., and T. Odean. 2008. All that Glitters: The Effect of Attention and News on the Buying Behavior of individual and Institutional Investors. Review of Financial Studies, 21:2 785-818. Barber, B.M., T. Odean, and L. Zheng. 2005. Out of Sight, Out of Mind: The Effects of Expenses on Mutual Fund Flows. Journal of Business, 78:6 2095-2119. Bertrand, M., D. Karlan, S. Mullainathan, E Shafir, and J. Zinman. 2006. What’s Psychology Worth? A Field Experiment in Consumer Credit Markets. Working Paper, Harvard University. Brown, J.R., N. Liang, and S. Weisbenner. 2007. Individual Account Investment Options and Portfolio Choice: Behavioral Lessons From 401(k) Plans. Journal of Public Economics, 91:10 1992-2013 Brueckner, J. 1986. The Pricing of Interest Rate Caps and Consumer Choice in the market for Adjustable-Rate Mortgages. Housing Finance Review, 16 119-136. Brueckner, J. 1993. Why Do We Have Arms? Journal of the American Real Estate and Urban Economics Association, 21:3 333-345. Brueckner, J. and J. Follain J. 1988. The Rise and Fall of the Arm - An Econometric- Analysis of 31 Mortgage Choice. Review of Economics and Statistics, 70:1 93-102. Brueckner, J. and J. Follain. 1989. Arms and the Demand for Housing. Regional Science and Urban Economics, 19:2 163-187. Braithwaite, D. 1928. The Economic Effects of Advertisement. Economic Journal, 38 16-37. Campbell, J. Y. 2006. Household Finance. Journal of Finance, LXI, 1553-1604 Campbell, J. Y., and Joao F. Cocco, 2003, Household risk management and optimal mortgage choice, Quarterly Journal of Economics, 118 1449-1494. Canner, G. B., Durkin, T.A., and C.A. Luckett. 1998. Recent developments in home equity lending, Federal Reserve Bulletin, 241-251. Chan, S. 1996. Residential Mobility and Mortgages. Regional Science and Urban Economics, 26:3-4 287-311. Chiang, R., Y. Chow, and M. Liu. 2002. Residential Mortgage Lending and Borrower Risk: The Relationship Between Mortgage Spreads and Individual Characteristics. Journal of Real Estate Finance and Economics, 25:1 5-32. Cronqvist, H. 2006. Advertising and Portfolio Choice. Working paper, The Ohio State University. Deng, Y., J.M. Quigley, and R. Van Order, 2000. Mortgage Terminations, Heterogeneity and the Exercise of Mortgage Options. Econometrica 68:2 310-331. Dhillon, U., J. Shilling, and C.F. Sirmans. 1987. Choosing Between Fixed and Adjustable Rate Mortgages. Journal of Money, Credit and Banking, 19 260-267. Gabriel, S. and S. Rosenthal. 1993. Adjustable-Rate Mortgages, Household Mobility, and Homeownership - A Simulation Study. Journal of Real Estate Finance and Economics, 7:1 29-41. Greenspan, A. 2004. Understanding household debt obligations at the Credit Union National Association 2004 Governmental Affairs Conference, Washington, D.C. (February 23, 2004) [http://www.federalreserve.gov/boardDocs/speeches/2004/20040223/default.htm]. Grullon, G., G. Kanatas, and J.P. Weston. 2004. Advertising, Breadth of Ownership, and Liquidity. Review of Financial Studies 17:2 439-461. Hartigan, J.A. 1985. Statistical Theory in Clustering. Journal of Classification, 2 63-76. Hong, H., J.D. Kubik, and J.C. Stein. 2005. Thy Neighbor’s Portfolio: Word-of-Mouth Effects in the Holdings and Trades of Money Managers. Journal of Finance, 60:6 2801-2824. Jain, P.C. and J.S. Wu. 2000. Truth in Mutual Fund Advertising: Evidence on Future Performance and Fund Flows. Journal of Finance 55:2 937-958. Kahneman,D. and A. Tversky. 1979. Prospect Theory: An Analysis of Decision under Risk. Econometrica 47:2 263-292. Kressel, L.M., G.G. Chapman, and E. Leventhal. 2007. The Influence of Default Options on the Expression of End-Of-Life Treatment Preferences in Advance Directives. Journal of General 32 Internal Medicine, 22:7 1007-1010. Koijen, R.S.J., O. van Hemert, and S. van Nieuwerburgh. 2006. Mortgage Timing. Working paper, Tilburg University. Madrain, B.C. and D.F. Shea, 2001. The Power of Suggestions: Inertia in 401(k) Participation and Savings Behavior. Quarterly Journal of Economics, 116:4 1149-1187. Mullainathan, S., J Schwartzstein, and A. Shliefer. 2006. Coarse Thinking and Persuasion. Working Paper, Harvard University. Mullainathan, S. and A. Shliefer. 2005. Persuasion in Finance. Working Paper, Harvard University. Nelson, P. 1974. Advertising as Information. The Journal of Political Economy 82:4 729-754. Nelson, P. 1975. The Economic Consequences of Advertising. The Journal of Business 48:2 213241. Ozga, S.A. 1960. Imperfect Markets Through Lack of Knowledge. Quarterly Journal of Economics, 74 29-52. Perry, V.G. and C.M. Motley. 2008. Reading the Fine Print: An Analysis of Mortgage Advertising Messages. Working Paper, George Washington University. Posey, L. and A. Yavas. 2001. Adjustable and Fixed Rate Mortgages as a Screening Mechanism for Default Risk. Journal of Urban Economics, 49:1 54-79. Reuter, J. and E. Zitzewitz. 2006. Do Ads Influence Editors? Advertising and Bias in the Financial Media. Quarterly Journal of Economics, 121:1 197-227. Robinson, J. 1933. Economics of Imperfect Competition, London: MacMillan and Co. Rosenthal, S. and P. Zorn. 1993. Household Mobility, Asymmetric Information, and the Pricing of Mortgage Contract Rates. Journal of Urban Economics 33 235-253. Russo, J.E., K.A. Carlson, and M.G. Meloy. 2006. Choosing an Inferior Alternative. Psychological Science, 17:10 899-904. Sa-Aadu, J. and J. Shilling. 1994. Tests of Borrower Perceptions in the Adjustable-Rate Mortgage Market - Do Borrowers View Arm Contracts as Distinct? Journal of Urban Economics, 36:1 8-22. Sa-Aadu, J. and C. Sirmans. 1995. Differentiated Contracts, Heterogeneous Borrowers, and the Mortgage Choice Decision. Journal of Money Credit and Banking, 27:2 498-510. Shapiro, J. 2006. A ‘memory-jamming’ theory of advertising. Working Paper, University of Chicago. Shiller, R.J. 2008. The Subprime Solution. Princeton, NJ: Princeton University Press. Shiller, R.J., and J. Pound. 1989. Survey Evidence on Diffusion of Interest and Information Among Investors. Journal of Economic Behavior and Organization, 12 47-66. Sirri, E.R. and P. Tufano. 1998. Costly Search and Mutual Fund Flows. Journal of Finance, 53:5 33 1589-1622. Stigler, G. 1961. The Economics of Information. Journal of Political Economy 69 213-225. Tversky, A. and D. Kahneman. 1986. Rational Choice and the Framing of Decisions. The Journal of Business 59:4 S251-S278. Tversky, A. and D. Kahneman. 1981. The Framing of Decisions and the Psychology of Choice. Science 211:4481 453-458. van Hemert, O. 2006. Life-Cycle Housing and Portfolio Choice with Bond Markets. Working paper, New York University. van Hemert, O., F. de Jong, and J. Driessen. 2005. Dynamic Portfolio and Mortgage Choice for Homeowners. Working paper, New York University. Vickery, J. 2006. Interest Rates and Consumer Choice in the Residential Mortgage Market. Working paper, Federal Reserve Bank of New York. 34 Table I: Summary Statistics for Direct Mail Customers Bank Solicitations Consumer Response Loans Lines Loans Lines Variables Mean Std Mean Std Mean Std Mean Std FICO Score 722.6 42.8 729.7 48.6 714.8 37.0 726.8 44.8 State MA 22.7% 30.2% 20.8% 28.6% 24.7% 28.6% 21.0% 27.2% State CT 10.8% 37.3% 13.0% 33.6% 10.8% 15.7% 14.9% 13.7% State ME 5.7% 8.7% 6.7% 9.7% 6.1% 6.2% 6.1% 4.9% 4.4% 20.4% 4.7% 21.2% 5.0% 14.7% 4.4% 14.7% State NH State NJ 8.4% 27.7% 10.3% 11.2% 7.1% 14.8% 10.7% 7.7% 35.5% 47.6% 33.0% 37.5% 33.7% 38.5% 30.7% 35.5% State NY State PA 4.9% 9.6% 4.2% 5.2% 4.5% 4.7% 4.5% 7.5% State RI 7.2% 25.7% 6.8% 25.0% 6.8% 24.7% 6.9% 17.6% Frequency .981 Million 2.072 Million 11,249 20,500 Note: This table shows the mean credit scores (FICO) for the recipients of the bank’s direct mail soliciation and the customers who responded to the solicitation. The table also reports the frequency distribution of mailings and responses by customer location (State). 35 Table II: Summary Statistics for Walk-In and Direct Mail Customers Variables Customer LTV Appraised LTV Borrower Estimated Home Value Appraised Home Value Requested Loan Amount Loan Amount Approved APR FICO Score Debt to Income Consumption Refinancing Years on the Job Income Borrower Age First Mortgage Balance Months at Address Self Employed Retired Home Maker Married Frequency Walk-In Direct Mail Mean Std 61.88 25.49 64.68 28.21 $329,521 $236,802 $318,491 $202,334 $66,664 $50,431 $67,279 $51,631 5.46 1.06 728.02 50.50 38.17 18.99 30% 42% 46% 48% 7.85 9.01 $122,241 $160,425 46.51 12.57 $143,361 $110,230 99.20 129.26 7% 26% 8% 27% 1% 11% 53% 50% 108,117 Mean Std 70.31 19.98 71.40 19.08 $299,334 $221,874 $286,330 $189,785 $60,291 $40,428 $61,123 $40,592 5.45 1.06 724.40 41.03 27.89 14.49 28% 42% 48% 48% 4.57 2.81 $110,694 $71,919 46.52 11.25 $132,991 $108,955 86.14 26.62 6% 23% 2% 13% 1% 10% 48% 50% 31,749 Note: Customer LTV is the loan-to-value ratio based on the requested loan amount and the borrower’s self-reported house value. Appraised LTV is the bank’s loan-to-value ratio based on the approved loan amount and independent property appraisal. APR is the effective contract interest rate on the loan product selected. FICO is the borrower’s credit quality score at application; Debtto-income is the ratio of the borrower’s debt to total income; Income is the borrower’s income at date of application; Consumption is a dummy variable indicating that the borrower intends to use the loan/line proceeds for consumption purposes; Refinancing is a dummy variable indicating that the borrower is using the loan/line proceeds to refinance existing debt; Self Employed, Retired, Home Maker, and Married are dummy variables indicating the borrower’s respective employment and marriage status. 36 37 Direct Mail (Response) Loans Lines 37.58 37.41 (5.33) (5.24) $50,698.97 $50,595.60 ($19,617.34) ($19,436.17) 50.32% 50.40% (2.53%) (2.53%) 7.08% 7.02% (13.70%) (13.66%) 4,577 4,605 Walk-Ins Loans Lines 37.53 37.90 (5.26) (5.34) $50,337.68 $50,597.95 ($19,622.13) ($19,829.23) 50.32% 50.08% (2.50%) (2.50%) 7.05% 7.13% (13.69%) (13.96%) 4,833 4,792 Note: This table reports the demographic characteristics’ means and standard deviations (in parentheses) for the zip codes corresponding to the resident locations of the walk-in customers and the direct-mail recipients based on whether the borrower selected a line-of-credit or a loan. The columns under “Direct Mail (Mailings)” refer to the zip codes (neighborhoods) associated with recipients of the bank’s direct mail solicitation. The columns under “Direct Mail (Response)” refer to the zip codes (neighborhoods) of the customers who responded to the solicitation. The columns under “Walk-Ins” refer to the zip codes (neighborhoods) of the walk-in customers. Number of Zip Codes % Black % Female Median Income Median Age Direct Mail (Mailings) Loans Lines 37.71 37.70 (5.38) (5.19) $50,400.83 $50,677.17 ($19,266.53) ($19,540.89) 50.86% 50.15% (2.57%) (2.56%) 6.99% 7.00% (13.65%) (13.55%) 4,622 4,668 Table III: Demographic Characteristics of Walk-In and Direct Mail Customers’ Locations Table IV: First-Stage Sample Selection Model Intercept Solicitation FICO Score Zip Code Fixed Effects Campaign Fixed Effects Number of Observations Pseudo-Rsq Coefficient Value -0.5381 -0.0005 Yes Yes 3.01 Million 17.92% Standard Error 0.0194 0.0002 P-value <.0001 0.0271 Marginal Impact -0.01% Notes: This table reports the maximum-likelihood parameter estimates for the first-stage bivariate probit model of whether the household responded to the direct mail solicitation. The dependent variable in is a dummy variable equal to 1 if the household responded to the solicitation and 0 otherwise. The independent variable, FICO, is the borrower’s credit quality score at the time of the bank’s solicitation. The Standard Errors are corrected for Heteroskadisticity. 38 Table V: Second Stage Bivariate Probit Model of Borrower Home-Equity Choice Coeff. Value -8.2318 Standard Error 0.2565 P-value <.0001 0.3542 0.2245 0.0179 0.0193 <.0001 <.0001 12.37% 13.25% Loan-to-Value Variables: Ln(Borrower Estimate of the House Value) Ln(Loan Amount Requested) 0.0053 0.0046 0.0017 0.0007 <.0001 <.0001 1.43% 1.56% Borrower State Use of Funds: Consumption Refinancing 1.1023 -0.9851 0.0239 0.0206 <.0001 <.0001 13.17% -16.76% Borrower Characteristics: Application FICO Debt to Income Ln(Income) Borrower Age Years on the Job Ln(First Mortgage Balance) Ln(Months at Address) Self Employed Retired Home Maker Married 0.0063 -0.0147 0.0659 0.0010 0.0539 0.0657 0.0272 0.4025 -0.4191 -0.3379 -0.1686 0.0002 0.0005 0.0187 0.0008 0.0009 0.0110 0.0081 0.0333 0.0429 0.0685 0.0163 <.0001 <.0001 0.0005 <.0001 <.0001 <.0001 0.0006 <.0001 <.0001 <.0001 <.0001 0.04% -0.03% 6.12% 0.18% 0.31% 4.84% 4.85% 2.27% -0.97% -0.32% -1.96% Direct Mail Solicitation Effects: DM Line Offer Dummy DM Loan Offer Dummy 0.2264 -0.2122 0.0556 0.0359 <.0001 <.0001 17.74% -14.67% DM Line * FICO Score DM Loan * FICO Score 0.0026 -0.0036 0.0005 0.0005 <.0001 <.0001 0.03% -0.05% DM Lines * Log(Income) DM Loans * Log(Income) 0.0049 -0.0058 0.0437 0.0730 0.9418 0.9940 0.02% -0.01% DM Line * Rate Difference DM Loan * Rate Difference -0.2102 -0.1934 0.1969 0.5202 0.5830 0.8971 -0.01% -0.02% DM Line * FRM APR % DM Loan * FRM APR% -0.2633 -0.2097 0.5337 0.2722 0.9360 0.4780 -0.01% -0.02% DM Line * Consumption DM Loan * Consumption DM Line * Refinancing DM Loan * Refinancing ρ Month Loan Application Dummies State Location Control Dummies Number of Observations Pseudo R-sq 0.1993 0.4709 -0.1085 -0.3333 -0.4251 Yes Yes 139,866 24.82% 0.0678 0.1942 0.0441 0.0878 0.2926 <.0001 0.0058 0.0291 <.0001 0.2714 6.83% 1.10% -1.90% -6.92% Intercept Economic Environment Variables: Rate Difference (FRM% - ARM%) FRM APR% Marginal Impact Notes: This table reports the maximum-likelihood parameter estimates for the bivariate probit model of borrower home-equity choice. The dependent variable equals one if the borrower selected a variable rate line-of-credit and zero otherwise. Application FICO is the borrower’s credit quality score at application; Rate Difference is the difference between the home-equity loan interest rate and the home-equity line-of-credit interest rate prevailing at the time of application; FRM APR is the fixed-rate home equity loan interest rate; Debt-to-income is the ratio of the borrower’s debt to total income; Income is the borrower’s income at date of application; Consumption is a dummy variable indicating that the borrower intends to use the loan/line proceeds for consumption purposes; Refinancing is a dummy variable indicating that the borrower is using the loan/line proceeds to refinance existing debt; Self Employed, Retired, Home Maker, and Married are dummy variables indicating the borrower’s respective employment and marriage status. Following Ai and Norton (2003), the marginal effects for the interaction variables as are calculated as: ∂ 2 Φ(·) ∂Ii ∂XiR 0 00 = πR Φ (·) + (β + πR Ii )(α + πR XiR )Φ (·) and The Standard Errors are corrected for Heteroskadisticity. 39 ∂ 2 Φ(·) ∂Ii ∂XiE 0 00 = πE Φ (·) + (β + πE Ii )(α + πE XiE )Φ (·). 40 Direct Mail Fixed-Rate Home Equity Loans Mean Std 74.89 20.78 75.84 19.99 $219,525 $159,651 $210,580 $134,598 $46,862 $29,024 $44,563 $26,247 8.13 1.18 716.14 34.42 31.88 11.93 18% 39% 77% 42% 3.25 2.53 $90,085 $66,596 46.18 12.10 $108,787 $83,558 78.89 86.16 6% 23% 2% 15% 1% 10% 48% 50% 8,874 - No Switch Variable Rate Home Equity Lines Mean Std 65.52 21.66 66.55 20.75 $338,120 $253,598 $323,056 $218,227 $66,193 $44,093 $68,842 $46,361 4.57 1.03 730.38 46.25 27.97 15.69 24% 43% 46% 50% 5.02 2.41 $112,312 $76,223 46.82 11.04 $136,660 $116,414 86.46 90.65 6% 24% 2% 13% 1% 10% 50% 50% 15,877 Note: Customer LTV is the loan-to-value ratio based on the requested loan amount and the borrower’s self-reported house value. Appraised LTV is the bank’s loan-to-value ratio based on the approved loan amount and independent property appraisal. APR is the effective contract interest rate on the loan product selected. FICO is the borrower’s credit quality score at application; Debt-to-income is the ratio of the borrower’s debt to total income; Income is the borrower’s income at date of application; Consumption is a dummy variable indicating that the borrower intends to use the loan/line proceeds for consumption purposes; Refinancing is a dummy variable indicating that the borrower is using the loan/line proceeds to refinance existing debt; Self Employed, Retired, Home Maker, and Married are dummy variables indicating the borrower’s respective employment and marriage status. Variables Customer LTV Appraised LTV Borrower Estimated Home Value Appraised Home Value Requested Loan Amount Loan Amount Approved APR FICO Score Debt to Income Consumption Refinancing Years on the Job Income Borrower Age First Mortgage Balance Months at Address Self Employed Retired Home Maker Married Frequency Direct Mail - Switch Fixed-Rate Variable Rate Home Equity Loans Home Equity Lines Mean Std Mean Std 76.59 20.75 82.48 12.78 77.51 19.57 83.95 11.92 $191,562 $144,906 $253,901 $179,175 $184,142 $112,317 $243,508 $153,031 $51,853 $35,112 $52,603 $38,481 $48,610 $29,132 $50,510 $34,981 8.24 1.13 5.61 1.06 705.72 38.71 719.24 44.72 30.25 11.68 23.34 13.06 15% 36% 53% 43% 81% 39% 24% 50% 3.16 2.33 4.48 4.67 $102,969 $53,361 $125,409 $65,256 46.60 12.12 45.69 11.02 $92,004 $81,711 $151,488 $111,018 83.16 74.71 94.87 101.67 4% 19% 5% 22% 5% 21% 1% 11% 1% 8% 1% 10% 45% 50% 45% 50% 2,375 4,623 Table VI: Summary Statistics for the Direct Mail Customers Who Did and Did Not Switch Products at Origination Table VII: Analysis of Decision to Switch Away from Product Offered in Direct Mail Solicitation Coeff. Val. Std. Err. P-value Marg Impact Intercept 8.2849 0.0701 <.0001 Loan-to-Value Variables: Ln(Borrower Estimate of the House Value) Ln(Loan Amount Requested) -0.0040 0.0020 0.0006 0.0004 <.0001 <.0001 1.45% 0.10% Borrower Stated Use of Funds: Consumption Refinancing -0.2410 0.0042 0.0450 0.0141 <.0001 0.7722 -5.64% 2.71% Borrower Characteristics: FICO 0.0209 0.0014 <.0001 0.48% Debt to Income -0.0238 0.0026 <.0001 0.00% Ln(Income) 0.0220 0.0098 0.0258 2.07% Borrower Age -0.1259 0.0123 <.0001 -3.29% Years on the Job 0.3505 0.0109 <.0001 0.10% Ln(First Mortgage Balance) -0.0132 0.0095 0.1694 -1.97% Ln(Months at Address) -0.1035 0.0080 <.0001 -0.63% Self Employed -0.0744 0.0024 <.0001 -0.04% Retired -0.0382 0.0050 <.0001 -0.17% Home Maker -0.0006 0.0002 0.0002 -0.06% Married 1.4110 0.0076 <.0001 0.00% Month Loan Origination Dummies Yes State Location Control Dummies Yes Number of Observations 31,749 Pseudo R-sq 7.93% Notes: This table reports the maximum-likelihood parameter estimates for the logit model of whether the customer selected the alternative product from the one contained in the direct mail solicitation. The dependent variable is a dummy variable equal to 1 if the customer switched and 0 otherwise. FICO is the borrower’s credit quality score at application; Debt-to-income is the ratio of the borrower’s debt to total income; Income is the borrower’s income at date of application; Consumption is a dummy variable indicating that the borrower intends to use the loan/line proceeds for consumption purposes; Refinancing is a dummy variable indicating that the borrower is using the loan/line proceeds to refinance existing debt; Self Employed, Retired, Home Maker, and Married are dummy variables indicating the borrower’s respective employment and marriage status. 41 Table VIII: Summary Statistics for the Matched Sample Walk-In (WI) and Direct Mail (DM) Customers Variables Customer LTV Appraised LTV Borrower Estimated Home Value Appraised Home Value Requested Loan Amount Loan Amount Approved APR FICO Score Debt to Income Consumption Refinancing Years on the Job Income Borrower Age First Mortgage Balance Months at Address Self Employed Retired Home Maker Married Frequency Walk-In Mean Std 71.36 24.16 72.37 26.45 $293,693 $137,769 $282,975 $155,315 $57,038 $19,188 $57,927 $17,540 5.87 1.16 722.80 39.35 31.34 20.27 26% 40% 50% 47% 4.41 7.77 $112,335 $70,896 47.24 11.40 $127,492 $68,007 87.59 110.17 7% 25% 5% 21% 1% 9% 50% 50% 31749 Direct Mean 70.31 71.40 $299,334 $286,330 $60,291 $61,123 5.45 724.40 27.89 28% 48% 4.57 $110,694 46.52 $132,991 86.14 6% 2% 1% 48% 31749 Mail Std 19.98 19.08 $221,874 $189,785 $40,428 $40,592 1.06 41.03 14.49 42% 48% 2.81 $71,919 11.25 $108,955 26.62 23% 13% 10% 50% Note: The walk-in sample was created using the nearest centroid sorting algorithm based on the Euclidean distances computed over all demographic and financial variables within a zip code. Customer LTV is the loan-to-value ratio based on the requested loan amount and the borrower’s self-reported house value. Appraised LTV is the bank’s loan-to-value ratio based on the approved loan amount and independent property appraisal. APR is the effective contract interest rate on the loan product selected. FICO is the borrower’s credit quality score at application; Debt-to-income is the ratio of the borrower’s debt to total income; Income is the borrower’s income at date of application; Consumption is a dummy variable indicating that the borrower intends to use the loan/line proceeds for consumption purposes; Refinancing is a dummy variable indicating that the borrower is using the loan/line proceeds to refinance existing debt; Self Employed, Retired, Home Maker, and Married are dummy variables indicating the borrower’s respective employment and marriage status. 42 43 Variable Rate Home Equity Lines Walk-In Direct Mail Mean Std Mean Std 70.7 23.5 69.10 19.79 72.3 26.6 70.22 18.88 $324,164 $138,623 $320,339 $237,885 $312,367 $159,017 $306,261 $204,462 $60,585 $19,372 $63,324 $42,908 $62,432 $16,725 $64,972 $43,958 5.09 1.15 4.79 1.03 725.8 38.0 728.02 45.93 30.4 20.7 26.99 15.14 30% 43% 30% 43% 40% 48% 41% 50% 4.7 7.5 4.90 2.89 $118,414 $71,895 $115,077 $73,908 46.6 11.3 46.58 11.04 $134,547 $69,539 $139,790 $115,274 90.6 106.4 88.24 92.98 8% 27% 6% 24% 5% 22% 2% 12% 1% 9% 1% 10% 49% 50% 49% 50% 22,728 20,500 Note: The walk-in sample was created using the nearest centroid sorting algorithm based on the Euclidean distances computed over all demographic and financial variables within a zip code. Customer LTV is the loan-to-value ratio based on the requested loan amount and the borrower’s self-reported house value. Appraised LTV is the bank’s loan-to-value ratio based on the approved loan amount and independent property appraisal. APR is the effective contract interest rate on the loan product selected. FICO is the borrower’s credit quality score at application; Debt-to-income is the ratio of the borrower’s debt to total income; Income is the borrower’s income at date of application; Consumption is a dummy variable indicating that the borrower intends to use the loan/line proceeds for consumption purposes; Refinancing is a dummy variable indicating that the borrower is using the loan/line proceeds to refinance existing debt; Self Employed, Retired, Home Maker, and Married are dummy variables indicating the borrower’s respective employment and marriage status. Variables Customer LTV Appraised LTV Borrower Estimated Home Value Appraised Home Value Requested Loan Amount Loan Amount Approved APR FICO Score Debt to Income Consumption Refinancing Years on the Job Income Borrower Age First Mortgage Balance Months at Address Self Employed Retired Home Maker Married Frequency Fixed-rate Home Equity Loans Walk-In Direct Mail Mean Std Mean Std 72.9 25.2 75.25 20.77 72.5 26.2 76.20 19.90 $216,922 $133,379 $213,621 $156,538 $208,924 $148,383 $204,998 $129,894 $48,102 $18,981 $47,916 $30,309 $46,579 $19,724 $45,417 $26,856 7.85 1.13 8.15 1.17 715.3 43.1 713.94 35.33 33.7 19.0 31.53 11.88 17% 33% 18% 38% 77% 44% 78% 41% 3.7 8.5 3.23 2.49 $97,020 $69,009 $92,805 $63,801 48.8 11.8 46.27 12.10 $109,717 $63,618 $105,244 $83,168 80.1 117.4 79.79 83.74 5% 22% 5% 23% 4% 20% 3% 16% 1% 7% 1% 9% 53% 50% 47% 50% 9,021 11,249 Table IX: Summary Statistics for the Matched Sample Walk-In (WI) and Direct Mail (DM) Customers Based On Credit Choice Table X: Consumer Choice Between Fixed- and Adjustable-Rate Home Equity Walk-In Consumers Coeff. Val. -9.6807 Std. Err. 1.4173 P-value <.0001 Marg Impact Economic Environment Variables: Rate Difference (FRM% - ARM%) FRM APR% 0.3060 0.1365 0.0472 0.0246 <.0001 <.0001 10.34% 14.29% Loan-to-Value Variables: Ln(Borrower Estimate of the House Value) Ln(Loan Amount Requested) 0.0024 0.0014 0.0002 0.0002 <.0001 <.0001 1.92% 1.79% Borrower Stated Use of Funds: Consumption Refinancing 0.8121 -1.2050 0.0896 0.0630 <.0001 <.0001 12.80% -20.26% 0.0062 -0.0167 0.0748 0.0036 0.0205 0.1943 0.2090 0.0468 -0.3269 -0.1392 -0.2489 Yes Yes 31,749 12.39% 0.0007 0.0020 0.0184 0.0012 0.0032 0.0415 0.0293 0.0879 0.1487 0.0326 0.0554 <.0001 <.0001 <.0001 <.0001 <.0001 <.0001 <.0001 0.8905 0.0173 <.0001 <.0001 0.06% -0.04% 8.73% 0.28% 0.26% 6.87% 4.99% 3.90% -1.44% -0.43% -1.69% Intercept Borrower Characteristics: FICO Debt to Income Ln(Income) Borrower Age Years on the Job Ln(First Mortgage Balance) Ln(Months at Address) Self Employed Retired Home Maker Married Month Loan Origination Dummies State Location Control Dummies Number of Observations Pseudo R-sq Notes: This table reports the maximum-likelihood parameter estimates for the logit model of whether the customer selected the adjustable-rate product. The model is estimated using the matched sample walk-in customers. The dependent variable is a dummy variable equal to 1 if the WI customer selected the ARM and 0 otherwise. FICO is the borrower’s credit quality score at application; Debt-to-income is the ratio of the borrower’s debt to total income; Income is the borrower’s income at date of application; Consumption is a dummy variable indicating that the borrower intends to use the loan/line proceeds for consumption purposes; Refinancing is a dummy variable indicating that the borrower is using the loan/line proceeds to refinance existing debt; Self Employed, Retired, Home Maker, and Married are dummy variables indicating the borrower’s respective employment and marriage status. 44 Table XI: Consumer Choice Between Fixed- and Adjustable-Rate Home Equity Direct Mail Consumers Coeff. Val. -4.0226 Std. Err. 0.5145 P-value <.0001 Marg Impact Economic Environment Variables: Rate Difference (FRM% - ARM%) FRM APR% -0.0784 -0.1026 0.3229 0.1480 0.8684 0.5548 -1.65% -0.92% Loan-to-Value Variables: Ln(Borrower Estimate of the House Value) Ln(Loan Amount Requested) 0.0022 0.0024 0.0008 0.0003 <.0001 <.0001 1.68% 1.24% Borrower Stated Use of Funds: Consumption Refinancing 1.0304 -0.7038 0.0889 0.0429 <.0001 <.0001 8.78% -8.25% FICO Debt to Income Ln(Income) Borrower Age Years on the Job Ln(First Mortgage Balance) Ln(Months at Address) Self Employed Retired Home Maker Married 0.0013 -0.0262 0.3434 0.0048 0.0092 0.0719 0.0338 0.2672 -0.6141 -0.0615 -0.1176 0.0005 0.0022 0.0870 0.0020 0.0060 0.0335 0.0028 0.0713 0.1268 0.0320 0.0705 <.0001 <.0001 0.0001 0.0119 0.1879 0.0277 <.0001 <.0001 <.0001 0.0399 0.0843 0.04% -0.04% 3.87% 0.27% 0.38% 5.68% 3.94% 1.27% -1.25% -0.22% -0.85% Direct Mail Solicitation Effects: Line Solicitation Month Loan Origination Dummies State Location Control Dummies Number of Observations Pseudo R-sq 2.1426 Yes Yes 31,749 10.37% 0.5740 <.0001 44.71% Intercept Borrower Characteristics: Notes: This table reports the maximum-likelihood parameter estimates for the logit model of whether the Direct Mail customer selected the adjustable-rate product. The dependent variable is a dummy variable equal to 1 if the DM customer selected the ARM and 0 otherwise. FICO is the borrower’s credit quality score at application; Debt-to-income is the ratio of the borrower’s debt to total income; Income is the borrower’s income at date of application; Consumption is a dummy variable indicating that the borrower intends to use the loan/line proceeds for consumption purposes; Refinancing is a dummy variable indicating that the borrower is using the loan/line proceeds to refinance existing debt; Self Employed, Retired, Home Maker, and Married are dummy variables indicating the borrower’s respective employment and marriage status. Line Solicitation is a dummy variable equal to 1 if the customer received a line-of-credit offer and 0 otherwise. 45 Table XII: Consumer Choice Between Fixed- and Adjustable-Rate Home Equity Direct Mail Switchers Coeff. Val. -5.0475 Std. Err. 1.5669 P-value <.0001 Marg Impact Economic Environment Variables: Rate Difference (FRM% - ARM%) FRM APR% 0.2707 0.1385 0.0476 0.0131 <.0001 <.0001 11.10% 8.15% Loan-to-Value Variables: Ln(Borrower Estimate of the House Value) Ln(Loan Amount Requested) 0.0062 0.0039 0.0002 0.0001 <.0001 0.0203 2.01% 1.09% Borrower Stated Use of Funds: Consumption Refinancing 1.1114 -0.7210 0.1131 0.0553 <.0001 <.0001 10.49% -12.20% 0.0015 -0.0288 0.1969 0.0041 0.0164 0.0866 0.0028 0.0093 -0.4596 -0.1987 -0.0636 Yes Yes 6,998 18.92% 0.0006 0.0027 0.0525 0.0025 0.0077 0.0373 0.0008 0.0849 0.1854 0.1987 0.0859 <.0001 <.0001 <.0001 0.0972 0.0192 0.0243 <.0001 0.9333 0.0147 0.4028 0.5821 0.02% -0.03% 4.05% 0.11% 0.13% 2.06% 0.94% 0.29% -1.31% -0.24% -0.24% Intercept Borrower Characteristics: FICO Debt to Income Ln(Income) Borrower Age Years on the Job Ln(First Mortgage Balance) Ln(Months at Address) Self Employed Retired Home Maker Married Month Loan Origination Dummies State Location Control Dummies Number of Observations Pseudo R-sq Notes: This table reports the maximum-likelihood parameter estimates for the logit model of whether the Direct Mail customers who switched products selected the adjustable-rate product. The dependent variable is a dummy variable equal to 1 if the DM ‘switch’ customer selected the ARM and 0 otherwise. FICO is the borrower’s credit quality score at application; Debt-to-income is the ratio of the borrower’s debt to total income; Income is the borrower’s income at date of application; Consumption is a dummy variable indicating that the borrower intends to use the loan/line proceeds for consumption purposes; Refinancing is a dummy variable indicating that the borrower is using the loan/line proceeds to refinance existing debt; Self Employed, Retired, Home Maker, and Married are dummy variables indicating the borrower’s respective employment and marriage status. 46 Table XIII: Analysis of Persuaded versus Informed Consumers Number Predicted FRM Predicted ARM Mailed Selected FRM Selected ARM Selected FRM Selected ARM Mailed ARM 15,877 ··· 12,336 ··· 3,541 50% 14% ··· ··· Mailed FRM 8,874 2,918 5,956 12% 24% Notes: This table reports the frequency of customers identified as either persuaded or informed. For all direct mail borrowers who did not switch products, we estimated the predicted product that they should select based on the coefficients from the walk-in mortgage choice model (Table X). We then compare the predicted choice to the actual choice based on whether the borrower received an adjustable-rate (ARM) or fixed-rate (FRM) solicitation. The persuaded borrowers selected the product opposite to the one predicted and are noted in bold. Informed borrowers selected the product that is consistent with the one predicted and are noted in italics. Borrowers following into the (· · ·) cells selected the predicted product and thus ignored the bank’s advertising. Table XIV: Walk-in Customer Prediction Error Rate Actual Selection ARM FRM Total Predicted ARM 48,345 85.5% Selection FRM 8,167 14.5% 3,056 15.4% 51,401 16,800 84.6% 24,967 Total 56,512 19,856 76,386 Notes: This table reports the predicted walk-in customer selection error rate. For all walk-in customers who were not included in the matched-sample analysis, we estimated the predicted product that they should select based on the coefficients from the matched sample walk-in mortgage choice model (Table X). We then compare the predicted choice to the actual choice. Effectively, this test provides an indication of the predictive accuracy of the estimated mortgage choice model using a hold-out sample. Table XV: Average Line-Of-Credit Takedown Rate Walk-in Direct Mail Month 0 58.9% 60.3% Month 12 63.2% 63.7% Month 24 67.1% 66.0% Notes: This table reports the average line-of-credit takedown (utilization) rate at origination, month 12, and month 24. The utilization rate is the amount of funds drawn expressed as a percent of total credit line available. 47 Table XVI: Prepayment Behavior of Home Equity Loans and Lines Intercept Log(FICO) LTV OPTION InMoney DSpread Persuaded Informed Complemented Other Controls State Dummies Time Dummies Number Prepayed Number of Observations Pseudo R-sq Coeff. Value -3.1833 0.1844 0.0318 0.0184 -0.0311 0.0188 0.4831 0.1758 0.1472 Yes Yes Yes 1,533 56,355 0.0173 Standard Error 0.3723 0.1373 0.0388 0.0178 0.1346 0.0691 0.1839 0.1592 0.1958 t-stat -8.55 1.34 0.82 1.03 -0.23 0.27 2.63 1.10 0.75 Marginal Effects 1.828 0.187 0.384 -0.157 0.066 3.992 0.950 0.738 Note: This table reports the maximum-likelihood parameter estimates for the logistic three-month prepayment model. The dependent variable equals one if the borrower prepaid the mortgage during the three month period following origination and zero otherwise. FICO is hte borrower’s credit quality score at application; LTV is the loan-to-value ratio at application; OPTION is the value of the borrower’s prepayment option reflecting the difference between the market rate of interest and the contract interest rate; InMoney indicates whether the prepayment option is ‘in-the-money’; DSpread is the interaction of InMoney and Option; Persuaded, Informed, and Complemented are indicator variables denoting whether the borrower was persuaded, informed or complemented by the bank’s direct mail solicitation (walk-in customers are the reference category.) Standard Errors are corrected for Heteroskadisticity. 48 6 5 APR (Percent) 4 3 2 FRM APR 1 0 1/2/2003 FRM-ARM Difference 4/2/2003 7/2/2003 10/2/2003 1/2/2004 4/2/2004 7/2/2004 10/2/2004 Months Figure 1: FRM and FRM-APR Interest Rate Differential Between January 2003 and December 2004 49 P(ARM - Consumption WI) P(ARM - Consumption DM Line Offer) P(ARM - Consumption DM Loan Offer) 100% 90% 80% ARM Percentage 70% 60% 50% 40% 30% 20% 10% 11/2/04 9/2/04 7/2/04 5/2/04 3/2/04 1/2/04 11/2/03 9/2/03 7/2/03 5/2/03 3/2/03 1/2/03 0% Months Figure 2: Probability of Choosing an ARM for Consumption Motive Borrowers with varying FRM APR and Rate Differences 50 P(ARM - Refinancing WI) P(ARM - Refinancing DM Line Offer) P(ARM - Refinancing DM Loan Offer) 100% 90% 80% ARM Percentage 70% 60% 50% 40% 30% 20% 10% 11/2/04 9/2/04 7/2/04 5/2/04 3/2/04 1/2/04 11/2/03 9/2/03 7/2/03 5/2/03 3/2/03 1/2/03 0% Months Figure 3: Probability of Choosing an ARM for Refinancing Borrowers with varying FRM APR and Rate Difference 51 Working Paper Series A series of research studies on regional economic issues relating to the Seventh Federal Reserve District, and on financial and economic topics. Firm-Specific Capital, Nominal Rigidities and the Business Cycle David Altig, Lawrence J. Christiano, Martin Eichenbaum and Jesper Linde WP-05-01 Do Returns to Schooling Differ by Race and Ethnicity? Lisa Barrow and Cecilia Elena Rouse WP-05-02 Derivatives and Systemic Risk: Netting, Collateral, and Closeout Robert R. Bliss and George G. Kaufman WP-05-03 Risk Overhang and Loan Portfolio Decisions Robert DeYoung, Anne Gron and Andrew Winton WP-05-04 Characterizations in a random record model with a non-identically distributed initial record Gadi Barlevy and H. N. Nagaraja WP-05-05 Price discovery in a market under stress: the U.S. Treasury market in fall 1998 Craig H. Furfine and Eli M. Remolona WP-05-06 Politics and Efficiency of Separating Capital and Ordinary Government Budgets Marco Bassetto with Thomas J. Sargent WP-05-07 Rigid Prices: Evidence from U.S. Scanner Data Jeffrey R. Campbell and Benjamin Eden WP-05-08 Entrepreneurship, Frictions, and Wealth Marco Cagetti and Mariacristina De Nardi WP-05-09 Wealth inequality: data and models Marco Cagetti and Mariacristina De Nardi WP-05-10 What Determines Bilateral Trade Flows? Marianne Baxter and Michael A. Kouparitsas WP-05-11 Intergenerational Economic Mobility in the U.S., 1940 to 2000 Daniel Aaronson and Bhashkar Mazumder WP-05-12 Differential Mortality, Uncertain Medical Expenses, and the Saving of Elderly Singles Mariacristina De Nardi, Eric French, and John Bailey Jones WP-05-13 Fixed Term Employment Contracts in an Equilibrium Search Model Fernando Alvarez and Marcelo Veracierto WP-05-14 1 Working Paper Series (continued) Causality, Causality, Causality: The View of Education Inputs and Outputs from Economics Lisa Barrow and Cecilia Elena Rouse WP-05-15 Competition in Large Markets Jeffrey R. Campbell WP-05-16 Why Do Firms Go Public? Evidence from the Banking Industry Richard J. Rosen, Scott B. Smart and Chad J. Zutter WP-05-17 Clustering of Auto Supplier Plants in the U.S.: GMM Spatial Logit for Large Samples Thomas Klier and Daniel P. McMillen WP-05-18 Why are Immigrants’ Incarceration Rates So Low? Evidence on Selective Immigration, Deterrence, and Deportation Kristin F. Butcher and Anne Morrison Piehl WP-05-19 Constructing the Chicago Fed Income Based Economic Index – Consumer Price Index: Inflation Experiences by Demographic Group: 1983-2005 Leslie McGranahan and Anna Paulson WP-05-20 Universal Access, Cost Recovery, and Payment Services Sujit Chakravorti, Jeffery W. Gunther, and Robert R. Moore WP-05-21 Supplier Switching and Outsourcing Yukako Ono and Victor Stango WP-05-22 Do Enclaves Matter in Immigrants’ Self-Employment Decision? Maude Toussaint-Comeau WP-05-23 The Changing Pattern of Wage Growth for Low Skilled Workers Eric French, Bhashkar Mazumder and Christopher Taber WP-05-24 U.S. Corporate and Bank Insolvency Regimes: An Economic Comparison and Evaluation Robert R. Bliss and George G. Kaufman WP-06-01 Redistribution, Taxes, and the Median Voter Marco Bassetto and Jess Benhabib WP-06-02 Identification of Search Models with Initial Condition Problems Gadi Barlevy and H. N. Nagaraja WP-06-03 Tax Riots Marco Bassetto and Christopher Phelan WP-06-04 The Tradeoff between Mortgage Prepayments and Tax-Deferred Retirement Savings Gene Amromin, Jennifer Huang,and Clemens Sialm WP-06-05 2 Working Paper Series (continued) Why are safeguards needed in a trade agreement? Meredith A. Crowley WP-06-06 Taxation, Entrepreneurship, and Wealth Marco Cagetti and Mariacristina De Nardi WP-06-07 A New Social Compact: How University Engagement Can Fuel Innovation Laura Melle, Larry Isaak, and Richard Mattoon WP-06-08 Mergers and Risk Craig H. Furfine and Richard J. Rosen WP-06-09 Two Flaws in Business Cycle Accounting Lawrence J. Christiano and Joshua M. Davis WP-06-10 Do Consumers Choose the Right Credit Contracts? Sumit Agarwal, Souphala Chomsisengphet, Chunlin Liu, and Nicholas S. Souleles WP-06-11 Chronicles of a Deflation Unforetold François R. Velde WP-06-12 Female Offenders Use of Social Welfare Programs Before and After Jail and Prison: Does Prison Cause Welfare Dependency? Kristin F. Butcher and Robert J. LaLonde Eat or Be Eaten: A Theory of Mergers and Firm Size Gary Gorton, Matthias Kahl, and Richard Rosen Do Bonds Span Volatility Risk in the U.S. Treasury Market? A Specification Test for Affine Term Structure Models Torben G. Andersen and Luca Benzoni WP-06-13 WP-06-14 WP-06-15 Transforming Payment Choices by Doubling Fees on the Illinois Tollway Gene Amromin, Carrie Jankowski, and Richard D. Porter WP-06-16 How Did the 2003 Dividend Tax Cut Affect Stock Prices? Gene Amromin, Paul Harrison, and Steven Sharpe WP-06-17 Will Writing and Bequest Motives: Early 20th Century Irish Evidence Leslie McGranahan WP-06-18 How Professional Forecasters View Shocks to GDP Spencer D. Krane WP-06-19 Evolving Agglomeration in the U.S. auto supplier industry Thomas Klier and Daniel P. McMillen WP-06-20 3 Working Paper Series (continued) Mortality, Mass-Layoffs, and Career Outcomes: An Analysis using Administrative Data Daniel Sullivan and Till von Wachter The Agreement on Subsidies and Countervailing Measures: Tying One’s Hand through the WTO. Meredith A. Crowley WP-06-21 WP-06-22 How Did Schooling Laws Improve Long-Term Health and Lower Mortality? Bhashkar Mazumder WP-06-23 Manufacturing Plants’ Use of Temporary Workers: An Analysis Using Census Micro Data Yukako Ono and Daniel Sullivan WP-06-24 What Can We Learn about Financial Access from U.S. Immigrants? Una Okonkwo Osili and Anna Paulson WP-06-25 Bank Imputed Interest Rates: Unbiased Estimates of Offered Rates? Evren Ors and Tara Rice WP-06-26 Welfare Implications of the Transition to High Household Debt Jeffrey R. Campbell and Zvi Hercowitz WP-06-27 Last-In First-Out Oligopoly Dynamics Jaap H. Abbring and Jeffrey R. Campbell WP-06-28 Oligopoly Dynamics with Barriers to Entry Jaap H. Abbring and Jeffrey R. Campbell WP-06-29 Risk Taking and the Quality of Informal Insurance: Gambling and Remittances in Thailand Douglas L. Miller and Anna L. Paulson WP-07-01 Fast Micro and Slow Macro: Can Aggregation Explain the Persistence of Inflation? Filippo Altissimo, Benoît Mojon, and Paolo Zaffaroni WP-07-02 Assessing a Decade of Interstate Bank Branching Christian Johnson and Tara Rice WP-07-03 Debit Card and Cash Usage: A Cross-Country Analysis Gene Amromin and Sujit Chakravorti WP-07-04 The Age of Reason: Financial Decisions Over the Lifecycle Sumit Agarwal, John C. Driscoll, Xavier Gabaix, and David Laibson WP-07-05 Information Acquisition in Financial Markets: a Correction Gadi Barlevy and Pietro Veronesi WP-07-06 Monetary Policy, Output Composition and the Great Moderation Benoît Mojon WP-07-07 4 Working Paper Series (continued) Estate Taxation, Entrepreneurship, and Wealth Marco Cagetti and Mariacristina De Nardi WP-07-08 Conflict of Interest and Certification in the U.S. IPO Market Luca Benzoni and Carola Schenone WP-07-09 The Reaction of Consumer Spending and Debt to Tax Rebates – Evidence from Consumer Credit Data Sumit Agarwal, Chunlin Liu, and Nicholas S. Souleles WP-07-10 Portfolio Choice over the Life-Cycle when the Stock and Labor Markets are Cointegrated Luca Benzoni, Pierre Collin-Dufresne, and Robert S. Goldstein WP-07-11 Nonparametric Analysis of Intergenerational Income Mobility with Application to the United States Debopam Bhattacharya and Bhashkar Mazumder WP-07-12 How the Credit Channel Works: Differentiating the Bank Lending Channel and the Balance Sheet Channel Lamont K. Black and Richard J. Rosen WP-07-13 Labor Market Transitions and Self-Employment Ellen R. Rissman WP-07-14 First-Time Home Buyers and Residential Investment Volatility Jonas D.M. Fisher and Martin Gervais WP-07-15 Establishments Dynamics and Matching Frictions in Classical Competitive Equilibrium Marcelo Veracierto WP-07-16 Technology’s Edge: The Educational Benefits of Computer-Aided Instruction Lisa Barrow, Lisa Markman, and Cecilia Elena Rouse WP-07-17 The Widow’s Offering: Inheritance, Family Structure, and the Charitable Gifts of Women Leslie McGranahan WP-07-18 Demand Volatility and the Lag between the Growth of Temporary and Permanent Employment Sainan Jin, Yukako Ono, and Qinghua Zhang WP-07-19 A Conversation with 590 Nascent Entrepreneurs Jeffrey R. Campbell and Mariacristina De Nardi WP-07-20 Cyclical Dumping and US Antidumping Protection: 1980-2001 Meredith A. Crowley WP-07-21 The Effects of Maternal Fasting During Ramadan on Birth and Adult Outcomes Douglas Almond and Bhashkar Mazumder WP-07-22 5 Working Paper Series (continued) The Consumption Response to Minimum Wage Increases Daniel Aaronson, Sumit Agarwal, and Eric French WP-07-23 The Impact of Mexican Immigrants on U.S. Wage Structure Maude Toussaint-Comeau WP-07-24 A Leverage-based Model of Speculative Bubbles Gadi Barlevy WP-08-01 Displacement, Asymmetric Information and Heterogeneous Human Capital Luojia Hu and Christopher Taber WP-08-02 BankCaR (Bank Capital-at-Risk): A credit risk model for US commercial bank charge-offs Jon Frye and Eduard Pelz WP-08-03 Bank Lending, Financing Constraints and SME Investment Santiago Carbó-Valverde, Francisco Rodríguez-Fernández, and Gregory F. Udell WP-08-04 Global Inflation Matteo Ciccarelli and Benoît Mojon WP-08-05 Scale and the Origins of Structural Change Francisco J. Buera and Joseph P. Kaboski WP-08-06 Inventories, Lumpy Trade, and Large Devaluations George Alessandria, Joseph P. Kaboski, and Virgiliu Midrigan WP-08-07 School Vouchers and Student Achievement: Recent Evidence, Remaining Questions Cecilia Elena Rouse and Lisa Barrow WP-08-08 Does It Pay to Read Your Junk Mail? Evidence of the Effect of Advertising on Home Equity Credit Choices Sumit Agarwal and Brent W. Ambrose WP-08-09 6