SEARCH

SEARCH BY CITATION

Abstract

  1. Top of page
  2. Abstract
  3. Decision theory
  4. Method
  5. Results
  6. Discussion
  7. References

A choice-based conjoint experiment was employed to identify consumer preferences for automobile loan attributes. Respondents prefer loans with low interest rates and moderate contract lengths. Less important are high rebates and moderate down payments. Rebates are most important to choice when down payment is high. Even when choosing among interest-free loans, respondents do not prefer long terms, conflicting with traditional financial rationality. Respondents appear to focus attention on the first digit of the monthly payment (payment, a function of the other attributes, was provided to respondents). This tendency was stronger among those with less education. Public policy recommendations are discussed.

Last year [2005], the Consumer Bankers Association reports, loans longer than 60 months accounted for 55.3 percent of its members’ new-vehicle loans …. The 2005 figure reflects a surge of 10 percentage points in one year (Allen 2006).

Over 70% of new automobile purchases in the United States are financed in part by vendor loans (Dasgupta, Siddarth, and Silva-Risso 2007), and personal car loans have amounted to 34% of monthly nonmortgage debt in the United States (Heitfield and Sabarwal 2004). The changing array of financing alternatives and rebate offers complicates the process of purchasing a car and may have significant implications for the well-being of buyers. In the summer of 2006, zero-interest loan offers with payment terms as long as seventy-two months presented compelling opportunities to the extent that prices for credit purchases were similar to prices for immediate cash. However, the tone of press coverage concerning long-term loan offers has often been negative. For example, one article suggests that car buyers considering long-term loans should “run from them like the plague” (Bankrate.com 2006). Columnists and some experts have emphasized the likelihood of buyers quickly becoming “upside-down” (when the value of a car is less than the amount owing on the loan). Faced with such issues, consumers may have a difficult time making a financially rational choice in the utility-maximizing sense.

Consumer confusion in the face of complex, multidimensional price information such as auto loan financing offers is indicated by empirical research (Estelami 2001; Herrmann and Wricke 1998; Stango and Zinman 2006; Thaler 1985). For example, Herrmann and Wricke (1998) found that when asked to rate the relative attractiveness of different auto loan financing offers, respondents at best used only linear functions of down payments, monthly payments, and contract lengths, without even calculating the product of monthly payment and number of payments, let alone using discounted values.

Given these indications of financial irrationality and the increasing reliance on credit to finance durable goods purchases, “understanding the psychology of consumer judgments of credit arrangements is … growing in importance from both a marketing and public policy perspective” (Estelami 2001, p. 63). Accordingly, this study aims to increase our understanding of how consumers choose among auto financing alternatives. More specifically, it attempts to answer the following research questions: (1) What are consumers’ preferred levels of loan interest rate, contract length, down payment, and rebate? Do interactions among these attributes matter? And which attributes are weighted most heavily in loan selection? (2) Are the answers to the questions in (1) consistent with rationality? (3) Do the findings for (1) and (2) vary by demographic subgroup? and (4) What implications do the findings have for consumer self-protection and financial literacy education efforts?

To address these questions, a choice-based conjoint1 experiment was designed to reveal consumer preferences for hypothetical automobile financing offers that varied with respect to interest rate charged, required down payment, contract length in months, and rebate amount. The experiment was part of an Internet questionnaire that was administered to a demographically diverse sample. The results are compared with predictions of the traditional expected utility model and alternative models of choice to determine whether consumers are financially rational when it comes to selecting a loan. A review of the decision theory and consumer choice literature in the next section helps to develop hypotheses contrasting “financially rational” decision makers with those who act in accordance with alternative models. The experimental design and analysis methods are then described, followed by the results and a discussion of their implications for public policy.

Decision theory

  1. Top of page
  2. Abstract
  3. Decision theory
  4. Method
  5. Results
  6. Discussion
  7. References

Traditional model of expected utility maximization

According to this model, individuals facing a choice of contract provisions for a loan of a given size will select the contract with the lowest present value of payments. For individuals with substantial liquid assets, the appropriate discount rate with which to calculate that present value equals the after-tax rate available on savings. For those who are already in debt, the discount rate would be the marginal rate on alternative sources of funds.2

Because everyone has a discount rate exceeding zero,3 an immediate implication is that when choosing among loans offering fixed zero-percent contractual rates, one should select a repayment schedule that delays payment as long as possible. In particular, one should select the lowest available down payment and longest available contract length. For an interest-free loan, the appeal of longer payment schedules should be clear to rational borrowers because they can invest the money they would otherwise use to make the higher payments on a shorter term loan. This choice requires no complicated arithmetic. Admittedly, there may be frictions in the market that play a role in such decisions, even for rational economic agents. For example, lenders require consumers to purchase comprehensive car insurance during the loan period. If the consumer views the benefit of that insurance to fall significantly short of the cost, then a shorter contract length may be preferable. This factor, however, is almost certainly of small order.

If one is instead choosing among loans with contractual rates above one’s appropriate discount rate, then rapid repayment (or rejection of all such loans) is the preferable course of action. Because typical car loan contracts involve no penalty for prepayment, however, there is no loss from accepting a long-term loan with a low down payment and then making extra payments on the loan. In fact, such a decision could be strictly optimal in an environment with substantial interest rate volatility because a loan with a long term could be held to maturity if market rates rose dramatically (or if one’s personal financial situation unexpectedly worsened) and could otherwise be paid off quickly as planned.4 This point, together with the fact that the time value of money is calculated with positive interest rates, is illustrated by Tables 1 and 2. Individuals who behave consistently with these predictions are characterized in this study as financially rational.

Table 1.  The Rationality of Preferred Loan Maturities
Loan Rate (Annual Percentage Rate) Relative to Discount RatePreferred Loan Term (Months)
ShortLong
  1. Note: This table presents a categorization of consumer preferences as rational or irrational. The reasoning assumes that a consumer chooses between multiple loans that all offer the same interest rate and other features but differ in contract length.

AboveFinancially rationalFinancially rational given prepayment option
BelowFinancially irrationalFinancially rational
Table 2.  The Rationality of Preferred Down Payments
Loan Rate (Annual Percentage Rate) Relative to Discount RatePreferred Down Payment
LowHigh
  1. Note: This table presents a categorization of consumer preferences as rational or irrational. The reasoning assumes that a consumer chooses between multiple loans that all offer the same interest rate and other features but differ in down payment.

AboveFinancially rationalFinancially rational given prepayment option
BelowFinancially irrationalFinancially rational

Mental accounting models

The predictions derived from the assumption of financial rationality are at odds with implications of some more recently developed frameworks for decision making. Prelec and Loewenstein (1998), in a model building upon Thaler’s (1985) work on mental accounting, claim that the satisfaction derived from consumption of a good is reduced by anticipation of future payments for the good. That is, there is some degree of coupling of costs with benefits. The disutility of making payments, on the other hand, is reduced by the anticipation of future consumption of the good being paid for. The tendency to focus on future rather than previous payments and consumption is described by Prelec and Loewenstein as prospective accounting. They also propose the assumption of prorating: the degree to which consumption enjoyment or payment pain is offset by anticipation of the counterpart depends on the ratio of remaining payments to remaining consumption. If, therefore, a durable good still has significant lifetime remaining, the happy anticipation of future consumption associated with making payments increases with each payment and is maximized at the final payment. Prelec and Loewenstein present evidence consistent with this view, namely that their survey respondents generally prefer making payments prior to scheduled vacations rather than paying afterward, even though discounted cash flow calculations would suggest the opposite preference. Further confirmation within the context of income tax withholding is provided by Ayers, Kachelmeier, and Robinson (1999). Such preferences might well drive automobile financing decisions.

Prelec and Loewenstein refer to the debt aversion resulting from coupling, prospective accounting, and prorating.5 In the case of durable goods purchases, they note that the motivation to fully prepay is weaker than it would be for short-lived goods, and many consumers are simply unable to pay cash for the cars they desire. Given the need to borrow, consumers may wish to complete payments on their loans as soon as feasible.

Moreover, issues of self-control may be sufficient that individuals will reject the flexibility of a long contract with prepayment option and instead constrain themselves to rapid repayment through the selection of relatively short-term contracts. Self-control would be a motive if, for example, discounting of utility were hyperbolic (e.g., Prelec and Loewenstein 1998).6 Constraining one’s self to a short repayment schedule ensures optimal payments for mental accounting purposes, but it may also ensure that one does not have future financial difficulties (e.g., loan default) due to the excessive consumption made possible by small loan payments. Mental accounting theory does not suggest that consumers always select the shortest possible loan length—present value considerations and inadequate savings may act counter to that outcome—but it does suggest that consumers tend to avoid loans with terms that stretch beyond the useful (and relatively repair-free) life of the automobile.

A similar conclusion follows from claims by Kamleitner and Kirchler (2006) based on survey analysis: cars are typically purchased earlier than necessary and hence are relatively hedonic (i.e., luxury purchases) in comparison with other durables. As such, they are harder to justify, and individuals are inclined toward short-term loans for items that are difficult to justify.

The notion of extremeness aversion discussed by Simonson and Tversky (1992) and by Kivetz (1999) may result in similar loan selection patterns. Consumers may tend to choose products or loans with attributes that are near the middle of their perceived ranges. Such decisions are made on the basis of ease of justification rather than utility maximization, even allowing for the mental accounting view of utility. The result would be a preference for down payments and contract lengths that are in the middle of their possible ranges, regardless of loan interest rate. For the purposes of our study, the foregoing discussion generates the following testable hypotheses against the null of financial rationality:

Hypothesis 1: Individuals selecting among loans will avoid loans with long contract lengths, even when such loans are offered at zero-percent interest.

Hypothesis 2: Individuals selecting among loans will avoid loans with zero down payment, even when such loans are offered at zero-percent interest.

Preference for odd prices

Consumers facing difficult choices involving multidimensional information may resort to heuristics to simplify decisions. Focusing on a low monthly payment is an example. Monthly payment may be a heuristic factor in loan decisions because many consumers think in terms of balancing monthly budgets.

Another such heuristic used by consumers may result in a phenomenon that is similar to the widely acknowledged consumer preference for “oddness” in quoted prices that is heavily exploited by sellers in setting price quotations such as $395 rather than $400. If examining three-digit monthly payments along with other loan attributes creates an overly heavy cognitive burden, consumers may tend to perceive car payments as if they were rounded down to the nearest multiple of $100 (Anderson and Simester 2003; Estelami 2001). If that is true, the preference for a loan payment such as $395 versus $400 will be stronger than the preference for $400 versus $405.

Boyes, Lynch, and Mounts (2007) pointed out that this outcome is consistent with predictions based on prospect theory (Kahneman and Tversky 1979), a theory that Thaler (1985) drew on in developing the conceptual basis for his mental accounting model. If individuals see a monthly payment of $395, they may actually round up to $400 for the most obvious reference price, the comparison price that anchors their thinking. Then, $395 represents a gain relative to the reference price. Utility is concave in gains according to prospect theory, so prices lower than $395, say $390, provide a lower marginal rate of appeal per dollar of gain. Boyes, Lynch, and Mounts (2007) provide evidence suggesting that people do round prices up when seeking reference prices, even rounding $16.47–$17.00 rather than $16.00, as an example.

The overall question of whether consumers use price comparison heuristics rather than financially rational valuation approaches is thus an empirically testable hypothesis as follows:

Hypothesis 3: Consumers strongly prefer loans with payments just below multiples of $100 over loans with payments at or just above multiples of $100, other factors held constant.

Weighting of loan attributes in decisions

Among both financially rational and irrational individuals, the relative importance placed on interest rate, down payment, contract length, and rebate can vary depending on factors including the individuals’ rates of time preference, current wealth, and existing borrowing and investment opportunities. In the absence of data on such factors, it is not normally possible to assess whether a consumer’s importance ranking is consistent with financial rationality. If interest rates were assigned very low importance, however, that would be evidence at least suggestive of excessive impatience and perhaps even irrationality.

The relative importance of various attributes across different demographic groups can provide evidence on another issue. As indicated by previous studies (e.g., Attanasio, Goldberg, and Kyriazidou 2008; Juster 1963), less affluent individuals and those otherwise likely to face financial constraints may more consistently avoid very short-term loans (and place correspondingly less weight on interest rates) than do unconstrained groups. At the same time, education may reduce the influence of the budget constraint on attribute importance weights or may otherwise increase the rationality of choices.

Method

  1. Top of page
  2. Abstract
  3. Decision theory
  4. Method
  5. Results
  6. Discussion
  7. References

Research design

Preferences for vendor-financed retail loan attributes were estimated in a choice-based conjoint7 experiment where the loan attributes subject to choice were contract length, interest rate, required down payment, and cash rebate.8 The design was full profile in the sense that all four of the attribute values were presented for each set of two loan alternatives that required a preferred loan choice by each respondent. The experiment obtained respondent choices on each of the twelve randomly generated screen displays, or choice tasks, and a fixed holdout screen, each of which presented two competing loan designs. The random displays were generated from 625 possibilities corresponding to five levels of four different loan attributes as presented in Table 3.

Table 3.  Attributes and Their Levels That Were Presented in Choice Tasks
AttributeAttribute Levels
  1. Note: This table shows the four loan contract attributes with their various levels. Each choice task presented two contracts with different levels for the attributes. For example, a choice task might involve electing between a contract of forty-eight months requiring a $2,500 down payment with a 4% interest rate and a $500 rebate versus a contract of sixty months requiring a $3,750 down payment with a 2% interest rate and a $3,000 rebate. Each respondent completed twelve such tasks with alternative contracts drawn from 1,025 possible contracts.

Contract length (months)3648607284
Down payment required ($)01,2502,5003,7505,000
Interest rate (%)02468
Cash rebate amount ($)3,0002,5001,0005000

The recorded choices on each of the twelve randomly generated choice tasks by each respondent were the inputs to a hierarchical Bayesian analysis that estimated the respondents’ utility functions across loan attributes. The responses to the fixed holdout screen were used to test the predictive power of the utility function estimates. The results formed the basis for determining preference shares of the respondents in simulations of competing loans.

The respondent sample

The respondents were solicited from a targeted population of U.S. adults who have purchased a new car on credit within the last five years or who planned to make such a purchase within the next two years. A random sample of individuals with these characteristics, balanced across the regions of the United States, was obtained from a commercial marketing research firm (Zoomerang, a division of MarketTools) that maintains an e-mail panel of over two million consumers. Their panel members opt in to respond to e-mail and Web-based surveys in return for a small fee and a chance to win a significant cash prize. An invitation to participate in the present study was prepared and e-mailed to the eligible panelists with brief background information on the subject of the study and the contingent reward system (cash prizes for completion of the questionnaire). The data were collected in January 2007. Of 2,721 individuals who received e-mail invitations to participate, 891 completed the online questionnaire, a response rate of 33%.

Experimental design

A fractional factorial,9 randomized experimental design was used to select the choice tasks for each respondent. A balanced overlap method employed random sampling with replacement for choosing loan contracts and permitted some attribute level overlap in screen displays (e.g., respondents would see two contracts that have the same contract length but differ with respect to interest rate, down payment, and rebate amount). The overlap increased the power of the test to detect attribute interactions (Chrzan and Orme 2000; Vriens, Oppewal, and Wedel 1998).

In addition to the twelve tasks that were randomly generated for each respondent as described above, a “practice” screen was placed as the first task in the questionnaire to allow respondents to gain familiarity with the type of questions that would follow. Data from this first screen were not used in the analysis of preference results. The same screen was also presented in the middle of the randomized choice tasks and used as a holdout task (not used in estimating the utility functions). Analysis of the responses to the holdout screen provided an indication of how well the utility values estimated from the twelve randomized tasks predicted each respondent’s actual holdout choices.

For each of the screen presentations, two different loans were presented side by side, and respondents were asked to indicate which they would choose if they were in the market to purchase a new car on credit that day, assuming these were the only two alternatives for financing their purchase. Respondents were also asked to assume that the price of the car in every alternative was $25,000 to remove any confounding effect of price on their choices. The instructions to the respondents and an example of a screen can be seen in Figure 1. The monthly payment that would apply was shown for each loan alternative in addition to the four attribute levels. The presentation order of the attributes for the loans remained the same for each respondent in order to make the information processing task easier and to limit the potential impact of fatigue on data quality. For the same reason, the number of tasks was limited to a practice screen plus the twelve randomly selected screens and the holdout screen described above.10

image

Figure 1. Example of a (Fixed) Choice Task

Download figure to PowerPoint

Presentation order was randomized across respondents so that presentation order should not have systematic effects such as primacy or recency on the experimental results.

Finally, the attribute value ranges included in the experimental design reflect typical car manufacturer offerings on these attributes as well as slightly more extreme values that are the subject of this investigation. The same number of levels was used for all attributes to achieve a balanced design.11 The validity of the results is enhanced to the extent that respondent choices are reflective of realistic values for important attributes that customers consider in their decision-making processes.

Use of the internet for data collection

Several factors recommended the use of the Internet to conduct the study. Computer-generated choice tasks are more easily randomized, and this form of data collection is much faster and easier for the respondents. Respondents required an average of twenty minutes to complete the online questionnaire. Several introductory screens were devoted to describing the purpose of the study, to assuring anonymity, and to explaining the choice tasks. An open-ended question was placed immediately after the choice screens to solicit respondent comments. This was followed by several questions asking respondents to self-report their creditworthiness (poor/satisfactory/excellent), types and levels of investments, and standard demographic information, all summarized in Table 4.

Table 4.  Sample Demographic Profile
Sample Characteristic
  1. Note: This table presents the demographic profile of the 891 respondents who completed the survey.

Gender (%)
 Male59
 Female41
Age (%)
 20–3412
 35–5446
 55–6430
 65 and over12
Marital status and homeownership (%)
 Married77
 Homeowners87
Education (%) 
 Bachelor’s degree or more54
 Some college or an associate’s degree36
Income (%)
 Less than $50,00023
  $50,000–$99,99948
  $100,000 or more29
Type of income (%)
 Salary/hourly wage54
 Some commission11
 Self-employed11
 Unemployed24
Average years in workforce since completing education25.8 (SD = 13)
Average number of dependents2.24 (SD = 1.28)
Mortgage/rent payment as percent of net income (take-home pay) (%)
 Less than 1%23
 1%–20%41
 21%–40%29
 41% or more7
Other monthly payments as percent of net income (take-home pay) (%)
 Less than 1%4
 1%–20%39
 21%–40%33
 41% or more23
Types of investments (%)
 No investments19
 Pension plans53
 Mutual funds49
 Individual retirement accounts53
 Bonds28
 Stocks44
Investment level (%)
 Less than $20,00029
  $20,000–$40,00013
 More than $40,00058
Monthly bill paying behavior (%)
 Always pay their bills on time83
 Occasionally late in paying bills15
Personal credit rating (self-report) (%)
 Excellent (700 points or higher)59
 Average/good (500–699 points)34
 Below average/poor (less than 499)7
Automobile buying behavior (%)
 Always purchase a car (vs. leasing)84
 Buy American cars44
 Buy Asian cars22
 Buy European cars5
 No preference29

Results

  1. Top of page
  2. Abstract
  3. Decision theory
  4. Method
  5. Results
  6. Discussion
  7. References

Sample characteristics

Table 4 presents a profile of the respondents. Over 70% of the sample was thirty-five to sixty-four years of age, and almost half reported an annual income of $50,000–$100,000. About 59% of the respondents were male, and 54% had at least a bachelor’s degree. The respondents had been in the workforce for an average of twenty-six years and had an average of 2.2 dependents. Investment levels were significant (58% had more than $40,000 invested), and investment types were diverse (over 40% of the respondents invested in pension plans, mutual funds, individual retirement account(s), and/or stocks). The majority of respondents stated that they always paid their bills on time (83%), and 60% believed their credit rating was excellent. Finally, respondents reported a preference to buy rather than lease an automobile (84% vs. 16%), and they tended to purchase American-made (44%) or Asian-made (22%) cars. The data indicate that the respondents were somewhat more educated and economically secure than the general population, so care should be taken in generalizing the study results to the U.S. population as a whole.

Multinomial logit and hierarchical Bayesian analysis of the experimental data

The twelve choices by each of the 891 respondents were the input to hierarchical Bayes (HB) estimations.12 The HB technique provided an indication of the heterogeneity in the population in the form of posterior means of the distributions for each individual’s part-worths and a more realistic fit for the data than an assumption that all individuals valued the various attributes equally. The individual-level heterogeneity that was captured facilitated analysis conditional on demographic characteristics of the respondents, although there were few major differences among demographic subgroups with regard to relative loan attribute importance (e.g., interest rate vs. length of the contract period) or preference for particular attribute levels (e.g., different interest rate levels or contract periods).

The HB analysis assumed that the vectors of individuals’ part-worths for various attribute levels and attribute interactions were drawn from a multivariate normal distribution. Given an individual’s vector of part-worths, βi, the probability that the individual selected a given loan option was multinomial logit, that is, inline image and inline image where pi1 is the probability that individual i chose loan 1 from among a pair of two loans. Further, the utility of loan j (j∈1, 2) was inline image where xjwas a vector of loan j’s attributes and βiwas individual i’s vector of part-worths. The mean vector and covariance matrix for the part-worths, along with the individual part-worths, were estimated iteratively. Initial estimates for the part-worths were approximately least squares estimates, and initial estimates for the parameters of the multivariate normal distribution were based on the distribution of initial part-worth estimates. The software conducted several thousand iterations in which the multivariate normal mean vector, covariance matrix, and the set of part-worths were each randomly updated conditional on the other current parameter estimates.

Both main and interaction effects models were examined to determine which one best fits the data because chi-square tests conducted prior to the HB analysis indicated that all main effects and three interaction terms were statistically significant in affecting the respondent choice (see Table 5). Root likelihood (RLH), a measure of goodness of fit, was calculated for each of the HB models. RLH for each individual was calculated as follows: the posterior means of an individual’s part-worths were used in the multinomial logit model to estimate the probability of each of the twelve choices made by that individual, and the geometric mean of those probabilities was taken. Hence, a model that placed 50% probability on all choices by a given individual would have RLH of .5. Model RLH was calculated as the arithmetic average of all the individual RLH13 values (hence it ignored the upper level normal distribution). RLH was 90% for the model that included the significant interaction terms compared to 82% for a model limited to main effects, and out-of-sample prediction was better for the model with interactions.

Table 5.  Relative Attribute Importances Derived from Hierarchical Bayes Estimation of Utilities
AttributeImportanceaWithin Attribute, χ2(df; p Value)b
  • a

    The relative importance of each attribute reflects how large an influence a product attribute has on choice of an auto finance package. Importance weights are calculated by computing the difference between the largest and the smallest part-worth for each respondent for each attribute, summing the differences, and normalizing to hundred. Attribute importances are ratio data. The importance weights do not add up to exactly hundred because the model includes three interaction terms and one attribute is deleted from this table (loan option, a dummy attribute).

  • b

    Chi-square values do not rely on hierarchical Bayes.

Interest rate, %40.061,141.65 (4; p < .01)
Contract length34.61177.91 (4; p < .01)
Cash rebate, dollars offered12.57112.053 (4; p < .01)
Down payment, dollars required12.7622.97 (4; p < .01)
Significant Second-Order Interactions
 Contract length × down payment ($) 74.99 (16; p < .01)
 Contract length × cash rebate ($) 50.64 (16; p < .01)
 Down payment × cash rebate ($) 43.19 (16; p < .01)

Relative attribute and attribute level importance

The relative importance of each product attribute is displayed in Table 5. The most important loan attribute was the interest rate, followed closely by the length of the contract period. The cash rebate and down payment values were each only one-third as important as the interest rate or contract length. Hypotheses about which loan attributes would be the most important in choice were not developed due to a lack of theoretical and empirical evidence, as noted earlier. However, the heavy weight placed on interest rates appears broadly consistent with rationality, necessitating more precise hypothesis tests to be discussed presently.

A simulation based on the HB-derived part-worth vectors for each respondent was used to estimate market choices (preferences shares) for different financing contract designs. Preference shares are defined as the percentage of respondents who would prefer (choose) each financing plan, given a specified set of attribute levels. The randomized first choice method (Huber, Orme, and Miller 1999) was used. It assumes that respondents tend to choose the products that provide them their highest overall utility (“first choice rule”), but it adds unique random error to the utilities in order to recognize the fact that individuals do not invariably choose the product that optimizes their utility. Each respondent is sampled many times to stabilize the share estimates (hundred thousand times in this study). The appropriateness of this method for the present study was validated using choice data collected from responses to the fixed holdout screen. Randomized first choice correctly estimated the aggregate revealed preferences in the holdout data within three percentage points.14

The results of these simulations reflect the average utilities of particular attribute levels as displayed in Table 6. The most strongly preferred or “ideal” car loan (i.e., the one with the greatest overall utility) across all respondents was one with no interest and a large cash rebate ($3,000), an intermediate contract length (sixty months), and a 10% down payment ($2,500). Respondent comments suggested that the preference for sixty-month contracts may have derived from a desire to avoid unusually long-term loans coupled with difficulty meeting the payments required for the very shortest term loans. A preference for shorter terms (or at least distaste for long-term loans) was one of the two decision-making factors most commonly cited by respondents along with lower rates. “I would NEVER consider 84 months since most cars don’t last that long” was a representative comment. The tension between short-term loans and financial reality is captured by another respondent’s statement, “I would like to pay off the loan in the shortest period possible without screwing up my budget.”

Table 6.  Ranking of Attribute Level Preferences by Average HB Estimated Utility
RankInterest Rate, % (Level [Part-Worth])Contract Length, Months (Level [Part-Worth])Cash Rebate, $ (Level [Part-Worth])Down Payment, $ (Level [Part-Worth])
  1. Note: These utilities are interval data that represent the average part-worths across all respondents. They are not the same as importance weights because the latter are computed for each respondent and then averaged across respondents (whereas a small range in these average part-worths for a given attribute could reflect the fact that some individuals seek high values of the attribute and others avoid it, while all view it as important). Within each attribute, utilities sum to zero. A negative part-worth for a level does not indicate that this level is unattractive but that it is less preferred than a level with a positive number. A main effects plus three specified two-way interactions model was used to generate the utilities presented here. This type of effects coding (zero-centered differences) allows assessment of the relative attractiveness of a loan by adding up the effects (or utilities) for its components.

10 (72.51)60 (26.77)3,000 (21.38)2,500 (8.24)
22 (37.62)48 (17.53)2,500 (14.85)1,250 (6.01)
34 (6.89)72 (10.32)1,000 (−5.07)0 (−1.36)
46 (−31.39)84 (−23.51)500 (−13.56)3,750 (−1.85)
58 (−85.63)36 (−31.11)0 (−17.60)5,000 (−11.05)

The impact of financial constraints on contract length choices was reflected in Figures 2 through 5, which graph average utilities for each attribute by demographic group. The rankings over attribute levels were very similar across groups, but loans with short terms and high down payments were more strongly disliked by groups that were more likely to have limited alternative sources of funds—those with no college education or with rent and other monthly obligations exceeding 50% of take-home pay. The opposite was true of those with significant amounts invested and, especially, those with all types of investments. These less constrained groups placed more importance on rebates and interest rates. Groups stratified by income displayed similar though somewhat weaker results. These differences in preferences were generally consistent with previous findings relating to individual differences in loan demand elasticity (Alessie, Hochguertel, and Weber 2005; Attanasio, Goldberg, and Kyriazidou forthcoming; Juster 1963; Karlan and Zinman 2008). Other characteristics, including gender, age, credit score, marital status, and time to complete the questionnaire, did not relate to substantial differences in utilities.

image

Figure 2. Utilities of Contract Lengths

Note: Average utilities as functions of contract length are displayed for the full sample (891 respondents), those with no college education (93), those with housing and other monthly obligations exceeding 50% of take-home pay (269), those with total investments above $40,000 (471), and those with investments in all five of the survey’s categories (112). Utilities are calculated using randomized first choice simulation on the hierarchical Bayes output.

Download figure to PowerPoint

image

Figure 3. Utilities of Down Payments

Note: Average utilities as functions of down payment are displayed for the full sample (891 respondents), those with no college education (93), those with housing and other monthly obligations exceeding 50% of take-home pay (269), those with total investments above $40,000 (471), and those with investments in all five of the survey’s categories (112). Utilities are calculated using randomized first choice simulation on the hierarchical Bayes output.

Download figure to PowerPoint

image

Figure 4. Utilities of Interest Rates

Note: Average utilities as functions of loan interest rate are displayed for the full sample (891 respondents), those with no college education (93), those with housing and other monthly obligations exceeding 50% of take-home pay (269), those with total investments above $40,000 (471), and those with investments in all five of the survey’s categories (112). Utilities are calculated using randomized first choice simulation on the hierarchical Bayes output.

Download figure to PowerPoint

image

Figure 5. Utilities of Rebates

Note: Average utilities as functions of rebate are displayed for the full sample (891 respondents), those with no college education (93), those with housing and other monthly obligations exceeding 50% of take-home pay (269), those with total investments above $40,000 (471), and those with investments in all five of the survey’s categories (112). Utilities are calculated using randomized first choice simulation on the hierarchical Bayes output.

Download figure to PowerPoint

Choices given zero-interest financing: Hypotheses 1 and 2

Combined with the lack of significant interaction effects found between interest rates and any of the other attributes, the implication of the preference rankings was that, even for zero-interest rate loans, consumers as a group prefer an intermediate contract length of sixty months and a 10% down payment of $2,500. Hypotheses 1 and 2 were supported by these results.

Because a lack of statistically significant interaction effects is not conclusive evidence of an absence of interaction, however, hierarchical Bayesian analysis and simulations were used while including all interaction effects to ascertain the influence of those interactions on point estimates of preference shares when long-term zero-interest loans were paired with short-term zero-interest loans. The results appear in Table 7. Setting the down payment and rebate at intermediate levels, 75% of individuals preferred sixty-month zero-interest loans over eighty-four-month zero-interest loans. This is only slightly different than the 78% preference share that holds in the model without interest rate interaction effects. Likewise, allowing for interactions, 71% of individuals would prefer to pay $2,500 down on a zero-interest loan rather than $0 down (not displayed in table).

Table 7.  Choice of Contract Length When Loans Are Interest Free
GroupContract Length (Months)Down Payment ($)Rebate ($)Interest Rate (%)Preference Share (%)SE (%)
  1. Note: Share of preference represents the percentage of respondents who would prefer or choose each auto loan “product,” assuming these were the only two choices available. Results are calculated using randomized first choice simulation on the hierarchical Bayes output.

Panel A: Only significant interaction effects included
Overall sample602,5001,000077.571.22
84 22.431.22
College educated602,5001,000083.951.41
84 16.051.41
Panel B: All interaction effects included
Overall sample602,5001,000074.731.32
84 25.271.32
College educated602,5001,000080.931.60
84 19.071.60

When considering only those respondents with at least an undergraduate degree, the preference share for the sixty-month loan is 81% when allowing for interest rate–contract length interaction effects and 84% without those interactions (see Table 7). This is consistent with the possibility that, among the few individuals who did choose the longer-term loan at zero interest, a large portion are relatively less educated, financially constrained individuals who choose the longer contract length as a means of easing constraints rather than as the optimal loan from a present value perspective. Simulations run on several other subgroups produced results consistent with that intuition, but the majority always preferred moderate loan terms.

These results confirm that, even allowing for interaction effects, financially irrational choices were made by a large majority of respondents, providing additional support for hypotheses 1 and 2. There was little in the respondents’ comments to contradict this conclusion.15

Preferences consistent with hypotheses 1 and 2 may result in consumers forgoing substantial value in interest-free borrowing. Given a zero-percent interest rate offer, the difference in present value of a sixty-month $2,500 down payment loan and an eighty-four-month $2,500 down payment loan is $1,061 (4.2% of the purchase price) when calculated at a 6% annual discount rate. Similarly, the present value difference between a sixty-month $2,500 down payment loan (including the $2,500 down) and a sixty-month zero down payment loan is $345 (1.4% of the purchase price).

Logit analysis of monthly payment: Hypothesis 3

In order to address the importance of monthly payment, total payments (including down payment less the rebate), and oddness of monthly payment, none of which were attributes in the choice tasks, logit analysis was employed, ignoring individual differences in preferences. A logistic regression was estimated for loan choice on down payment, rebate, contract length, interest rate, the squares of those variables, the products of each pair of those variables, monthly payment (M), total payments minus rebate, and a discrete variable (D) describing the oddness of M according to

  • image
  • image
  • image

for some integer I. Several of the quadratic and interaction terms were insignificant in the full regression, as was total payments minus rebate. A lack of significance for total payments was mildly surprising in light of the fact that approximately 60 respondents (from among 763 who provided any comments) made statements to the effect that their primary decision-making rule was to select the loan with lowest “total cost.” (In some cases, it was difficult to decipher whether this was the calculation that a respondent intended.) These individuals generally gave no indication that the payment totals were only an approximation of present value; rather, they referred to the total as “the bottom line,” or, to quote another respondent, “If a car costs me a total of $27,000, against a total of $29,000, all the other parts are not interesting!”

In order to focus on the significant effects, the logistic regression was run using only the variables that were significant in the larger regression. The results of this second regression appear in Table 8. The coefficient on oddness of monthly payment was positive and significant (p = 0.002). The coefficient on monthly payment itself was negative and even more significant statistically, acting as a strong control variable along with other controls because monthly payment is correlated with present value. The oddness result suggests that individuals did pay attention to monthly payment and were drawn to payments just below even amounts, providing support for hypothesis 3. The overall importance of monthly payments was confirmed by the respondent comments; it was the third most frequently mentioned decision-making factor (after interest rate and loan term).

Table 8.  Logistic Regression of Loan Choice on Loan Characteristics
Loan PropertyOverall SampleAge At Least 55 YearsTime At Least 5 MinutesBills At Least 50% IncomeAt Least College DegreeExcellent CreditMenIncome Over $ 75,000
  1. Note: Coefficients are bolded and standard errors are shown in parentheses. Choice of loan was regressed on the characteristics shown. Coefficients were assumed to be the same for all individuals in the overall sample regression and for all individuals in a given subgroup for the corresponding subgroup regressions.

  2. * p < .1, **p < .05, ***p < .01.

Contract length −0.065 * * * (0.003) −0.073 * * * (0.005) −0.072 * * * (0.004) −0.057 * * * (0.006) −0.072 * * * (0.005) −0.077 * * * (0.004) −0.061 * * * (0.004) −0.078 * * * (0.005)
Down payment −0.068 * * (0.028) −0.200 * * * (0.019) −0.082 * * * (0.032) −0.143 * * * (0.052) −0.192 * * * (0.016) −0.201 * * * (0.016) −0.178 * * * (0.015) −0.206 * * * (0.017)
Interest rate −14.642 * * * (0.635) −18.776 * * * (1.021) −16.447 * * * (0.708) −13.384 * * * (1.152) −18.452 * * * (0.889) −18.225 * * * (0.854) −17.304 * * * (0.836) −17.085 * * * (0.908)
Rebate0.163 * * * (0.012)0.171 * * * (0.019)0.181 * * * (0.013)0.127 * * * (0.022)0.204 * * * (0.017)0.202 * * * (0.016)0.176 * * * (0.016)0.181 * * * (0.017)
Monthly payment −0.917 * * * (0.096) −0.875 * * * (0.071)0.954***(0.049) −0.982 * * * (0.081) −0.905 * * * (0.062) −0.967 * * * (0.06) −0.809 * * * (0.057) −0.959 * * * (0.064)
Oddness0.096 * * * (0.035)0.196 * * * (0.056)0.111 * * * (0.039)0.121 * (0.065)0.015 (0.048)0.067 (0.047)0.087 * (0.046)0.060 (0.050)
[Down payment]2 −0.03 * * * (0.005)0.002 (0.002) −0.03 * * * (0.006) −0.023 * * (0.009) −0.001 (0.002)0.002 (0.002) −0.002 (0.002)0.002 (0.002)
Interest rate × rebate0.07 * * * (0.027)0.091 * * (0.043)0.078 * * * (0.03)0.096 * (0.049) −0.010 (0.038) −0.082 * * (0.036) −0.035 (0.035)0.069 * (0.039)
Contract Length × rebate −0.006 * * * (0.002) −0.01 * * * (0.004) −0.007 * * * (0.003) −0.008 * (0.005) −0.005 (0.003)0.001 (0.003) −0.002 (0.003) −0.008 * * (0.004)
Constant0.004 (0.022)0.000 (0.035)0.001 (0.024)0.000 (0.04)0.020 (0.03) −0.008 (0.029) −0.039 (0.028)0.017 (0.031)

Table 8 also contains results for several subgroups of interest. These groups include: (1) respondents at least fifty-five years of age (42% of the sample), (2) those who devoted at least five minutes to the questionnaire (84%), (3) those with over 50% of income devoted to debt and other bills (30%), (4) those with at least an undergraduate degree (54%), (5) those with self-reported excellent credit (59%), (6) those with incomes above $75,000 (50%), and (7) men (59%). Regression coefficients for the most interesting variables were not substantially different than those for the overall sample, but college-educated respondents had an oddness variable coefficient of only 1.5% compared to the 9.6% coefficient for the entire sample. Hence, the oddness factor in determining loan choice was stronger for less educated individuals.

Other findings

Because of the significance of interaction effects between rebate and down payment, product shares were simulated to investigate whether rebates might be more important in the presence of high down payments. This result could follow from either financial constraints or psychological factors linking rebates and down payments. The results in Table 9 are consistent with these ideas. Changing down payment from $0 to $2,500 and then $5,000, the high-rebate, high-interest contract became progressively more preferred relative to the lower-interest, zero-rebate choice. In the words of one respondent, “I lean toward nothing down … or a rebate that covers the amount down.” The interaction between down payment and rebate amount had the same sign and similar magnitude in each of the several demographic subgroup simulations.

Table 9.  Importance of Rebates Relative to Interest Rates as a Function of Down Payments
Contract Length (Months)Down Payment ($)Rebate ($)Interest Rate (%)Preference Share (%)
  1. Note: Share of preference represents the percentage of respondents who would prefer or choose each auto loan product, assuming these were the only two choices available. Results are calculated using randomized first choice simulation on the hierarchical Bayes output.

6003,000425.33
 0274.67
602,5003,000436.69
 0263.31
605,0003,000448.27
 0251.73

Discussion

  1. Top of page
  2. Abstract
  3. Decision theory
  4. Method
  5. Results
  6. Discussion
  7. References

This study has shown, at least among respondents and given the range of values offered, that contractual interest rate and contract length are the key determinants of consumer loan choice; these attributes are three times as important as rebate and down payment. Respondents preferred low interest rates, high rebates, intermediate contract lengths, and moderate down payments. The revealed preference for intermediate (sixty-month) contract lengths may in some cases have reflected personal preference for very short contract lengths coupled with budgetary considerations (binding financial constraints) that necessitate a longer loan term. This possibility is supported by the fact that respondents with substantial and varied investments preferred short-term loans more than did those without such investments or those with high-existing monthly obligations.

While the evidence demonstrates that the respondents as a group understood that high interest rates are undesirable, they generally had a weak sense of the time value of money and compound interest. They avoided long-term zero down payment loans, even when these loans were interest free. These preferences are consistent with hypotheses 1 and 2, derived from the predictions of mental accounting theory (Thaler 1985). In particular, the respondents behaved as though they chose contract lengths and down payments with the goal of minimizing the size and duration of future payments. This may be due to their mental coupling of future benefits from cars with their loan payments (Prelec and Loewenstein 1998); however, they were financially unable to manage very high amounts down or very rapid repayment schedules. The down payment and contract length preferences were also consistent with extremeness aversion (Kivetz 1999; Simonson and Tversky 1992). There is no clear evidence, however, that one of these explanations is solely responsible for the observed preferences, although there is (as expected) no apparent aversion to extremely low interest rates or extremely high rebates. Many respondents calculated the total undiscounted cost of alternative loan choices and believed that such a metric was the appropriate basis for selecting a loan. Respondents were also attracted to loans with monthly payments that fell just short of multiples of $100, providing support for hypothesis 3. Moreover, oddness of payment influenced those without college degrees more than it did the college educated. Once again, these results are consistent with multiple possible thought processes. Individuals may effectively round down monthly payments by focusing on the first digit (Anderson and Simester 2003; Estelami 2001), or they may round up and use the higher numbers as reference prices against which gains are perceived and weighted according to prospect theory (Kahneman and Tversky 1979). In either case, one would expect more capable or educated individuals to be less prone to rounding, so that the theories cannot be disentangled. However, the results are inconsistent with financial rationality.

Implications for consumers’ financial welfare

Taken together, our findings suggest that consumer financial welfare is less than it would be if decisions were made on a financially rational basis. The differential coefficients on oddness, moreover, suggest that there is not a level playing field; college-educated consumers may fare better in their loan choices than those with less education. These results should be of interest to financial educators and counselors as they develop their curricula and target the population in need of knowledge in pursuit of financial rationality.

Mathematical education at any level should serve to mitigate excessive attraction to payments ending in 9s (or 90s). That holds if individuals are simply ignoring digits after the first, but it likely holds even if the appeal of odd prices is because of perceived gains relative to a higher reference price. An individual who is better skilled in arithmetic is plausibly less anchored to even reference prices. Other benefits from financial education, however, may depend on the reason that consumer behavior deviates from financial rationality. Individuals who choose short-term loans because of a strong mental coupling between payments and consumption would evidently experience psychic pain each time a loan payment was made on a car that was six years old and becoming too costly or unsafe to continue operating. Comments from the respondents are certainly consistent with this possibility. There is no known evidence addressing whether the undesirable psychological features of mental accounting can be eliminated through education. And as Thaler (1999, p. 203) comments, “It is not possible to say that the system is flawed without knowing how to fix it … repairing one problem may create another.”

If rejection of long-term loans is instead primarily a means of self-control to prevent excessive personal spending, then financial education may be more promising. Education should not deny the value of self-control mechanisms (because the willpower problems that necessitate such mechanisms are not easily solved), but it should make individuals aware of low-cost approaches to self-control. In the case of a car loan with zero interest rate, a rational consumer could take the larger, longer-term loan and set up automatic deposits to a savings or money market account that was dedicated to future automobile purchases. With that in place, self-control would be largely achieved and financial problems would be unlikely to occur. Even if long-term zero-interest offers may come at the cost of a higher contractual price and may therefore not always be desirable in the real world, effective financial self-control skills should be generally useful. In those instances where individuals have the skills but lack the access, financial institutions could be encouraged to offer inexpensive dedicated accounts that are easy to set up.

This example illustrates the general point that consumers should be advised to consider the (after-tax) rates paid or earned on all their sources and uses of funds when making financial decisions. Ideally, they should calculate present values, but an awareness of rates of return on alternative borrowing and lending opportunities would likely lead to nearly as much welfare improvement as would a fully rigorous optimization. This is relevant not only for individuals who pass up desirable loans but also for those who maintain balances on high-rate credit cards while at the same time holding large sums of cash in their checking or savings accounts.

Of course, even if borrowers do have all the relevant knowledge, the question remains whether the best financial decision will be made. Future research could address this question by investigating the extent to which loan choices would change if consumers were motivated to acquire and use information regarding effective interest rates on loans, returns on investment opportunities, present value, and future value calculations.

It should be noted that financial education is not a panacea. This study does not address the efficacy or cost-effectiveness of financial education in general, which is an appropriate topic for ongoing study.16 Further, there are other factors that negatively affect consumers’ abilities to understand the different financial packages offered by auto dealers and finance companies. First, loan terms are frequently so complex that even those with a college education cannot completely protect themselves merely by reading loan documents. Second, weak disclosure requirements and insufficient enforcement of Truth in Lending Laws mean that consumers are not always informed about all of a loan’s cost dimensions. These problems suggest a role for credit reform and regulation, although more research is needed in order to determine what information should be required in auto financing offers and the format in which it should be presented.

Footnotes
  • 1

    Conjoint is a multivariate statistical technique that analyzes preferences for various combinations of attributes (derived from considering jointly). “Choice-based” conjoint uses survey respondents’ choices among products to infer preferences; it can be distinguished from other conjoint methods in which respondents might rate several products, considered individually, with respect to purchase likelihood; provide complete rankings over a group of possible products; or rate the strength of preference for one option over another in a series of comparisons.

  • 2

    For an individual with neither significant borrowing nor savings, owing to a rate of time preference for consumption that fell between available rates on savings and borrowing, the rate of time preference would be the appropriate discount rate for the present value calculation. Also, because a car purchase is a large expenditure, rates on both savings and alternative borrowing may be relevant for some individuals who have limited assets.

  • 3

    If there were unusual circumstances in which a consumer had an alternative borrowing source with a negative interest rate, the consumer would optimally exhaust that borrowing source, investing any proceeds not desired for current consumption. After the subsidized borrowing source was exhausted, the relevant alternative rate would exceed zero.

  • 4

    Another reason that rational individuals might prefer long-term loans given a high fixed interest rate is that long-term investments (e.g., thirty-year bonds) have in many eras offered higher returns than have similar short-term investments. The term structure of interest rates was essentially flat, however, when this survey was conducted.

  • 5

    Debt aversion could alternatively arise from moral or religious concerns.

  • 6

    Consumers with hyperbolic discounting essentially apply a much larger discount rate between today (whatever day today is) and tomorrow than between any two subsequent days, and as a result, their preferences are inconsistent across time. They always desire to push off payments just a little further.

  • 7

    Choice-based conjoint is defined in footnote 1.

  • 8

    A pilot study (n = 96) was conducted to confirm the importance of particular loan attributes identified by previous research as being the most important (Beltramini and Chapman 2003; Estelami 2001; Heitfield and Sabarwal 2004; Herrmann and Wricke 1998).

  • 9

    A factorial design presents each respondent with all possible combinations of the attribute levels; a fractional factorial design still involves all the attributes but presents each survey respondent with only a small subset of the possible combinations. Program modules developed by Sawtooth Software, Inc. were used to implement the factorial design.

  • 10

    Johnson and Orme (1996), in a study analyzing surveys with between three and six attributes, report that at least twenty choice tasks generally can be employed without deterioration in data late in a survey.

  • 11

    An unequal number of attribute levels could bias the estimation of the importance of one attribute relative to another, as reported in our Table 5, in respondents’ decisions. In particular, the importance of attributes with greater numbers of levels tends to be overestimated (Wittink et al. 1992).

  • 12

    Sawtooth Software, Inc.’s HB estimator was used. Simulations described below were also conducted using Sawtooth programs.

  • 13

    Individual RLH is sometimes used as a criterion for deleting respondents from analysis. Low RLH could suggest random answers, though it could also suggest preferences that were very far from the rest of the sample (in levels and in relations among attribute preferences) and therefore poorly fit due to the multivariate normal restriction on part-worths. For the interactions model, minimum individual RLH was .737 (.470 in the model without interactions). No individuals were deleted from the sample.

  • 14

    The two fixed choice tasks remained the same for all respondents and presented the two alternative “products” displayed in Figure 1. Option 1 was $0 down, sixty-month contract, 6% interest, and $2,500 rebate, and option 2 was $2,500 down, forty-eight month contract, 0% interest, and no rebate. For fixed task #1/fixed task #2, 24.5%/26.8% of the respondents chose option 1 and 75.5%/73.2% chose option 2. In the simulation, option 1 had a 24.2% and option 2 had a 75.8% share of preference; the simulated share of preference values were within three percentage points of the actual preferences indicated in the holdout data.

  • 15

    One respondent did mention the added insurance requirements associated with having a loan outstanding, but the issue did not appear to be determinative for that individual.

  • 16

    Lyons et al. (2006) and Fox, Bartholomae, and Lee (2005) describe the state of the literature and discuss the challenges of evaluating financial education in its varied forms; Elliehausen, Lundquist, and Staten (2007) provide recent evidence on benefits of credit counseling.

References

  1. Top of page
  2. Abstract
  3. Decision theory
  4. Method
  5. Results
  6. Discussion
  7. References
  • Alessie, Rob, Stefan Hochguertel, and Guglielmo Weber. 2005. Consumer Credit: Evidence from Italian Micro Data. Journal of the European Economic Association, 3 (1): 144178.
  • Allen, Leslie J. 2006. UP TO 8 Years, Auto Loans Getting Longer, Debts Bigger, ‘Upside-Down’ Buyers Owe More than Cars Are Worth. Autoweek, Newsday, Long Island, NY. July 9.
  • Anderson, Eric and Duncan Simester. 2003. Effects of $9 Price Endings on Retail Sales: Evidence from Field Experiments. Quantitative Marketing and Economics, 1 (1): 93110.
  • Attanasio, Orazio, Pinelopi Goldberg, and Ekaterini Kyriazidou. 2008. Credit Constraints in the Market Consumer Durables: Evidence from Micro Data on Car Loans. International Economic Review, 49 (2).
  • Ayers, Benjamin, Steven Kachelmeier, and John Robinson. 1999. Why Do People Give Interest-Free Loans to the Government? An Experimental Study of Interim Tax Payments. Journal of the American Taxation Association, 21 (Fall): 5574.
  • Bankrate.com. 2006. Where To Get the Money. http://www.bankrate.com/cnn/green/auto/basics5-1a.asp?caret=35.
  • Beltramini, Richard and Patricia Chapman. 2003. Do Consumers Believe in Automobile Industry Rebate Incentives? Journal of Advertising Research, 33 (1): 1624.
  • Boyes, William, Allen Lynch, and William Steward Mounts, Jr. 2007. Why Odd Pricing? Journal of Applied Social Psychology, 37 (May): 11301140.
  • Chrzan, Keith and Bryan Orme. 2000. An Overview and Comparison of Design Strategies for Choice-Based Conjoint Analysis. Sawtooth Software Research Paper Series. http://www.sawtoothsoftware.com/download/techpap/desgncbc.pdf 20 (retrieved May 2003).
  • Dasgupta, Srabana, Sivaramakrishnan Siddarth, and Jorge Silva-Risso. 2007. To Lease or Buy? A Structural Model of a Consumer’s Vehicle and Contract Choice Decisions. Journal of Marketing Research, 44 (August): 490502.
  • Elliehausen, Gregory, E. Christopher Lundquist, and Michael E. Staten. 2007. The Impact of Credit Counseling on Subsequent Borrower Behavior. The Journal of Consumer Affairs, 41 (1): 128.
  • Estelami, Hooman. 2001. Determinants of Discount Rates in Consumer Credit Decisions. Journal of Marketing Theory and Practice, 9 (1): 6373.
  • Fox, Jonathon, Suzanne Bartholomae, and Jinkook Lee. 2005. Building the Case for Financial Education. The Journal of Consumer Affairs, 39 (1): 195214.
  • Heitfield, Erik and Tarun Sabarwal. 2004. What Drives Default and Prepayment on Subprime Auto Loans? Journal of Real Estate Finance and Economics, 29 (4): 457477.
  • Herrmann, Andreas and Martin Wricke. 1998. Evaluating Multidimensional Prices. Journal of Product and Brand Management, 7 (2): 161169.
  • Huber, Joel, Bryan Orme, and R. Miller. 1999. Dealing with Product Similarity in Conjoint Simulations. Sawtooth Software Conference Proceedings, pp. 253266. Sequim, WA: Sawtooth Software.
  • Johnson, Richard and Bryan Orme. 1996. How Many Questions Should You Ask in Choice-Based Conjoint Studies? Advanced Research Techniques Forum.
  • Juster, Thomas. 1963. Consumer Sensitivity to the Price of Credit. The Journal of Finance, 19 (2): 139214.
  • Kahneman, Daniel and Amos Tversky. 1979. Prospect Theory: An Analysis of Decision under Risk. Econometrica, 47 (March): 263291.
  • Kamleitner, Bernadette and Erich Kirchler. 2006. Personal Loan Users’ Mental Integration of Payment and Consumption. Marketing Letters, 17 (4): 281294.
  • Karlan, Dean and Jonathan Zinman. 2008. Credit Elasticities in Less-Developed Economies: Implications for Microfinance. American Economic Review, 98 (3).
  • Kivetz, Ran. 1999. Advances in Research on Mental Accounting and Reason-Based Choice. Marketing Letters, 10 (3): 249266.
  • Lyons, Angela, Lance Palmer, Koralalage Jayaratne, and Erick Scherpf. 2006. Are We Making the Grade? A National Overview of Financial Education and Program Evaluation. The Journal of Consumer Affairs, 40 (2): 208235.
  • Prelec, Drazen and George Loewenstein. 1998. The Red and the Black: Mental Accounting of Savings and Debt. Marketing Science, 17 (1): 428.
  • Simonson, Itamar and Amos Tversky. 1992. Choice in Context: Tradeoff Contrast and Extremeness Aversion. Journal of Marketing Research, 29 (August): 281295.
  • Stango, Victor and Jonathan Zinman. 2006. How a Cognitive Bias Shapes Competition: Evidence from Consumer Credit Markets. Working Paper. http://emlab.berkeley.edu/users/webfac/gilbert/e221-f06/stango.pdf (retrieved May 14, 2007).
  • Thaler, Richard. 1985. Mental Accounting and Consumer Choice. Marketing Science, 4 (3): 199214.
  • Vriens. 1999. Mental Accounting Matters. Journal of Behavioral Decision Making, 12 (3): 183206.
  • Vriens, Marco, Harmen Oppewal, and Michel Wedel. 1998. Ratings-Based versus Choice-Based Latent Class Conjoint Models—An Empirical Comparison. Journal of the Market Research Society, 40 (1): 237248.
  • Wittink, Dick, Joel Huber, Peter Zandan, and Richard Johnson. 1992. The Number of Levels Effect in Conjoint: Where Does It Come from and Can It Be Eliminated? Sawtooth Software Conference Proceedings.