An Analysis of Differences Between Consumers of Variable-Rate and Fixed-Rate Residential Mortgages

ABSTRACT - This study examines which borrower characteristics, attitudes, and shopping behavior could best discriminate between consumers financing their homes with a variable-rate mortgage from those financing with a fixed-rate mortgage.

Citation:

Gerald Albaum and Del Hawkins (1980) ,"An Analysis of Differences Between Consumers of Variable-Rate and Fixed-Rate Residential Mortgages", in NA - Advances in Consumer Research Volume 07, eds. Jerry C. Olson, Ann Abor, MI : Association for Consumer Research, Pages: 692-695.

Advances in Consumer Research Volume 7, 1980     Pages 692-695

AN ANALYSIS OF DIFFERENCES BETWEEN CONSUMERS OF VARIABLE-RATE AND FIXED-RATE RESIDENTIAL MORTGAGES

Gerald Albaum, University of Oregon

Del Hawkins, University of Oregon

ABSTRACT -

This study examines which borrower characteristics, attitudes, and shopping behavior could best discriminate between consumers financing their homes with a variable-rate mortgage from those financing with a fixed-rate mortgage.

INTRODUCTION

Among the many purchase decisions that consumers make, perhaps the most important is that concerning buying a house. In addition, most consumers also must select a lender from which to acquire a mortgage. In the past, the mortgage decision involved simply comparing terms offered by different lenders for the same mortgage plan-a fixed rate level payment mortgage (FRM). Since a number of innovative alternative mortgage instruments have been developed in recent years, the mortgage decision has been broadened to include type of mortgage as well as lending institution (Kaplan, 1977).  A major alternative residential mortgage plan is the variable rate mortgage (VRM), which differs from the FRM in that the interest rate of the mortgage is allowed to change over the term of the loan, usually within bounds specified by federal and/or state regulations regarding mortgage loans. The interest rate is pegged to some reference index (i.e., market interest rate) that fluctuates over the life of the loan.

The purpose of this paper is to examine mortgage features, borrower characteristics, borrower attitudes and borrower shopping behaviors that discriminate individuals financing their homes with a VRM from those financing their homes with the more traditional FRM.  There are important reasons for analyzing the differences between VRM and FRM homeowners. Although VRMs have been offered by a small number of lenders in various parts of the country for many years, their only widespread use in residential lending has been in California. Because of its relative newness, the industry itself does not have a well developed profile of the differences and similarities between VRM and FRM homeowners. Consequently, lenders do not knew whether there are distinct market segments for each of the mortgage instruments.

A second important reason for examining these two groups of borrowers is the controversial nature of the VRM. Those opposed to the variable rate mortgage have argued that residential mortgage rates would be raised, that low income and minority groups would be discriminated against, that VRMs are more complex and difficult to understand than FRMs, and that VRMs would increase the borrower's risk (Kaufman, 1976). Industry representatives deny that the first two situations would exist. With respect to the last two areas of criticism, proponents of the VRM argue that, although they are true, they are more than offset by such favorable features as the greater availability of mortgage funds and increased transferability of existing mortgages thereby making it easier to sell existing homes. Thus, information on the characteristics of VRM and FRM homeowners is required to shed some light on the controversies.

The extent to which a lender may implement a VRM depends upon the type of institution (i.e., commercial bank, savings and loan, etc.) and existing federal and/or state regulations regarding mortgage loans. For instance, some states explicitly prohibit the VRM, some states explicitly allow it, but the majority of states have no explicit regulation concerning the VRM. Also, Federally-chartered savings and loan associations nationwide are authorized to make, purchase, and participate in VRMs.

METHODOLOGY

The population sampled was homeowners who had acquired conventional first mortgage loans (VRM or FRM) from nine large California savings and loan associations. At the time of data collection, two of the associations offered VRMs exclusively, three associations offered both VRMs and FRMs, and four associations offered only FRMs.

The data were collected using mail survey techniques with one follow-up. Questionnaires were sent to a stratified random sample of 1,726 homeowners. Stratification was by type of mortgage. Responses were received from 367 VRM and 380 FRM homeowners, representing an overall response rate of 45 percent of those who actually received questionnaires. Nonresponse validation was conducted by obtaining information on selected demographic and socioeconomic characteristics by telephone interview from a sample of 20 nonrespondents from each of the sample groups. There were no significant differences between respondents and nonrespondents in the distribution of major characteristics (e.g., age, income, extent of education) nor did there appear to be any unique population sub-groups omitted from the obtained sample.

The data were analyzed by the use of a series of step-wise discriminant analyses. For each analysis, the variables are presented in the order in which they entered the discriminant model. Although standardized coefficients are indicators of the importance of variables, they are not appropriate in assessing the relative discriminatory power of the variables in a linear discriminant function (Joy and Tollefson, 1975). An appropriate measure of relative discriminating power is one used by Mosteller and Wallace (1963). This measure is:

I_j = |b_j(X_j1-X_j2)|    (1)

where

I_j = the importance value of the j^th variable,

b_j = unstandardized discriminant function coefficient for the j^th variable,

X_jk = mean of the j^th variable for the k^th group.

The relative importance weights may be interpreted as the portion of the discriminant score separation between the groups that is attributable to the j^th variable. Since a relative importance value shows the importance value of a particular variable relative to the sum of the importance values of all variables, the relative importance of the j^th variable (R.) is given by (Awh and Waters, 1974)

EQUATION    (2)

The statistical significance of each of the models de- rived is tested by the Wilks' lambda statistic. The larger lambda is, the less discriminating power is pre- sent. Also shown for each function is its associated canonical correlation, which is a measure of association between the discriminant function and the group variable. A third test of the adequacy of each of the derived discriminant functions is its classificatory power. For each model, the percent of homeowners correctly classified by the model is presented. This measure can be compared to the proportion correct that would be expected by chance alone. For this study, the proportion correct expected by chance is 50%, using the C_pro measure proposed by Morrison (1969). Since the primary purpose of the study was analysis of differences between the two groups of homeowners rather than classification, the same observations used in developing the model were used to test the classificatory power, which usually leads to an inflated percentage of classified observations (Morrison, 1969).

RESULTS AND DISCUSSION

The following discussion pertains to separate discriminant models based on selected economic expectations of homeowners, demographic and socioeconomic characteristics of homeowners, mortgage shopping behavior, perceptions of importance of features of mortgage plans, and attitudes toward variable rate mortgages. In addition, an overall analysis was run incorporating all of the dimensions together with additional variables.

Economic Expectations

Economists often argue that for certain problems economic expectations of individual people would affect choice behavior in an economic setting. Because they contain different features, VRMs and FRMs may be expected to appeal to borrowers with differing expectations concerning the future course of interest rates and housing prices. For example, the VRM may be preferred by borrowers who believe interest rates will decline over the life of their mortgages while the FRM should be preferred by borrowers who believe interest rates will rise. Another expectation variable that should affect mortgage choice is the length of time one expects to own his or her home.

Future expectations regarding four variables--interest rates, house prices in general, value of the individual's home, and length of time one expected to own a home--were analyzed in a discriminant analysis. The results were such that none of the variables were significant at the level for inclusion, and a model could not he derived. Thus, these particular variables by themselves were not useful discriminators of VRM and FRM borrowers, and would not be useful guides to actual behavior.

Demographic and Socioeconomic Characteristics

It is frequently assumed that demographic and socioeconomic variables are highly correlated with various aspects of shopping behavior. Therefore, a discriminant analysis was performed on 18 demographic and socioeconomic variables. The result of this analysis is shown in Table 1.

While this model works better than the "economic expectations'' model, it only classified 52 percent correctly. This is not substantially better than chance. Although Wilks' lambda was significant at p < .09, the canonical correlation indicated a low association between the function and the groups. Moreover, it will be noted that almost two-thirds of the discrimination between the two groups was accounted for by only two variables, income and age.

TABLE 1

DEMOGRAPHIC AND SOCIOECONOMIC VARIABLES DISCRIMINANT FUNCTION

Shopping Behavior

Fifteen shopping behavior variables were identified for analysis, and the resulting model is shown in Table 2. Wilks' lambda was highly significant, and there was a moderate correlation between the function and the groups. In terms of classification power, the model performed reasonably well, correctly classifying 66.4 percent of the homeowners.

TABLE 2

SHOPPING VARIABLE DISCRIMINANT FUNCTION

Although eight variables were included in the model, two of them accounted for over three-fourths of total discrimination. The primary variable, accounting for almost one-half of discrimination, was whether or not the homeowner considered both types of mortgages. Consumers tended to select the VRM when both types were considered. Whether this implies an inherent superiority in the VRM or an increased tendency for VRM borrowers to shop is unclear. The latter seems unlikely since whether or not a person shopped for a mortgage accounted for only 1.2 percent of discrimination. The second major variable entered in the model was whether or not there was a difference in the initial interest rates for the two mortgage plans when a person was in fact offered both types. Homeowners were more likely to select the VRM when there was a difference and the rate for the VRM was lower. When the VRM was first offered in California, some VRM lenders, but not all, did in fact price the initial mortgage interest rate up to 1/2 percentage point lower than the market FILM rate.

Importance of Mortgage Features

In addition to the variable interest rate feature, VRMs differ from FRMs in the following ways:

1.  There is no fee if prepayment is made within a specified period of time after notification of an increase in the interest rate.

2.  A qualified purchaser can often assume the seller's existing VRM at the interest rate in effect at the time of sale, resulting in the seller being guaranteed that financing is available for a buyer meeting the lender's credit standards.

3.  Some lenders have provided an open-end line of credit on VRMs by which the mortgages may borrow on the equity, at the mortgage rate, for any purpose.

Nine features of mortgages together with the homeowner's expectations about how long he or she expected to own the home were included in this analysis. To reduce multicollinearity which can distort the interpretation of a discriminant function, these data were factor analyzed, and factor scores used instead of the raw data. Three factors were extracted, accounting for 39 percent of the total variance.

The discriminant model utilized all three factors (Table 3), and each accounted for a meaningful share of the total discrimination. However, this model was not particularly good since it classified correctly slightly more than one-half of the homeowners, the canonical correlation was very low, and Wilks' lambda was relatively large, even though it was significant.

TABLE 3

IMPORTANCE OF FEATURES MODEL

Attitudes

Consumer behavior theorists frequently utilize attitudes as a predictive variable. However, they also recognize that attitudes often shift after behavior to become more consistent with the completed action. Therefore, one would expect that a discriminant model developed from measures of attitude would differentiate the two groups. A set of 18 Likert-type scale statements were used to elicit attitude toward variable rate mortgages. These data were factor analyzed, and three factors accounting for 35 percent of the total variance were extracted. Factor scores were used in the discriminant function.

As shown in Table 4, all factors were included in the model. However, 84 percent of the discrimination was accounted for by the factor "macro evaluation of VRMs." The factor includes such variables as whether VRMs are alternatives worthy of consideration, are good for the economy, are beneficial to minorities, and are good for the lending industry. As would be expected, VRM borrowers had more positive attitudes and were more knowledgeable. This model seems to be fairly effective, having high classificatory power, a significant lambda, and moderate correlation between the function and groups. However, one cannot be sure of whether the attitudes preceded and influenced the behavior of if they were subsequent to the behavior and were themselves influenced by the behavior.

TABLE 4

ATTITUDE MODEL

Overall Model

Prior to the analysis of the comprehensive basic data set, steps were taken to reduce multicollinearity among the many variables involved. First, the responses to attitudes and importance of features were factor analyzed and the factor scores used. In addition, a correlation matrix was computed for all remaining variables. All pairs of variables with a high r value had one member of the pair deleted. The discriminant analysis was performed on the reducted data set, which amounted to 38 variables. In the overall analysis, 21 variables met the entry criterion. The results are shown in Table 5. Although many variables were included in the model, three of them accounted for 60 percent of the total discrimination.

The first and most important variable to enter the model is related to the evoked set. That is, if the respondent considered both types of mortgages, as opposed to only one type, they were much more likely to choose the VRM. This implies (a) those who have decided to choose VRM look at FRMs just to be sure (b) in a comparison VRMs tend to win, (c) people who tend to shop also tend to prefer the features of a VRM or (d) those who eventually secured a VRM tried to secure an FRM but were unable to do so. A look at the other variables in the model makes option "c" above seem unlikely. The second most important variable was one of the dimensions of attitude--the more favorable the overall evaluation of VRMs, the more likely the individual is to own one. The third attribute is the actual down payment. VRM owners paid significantly less down than FRM owners (the actual averages for this were $10,642 for VRMs and $15,470 for FRMs, t = 7.29, p < .001).

TABLE 5

DISCRIMINANT FUNCTION BASED ON ALL VARIABLES

This comprehensive overall model performed the best of all the models. The percent of homeowners correctly classified and canonical correlation were the highest, and Wilks' lambda the lowest.

CONCLUSIONS

Perhaps the strongest conclusion that one can derive from this data is that VRM and FRM borrowers are remarkably similar. The only consistently powerful discriminator was whether or not the individual had considered both types of mortgages. This is, of course, an after-the-fact variable. Forward looking variables such as demographics and economic expectations did not differentiate the two groups.

Several actions seem implied by the results reported above. First, it does not seem realistic to pinpoint a unique market segment for VRMs based on this analysis at this time; for example, the demographic and socioeconomic model predicted at no more than the chance level. The most useful approach to expanding the VRM market would be to improve attitudes toward VRMs in general and to strongly encourage a comparison between VRMs and FRMs. From a public policy standpoint, there is a great need to educate people on how to select a mortgage type. In other words, individuals should relate specific mortgage features on their own future plans and expectations. This does not appear to be the decision process currently being used. FRM and VRM mortgages offer unique advantages and disadvantages. Yet current borrowers do not appear to be able to relate the unique features of each mortgage type to their own unique situation and expectations. Since VRM lenders in California have followed a push strategy in marketing the VRM concept by concentrating on real estate agents and brokers, there appears to be a need for paying more direct attention to borrowers, thus engaging in a pull strategy.

REFERENCES

Awh, R. Y. and Waters, D. (1974), "A discriminant analysis of economic, demographic, and attitudinal characteristics of bank charge-card holders: A case study," The Journal of Finance, 29, 973-980.

Joy, O. M., and Tollefson, J. O. (1975), "On the financial applications of discriminant analysis," Journal of Financial and Quantitative Analysis, 10, (December), 723-739.

Kaplan, D. M. (1977), Alternative Mortgage Instruments Research Study, Washington, D. C.: Federal Home Loan Bank Board.

Kaufman, G. G. (1976), Financial Intermediaries and Variable Rate Mortgages, San Francisco: Federal Reserve Bank of San Francisco.

Morrison, D. G. (1969), "On the interpretation of discriminant analysis," Journal of Marketing Research, 6, (May), 156-163.

Mosteller, F. and Wallace, D. F. (1963), "Influence in an authorship problem," Journal of the American Statistical Association, 58, (June), 275-309.

----------------------------------------