Incorporating credit in demand analysis.
Analysis of consumer demand has evolved along two related but distinct lines of thought. Theorists typically begin with the twin assumptions that individuals choose the consumption bundle that maximizes utility, subject to a budget constraint, and that data are available to test the validity of the results. Attention is focused on restrictions such as Slutsky symmetry and homogeneity that are either implicitly or explicitly imposed. For theoretical economists, imposing these restrictions is what differentiates a model of consumer demand from a mere description of the correlation between the exogenous and endogenous variables (Philips 1974).
The second school of thought comprises those who must gather the relevant data and use them to make predictions and to analyze policies. These researchers quickly learn that the data are deficient in both the number of observations and in the definitions employed by the data-collection agency. For this group, imposition of restrictions tends to be a little unsettling because they are often rejected by the data. Restrictions tend to be imposed only when they result in parameters that conform to those predicted by the theory, even when the imposition of the restrictions causes a sign reversal in the estimated parameters. The conformation of estimated results to those predicted by the theory is regarded more as a test of the model than a test of the theory (Chalfant 1987).
Where the data are weak, it is more logical to build models that take into account data limitations than to build theoretical models that can never be tested. The example used in this paper is estimation of the impact of credit use on consumer expenditure patterns. There is a fundamental problem with publicly available statistics as they relate to this area. Governments and corporations define expenditures by the date when the consumer agrees to purchase rather than the date when the consumer must find the required funds. The monthly U.S. per capita expenditures on durables, nondurables, services, and savings sum to the per capita disposable income. Yet Americans spend almost 18 percent of their disposable incomes repaying loans, an item that does not appear as an expenditure item (Paquette 1986). If one purchases a new house, the entire value of the transaction is recorded as a one-period increase in housing expenditures. To the individual who has made the purchase, however, it represents an agreement to consign future expected income to a budgetary item called mortgage repayments.
Periodically, the government does publish surveys of actual consumer expenditures, but these are useless if it is necessary to examine the impact of changing prices, interest rates, income shocks, and other economic conditions on expenditure patterns because all consumers face the conditions prevalent when the survey was undertaken. The economist has three options. The first is to assume that expenditure figures measure how individuals spend money (rather than measuring sales made by corporations) and estimate a demand system by using available data. This is equivalent to assuming that consumers pay cash for all items. The second option is to assume that correct data exist and to examine the intertemporal optimization decision within a dynamic demand system. The lack of suitable data means that these models can never be tested.1
The third possible option, discussed in this paper, is to begin with a description of data limitations, to discuss which aspects of the theory are relevant to the particular data set, and to estimate the model within the confines of the relevant theory. The test of any hypothesis regarding consumer behavior is, therefore, a joint test of both the underlying theoretical model and of the specified hypothesis itself. Hence, the theory used is an unrestrictive as possible. 1 In fact, most intertemporal demand equations that have been estimated assume that the intertemporal consumer choice can be decentralized. The consumer first allocates his budget to different periods and then to different goods and services within the period, with each period's budget acting as a binding constraint. Again, this makes the analysis of credit use impossible.
In a study published in 1986, Johnson and Widdows examined the emergency fund levels of households. They reported that the total value of all emergency funds (checking and savings accounts, certificates of deposit, and savings certificates) had fallen dramatically in the six years preceding 1983. The median family had an emergency fund level equal to 16 percent of annual pretax income in 1977. By 1983 this had fallen to 7 percent. In 1983, only 77 percent of families surveyed had sufficient funds to maintain consumption over a typical period of unemployment. Recent evidence indicates that the level of emergency funds has fallen even more since 1983. The savings rate fell from 6 percent at the end of 1983 to less than 3 percent in August 1987 (Federal Reserve Bank of Cleveland 1988).
This decline in readily available funds does not seem to have adversely affected the credit worthiness of the typical consumer (as measured by the willingness of lending institutions to extend credit). In fact, the ratio of consumer installment and home mortgage credit outstanding to disposable income has been trending upward on a path that appears the mirror image of the savings rate (Paquette 1986).2
The ratio of debt service repayments to disposable income has also trended upwards. This was particularly evident during 1985. It reached 17.6 percent in January 1986, making this item more important than food in consumers' budgets (Paquette 1986).
The media have been quick to use figures similar to those just cited to predict an ever-increasing delinquency rate (Rudolph 1986). There is, however, no proven relationship between the debt service load and the delinquency rate (Eastwood and Sencindiver 1985). Debt service payments can be thought of as expenditures that must be paid like a tax and the income remaining after such payments considered to be discretionary income. Although it is true that increasing dependence on credit must eventually necessitate an increase in the proportion of disposable income used to service debt, it does not follow that the absolute amount of real discretionary income will fall. Just as a real income increase can negate the impact of a tax increase, leaving the consumer with more disposable income, a real income increase can compensate for an increased debt service load, leaving the consumer with more discretionary income. Even if real income should fall, a high debt service obligation does not imply increased delinquencies. Consumers can continue their repayments by reducing cash expenditures on nonessentials and, if the income fall is viewed as transitory, by increasing their credit use. 2 An interesting, but seemingly untested, hypothesis is that the decline in the U.S. savings rate is a direct result of the increasing availability of credit. This would occur if consumers used credit in lieu of savings to purchase big budget items or felt that they could rely on credit cards to stretch their budgets in case of an emergency.
An increased reliance on credit does, however, have important implications for the way in which we construct and interpret models of consumer expenditure by using available expenditure information. Consider how the ready availability of credit influences the consumer's budget constraint, which is used to derive both the Slutsky symmetry and homogeneity conditions. With interest rates of 18 to 22 percent on credit cards, banks are more than willing to extend credit to consumers who would ordinarily be considered credit risks. A recent Nielsen survey indicates that the average credit card holder has access to seven credit cards (Rudolph 1986). The absence of cross-checking with other lenders has led to a situation in which some consumers have access to almost unlimited short- to medium-term credit. As long as this is the case, the budget constraint is binding only to the extent that consumers are unwilling to consign a significant proportion of expected future income. Paradoxically, the more dependent consumers become on credit, the less willing they are to ruin their credit record by defaulting.
The absence of a budget constraint invalidates all restrictions based upon it when the expenditure data on which the restrictions are imposed are those provided by the government. An example of the importance of this problem is evident in what has become a fall tradition--low finance rates on cars. Consider the imposition of Slutsky symmetry between any other expenditure item and automobiles in a monthly or quarterly model. A drop in car prices results in a large increase in reported automobile expenditures, but this is reflected in other categories only to the extent of the down payments made on automobiles. Any estimate of the compensated cross-price elasticity between cars and other categories will be biased downward because consumers can spend more than they earn. This bias will remain as long as consumers are willing to increase their reliance on credit. With increasing real incomes, this process can continue indefinitely, so it can be expected that Slutsky symmetry would be rejected by the data.
A second theoretical result that bears further scrutiny is the impact of transitory income shocks on consumption patterns. The Life-Cycle Permanent Income Hypothesis (LCPIH), which is now generally accepted as the correct application of theory to explain the allocation of consumer spending through time, predicts that transitory shocks have a minimal impact on consumption patterns. Most quantitative tests of the LCPIH have concluded that consumption is sensitive to transitory shocks and have rejected the hypothesis. Flavin (1985), in a paper entitled "Excess Sensitivity of Consumption to Current Income: Liquidity Constraints or Myopia?" reports studies by Blinder (1981), Hall and Mishkin (1982), Hayashi (1985), Sargent (1978), and Flavin (1981) that reject the LCPIH.
In addition to ignoring the problem of credit repayments and assuming that expenditure data exist, the LCPIH also assumes that capital markets are perfect in the sense that all borrowing and lending occurs at the same riskless rate (Hall 1978), that the rate of return on assets is constant and expected to remain so (Ando and Modigliani 1983), that the assumption of a utility function separable in the major categories of consumption--durables, nondurables, and services--allows the estimation of one of these groups alone as the consumption concept (Flavin 1981), and that a liquidity constraint exists that is an upper bound beyond which consumption cannot occur (Flavin 1981; Hayashi 1985). The combined impact of these assumptions is to assume away the reasons for and problems associated with credit use. For a detailed account of the impact of these assumptions see Hayes (1986), where it is shown that when a more realistic attempt is made to incorporate the credit market as it currently exists, consumption can be expected to be sensitive to transitory shocks.
Intuitively, this theory can be argued as follows. Consider the impact of a negative transitory shock on a heavily indebted consumer. Additional credit may be available, but only at considerable expense and with some delay. In the interim, cash would be scarce. Items that cannot be purchased on credit (food and some services) bear the brunt of the adjustment while other categories suffer only to the extent that credit is unavailable. The attempt to stretch the budget to the next paycheck alters all the parameters of the demand system. The consumer is more price conscious and purchases food more on the basis of nutrition than taste. The impact of a positive transitory shock is not symmetric, because additional income may be used as a down payment on a budget item if the consumer feels that the repayments can be made from expected future income. The analysis of this decision is complex but depends on the relative interest rates on credit and savings, the expected change in the price of durables, and the ability to obtain additional credit to counteract the inflexibility that such additional repayments introduce into future budgeting decisions.
Finally, one must question the estimates of income elasticity of demand that would be forthcoming from a model that ignores these factors. If expenditures are used as the correct measure of income, the estimated elasticity would depend on the elasticity with respect to earned income and with respect to the "income" that could be achieved by using credit. It must also be determined whether the correct measure of income is disposable or discretionary and what the proportions of disposable income are that are transitory and permanent. The choice and estimation of a model that avoids these pitfalls is the subject of this study.
SELECTING THE UNDERLYING FUNCTIONAL FORM
The first step in model selection must be choosing the underlying functional form. The desirable properties of the static model are 1. it should be amendable to dynamic formulation, 2. it should impose no a priori restrictions on the relationships between estimated parameters, 3. it should not be derived from the assumption of maximization subject to a budget constraint, 4. no restrictions (other than adding up) should be placed on the parameters, 5. one should be able to test for a different reaction to economic conditions, depending on whether credit is or is not available (this entails estimating a model with both expenditures for durables and nondurables), and 6. the period between observations should be as short as possible to avoid problems with the simultaneous determination of supply and demand and of income and expenditures. (Monthly data are also required to allow an accurate estimation of the size of transitory shocks.)
The Almost Ideal Demand System (AIDS) (Deaton and Muellbauer 1980) meets specifications 1, 2, and 4. In addition, the AIDS aggregates perfectly over consumers. For condition 3, the AIDS demand function can provide a first-order approximation without maximizing assumptions (Deaton and Muellbauer 1980).
The AIDS demand system can be written w sub i = alpha sub i + summation of j gamma sub ij log P sub j + Beta sub i log (X/P) (1) where the budget share for commodity i (w sub i) depends on the prices of commodity groups 1 to j and total expenditure X. Alpha sub i may be interpreted as the average amount of the ith budget share when prices and real income are normalized to one. The parameter Y sub ij measures the change in the ith budget share corresponding to a one percent change in P sub j, ceteris paribus, and Beta sub i measures the change in the ith budget share caused by a one percent change in log X, ceteris paribus. P is a price index that can be approximated by: log P = summation of i w sub j log P sub j. This approximation, termed the Linear Approximation to AIDS, has been used successfully in Anderson and Blundell (1983) and in the seminal paper by Deaton and Muellbauer (1980).
Property 6 was achieved by estimating a monthly model from January 1959 to 1983.
Property 5 depends on obtaining an estimate of deviations between expected and actual income. Any test performed on the final model will be a joint test of the method used to determine transitory shocks. Considerable effort was, therefore, placed on specification of this variable. Hall (1978) provides an intuitively appealing motivation for including only consumption in t to predict consumption in t + 1. His model is, however, based on the Rational Expectations Permanent Income Hypothesis (REPIH). Other variables available are the consumer confidence index, the index of consumer sentiment, the lagged value of real income in previous months, and the increase in expenditures in previous months.
In the most general form, if ^/x sub t is the expected value of the entire x vector income and price variables and assuming that agents predict economic behavior on the current and past realizations over T periods of both prices and income, then we can write: ^/x sub t = omega x sub t-i i = 1 ... T.
An estimate, ^/x sub t, can be found by using the regression: ^/x sub t = omega x sub t-1 ... x sub t-T c sub t-1 ... c sub t-T CCI sub t-1 ... CCI sub t-T]' + equation sub t
(3) where CCI is a measure of consumer confidence, and c sub t is actual consumption in t. The inclusion of zeros in the omega matrix allows several alternative tests. In the results presented, estimates were not sensitive to restrictions on omega as long as the estimated equations had high R raise to 2 s.
If positive error terms are defined from regression (3) as negative transitory shocks and negative terms as positive shocks, then a model capable of estimating the impact of deviations from current period expectations can be built. This is shown schematically in Figure 1.
In Figure 1, line a is the actual path of real income, line b is the OLS estimate, and the vertical distance, c sub t, is the size of the transitory shock. The dummy variable delta allows for an asymmetric response to transitory shocks in the dynamic model outlined in the next section. Let D = x sub t - ^/x sub t Define delta = 0 if D greater than or equal to 0 delta = 1 otherwise In other words, delta is a relevant component of omega when predicted real income is greater than actual income or when consumers are faced with a negative transitory (unexpected) shock.
CHOOSING A DYNAMIC FORMULATION
The second stage is to extend the static model to capture dynamic adjustments to macroeconomic shocks. In addition to the properties specified for the static model, the dynamic model should 7. provide both short- and long-term elasticities, 8. have sufficient parameters to provide estimates of elasticities for different vectors of price and income, and 9. not impose any specific adjustment process.
Anderson and Blundell (1983) have extended the AIDS model to satisfy these restrictions. Consider the long-run equilibrium structure: w = f(Z,teta),
(4) where teta is a vector of the underlying preference parameters and Z is a vector of income and price variables. The AIDS model allows us to write: w = pi(teta)|/x,
(5) where |/x is a vector of transformed income and price variables, and pi is some constant matrix function of the underlying preference parameters.
Anderson and Blundell (1983) define a general first-order dynamic model as: delta w raise to n sub t = A delta |/x'sub t - B(W raise to n sub t-1 - pi raise to n (teta) x sub t-1 + equation sub t,
(6) where delta represents the first-difference operator, |/x' sub t, represents the |/x vector with the constant term excluded, and n defines an operator that deletes the nth row of any vector or matrix. Deletion of the last row of the share vector and the long-run parameter matrix pi(teta) is essential, because the budget shares sum to one. For a discussion of the econometric properties and the maximum likelihood estimation technique for this system, see Green and Johnson (1986).
The elements of the short-run parameter matrix A show how shares respond to month-to-month changes in economic conditions. The parameters of the n X n - 1 matrix B measure how much of the adjustment between the actual share and the share that is desired--given economic conditions--is made in one period.
Notice that in a long-run equilibrium, the diagonal terms of the matrix B will equal one, while the off-diagonal terms will equal zero. Also, delta w and delta x will equal zero; therefore, the equation becomes the static AIDS model. For purposes of this paper, the model can be extended to: Delta w raise to n sub t = (A + delta A*) delta |/x' - B (w raise to n sub t-1 - pi(teta) raised to n x sub t-1) + equation sub t
(7) where delta is the dummy variable defined above.
The income variables used in the delta |/x vector change the interpretation of the A matrices, but not the estimates of the long-run parameter matrix. Hence, we may include the measure of unexpected change in income from above.
The elements of A* represent the response of each share to shortrun negative shifts in the explanatory variables. The elements of (A + A*) represent the overall response to changes in these parameters. If inclusion of A* fails to increase the overall fit of the model significantly, it should be dropped. This is equivalent to rejection of the hypothesis positing dependency of consumer reaction to the sign of the movement in real or transitory income. Some coefficients of interest in the A matrix are a sub hh, a sub fh, and a sub ff. These terms are the reaction of the housing share to a one percent change in housing prices, the reaction of the food share to a one percent change in housing prices, and the reaction of the food share to a one percent increase in food costs, respectively. These are the average amounts by which consumers adjust to a one-month change in the dependent variables.
The off-diagonal elements of the B matrix represent a change in share i caused by a deviation from the desired level of share j where i = 1...n - 1 and j = 1...n. Typical elements would be b sub fh, b sub fe, and b sub ff. These terms may be interpreted as the deviation from the optimal level of the food share caused by a one percent difference between the actual and desired levels of the share devoted to housing, energy, and food, respectively. For example, a value of b sub fh = -.3 would imply that, in months in which the starting value of housing consumption was ten percent above the value dictated by the parameters of the long-run system, food consumption would, on average, be three percent below the desired figure.
Notice that the parameters of the long-run system need not be similar to those that would have been estimated in a static model. In Anderson and Blundell (1983), the parameters of the static model are in every case rejected in favor of the dynamic representation.
The elements of the parameter matrix pi(teta) are those that would be estimated in a static model if consumers could make all the desired adjustments instantaneously and if the data reflected these changes. Hence, we may impose the theoretical restrictions of Slutsky symmetry, homogeneity, and adding up on these parameters. The parametric restrictions that lead to the imposition of adding up, homogeneity, and symmetry, respectively, are expressed in equations 8, 9, and 10, respectively: summation of i from 1 to n alpha sub i = 1, summation of i from 1 to n gamma sub ij = 0,
summation of i from 1 to n beta sub i = 0, (8) summation of j gamma sub ij = 0, (9) gamma sub ij = gamma sub ji. (10)
The Marshallian and Hicksian measures of elasticities can be computed from the estimated long-run parameters as:
Epsilon sub ij = -1 + gamma sub ij/w sub i - Beta sub i, (11) Epsilon sub ij = gamma sub ij/w sub i - Beta sub i (w sub j/w sub i), (12) phi sub ii = -1 + gamma sub ii/w sub i + w sub i, (13) phi sub ij = gamma sub ij/w sub i + w sub j, (14)
where epsilon denotes Marshallian elasticities and phi denotes the income compensated or Hicksian measure. Expenditure elasticities can be obtained from the expression: n sub i = 1 + Beta sub i/w sub i.
These formulas can be derived by taking the derivative of the AIDS share equation with respect to prices or expenditures. Note that month-to-month variation in shares will cause the estimated elasticities to be different at each data point.
APPLYING THE MODEL TO U.S. DATA
Monthly expenditures figured on each of the categories listed in Table 1 are available from January 1959 in the Wharton data base. Price indices for each of these categories are either publicly available or can be constructed. The consumer confidence index before 1969 is unavailable. A proxy for consumer confidence, the actual increase in total spending for the previous month, allowed the estimation to begin in January 1959. The eleven-equation system was estimated by using the nonlinear routines of SAS.3 A full description of all results would greatly extend the paper. The emphasis of the reported results is on food sales and on credit use. Only the adding-up restriction has been imposed. Table 1 shows the effect on the various consumption shares of negative income shocks, measured in billions of dollars; these are the parameters of the A* matrix. For economists accustomed to thinking of food as a necessity, the decline in the food share seems counterintuitive. The coefficient on the food share was significant only at the 90 percent level. However, when a similar model was estimated using the share of food sales in total retail sales, this figure was negative and significant at the 95 percent level. 3 Durable goods are modeled as if they were one-use goods. Although this is obviously untrue, it is implicity assumed in the manner that the data on durable goods expenditures are presented. The alternatives are either to ignore this important segment or to assume falsely that the data on new durable purchases reflect the flow of services from the existing stock of durables.
It is commonly felt that American consumers first decide how much they will spend on necessities (that is, food and shelter) and then allocate the remainder of their income among other less important or luxury items. Any unexpected reduction in real income would be expected to have an impact first on savings and then on luxury items, durable goods, travel, and services. This reasoning best explains consumer behavior in economic conditions more stable, less complex, and with lower standards of living than those in this economy. In reality, food, as defined by government statistics or as seen in a typical shopping basket, contains items either superfluous to or of a quality and price well above what is required solely for nutritional needs. For most shoppers the marginal dollar spent on food seems to be regarded as something that can be foregone in the event of a budgetary squeeze. Other researchers have also found this to be the case (Hinson and Brooker 1981; Capps and Havlicek 1980).
In addition, food is one of the very few items that cannot be purchased on credit. In the event of negative transitory shocks, it is cash, not credit availability, that is scarce. Substituting lower-priced meats for more expensive cuts seems a painless way to stretch the cash budget to the end of the month. The increase in the share allocated to clothing and shoes, motor vehicles and parts, and other durables seems to indicate that consumers use credit cards to maintain consumption of these items when the shock is transitory. This makes perfect sense as long as the decline in living standards is viewed as temporary. If the above results have validity, they demonstrate the importance of including the differential availability of credit in demand analysis.
The results in Table 1 have implications for other retail sectors as well. Unexpected negative income shocks are felt also in gasoline and oil, other nondurables, household operations, and transport services. One possible explanation for this behavior is that people cut back somewhat on travel and household upkeep when liquidity is low. It is also possible, however, that fewer of the purchases made in these sectors can be made with credit cards. This latter explanation would also explain why sectors such as clothing, motor vehicles, other durables, and housing services--goods normally considered as luxuries--should increase their share during periods of tight liquidity. Whenever one share falls, however, at least one other must rise. The results do not necessarily imply that people spend more on these goods, only that they fail to reduce purchases as deeply as they do for those commodities whose shares fall. The figures for clothing and shoes are influenced by new purchases, once again demonstrating the importance of credit availability in maintaining sales.
The coefficients reported in Table 2 measure the short-term response to deviations in shares from their optimal level; these are the diagonal elements of B. Interpretation of these coefficients is complicated. It is impossible to differentiate between the agents' inability to adjust shares to their new desired levels and their desire to do so. If the level of expenditure in one expenditures category was constrained above equilibrium, some other share would, by definition, be below the desired level. Despite this, some general comments can be made.
The response for durable goods is, in general, lower than that for nondurables. Housing services have the lowest estimate. The low value for transport services may reflect the inability of public transport commuters to react to increases in ticket prices, whereas that for other services may be due to the large labor content of the commodity. The response time for food is low relative to other non-durables. Examination of the cross-price elasticities between food and housing seems to indicate that the response of the food share is retarded by its use as a counterbalance to the housing share.
THE INFLUENCE OF PRICES AND INCOME ON FOOD DEMAND
The dynamic nature of the model allows for the simultaneous estimation of the impact on the food share of both short- and long-term changes in its own-price and the price of competing products. For example, an increase in the price of housing may have two measurable effects on food demand. The immediate effect would be an increase in the housing share at the expense of the shares of the more flexible areas. Given sufficient time, however, people can gradually move expenditures away from housing and back into competing areas. The parameters estimated for the AIDS are those that occur in the absence of inflexibilities. They are, in essence, the response that agents would like to make to changes in economic conditions. It can, therefore, be expected that elasticities measured from these will be larger than those arising from models in which agents are assumed to adjust instantaneously. There is no a priori reason to expect the direction of response to be similar in both the long and short run. As can be seen from Tables 3 and 4, the effect of price changes in housing services on food demand depends greatly on whether the short or long run is being considered.
The estimates in Table 3 are the short-run estimates of a percentage change in the food share due to a one percent change in the prices of each expenditures category. The negative signs for transport and housing services may reflect the inability of agents to respond to increases in the prices of these services.
The long-run, or unconstrained, cross-price elasticities between food and the other expenditure categories are listed in Table 4. The structure of the AIDS allows for measurement of the long-run elasticities for all data points. Table 4 presents only the extremes of the estimated cross-price elasticities. Note that, although a short-run increase in the cost of household services tends to decrease the food share, this is not true in the long run. This implies that the reduction in the food share that occurs immediately after an increase in this cost is forced on consumers via the inflexibility of the housing market. The negative sign for the last three expenditure categories implies that, even in the absence of constraints, consumers maintain consumption of these services at the expense of the food share. These variables tend to be labor intensive. Increases in real national income through time have increased the relative price of these categories. Consumers have reduced the food share to increase their service-related expenditures. This trend seems likely to continue as long as real income rises.
Figures 2 and 3 plot the behaviors of the own-price and income elasticities of demand for food. These estimates are taken from the long-run model. The absolute values of both are high, reflecting the unconstrained nature of the estimate. Both measures have become inelastic and bahave countercyclically through time. The income elasticity of demand for food in 1974 was higher than at any point duringg the previous decade. As hypothesized, these measures are dependent on the direction of changes in real income. A given percentage decrease in real income causes a larger change in food expenditures than does a similar increase in real income.
A model has been constructed and estimated that acknowledges the limitations of published statistics and allows for the effect of credit on patterns of consumer expenditure. This model departs from the tradition of estimating demand systems only for nondurable goods. The results indicate that consumption patterns depend on transitory shocks, that consumers adjust slowly to movements in both price and permanent income, and that the availability of credit can be determinant in the short-term reaction to shocks.
The ready availability of credit has been ignored in most empirical application of demand analysis. The recent increase in credit dependence and reduction in the emergency fund level of U.S. households will be increasingly detrimental to the performance of models that ignore these developments. The results provide support for the hypothesis that the refusal of most retail food stores to accept credit occasionally reduces sales of food items that can be substituted for with less expensive foods.
Some tentative policy conclusions can also be drawn from the results. Many consumers now seem to use credit to smooth out the effects of temporary economic fluctuations and to maintain living standards. Should this trend continue, the ability of fiscal policy to change economic conditions will be reduced. Conversely, the ability of the monetary authorities to influence the economy--via interest rates--will increase.
If credit use continues to grow at a rate faster than real per capita incomes grow, consumers will lose the ability to react to recessions should it become obvious that the economy has entered a long recessionary phase. Many consumers will be forced to spend a greater portion of their income for credit repayments than they desire. This reduction in discretionary income coupled with the reticence of consumers to take out new credit to cover the shortfall could exacerbate the effects of a recession.
These results indicate three fascinating avenues for further research. The first would be to build a demand system that categorizes credit repayment as an expenditure item and credit taken out as a form of income. The coefficients on these variables would provide useful information about factors determining credit use and the impact of credit repayment in expenditure patterns. The second possibility would be to collect time-series data on meat sales from one of the few retail food chains accepting credit. These sales could be compared with meat sales from similar stores within the same chain that do not accept credit. The effect--if any--of credit availability on the sales of better-quality meat cuts would be of interest to both food retailers and livestock producers. Finally, the results indicate that the possible impacts of credit availability for food purchases and the aims of the food stamp program--to deal with transitory poverty and raise farm prices--are very similar. This raises the possibility that the government may find it beneficial to provide a guarantee of credit for food purchases in place of food stamps. Because many consumers would eventually be in a position to repay this credit, society might obtain greater benefits for each dollar spent on a credit guarantee program than for a dollar spent on foods stamps.
Table : 1 The Short-Term (Monthly) Impact of an Unexpected One Percent Negative Deviation Between Actual and Expected National Disposable Income on the Share of Each Expenditure Category
Table : 2 Measured Short-Term Response to Deviation in Shares From Their Optimal Level
Table : 3 The Short-Term Response of the Food Share to Price Changes in Expenditure Category
Table : 4 Uncontrained Cross-Price Elasticities for Food
|Printer friendly Cite/link Email Feedback|
|Author:||Hayes, Dermont J.|
|Publication:||Journal of Consumer Affairs|
|Date:||Jun 22, 1989|
|Next Article:||Individual consumption within the household: a study of expenditures on clothing.|