The impact of advertising on aggregate consumption function ([dagger]).

Khan, Ashfaque H. ; Siddiqui, Rizwana

I. INTRODUCTION

Advertising plays an important role in the economic life of almost every market economy. The informative role that advertising plays in the modern world in motivating potential customers to increase consumption at the expense of saving is hardly debatable. At the time of writing The General Theory, Keynes listed several factors that might influence the propensity to consume but ignored advertising. Given the level of maturity of economic activity half a century ago the exclusion of advertising as a potential determinent of consumption is understandable. However, the same cannot be justified in recent years where advertising plays an important role in bringing new goods and services to the attention of consumers.

Two opposing views have emerged in recent years regarding the effects of advertising on consumption. Schmalensee (1972) did not find any evidence that advertising affects private consumption. Thus, expenditures on advertising can be regarded as economic waste. It simply reallocates the market positions of existing firms with no effect on private consumption or propensity to consume. Taylor and Weiserbs (1972); Metwally and Tamaschke (1972) and Parsons et at. (1979) on the other hand, found a positive impact of advertising on consumption. Hence, the informative role of advertising appears to be strong enough to motivate potential consumers to increase their consumption expenditure. Given its importance in industrialized countries it has been surprising that the impact of advertising on aggregate consumption has not yet received much attention in the literature. (1) This issue has been completely neglected in the case of developing countries.

The purpose of this paper is to make a modest yet not insignificant attempt to examine the impact of advertising on aggregate consumption for a developing country like Pakistan using time-series data for the period 1969-70 to 1987-88. Advertising expenditure in Pakistan has increased at a compound growth rate of 19.0 percent and has averaged 0.05 percent of GNP during the period under analysis. (2) Hence a study of Pakistan provides an opportunity to examine advertising's impact on aggregate consumption. In so doing, however, it is emphasized that the relationship between advertising and consumption is not unidirectional but simultaneous. An increase in advertising expenditure will stimulate private consumption expenditure (if advertising is not regarded as economic waste). This. will, in turn, attract new entrants into the market. These newcomers will use advertising not only to establish themselves but to gain market share. The encumbent will also use advertising to protect their market share. Thus, an increase in private consumption will also increase advertising expenditure.

The plan of the paper is as follows. In Section II we discuss methodology and data. The results corresponding to a single-equation model as well as simultaneous-equation model are presented and discussed in Section III. Concluding remarks are contained in the final section.

II. METHODOLOGY AND DATA

To achieve the objective of the paper we specify an aggregate consumption function with advertising as an argument.

[C.sub.t] = [[alpha].sub.0] + [[alpha].sub.1] [y.sub.t] + [[alpha].sub.2] [A.sub.t] + [[alpha].sub.3] [C.sub.t-1] ... ... ... (1)

where:

[C.sub.t] = Real private consumption;

[y.sub.t] = Real disposable income adjusted for remittances; (3)

[A.sub.t] = Advertising expenditure; and

[C.sub.t-1] = Last year consumption used as a proxy for wealth.

Equation (1) is not suitable for estimation purposes because of the following three reasons. First, it is the advertising intensity rather than advertising levels which is important for businessmen because they think more of advertising/sales ratio rather than the actual amount spent on advertising. Second, the high correlation between [Y.sub.t] and [A.sub.t] leads to the problem of multi-collinearity which makes our estimates inefficient. Third, since we use annual time series-data, advertising intensity is less likely to have its full effect on the propensity to consume for a period less than one year. As such, we re-specify the aggregate consumption function as:

[(C/y).sub.t] = [[alpha].sub.0] + [[alpha].sub.1] [y.sub.t] + [[alpha].sub.2] [(A/C).sub.t] + [[alpha].sub.3] [(C/y).sub.t-1] ... ... (2)

where:

[(C/y).sub.t] = The average propensity to consume;

[(A/C).sub.t] = Advertising intensity; and

[(C/y).sub.t-1] = Lagged average propensity to consume used as a proxy for wealth variable.

It may be pointed out that instead of consumption expenditure we use average propensity to consume as an endogenous variable because we believe that advertising intensity will influence the propensity to consume more than the consumption level itself.

As pointed out in the preceding section the relationship between advertising and consumption is bidirectional therefore, advertising itself is an endogenous variable and is determined by its own previous level as wall as private consumption expenditure.

[(A/C).sub.t] = + [[beta].sub.0] + [[beta].sub.1] [(C/Y).sub.t] + [[beta].sub.2] [(A/C).sub.t-1) ... ... (3)

The bidirectional relationship between consumption and advertising requires that Equations (2) and (3) must be estimated simultaneously using an appropriate estimation technique. It may be pointed that Equations (2) and (3) form a system of structural equations which is mathematically complete in the sense that there are two equations for two endogenous variables C/Y and A/C The use of Ordinary Least Squares (OLS) to estimate these two equations would yield biased and inconsistent estimates. One way to circumvent the problem of the simultaneous equation system is to use the Two-Stage Least Squares (2SLS) method. However, 2SLS is essentially limited-information estimators where in the estimation of any structural equation complete information on all other structural equations in the model is not taken into account. The method of Three-Stage Least Squares (3SLS) is the full-information method which estimates the entire system simultaneously rather than each equation separately. If the specification of each equation in the system is correct then the 3SLS will be asymptotically more efficient than 2SLS. In this paper we use 3SLS to estimate Equations (2) and (3) simultaneously. (4)

Before we close this section a few words regarding data and its sources are in order. The data pertaining to real private consumption are taken from Statistical Supplement Pakistan Economic Survey 1987-88. Real disposable income is defined as real GNP minus income and corporation taxes adjusted for subsidies as well as remittances. This information is taken from the above-mentioned source. The data for advertising expenditure, as mentioned earlier, corresponds to advertising through television only; these are obtained from Pakistan Television Corporation. (5)

III. RESULTS

Having discussed the model and superiority of the estimation technique in the preceding sections we now turn to present results. Following the general practice we estimate a single-equation model comprising Equation (2) to examine the impact of advertising intensity on average propensity to consume with the help of OLS. (6) To check on for simultaneous equation bias we estimate Equations (2) and (3) simultaneously with the help of 3SLS.

Single Equation Results: Ordinary Least Squares

Equation (2) which specifies the average propensity to consume dependent upon the level of real disposable income, advertising intensity and lagged propensity to consume as arguments is estimated and the results are reported below. In order to make our analysis rich we estimated several variants of Equation (2) but report the results corresponding to only three.

III-1 [(C/Y).sub.t] = 0.225 - 0.000000339[y.sub.t] + 0.000283[A.sub.t]
 (1.27) (-0.43) (0.40)
+ 0.782 [(C/y).sub.t-1]
 (3.942)*

[[bar.R].sup.2] = 0.51; h = 0.87; F = 7.00

III-2 [(C/y).sub.t] = 0.243 - 0.000000285[y.sub.t] +
 (1.45) (1.035)

70.58 [(A/C)].sub.t] + 0.736 [(C/y).sub.t-1]
 (1.0613) (3.75)*

[[bar.R].sup.2] = 0.545; h = 0.167; F = 7.80

III-3 [(C/y).sub.t] = 0.261 - 0.000000372[y.sub.t] + 103.77
 (1.66) (-1.60)** (1.75)**
[(A/C).sub.t-1] + 0.713 [(C/y).sub.t-1]
 (3.86)*

[[bar.R].sup.2] = 0.60; h = 1.11; F = 9.39

In Equation III-1 average propensity to consume is specified to depend upon the level of real disposable income, level of advertising expenditure and lagged propensity to consume. It can be seen that the advertising variable which possesses a positive sign failed to reach the traditional level of significance. The level of disposable income, though bearing a correct sign, also failed to reach the significance level. Hence, the result in Equation III-1 suggests that advertising has no significant influence on the propensity to consume.

In Equation III-2, when the level of advertising expenditure is replaced with advertising intensity, the estimates improved slightly in terms of [[bar.R].sup.2] and t-statistics. However, advertising intensity once again failed to reach the significance level thereby suggesting that advertising intensity has no significant impact on the propensity to consume.

In Equation III-3, we introduced a reaction lag under the assumption that advertising intensity will influence propensity to consume, with one year lag. It can be seen that estimates are improved further with the introduction of reaction lag in advertising intensity. Both the level of income and advertising intensity are found to be statistically sigfnificant at the 10 percent level of significance. Thus, the results in Equation III-3 suggest that advertising intensity does influence the propensity to consume with a lag of one year, though the effects are statistically significant only at the 10 percent level.

The overall results corresponding to a single equation are therefore, not very encouraging. On the basis of these results one cannot reach a definite conclusion about the relationship between the advertising intensity and the propensity to consume. The only variable which is found to be statistically significant in all the three" equations is the lagged propensity to consume which is used as a proxy for the wealth variable.

Simultaneous Equation Results : Three Stage Least Squares

Next, a simultaneous equation system consisting of Equations (2) and (3) is estimated with the help of 3SLS method. We argue that the relationship between advertising and consumption is interdependent and must be estimated simultaneously. It is not uncommon in macroeconometric studies to make compromises in terms of variables to include or exclude as arguments in the specific functions while going from the specification to the estimation stage. This has also been true in the present case. We estimated several variants of Equations (2) and (3) simultaneously but report the results corresponding to only two exercises.

III-4 [(C/y).sub.t] = 0.40 - 0.0000009[y.sub.t] + 286.02 [(A/C).sub.t]
 (2.29)* (2.07)* (2.19)*
+ 0.54 [(C/y).sub.t-1]
 (2.73)*

[[bar.R].sup.2 = 0.99; DW = 1.60; SER = 0.027

[(A/C).sub.t] = -0.0009 + 0.0012 [(C/y).sub.t] + 0.79 [(A/C).sub.t-1]
 (1.19) (1.30) (5.83)

[[bar.R].sup.2] = 0.79; DW = 1.50; SER = 0.0009

III-5 [(C/y).sub.t] = 0.64 - 0.0000013[y.sub.t] + 363.46 [(A/C).sub.t]
 (5.13)* (4.53)* (4.36)*
+ 0.28 [(C/y).sub.t-1]
 (2.04)*

[[bar.R].sup.2] = 0.99; DW = 1.40; SER = 0.031

[A.sub.t] = -60.7 + 0.00091[C.sub.t] + 0.25[A.sub.t-1]
 (6.60)*(7.68)* (2.31)*

[[bar.R].sup.2] = 0.98; DW = 1.40; SER = 7.96

A cursory look at Equation III-4 is sufficient to see that advertising intensity is found to be statistically significant with the positive sign. (7) This suggests that advertising intensity is an important determinant of the propensity to consume. Furthermore, an increase in advertising intensity leads to an increase in the propensity to consume. What follows from this result is the fact that advertising in Pakistan cannot be regarded as economic waste. Its significant impact on the propensity to consume clearly shows the strong informative role that advertising has been playing in Pakistan to motivate potential consumers to increase their consumption expenditure. The income variable is found to be statistically significant with the negative sign. This result suggests that as income increases the average propensity to consume falls. This is consistent with the fact that the marginal propensity to consume (MPC) is less than unity. (8) The lagged propensity to consume used as a proxy for wealth is also found to be statistically significant with the positive sign.

In the second Equation of III-4, the average propensity to consume used as an explanatory variable, though bearing a positive sign, failed to reach the significance level. This suggests that the effect of an increase in the propensity to consume on advertising intensity is not strong. However, when the advertising intensity is changed to the advertising level in the second Equation of III-5, the overall results of the model improved significantly. In the first Equation, not only did the magnitude of the coefficients of income and advertising intensity increased considerably but also the significance level of the coefficient also improved (t-statistics of the coefficients are almost double compared with the first Equation in III-4). The advertising intensity is found to have a strong effect on the propensity to consume. In the second Equation of III-5, private consumption is found to be statistically significant with the positive sign. The result suggests that an increase in private consumption also increases advertising expenditure. The results in Equation III-5 confirm our earlier assertion that the relationship between advertising and consumption is bidirectional. (8) That is, an increase in advertising intensity increases the propensity to consume, and an increase in consumption increases advertising expenditure.

It may be noted that our results improved considerably when we moved from the single-equation estimation to the simultaneous equation system. The results suggest the presence of simultaneous equation bias in the estimation of the single-equation model using the OLS estimation technique which biased the coefficient of advertising intensity (A/C) downward.

IV. CONCLUDING REMARKS

The purpose of this paper has been to examine the impact of advertising on aggregate consumption. Despite its limitation, this paper has some important findings. It is found that advertising intensity is an important determinant of the propensity to consume. Furthermore, an increase in private consumption increases advertising expenditure. What follows from these results is the fact that advertising seems to have a beating on aggregate spending and as such, it cannot be regarded as economic waste. Its informative role in motivating a potential consumer to increase his/her spending is confirmed. If this is the case, then advertising deserves special research and policy attention. Another implication that stems out from the result is the fact that advertising should be considered as an important factor which influences the aggregate consumption function.

Authoers' Note: We wish to thank Drs Asghar Qadir, Nadeem A. Burney and Eatzaz Ahmed for helpful discussion. Editorial assistance from Ali Zafar is acknowledged.

REFERENCES

Ashley, R., C. W. J. Granger and R. Schmalensee (1980) Advertising and Aggregate Consumption: An Analysis of Causality. Econometrica 48 : 5.

Della Valle, P. A, and N. Oguchi (1976) Distribution, the Aggregate Consumption Function and the Level of Economic Development: Some Cross-Country Results. Journal o/Political Economy 84 : 6.

Goldstein, M., and Mohsin S. Khan (1978) The Supply and Demand for Exports: A Simultaneous Approach. The Review of Economics and Statistics 60 : 2.

Khan, Ashfaque H. (1984) Inflation, its Variability and Consumption: Additional Evidence from Asian Countries. Submitted for publication in Savings and Development.

Khan, Ashfaque H. (1987) A Macroeconometric Study of the Growth Prospects of Pakistan's Economy. Unpublished Ph.D. dissertation. Baltimore: The Johns Hopkins University.

Khan, Ashfaque H. (1987a)Aggregate Consumption Function and Income Distribution Effect: Some Evidence from Developing Countries. World Development 15 : 10/11.

Metwally, M. M., and H. U. Tamaschke (1981) Advertising and The Propensity to Consume. Oxford Bulletin of Economics and Statistics 43 : 3.

Naqvi, Syed Nawab Haider, Ashfaque H. Khan, N. M. Khilji and A. M. Ahmed (1983) The P.I.D.E. Macro-econometric Model of Pakistan's Economy. Islamabad: Pakistan Institute of Development Economics.

Parsons, L J., R. L. Schultz and T. L. Pilon (1979) The Impact of Advertising on the Aggregate Consumption Function. Paper No. 695 Krannert Graduate School of Management, Purdue University, May.

Schmalensee, R. (1972) The Economics of Advertising (Vol. 80, Contributions to Economic Analysis) Amsterdam: North-Holland.

Taylor, L D., and D. Weiserbs (1972) Advertising and The Aggregate Consumption Function. American Economic Review 62 : 4 September.

([dagger]) Comments on this paper have not been received.

(1) To the best of our knowledge only Schmalensee (1972); Taylor and Weiserbs (1972); Parsons et al. (1979); Ashley et al. (1980) and Metwally and Tamaschke (1982) have examined the effects of advertising on aggregate consumption for the U.S. and Australia.

(2) Advertising expenditure in this paper includes only advertising through television. Ideally, besides television, advertising expenditure should also have included advertising through press, radio, cinema and outdoor. However, data constraints have restricted us to include only advertising through television.

(3) When real disposable income was used the MPC always exceeded unity. Disposable income adjusted for remittances has also been used by Naqvi et al., (1983) and Khan (1984, 1987).

(4) We also estimated Equations (2) and (3) simultaneously by using the instrument variable variant of the 2SLS but the results were not satisfactory.

(5) We are grateful to Mr Matee-ur-Rehman of PTV for supplying the necessary data on advertising to us.

(6) Metwally and Tamaschke (1982) and Taylor and Weiserbs (1972) have also estimated single Equation model to examine the relationship between Advertising and Consumption.

(7) The t-statistics reported in the parentheses of each estimated coefficients can be interpreted as following a "quasi-T" distribution, though strictly speaking its distribution is asymptotic normal. The meaning of [[bar.R].sup.2] in simultaneous system is also at best ambiguous because it is not bounded (0, 1) but (- [infinity], 1). Hence low values of [[bar.R].sup.2] are not an indication of a "poor fit". Similarly, the use of the DW statistic to test the absence or presence of serial correlation is not clear when one is using a full-information estimator [see Goldstein and Khan (1978)].

(8) Our result is consistent with Khan (1987a) but differs from Della Valle and Oguchi (1976), who have made an assertion that despite changes in the level of income APC does not change.

(9) Our results differ with Taylor and Weiserbs (1972) who find that considering advertising as endogenous diminishes its impact on consumption. In our case it makes the impact of advertising stronger on consumption.

ASHFAQUE H. KHAN and RIZWANA SIDDIQUI *

* The authors are Senior Research Economist and Staff Economist at the Pakistan Institute of Development Economics, Islamabad.