# Advertising, concentration and profitability in manufacturing.

ADVERTISING, CONCENTRATION AND PROFITABILITY IN MANUFACTURING

I. INTRODUCTION

Economists have devoted much effort to empirically testing the hypothesized relationships between industrial concentration, profits and advertising. Twenty years ago Lester Telser concluded that there is little empirical support for an inverse association between advertising and competition [1964!. Later, other economists studying the relationship between advertising and concentration discovered a positive correlation between the four-firm concentration ratio, a measure of industry concentration, and advertising intensity. (1) Strickland and Weiss [1976! went further to show that there is a causal relationship between these elements: increases in concentration cause advertising to intensify, and, conversely, a rise in advertising intensity causes an increase in the level of concentration.

In recent years the empirical literature on advertising, concentration and profitability has atrophied because of dissatisfaction with its theoretical basis. As a new approach I adopt the theory of advertising developed by Stigler and Becker [1977! and extend it to incorporate the relationships among advertising, concentration and profitability. I then test its most important implications using data from the 1977 Census of Manufactures and the Detailed Input-Output Structure of the U.S. Economy. (2)

Section II examines the relationship between concentration and advertising intensity. Conventional theory, known as the inverted-U hypothesis, asserts that advertising is expected to increase with concentration at first, but to decrease at very high levels of concentration because collusion to avoid advertising costs becomes easier. Extending the Stigler and becker theory, I hypothesize that when concentration increases, even if the large firms do not collude and continue to act independently, they will eventually decrease their advertising intensity at high levels of concentration provided that the demand curve for the industry's products is relatively inelastic. If the industry's demand curve is relatively elastic, advertising intensity increases monotonically with concentration at all levels of concentration. This hypothesis is supported by the 1977 data taken from the Census and the Input-Output Structure of the U.S. Economy.

Section III focuses on the relationship between advertising and profitability. According to conventional theories, advertising intensity promotes product differentiation, leading to higher barriers to entry and higher profit margins. In the past, researchers, e.g., Strickland and Weiss [1976!, realized that, contrary to the implications of barriers-to-entry theory, the coefficient of advertising in a profit-margin equation was longer for a sample of industries producing producer-goods than for industries producing consumer-goods. They viewed this result as an anomaly, inconsistent with both their theoretical reasoning and with other important empirical results already published. I obtain a similar result which I am able to link to a theoretical structure, thus contributing to the refutation of the barriers-to-entry hypothesis.

I begin by extending the Stigler and Becker theory to treat advertising and research and development (R&D) as two complementary intangible capital goods. The sample is divided into two groups: producers of homogeneous goods, who spend relatively more on R&D, and producers of heterogeneous goods, who spend relatively more on advertising. Since we are unable to measure R&D costs at the appropriate industry level, it is not included among the regression variables. If R&D and advertising intensity are correlated, we should expect the coefficient of advertising in the profitability equation in the homogeneous-goods subsample to exceed the coefficient of advertising in the profitability equation in the heterogeneous-goods subsample. In this case, the omission of the R&D variable provides an implication for testing the new hypothesis.

The data and the empirical results of testing the two hypotheses are described in Section IV.

II. THE THEORY OF THE RELATIONSHIPS

BETWEEN CONCENTRATION AND ADVERTISING

Following Stigler and Becker [1977!, assume that the output x of a certain firm and its advertising A are the inputs into a commodity Z produced and consumed by households. If we let g = g(A) reflect the impact of advertising on X when it is transformed into Z, we may write, as did Stigler and Becker,

(1) Z = g(A)X

where the multiplicative form is a matter of mathematical convenience. Stigler and Becker show that when the level of advertising is held constant, the firm perceives the elasticity of the demand curve to be the same for product X and commodity Z. It thus becomes obvious that advertising is more efficient when the firm's demand curve is more elastic. At an extreme, if the firm's demand is perfectly elastic, doubling g (by increasing A) would enable the firm to double its price for X. If the demand curve confronting the firm is elastic but the elasticity is close to unity, doubling g would enable the firm to increase the price of X by only a little bit.

In the Stigler and Becker model, the marginal revenue from additional advertising by the ith firm is equal to the marginal cost of advertising as follows:

(2) ([P.sub.X.sub.i!g'/g)1 + 1/[eta.sub.i!) = [P.sub.a[).

Assume that the total value of shipments in a certain industry is R, and firm i contributes [R.sub.i! to the total. Then if there are N firms in the industry, we have

[Mathematical Expression Omitted!

(4) [s.sub.i! = [R.sub.i!/R = ([P.sub.x.X.sub.i!)/R.

If firms act independently of each other, and we assume a Cournot mode of oligopolistic behavior, equation (2) can be rewritten for the ith firm as follows:

(5) (Rg'/g)[s.sub.i!(1 + [sub.i.!/eta) = [P.sub.a!.

In equation (5) eta is the elasticity of demand for the industry's product ([eta.sub.i! = [eta!/[S.sub.i!).

There is ample evidence that the concentration level in an industry may increase when the leading firms invest in the development of new technology. By lowering their marginal costs, the new technology enables firms to lower the price to consumers and increase their market share. (3) For simplicity assume that a leading firm invests in the development of a new technology and increases its market share. If [eta! Equation (5) is equal to -1, then as long as the share of the leading firm is less than 0.5, an increase in its market share would cause its marginal revenue from advertising (the left-hand side of equation (5)) to increase, leading the firm to intensify its level of advertising. This can be shown by differentiating the lefthand side of equation (5) with respect to [s.sub.i! and equating it to zero. We obtain the result that the marginal revenue from advertising increases up to the level where [s.sub.i! = [-eta/2!, and thereafter it decreases. Note that if [eta! = -0.5, marginal revenue from advertising increases with [s.sub.i! up to a level of [s.sub.i! = 0.25, and then decreases. If \[eta!\ [is greater than or equal to ! 2, the marginal revenue from advertising increases monotonically with [s.sub.i!.

A testable implication emerges from this partial-equilibrium analysis: as the share of the leading firm increases, it increases its advertising intensity. The small firms whose shares decrease cut back their level of advertising. While at a very low level of concentration the small firms may dominate, because the share of the large firm increases at the expense of the small firms, it is likely that as concentration increases the industry's advertising intensity also increases. But when the share of the large firm exceeds the critical level of [s.sub.i! = -[eta!/2, the large firm reduces its level of advertising, and, as a result, the advertising intensity of the industry as a whole decreases. Early studies, such as Strickland and Weiss [1976!, promoted the hypothesis that advertising is expected to increase with concentration at first, but to decrease at very high levels of concentration because collusion to avoid advertising costs becomes easier. This is known as the inverted-U hypothesis. My hypothesis, which draws heavily on Stigler and Becker, is different. First, I assume that firms act independently of each other. Second, the inverted-U is limited to industries with relatively less elastic demand curves (\[eta!\ 2). For industries with relatively very elastic demand curves, advertising is expected to increase with concentration even at very high levels of concentration. If the elasticity of the demand curves facing highly concentrated industries varies across a wide range of values, I would expect advertising to be highly correlated with concentration at first, but to be uncorrelated with concentration at very high levels of concentration.

III. THE THEORY OF THE RELATIONSHIPS

BETWEEN ADVERTISING AND PROFITABILITY

Advertising effects last longer than one year. When advertising is aimed at reminding old customers, it is akin to maintenance costs for capital equipment. When advertising is aimed at reaching potential new customers, it is akin to investment in

new capital. The dollar value of the intangible advertising capital stock can in principle be derived by accumulating past advertising expenditures net of the associated depreciations. Coupled to the problem of determining how much of current advertising expenditures should be treated as maintenance costs, and how much is actually allocated to creating new intangible capital, there is the problem of determining the appropriate depreciation rate for advertising. When Bloch [1974! and Ayanian [1975! treat advertising as capital amortized at a 5 percent rate, they find profitability is not highly correlated with advertising. Weiss [1969! assumes that advertising capital should be depreciated over a much shorter six-year life, and he finds that advertising and profitability remain positively correlated. I do not pretend to know what is the true life of advertising. I am inclined to think that it is rather short.

At this point I would like to offer a plausible hypothesis that draws heavily on Stigler and Becker [1977!. Equation (1), namelt Z =g(A)X is familiar not only to students of industrial organization, but also to students of technical change in production. The coefficient g(A) could be viewed as corresponding to Solow's [1957! neutral production-function shifter. This coefficient can embody in varying proportions both technical change and advertising. Let us consider a hypothetical case where a firm that makes light bulbs decides to invest in a new technology that would double the hours of light per light bulb. Hours of light are denoted by Z, and Z is measured along the horizontal axis in Figure 1. As a result, the marginal cost of Z would shift downwards from [MC.sub.0! to [MC.sub.1!. (Let L represent any variable input, say labor. Then [MC = P.sub.l!/([derivative!Z/[derivative!L) and [derivative!Z/[derivative!L = g(A) ([derivative!X/[derivative!L). q.e.d.) The firm would move from point G to point H, and its producer's surplus would increase by the shaded area. Two additional points must be mentioned at this juncture. First, if the firm decided to embark on the change in technology, the expected shaded area must exceed the amortized investment in the new technology, i.e., its profit margin has to increase. SEcond, undertaking such an investment in new technology is useless without advertising: consumers must be informed that the new light bulbs provide twice as many hours of light as the old light bulbs. This line of argument suggests that technical change, particularly if it affects the quality of a product, is likely to be highly correlated with advertising intensity. It seems that in corss-sectional studies in which profitability is regressed on advertising intensity, the latter may capture part of the effect of technical change. In fact, if the association between advertising intensity and technical change is strong in both the production of homogeneous and heterogeneous goods, the impact of advertising on profitability should appear stronger in the case of homogeneous products. The following numerical example will demonstrate this.

Let A stand for advertising and R&D capital. there are two firms, the first producing a homogeneous product and the second producing a heterogeneous product. Each increases its combined amortized advertising and R&D capital by $100 per unit of time. As a result, the producer's surplus of each firm increases by $200 per unit of time. (R&D is not accounted for as a separate capital item.) Let the share of advertising in total A be 25 percent for the first firm and 75 percent for the second firm. It then appears that increasing advertising expenditures by $25 yielded an additional profit of $100 for the first firm. For the second firm, the same additional profit of $100 required an additional advertising expenditure of $75. If profit is a function of advertising effort, this result translates into a slope ([Delta! profit/[Delta! advertising) of 4 for the firm producing the homogeneous product, versus a slope of only 1.33 for the firm producing the heterogeneous product.

FIGURE 1

A Geometric Illustration of the Impact of an Increase in g(A) on the Firm

Advertising expenditures are correlated with intangible advertising and R&D capital intensity. Thus, my model predicts the not terribly surprising result that advertising expenditures will have a significant positive regression coefficient in the profitability equation. But, as in the discussion above, my model (a) predicts that the regression coefficient of advertising in the profit-margin equation should be larger in the homogeneous-goods subsample than in the heterogeneous-goods subsample, and (b) implies that such a result is due to the ommission of a variable. This point attracted the attention of researchers in the past, and it will be discussed in more detail below.

In the next section I test the implications of the hypotheses developed here and in the previous section.

IV. EMPIRICAL RESULTS

The data employed in this study are drawn from the 1977 Census of Manufactures, the 1977 Input-Output Structure of the U.S. Economy, and the Federal Trade Commission report by Weiss and Pascoe [1986!. A detailed description of the data, which contain 445 SIC-4 industries, is provided in appendix A.

The advertising intensity equation is

(6) Ad/S + [[alpha!.sub.0! + [[alpha!.sub.1!C + [[alpha!.sub.2!(Prof/S) + [[alpha!.sub.3!Hetro,

where Ad/S is advertising expenditures divided by the value of shipments (taken from the input-output tables), C is the four-firm concentration ratio, Prof/S is profit divided by the value of shipments (taken from the Census) and Hetro is a dummy variable that assumes a value of one for 174 industries producing heterogeneous products and a value of zero for 271 industries producing homogeneous products.

The concetration equation is

(7) C= [[beta!.sub.0! + [b.sub.1!(Ad/S) + [[[beta!.sub.2!(Ener/S),

where Ener/S is energy expenditures divided by the value of shipments (taken from the Census). I included Ener/S in the concentration equation because the year 1977 was in the midst of the so-called" energy crisis." It is likely that the rise in the cost of energy forced small marginal firms to shut down thus leading to a rise in the concentration ratio in industries that are relatively heavy users of energy.

The profitability equation is

(8) Prof/S = [[gamma!.sub.0! + [[gamma!.sub.1!C + [[gamma!.sub.2!(Ad/S), + [[gamma!.sub.3!Gro,

where Gro is the annual percentage rate of growth of real output for the period 1972-1977. Gro is included as a backdrop variable to account for the impact of either demand shifts, or marginal-cost shifts, on profitability.

Note that the Hetro variable is similar to the share of total sales going to consumers that was adopted by Strickland and Weiss [1976!. Thus my advertising intensity equation and concentration equation are almost identical to those employed by Strickland and Weiss. Below I introduce a plant-size proxy as an additional predetermined variable into the concentration equation, although I do not believe that the minimum efficiency scale can be reasonably estimated. This variable, denoted by Firms (firm size) is defined as follows:

(9) Firms = [Value of Shipments (1 - C)!

/(Number of Companies - 4).

By deleting the four industry leaders from this definition I hope to separate cause from effect. For if the cause is the firm scale and the effect is the level of concentration, which itself is measured by the size of the four leaders, failure to exclude the large firms from the definition of firm size renders indistinguishable cause and effect.

My profitability variable (Prof/S) is defined as value added less the cost of production and advertising expenditures, divided by the value of shipments. (4) In addition to payroll, the cost of production includes the average assets and inventories at the beginning and end of the year multiplied by the rate of interest, depreciation charges and rental payments. Thus, since the cost of capital is excluded from the profit margin, I do not include a measure of capital intensity as a predetermined variable in the profitability equation. If there are industries in which vast amounts of capital are required before the typical firm can produce efficiently, and if talent to amass such amounts of capital is scarce, then capital intensity should perhaps be included in the profitability equation. However, since such a relationship is surely captured by a higher level of concentration, the inclusion of a measure of capital intensity seems to be superfluous. Advertising expenditures are calculated by multiplying Ad/S, taken from the Input-Output Structure, by the value of shipments, taken from the Census.

The data for concentration taken from the 1977 Census reports are somewhat flawed because they do not reflect the market structure correctly. For goods like cement or ice, high costs of shipment render the markets regional or strictly local. In the case of automobiles, the impact of market internationalization is not accounted for by the concentration data. Fortunately, adjusted concentration ratios in manufacturing were calculated by Weiss and Pascoe [1986!: five different adjustments were made in order to move the concentration ratios in the direction of the real world. In particular, concentration ratios were adjusted for industries where markets are geographically fragmented, e.g., cement, and for industries facing substantial competition from imports, e.g., automobiles. For example, the concentration ratio of cement (SIC 3241) was adjusted upwards from 24 to 71 percent. The concentration ratio of motor vehicles (SIC 3711) was adjusted downwards from 93 to 80 percent. In what follows, I estimate the various models first using concentration ratios taken from the 1977 Census, and then using the concentration ratios adjusted by Weiss and Pascoe [1986!. As expected, the adjusted data yield somewhat more robust results.

The results shown in Table VII in appendix B are used to determine if Equation (7) belongs at all in the model. Table VII first shows both ordinary-least squares (OLS) and two-stage least squares (2SLS(I)) estimates of equations (6), (7) and (8). Notice that the coefficient of C in the Ad/S equation is significant in the OLS estimate (t = 2.85) but not in the 2SLS(I) estimate (t = 0.26). Moreover, the coefficient of C in the Prof/S equation is positive and mildly significant (t = 2.02) in the OLS estimate. It has the wrong sign and is not significant (t = 1.01) in the 2SLS(I) estimate.

To further test for potential simultaneity in model 2SLS(I), I applied the Wu-Hausman test to the three equations. The null hypothesis is that the regressors are oxogenous. The results for equations (6), (7) and (8) are 5.96, 0.23 and 2.28. For equations (6) and (8) the critical value of the [Chi!(4) is 9.49, and for euation (7) the critical value of the [Chi!(3) is 7.81 at the 5 percent level. From these results I conclude that there is no justification for treating the concentration variable C as an endogenous variable and equation (7) should be omitted from the model. Omitting equation (7) from the two-stage least squares estimates resulted in the third set of estimates in Table VII (denoted by 2SLS(II)). The results are very similar to the results of the Ad/S and Prof/S equations in the OLS model. I again applied the Wu-Hausman test for simulaneity to Equations (6) and (8). For equation (6), the result was 6.63, which is significant only at the 25 percent level. The critical level of the [Chi!(4) is 9.49 at the 5 percent level. For equation (8), the result was a negative number (-19.47). Moreover, three of the four scalars in the diagonal of the inverse of the covariance-difference matrix were negative. Given such inconclusive results, although it appears that the OLS is the correct econometric model, in what follows I opt to present the 2SLS(II) estimates which do not include equation (7). If Profit/S is regressed on the exogenous variables, including C, its estimated value is highly correlated with C. To get around this problem, C is replaced by Ener/S as an instrument. (The instruments are Hetro, Gro and Ener.)

Before moving on, notice that in Table VII there are two versions of the OLS estimate of equation (7), one with Firms and one without Firms. Adding Firms as another backdrop variable does not alter the results when estimating equation (7). The regression coefficient of Firms is positive and significant at the 0.4 percent level. This is the expected result. The other backdrop variable, Ener/S, is positive and significant with a t-ratio of 5.23 and 4.18 for the OLS estimates of the two C equations. A possible explanation for such a robust result is that the increase in the price of energy during the 1970s was less than merciful to small firms.

Table I contains both the ordinary least-squares (OLS) and the two-stage least squares (2SLS) estimates of equations (6) and (8). Each model is estimated twice: first with the Census concentration ratios, and second with the adjusted concentration ratios. In all regressions the coefficient of C in the Ad/S equation is highly significant, and more significant when adjusted concentration ratios are used (t = 3.05 versus t = 2.85 t= 3.40 versus t = 3.08). The coefficient of C is significant in the Prof/S equation. It is larger when adjusted concentration ratios are used (0.00068 versus 0.00037 0.00084 versus 0.00050). Also, as expected, the coefficient of C is more significant when the adjusted concentration ratios are used (t = 3.36 versus t = 2.02 t = 4.19 versus t = 2.78). A glance at the [R.sup.2! and F-ratios tells us that using the adjusted data yields somewhat more robust results.

The Relationship Between Advertising and Concentration. To test the hypothesis developed in section II, I first estimated an advertising equation containing both C and [C.sup.2!, and obtained the following result (t-ratios in parentheses):

(10) Ad/S = -0.0110 + 0.00018C (-2.39) (0.92)

[-2.7415C.sup.2! + 0.0608 Prof/S + 0.0123 Hetro. (-0.13) (4.22) (5.88)

Notice that in Table I the coefficient of C is significant (t = 2.85). Adding the [C.sup.2! variable to the advertising equation renders the coefficient of C insignificant (t = 0.92) and, moreover, yields a very poor result for the term [C.sup.2! (t = -0.13). This should not

surprise us given that the correlation coefficient between C and [C.sup.2! is 0.97. To get around this problem of multicollinearity I partitioned the sample of 445 industries into two subsamples, the first containing 379 industries (389 industries using the adjusted concentration ratios) with C [is less than or! 60 percent and the second 66 industries (56 industries) with C 60 percent. (5) The results of estimating the two structural equations with the first subsample are given in Table II, and they are similar to the results reported in Table I. In particular, when C [is less than or equal! 60 percent, the coefficient of C in the advertising intensity equation is significant in the OLS estimates (t = 2.47 for the unadjusted data, and t = 2.77 for the adjusted data). In the 2SLS estimates the coefficient of C is significant for the unadjusted data (t = 1.98). The coefficient of C has the expected positive sign, but it is insignificant for the adjusted data (t = 0.40). The results of estimating the two structural equations with the second subsample are presented in Table III. For

C 60 percent, the coefficient of C in the advertising intensity equation is positive and significant only at the 8 percent level (t = 1.75) in the OLS estimates for the unadjusted data. It is negative and insignificant (t = -0.41) in the OLS estimates for the adjusted data. It is negative and insignificant (t = -0.35 t = -.042) in the 2SLS estimates. The inverted-U hypothesis (or rather the inverted-V hypothesis) is not supported by the positive and mildly significant regression coefficient of C in one of the two OLS estimates. Nor is it supported by the negative and insignificant coefficient of C in the 2SLS estimates of the advertising intensity equation for the industries with C 60 percent. The results in Tables II and III lend support to the hypothesis that up to a point advertising intensity increases when concentration increases. As indicated by equation (5), for industries facing demand curves with low elasticities, advertising intensity reaches a peak and then falls. For industries facing relatively elastic demand curves, advertising intensity increases monotonically with concentration. In particular, the inverted-U hypothesis, that advertising is expected to decrease at very high levels of concentration because collusion to avoid advertising costs becomes easier, would be supported only if the coefficient of C in the Ad/S equation were both negative and significant for C 60. The results in Table

III tell us a different story.

In Table I the regression coefficient of Hetro in the advertising intensity equation is 0.01 and significant in both the OLS and 2SLS estimates for the unadjusted and adjusted concentration ratios. This result indicates that industries producing heterogeneous products tend to spend on advertising one percent of total sales more than do industries producing homogeneous products. The mean value of Ad/S for industries producing heterogeneous products is 0.017, and it is 0.0004 for industries producing homogeneous products. If we assume that the variances of the two populations are equal, then the difference between the two means is significantly positive with a T-ratio of 7.22 and 443 degrees of freedom. If we assume that the two variances are unequal, the difference is again positive with a t-ratio of 5.99 and approximatly 193 degrees of freedom. (6) This finding, coupled with the suspicious hypothesis that advertising intensity leads to higher concentration levels, implies a significantly higher concentration level for industries producing heterogeneous products. However, the mean unadjusted concentration ratio for the 174 industries producing heterogeneous products in 1977 is 39.14 percent and for the 271 industries producing homogeneous products it is 37.80 percent. The difference between the two means is not significantly different from zero with a t-ratio of 0.67 and 443 degrees of freedom. The mean adjusted concentration ratio for 174 industries producing heterogeneous products in 1977 is 36.00 percent and for the 271 industries producing homogeneous products is 38.15 percent. The difference between the two means is negative and not significantly different from zero with a t-ratio of -1.20 and 443 degrees of freedom. Thus the comparison of the two means of the two subsamples does not confirm the hypothesis that causation runs from advertising to concentration.

At this point I wondered if the inter-temporal change in the level of concentration was different in industries producing heterogeneous products as compared to those producing homogeneous products. In order to compare the means of subsamples for the Census years 1963, 1972 and 1977, I had to sacrifice 136 four-digit industries that had been reclassified over the years. (7) The results are presented in Table IV. While the mean concentration level of all 309 industries hardly changed at all from 1963 to 1977, the mean concentration level of the 127 industries producing heterogeneous products increased by about three percentage points and the mean concentration level of the 182 industries producing homogeneous products decreased by one percentage point. The difference in the means between 1963 and 1977 insignificant for the sample of all 309 industries and for the sample of 182 industries producing homogeneous products. It is almost significant (one-tail test) at the 10 percent level for the sample of 127 industries producing heterogeneous products.

In the framework of the theory formulated by Stigler and Becker, particularly as summarized by equation (2), increased concentration benefits the large firm by enabling it to realize economies of scale in advertising. This benefit is only partly offset by the effect of a less elastic demand curve associated with higher levels of concentration. In other words, where advertising is important it gives the already large firm a slight extra edge.

The Relationship Between Advertising and Profitability. We now turn our attention to the hypothesis developed in section III, namely that advertising expenditures are a proxy for intangible capital created by advertising and research and development. The OLS regression coefficients on Ad/S in the profitability equation in Tables I, II and III are 0.86, 0.84 and 0.95 (0.83, 0.82 and 0.90) with t-ratios of 4.55, 4.33 and 1.53 (4.42, 3.89 and 2.09), respectively. The 2SLS regression coefficients are 0.93, 0.87 and 1.14 (1.12, 0.77 and 1.76) with t-ratios of 1.64, 1.47 and 0.65 (1.98, 1.15 and 2.05), respectively. Notice that the value added as defined by the Census does not exclude the cost of services purchased from other business firms. Hence the cost of advertising was included in the profit margin in some early studies, e.g., Strickland and Weiss [1976!. I solved this problem by multiplying the ratio Ad/S, taken from the Input-Output Structure of the U.S. Economy, by the value of shipments, taken from the Census, and subtracting the result in dollar amounts from the value added.

In the literature advertising intensity serves as a proxy for product-differentiation barriers. The hypothesis of the product-differentiation barriers is accepted if the regression coefficient on Ad/S is positive and significant. (8) Logic dictates that the effect of a product-differentiation barrier, if it exists at all, should be stronger in industries producing heterogenous products than in industries producing homogeneous products. To test this hypothesis I partitioned the sample of 445 industries into two subsamples: 174 industries producing heterogeneous products and 271 industries producing homogeneous products. Partitioning was based on personal judgment. (See Appendix A for more detail.) The results of estimating equations (6) and (8) are presented in Table V. The coefficient on Ad/S in the profitability equation for the first subsample is 0.82 for the Census data and 0.70 for the revised data for the second subsample it is 1.26 for the Census data and 1.3k for the revised data. With t-ratios of 30.9 and 2.59 (and 2.62 and 2.79 for the revised data) both coefficients are highly siginificant. The puzzling higher coefficient in industries producing homogeneous products than in industries producing heterogeneous products suggests that the product-differentiation barrier-to-entry hypothesis is not at all supported by the data, and lends support to the hypothesis developed in section III.

As explained in section III, there is a possible solution to this puzzle: the prospective returns to intangible R&D capital are reduced without advertising, and hence the two highly correlated. If so, statistical studies which do not include R&D as a separate independent variable in the profitability equation, are flawed by an omitted variables bias. The explanatory power of the R&D variable is incorrectly attributed to the advertising variable. Thus, the higher coefficient on Ad/S for homogeneous goods industries may reflect nothin more than a statisticaly bias due to the omission of the R&D variable.

Note finally that the positive and significant regression coefficient on Prof/S in the Ad/S equation could be accounted for by two factors. First, a firm may stumble on a technological innovation which increases its profitability. The firm may find that its profit would increase even more following an advertising campaign to inform customers about its new innovation. Second, even if causation strictly runs from Ad/S to Prof/S, since the two are highly correlated Prof/S is expected to have a positive and significant regression coefficient in the Ad/S equation.

V. SUMMARY

The old empirical literature on concentration, advertising and profitability has atrophied because of dissatisfaction with the theoretical basis for the econometric estimates offered. In an attempt to remedy this situation, I adopt the theory of advertising developed by Stigler and Becker [1977! and extend it in two directions. First, I extend their concept of the marginal revenue from advertising so that it includes the firm's share in total shipments and the elasticity of the industry's demand curve. Under the assumption that the large firms behave like independent oligopolists, the new theory predicts an inverted-U relationship only for industries facing relatively less elastic demand curves. For industries facing highly elastic demand curves, the new theory predicts a positive relationship between concentration and advertising at all levels of concentration. The positive and significant (in three out of four equations) regression coefficient on concentration in the advertising equation for C [is less than or equal to! 60 percent, and the mixed results (a positive sign in one of the two OLS equations and a negative sign in the 2SLS equations) and insignificant regression coefficients in the advertising equations for C 60, rejects the inverted-U theory and lends support to the new theory.

Given the inconclusive evidence that from 1963 to 1977 concentration increased slightly in the heterogeneous subsample and decreased slightly in the homogeneous subsample, I find only limited support for the plausible hypothesis that already-large firms get an extra push from advertising. The advantage from economies of scale in advertising slightly exceeds the disadvantage of facing a less elastic demand curve.

Second, I extend the Stigler and Becker theory to treat advertising and R&D as two complementary intangible capital goods. Expenditures for R&D are not available at the four-digit SIC level. As a result, the omission of an R&D variable from the profitability equation provides a testable implication: the regression coefficient of advertising intensity in the profitability equation should be larger in the homogeneous-goods subsample than in the heterogeneous-goods subsample. The statistical tests confirm that the effect of advertising on profitability is significant, and greater in the homogeneous subsample than in the heterogeneous subsample. This result casts doubt on the yet-to-be confirmed product-differentiation-barrier doctrine. It is a direct confirmation of the extended Stigler and Becker hypothesis.

APPENDIX A

The Data

Data are taken from three sources: The U.S. Bureau of the Census, Census of Manufactures, 1977, vol. 1 the U.S. Bureau of Economic Analysis, The Detailed Input-Output Structure of the U.S. Economy, 1977, vol. 1, both from the Department of Commerce and the Federal Trade Commission report Concentration Ratios in Manufacturing, by Weiss and Pascoe [1986!.

Number of companies, unrevised concentration ratios and total shipments are obtained from the 1977 Census and Concentration Ratios in Manufacturing, Table 7. Revised concentration ratios are obtained from Weiss and Pascoe [1986!. Value-added by manufacture, rather than value of shipments, are given for industries 2011, 2013, 2271, 3312, 3331, 3332, 3575, and 3585, because their values of shipment contain substantial duplication.

Labor cost figures are obtained from the 1977 Census of Manufactures General Summary, Table 2.

Value-added and inventories are taken from the 1977 Census of Manufactures General Summary, Industry Statistics.

Assets, depreciation charges and rental payments are obtained from Gross Book Value of Depreciable Assets, Capital Expenditures, Retirements, Depreciation and Rental Payments, 1977 Census of Manufactures General Summary, Table 2.

The sample size of 445 four-digit SIC industries was obtained by excluding industries 2111, 2823, 3572, 3674 and 3661 from an exhaustive sample of 450 industries. The 1977 Census does not provide the four-firm concentration ratio for these industries. Industry 3332 has a concentration ratio of 100 percent.

Partitioning the sample of 445 industries into a subsample of 174 industries producing a heterogeneous product and 271 industries producing a homogeneous product was based on judgment. For example, almost the entire food and kindred industry (20) was classified as heterogeneous, and the entire primary metal industry (33) was classified as homogeneous furniture and fixture industries (25) were classified as heterogeneous if they sell to households, and homogeneous if they sell office furniture to firms.

Spending for energy is the cost of purchased fuels and electric energy used for heat and power by industry group. These data are taken from the 1977 Census of Manufactures Subject Statistics, vol. 1.

Growth rates are based on indexes of production with 1972-77 cross weights and value-added weights. Indexes of production data are taken from the 1977 Census of Manufacturers Indexes of Production, MC77-SR-14, Subject Series. The annual rate of growth is calculated by

r = [exp[(In(index of production))/5! - 1!100

where the index of production is for 1977, 1972 = 1.

The use of advertising services in dollars is taken from The Detailed Input-Output Structure of the U.S. Economy, vol. 1, Table 1. Advertisement is classified as 73.0200. To obtain the ration Ad/S the use of advertising services in dollars is divided by total industry output in the same table. (Example: SIC 2011 is numbered 14.010 in the input-output table. The ratio for this industry is calculated as 0.00147 = 42.9/29,112.7.) In cases in which the input-output classification covered several SIC-4 industries, the same ratio was used for this group of industries. Advertising cost is obtained by multiplying the Ad/S ratio by the value of shipments taken from the Census.

Means and standard deviations are reported in Table VI.

(*1) The University of New Mexico. I am grateful to Bruce Williamson, Peter Gregory, Tom Goodwin, Tim Sass, Raymond Sauer, Alok Bohara and two anonymous referees for helpful comments.

(1.) See Strickland and Weiss [1976, 1111!. It is interesting to note that economists were interested more in the concentration-profitability hypothesis than in the concentration-advertising hypothesis. For example, Weiss [1974! surveyed forty-six studies and concluded that, taken together, these studies "show a significant positive effect of concentration on profits or margins" [1974, 202!.

(2.) As explained below, I use concentration ratios taken from the 1977 Census, as well as adjusted concentration ratios taken from an FTC report by Weiss and Pascoe [1986!.

(3.) For Example, see Peltzman [1977! and Gisser [1984!.

(4.) In Gisser [1986!, I standardized the profit variable by dividing it by value added. Here, I standardized the profit variable by dividing it by the value of shipments in order to render my study more comparable with Strickland and Weiss [1976!.

(5.) Strickland and Weiss [1976! concluded that advertising intensity reaches its peak when C is somewhere between 49 and 57 percent. In another study [1984!, I argued that the cutoff point between the low-concentration industries and the high-concentration industries is in the 60-70 percent range. Thus, the 60 percent cutoff point is arbitrary, but reasonable.

(6.) For unequal variances the approximate solution for the t-ratio (and degrees of freedom) is known as the Behrens-Fisher problem. It is described in Wine [1964, 264!.

(7.) Note that reclassified industries differ in important ways from the unreclassified ones. There is evidence in Telser [1987, ch. 8! that there is more innovation among the reclassified than the unreclassified industries. If so, it is likely that the concentration levels in the reclassified industries increase faster over time than those of the unreclassified industries. If industries producing heterogeneous products are relatively more innovative than the other industries, the comparison of the means of subsamples for the Census years 1963, 1972 and 1977 might be biased.

(8.) In early studies, since advertising cost was not substracted from the value added, the hypothesis of the product-differentiation barrier was accepted if the regression coefficient of Ad/S significantly exceeded one.

REFERENCES

Ayanian, Robert. "Advertising and Rates of Return." Journal of Law and Economics, October 1975, 479-506.

Bloch, HArry. "Advertising and Profitability: A Reappraisal." Journal of Political Economy, Part I, March-April 1974, 267-86.

Gisser, Micha. "Price Leadership and Dynamic Aspects of Oligopoly in U.S. Manufacturing." Journal of Political Economy, December 1984, 1035-48.

_____. "Price Leadership and Welfare Losses in U.S. Manufacturing." American Economic Review, september 1986, 756-67.

Peltzman, Sam. "Gains and Losses from Industrial Concentration." Journal of Law and Economics, October 1977, 229-63.

Solow, Robert M. "Technical Change and the Aggregate Production Function." Review of Economics and Statistics, August 1957, 312-20.

Stigler, George J., and Gary S. Becker. "De Gustibus Non Est Disputandum." American Economic Review, March 1977, 76-90.

Strickland, Allyn D., and Leonard W. Weiss. "Adverising, Concentration, and Price-Cost Margins." Journal of Political Economy, October 1976, 1109-21.

Telser, Lester G. A Theory of Efficient Cooperation and Competition. Cambridge: Cambridge University Press, 1987.

_____. "Advertising and Competition." Journal of Political Economy, December 1964, 537-62.

U.S. Department of Commerce, Bureau of the Census. Census of Manufactures, Washington, D.C.: U.S. Government Printing Office, 1977.

U.S. Department of Commerce, Bureau of Economic Analysis. The Detailed Input-Output Structure of the U.S. Economy, vol. 1. Washington, D.C.: U.S. Government Printing Office, 1977.

Weiss, Leonard W. "Advertising, Profits and Corporate Taxes." Review of Economics and Statistics, November 1969, 421-30.

_____. "The Concentration-Profits Relationship and Antitrust," in Industrial Concentration: The New Leaning, edited by H.J. Goldschmid, H.M. Mann and J.F. Weston. Boston: Little, Brown and Company, 1974, 184-233.

Weiss, Leonard W., and George A. Pascoe, Jr. Adjusted Concentration Ratios in Manufacturing, 1972 and 1977. Statistical Report of the Bureau of Economics and the Federal Trade Commission, Fiche, June 1986.

Wine, R. Lowell. Statistics for Scientists and Engineers. Englewood Cliffs, N.J.: Prentice-Hall, 1964.

I. INTRODUCTION

Economists have devoted much effort to empirically testing the hypothesized relationships between industrial concentration, profits and advertising. Twenty years ago Lester Telser concluded that there is little empirical support for an inverse association between advertising and competition [1964!. Later, other economists studying the relationship between advertising and concentration discovered a positive correlation between the four-firm concentration ratio, a measure of industry concentration, and advertising intensity. (1) Strickland and Weiss [1976! went further to show that there is a causal relationship between these elements: increases in concentration cause advertising to intensify, and, conversely, a rise in advertising intensity causes an increase in the level of concentration.

In recent years the empirical literature on advertising, concentration and profitability has atrophied because of dissatisfaction with its theoretical basis. As a new approach I adopt the theory of advertising developed by Stigler and Becker [1977! and extend it to incorporate the relationships among advertising, concentration and profitability. I then test its most important implications using data from the 1977 Census of Manufactures and the Detailed Input-Output Structure of the U.S. Economy. (2)

Section II examines the relationship between concentration and advertising intensity. Conventional theory, known as the inverted-U hypothesis, asserts that advertising is expected to increase with concentration at first, but to decrease at very high levels of concentration because collusion to avoid advertising costs becomes easier. Extending the Stigler and becker theory, I hypothesize that when concentration increases, even if the large firms do not collude and continue to act independently, they will eventually decrease their advertising intensity at high levels of concentration provided that the demand curve for the industry's products is relatively inelastic. If the industry's demand curve is relatively elastic, advertising intensity increases monotonically with concentration at all levels of concentration. This hypothesis is supported by the 1977 data taken from the Census and the Input-Output Structure of the U.S. Economy.

Section III focuses on the relationship between advertising and profitability. According to conventional theories, advertising intensity promotes product differentiation, leading to higher barriers to entry and higher profit margins. In the past, researchers, e.g., Strickland and Weiss [1976!, realized that, contrary to the implications of barriers-to-entry theory, the coefficient of advertising in a profit-margin equation was longer for a sample of industries producing producer-goods than for industries producing consumer-goods. They viewed this result as an anomaly, inconsistent with both their theoretical reasoning and with other important empirical results already published. I obtain a similar result which I am able to link to a theoretical structure, thus contributing to the refutation of the barriers-to-entry hypothesis.

I begin by extending the Stigler and Becker theory to treat advertising and research and development (R&D) as two complementary intangible capital goods. The sample is divided into two groups: producers of homogeneous goods, who spend relatively more on R&D, and producers of heterogeneous goods, who spend relatively more on advertising. Since we are unable to measure R&D costs at the appropriate industry level, it is not included among the regression variables. If R&D and advertising intensity are correlated, we should expect the coefficient of advertising in the profitability equation in the homogeneous-goods subsample to exceed the coefficient of advertising in the profitability equation in the heterogeneous-goods subsample. In this case, the omission of the R&D variable provides an implication for testing the new hypothesis.

The data and the empirical results of testing the two hypotheses are described in Section IV.

II. THE THEORY OF THE RELATIONSHIPS

BETWEEN CONCENTRATION AND ADVERTISING

Following Stigler and Becker [1977!, assume that the output x of a certain firm and its advertising A are the inputs into a commodity Z produced and consumed by households. If we let g = g(A) reflect the impact of advertising on X when it is transformed into Z, we may write, as did Stigler and Becker,

(1) Z = g(A)X

where the multiplicative form is a matter of mathematical convenience. Stigler and Becker show that when the level of advertising is held constant, the firm perceives the elasticity of the demand curve to be the same for product X and commodity Z. It thus becomes obvious that advertising is more efficient when the firm's demand curve is more elastic. At an extreme, if the firm's demand is perfectly elastic, doubling g (by increasing A) would enable the firm to double its price for X. If the demand curve confronting the firm is elastic but the elasticity is close to unity, doubling g would enable the firm to increase the price of X by only a little bit.

In the Stigler and Becker model, the marginal revenue from additional advertising by the ith firm is equal to the marginal cost of advertising as follows:

(2) ([P.sub.X.sub.i!g'/g)1 + 1/[eta.sub.i!) = [P.sub.a[).

Assume that the total value of shipments in a certain industry is R, and firm i contributes [R.sub.i! to the total. Then if there are N firms in the industry, we have

[Mathematical Expression Omitted!

(4) [s.sub.i! = [R.sub.i!/R = ([P.sub.x.X.sub.i!)/R.

If firms act independently of each other, and we assume a Cournot mode of oligopolistic behavior, equation (2) can be rewritten for the ith firm as follows:

(5) (Rg'/g)[s.sub.i!(1 + [sub.i.!/eta) = [P.sub.a!.

In equation (5) eta is the elasticity of demand for the industry's product ([eta.sub.i! = [eta!/[S.sub.i!).

There is ample evidence that the concentration level in an industry may increase when the leading firms invest in the development of new technology. By lowering their marginal costs, the new technology enables firms to lower the price to consumers and increase their market share. (3) For simplicity assume that a leading firm invests in the development of a new technology and increases its market share. If [eta! Equation (5) is equal to -1, then as long as the share of the leading firm is less than 0.5, an increase in its market share would cause its marginal revenue from advertising (the left-hand side of equation (5)) to increase, leading the firm to intensify its level of advertising. This can be shown by differentiating the lefthand side of equation (5) with respect to [s.sub.i! and equating it to zero. We obtain the result that the marginal revenue from advertising increases up to the level where [s.sub.i! = [-eta/2!, and thereafter it decreases. Note that if [eta! = -0.5, marginal revenue from advertising increases with [s.sub.i! up to a level of [s.sub.i! = 0.25, and then decreases. If \[eta!\ [is greater than or equal to ! 2, the marginal revenue from advertising increases monotonically with [s.sub.i!.

A testable implication emerges from this partial-equilibrium analysis: as the share of the leading firm increases, it increases its advertising intensity. The small firms whose shares decrease cut back their level of advertising. While at a very low level of concentration the small firms may dominate, because the share of the large firm increases at the expense of the small firms, it is likely that as concentration increases the industry's advertising intensity also increases. But when the share of the large firm exceeds the critical level of [s.sub.i! = -[eta!/2, the large firm reduces its level of advertising, and, as a result, the advertising intensity of the industry as a whole decreases. Early studies, such as Strickland and Weiss [1976!, promoted the hypothesis that advertising is expected to increase with concentration at first, but to decrease at very high levels of concentration because collusion to avoid advertising costs becomes easier. This is known as the inverted-U hypothesis. My hypothesis, which draws heavily on Stigler and Becker, is different. First, I assume that firms act independently of each other. Second, the inverted-U is limited to industries with relatively less elastic demand curves (\[eta!\ 2). For industries with relatively very elastic demand curves, advertising is expected to increase with concentration even at very high levels of concentration. If the elasticity of the demand curves facing highly concentrated industries varies across a wide range of values, I would expect advertising to be highly correlated with concentration at first, but to be uncorrelated with concentration at very high levels of concentration.

III. THE THEORY OF THE RELATIONSHIPS

BETWEEN ADVERTISING AND PROFITABILITY

Advertising effects last longer than one year. When advertising is aimed at reminding old customers, it is akin to maintenance costs for capital equipment. When advertising is aimed at reaching potential new customers, it is akin to investment in

new capital. The dollar value of the intangible advertising capital stock can in principle be derived by accumulating past advertising expenditures net of the associated depreciations. Coupled to the problem of determining how much of current advertising expenditures should be treated as maintenance costs, and how much is actually allocated to creating new intangible capital, there is the problem of determining the appropriate depreciation rate for advertising. When Bloch [1974! and Ayanian [1975! treat advertising as capital amortized at a 5 percent rate, they find profitability is not highly correlated with advertising. Weiss [1969! assumes that advertising capital should be depreciated over a much shorter six-year life, and he finds that advertising and profitability remain positively correlated. I do not pretend to know what is the true life of advertising. I am inclined to think that it is rather short.

At this point I would like to offer a plausible hypothesis that draws heavily on Stigler and Becker [1977!. Equation (1), namelt Z =g(A)X is familiar not only to students of industrial organization, but also to students of technical change in production. The coefficient g(A) could be viewed as corresponding to Solow's [1957! neutral production-function shifter. This coefficient can embody in varying proportions both technical change and advertising. Let us consider a hypothetical case where a firm that makes light bulbs decides to invest in a new technology that would double the hours of light per light bulb. Hours of light are denoted by Z, and Z is measured along the horizontal axis in Figure 1. As a result, the marginal cost of Z would shift downwards from [MC.sub.0! to [MC.sub.1!. (Let L represent any variable input, say labor. Then [MC = P.sub.l!/([derivative!Z/[derivative!L) and [derivative!Z/[derivative!L = g(A) ([derivative!X/[derivative!L). q.e.d.) The firm would move from point G to point H, and its producer's surplus would increase by the shaded area. Two additional points must be mentioned at this juncture. First, if the firm decided to embark on the change in technology, the expected shaded area must exceed the amortized investment in the new technology, i.e., its profit margin has to increase. SEcond, undertaking such an investment in new technology is useless without advertising: consumers must be informed that the new light bulbs provide twice as many hours of light as the old light bulbs. This line of argument suggests that technical change, particularly if it affects the quality of a product, is likely to be highly correlated with advertising intensity. It seems that in corss-sectional studies in which profitability is regressed on advertising intensity, the latter may capture part of the effect of technical change. In fact, if the association between advertising intensity and technical change is strong in both the production of homogeneous and heterogeneous goods, the impact of advertising on profitability should appear stronger in the case of homogeneous products. The following numerical example will demonstrate this.

Let A stand for advertising and R&D capital. there are two firms, the first producing a homogeneous product and the second producing a heterogeneous product. Each increases its combined amortized advertising and R&D capital by $100 per unit of time. As a result, the producer's surplus of each firm increases by $200 per unit of time. (R&D is not accounted for as a separate capital item.) Let the share of advertising in total A be 25 percent for the first firm and 75 percent for the second firm. It then appears that increasing advertising expenditures by $25 yielded an additional profit of $100 for the first firm. For the second firm, the same additional profit of $100 required an additional advertising expenditure of $75. If profit is a function of advertising effort, this result translates into a slope ([Delta! profit/[Delta! advertising) of 4 for the firm producing the homogeneous product, versus a slope of only 1.33 for the firm producing the heterogeneous product.

FIGURE 1

A Geometric Illustration of the Impact of an Increase in g(A) on the Firm

Advertising expenditures are correlated with intangible advertising and R&D capital intensity. Thus, my model predicts the not terribly surprising result that advertising expenditures will have a significant positive regression coefficient in the profitability equation. But, as in the discussion above, my model (a) predicts that the regression coefficient of advertising in the profit-margin equation should be larger in the homogeneous-goods subsample than in the heterogeneous-goods subsample, and (b) implies that such a result is due to the ommission of a variable. This point attracted the attention of researchers in the past, and it will be discussed in more detail below.

In the next section I test the implications of the hypotheses developed here and in the previous section.

IV. EMPIRICAL RESULTS

The data employed in this study are drawn from the 1977 Census of Manufactures, the 1977 Input-Output Structure of the U.S. Economy, and the Federal Trade Commission report by Weiss and Pascoe [1986!. A detailed description of the data, which contain 445 SIC-4 industries, is provided in appendix A.

The advertising intensity equation is

(6) Ad/S + [[alpha!.sub.0! + [[alpha!.sub.1!C + [[alpha!.sub.2!(Prof/S) + [[alpha!.sub.3!Hetro,

where Ad/S is advertising expenditures divided by the value of shipments (taken from the input-output tables), C is the four-firm concentration ratio, Prof/S is profit divided by the value of shipments (taken from the Census) and Hetro is a dummy variable that assumes a value of one for 174 industries producing heterogeneous products and a value of zero for 271 industries producing homogeneous products.

The concetration equation is

(7) C= [[beta!.sub.0! + [b.sub.1!(Ad/S) + [[[beta!.sub.2!(Ener/S),

where Ener/S is energy expenditures divided by the value of shipments (taken from the Census). I included Ener/S in the concentration equation because the year 1977 was in the midst of the so-called" energy crisis." It is likely that the rise in the cost of energy forced small marginal firms to shut down thus leading to a rise in the concentration ratio in industries that are relatively heavy users of energy.

The profitability equation is

(8) Prof/S = [[gamma!.sub.0! + [[gamma!.sub.1!C + [[gamma!.sub.2!(Ad/S), + [[gamma!.sub.3!Gro,

where Gro is the annual percentage rate of growth of real output for the period 1972-1977. Gro is included as a backdrop variable to account for the impact of either demand shifts, or marginal-cost shifts, on profitability.

Note that the Hetro variable is similar to the share of total sales going to consumers that was adopted by Strickland and Weiss [1976!. Thus my advertising intensity equation and concentration equation are almost identical to those employed by Strickland and Weiss. Below I introduce a plant-size proxy as an additional predetermined variable into the concentration equation, although I do not believe that the minimum efficiency scale can be reasonably estimated. This variable, denoted by Firms (firm size) is defined as follows:

(9) Firms = [Value of Shipments (1 - C)!

/(Number of Companies - 4).

By deleting the four industry leaders from this definition I hope to separate cause from effect. For if the cause is the firm scale and the effect is the level of concentration, which itself is measured by the size of the four leaders, failure to exclude the large firms from the definition of firm size renders indistinguishable cause and effect.

My profitability variable (Prof/S) is defined as value added less the cost of production and advertising expenditures, divided by the value of shipments. (4) In addition to payroll, the cost of production includes the average assets and inventories at the beginning and end of the year multiplied by the rate of interest, depreciation charges and rental payments. Thus, since the cost of capital is excluded from the profit margin, I do not include a measure of capital intensity as a predetermined variable in the profitability equation. If there are industries in which vast amounts of capital are required before the typical firm can produce efficiently, and if talent to amass such amounts of capital is scarce, then capital intensity should perhaps be included in the profitability equation. However, since such a relationship is surely captured by a higher level of concentration, the inclusion of a measure of capital intensity seems to be superfluous. Advertising expenditures are calculated by multiplying Ad/S, taken from the Input-Output Structure, by the value of shipments, taken from the Census.

The data for concentration taken from the 1977 Census reports are somewhat flawed because they do not reflect the market structure correctly. For goods like cement or ice, high costs of shipment render the markets regional or strictly local. In the case of automobiles, the impact of market internationalization is not accounted for by the concentration data. Fortunately, adjusted concentration ratios in manufacturing were calculated by Weiss and Pascoe [1986!: five different adjustments were made in order to move the concentration ratios in the direction of the real world. In particular, concentration ratios were adjusted for industries where markets are geographically fragmented, e.g., cement, and for industries facing substantial competition from imports, e.g., automobiles. For example, the concentration ratio of cement (SIC 3241) was adjusted upwards from 24 to 71 percent. The concentration ratio of motor vehicles (SIC 3711) was adjusted downwards from 93 to 80 percent. In what follows, I estimate the various models first using concentration ratios taken from the 1977 Census, and then using the concentration ratios adjusted by Weiss and Pascoe [1986!. As expected, the adjusted data yield somewhat more robust results.

The results shown in Table VII in appendix B are used to determine if Equation (7) belongs at all in the model. Table VII first shows both ordinary-least squares (OLS) and two-stage least squares (2SLS(I)) estimates of equations (6), (7) and (8). Notice that the coefficient of C in the Ad/S equation is significant in the OLS estimate (t = 2.85) but not in the 2SLS(I) estimate (t = 0.26). Moreover, the coefficient of C in the Prof/S equation is positive and mildly significant (t = 2.02) in the OLS estimate. It has the wrong sign and is not significant (t = 1.01) in the 2SLS(I) estimate.

To further test for potential simultaneity in model 2SLS(I), I applied the Wu-Hausman test to the three equations. The null hypothesis is that the regressors are oxogenous. The results for equations (6), (7) and (8) are 5.96, 0.23 and 2.28. For equations (6) and (8) the critical value of the [Chi!(4) is 9.49, and for euation (7) the critical value of the [Chi!(3) is 7.81 at the 5 percent level. From these results I conclude that there is no justification for treating the concentration variable C as an endogenous variable and equation (7) should be omitted from the model. Omitting equation (7) from the two-stage least squares estimates resulted in the third set of estimates in Table VII (denoted by 2SLS(II)). The results are very similar to the results of the Ad/S and Prof/S equations in the OLS model. I again applied the Wu-Hausman test for simulaneity to Equations (6) and (8). For equation (6), the result was 6.63, which is significant only at the 25 percent level. The critical level of the [Chi!(4) is 9.49 at the 5 percent level. For equation (8), the result was a negative number (-19.47). Moreover, three of the four scalars in the diagonal of the inverse of the covariance-difference matrix were negative. Given such inconclusive results, although it appears that the OLS is the correct econometric model, in what follows I opt to present the 2SLS(II) estimates which do not include equation (7). If Profit/S is regressed on the exogenous variables, including C, its estimated value is highly correlated with C. To get around this problem, C is replaced by Ener/S as an instrument. (The instruments are Hetro, Gro and Ener.)

Before moving on, notice that in Table VII there are two versions of the OLS estimate of equation (7), one with Firms and one without Firms. Adding Firms as another backdrop variable does not alter the results when estimating equation (7). The regression coefficient of Firms is positive and significant at the 0.4 percent level. This is the expected result. The other backdrop variable, Ener/S, is positive and significant with a t-ratio of 5.23 and 4.18 for the OLS estimates of the two C equations. A possible explanation for such a robust result is that the increase in the price of energy during the 1970s was less than merciful to small firms.

Table I contains both the ordinary least-squares (OLS) and the two-stage least squares (2SLS) estimates of equations (6) and (8). Each model is estimated twice: first with the Census concentration ratios, and second with the adjusted concentration ratios. In all regressions the coefficient of C in the Ad/S equation is highly significant, and more significant when adjusted concentration ratios are used (t = 3.05 versus t = 2.85 t= 3.40 versus t = 3.08). The coefficient of C is significant in the Prof/S equation. It is larger when adjusted concentration ratios are used (0.00068 versus 0.00037 0.00084 versus 0.00050). Also, as expected, the coefficient of C is more significant when the adjusted concentration ratios are used (t = 3.36 versus t = 2.02 t = 4.19 versus t = 2.78). A glance at the [R.sup.2! and F-ratios tells us that using the adjusted data yields somewhat more robust results.

The Relationship Between Advertising and Concentration. To test the hypothesis developed in section II, I first estimated an advertising equation containing both C and [C.sup.2!, and obtained the following result (t-ratios in parentheses):

(10) Ad/S = -0.0110 + 0.00018C (-2.39) (0.92)

[-2.7415C.sup.2! + 0.0608 Prof/S + 0.0123 Hetro. (-0.13) (4.22) (5.88)

Notice that in Table I the coefficient of C is significant (t = 2.85). Adding the [C.sup.2! variable to the advertising equation renders the coefficient of C insignificant (t = 0.92) and, moreover, yields a very poor result for the term [C.sup.2! (t = -0.13). This should not

surprise us given that the correlation coefficient between C and [C.sup.2! is 0.97. To get around this problem of multicollinearity I partitioned the sample of 445 industries into two subsamples, the first containing 379 industries (389 industries using the adjusted concentration ratios) with C [is less than or! 60 percent and the second 66 industries (56 industries) with C 60 percent. (5) The results of estimating the two structural equations with the first subsample are given in Table II, and they are similar to the results reported in Table I. In particular, when C [is less than or equal! 60 percent, the coefficient of C in the advertising intensity equation is significant in the OLS estimates (t = 2.47 for the unadjusted data, and t = 2.77 for the adjusted data). In the 2SLS estimates the coefficient of C is significant for the unadjusted data (t = 1.98). The coefficient of C has the expected positive sign, but it is insignificant for the adjusted data (t = 0.40). The results of estimating the two structural equations with the second subsample are presented in Table III. For

C 60 percent, the coefficient of C in the advertising intensity equation is positive and significant only at the 8 percent level (t = 1.75) in the OLS estimates for the unadjusted data. It is negative and insignificant (t = -0.41) in the OLS estimates for the adjusted data. It is negative and insignificant (t = -0.35 t = -.042) in the 2SLS estimates. The inverted-U hypothesis (or rather the inverted-V hypothesis) is not supported by the positive and mildly significant regression coefficient of C in one of the two OLS estimates. Nor is it supported by the negative and insignificant coefficient of C in the 2SLS estimates of the advertising intensity equation for the industries with C 60 percent. The results in Tables II and III lend support to the hypothesis that up to a point advertising intensity increases when concentration increases. As indicated by equation (5), for industries facing demand curves with low elasticities, advertising intensity reaches a peak and then falls. For industries facing relatively elastic demand curves, advertising intensity increases monotonically with concentration. In particular, the inverted-U hypothesis, that advertising is expected to decrease at very high levels of concentration because collusion to avoid advertising costs becomes easier, would be supported only if the coefficient of C in the Ad/S equation were both negative and significant for C 60. The results in Table

III tell us a different story.

In Table I the regression coefficient of Hetro in the advertising intensity equation is 0.01 and significant in both the OLS and 2SLS estimates for the unadjusted and adjusted concentration ratios. This result indicates that industries producing heterogeneous products tend to spend on advertising one percent of total sales more than do industries producing homogeneous products. The mean value of Ad/S for industries producing heterogeneous products is 0.017, and it is 0.0004 for industries producing homogeneous products. If we assume that the variances of the two populations are equal, then the difference between the two means is significantly positive with a T-ratio of 7.22 and 443 degrees of freedom. If we assume that the two variances are unequal, the difference is again positive with a t-ratio of 5.99 and approximatly 193 degrees of freedom. (6) This finding, coupled with the suspicious hypothesis that advertising intensity leads to higher concentration levels, implies a significantly higher concentration level for industries producing heterogeneous products. However, the mean unadjusted concentration ratio for the 174 industries producing heterogeneous products in 1977 is 39.14 percent and for the 271 industries producing homogeneous products it is 37.80 percent. The difference between the two means is not significantly different from zero with a t-ratio of 0.67 and 443 degrees of freedom. The mean adjusted concentration ratio for 174 industries producing heterogeneous products in 1977 is 36.00 percent and for the 271 industries producing homogeneous products is 38.15 percent. The difference between the two means is negative and not significantly different from zero with a t-ratio of -1.20 and 443 degrees of freedom. Thus the comparison of the two means of the two subsamples does not confirm the hypothesis that causation runs from advertising to concentration.

At this point I wondered if the inter-temporal change in the level of concentration was different in industries producing heterogeneous products as compared to those producing homogeneous products. In order to compare the means of subsamples for the Census years 1963, 1972 and 1977, I had to sacrifice 136 four-digit industries that had been reclassified over the years. (7) The results are presented in Table IV. While the mean concentration level of all 309 industries hardly changed at all from 1963 to 1977, the mean concentration level of the 127 industries producing heterogeneous products increased by about three percentage points and the mean concentration level of the 182 industries producing homogeneous products decreased by one percentage point. The difference in the means between 1963 and 1977 insignificant for the sample of all 309 industries and for the sample of 182 industries producing homogeneous products. It is almost significant (one-tail test) at the 10 percent level for the sample of 127 industries producing heterogeneous products.

In the framework of the theory formulated by Stigler and Becker, particularly as summarized by equation (2), increased concentration benefits the large firm by enabling it to realize economies of scale in advertising. This benefit is only partly offset by the effect of a less elastic demand curve associated with higher levels of concentration. In other words, where advertising is important it gives the already large firm a slight extra edge.

The Relationship Between Advertising and Profitability. We now turn our attention to the hypothesis developed in section III, namely that advertising expenditures are a proxy for intangible capital created by advertising and research and development. The OLS regression coefficients on Ad/S in the profitability equation in Tables I, II and III are 0.86, 0.84 and 0.95 (0.83, 0.82 and 0.90) with t-ratios of 4.55, 4.33 and 1.53 (4.42, 3.89 and 2.09), respectively. The 2SLS regression coefficients are 0.93, 0.87 and 1.14 (1.12, 0.77 and 1.76) with t-ratios of 1.64, 1.47 and 0.65 (1.98, 1.15 and 2.05), respectively. Notice that the value added as defined by the Census does not exclude the cost of services purchased from other business firms. Hence the cost of advertising was included in the profit margin in some early studies, e.g., Strickland and Weiss [1976!. I solved this problem by multiplying the ratio Ad/S, taken from the Input-Output Structure of the U.S. Economy, by the value of shipments, taken from the Census, and subtracting the result in dollar amounts from the value added.

In the literature advertising intensity serves as a proxy for product-differentiation barriers. The hypothesis of the product-differentiation barriers is accepted if the regression coefficient on Ad/S is positive and significant. (8) Logic dictates that the effect of a product-differentiation barrier, if it exists at all, should be stronger in industries producing heterogenous products than in industries producing homogeneous products. To test this hypothesis I partitioned the sample of 445 industries into two subsamples: 174 industries producing heterogeneous products and 271 industries producing homogeneous products. Partitioning was based on personal judgment. (See Appendix A for more detail.) The results of estimating equations (6) and (8) are presented in Table V. The coefficient on Ad/S in the profitability equation for the first subsample is 0.82 for the Census data and 0.70 for the revised data for the second subsample it is 1.26 for the Census data and 1.3k for the revised data. With t-ratios of 30.9 and 2.59 (and 2.62 and 2.79 for the revised data) both coefficients are highly siginificant. The puzzling higher coefficient in industries producing homogeneous products than in industries producing heterogeneous products suggests that the product-differentiation barrier-to-entry hypothesis is not at all supported by the data, and lends support to the hypothesis developed in section III.

As explained in section III, there is a possible solution to this puzzle: the prospective returns to intangible R&D capital are reduced without advertising, and hence the two highly correlated. If so, statistical studies which do not include R&D as a separate independent variable in the profitability equation, are flawed by an omitted variables bias. The explanatory power of the R&D variable is incorrectly attributed to the advertising variable. Thus, the higher coefficient on Ad/S for homogeneous goods industries may reflect nothin more than a statisticaly bias due to the omission of the R&D variable.

Note finally that the positive and significant regression coefficient on Prof/S in the Ad/S equation could be accounted for by two factors. First, a firm may stumble on a technological innovation which increases its profitability. The firm may find that its profit would increase even more following an advertising campaign to inform customers about its new innovation. Second, even if causation strictly runs from Ad/S to Prof/S, since the two are highly correlated Prof/S is expected to have a positive and significant regression coefficient in the Ad/S equation.

V. SUMMARY

The old empirical literature on concentration, advertising and profitability has atrophied because of dissatisfaction with the theoretical basis for the econometric estimates offered. In an attempt to remedy this situation, I adopt the theory of advertising developed by Stigler and Becker [1977! and extend it in two directions. First, I extend their concept of the marginal revenue from advertising so that it includes the firm's share in total shipments and the elasticity of the industry's demand curve. Under the assumption that the large firms behave like independent oligopolists, the new theory predicts an inverted-U relationship only for industries facing relatively less elastic demand curves. For industries facing highly elastic demand curves, the new theory predicts a positive relationship between concentration and advertising at all levels of concentration. The positive and significant (in three out of four equations) regression coefficient on concentration in the advertising equation for C [is less than or equal to! 60 percent, and the mixed results (a positive sign in one of the two OLS equations and a negative sign in the 2SLS equations) and insignificant regression coefficients in the advertising equations for C 60, rejects the inverted-U theory and lends support to the new theory.

Given the inconclusive evidence that from 1963 to 1977 concentration increased slightly in the heterogeneous subsample and decreased slightly in the homogeneous subsample, I find only limited support for the plausible hypothesis that already-large firms get an extra push from advertising. The advantage from economies of scale in advertising slightly exceeds the disadvantage of facing a less elastic demand curve.

Second, I extend the Stigler and Becker theory to treat advertising and R&D as two complementary intangible capital goods. Expenditures for R&D are not available at the four-digit SIC level. As a result, the omission of an R&D variable from the profitability equation provides a testable implication: the regression coefficient of advertising intensity in the profitability equation should be larger in the homogeneous-goods subsample than in the heterogeneous-goods subsample. The statistical tests confirm that the effect of advertising on profitability is significant, and greater in the homogeneous subsample than in the heterogeneous subsample. This result casts doubt on the yet-to-be confirmed product-differentiation-barrier doctrine. It is a direct confirmation of the extended Stigler and Becker hypothesis.

APPENDIX A

The Data

Data are taken from three sources: The U.S. Bureau of the Census, Census of Manufactures, 1977, vol. 1 the U.S. Bureau of Economic Analysis, The Detailed Input-Output Structure of the U.S. Economy, 1977, vol. 1, both from the Department of Commerce and the Federal Trade Commission report Concentration Ratios in Manufacturing, by Weiss and Pascoe [1986!.

Number of companies, unrevised concentration ratios and total shipments are obtained from the 1977 Census and Concentration Ratios in Manufacturing, Table 7. Revised concentration ratios are obtained from Weiss and Pascoe [1986!. Value-added by manufacture, rather than value of shipments, are given for industries 2011, 2013, 2271, 3312, 3331, 3332, 3575, and 3585, because their values of shipment contain substantial duplication.

Labor cost figures are obtained from the 1977 Census of Manufactures General Summary, Table 2.

Value-added and inventories are taken from the 1977 Census of Manufactures General Summary, Industry Statistics.

Assets, depreciation charges and rental payments are obtained from Gross Book Value of Depreciable Assets, Capital Expenditures, Retirements, Depreciation and Rental Payments, 1977 Census of Manufactures General Summary, Table 2.

The sample size of 445 four-digit SIC industries was obtained by excluding industries 2111, 2823, 3572, 3674 and 3661 from an exhaustive sample of 450 industries. The 1977 Census does not provide the four-firm concentration ratio for these industries. Industry 3332 has a concentration ratio of 100 percent.

Partitioning the sample of 445 industries into a subsample of 174 industries producing a heterogeneous product and 271 industries producing a homogeneous product was based on judgment. For example, almost the entire food and kindred industry (20) was classified as heterogeneous, and the entire primary metal industry (33) was classified as homogeneous furniture and fixture industries (25) were classified as heterogeneous if they sell to households, and homogeneous if they sell office furniture to firms.

Spending for energy is the cost of purchased fuels and electric energy used for heat and power by industry group. These data are taken from the 1977 Census of Manufactures Subject Statistics, vol. 1.

Growth rates are based on indexes of production with 1972-77 cross weights and value-added weights. Indexes of production data are taken from the 1977 Census of Manufacturers Indexes of Production, MC77-SR-14, Subject Series. The annual rate of growth is calculated by

r = [exp[(In(index of production))/5! - 1!100

where the index of production is for 1977, 1972 = 1.

The use of advertising services in dollars is taken from The Detailed Input-Output Structure of the U.S. Economy, vol. 1, Table 1. Advertisement is classified as 73.0200. To obtain the ration Ad/S the use of advertising services in dollars is divided by total industry output in the same table. (Example: SIC 2011 is numbered 14.010 in the input-output table. The ratio for this industry is calculated as 0.00147 = 42.9/29,112.7.) In cases in which the input-output classification covered several SIC-4 industries, the same ratio was used for this group of industries. Advertising cost is obtained by multiplying the Ad/S ratio by the value of shipments taken from the Census.

Means and standard deviations are reported in Table VI.

(*1) The University of New Mexico. I am grateful to Bruce Williamson, Peter Gregory, Tom Goodwin, Tim Sass, Raymond Sauer, Alok Bohara and two anonymous referees for helpful comments.

(1.) See Strickland and Weiss [1976, 1111!. It is interesting to note that economists were interested more in the concentration-profitability hypothesis than in the concentration-advertising hypothesis. For example, Weiss [1974! surveyed forty-six studies and concluded that, taken together, these studies "show a significant positive effect of concentration on profits or margins" [1974, 202!.

(2.) As explained below, I use concentration ratios taken from the 1977 Census, as well as adjusted concentration ratios taken from an FTC report by Weiss and Pascoe [1986!.

(3.) For Example, see Peltzman [1977! and Gisser [1984!.

(4.) In Gisser [1986!, I standardized the profit variable by dividing it by value added. Here, I standardized the profit variable by dividing it by the value of shipments in order to render my study more comparable with Strickland and Weiss [1976!.

(5.) Strickland and Weiss [1976! concluded that advertising intensity reaches its peak when C is somewhere between 49 and 57 percent. In another study [1984!, I argued that the cutoff point between the low-concentration industries and the high-concentration industries is in the 60-70 percent range. Thus, the 60 percent cutoff point is arbitrary, but reasonable.

(6.) For unequal variances the approximate solution for the t-ratio (and degrees of freedom) is known as the Behrens-Fisher problem. It is described in Wine [1964, 264!.

(7.) Note that reclassified industries differ in important ways from the unreclassified ones. There is evidence in Telser [1987, ch. 8! that there is more innovation among the reclassified than the unreclassified industries. If so, it is likely that the concentration levels in the reclassified industries increase faster over time than those of the unreclassified industries. If industries producing heterogeneous products are relatively more innovative than the other industries, the comparison of the means of subsamples for the Census years 1963, 1972 and 1977 might be biased.

(8.) In early studies, since advertising cost was not substracted from the value added, the hypothesis of the product-differentiation barrier was accepted if the regression coefficient of Ad/S significantly exceeded one.

REFERENCES

Ayanian, Robert. "Advertising and Rates of Return." Journal of Law and Economics, October 1975, 479-506.

Bloch, HArry. "Advertising and Profitability: A Reappraisal." Journal of Political Economy, Part I, March-April 1974, 267-86.

Gisser, Micha. "Price Leadership and Dynamic Aspects of Oligopoly in U.S. Manufacturing." Journal of Political Economy, December 1984, 1035-48.

_____. "Price Leadership and Welfare Losses in U.S. Manufacturing." American Economic Review, september 1986, 756-67.

Peltzman, Sam. "Gains and Losses from Industrial Concentration." Journal of Law and Economics, October 1977, 229-63.

Solow, Robert M. "Technical Change and the Aggregate Production Function." Review of Economics and Statistics, August 1957, 312-20.

Stigler, George J., and Gary S. Becker. "De Gustibus Non Est Disputandum." American Economic Review, March 1977, 76-90.

Strickland, Allyn D., and Leonard W. Weiss. "Adverising, Concentration, and Price-Cost Margins." Journal of Political Economy, October 1976, 1109-21.

Telser, Lester G. A Theory of Efficient Cooperation and Competition. Cambridge: Cambridge University Press, 1987.

_____. "Advertising and Competition." Journal of Political Economy, December 1964, 537-62.

U.S. Department of Commerce, Bureau of the Census. Census of Manufactures, Washington, D.C.: U.S. Government Printing Office, 1977.

U.S. Department of Commerce, Bureau of Economic Analysis. The Detailed Input-Output Structure of the U.S. Economy, vol. 1. Washington, D.C.: U.S. Government Printing Office, 1977.

Weiss, Leonard W. "Advertising, Profits and Corporate Taxes." Review of Economics and Statistics, November 1969, 421-30.

_____. "The Concentration-Profits Relationship and Antitrust," in Industrial Concentration: The New Leaning, edited by H.J. Goldschmid, H.M. Mann and J.F. Weston. Boston: Little, Brown and Company, 1974, 184-233.

Weiss, Leonard W., and George A. Pascoe, Jr. Adjusted Concentration Ratios in Manufacturing, 1972 and 1977. Statistical Report of the Bureau of Economics and the Federal Trade Commission, Fiche, June 1986.

Wine, R. Lowell. Statistics for Scientists and Engineers. Englewood Cliffs, N.J.: Prentice-Hall, 1964.

Printer friendly Cite/link Email Feedback | |

Author: | Gisser, Micha |
---|---|

Publication: | Economic Inquiry |

Date: | Jan 1, 1991 |

Words: | 7007 |

Previous Article: | Firm participation in steel industry lobbying. |

Next Article: | Firm specific training, layoffs, and unreliable workers. |

Topics: |