R&D rivalry in the U.S. automobile industry: a simultaneous equation model appraoch to Bain's hypothesis.

Ramrattan, Lall B.

Background and Introduction

R&D efforts in the U.S. Auto industry are channeled into a variety of processes such as stamping, casting, machining, and assembling. Within the time-frame of our investigation, R&D efforts had to embrace sudden changes in taste toward small, fuel-efficient cars due to the OPEC crises [12] pp. 2-3 and other concerns. For instance, compliances with environmental regulation such as the Vehicle Air Pollution Control Act of 1965, the Air Quality Act of 1967, the Clear Air Amendment of 1970, and a few other environmental laws cost all industries about 15.4 percent loss in productivity annually between 1974 to 1977 [21].

The energy shocks of the 1970s precipitated a decline in U. S. Auto firms market share, which initiated reactions from several fronts. General Motors, Ford, and Chrysler invested more in production technologies that made greater use of robotics. They developed product lines that competed directly with foreign makes and models [23, 9] They have increased product performance and new design in all regional markets. [7] They also have changed their industrial relation practice, making layoffs a function of declining sales rather than technological changes, outsourcing, productivity, and reorganizations. [20] p. 687.

R&D reaction patterns were paradoxical during the time-frame we studied. As in the race between the hare and the tortoise, General Motors had a head start, channeling R&D expenditures into more fuel efficient cars approximately two years before the first OPEC cartel in 1973. Ford joined the race in 1976, and Chrysler in 1979. [9] p. 130. But in the 1980s, the winner appeared to be Chrysler Corporation.

This paper provides a test of Joe Bain's R&D rivalry hypothesis for firms in the U.S. Auto industry. Studying the structure-performance paradigm of the automobile industry, Joe Bain gave the lead rival role to sales promotion and the collusive role to price. [2, 3, 4, 26] R&D has a secondary albeit competitive role that is foiled with advertising. While investment in R&D has given a distinct appearance to autos, advertising informs customers of the new features. [3], p. 333. Inspire of advertising rivalry, firms still engage in protective imitation in brands, styles, accessories, and engines, [4] p. 298 which characterize R&D rivalry.

Scherer echoed Bain's concerted R&D rivalry view from the market sharing (product innovation via new processes) and the new market (new product) angles. For him, innovation increases demand through technical and style changes. [28] R&D expenditures may escalate as rivals compete to be the first to get a patent on a new product. Subject to diminishing returns, the expected completion date of a project can be shortened by increasing R&D expenditures. Scherer shared Bain's paradigm that advertising makes known the availability of new products, widens the market for innovators, and creates incentive for investment in innovation. [29] p. 378.

Bain's advertising rivalry and price collusion hypotheses were not falsified by some recent studies. [25, 26, 27] Using a single equation model specified by Grabowski and Baxter, [13] for Chemical firms, a recent investigation of R&D rivalry for the auto industry did not find any significant relationships. [25] However, that study used only the sample size of 1970-1982. We have repeated that test for the period 1970-1993, a larger sample, and still found insignificant results. The insignificant results of the single equation specification on two accounts propelled us to re-specify our model to take account for feedbacks among advertising, dividend, investment, finance and R&D in a system of equations environment.

We created a hybrid system model from several works to test Bain's R&D rivalry hypothesis. Dhrymes and Kurz [11, 10] were the pioneers of the first such model which they applied to a cross-section of industries. Switzer [31] and Guerard and Vaught [18] have developed different strands of that model, but limited their applications also to cross-section data. Ours is the first time-series industry application of Dhrymes and Kurz's model. We found that it failed to falsify Bain's lead role advertising hypothesis, bringing out the coordinated role of R&D rivalry among advertising rivalry, given the firm's divided policy, investment expenditures, and financial ratios.

Specification

We will adopt the single equation specification of Grabowski and Baxter [13] for R&D rivalry, and the single equation specification of Grabowski and Mueller [17, 16] for Advertising Rivalry. We will set these equations within a version of the Modigliani and Miller's [24] perfect market hypothesis where given the budget constraint facing the firm, management considers expenditure in R&D, advertising, investment, dividend disbursement, and financial ratios as interdependent. We will adopt the system of equations approach pioneered by Dhrymes and Kurz [11], henceforth referred to as D-K, primarily to accomplish that task. The perfect market hypotheses will stand out merely from the significant results of the system of equations vs. the poor performance of the single equations.

Grabowski and Mueller [16] corroborated the single equation advertising rivalry for the cigarette firms. A parallel finding was made by Ramrattan [25] for the automobile industry. Similarly, Grabowski and Baxter [13] have demonstrated for 8 firms in the U. S. chemical industry that rivals do react to the R&D outlay of other firms, given the state of their cashflow, the other significant determinant. They argued that firms do not match R&D expenditures as precisely as they do advertising, even though an earlier work of one of the authors [14] for three industries; viz., chemical, petroleum and drags, authenticates a one period lag response for profits as a significant determinant of R&D outlay. The complete system specification is as follows:

Ln[A.sub.i,t] = Ln[Alpha] + [Beta]Ln[A.sub.t-1,j] + [Gamma]Ln[P.sub.t-1, i] + [Lambda]Opec 1

[Mathematical Expression Omitted] 2

[Mathematical Expression Omitted] 3

[Mathematical Expression Omitted] 4

[Delta][R.sub.i,t] = [c.sub.1][Delta][D.sub.i,t] + [c.sub.2][Delta][R.sub.j,t-1] + [c.sub.3][Delta][P.sub.i,t] + [c.sub.4][Delta][V.sub.i,t] + [c.sub.5][Delta][R.sub.i,t-1] 5

[Inv.sub.i,t] = [F.sub.i,t] + N[I.sub.i,t] - [Div.sub.i,t] - [R.sub.i,t] - [A.sub.i,t] + [Dep.sub.i,t] 6

Where A = Advertising Expenditure.

P = Cashflow (NI + Depreciation).

NI = Net Income.

i,j = [i.sup.th] and [j.sup.th] firm.

Coefficients = [Alpha], [Beta], [Gamma], [Lambda], , a, b, c, g are estimates.

t = Time.

Opec = Dummy (zero for 1970-74, one otherwise).

[Delta] = Change.

[Delta][D.sub.i,t] = Cyclical Dummy: one when both sales and cashflow fall and zero otherwise.

Ln = [log.sub.e].

Inv = Investment in Plant and Equipment.

Div = Common Stock Dividend Disbursements.

S = Sales.

[S.sup.*] = Capacity Accelerator = ([Sales.sub.t] - [Sales.sub.t-3]) / [Sales.sub.t-3]

K = Total Invested Capital.

N = Short-term position: Excess of inventories, cash, short-term securities, and accounts receivables over accounts payable and other short term liabilities.

V = Firms outstanding debt and money value of common and preferred stock.

R = A firms R&D outlays.

F = External Bond Financing: First Difference of LTD.

LTD = Long Term Debt outstanding.

r = An appropriate interest rate to LTD.

Equation 5 is the Grabowski and Baxter [13] model without the constant term. The first difference form is fitted because of a demonstrated time trend in firm's R&D expenditures. [25, 13] Equation 1 is the advertising rivalry specification for Bain's hypothesis that was corroborated in two studies. [25, 27] The equations in between, viz., 2-4, are the system specifications of D-K. [10, 11] Equation 6 is an identity for the firm's budget constrain, which is an expansion of the D-K constraints to include advertising and R&D expenditures while staying in the perfect market hypothesis. Other specification and data source considerations are taken up systematically below.

A priori Specifications

We expect all the advertising estimates to be positive. An advertising coefficient close to one will indicate strong rivalry. Similarly, we expect the R&D coefficient to be positive. When a firm increases its R&D expenditures, it intends to be more inventive, which in turn causes innovation over time indicated by an upward shift in its production function. Other firms will not stand by idle and see their market share eroded away by their rivals becoming more efficient and perhaps more capital intensive over time. A reaction to neutralize, negate, or better an opponent's R&D outlays implies a positive coefficient. The introduction of a new. technology may introduce uncertainty in the firm's return, perhaps in the form of a random arrival time of the new invention. The consequence might be that firms will continue vigorous rivalry, which would indicate a positive coefficient also. [22] p. 288.

Regarding the divided and investment equations, we expect a positive coefficient when a source variable is regressed against a use variable, and a negative coefficient when regressed against another source. This is an expectation that pivots on the constraint facing the firms. [31] p. 164.

While a leadership-follower pattern was weakly demonstrated in an earlier advertising rivalry study [27], we expect even weaker results for that hypothesis in R&D rivalry because firms continually develop new models, thereby depriving each other of the first move advantage. [5] If firms follow a target return strategy, we would also expect to see disruption in a leader-follower relationship, Finally, small firms have been known to follow a go-it-alone tendency during crises periods in the industry, [25] showing their disinclination to follow a leader.

The above specification differs from D-K's [11], Switzer [31], and Guerard and Vaught [18] in two major respects. D-K have assumed that "technological and marketing constraints are exogenous to the system and predate the decision process we wish to study," [11] pp. 431-432. By explicitly specifying equations for advertising and R&D, we have removed those conditions. However, the R&D equation has only partially reduced the technological constraint to the extent that it drives invention, which is only one source of innovation [22]. Meanwhile, the literature has validated a residual approach to technological advance, not the difference equation form adopted here. [30] Finally, the common circumstances between D-K's and our results are limited by the phenomenon that our model is for one industry in a time series domain, while D-K's model is a cross-section study by industries.

Data Sources and Estimation Procedures

Our observation period is 1970-1993. The source of data for the advertising model is the same as Ramrattan's model [27] where, as Bain [4] observed, only "transaction type" data are available on advertising. The advertising data came from annual "100 Leading Advertisers" issues of the Advertising Age magazine. The other data were taken from generally available sources such as S&P Compustat Tapes, S&P 500 Stock Market Encyclopedia, Datext, Infotrac, Moody's Industrial Manuals, and firm's 10Ks and Annual reports.

D-K [11] and Switzer [31] got their best results using a 3-stage-least-square estimation technique, which extracts information from the over-identification of the equations, and inter-correlation of their errors. Our study underscores that finding over OLS, 2-stage-least-square, or even 3-stage-least-square with iteration.

The consistent set of instrumental variable that yielded significant R&D coefficients is the independent variable set exclusive of the predetermined rivals R&D outlays. The debate as whether or not to include a lagged dividend variable in the dividend equation does not benefit our results, corroborating the finding of Switzer [31] in that regard. Finally, the theme of the estimating procedure revolved around the Grabowski and Baxter [13] R&D rivalry specification using a one period change model without the constant term embedded in a system of equations model.

Results

The single equation results of Table 1 do not establish R&D rivalry among firms. The coefficients for [Delta][R.sub.j,t - 1] are insignificant. We also have significant coefficients for other independent variables beside the rival's previous period outlay of R&D. Overall, we get more significant estimates for equations without the constant term. Grabowski's and Baxter's [13] recommendation for dropping the constant term is feasible since only one constant term is significant. The results of Table 1 that incorporated a larger sample size than an earlier study [25] do not improve the single equation estimates. It beckons a system of equations model, which purports to improve the model's specification.

Tables 2-7 presents the system of equations results for firms taken two at a time. The 3-stage-least-square techniques recommended by Dhrymes and Kurz [11] yielded the best results over 2-stage-least-square, SUR, and OLS. We now gain substantial insight into the nature of R&D rivalry over the zero level information provided by its single equation counterpart. Three of the six R&D coefficients are now significant, indicating imitation between Ford vs. General Motors, General Motors vs. Ford, and General Motors vs. Chrysler. The cyclical dummy representing simultaneous decline in cash flow and sales performed poorly. The Ford vs. General Motors rivalry equation performed the best, with all independent variables save the dummy displaying significant coefficients. Cashflow did not come out as expected in the two significant General Motors responses, and the market valuation variable ranked the lowest in the General Motors vs. Chrysler equation.

The results posit R&D rivalry mainly between the two largest firms. General Motors and Ford corporations react to changes in each other R&D expenditures at a level of 0.96, and 0.23, respectively. The reaction of General Motors to Chrysler, the smallest firm, appears anomalous, given that the size of the coefficient is 2.29. One perhaps not so idle reconciliating factor is that Ford does not have to watch Chrysler if General Motors does. It has that degree of freedom, which is a gain if the information of Chrysler's R&D behavior is already incorporated into General Motors responses. A more active reconciliation is that General Motors watches Chrysler's R&D expenditures with jaundice eyes because its R&D reaction does not depend on the other independent variables whose coefficients are insignificant. Therefore, an inference for the R&D rivalry hypothesis in the U.S. auto industry is that only the big firms react to changes in their previous outlays. The leadership-follower hypothesis is dead in the R&D [TABULAR DATA FOR TABLE 1 OMITTED] area, and would be further killed in the advertising area below.

The Advertising Equation

The advertising equations, within the system of equations environment, indicate similar conclusion as found in a recent single equation study. [27] Comparing only significant coefficients, the system model reveals higher average reaction coefficients, viz., 0.85, 0.72, and 0.56 vs. 0.74, 0.63, and 0.5 of Ramrattan's [27] best results with multicolinearity correction for advertising, cashflow and the opecdummy, respectively. The higher coefficients are partly a consequence of a larger sample, and partly the account of mutual interactions of decision variables facing the firms.

The system results seem to be further killing a dead horse by showing lesser support for the leader-follower hypothesis in advertising rivalry. Where Ramrattan found that "Overall, the model with advertising, cashflow, and sales indicate that firms do not follow General Motors alone in setting their advertising policies" [27], this study finds more significant reactions of large firms to small firms outlay of advertising, even without substituting sales for cashflow. In fairness, the former study [27] included the now defunct American Motors, and this study does not take into account its merger with Chrysler. However, we have attempted to exorcise the ghost of American Motors by specifying a [TABULAR DATA FOR TABLE 2 OMITTED] [TABULAR DATA FOR TABLE 3 OMITTED] [TABULAR DATA FOR TABLE 4 OMITTED] [TABULAR DATA FOR TABLE 5 OMITTED] [TABULAR DATA FOR TABLE 6 OMITTED] [TABULAR DATA FOR TABLE 7 OMITTED] dummy variable equal to one from 1987 onwards and zero otherwise in the Chrysler rivalry relationships. The results were not improved for Chrysler rivalry relationships, further weakening the leader-followed hypothesis.

The results suggest that advertising performed well in both the single and system of equations, models, and the R&D behavior is significant only in the latter, giving credence to the idea that advertising sells innovations. The system model brings out the coordinated role of R&D, and the lead role hypothesis of advertising in Joe Bain's paradigm.

The Dividend, Investment and Finance Equations

The other equations in the system model are mostly consistent with D-K's [11] findings. Our model has six replications representing different rivalry situations; Dhrymes and Kurz [11] model, ten replications representing the years, 1951-1960. In the dividend equation, the per capital variables - profit and net short-run position, did not perform well. The profit variable was not significant, and while we got two of six significant net short-run position coefficients, they are of positive rather than negative sign. On the other hand, the per sales variables-investment and finance are mostly of the correct signs and performed about the same or better than in the D-K study in numbers of significant coefficients, i.e. 3 of 6 vs. 5 of 10 for investment, and 1 of 6 vs. 1 of 10 for finance, respectively. An attempt to improve the dividend equation by introducing a predetermined dividend variable yielded poorer results, confirming Dhrymes's and Kurz's admonition to Resek [11], p. 485, that multicollinearity exists between [(D/S).sub.t-1] and [(P/K).sub.t-1].

In the investment equation, the significant capacity accelerator coefficients were mostly negative. They were dominated by the profit version of investment model coefficients. In our model, the coefficients level of the [(P/K).sub.t-1] variable appears slightly more stable, ranging from -0.25 to 0.20, while in the D-K study, they ranged from -0.28 to 2.5. Also, 5 of 6 of our coefficients were significant vs. 4 of 10 in the D-K study. All the net short-term position coefficients is significant, and 2 of 6 finance coefficients are significant. The dividend variable performed poorly in this equation.

In the finance equation, all the coefficients save the investment per sales performed strongly compared to the D-K study. The number of significant coefficients were 2 of 6 vs. 2 of 10 for profits per capital, 2 of 6 vs. 4 of 10 for dividend per sales, 2 of 6 vs. 3 of 10 for risk, 4 of 6 vs. 3 of 10 for interest, and 5 of 6 vs. 3 of 10 for depreciation in ours and D-K study, respectively.

Conclusion

This paper gives a third home run to Bain's paradigm of a differentiated oligopolistic auto industry. Two of his theses regarding price collusion and advertising rivalry in the domestic auto industry were corroborated by several studies. [25, 26, 27] The failure of a standard R&D rivalry hypothesis via a single equation model of the Grabowski and Baxter [13] specification brought Bain's hypothesis to the front burner. To guard against naive falsificationism of the single equation approach, we first re-estimated the single equation specification with a larger sample. However, the results of Table 1 show that the larger sample also fails to turn up significant R&D rivalry within the Grabowski and Baxter research program.

Using simultaneous equations, this study corroborates Bain's R&D rivalry hypothesis in a joint relationship with advertising, dividend, investment and finance. We found reciprocal rivalry between the large firms, and a one-way reaction from General Motors to Chrysler for which we expressed some scepticism.

The mostly favorable significant estimates of the dividend, investment, and finance equations strengthened our rejection of the null-hypothesis that firms do not compete in R&D expenditure. The favorable results increase the robustness of the extended D-K specification by extending its empirical justification from a cross-section alone to a time-series rivalrous domain.

The validation of the hypothesis that advertising sells R&D has some implications for future studies. On the one hand, it justifies the search for an optimal level of advertising outlays of a firm, which should be done at the media and brands levels for the improved targeting of advertising. On the other hand, the coordination of R&D with other expenditures should be viewed from an appropriate social-welfare function that balances regulation between fuel economy, safety, comfort, environmental concern, and basic research in order to avoid adverse effects on sales volume and profits that constrains competition.

References

1. Abernathy, William J, The Productivity Dilemma [Baltimore: John Hopkins University Press, 1978].

2. Bain, Joe S., Industrial Organization [New York: John Wiley and Sons, Inc., 1959] Third printing, 1967].

3. Bain, Joe S., Price Theory [N.Y.: John Wiley and Sons, Inc., 1952]

4. Bain, Joe S., Barriers to New Competition [M.A.: Harvard University Press, 1956].

5. Brander, J. A., and Spencer, B. J., Strategic Commitment with R&D: The Symmetric Case, The Bell Journal of Economics, Vol. 14, No. 1, Spring 1983, p. 226.

6. Brown, Donaldson, Pricing Policy in Relation to Management Control, Management and Administration, February, 1924.

7. Clark, Kim B. et al., Product Development in the World Auto Industry, Brookings Papers on Economic Activity, Vol. 3, 1987, p. 730, and 732.

8. Cohen, Wesley M., and Klepper, Steven, The Anatomy of Industry R&D Intensity Distributions, AER, September, 1992.

9. Crandall, Roben W., et al., Regulating the Automobile [Washington: The Brookings Institution, 1986].

10. Dhrymes, Phoebus J., Econometrics (New York: Springer-Verlag, 1974].

11. Dhrymes, Phoebus J., Kurz, Mordecia, Investment, Dividend, and External Finance Behavior of Firms, in Robert Farber (ed.), Determinants of Investment Behavior [New York: National Bureau of Economic Research, 1967].

12. Friedlander, Ann F., Winston, Clifford, and Wang, Kung, Cost, technology, and productivity in the U.S. automobile industry, The Bell Journal of Economics Vol. 14, No. 1, Spring 1983.

13. Grabowski, H., and Baxter, N. D. Rivalry in Industrial Research and Development, The Journal of Industrial Economics, Vol. XXI, No. 3, 1973.

14. Grabowski, H., The Determinants of R and D in Three Industries, J. P. E., Vol. 76, 1968.

15. Grabowski, Henry G., and Mueller, Dennis C., Imitative Advertising in The Cigarette Industry, The Antitrust Bulletin. Vol. XVI (Summer 1971) pp. 257-292.

16. Grabowski, H., and Mueller, D. C., "Non-Price Competition in the Cigarette Industry," The Antitrust Bulletin, XIV [1969]. pp. 607-628.

17. Grabowski, H., and Mueller, D. C., "Non-Price Competition in the Cigarette Industry: A Comment," The Antitrust Bulletin, XIV [1970]. pp. 679-686.

18. Guerard, John, and Vaught, H. T., The Handbook of Financial Modeling [Chicago: Probus Publishing Company, 1989.]

19. Johnson, J., Econometric Methods [New York: McGraw-Hill Book Company, Second Edition, 1972]. See similar discussions in the Third Edition, 1984, pp. 450-451.

20. Katz, Harry C. et al., Industrial Relations and Productivity in the U.S. Automobile Industry, Brookings Papers on Economic Activity, Vol. 3, 1987, p. 687.

21. Link, Albert N., Productivity Growth, Environmental Regulation and the composition of R&D, The Bell Journal of Economics, Vol. 13, No. 2, Autumn 1982.

22. Lippman, Steven A., and McCardle, Kevin, Dropout behavior in R&D races with learning, Rand Journal of Economics, Vol. 18, No. 2, 1987.

23. Mannering, Fred, and Winston, Clifford, A Dynamic Empirical Analysis of Household Vehicle Ownership and Utilization, Rand Journal of Economics, Vol, 16, No. 2, Summer 1985.

24. Miller, M., and Modigliani, F., Dividend Policy, Growth, and the Valuation of Shares, Journal of Business, Vol. 34, 1961, pp. 411-433.

25. Ramrattan, Lall B., Rivalry in the U.S. Automobile Industry: An Advertising Approach, 1986. Unpublished Phd. Dissertation, The New School For Social Research.

26. Ramrattan, Lall B., The Impact of Imports on Prize Competition in the Automobile Industry, The American Economist, (Fall 1991).

27. Ramrattan, Lall B., Advertising Rivalry In The U.S. Automobile Industry: A Test of Bain's Hypothesis, The American Economist, (Fall 1994).

28. Scherer, F. M., Research and Development Resource Allocation Under Rivalry, The Quarterly Journal of Economics, Vol. LXXXI, No. 3, August 1967.

29. Scherer, F. M., Industrial Market Structure and Economic Performance, Second Edition [Chicago: Rand McNally College Publishing Company, 1980].

30. Solow, Robert M., Technical Change and the Aggregate Production Function, The Review of Economics and Statistics, Vol. 39, No. 3, August 1957, pp. 312-20.

31. Switzer, Lorne, The Determinants of Industrial R&D: A Funds Flow Simultaneous Equation Approach, The Review of Economics and Statistics, Vol. 66, 1984.

Lall B. Ramrattan, Economist at U. S. Dept. of HUD, Adjunct Associate Professor of Economics at Golden Gate University, and Lecturer at California State University, Hayward