Do Firms Differ Much?

THE JOURNAL OF INDUSTRIAL ECONOMICS Volume XXXIX March 1991

0022-1821 S2.00 No. 3

DO FIRMS DIFFER MUCH?* D E A N AMEL AND L U K E FROEB

With finn profitability data for a cross-section of geographic markets, it is possible to determine the relative importance of firm and market effects on profitability. Analysis of variance from a panel of multihank holditig companies in Texas suggests that finn effects are more important than market effects in determining profitability, and that the magnitudes of the effects vary over time.

I.

INTRODUCTION

IN A 1985 article, Schmalensee takes a new approach to analysis of the structure-conduct-performance paradigm. Schmalensee chooses a purely descriptive approach to distinguish among the "classical" tradition of Bain [1956], which takes markets as the basic unit of analysis, with a particular focus on market concentration; the "revisionist" views of Demsetz [1973] and Peltzman [1977], who view the observed relation between profitability and concentration as reflecting underlying inter-firm profitability differences; and the "managerial" view, that profitability is determined by firm-wide management practices, but not necessarily by differences in market share. This descriptive approach has two advantages over a structural analysis. A descriptive analysis avoids imposition of the troublesome maintained hypotheses inherent in structural models. Further, as Scott and Pascoe [1986] point out, the explanatory power of structural models is generally only a small fraction of that of descriptive models. Schmalensee uses FTC Line of Business data, where firm profitabihty data are broken down by line of business, to calculate the relative importance of firm and market effects on profitabihty. While a descriptive approach cannot shed any hght on structural parameters, it can yield useful insights into the structure-conduct-performance controversy. In particular, if firm effects are economically unimportant (as Schmalensee concludes), then managerial effects are not a substantive cause of differences in profitability. However, as pointed out by Shepherd [1972] and Harris [1988], the presence of significant firm effects does not contradict the classical hypothesis. The purpose of this paper is to investigate the robustness of Schmalensee's results with respect to a different data set, where firm profitability is broken * The authors thank Steve Bumbaugh for excellent research assistance. Tim Brennan, Timothy Hannan, Nellie Liang, Stephen Rhoades and anonymous referees made useful comments on earlier drafts. The views of the authors do not necessarily reflect the views of their respective institutions.

323

324

DEAN AMEL AND LUKE FROEB

down by geographic market rather than by line of business. Such a reexamination is appropriate because Schmalensee's results contradict findings of earlier studies and results derived by Scott and Pascoe from an expanded sample of FTC lines of business. The data set is a panel of multimarket banking firms in Texas over the period 1982-87 for which market-specific profit data are available. The use of geographic markets within a well-defined industry may yield better approximations to what most economists think of as markets than use of a cross-section of Standard Industrial Classification (SIC) industries.' We find that firm effects are more important than market effects in explaining variation in profitability. The explanatory power of the firm effects varies over the sample period as economic conditions in the industry change. In addition, market share and market concentration have little, if any, explanatory power. The results provide support for the hypothesis that firm-wide managerial practices are important determinants of profitability, but do not provide evidence supporting the hypotheses that market share or market concentration are important determinants of performance. n. BANKING FIRMS AND BANKING MARKETS

Banking markets provide more precise, consistent and economically meaningful measures of economic integration than are available for the industrial sector. The drawbacks of defining economic markets by SIC codes are well known (e.g. Werden [1988]). Studies using data aggregated by SIC code are vulnerable to criticism that the aggregation is either masking important relationships or creating spurious ones. In addition, FTC data are available for only four years and are now somewhat dated. Benston [1985] recommends a course for future research: "detailed company and industry studies, which require knowledge of the specifics and dynamics of events and institutions more than of econometric techniques are required" (p. 65). We follow his recommendation with a study of the Texas banking industry. By using cross-sectional data on local geographic markets within one industry, and by covering six years during which the industry faced rapidly changing economic conditions, this study attempts to avoid some of the shortcomings associated with the use of SIC data. Data from a number of consumer and business surveys suggest that the markets for at least some banking services are local in nature. While large businesses typically have access to regional or national markets for credit, smaller businesses are generally limited to local credit markets. In addition, the markets for many consumer banking products are restricted to local areas ' Studies using banking data generally find a significant positive relationship between profitability and market concentration similar to that found in auinufacturing industries. See Gilbert [1984] and Rhoades [1982] for surveys.

DO FIRMS DIFFER MUCH?

325

(King [1982], Whitehead [1982]). Federal bank regulators and the Department of Justice, in antitrust analyses of bank mergers and acquisitions, define banking markets as local geographic areas. Following common practice in cross-sectional studies, we estimate urban Texas geographic banking markets by Metropolitan Statistical Areas.^ The courts have traditionally defined the product market for banking to be the cluster of banking services.^ Such a definition limits the firms in the market to commercial banks. However, deregulatory legislation in 1980 and 1982 and recent technological developments have caused many to question this traditional product market definition. In particular, thrift institutions, which now have virtually the same legal powers as banks, are often thought to compete in the same markets as banks, and antitrust authorities typically give thrifts some weight in analyses of banking mergers. Nevertheless, the activities of banks and thrifts still differ substantially. Most thrifts have not become aggressive suppliers of products, such as commercial loans, that they were prohibited from supplying before 1980, and, compared to banks, thrift balance sheets show a larger focus on mortgages and real estate investments. Consequently, the sample is restricted to commercial banks. Banks are required to report their profits annually to bank regulators. Some states, including Texas, have restrictive branching laws that prohibit multimarket branching, so that each bank is limited to operating in a single market.* However, to circumvent branching restrictions, banking organizations form multibank holding companies (MBHCs) that own all of the stock of a number of banks. While each bank within a MBHC is legally a separate firm, we consider the MBHC to be the firm in this paper. One would expect all banks in an MBHC to be run so as to maximize their joint profi^ts because of their common ownership. Cornyn, et al. [1986] and Flannery [1986] review empirical evidence suggesting that market participants view the MBHC, not the individual banks, as the firm. Bank regulators and financial analysts treat MBHCs as firms for most purposes. In this paj>er, however, we utilize data on individual bank subsidiaries of MBHCs so that we can observe marketspecific perfonnance of MBHCs. The yearly samples were developed to yield data for bank subsidiaries of all MBHCs operating in two or more of the 27 urban banking markets in

^Primary Metropolitan Statistical Areas are used within Consolidated Metropolitan Statistical Areas. Thus, Dallas and Fort Worth are treated as separate banking markets, as are Houston, Galveston and Brazoria. ^ US V. Philadelphia National Bank, 374 US 321 [1963]; US v. Connecticut National Bank, 418

US 656 [1974]. *The Texas state constitutioti prohibited branching until 1987, when a constitutional amendment took effect allowing branching within a county. Since local markets are defined as Metropolitan Statistical Areas (MSAs), and MSAs are composed of counties, branching between markets is still prohibited.

326

DEAN AMEL AND LUKE FROEB TA B U - 1 DATA DESCRIPTION MEANS, (STANDARD DEVIATIONS), [MINIMAL], AND {MAXIMA}

Variable Profit rate

Concentration

Market share

No. ofobs. No. of firms No. of markets

1982

1983

1984

1.15 (0.44) [-0.49] {2.32} 1549 (738) [831] {5270} 11.6 (9.9) [0.1] {52.4} 107 18 26

0.54 (0.32) [-0.80] {1.95} 1500 (628) [814] {3880} 11.1 (9.9) [0.0] {51.9} 133 23 26

0.46 (0.43) [-2.74] {1.09} 1384 (554) [788] {3632} 10.9 (10.3) [0.0] (50.6} 144 26 26

1985

1986

1987

-1.40 -0.70 0.28 (2.03) (2.49) (2.96) [-15.20] [-21.31] [-16.95] {2.12} {1.63} {1.40} 1377 1371 1493 (553) (593) (586) [767] [734] [734] {3634} {3604} {3651} 10.6 10.4 10.7 (10.1) (9.9) (10.6) [0.0] [0.0] [0.0] {47.6} {47.8} {45.7} 153 157 156 30 31 33 27 27 27

Texas.' If an MBHC controls two or more bank subsidiaries within the same market, the deposit and profit data for those banks are consolidated into one observation for that market by summing bank balance sheets. Because of this within-market summing, in each year the number of observations on an MBHC is equal to the number of markets in which its subsidiary banks operate. The number of MBHCs in the sample rises from 18 in 1982 to 33 in 1987 while the number of observations on MBHC bank subsidiaries rises from 107 to 156. The increases are due primarily to mergers and acquisitions by one-market banking organizations that increase the number of multimarket MBHCs and to acquisitions of banks by existing MBHCs. The data are summarized in Table I for each ofthe six years in the sample.'' Profits are measured by the ratio of net income after taxes to assets.'' Use of the ratio of net income after taxes and securities gains and losses to assets as an altemative profit measure yielded very similar results, which are not reported. Market concentration, as measured by the Herfindahl Index, and * Texarkana hanks are excluded from the sample because part of the Texarkana market is in Arkansas. Texarkana banks face different branching restrictions and competitive conditions than hanks in other Texas metropolitan areas. * The number of markets covered rises from 26 over 1982-84 to 27 over 1985-87 because the Laredo market contains no banks owned by multimarket MBHCs prior to 1985. 'Intra-MBHC transfers could bias the profitability figures for individual banks within a MBHC. Howevffl:, this bias is controlled by regulations that require all svKh transfers to be at fair market values. Transfers among banks designed to aid failing affiliates are supposed to come through bank dividends paid to the parent holding company, then transferred to the failing bank subsidiary. Such dividends and transfers do not affect the profitability data of individual banks. Similarly, purchase premia paid for acquisitions of banks would come from dividends paid to the holding company parent and would not affect the profitability of banking subsidiaries.

DO FIRMS DIFFER MUCH?

327

market share are computed using the total assets of all banks operating in a market. in. ANALYSIS OF VARIANCE IN PROFITABILITY

Multimarket data are particularly suited to analyzing the relationship between market structure and performance. With such data, it is possible to decompose variance in profitability into two orthogonal parts: betweenmarket variation and within-market variation over firms. The classical approach predicts large between-market variation, and uses market concentration, along with other industry-level variables, as a proxy for the market effects, while the revisionist approach predicts large within-market variation, and uses market share as a proxy for firm effects within a market. The managerial view has implications for a different variance decomposition: between-firm variation and withiil-firm variation over markets. This view holds that management at the firm level is the most important

TABLE II ANALYSIS OF VARIANCE OF PROFITABILITY ADJUSTED i?-sQUARED MEASURES' (MARGINAL SIGNIFICANCE LEVELS IN PARENTHESES)

Effea

1982

1983

1984

1982-84"

1985

1986

1987

0.9 (0.41) 0.9 (0.14) 6.6 (0.12) 2.1* (0.05) 7.0 (0.21) 2.0 (0.34) 5.9 (0.14) 1.6 (0.12) 0.4 (0.45) 6.1 (0.21)

35.1** (0.00) 6.7** (0.00) 0.3 (0.02) 3.0* (0.02) 0.3 (0.10) 9.6** (0.00) 6.9** (0.00) 3.5** (0.00) 0.3 (0.16) 3.9** (0.00) 0.2* (0.01)

80.7** (0.00) 1.7 (0.06) 0.0 (0.67) 0.0 (0.71) 84.0** (0.00) 80.6** (0.00) 0.4 (0.44) 1.5 (0.12) 84.0** (0.00) 3.3** (0.01)

41.5** (0.00) 0.1 (0.27) 7.1 (0.09) 0.0 (0.55) 51.7** (0.00) 41.1** (0.00) 7.0 (0.10) 0.0 (0.55) 44.6** (0.00) 10.2** (0.00)

48.0** (0.00) 2.7* (0.02) 10.9* (0.02) 3.5* (0.01) 60.1** (0.00) 49.7** (0.00) U.O* (0.03) 4.0* (0.02) 49.2** (0.00) 12.1* (0.00)

YEAR FIRM SHARE MKT HERF FIRM + MKT FIRM+HERF SHARE-f-MKT SH ARE+HERF FIRM/MKr MKT/FIRM'

3.5 (0.26) 0.0 (0.94) 14.1* (0.04) 0.0 (0.83) 15.4 (0.08) 2.4 (0.32) 16.7* (0.02) 0.0 (0.98) 1.3 (0.39) 11.9 (0.10)

25.3** (0.00) 0.4 (0.23) 4.4 (0.22) 0.3 (0.45) 31.0** (0.00) 25.9** (0.00) 5.7 (0.17) 0.0 (0.45) 26.6** (0.00) 5.7 (0.15)

' Negative values of the adjusted R-squared measure are set equal to ZKTO. * denotes significance at the fi»e-percent level, and ** denotes significance at the one-perctmt level. The adjusted K-squared measures are calculated conditioned on the year efFects; i.e. the adjusted K-squared measures are the increase in adjusted S-sqiiared ahove the adjusted A-squared ofthe year effects. ° The conditional notation means that thefirmadjusted «-squared measures and the marginal significance levels are calculated conditional on the market effects, and vice versa for market and firm.

328

DEAN AMEL AND LUKE FROEB

determinant of profitability, independent of the markets in which a firm operates. The managerial approach predicts large between-firm variation and small within-firm variation across markets.^ Table II presents tests of the predictions of these three hypotheses. Each row of the table presents a fixed-effects model specification, and the columns give the samples over which each model is estimated. All ofthe samples cover single years except for 1982-84, which is the only period over which a Chow stability test is accepted, conditional on year effects. The entries in the table are adjusted R-squared measures, with the marginal significance level of the effects listed in parentheses below the measures. Adjusted /{-squared measures are used because we are comparing the explanatory power of models with different numbers of regressors. For the aggregated sample, 1982-84, table entries are the increase in adjusted R-squared over the adjusted R-squared of the year effects; likewise, the significance tests are performed conditional on the year effects. The variance decomposition suggested by the classical view can be seen by examining the MKT effects in the fourth row. The adjusted R-squared of the market effects (i.e. of a set of market dummy variables',, measures the amount of between-market variation in the sample. The residual is within-market variation. In only two ofthe seven years are market effects significant, in t982 and 1987, explaining 14 percent and 11 percent ofthe variation in profitability. For the aggregate 1982-84, market effects are also significant, explaining about 3 percent of the variation above that explained by year effects alone. The residual within-market variation is much larger than the betweenmarket variation, lending some support to the revisionist critique of the classical view.' However, support for the revisionist view is ambiguous. The market share variable does not pick up much of the residual within-market variation. This can be seen by examining the third row, denoted SHARE. The market share variable is insignificant in all but the year 1987, where it explains only 2.7 percent of the adjusted variation. The effects of market share can also be calculated conditional on the market effects. Looking at the two SHARE + MKT, it is possible to calculate the additional explanatory power of market share by comparing it to the MKTrow above. At most, in 1982, market share adds 2.6 percent to the explanatory power ofthe market effects. Likewise, market concentration is significant in only two ofthe seven years, ' Finn effects will capture any differences in profits due to differences in firm behavior. In particular, the effects of differences in portfolio composition, which reflect differences in managerial philosophy, will be captured by firm-specific dummy variables. ' In some ways this is a more stringent test of the classical hypotheis than interindustry studies using manufacturing data. Structural variables other than market concentration, such as entry barriers, do not vary as substantially across markets within an industry as they do in interindustry studies. However, as noted in the first footrsote, maritet concentration and profitability have been found to have a significant positive correlation in many studies of the banking industry.

DO FIRMS DIFFER MUCH?

329

1984 and 1987, and explains only 2.1 and 3.5 percent of the variation in these two years. Conditioning on firm effects, as calculated by comparing the FIRM + HERF row to the FIRM row, does not add to the explanatory power of concentration.^" Turning to the second variance decomposition, that suggested by the managerial view of the firm and given in the FIRM row of Table II, we see that firm effects (i.e. a set of firm dummy variables) are relatively large and significant in four of six years. " F i r m eifects are also significant for the period 1982-84. In the latter years of the sample, 1985, 1986 and 1987, the between-firm variation becomes very large. During 1986 and 1987, many Texas banks lost a great deal of money, as can be seen by examining the average profitability rates listed in Table I. Profit rates fell throughout the sample period, from a high in 1982 of 1.15 percent, to a low of —1.40 percent in 1987." In the last two rows of Table II, we attempt to measure the relative importance of firm and market effects. Since the models are not nested,'* statistical hypothesis tests cannot speak directly to this issue, although they can provide some evidence. The two rows, FIRM/MKT and MKT/FIRM, have entries calculated conditional on the market and firm effects. Specifically, the increase in adjusted i?-squared is measured and the hypothesis tests are calculated for the additional explanatory power of the effect over and above the explanatory power of the conditioned effect. These two rows clear up any ambiguity about the relative importance of firm and market effects that couid be caused by a correlation between the effects. Conditionally and unconditionally, firm effects are more important than market effects in 1983 and in the last three years of the sample. In 1982 and 1984, ihe market effects are larger than the firm effects, but not for the 1982-84 aggregate. Measuring the firm effects conditional on the market effects, and vice versa does not affect either the statistical significance or the size of the efFects. rV. CONCLUSIONS

We find support for the following four empirical propositions: (i) During the recession in the Texas banking industry, firm effects are large and significant. '" Results of the market share and market concentration variables are consistent with those of Scott and Pascoe [1986]. "Because of systematic differences in portfolio composition between banks and thrifts, the addition of thrifts to the sample would likely increase the importance of firm effects relative to market efFects and would not substantively alter the paper's conclusions. "These profit figures are simple averages of all observations in the sample, and are not weighted by assets. " T h e models are not nested because one consists of a set of firm dummy variables and the other of a set of market dummy variables.

330

DEAN AMEL AND LUKE FROEB

(ii) Prior to the recession, firm effects and market effects are much smaller, though statistically significant, (iii) Market share effects are very small, (iv) Market concentration effects are very small. In so far as they can be interpreted with respect to the classical, revisionist and managerial hypotheses, the results offer the most support to the managerial view. The observed increase in the variability of firm performance during times of stress is what one would expect if managerial ability is the primary determinant of firm performance. However, data for the earlier, more prosperous part of the sample period do not give as much support to the managerial hypothesis. The classical emphasis on the market as the unit of observation does not find any support in the data, but its alternative, the revisionist critique, finds only partial support. These results agree more closely with those of Scott and Pascoe than with those of Schmalensee. Further, the results demonstrate the degree to which macroeconomic conditions can affect the structure-performance relationship. Finally, the paper indicates that industry-specific panel data are a usable alternative to the somewhat-dated FTC Line of Business data. DEAN AMEL,

ACCEPTED MARCH 1990

Federal Reserve Board, Washington, DC, USA AND LUKE FROEB,

University of Chicago, and US Department of Justice, Washington, DC, USA.

REFERENCES

BAIN, J. S., 1956, Barriers to New Competition, Cambridge: Harvard University Press. BENSTON, G . J., 1985, 'The Validity of Profits-Structure with Particular Reference to the FTC's Line of Business Data', American Economic Review, 75, March, pp. 37-67. CoRNYN, A., HANWECK, G . , RHOADES, S. and ROSE, J., 1986, 'An Analysis of the

Concept of Corporate Separateness in BHC Regulation from an Economic Perspective', in Bank Structure and Competition, Federal Reserve Bank of Chicago, Chicago, pp. 174-212. DEMSETZ, H., 1973, 'Industry Structure, Market Rivalry, and Public Policy', Journal of Law and Economics, 16, April, pp. 1 -10. FLANNERY, M . J., 1986, 'Contagious Bank Runs, Financial Structure and Corporate Separateness within a Bank Holding Company', in Bank Structure and Competition, Federal Reserve Bank of Chicago, Chicago, pp. 213-230.

DO FIRMS DIFFER MUCH?

331

GILBERT, R. A., 1984, 'Bank Market Structure and Competition: A Survey', Journal of Money, Credit and Banking, 16, November, pp. 617-645. GuNTHER, J. W., 1989, Texas Banking Conditions: Managerial versus Economic Factors', Financial Industry Studies (Federal Reserve Bank of Dallas), October. HARRIS, F., 1988, 'Testable Competing Hypotheses from Structure-Performance Theory: Efficient Structure Versus Market Power', Joumal of Industrial Economics, 36, March, pp. 267-280. KING, B. F., 1982, 'Review of Empirical Literature', Economic Review (Federal Reserve Bank of Atlanta), April, pp. 35-40. PELTZMAN, S., 1977, 'The Gains and Losses from Industrial Concentration', Journal of Law and Economics, 20, October, pp. 229-263. RHOABES, S. A., 1982, 'Structure-Performance Studies in Banking: An Updated Summary and Evaluation', Staff Study 119 (Board of Governors of the Federal Reserve System), August. ScHMALENSEE, R.. 1985, 'Do Markets Diifer Much?' American Economic Review, 75, June, pp. 341-351. SCOTT, J. T. and PASCOE, G., 1986, 'Beyond Firm and Industry Effects on Profitability in Imperfect Markets', Review of Economics and Statistics, 68, May, pp. 284-292. SHEPHERD, W . G., 1972, 'The Elements of Market Structure', Review of Economics and Statistics, 54, February, pp. 25-37. WEISS, L. W., 1974, 'The Concentration-Profits Relationship and Antitrust', in H. J. GoLDSCHMiD et al, (eds.). Industrial Concentration and the New Learning, LittleBrown, Boston. WERDEN, G., 1988, 'The Divergence of SIC Industries from Antitrust Markets: Some Evidence from Price Fixing Cases', Economic Letters, 28, pp. 193-200. WHITEHEAD, D . D., 1982, 'The Sixth District Survey of Small Business Credit', Economic Review (Federal Reserve Bank of Atlanta), April, pp. 42-48.