# Imperfect competition and business cycles: an empirical investigation.

IMPERFECT COMPETITION AND BUSINESS CYCLES: AN EMPIRICAL INVESTIGATION

Does imperfect competition increase the magnitude of business cycles? If so, the

variability of an industry's employment and output should be positively related to the

market power of firms in that industry. This paper demonstrates that the opposite is

true: U.S. manufacturing industries with high price-cost margins display less

employment variability than do low-markup industries. These high-markup industries

display less price variability as well. Highly concentrated industries, however, do

display more employment variability. To some degree, markups may reflect labor

hoarding rather than market power; this may account for part, but not all, of the negative

correlation between markups and variability.

I. INTRODUCTION

Recent research has investigated whether deviations from perfect competition, either in goods or labor markets, help explain the existence and magnitude of business cycles. In particular, both theoretical and empirical evidence suggests that imperfectly competitive firms hold prices more rigid in the face of demand shocks than do more competitive firms.(1)

This finding raises the question of whether imperfectly competitive industries also exhibit larger fluctuations in output and employment. This paper examines employment and output data for 446 U.S. manufacturing industries over the years 1958 to 1984. If employment and output fluctuations have been smaller in more competitive industries, this would suggest that imperfect competition may indeed lead to larger business cycle fluctuations.

In fact, my results suggest the opposite: industries with larger price-marginal cost margins display less variability in employment and output (although the output results are not robust). The finding for employment is remarkably robust to a wide range of specifications; in particular, it holds for two different methods of estimating the price-marginal cost margin. In contrast, I find that highly concentrated industries--as opposed to those with high price-marginal cost margins--do display more variability in output and employment than do less concentrated industries. In addition, I confirm the finding of previous studies that industries with high price-cost margins display less price variability. Thus it appears that imperfect competition may reduce variability in both prices and quantities.

The paper proceeds as follows. Section II outlines the theoretical literature and surveys previous empirical work in this area. Section III describes the methodology of the paper, and section IV describes the estimation of the price-marginal cost margin. Section V presents the results, and section VI discusses possible explanations. Section VII concludes.

II. THEORETICAL LITERATURE AND PREVIOUS WORK

Many theoretical models suggest that imperfectly competitive industries ought to display more cyclical variability of output than perfectly competitive industries. The common thread in all these theories is real price rigidity.(2) Some reason is offered why an imperfectly competitive firm does not adjust its (relative) price in response to a demand shock as much as would a perfectly competitive firm. The less prices adjust, the more output (and employment) must adjust. Therefore, the argument goes, imperfect competition may exacerbate the size of business cycle fluctuations.(3) This basic argument dates back to Means [1935].

Prices may be relatively acyclical or even countercyclical in industries with market power for one of two reasons: either marginal cost is relatively flat, or markups are relatively countercyclical. The following theories of real price rigidity may be classified as falling into one of these two categories:

* Firms with market power are assumed

to maintain excess capacity, perhaps as

a barrier to entry by new firms. A by-product

of this excess capacity is that

these firm's marginal cost curves are

relatively flat (Hall [1986]). * The gains from cheating on a collusive

agreement between oligopolists may be

larger in a boom than in a recession

(Rotemberg and Saloner [1986]). This

implies that markups will be relatively

more countercyclical in concentrated

industries. * Oligopolists may act as though they

face "kinked" demand curves, which

would imply stable prices in response

to small shocks (Woglom [1982]). Thus

markups may be countercyclical in

these industries. * Firms are more likely to go bankrupt in

recessions. This may imply that

concentration and, therefore, market power

and price-cost margins vary

countercyclically.

Most empirical work to date has focused on the relationship between market structure and prices, and this literature has found substantial support for the above theories.(4) In this paper, I focus on quantities--output and employment--rather than prices, since quantities are of more direct interest. Further, prices are by no means a perfect indicator of what happens to quantities. While previous research has shown prices are relatively stable in industries with market power, this paper demonstrates that employment and output are also relatively stable in these industries.

Some studies have also examined the connection between market structure and employment variability, with results tending to provide weak evidence that more concentrated industries display more employment variability.(5) The primary difference between those studies and this paper is that they used concentration ratios as the only measure of market structure, whereas I estimate and use the margin of price over marginal cost, a more direct measure of market power. Further, this paper is broader in scope: it examines the variability of production as well as that of employment. Furthermore, it uses a much larger sample (446 four-digit manufacturing industries) over a longer period of time (1958 to 1984).

III. METHODOLOGY

I estimate a single cross-industry equation in which the industry's variability of output or employment (V) is explained by the industry's market power, as well as by other variables thought to be correlated with variability. The dependent variable measures the variability of either employment (the hours of production employees) or production (total industry output).(6) Variability is measured as the covariance of the series with real GNP. This covariance is constructed as in Lustgarten and Mendelowitz [1979], by taking the coefficient V from the equation(7) [Delta] log(series) = a + V [Delta] logGNP.

Two measures of market power are used. First, and most important, is the margin of price over marginal cost (PMCM), defined as (P-MC)/P. This measure is estimated for each industry twice, using two different techniques. The first, which will receive primary emphasis in this paper, was originally proposed by Hall [1988] and uses the procyclicality of total factor productivity to measure market power. The alternative is to estimate a translog cost function. The second measure of market power is an average of the four-firm concentration ratios for the years 1972 and 1977, as modified by Weiss and Pascoe [1986].(8) The price-marginal cost margin being the more direct measure of market power, is expected to be the more interesting of the two variables.

The equation also contains several control variables. Most important, it is necessary to control for the magnitude of shocks hitting each industry. It is well known that demand for durable consumer goods is highly procyclical; similarly, there may be systematic differences in the variability of producer goods from consumer goods. I therefore use dummy variables to separate durable goods from nondurables and producer goods from consumer goods.(9) If either of these dummies are correlated with market power, their omission would bias the results. Other control variables included in the cross-industry regression are the size of the industry (measured as the average real value added over time), the capital intensity of the industry (measured as the average value of capital stock divided by value added), and the growth rate of total industry output.

The data employed cover 446 four-digit U.S. manufacturing industries over the years 1958 to 1984. The primary sources of data are the quinquennial Census of Manufactures and the Annual Survey of Manufactures. The data are more fully described in the data appendix.

An econometric problem must be resolved before estimating the cross-industry equation. An estimated parameter (the price-marginal cost margin, estimated from a time-series regression for each industry) is among the independent variables. This introduces a measurement error problem because the true price-marginal cost margin for each industry is "observed" (estimated) with error. The cross-industry equation is given by equation (1), in which the covariance of industry output with GNP (V) is a function of the true regressors (X*): (1) V = X* [Beta] + u.

Unfortunately, though, we only observe the noisy estimate X = X* + [Delta]. Therefore, the equation actually estimated is (1a) V = X [Beta] + u - [Delta] [Beta].

Because of the "errors in variables" problem (X is correlated with [Delta]), ordinary least squares is inconsistent. A solution is to estimate this equation using instrumental variables. The instrument I use for the price-marginal cost margin is the (average over time of the) margin of price over average variable cost (PCM), calculated according to the formula:

PCM = (Value added -- payroll)

/ (Value added + materials costs). The variable PCM is of course primarily a measure of profit rates. It satisfies the necessary criteria for a valid instrument: as long as the equation is properly specified such that the true price-marginal cost margin is uncorrelated with the error term u, the variable PCM will also be uncorrelated with u. Further discussion of the properties of this instrument, as well as of other econometric issues, is available from the author in an econometric appendix.

IV. ESTIMATION OF THE PRICE-MARGINAL COST MARGIN

Estimation Using Procyclical Productivity

The first method relates the price-marginal cost margin to the degree of procyclical behavior in total factor productivity. The technique is based on Hall's [1988] insight that the procyclical residual of a "Solow" productivity equation (Solow [1957]) can be explained by the existence of market power (i.e., by a positive price-marginal cost margin), and need not imply procyclical productivity.(10)

The basic technique is to use time series data on each industry to observe changes in output and costs that are not associated with productivity changes. Marginal cost (MC) may be seen to equal: (2) MC = (w [Delta] N + r [Delta] K + [Tau] [Delta] M)

/([Delta] Q - [Theta] [prime] Q) = P(1 - [Mu]), where w, r, and [Tau] represent wages, the rental rate on capital, and materials prices; [Theta] [prime] represents (Hicks neutral) technical progress and is assumed to equal the sum of a constant term [Theta] and random component [Epsilon]; Q, N, K and M represent real output, labor hours, capital and materials; and [Mu] is the price-marginal cost margin, assumed to be constant.(11) Since the rental rate on capital is not directly observable, I assume constant returns to scale, which implies that the shares of labor, capital and materials in total output (valued at marginal cost, not at price) sum to unity. This allows the derivation of the equation to be estimated: (3) [Delta] log q - [[Alpha].sub.n] [Delta] log n - [[Alpha] sub.m] [Delta] logM

= [Mu] [Delta] log q + (1 - [Mu]) [Theta] + (1 - [Mu]) [Epsilon], where lowercase q, n and m represent variables as a fraction of K, and [[Alpha].sub.n] and [[Alpha].sub.m] represent the shares of labor and materials in output (valued at price). The left-hand side of equation (3), total factor productivity (or the "Solow residual"), is procyclical for most industries. As this equation makes clear, when price is greater than marginal cost (i.e., [Mu] > 0) total factor productivity will be procyclical even if the true productivity shock ([Epsilon]) is not.

I use current and lagged values of (the growth rate of) real defense spending and of the relative price of imported oil as instruments. It may be plausibly asserted that these variables are uncorrelated with the current productivity shock ([Epsilon]). That being so, the parameter estimates will be consistent even in the presence of aggregate productivity shocks, as is assumed, for example, in real business cycle theory.(12)

Estimation Using a Translog Cost Function

The second method used to calculate the price-marginal cost margin is the estimation of a translog cost function. The procedure follows Christensen and Greene [1976], and will be only briefly sketched here. The cost function is written as a function of output and factor prices as follows: (4) [Mathematical Expression Omitted] where the [w.sub.i] are input prices and the subscript i ranges over the three inputs: labor, capital, and materials. T is a time trend.

From Shephard's Lemma, the derivative of the cost function with respect to input prices gives factor demand equations. Therefore, the share of input i in total cost ([S.sub.i]) is given by (5) [Mathematical Expression Omitted]

Equation (4) is estimated together with two share equations (5) (those of labor and materials), imposing as restrictions all implied cross-equation restrictions and homogeneity of degree one of the cost function in input prices. Marginal cost is then obtained from differentiation of the cost function: (6) [Mathematical Expression Omitted] and is evaluated at industry averages of the given variables. The price-marginal cost margin is constructed using this marginal cost estimate.

Results of the PMCM Estimation

Table I presents the results of estimating the price-marginal cost margin, and compares them to the results obtained by Domowitz, Hubbard and Petersen [1988] (hereafter Domowitz et al.) and by Hall [1988]. Presented are two-digit averages of the price-marginal cost margin estimates for each four-digit industry (weighted by real value added). Columns a and b present estimates using the procyclical productivity and cost function techniques, respectively, while columns c and d present the estimates by Domowitz et al. and Hall, both of whom employ the procyclical productivity technique. My price-marginal cost margin estimates tend to be larger than those of Domowitz et al. and Hall. This difference appears to result from my estimating the price-marginal cost margin at the four-digit industry level, while Domowitz et al. estimate them using a panel of four-digit industries in two-digit subgroups, and Hall carries out the estimation directly on two-digit industries.

The correlation coefficients between the price-marginal cost margin estimates and the other measure of market power, the concentration ratio, are presented in Table II. The price-marginal cost margin estimates are positively correlated with concentration ratios, and although these correlations are not particularly large, they are strongly significant. The correlation between the two sets of price-marginal cost margins, while positive and strongly significant, is not as large as one might have hoped. However, if the sample is restricted to the 390 industries for which both methodologies give price-marginal cost estimates in the (0,1) range, this correlation increases from 0.411 to 0.546.

V. RESULTS

Three results will be highlighted. The first is that market power is inversely related to the variability of employment and output, contrary to the predictions of the theories discussed in section II. This is evident from the summary statistics presented in Table III. This table presents average values of the covariance of employment and output with GNP for subsamples of the industries divided into high and low price-marginal cost margin groups. This result is further supported by the direct correlation coefficient of the price-marginal cost margin with Cov(L,GNP) (-0.210) and with Cov (Q,GNP) (-0.137), both of which are significant at the 99 percent confidence level.

This result is documented more formally in the first two rows of Table IV. This table reports results from the instrumental variables estimation of equation (1a), where the price-marginal cost variable is estimated using the procyclical productivity technique.(13) Line 1 presents the employment variability results, and line 2 does the same for output. The negative and significant coefficient on PMCM in the Cov(L,GNP) equation indicates that low variability of employment tends to occur in industries that have a high degree of market power. The magnitude of this effect is fairly large: an increase in the price-marginal cost margin from 0.4 to 0.5 will decrease the covariance of employment with GNP from its median value of 1.9 to 1.6, a 15 percent decline. As shown in line 2, however, PMCM appears to be uncorrelated with Cov(Q,GNP). However, we will see that for many other specifications this correlation is negative, as it is for employment.

The second main result is that the concentration ratio is positively correlated with production and employment variability. This does lend some degree of support to the theories of section II, and is broadly consistent with the results of previous work. The coefficient is significant, however, only for Cov(L,GNP), and even in this case the magnitude of the effect is moderate: an increase in the concentration ratio from 0.4 to 0.5 would increase the median covariance by 5 percent, from 2.0 to 2.09. Since concentration ratios are often considered to be a measure of market power, as are price-cost margins, the differing correlations of these two variables with employment and output variability are puzzling.

Regarding the rest of the independent variables, the dummies behave as expected, with very large and strongly significant coefficients indicating more variability in the durable goods and producer goods industries. The size and capital intensity of the industry have insignificant effects on variability. The growth rate of output is not correlated with the variability of labor, but has a positive and nearly significant correlation with the variability of output.

To shed further light on the relationship between the empirical results presented here and those from the empirical literature on pricing (in particular, from Domowitz, Hubbard, and Petersen [1987]), line 3 of Table IV presents results examining the variability of prices. The dependent variable is the covariance of the output deflator with GNP. The conclusion of Domowitz and his coauthors was that highly concentrated industries display more price stability than do other industries. This result is confirmed here, as the coefficient on concentration ratio is negative and approaches significance. That is, prices are relatively countercyclical in concentrated industries, and this would lead one to predict that output and employment should be more variable in those industries.

The coefficient on PMCM is also negative and is strongly significant, indicating that industries with higher markups have prices that are more countercyclical. These price results suggest that high markups should be correlated with more variability of output and employment, but as we saw above, the opposite was the case. This apparent paradox is the third result of the paper.

As a check of robustness, Table V uses price-cost margin estimates from the cost function (eq. 6) in place of those from the procyclical productivity estimation. The results are virtually identical to those presented in Table IV. Given the great differences between the two methods of estimating the price-marginal cost margin, this is particularly strong evidence of the robustness of the basic results of this paper.(14)

I reported above that the price-marginal cost margin is negatively correlated with Cov(L,GNP) but is uncorrelated with Cov(Q,GNP). There is evidence that the output result may not be robust. In particular, two alternative specifications (OLS, and estimation using as the dependent variable the industry variance rather than the covariance with GNP) result in the price-marginal cost margin having a negative and significant coefficient in the output equation. Further, the output equation is heavily influenced by outliers: a negative coefficient on PMCM does occur in particular subsamples of industries, whereas virtually all other results of this paper tend to be robust to such changes. These points all give reason to believe that the variability of output as well as labor may be negatively correlated with price-marginal cost margins.

VI. EXPLANATIONS

As discussed above, there are three results to be explained. First, high price-marginal cost margins are associated with less variability of employment and (perhaps) output. Second, variability has differing correlations with the price-marginal cost margin and with concentration ratios. Finally, there is the apparent paradox that markups are correlated both with price stability and with quantity stability.

This section discusses four potential explanations for these results. For more detail on these and other explanations, see Lebow [1989].

Endogeneity of Market Power

One possible explanation for these results could come from the endogeneity of market structure and market power. It may be that firms in inherently stable industries are more likely to possess market power. The simplest such explanation, however, would be contrary to the results presented here: industries that are highly variable may require higher average profits to induce lending, which would suggest that variability leads to larger margins, not smaller. Moreover, this paper explains variability by average price-marginal cost margins and concentration ratios over time, so any possible argument based on the endogeneity of market power must also explain why this additional profitability is not eventually dissipated.

Magnitude of Shocks

A second logically possible explanation is that industries with market power systematically face smaller shocks than do more competitive industries. If true, this could easily explain why both prices and quantities appear to be more stable in the presence of market power. The use of dummy variables to separate durables from nondurables and producer goods from consumer goods was an admittedly imperfect attempt to address this issue.

There are two possible reasons to suggest that this phenomenon may occur. The first is the Galbraithian idea that firms with market power can "control" their environment, perhaps by using advertising to regulate demand. The second is the positive correlation between price and income elasticities of demand noted by Deaton and Muellbauer [1980], and presumably related to the fact that necessities tend to have few substitutes. Suppose that market power were determined mainly through industry demand elasticities (rather than, e.g., through the number of firms in different industries). Then this correlation would indeed suggest that industries with market power (selling goods with low price elasticities) face relatively stable demand as income fluctuates over the business cycle. I have no reason to believe that either of these two explanations is quantitatively important, however.

Inventories

A third possible explanation, which will be shown to have no empirical support, is that industries with market power are able to make better use of inventories. Blinder [1982] has demonstrated that storable goods ought to display more stability of both price and output than should non-storable goods. Thus, if it were indeed the case that firms with market power make better use of inventories than do more competitive firms, this could explain the first and third results of the paper. This explanation could not speak to the difference between results using price-cost margins and those using concentration ratios as measures of market power, however.

The preceding discussion suggests that equation (1) may be misspecified and should include among the explanatory variables a measure of inventory use (I), where I is positively correlated with X* (the true degree of market power) and is negatively correlated with the error term. This would impart a negative bias to the estimate of [Beta] and could, therefore, explain the negative coefficient obtained.

To examine the importance of inventories empirically, it is necessary to specify the variable I, and to include this variable in the cross-industry regression. This is done in Table VI, which replicates the basic equation including as explanatory variable one of three different measures of (finished goods) inventory use: the inventory-shipments ratio, the covariance of inventory levels with GNP, and the ratio of the covariance with GNP of output to that of shipments. In none of these regressions does inclusion of the inventory measure greatly alter the coefficient on PMCM. Further, the former two measures of inventory use are positively correlated with employment and output variability, which contradicts the argument that industries that make use of inventories have smoother output. This could reflect the well-known problems of the production smoothing model of inventory behavior, or it could simply be the case that industries with more variable demand have more incentive to use inventories. In any case, Table VI provides evidence that, however theoretically plausible, inventories are not an important part of the explanation for the results of this paper.

Labor Hoarding

A final possible explanation (based on work by Rotemberg and Summers [1990]) is that the price-marginal cost margin may be a poor measure of market power and may instead be primarily a measure of labor-hoarding behavior.(15) Since firms that hoard labor naturally will have more stable labor forces, this could explain the correlation between "market power" and employment variability. This argument depends critically on price stickiness as the reason why price remains above marginal cost.

Rotemberg and Summers do not suggest that the Hall markup estimates are biased; they question the interpretation that these estimates are indicative of market power. The basic Rotemberg-Summers model is presented in Figure 1. In this model, production requires capacity costs in the form of overhead labor, and marginal cost is constant up to capacity (Y). Overhead labor is directly proportional to capacity; therefore, the presence of overhead labor does not violate long-run constant returns to scale, and competitive equilibrium is possible. Demand fluctuates between a high level and a low level.

If prices instantaneously cleared the market, we would have a simple peak load pricing model, with equilibrium given by points A. When demand is low, price is equal to marginal cost; and when demand is high, firms produce at capacity and price is "whatever the market will bear." In this case, total factor productivity is not procyclical, because price is equal to marginal cost when production is below capacity, and quantities are fixed when production is at capacity.

Suppose, however, that prices are fixed at P* over the entire business cycle (referred to as "Keynesian" price rigidity). Then equilibrium at high and low demand is given by points B. Price is above marginal cost when demand is low, and consumers are rationed when demand is high. In this case, total factor productivity is procyclical because of the positive price-cost margin, even though firms have no market power. This is the basic Rotemberg-Summers result.(16)

The Rotemberg and Summers model offers a possible explanation for the results of this paper. Firms that practice a relatively large degree of labor hoarding will naturally have more stable employment, and this model suggests that they would give larger price-cost markup estimates as well. In the notation of section III, the true price-cost margin (X*) and the instrument are both negatively correlated with the error term (u) because they measure labor hoarding, an omitted variable. Further, this model says nothing about concentration ratios, so these may have any correlation at all with output variability. For example, it could be argued that concentration ratios measure market power, and their positive (if often insignificant) correlation with variability lends some support to the real rigidity theories presented in section II; while the price-cost margin estimates measure sticky prices and labor hoarding and have no bearing on these theories.

Do markups indicate market power or labor hoarding combined with sticky prices? Following Rotemberg and Summers, who argue that nonproduction workers tend to be hoarded in recessions, I construct a measure of labor hoarding for each industry as the ratio of nonproduction workers to total employment. Table VII presents results from a regression of both the estimated price-marginal cost margin (PMCM) and the price-average variable cost margin (PCM) against the nonproduction worker ratio (a measure of labor hoarding) and against the adjusted concentration ratio (a measure of market power). These results show a strong correlation of price-cost margins with the nonproduction worker ratio, and also show a positive and, for PCM, significant correlation with concentration ratios. Therefore, making the assumption that concentration ratios are a meaningful measure of market power, these results suggest that price-cost margins are indicative of both market power and labor hoarding.(17)

In order to examine the effect of labor hoarding on the variability-market power relationship, Table VIII again replicates the basic regression of Table IV, with the labor-hoarding measure, the ratio of nonproduction labor to total labor, included as a regressor. This nonproduction ratio enters into each equation with a negative and significant coefficient. That is, as expected, industries with a high ratio of nonproduction workers display more stability of output and employment. If the price-marginal cost margin were purely a measure of labor hoarding, however, we would expect that including the nonproduction ratio would eliminate all correlation of the price-cost margin with variability. In fact, in the employment equation the coefficient on PMCM declines by about one-third, from -3.47 to -2.35, and remains strongly significant. This suggests that about one-third of the negative correlation between price-cost margins and employment variability is due to labor hoarding, and the rest is due to market power.

VII. CONCLUSIONS

This paper examines whether the variability over the business cycle of output and employment in U.S. manufacturing industries is related to the degree of market power of firms in those industries, and presents a number of interesting results. First, contrary to the predictions of much theoretical literature, industries in which firms have higher price-marginal cost margins tend to display less variability of employment and (perhaps) output. The result for employment is robust to two different methods of estimating the price-marginal cost margin. In contrast, highly concentrated industries do display more variability of employment and output. The different correlations of price-marginal cost margins and concentration ratios with cyclical variability is an important and unexplained result of this paper. Finally, I confirm the finding of previous studies that industries with high price-cost margins display less price variability, which implies that these industries display less variability of both prices and quantities. It is possible that to some degree price-cost markups may reflect labor hoarding rather than market power; this explanation may account for approximately one-third of the negative correlation between markups and employment variability. A completely convincing explanation, however, remains elusive.

DATA APPENDIX

Most of the data are from a panel of 446 four-digit manufacturing industries (according to the 1982 classification) over the years 1958 to 1984 published by the Commerce department. The sources of these data are the quinquennial Census of Manufactures and the Annual Survey of Manufactures.

The panel includes the following series: total number of employees (excluding workers in central administrative offices); total payroll; number, hours and wages of production workers; value added; materials costs; real and nominal values of output and shipments; and net real capital stock. I constructed real materials costs using the producer price index on intermediates to manufacturing, from Citibase. This same deflator is used for all industries.

Real defense expenditures are from Citibase. Imported oil prices and the U.S. GNP deflator are from IMF's International Financial Statistics.

Because of industry reclassifications, it is important to check that the industries in the panel are defined consistently over time. In many cases, there is enough information in the published Census of Manufactures to see that the industries are defined consistently; in other cases this consistency is not certain. Further, the net real capital stock series was not available from the Commerce Department at the four-digit level in a consistent form (I obtained the data from Wayne Gray). However, the results in the paper hold when the sample is limited to those industries that were never redefined.

TABLE II

Correlations between Market Power Estimates

productivity

ratio Significance levels are in parentheses.

TABLE III

Summary Statistics: Average values of Cov(L,GNP) and Cov(Q,GNP)

TABLE VI

Inventories and the PMCM-Variability Correlation

(Instrumental Variables Estimation) Dependent

Notes: Industries with PMCM in (0,1) (399 Industries).

I is the level of real finished goods inventories.

S is real shipments.

Regressions also include concentration ratio, consumer/producer and durable/nondurable

dummies, industry size, capital intensity, and output growth rate.

T-statistics in parentheses.

TABLE VII

Explaining Price-Cost Margins

(Least-Squares Estimation)

Notes: Nonproduction ratio is nonproduction workers divided by total employment.

T-statistics in parentheses.

TABLE VIII

Labor Hoarding and the PMCM-Variability Correlation

(Instrumental Variables Estimation)

Notes: Industries with PMCM in (0.1) (399 Industries)

Regressions also include consumer/producer and durable/non-durable dummies,

industry size, capital intensity, and growth rate.

T-statistics in parentheses. [Tabular Data I, IV and V Omitted] [Figure 1 Omitted]

(1)See, among others, Hall [1986], Domowitz, Hubbard and Petersen [1987], Rotemberg and Saloner [1986], and Carlton [1989]. (2)Ball and Romer [1987] discuss the importance of real rigidities in theories of the business cycle. Clearly, a nominal shock can have real effects only if a nominal rigidity is present; the theories discussed in the text cannot themselves explain such nominal effects. Given some nominal rigidity, however, real rigidities can help to explain the size of economic fluctuations. (3)The connection between small price adjustments and large quantity adjustments in response to a given demand shock is subject to three caveats. First, the connection assumes that production occurs on the demand schedule; the possibility that this may not be true because of inventories is discussed in section VI. Second, the connection holds for industries initially producing at the same point on the demand curve. For industries producing at different points--as they would if one had more market power-results are more complicated; they depend on the elasticity of demand in the two relevant ranges of the demand curve and on the magnitude of the demand shock in these two ranges. But the connection does hold without qualification for some simple ways of specifying demand shocks; for example, it holds if shocks are defined as proportional shifts of an iso-elastic demand curve. Finally, because of "spillover" effects the connection may hold for specific industries but not for the economy as a whole. The fact that one industry responds to a negative demand shock largely through output rather than price need not imply that aggregate output falls, as this industry's behavior may induce other industries to increase output at the same time. Examples of such spillover effects include destabilizing price flexibility (De Long and Summers [1986]) and strategic complementarity (Cooper and John [1988]). The empirical framework of this paper measures only direct effects and ignores spillover effects. (4)See Qualls [1975], Domowitz, Hubbard and Petersen [1987]. (5)See Smith [19710, Lustgarten and Mendelowitz [1979], Scherer [1980], or Feinberg [1979]. (6)Results using other measures of employment (total number of employees, number of employees in production) and other measures of production (value added, total industry shipments) are similar to those presented in the text. (7)Results using an alternative measure of variability, the variance of the series, are similar to those presented in the text. The covariance measure is the more relevant, as the theories described above generally relate to demand shocks rather than supply shocks. Since demand shocks are more likely to be aggregate in nature than are supply shocks (see Lebow [1990] for evidence on productivity shocks), the covariance with GNP more nearly measures the industry response to demand shocks, whereas the variance is contaminated with industry-specific supply shocks. (8)The four-firm concentration ratio is the percetage of total industry sales made by the leading four firms. The modifications are intended to correct the simple concentration ratios for a number of problems, in order to make them more comparable across industries. These problems include geographically fragmented markets, non-competing subproducts, and levels of import competition. (9)The dummies were constructed as in Ornstein [1975], using the Commerce Department input-output tables for 1977. Goods were classified as consumer goods if more than 50 percent of output went to consumption (private or government) and as producer goods if more than 50 percent went to fixed investment or to intermediate use. Among producer goods, those in which output went primarily to fixed investment were classified as durables, and those in which output went primarily to intermediate use were classified as nondurables. Consumption goods were divided by inspection into durables and nondurables. (10)This paper follows the modification of Hall's methodology by Domowitz, Hubbard and Petersen [1988]. (11)Using a time-varying parameter model, Domowitz et al. [1988] find evidence that the price-marginal cost margin is not constant, but tends to be countercyclical. Because this is a cross-industry study, however, I only consider an average price-marginal cost margin for each industry. (12)These are the same instruments used by Hall [1988] and Domowitz et al. [1988]. Since the materials data include energy input, the assumption that oil prices do not affect productivity is equivalent to an assumption that factor prices do not shift production functions in the short run. Some have argued (e.g., Jorgenson [1988]) that energy prices do play a role in longer-term productivity movements. In any case, estimation of the price-marginal cost margin either by omission of oil prices from the instrument set, or by OLS estimation, do not affect the results of this paper. (13)This estimation omits those industries in which the price-cost margin estimate falls outside the range (0,1)--an estimate outside this range implies that price is less than marginal cost (or that marginal cost is negative). There are forty-seven industries in this category. An alternative approach, which gives very similar results, is to use the standard errors on the price-cost margin estimates to form an efficient weighting matrix, and then construct a weighted instrumental variables estimator. This approach is described in an econometric appendix available from the author. (14)Other robustness checks include estimation by ordinary least squares, and use of the industry variance, rather than the covariance with GNP, as the measure of variability. Most results are robust to these variations (but note the discussion below on the correlation of the price-cost margin with output variability). (15)The idea that labor hoarding can cause measured productivity to be procyclical obviously did not originate with Rotemberg and Summers. See, for example, Gordon [1979]. (16)In fact, Rotemberg and Summers argue that this result requires only a much milder form of price rigidity than suggested in this simple model. They present a model in which at any known level of aggregate demand, demand to the firm is uncertain, and even in a recession some states of nature will generate high enough demand that the firm can produce at capacity. Then as long as the firm must set its price before this state is realized (referred to as "micro" price rigidity), the firm will set price greater than marginal cost. (17)Using two-digit data, Rotemberg and Summers obtain a positive and significant coefficient on the concentration ratio. They argue that this provides support for the labor hoarding interpretation of markups.

REFERENCES

Ball, Laurence, and David Romer. "Real Rigidities and the non-Neutrality of Money." National Bureau of Economic Research Working Paper No. 2476, 1987. Bils, Mark. "The Cyclical Behavior of Marginal Cost and Price." American Economic Review, December 1987, 838-55. Blinder, Alan S. "Inventories and Sticky Prices: More on the Microfoundations of Macroeconomics." American Economic Review, June 1982, 334-48. Carlton, Dennis W. "The Theory and the Facts of How Markets Clear: Is Industrial Organization Valuable for Understanding Macroeconomics?" in Handbook of Industrial Organization, edited by R. Schmanlensee and R. Willig. New York: North Holland, 1989. Christensen, Laurits R., and William H. Greene. "Economies of Scale in U.S. Electric Power Generation." Journal of Political Economy, August 1976, 655-76. Cooper, Russel, and Andrew John. "Coordinating Coordination Failures in Keynesian Models." Quarterly Journal of Economics, August 1988, 441-63. Deaton, Angus S., and John Muellbauer. The Economics of Consumer Behavior. Cambridge: Cambridge University Press, 1980. De Long, J. Bradford, and Lawrence H. Summers. "Is Increased Price Flexibility Destabilizing?" American Economic Review, December 1986, 1031-44. Domowitz, Ian, R. Glenn Hubbard, and Bruce C. Petersen. "Market Structure, Durable Goods, and Cyclical Fluctuations in Markups." Mimeo, 1987. __. "Market Structure and Cyclical Fluctuations in U.S. Manufacturing." Review of Economics and Statistics, February 1988, 55-66. Feinberg, Robert M. "Market Structure and Employment Instability." Review of Economics and Statistics, November 1979, 497-505. Gordon, Robert J., "The |End-of-Expansion' Phenomenon in Short-Run Productivity Behavior." Brookings Papers on Economic Activity 2, 1979, 447-61. Hall, Robert E. "Market Structure and Macroeconomic Fluctuations." Brookings Papers on Economic Activity 2, 1986, 285-322. __. "The Relationship between Price and Marginal Cost in U.S. Industry." Journal of Political Economy, October 1988, 921-47. Jorgenson, Dale W. "Productivity and Postwar U.S. Economic Growth." Journal of Economic Perspectives, Fall 1988, 23-41. Lebow, David E. "Market Power, Productivity, and Business Cycles: Essays in Empirical Macroeconomics." Ph. D. Dissertation, Princeton University, 1989. __. "The Covariability of Productivity Shocks Across Industries." Economic Activity Working Paper No. 102. Washington: Board of Governors of the Federal Reserve System, 1990. Lustgarten, Steven, and Allan I. Mendelowitz. "The Covariability of Industrial Concentration and Employment Fluctuations." Journal of Business, April 1979, 291-305. Means, Gardiner. "Industrial Prices and their Relative Inflexibility." 74th Congress, 1st Session, Senate Document 13. Washington: Government Printing Office, January 1935. Ornstein, S. I. "Empirical Uses of the Price-Cost Margin." Journal of Industrial Economics, December 1975, 105-17. Qualls, David. "Price Stability in Concentrated Industries." Southern Economic Journal, October 1975, 294-8. Rotemberg, Julio J., and Garth Saloner. "A Supergame Theoretic Model of Business Cycles and Price Wars During Booms." American Economic Review, June 1986, 390-407. Rotemberg, Julio J., and Lawrence H. Summers. "Inflexible Prices and Procyclical Productivity." Quarterly Journal of Economics, November 1990, 851-74. Scherer, F. M. Industry Market Structure and Economic Performance. Chicago: Rand McNally, 1980. Smith, D.S. "Concentration and Employment Fluctuations." Western Economic Journal, September 1971, 267-77. Solow, Robert M. "Technical Change and the Aggregate Production Function." Review of Economics and Statistics, August 1957, 312-20. Weiss, Leonard W., and George A. Pascoe, Jr. "Adjusted Concentrated Ratios in Manufacturing, 1972 and 1977." Federal Trade Commission, June 1986. Woglom, Geoffrey. "Underemployment Equilibrium with Rational Expectations." Quarterly Journal of Economics, February 1982, 89-107.

DAVID E. LEBOW, Economist, Board of Governors of the Federal Reserve System. This is a revised version of chapters 1 and 2 of my doctoral dissertation. I would like to thank Larry Ball, Ben Bernanke, Alan Blinder, David Romer, David Wilcox, and two anonymous referees for helpful comments and discussions, and Wayne Gray for providing me with data. The opinions expressed here are my own and are not necessarily those of the Board of Governors of the Federal Reserve System or its staff.

Does imperfect competition increase the magnitude of business cycles? If so, the

variability of an industry's employment and output should be positively related to the

market power of firms in that industry. This paper demonstrates that the opposite is

true: U.S. manufacturing industries with high price-cost margins display less

employment variability than do low-markup industries. These high-markup industries

display less price variability as well. Highly concentrated industries, however, do

display more employment variability. To some degree, markups may reflect labor

hoarding rather than market power; this may account for part, but not all, of the negative

correlation between markups and variability.

I. INTRODUCTION

Recent research has investigated whether deviations from perfect competition, either in goods or labor markets, help explain the existence and magnitude of business cycles. In particular, both theoretical and empirical evidence suggests that imperfectly competitive firms hold prices more rigid in the face of demand shocks than do more competitive firms.(1)

This finding raises the question of whether imperfectly competitive industries also exhibit larger fluctuations in output and employment. This paper examines employment and output data for 446 U.S. manufacturing industries over the years 1958 to 1984. If employment and output fluctuations have been smaller in more competitive industries, this would suggest that imperfect competition may indeed lead to larger business cycle fluctuations.

In fact, my results suggest the opposite: industries with larger price-marginal cost margins display less variability in employment and output (although the output results are not robust). The finding for employment is remarkably robust to a wide range of specifications; in particular, it holds for two different methods of estimating the price-marginal cost margin. In contrast, I find that highly concentrated industries--as opposed to those with high price-marginal cost margins--do display more variability in output and employment than do less concentrated industries. In addition, I confirm the finding of previous studies that industries with high price-cost margins display less price variability. Thus it appears that imperfect competition may reduce variability in both prices and quantities.

The paper proceeds as follows. Section II outlines the theoretical literature and surveys previous empirical work in this area. Section III describes the methodology of the paper, and section IV describes the estimation of the price-marginal cost margin. Section V presents the results, and section VI discusses possible explanations. Section VII concludes.

II. THEORETICAL LITERATURE AND PREVIOUS WORK

Many theoretical models suggest that imperfectly competitive industries ought to display more cyclical variability of output than perfectly competitive industries. The common thread in all these theories is real price rigidity.(2) Some reason is offered why an imperfectly competitive firm does not adjust its (relative) price in response to a demand shock as much as would a perfectly competitive firm. The less prices adjust, the more output (and employment) must adjust. Therefore, the argument goes, imperfect competition may exacerbate the size of business cycle fluctuations.(3) This basic argument dates back to Means [1935].

Prices may be relatively acyclical or even countercyclical in industries with market power for one of two reasons: either marginal cost is relatively flat, or markups are relatively countercyclical. The following theories of real price rigidity may be classified as falling into one of these two categories:

* Firms with market power are assumed

to maintain excess capacity, perhaps as

a barrier to entry by new firms. A by-product

of this excess capacity is that

these firm's marginal cost curves are

relatively flat (Hall [1986]). * The gains from cheating on a collusive

agreement between oligopolists may be

larger in a boom than in a recession

(Rotemberg and Saloner [1986]). This

implies that markups will be relatively

more countercyclical in concentrated

industries. * Oligopolists may act as though they

face "kinked" demand curves, which

would imply stable prices in response

to small shocks (Woglom [1982]). Thus

markups may be countercyclical in

these industries. * Firms are more likely to go bankrupt in

recessions. This may imply that

concentration and, therefore, market power

and price-cost margins vary

countercyclically.

Most empirical work to date has focused on the relationship between market structure and prices, and this literature has found substantial support for the above theories.(4) In this paper, I focus on quantities--output and employment--rather than prices, since quantities are of more direct interest. Further, prices are by no means a perfect indicator of what happens to quantities. While previous research has shown prices are relatively stable in industries with market power, this paper demonstrates that employment and output are also relatively stable in these industries.

Some studies have also examined the connection between market structure and employment variability, with results tending to provide weak evidence that more concentrated industries display more employment variability.(5) The primary difference between those studies and this paper is that they used concentration ratios as the only measure of market structure, whereas I estimate and use the margin of price over marginal cost, a more direct measure of market power. Further, this paper is broader in scope: it examines the variability of production as well as that of employment. Furthermore, it uses a much larger sample (446 four-digit manufacturing industries) over a longer period of time (1958 to 1984).

III. METHODOLOGY

I estimate a single cross-industry equation in which the industry's variability of output or employment (V) is explained by the industry's market power, as well as by other variables thought to be correlated with variability. The dependent variable measures the variability of either employment (the hours of production employees) or production (total industry output).(6) Variability is measured as the covariance of the series with real GNP. This covariance is constructed as in Lustgarten and Mendelowitz [1979], by taking the coefficient V from the equation(7) [Delta] log(series) = a + V [Delta] logGNP.

Two measures of market power are used. First, and most important, is the margin of price over marginal cost (PMCM), defined as (P-MC)/P. This measure is estimated for each industry twice, using two different techniques. The first, which will receive primary emphasis in this paper, was originally proposed by Hall [1988] and uses the procyclicality of total factor productivity to measure market power. The alternative is to estimate a translog cost function. The second measure of market power is an average of the four-firm concentration ratios for the years 1972 and 1977, as modified by Weiss and Pascoe [1986].(8) The price-marginal cost margin being the more direct measure of market power, is expected to be the more interesting of the two variables.

The equation also contains several control variables. Most important, it is necessary to control for the magnitude of shocks hitting each industry. It is well known that demand for durable consumer goods is highly procyclical; similarly, there may be systematic differences in the variability of producer goods from consumer goods. I therefore use dummy variables to separate durable goods from nondurables and producer goods from consumer goods.(9) If either of these dummies are correlated with market power, their omission would bias the results. Other control variables included in the cross-industry regression are the size of the industry (measured as the average real value added over time), the capital intensity of the industry (measured as the average value of capital stock divided by value added), and the growth rate of total industry output.

The data employed cover 446 four-digit U.S. manufacturing industries over the years 1958 to 1984. The primary sources of data are the quinquennial Census of Manufactures and the Annual Survey of Manufactures. The data are more fully described in the data appendix.

An econometric problem must be resolved before estimating the cross-industry equation. An estimated parameter (the price-marginal cost margin, estimated from a time-series regression for each industry) is among the independent variables. This introduces a measurement error problem because the true price-marginal cost margin for each industry is "observed" (estimated) with error. The cross-industry equation is given by equation (1), in which the covariance of industry output with GNP (V) is a function of the true regressors (X*): (1) V = X* [Beta] + u.

Unfortunately, though, we only observe the noisy estimate X = X* + [Delta]. Therefore, the equation actually estimated is (1a) V = X [Beta] + u - [Delta] [Beta].

Because of the "errors in variables" problem (X is correlated with [Delta]), ordinary least squares is inconsistent. A solution is to estimate this equation using instrumental variables. The instrument I use for the price-marginal cost margin is the (average over time of the) margin of price over average variable cost (PCM), calculated according to the formula:

PCM = (Value added -- payroll)

/ (Value added + materials costs). The variable PCM is of course primarily a measure of profit rates. It satisfies the necessary criteria for a valid instrument: as long as the equation is properly specified such that the true price-marginal cost margin is uncorrelated with the error term u, the variable PCM will also be uncorrelated with u. Further discussion of the properties of this instrument, as well as of other econometric issues, is available from the author in an econometric appendix.

IV. ESTIMATION OF THE PRICE-MARGINAL COST MARGIN

Estimation Using Procyclical Productivity

The first method relates the price-marginal cost margin to the degree of procyclical behavior in total factor productivity. The technique is based on Hall's [1988] insight that the procyclical residual of a "Solow" productivity equation (Solow [1957]) can be explained by the existence of market power (i.e., by a positive price-marginal cost margin), and need not imply procyclical productivity.(10)

The basic technique is to use time series data on each industry to observe changes in output and costs that are not associated with productivity changes. Marginal cost (MC) may be seen to equal: (2) MC = (w [Delta] N + r [Delta] K + [Tau] [Delta] M)

/([Delta] Q - [Theta] [prime] Q) = P(1 - [Mu]), where w, r, and [Tau] represent wages, the rental rate on capital, and materials prices; [Theta] [prime] represents (Hicks neutral) technical progress and is assumed to equal the sum of a constant term [Theta] and random component [Epsilon]; Q, N, K and M represent real output, labor hours, capital and materials; and [Mu] is the price-marginal cost margin, assumed to be constant.(11) Since the rental rate on capital is not directly observable, I assume constant returns to scale, which implies that the shares of labor, capital and materials in total output (valued at marginal cost, not at price) sum to unity. This allows the derivation of the equation to be estimated: (3) [Delta] log q - [[Alpha].sub.n] [Delta] log n - [[Alpha] sub.m] [Delta] logM

= [Mu] [Delta] log q + (1 - [Mu]) [Theta] + (1 - [Mu]) [Epsilon], where lowercase q, n and m represent variables as a fraction of K, and [[Alpha].sub.n] and [[Alpha].sub.m] represent the shares of labor and materials in output (valued at price). The left-hand side of equation (3), total factor productivity (or the "Solow residual"), is procyclical for most industries. As this equation makes clear, when price is greater than marginal cost (i.e., [Mu] > 0) total factor productivity will be procyclical even if the true productivity shock ([Epsilon]) is not.

I use current and lagged values of (the growth rate of) real defense spending and of the relative price of imported oil as instruments. It may be plausibly asserted that these variables are uncorrelated with the current productivity shock ([Epsilon]). That being so, the parameter estimates will be consistent even in the presence of aggregate productivity shocks, as is assumed, for example, in real business cycle theory.(12)

Estimation Using a Translog Cost Function

The second method used to calculate the price-marginal cost margin is the estimation of a translog cost function. The procedure follows Christensen and Greene [1976], and will be only briefly sketched here. The cost function is written as a function of output and factor prices as follows: (4) [Mathematical Expression Omitted] where the [w.sub.i] are input prices and the subscript i ranges over the three inputs: labor, capital, and materials. T is a time trend.

From Shephard's Lemma, the derivative of the cost function with respect to input prices gives factor demand equations. Therefore, the share of input i in total cost ([S.sub.i]) is given by (5) [Mathematical Expression Omitted]

Equation (4) is estimated together with two share equations (5) (those of labor and materials), imposing as restrictions all implied cross-equation restrictions and homogeneity of degree one of the cost function in input prices. Marginal cost is then obtained from differentiation of the cost function: (6) [Mathematical Expression Omitted] and is evaluated at industry averages of the given variables. The price-marginal cost margin is constructed using this marginal cost estimate.

Results of the PMCM Estimation

Table I presents the results of estimating the price-marginal cost margin, and compares them to the results obtained by Domowitz, Hubbard and Petersen [1988] (hereafter Domowitz et al.) and by Hall [1988]. Presented are two-digit averages of the price-marginal cost margin estimates for each four-digit industry (weighted by real value added). Columns a and b present estimates using the procyclical productivity and cost function techniques, respectively, while columns c and d present the estimates by Domowitz et al. and Hall, both of whom employ the procyclical productivity technique. My price-marginal cost margin estimates tend to be larger than those of Domowitz et al. and Hall. This difference appears to result from my estimating the price-marginal cost margin at the four-digit industry level, while Domowitz et al. estimate them using a panel of four-digit industries in two-digit subgroups, and Hall carries out the estimation directly on two-digit industries.

The correlation coefficients between the price-marginal cost margin estimates and the other measure of market power, the concentration ratio, are presented in Table II. The price-marginal cost margin estimates are positively correlated with concentration ratios, and although these correlations are not particularly large, they are strongly significant. The correlation between the two sets of price-marginal cost margins, while positive and strongly significant, is not as large as one might have hoped. However, if the sample is restricted to the 390 industries for which both methodologies give price-marginal cost estimates in the (0,1) range, this correlation increases from 0.411 to 0.546.

V. RESULTS

Three results will be highlighted. The first is that market power is inversely related to the variability of employment and output, contrary to the predictions of the theories discussed in section II. This is evident from the summary statistics presented in Table III. This table presents average values of the covariance of employment and output with GNP for subsamples of the industries divided into high and low price-marginal cost margin groups. This result is further supported by the direct correlation coefficient of the price-marginal cost margin with Cov(L,GNP) (-0.210) and with Cov (Q,GNP) (-0.137), both of which are significant at the 99 percent confidence level.

This result is documented more formally in the first two rows of Table IV. This table reports results from the instrumental variables estimation of equation (1a), where the price-marginal cost variable is estimated using the procyclical productivity technique.(13) Line 1 presents the employment variability results, and line 2 does the same for output. The negative and significant coefficient on PMCM in the Cov(L,GNP) equation indicates that low variability of employment tends to occur in industries that have a high degree of market power. The magnitude of this effect is fairly large: an increase in the price-marginal cost margin from 0.4 to 0.5 will decrease the covariance of employment with GNP from its median value of 1.9 to 1.6, a 15 percent decline. As shown in line 2, however, PMCM appears to be uncorrelated with Cov(Q,GNP). However, we will see that for many other specifications this correlation is negative, as it is for employment.

The second main result is that the concentration ratio is positively correlated with production and employment variability. This does lend some degree of support to the theories of section II, and is broadly consistent with the results of previous work. The coefficient is significant, however, only for Cov(L,GNP), and even in this case the magnitude of the effect is moderate: an increase in the concentration ratio from 0.4 to 0.5 would increase the median covariance by 5 percent, from 2.0 to 2.09. Since concentration ratios are often considered to be a measure of market power, as are price-cost margins, the differing correlations of these two variables with employment and output variability are puzzling.

Regarding the rest of the independent variables, the dummies behave as expected, with very large and strongly significant coefficients indicating more variability in the durable goods and producer goods industries. The size and capital intensity of the industry have insignificant effects on variability. The growth rate of output is not correlated with the variability of labor, but has a positive and nearly significant correlation with the variability of output.

To shed further light on the relationship between the empirical results presented here and those from the empirical literature on pricing (in particular, from Domowitz, Hubbard, and Petersen [1987]), line 3 of Table IV presents results examining the variability of prices. The dependent variable is the covariance of the output deflator with GNP. The conclusion of Domowitz and his coauthors was that highly concentrated industries display more price stability than do other industries. This result is confirmed here, as the coefficient on concentration ratio is negative and approaches significance. That is, prices are relatively countercyclical in concentrated industries, and this would lead one to predict that output and employment should be more variable in those industries.

The coefficient on PMCM is also negative and is strongly significant, indicating that industries with higher markups have prices that are more countercyclical. These price results suggest that high markups should be correlated with more variability of output and employment, but as we saw above, the opposite was the case. This apparent paradox is the third result of the paper.

As a check of robustness, Table V uses price-cost margin estimates from the cost function (eq. 6) in place of those from the procyclical productivity estimation. The results are virtually identical to those presented in Table IV. Given the great differences between the two methods of estimating the price-marginal cost margin, this is particularly strong evidence of the robustness of the basic results of this paper.(14)

I reported above that the price-marginal cost margin is negatively correlated with Cov(L,GNP) but is uncorrelated with Cov(Q,GNP). There is evidence that the output result may not be robust. In particular, two alternative specifications (OLS, and estimation using as the dependent variable the industry variance rather than the covariance with GNP) result in the price-marginal cost margin having a negative and significant coefficient in the output equation. Further, the output equation is heavily influenced by outliers: a negative coefficient on PMCM does occur in particular subsamples of industries, whereas virtually all other results of this paper tend to be robust to such changes. These points all give reason to believe that the variability of output as well as labor may be negatively correlated with price-marginal cost margins.

VI. EXPLANATIONS

As discussed above, there are three results to be explained. First, high price-marginal cost margins are associated with less variability of employment and (perhaps) output. Second, variability has differing correlations with the price-marginal cost margin and with concentration ratios. Finally, there is the apparent paradox that markups are correlated both with price stability and with quantity stability.

This section discusses four potential explanations for these results. For more detail on these and other explanations, see Lebow [1989].

Endogeneity of Market Power

One possible explanation for these results could come from the endogeneity of market structure and market power. It may be that firms in inherently stable industries are more likely to possess market power. The simplest such explanation, however, would be contrary to the results presented here: industries that are highly variable may require higher average profits to induce lending, which would suggest that variability leads to larger margins, not smaller. Moreover, this paper explains variability by average price-marginal cost margins and concentration ratios over time, so any possible argument based on the endogeneity of market power must also explain why this additional profitability is not eventually dissipated.

Magnitude of Shocks

A second logically possible explanation is that industries with market power systematically face smaller shocks than do more competitive industries. If true, this could easily explain why both prices and quantities appear to be more stable in the presence of market power. The use of dummy variables to separate durables from nondurables and producer goods from consumer goods was an admittedly imperfect attempt to address this issue.

There are two possible reasons to suggest that this phenomenon may occur. The first is the Galbraithian idea that firms with market power can "control" their environment, perhaps by using advertising to regulate demand. The second is the positive correlation between price and income elasticities of demand noted by Deaton and Muellbauer [1980], and presumably related to the fact that necessities tend to have few substitutes. Suppose that market power were determined mainly through industry demand elasticities (rather than, e.g., through the number of firms in different industries). Then this correlation would indeed suggest that industries with market power (selling goods with low price elasticities) face relatively stable demand as income fluctuates over the business cycle. I have no reason to believe that either of these two explanations is quantitatively important, however.

Inventories

A third possible explanation, which will be shown to have no empirical support, is that industries with market power are able to make better use of inventories. Blinder [1982] has demonstrated that storable goods ought to display more stability of both price and output than should non-storable goods. Thus, if it were indeed the case that firms with market power make better use of inventories than do more competitive firms, this could explain the first and third results of the paper. This explanation could not speak to the difference between results using price-cost margins and those using concentration ratios as measures of market power, however.

The preceding discussion suggests that equation (1) may be misspecified and should include among the explanatory variables a measure of inventory use (I), where I is positively correlated with X* (the true degree of market power) and is negatively correlated with the error term. This would impart a negative bias to the estimate of [Beta] and could, therefore, explain the negative coefficient obtained.

To examine the importance of inventories empirically, it is necessary to specify the variable I, and to include this variable in the cross-industry regression. This is done in Table VI, which replicates the basic equation including as explanatory variable one of three different measures of (finished goods) inventory use: the inventory-shipments ratio, the covariance of inventory levels with GNP, and the ratio of the covariance with GNP of output to that of shipments. In none of these regressions does inclusion of the inventory measure greatly alter the coefficient on PMCM. Further, the former two measures of inventory use are positively correlated with employment and output variability, which contradicts the argument that industries that make use of inventories have smoother output. This could reflect the well-known problems of the production smoothing model of inventory behavior, or it could simply be the case that industries with more variable demand have more incentive to use inventories. In any case, Table VI provides evidence that, however theoretically plausible, inventories are not an important part of the explanation for the results of this paper.

Labor Hoarding

A final possible explanation (based on work by Rotemberg and Summers [1990]) is that the price-marginal cost margin may be a poor measure of market power and may instead be primarily a measure of labor-hoarding behavior.(15) Since firms that hoard labor naturally will have more stable labor forces, this could explain the correlation between "market power" and employment variability. This argument depends critically on price stickiness as the reason why price remains above marginal cost.

Rotemberg and Summers do not suggest that the Hall markup estimates are biased; they question the interpretation that these estimates are indicative of market power. The basic Rotemberg-Summers model is presented in Figure 1. In this model, production requires capacity costs in the form of overhead labor, and marginal cost is constant up to capacity (Y). Overhead labor is directly proportional to capacity; therefore, the presence of overhead labor does not violate long-run constant returns to scale, and competitive equilibrium is possible. Demand fluctuates between a high level and a low level.

If prices instantaneously cleared the market, we would have a simple peak load pricing model, with equilibrium given by points A. When demand is low, price is equal to marginal cost; and when demand is high, firms produce at capacity and price is "whatever the market will bear." In this case, total factor productivity is not procyclical, because price is equal to marginal cost when production is below capacity, and quantities are fixed when production is at capacity.

Suppose, however, that prices are fixed at P* over the entire business cycle (referred to as "Keynesian" price rigidity). Then equilibrium at high and low demand is given by points B. Price is above marginal cost when demand is low, and consumers are rationed when demand is high. In this case, total factor productivity is procyclical because of the positive price-cost margin, even though firms have no market power. This is the basic Rotemberg-Summers result.(16)

The Rotemberg and Summers model offers a possible explanation for the results of this paper. Firms that practice a relatively large degree of labor hoarding will naturally have more stable employment, and this model suggests that they would give larger price-cost markup estimates as well. In the notation of section III, the true price-cost margin (X*) and the instrument are both negatively correlated with the error term (u) because they measure labor hoarding, an omitted variable. Further, this model says nothing about concentration ratios, so these may have any correlation at all with output variability. For example, it could be argued that concentration ratios measure market power, and their positive (if often insignificant) correlation with variability lends some support to the real rigidity theories presented in section II; while the price-cost margin estimates measure sticky prices and labor hoarding and have no bearing on these theories.

Do markups indicate market power or labor hoarding combined with sticky prices? Following Rotemberg and Summers, who argue that nonproduction workers tend to be hoarded in recessions, I construct a measure of labor hoarding for each industry as the ratio of nonproduction workers to total employment. Table VII presents results from a regression of both the estimated price-marginal cost margin (PMCM) and the price-average variable cost margin (PCM) against the nonproduction worker ratio (a measure of labor hoarding) and against the adjusted concentration ratio (a measure of market power). These results show a strong correlation of price-cost margins with the nonproduction worker ratio, and also show a positive and, for PCM, significant correlation with concentration ratios. Therefore, making the assumption that concentration ratios are a meaningful measure of market power, these results suggest that price-cost margins are indicative of both market power and labor hoarding.(17)

In order to examine the effect of labor hoarding on the variability-market power relationship, Table VIII again replicates the basic regression of Table IV, with the labor-hoarding measure, the ratio of nonproduction labor to total labor, included as a regressor. This nonproduction ratio enters into each equation with a negative and significant coefficient. That is, as expected, industries with a high ratio of nonproduction workers display more stability of output and employment. If the price-marginal cost margin were purely a measure of labor hoarding, however, we would expect that including the nonproduction ratio would eliminate all correlation of the price-cost margin with variability. In fact, in the employment equation the coefficient on PMCM declines by about one-third, from -3.47 to -2.35, and remains strongly significant. This suggests that about one-third of the negative correlation between price-cost margins and employment variability is due to labor hoarding, and the rest is due to market power.

VII. CONCLUSIONS

This paper examines whether the variability over the business cycle of output and employment in U.S. manufacturing industries is related to the degree of market power of firms in those industries, and presents a number of interesting results. First, contrary to the predictions of much theoretical literature, industries in which firms have higher price-marginal cost margins tend to display less variability of employment and (perhaps) output. The result for employment is robust to two different methods of estimating the price-marginal cost margin. In contrast, highly concentrated industries do display more variability of employment and output. The different correlations of price-marginal cost margins and concentration ratios with cyclical variability is an important and unexplained result of this paper. Finally, I confirm the finding of previous studies that industries with high price-cost margins display less price variability, which implies that these industries display less variability of both prices and quantities. It is possible that to some degree price-cost markups may reflect labor hoarding rather than market power; this explanation may account for approximately one-third of the negative correlation between markups and employment variability. A completely convincing explanation, however, remains elusive.

DATA APPENDIX

Most of the data are from a panel of 446 four-digit manufacturing industries (according to the 1982 classification) over the years 1958 to 1984 published by the Commerce department. The sources of these data are the quinquennial Census of Manufactures and the Annual Survey of Manufactures.

The panel includes the following series: total number of employees (excluding workers in central administrative offices); total payroll; number, hours and wages of production workers; value added; materials costs; real and nominal values of output and shipments; and net real capital stock. I constructed real materials costs using the producer price index on intermediates to manufacturing, from Citibase. This same deflator is used for all industries.

Real defense expenditures are from Citibase. Imported oil prices and the U.S. GNP deflator are from IMF's International Financial Statistics.

Because of industry reclassifications, it is important to check that the industries in the panel are defined consistently over time. In many cases, there is enough information in the published Census of Manufactures to see that the industries are defined consistently; in other cases this consistency is not certain. Further, the net real capital stock series was not available from the Commerce Department at the four-digit level in a consistent form (I obtained the data from Wayne Gray). However, the results in the paper hold when the sample is limited to those industries that were never redefined.

TABLE II

Correlations between Market Power Estimates

PMCM: PMCM: Procyclical Cost Concentration productivity function ratio PMCM:Procyclical 1 .411 (.000) .132 (.005)

productivity

PMCM: Cost function 1 1.48 (.002) Concentration 1

ratio Significance levels are in parentheses.

TABLE III

Summary Statistics: Average values of Cov(L,GNP) and Cov(Q,GNP)

Cov(L,GNP) Cov(Q,GNP) Low PMCM 2.25 2.16 High PMCM 1.58 1.82

TABLE VI

Inventories and the PMCM-Variability Correlation

(Instrumental Variables Estimation) Dependent

Variable PMCM Inventory Measure Cov(L,GNP) -3.47 (-3.7) -3.59 (-3.8) 2.42 (1.2) I/S (avg.) -3.32 (-3.5) .074 (2.1) Cov(I,GNP) -3.42 (-3.7) -.047 (-1.3) Cov(Q,GNP)/Cov(S,GNP) Cov(Q,GNP) -0.50 (-0.5) -0.65 (-0.6) 2.84 (1.3) I/S (avg.) -0.15 (-0.1) .176 (4.5) Cov(I,GNP) -0.48 (-0.5) -.017 (-0.4) Cov(Q,GNP)/Cov (S,GNP)

Notes: Industries with PMCM in (0,1) (399 Industries).

I is the level of real finished goods inventories.

S is real shipments.

Regressions also include concentration ratio, consumer/producer and durable/nondurable

dummies, industry size, capital intensity, and output growth rate.

T-statistics in parentheses.

TABLE VII

Explaining Price-Cost Margins

(Least-Squares Estimation)

Dependent Nonproduction Concentration Variable Ratio Ratio R-square PMCM .801 (5.2) .131 (1.5) .074 PCM .215 (6.9) .102 (5.7) .193

Notes: Nonproduction ratio is nonproduction workers divided by total employment.

T-statistics in parentheses.

TABLE VIII

Labor Hoarding and the PMCM-Variability Correlation

(Instrumental Variables Estimation)

Dependent Concentration Nonproduction Variable PMCM Ratio Ratio R-square Cov(L,GNP) -3.47 0.90 .16 (3.7) (2.3) -2.35 1.09 -3.03 .22 (-2.1) (3.0) (-3.0) Cov(Q,GNP) -0.50 0.48 .20 (-0.5) (1.1) 0.60 0.67 -2.97 .21 (0.5) (1.6) (-2.5)

Notes: Industries with PMCM in (0.1) (399 Industries)

Regressions also include consumer/producer and durable/non-durable dummies,

industry size, capital intensity, and growth rate.

T-statistics in parentheses. [Tabular Data I, IV and V Omitted] [Figure 1 Omitted]

(1)See, among others, Hall [1986], Domowitz, Hubbard and Petersen [1987], Rotemberg and Saloner [1986], and Carlton [1989]. (2)Ball and Romer [1987] discuss the importance of real rigidities in theories of the business cycle. Clearly, a nominal shock can have real effects only if a nominal rigidity is present; the theories discussed in the text cannot themselves explain such nominal effects. Given some nominal rigidity, however, real rigidities can help to explain the size of economic fluctuations. (3)The connection between small price adjustments and large quantity adjustments in response to a given demand shock is subject to three caveats. First, the connection assumes that production occurs on the demand schedule; the possibility that this may not be true because of inventories is discussed in section VI. Second, the connection holds for industries initially producing at the same point on the demand curve. For industries producing at different points--as they would if one had more market power-results are more complicated; they depend on the elasticity of demand in the two relevant ranges of the demand curve and on the magnitude of the demand shock in these two ranges. But the connection does hold without qualification for some simple ways of specifying demand shocks; for example, it holds if shocks are defined as proportional shifts of an iso-elastic demand curve. Finally, because of "spillover" effects the connection may hold for specific industries but not for the economy as a whole. The fact that one industry responds to a negative demand shock largely through output rather than price need not imply that aggregate output falls, as this industry's behavior may induce other industries to increase output at the same time. Examples of such spillover effects include destabilizing price flexibility (De Long and Summers [1986]) and strategic complementarity (Cooper and John [1988]). The empirical framework of this paper measures only direct effects and ignores spillover effects. (4)See Qualls [1975], Domowitz, Hubbard and Petersen [1987]. (5)See Smith [19710, Lustgarten and Mendelowitz [1979], Scherer [1980], or Feinberg [1979]. (6)Results using other measures of employment (total number of employees, number of employees in production) and other measures of production (value added, total industry shipments) are similar to those presented in the text. (7)Results using an alternative measure of variability, the variance of the series, are similar to those presented in the text. The covariance measure is the more relevant, as the theories described above generally relate to demand shocks rather than supply shocks. Since demand shocks are more likely to be aggregate in nature than are supply shocks (see Lebow [1990] for evidence on productivity shocks), the covariance with GNP more nearly measures the industry response to demand shocks, whereas the variance is contaminated with industry-specific supply shocks. (8)The four-firm concentration ratio is the percetage of total industry sales made by the leading four firms. The modifications are intended to correct the simple concentration ratios for a number of problems, in order to make them more comparable across industries. These problems include geographically fragmented markets, non-competing subproducts, and levels of import competition. (9)The dummies were constructed as in Ornstein [1975], using the Commerce Department input-output tables for 1977. Goods were classified as consumer goods if more than 50 percent of output went to consumption (private or government) and as producer goods if more than 50 percent went to fixed investment or to intermediate use. Among producer goods, those in which output went primarily to fixed investment were classified as durables, and those in which output went primarily to intermediate use were classified as nondurables. Consumption goods were divided by inspection into durables and nondurables. (10)This paper follows the modification of Hall's methodology by Domowitz, Hubbard and Petersen [1988]. (11)Using a time-varying parameter model, Domowitz et al. [1988] find evidence that the price-marginal cost margin is not constant, but tends to be countercyclical. Because this is a cross-industry study, however, I only consider an average price-marginal cost margin for each industry. (12)These are the same instruments used by Hall [1988] and Domowitz et al. [1988]. Since the materials data include energy input, the assumption that oil prices do not affect productivity is equivalent to an assumption that factor prices do not shift production functions in the short run. Some have argued (e.g., Jorgenson [1988]) that energy prices do play a role in longer-term productivity movements. In any case, estimation of the price-marginal cost margin either by omission of oil prices from the instrument set, or by OLS estimation, do not affect the results of this paper. (13)This estimation omits those industries in which the price-cost margin estimate falls outside the range (0,1)--an estimate outside this range implies that price is less than marginal cost (or that marginal cost is negative). There are forty-seven industries in this category. An alternative approach, which gives very similar results, is to use the standard errors on the price-cost margin estimates to form an efficient weighting matrix, and then construct a weighted instrumental variables estimator. This approach is described in an econometric appendix available from the author. (14)Other robustness checks include estimation by ordinary least squares, and use of the industry variance, rather than the covariance with GNP, as the measure of variability. Most results are robust to these variations (but note the discussion below on the correlation of the price-cost margin with output variability). (15)The idea that labor hoarding can cause measured productivity to be procyclical obviously did not originate with Rotemberg and Summers. See, for example, Gordon [1979]. (16)In fact, Rotemberg and Summers argue that this result requires only a much milder form of price rigidity than suggested in this simple model. They present a model in which at any known level of aggregate demand, demand to the firm is uncertain, and even in a recession some states of nature will generate high enough demand that the firm can produce at capacity. Then as long as the firm must set its price before this state is realized (referred to as "micro" price rigidity), the firm will set price greater than marginal cost. (17)Using two-digit data, Rotemberg and Summers obtain a positive and significant coefficient on the concentration ratio. They argue that this provides support for the labor hoarding interpretation of markups.

REFERENCES

Ball, Laurence, and David Romer. "Real Rigidities and the non-Neutrality of Money." National Bureau of Economic Research Working Paper No. 2476, 1987. Bils, Mark. "The Cyclical Behavior of Marginal Cost and Price." American Economic Review, December 1987, 838-55. Blinder, Alan S. "Inventories and Sticky Prices: More on the Microfoundations of Macroeconomics." American Economic Review, June 1982, 334-48. Carlton, Dennis W. "The Theory and the Facts of How Markets Clear: Is Industrial Organization Valuable for Understanding Macroeconomics?" in Handbook of Industrial Organization, edited by R. Schmanlensee and R. Willig. New York: North Holland, 1989. Christensen, Laurits R., and William H. Greene. "Economies of Scale in U.S. Electric Power Generation." Journal of Political Economy, August 1976, 655-76. Cooper, Russel, and Andrew John. "Coordinating Coordination Failures in Keynesian Models." Quarterly Journal of Economics, August 1988, 441-63. Deaton, Angus S., and John Muellbauer. The Economics of Consumer Behavior. Cambridge: Cambridge University Press, 1980. De Long, J. Bradford, and Lawrence H. Summers. "Is Increased Price Flexibility Destabilizing?" American Economic Review, December 1986, 1031-44. Domowitz, Ian, R. Glenn Hubbard, and Bruce C. Petersen. "Market Structure, Durable Goods, and Cyclical Fluctuations in Markups." Mimeo, 1987. __. "Market Structure and Cyclical Fluctuations in U.S. Manufacturing." Review of Economics and Statistics, February 1988, 55-66. Feinberg, Robert M. "Market Structure and Employment Instability." Review of Economics and Statistics, November 1979, 497-505. Gordon, Robert J., "The |End-of-Expansion' Phenomenon in Short-Run Productivity Behavior." Brookings Papers on Economic Activity 2, 1979, 447-61. Hall, Robert E. "Market Structure and Macroeconomic Fluctuations." Brookings Papers on Economic Activity 2, 1986, 285-322. __. "The Relationship between Price and Marginal Cost in U.S. Industry." Journal of Political Economy, October 1988, 921-47. Jorgenson, Dale W. "Productivity and Postwar U.S. Economic Growth." Journal of Economic Perspectives, Fall 1988, 23-41. Lebow, David E. "Market Power, Productivity, and Business Cycles: Essays in Empirical Macroeconomics." Ph. D. Dissertation, Princeton University, 1989. __. "The Covariability of Productivity Shocks Across Industries." Economic Activity Working Paper No. 102. Washington: Board of Governors of the Federal Reserve System, 1990. Lustgarten, Steven, and Allan I. Mendelowitz. "The Covariability of Industrial Concentration and Employment Fluctuations." Journal of Business, April 1979, 291-305. Means, Gardiner. "Industrial Prices and their Relative Inflexibility." 74th Congress, 1st Session, Senate Document 13. Washington: Government Printing Office, January 1935. Ornstein, S. I. "Empirical Uses of the Price-Cost Margin." Journal of Industrial Economics, December 1975, 105-17. Qualls, David. "Price Stability in Concentrated Industries." Southern Economic Journal, October 1975, 294-8. Rotemberg, Julio J., and Garth Saloner. "A Supergame Theoretic Model of Business Cycles and Price Wars During Booms." American Economic Review, June 1986, 390-407. Rotemberg, Julio J., and Lawrence H. Summers. "Inflexible Prices and Procyclical Productivity." Quarterly Journal of Economics, November 1990, 851-74. Scherer, F. M. Industry Market Structure and Economic Performance. Chicago: Rand McNally, 1980. Smith, D.S. "Concentration and Employment Fluctuations." Western Economic Journal, September 1971, 267-77. Solow, Robert M. "Technical Change and the Aggregate Production Function." Review of Economics and Statistics, August 1957, 312-20. Weiss, Leonard W., and George A. Pascoe, Jr. "Adjusted Concentrated Ratios in Manufacturing, 1972 and 1977." Federal Trade Commission, June 1986. Woglom, Geoffrey. "Underemployment Equilibrium with Rational Expectations." Quarterly Journal of Economics, February 1982, 89-107.

DAVID E. LEBOW, Economist, Board of Governors of the Federal Reserve System. This is a revised version of chapters 1 and 2 of my doctoral dissertation. I would like to thank Larry Ball, Ben Bernanke, Alan Blinder, David Romer, David Wilcox, and two anonymous referees for helpful comments and discussions, and Wayne Gray for providing me with data. The opinions expressed here are my own and are not necessarily those of the Board of Governors of the Federal Reserve System or its staff.

Printer friendly Cite/link Email Feedback | |

Author: | Lebow, David |
---|---|

Publication: | Economic Inquiry |

Date: | Jan 1, 1992 |

Words: | 7152 |

Previous Article: | Taxation aggregate activity and economic growth; further cross-country evidence on some supply-side hypotheses. |

Next Article: | On Journalists' use of macroeconomic concepts. |

Topics: |