determinants of individual-firm markup in japan: market concentration, market share, and ftc...

27
Journal of the Japanese and International Economies 13, 424–450 (1999) Article ID jjie.1999.0430, available online at http://www.idealibrary.com on Determinants of Individual-Firm Markup in Japan: Market Concentration, Market Share, and FTC Regulations * Kenn Ariga Institute of Economic Research, Kyoto University, Yoshida-Honmachi, Sakyo-ku, Kyoto 606-8501, Japan E-mail: [email protected] Yasushi Ohkusa Institute for Social and Economic Research, Osaka University and Kiyohiko G. Nishimura Faculty of Economics, University of Tokyo Received February 7, 1999; revised July 13, 1999 Ariga, Kenn, Ohkusa, Yasushi, and Nishimura, Kiyohiko G.—Determinants of Individual- Firm Markup in Japan: Market Concentration, Market Share, and FTC Regulations This paper estimates individual firm level markup for more than 400 major manufacturing firms in Japan. Our estimates suggest the presence of significant market power for most of these firms, due not only to market concentration but also to the firms’ own market shares, as well as advertizing and sales promotion efforts. The paper then goes on to assess system- atically the impact on estimated markups of regulatory measures taken by the Fair Trade Commission (FTC) of the Japanese Government. We find that non-punitive FTC activities are directed toward the right targets and are reasonably effective, whereas injunctions, the strongest measure endowed to the FTC, has essentially no effect on the markups of firms in our sample. J. Japan. Int. Econ., Dec. 1999, 13(4), pp. 424–450. Institute of Economic Research, Kyoto University, Yoshida-Honmachi, Sakyo-ku, Kyoto 606-8501, Japan; In- stitute for Social and Economic Research, Osaka University; and Faculty of Economics, University of Tokyo. c 1999 Academic Press Journal of Economic Literature Classification Numbers: L13, L41. * An earlier version of the paper was presented at the NBER-CEPR-TCER Conference on Competi- tion Policy, December 18–19, 1998, International House, Tokyo, Japan. We benefited from comments by Professors Tim Bresnahan and Noriyuki Yanagawa and other participants of the conference. The research reported here is partially supported by a Grant-in-Aid from the Ministry of Education. 424 0889-1583/99 $30.00 Copyright c 1999 by Academic Press All rights of reproduction in any form reserved.

Upload: kutukankutu

Post on 27-Jul-2015

105 views

Category:

Documents


2 download

DESCRIPTION

Journal of the Japanese and International Economies 13, 424–450 (1999) Article ID jjie.1999.0430, available online at http://www.idealibrary.com onDeterminants of Individual-Firm Markup in Japan: Market Concentration, Market Share, and FTC Regulations∗Kenn ArigaInstitute of Economic Research, Kyoto University, Yoshida-Honmachi, Sakyo-ku, Kyoto 606-8501, Japan E-mail: [email protected] OhkusaInstitute for Social and Economic Research, Osaka Universityand Kiyohiko G. Nishimu

TRANSCRIPT

Page 1: Determinants of Individual-Firm Markup in Japan: Market Concentration, Market Share, and FTC Regulations

Journal of the Japanese and International Economies13, 424–450 (1999)Article ID jjie.1999.0430, available online at http://www.idealibrary.com on

Determinants of Individual-Firm Markup in Japan: MarketConcentration, Market Share, and FTC Regulations∗

Kenn Ariga

Institute of Economic Research, Kyoto University, Yoshida-Honmachi, Sakyo-ku,Kyoto 606-8501, Japan

E-mail: [email protected]

Yasushi Ohkusa

Institute for Social and Economic Research, Osaka University

and

Kiyohiko G. Nishimura

Faculty of Economics, University of Tokyo

Received February 7, 1999; revised July 13, 1999

Ariga, Kenn, Ohkusa, Yasushi, and Nishimura, Kiyohiko G.—Determinants of Individual-Firm Markup in Japan: Market Concentration, Market Share, and FTC Regulations

This paper estimates individual firm level markup for more than 400 major manufacturingfirms in Japan. Our estimates suggest the presence of significant market power for most ofthese firms, due not only to market concentration but also to the firms’ own market shares,as well as advertizing and sales promotion efforts. The paper then goes on to assess system-atically the impact on estimated markups of regulatory measures taken by the Fair TradeCommission (FTC) of the Japanese Government. We find that non-punitive FTC activitiesare directed toward the right targets and are reasonably effective, whereas injunctions, thestrongest measure endowed to the FTC, has essentially no effect on the markups of firmsin our sample.J. Japan. Int. Econ., Dec. 1999,13(4), pp. 424–450. Institute of EconomicResearch, Kyoto University, Yoshida-Honmachi, Sakyo-ku, Kyoto 606-8501, Japan; In-stitute for Social and Economic Research, Osaka University; and Faculty of Economics,University of Tokyo. c© 1999 Academic Press

Journal of Economic LiteratureClassification Numbers: L13, L41.

∗ An earlier version of the paper was presented at the NBER-CEPR-TCER Conference on Competi-tion Policy, December 18–19, 1998, International House, Tokyo, Japan. We benefited from commentsby Professors Tim Bresnahan and Noriyuki Yanagawa and other participants of the conference. Theresearch reported here is partially supported by a Grant-in-Aid from the Ministry of Education.

424

0889-1583/99 $30.00Copyright c© 1999 by Academic PressAll rights of reproduction in any form reserved.

Page 2: Determinants of Individual-Firm Markup in Japan: Market Concentration, Market Share, and FTC Regulations

INDIVIDUAL -FIRM MARKUP IN JAPAN 425

1. INTRODUCTION

For most economists, in and outside of Japan, competition policy in Japan islargely a neglected cousin of Japan’s industrial policy. Although the Japanesegovernment in recent years has made some attempts to place more emphasis onthe enforcement of anti-monopoly laws, the Fair Trade Commission (FTC), anindependent (administrative) body of the government, is tiny, seriously under-manned, reportedly weak in political power, and largely neglected, at least untilquite recently.

The lack of resources is not limited to the government. There has been littlesystematic research on the overall competitiveness of Japanese markets that iscomparable and consistent with the research on the United States and Europe.1

Among other things, in spite of its obvious importance, we know very little aboutthe overall magnitude, cross-section distribution, and temporal variations of pricemarkups in Japan.2

This paper attempts to fill this gaping hole in empirical studies of Japanesemarkets and, in particular, of the pricing behavior of firms. We employ the estima-tion procedure developed in Nishimuraet al. (1999) (hereafter NOA)3 to obtainestimates of individual firm markup over marginal cost4 for roughly 450 majormanufacturing firms5 located in 130 four-digit industries.6 The average estimatedmarkup of these firms is 1.29, and roughly 90% of the sample firms have markupslarger than unity. The markups are 1.5 or higher at 74 firms. Our estimation resultsthus suggest the significant market power of these firms located in a large varietyof manufacturing sectors.

After the estimation of individual firm markup, we conduct a series of regres-sion analyses on the determinants of these markups. The paper addresses twomain issues in these exercises. First of all, given the large diversity in the empiricalstudies on the magnitude and consequences of the market power of firms in indus-trialized economies, our study attempts to offer a comprehensive assessment ofthe magnitude and major sources of the market power of Japanese firms. Another

1 Bresnahan (1989) surveys major empirical studies in this field.2 Odagiri and Yamashita (1987), Baba (1995), and our own previous paper (see below) are the few

studies that estimate markups (price–cost margins) in Japanese industries.3 This paper differs from NOA in three respects. First, the other paper tries to get a general picture

of Japanese firms and estimates the markup over marginal cost of 1643 firms, including construction,wholesale, retail, and land transportation, firms, while this paper is concerned with only manufacturing.Second, NOA uses the value-added production function framework, while this paper utilizes the grossproduction function incorporating material inputs. Third, this paper analyzes the determinants of themarkup, while the other examines whether the standard concept of an industry is statistically meaningfulby testing the homogeneity of coefficients within the industry.

4 Our estimates include intermediate, material, as well as factor, inputs so that the estimated markupsare the ratios of prices over marginal costs inclusive of these inputs.

5 As we see below, most of the firms produce a variety of products across these industries and theyare located in more than one industry.

6 The JSIC lists roughly 550 four-digit manufacturing industries.

Page 3: Determinants of Individual-Firm Markup in Japan: Market Concentration, Market Share, and FTC Regulations

426 ARIGA, OHKUSA, AND NISHIMURA

important issue that we address is the role of competition policy in Japan. We focuson econometric investigation of various regulatory and punitive measures takenby the Fair Trade Commission of Japan (Kousei Torihiki Iinkai).

This paper addresses these issues by estimating individual firms’ average ofproduct-specific markups over marginal cost for different products. The uniquestructure of the data and the estimate of markup for individual firms enable us toaddress issues that cannot be analyzed easily in cross-sectional studies based uponindustry-level data. Our estimation of markup developed in NOA also providessimultaneously the estimate for the magnitude of the fixed cost and the effect ofadjustment costs due to the firm’s growth. Using these estimates, we also obtainthe local curvature of a U-shaped marginal cost curve and are able to examine thetechnological properties of various industries.

The data on individual markets are also unique. Our data include informationon sales (production) shares of major producers for more than 800 industrial prod-ucts. This data set can be used to calculate concentration and own-market-sharemeasures, and they can be utilized in analyzing the past history of anti-competitiveconduct and disciplinary actions documented and enforced by the FTC.

The paper is organized as follows. In the next section, we briefly review theunderlying theoretical model developed in NOA for the estimation of individualfirm markup. Section 3 then extends the model to address the issues of our interestin this paper. Section 4 reports major estimation results and investigates the effectof the FTC’s activities on markups. Concluding remarks are given in Section 5.

2. A MODEL OF A MULTI-PRODUCT FIRM

In this section, we derive the basic relation that can be utilized in the empiricalanalysis of markup over marginal cost. This is the relationship among the elasticityof output to inputs, the markup rate over marginal cost, and factor shares. Using thisrelationship, we can identify the markup rate from factor and other inputs sharesand information about production technology, without knowing a firm’s input andoutput prices, its demand conditions, or the strategic interaction among firms.

In estimating markup, we utilize a method developed in NOA. In that paper,we derived a fundamental relationship between factor shares, a markup, and tech-nological parameters determining output elasticity, and we estimated a markuputilizing this relationship. We assumed a value-added production function and ig-nored multiplicity of products. In this section, we extend this method and apply itto the case of a gross production function with material inputs and multiple prod-ucts. We proceed in three steps. In the following, we utilize the following identifierconventions:

i firmk productt period.

Page 4: Determinants of Individual-Firm Markup in Japan: Market Concentration, Market Share, and FTC Regulations

INDIVIDUAL -FIRM MARKUP IN JAPAN 427

The first step is to determine a fundamental relation among factor shares, themarkup, and the output elasticity for each product. Here, we briefly summarizethe results given in NOA, which contains detailed discussion of the assumptionsand implications of this approach. LetK ik

t , Likt , andJik

t be, respectively, firmi ’scapital service input, labor input, and material input to produce thekth product inthet th period. Firmi ’s production function of thekth product is

Qikt = Fik

(K ik

t , Likt , Jik

t

∣∣Äi). (1)

Here,Äi is the set of firmwide parameters which are given to the production ofthekth product.

We make two sets of assumptions. First, we assume that (1) the production func-tion Fik(K ik

t , Likt , Jik

t ) is homothetic in (K ikt , Lik

t , Jikt ) and (2) the firm minimizes

production costs, taking input prices as given. Under the homotheticity assump-tion, the elasticity of output with respect to capital, labor, and material,εik

Q , is welldefined by

εikQ ≡

K ikt

Qikt

[Fik

K

]t+ Lik

t

Qikt

[Fik

L

]t+ Jik

t

Qikt

[Fik

J

]t, (2)

where [g]t is the value of functiong evaluated att . Next, letµikt be the markup

over the marginal cost such that

pikt = µik

t λikt , (3)

in which pikt is firm i ’s price for itskth product,λik

t is firm i ’s marginal cost forproducing thekth product:

λikt =

∂Cik

∂Qik. (4)

Here,Cik is the cost function such that

Cik(r ik

t , wikt , z

ikt , Qik

t

) = Min[r ik

t K ikt + wik

t Likt + zik

t Jikt

]subject to (1),

wherer ikt ,wik

t , andzikt are the rental price of capital, the wage rate, and the price of

material, respectively, in firmi ’s production of thekth product in periodt . Finally,the termαik

X (X= K , L , J) is the productwise input share such that

[αik

K

]t =

r ikt K ik

t

pikt Qik

t

;[αik

L

]t =

wikt Lik

t

pikt Qik

t

;[αik

J

]t =

zikt Jik

t

pikt Qik

t

. (5)

Using the above definitions, we have the following simple relationship between

Page 5: Determinants of Individual-Firm Markup in Japan: Market Concentration, Market Share, and FTC Regulations

428 ARIGA, OHKUSA, AND NISHIMURA

the output elasticity, the markup, and the input shares7:

εikQ = µik

t

([αik

K

]t +

[αik

L

]t +

[αik

J

]t

). (6)

Second, we further assume that (i) there exists a fixed cost in production, (ii) themarginal cost is constant if the firm’s size is unchanged, and (iii) the marginal costis increasing if the firm as a whole is expanding or contracting. Property (iii) isone representation of the adjustment cost in firm growth. In order to incorporatethe above three assumptions, we assume the production function

Fik

(K ik

t , Likt , Jik

t

∣∣∣∣ 1Sit

Sit

, Qi Nt

)=Max

[0,(1+ γ ik

0

)f ik

(K ik

t , Likt , Jik

t

∣∣∣∣ 1Sit

Sit

)− γ ik Qi N

t

], (7)

whereγ ik andγ ik0 are technological parameters,Si

t is firm i ’s size, andQi Nt is its

normal level of total output. (The empirical counterparts of the latter two variableswill be discussed in the next section.)

In this formulation,γ ik0 andγ ik Qi N

t represent the existence of the fixed cost.Before the firm ever produces positive output, it has to input sufficient capital, labor,and material [(K ik

t , Likt , Jik

t ) such that (1+ γ ik0 ) f ik(K ik

t , Likt , Jik

t ) = γ ik Qi Nt )].

The cost of such capital, labor, and material inputs is the fixed cost. A major part ofthe fixed cost is overhead, so that the fixed cost depends on the firmwide plannedscale of production, which we call a normal level of total output,Qi N

t . However,the magnitude of the fixed cost is likely to differ among firms and products, so thatthe coefficientsγ ik

0 andγ ik may differ among firms and products. Our approach isdifferent from the conventional approach of characterizing the fixed cost as a resultof fixed factors of production: we assume that production organization planned fora normal level of output is fixed in the short run, but factors of production arevariable and substitutable.

Function f ik is assumed to be homogeneous of degree 1− δik(1Sit /S

it )

2 inits inputs. Thus, marginal cost is constant (f ik is linear homogeneous) if thereis no change in the size of the firm (1Si

t /Sit = 0). However, the production ef-

ficiency is reduced if the firm expands and contracts, since firmwide productionre-organization due to expansion and contraction necessarily strains the firm’smanagement resources. This is the essence of the adjustment cost discussed in theliterature of firm growth, often called the Penrose effect. The effect of the adjust-ment cost may differ among firms and products, andδik represents the magnitudeof this effect.

7 Cost minimization impliesFK = r/p, FL =w/p, andFJ = z/p (superscripts are ignored for ex-pository simplicity in this footnote). Then, it is straightfoward to show this relationship by first substitut-ing FK = r/p, FL =w/p, andFJ = z/p into the definition of the input elasticity and then rearrangingterms using the definition of input shares.

Page 6: Determinants of Individual-Firm Markup in Japan: Market Concentration, Market Share, and FTC Regulations

INDIVIDUAL -FIRM MARKUP IN JAPAN 429

Under (7), it is straightforward to show (see NOA)

εikQ =

{1+ γ ik

(Qi N

t

Qit

)}{1− δik

(1Si

t

Sit

)2}≈ 1+γ ik

(Qi N

t

Qit

)− δik

(1Si

t

Sit

)2

.

(8)

Here we assume thatγ ik andδik are sufficiently small, so that the cross-term isignored.

Combining (6) and (8), we have

µikt

([αik

K

]t +

[αik

L

]t +

[αik

J

]t

) ≈ 1+ γ ik

(Qi N

t

Qit

)− δik

(1Si

t

Sit

)2

. (9)

This is the fundamental relationship from which we can identify the markuprateµik

t .The second step is to obtain the firmwide relation among input shares, markups,

and determinants of output elasticities. Through tedious but straightforward cal-culation (see Appendix A), we have the multi-product counterpart of (9) suchthat

µit

([αi

K

]t+ [αi

L

]t+ [αi

J

]t

) ≈ 1+∑

k

ωikt γ

ik

(Qi N

t

Qit

)−∑

k

ωikt δ

ik

(1Si

t

Sit

)2

.

(10)

On the right-hand side of (10),ωikt is the sales share of thekth product,

ωikt =

pikt Qik

t(∑k pik

t Qikt

) . (11)

On the left-hand side,µit is firm i ’s mean marginal markup rate, which is the

cost-share-weighted average of the marketwise marginal-cost markup, such that

µit =

∑k

ωikt µ

ikt , whereωik

t =(r ik

t K ikt + wik

t Likt + zik

t Jikt

)∑k

(r ik

t K ikt + wik

t Likt + zik

t Jikt

) , (12)

andαiX (X = K , L , J) is the firmwide factor shares such that

[αi

K

]t =

∑k r ik

t K ikt∑

k pikt Qik

t

;[αi

L

]t =

∑kw

ikt Lik

t∑k pik

t Qikt

;[αi

J

]t =

∑k zik

t Jikt∑

k pikt Qik

t

. (13)

Page 7: Determinants of Individual-Firm Markup in Japan: Market Concentration, Market Share, and FTC Regulations

430 ARIGA, OHKUSA, AND NISHIMURA

Taking the logarithm of both sides of (10) and rearranging terms, we have

log([αi

K

]t+ [αi

L

]t+ [αi

J

]t

)≈− logµi

t +∑

k

ωikt γ

ik

(Qi N

t

Qit

)−∑

k

ωikt δ

ik

(1Si

t

Sit

)2

, (14)

where the relation log(1+ x) ≈ x for x with small|x| is utilized.The third step is to obtain an empirically estimable relation between observable

variables, based on (14). Note thatγ ik andδik are the technological parametersnot directly observable, whileωik

t , the sales share variable, is not available forall sample periods in the empirical study presented below. In order to circumventthese problems, we make additional assumptions. We assume that firms in thesame product market have the same fixed cost parameter and the same Penroseeffect,8

γ ik = γ k for all i (15)

δik = δk for all i . (16)

Let the time averageωik , of ωikt , be such that

ωik = limT→∞

1

T

T∑t=0

ωikt . (17)

Then, under the additional assumptions (15) and (16), it is straightforward to showthat (14) can be rewritten as

log( [αi

K

]t +

[αi

L

]t +

[αi

J

]t

)≈− logµi

t +∑

k

ωikγ k

(Qi N

t

Qit

)−∑

k

ωikδk

(1Si

t

Sit

)2

+ wit , (18)

where

wit =

(Qi N

t

Qit

)∑k

[ωik

t − ωik]γ k −

(1Si

t

Sit

)2∑k

[ωik

t − ωik]δk. (19)

It is natural to assume that the sales-share deviation from its long-run averageωik

t − ωik of a particular product is uncorrelated to the firmwide normal-to-actual

8 It is desirable to test this and following homogeneity assumptions onγ andδ statistically. Unfor-tunately however, this practically impossible since we do not have a sufficient degree of freedom inour data set.

Page 8: Determinants of Individual-Firm Markup in Japan: Market Concentration, Market Share, and FTC Regulations

INDIVIDUAL -FIRM MARKUP IN JAPAN 431

total output ratioQi Nt /Qi

t and firm-level growth1Sit /S

it . Keeping this in mind,

we assume

limT→∞

1

T

T∑t=0

wit = 0,

so thatwit can be treated as an error term. Equation (18) is the base equation that

we use in the following estimation.

3. AN EMPIRICAL MODEL OF MARKUP AND MARKET STRUCTURE

The model that we presented in the previous section relies only upon the spec-ification of production technology and cost minimization. As such, the modelcan accommodate a variety of specifications of production and pricing decisions,including that of markups. We take advantage of this structure and exploit the inter-firm and inter-market variations in the variety of characteristics related in one wayor another to (1) seller concentrations, (2) market shares, and (3) the nature andpossible degree of collusive behavior of firms within a market.

3.1. Determinants of markups: Conjectural Variations and Pricing Policy

Since the base Eq. (18) above is derived essentially as the cost minimization,(18) can be combined to a model of product price determination. Consider arepresentative firm’s pricing policy, facing demand scheduleDik in marketk,

pikt = arg max

[pik

t Qikt − Cik

(r ik

t , wikt , z

ikt , Qik

t

)],

where

Qikt = Dik

(pik

t ,{

pjkt : j = 1, . . .m

};2k

t

),

in which2kt denotes the vector of exogenous variables. The conventional short-run

profit maximization condition is given by

µikt =

ηikt −

∑j η

i jkt 1

i jkt

ηikt −

∑j η

i jkt 1

i jkt − 1

,

whereηikt andηi jk

t are own and cross-price elasticities, respectively, and1i jkt is

the conjectural variation such that

1i jkt ≡

pikt

p jkt

∂pjkt

∂pikt

.

Page 9: Determinants of Individual-Firm Markup in Japan: Market Concentration, Market Share, and FTC Regulations

432 ARIGA, OHKUSA, AND NISHIMURA

Then, the firm-level markup is (see (12))

µit =

∑k

ωikt

ηikt −

∑j η

i jkt 1

i jkt

ηikt −

∑j η

i jkt 1

i jkt − 1

.

A variety of variables may influence the firmwide markup through their effectson two determinants,ηik

t and∑

j ηi jkt 1

i jkt . Since our principal interest lies in the

long-run cross-sectional variations in the markup, we need to filter the cyclicalfluctuations.9 After netting out cyclical fluctuations, we consider three sets ofvariables that potentially influence one of the two determinants of the markup.

1. Product Characteristics: These are likely to influence price elasticities ofdemand. We use advertisement and sales promotion expenditure to represent thedegree of product differentiation.

2. Market Characteristics: To the extent that higher seller concentration is con-ducive to collusive pricing behavior, we use conventional concentration measuresto represent market characteristics, as well as individual-firm market shares torepresent the potential effect of market dominance.

3. FTC’s Policy Measures: These variables may influence the markup ratethrough their impacts on collusive conducts, either by deterrent effects or by directinjunctions to prohibit specific conducts. It is also conceivable that some policymeasures may directly influence the market characteristics; for example, the lawprohibits overly speculative sales promotion prizes, thereby possibly influencingthe nature of product differentiations in markets.

3.2. Data and Econometric Issues

The firm-level data that we use are based upon official financial statements ofmajor manufacturing firms submitted to the Ministry of Finance. This data set canbe considered as the most broadly based panel of large Japanese manufacturingfirms. The data set includes all the firms listed in major stock exchanges, andit also includes major non-listed firms. It contains information about the sales,the operating profits, the employment compensation, the depreciation allowance,the investment expenditure, the total asset value, and the book value of existingcapital stocks of each firm, as well as other information. The data are annual, andthe sample period is 1975–1994.

9 The direction of the cyclicality of the markup has been intensively discussed in the literature (seeRotemberg and Woodford, 1991). In theory, it can go either way, depending on information availabilityand other factors, even though underlying market structure is the same. For example, Green and Porter(1984) suggest a procyclical markup under imperfect information and Rotemberg and Saloner (1986)stress a counter-cyclical one under perfect information although both assume implicit collusion amongfirms. Similarly, Bils (1987) and Gottfries (1991) derive a counter-cyclical markup in the customermarket, but Nishimura (1989) implies a procyclical markup. The above argument suggests that whetherthe markup is procyclical or not is essentially an empirical problem.

Page 10: Determinants of Individual-Firm Markup in Japan: Market Concentration, Market Share, and FTC Regulations

INDIVIDUAL -FIRM MARKUP IN JAPAN 433

TABLE ISummary Statistics (1), Firms: Sample Firms= 451,

Firm-Observation Years= 8921

Mean Std. Min Max

Assetsa 131.203 315.147 0.447 4130.666Salesa 119.604 271.966 0.402 4550.086NORCUR 0.0484 0.0598 −0.368 1.105KGROWTH 0.0493 0.298 −0.969 11.310CFAS 0.0484 0.059 −0.368 1.105

Note. VariablesNORCUR KGROWTH, andCFASand are de-fined in the text and are used in the base-line regression (13).

a Figures in billion yen.

The second set of data on product markets is taken from “NihonMarket ShareJiten” (Yearbook of Market Shares in Japan) (Yano Research Institute, 1995, here-after Yano), which collects data on the production of finely classified individ-ual products and services. In the book, roughly 1500 products and services arecovered.10 For each individual market, information about major producers andtheir sales volumes is collected.11 We used all of the data on manufactured prod-ucts, comprising roughly 720 product lines, covering virtually the entire universeof the important manufactured products in Japan.12 In order to mesh the productdata with the firm-level data, we looked for the entry of sample firms in thesemarket data. We also classified these products into 130 four-digit manufacturingindustries, following the JSIC (Japan Standard Industry Classification). We treateach of these four-digit sectors comprising the relevant product market.13 Hence,each of the 720 or so product market data is grouped into one of the 130 four-digitmanufacturing industries.14 Rather than assigning each firm a unique four-digitindustry code affiliation, we consider each sample firm as a multi-product firmwhose lines of products are picked up from Yano (1995).

Table I shows the panel of firms used in the analysis. In the data set, there are451 firms with 8921 firm-year observations. The sample period is between 1975and 1994. Most of the sampled firms continue to exist and their data are available.The way these variables are constructed is explained in Appendix B, and Table I

10 Most of the data collected in Yano (1995) are based on 1994 surveys.11 When production shares are given in quantities, we used the unit price of the corresponding product

in theCensus of Manufacturers, 1995.12 One major shortcoming of Yano (1995) is the lack of data on imports.13 TheCensus of Manufacturers, the most comprehensive establishment-level survey of manufactur-

ing in Japan, lists roughly 1800 JSIC six-digit individual industrial products. Yano (1995), on the otherhand, covers roughly 850 industrial goods. The latter, however, does not cover the whole spectrumof industries and instead focuses more narrowly on important industrial products. For the productscovered, Yano (1995) classifies individual products into groups finer than the JSIC six-digit level.

14 Among 130 four-digit industries, 49 of them have only one entry in Yano (1995) and 22 industrieshave two entries. The number of industries with 3 to 10 entries are 10, 7, 9, 8, 6, 5, 3, and 4, respectively.One industry has 49 entries.

Page 11: Determinants of Individual-Firm Markup in Japan: Market Concentration, Market Share, and FTC Regulations

434 ARIGA, OHKUSA, AND NISHIMURA

TABLE IISummary Statistics (2), Markets: Number of Individual Commodities, 723,

Number of Four-Digit Industries, 130, Year= 1994

Mean Std. Min Max

Salesa 473.03 1010.06 0.007 9573.58Herfindahl index 0.285 0.184 0.072 1.00Three-firm concentration ratio 0.776 0.149 0.483 1.00

a Figures in billion yen.

gives descriptive summary statistics for these variables in our sample. It mightbe noted that the sampled firms represent roughly 20% of all the listed firms inJapan, and roughly 35% of all listed manufacturing firms. Summary statistics onmarkets are shown in Table II. The aggregate size of the markets covered in ourdata set is 89 trillion yen in 1994, which corresponds to roughly a third of the totalgross shipments (266 trillion yen) of all manufactured goods in Japan. As the datashows, the average three-firm concentration ratio is 0.77, considerably higher thanthe average from previous selective studies done by the FTC.15

3.3. Model Specifications

In the previous section, we obtained a basic relation to be estimated for individualfirm i (replacing≈ with = for notational simplicity),

log([αi

K

]t +

[αi

L

]t +

[αi

J

]t

)=− logµi

t +∑

k

ωikγ k

(Qi N

t

Qit

)−∑

k

ωikδk

(1Si

t

Sit

)2

+ wit .

To estimate this equation, we use proxies for the normal-to-actual total output ratioQi N

t /Qit and the change in the firm size,1Si

t /Sit . The variableQi N

t /Qit is proxied

by the ratio of the firm’s trend total output to the current total output,NORCURit .

For1Sit /S

it , we use the growth rate of the firm’s capital stock,KGROWTHi

t as itsproxy.

Now we have to filter out cyclical fluctuations in the markup. To do this, weassume that (a) the individual-product markup rate depends on the condition ofthe own-product market, (b) the sensitivity of the markup to market conditions,

15 According to a study conducted by the FTC in 1979, the three-firm concentration was 0.7 or higherin 86 cases, 0.5 to 0.7 in 80 cases, 0.3 to 0.5 in 92 cases, and less than 0.3 for 40 cases. On the other hand,there are reasons to believe that the markets used in our analysis are fairly representative and the dataare more accurate. First of all, the corresponding firms are the largest manufacturing firms in Japan,and it is well known that larger firms tend to produce in more concentrated industries. Second, Yano(1995) segments markets into finer categories than most of the previous studies, focusing narrowly onthe important industrial products. Finer classifications of products naturally increases average marketconcentration.

Page 12: Determinants of Individual-Firm Markup in Japan: Market Concentration, Market Share, and FTC Regulations

INDIVIDUAL -FIRM MARKUP IN JAPAN 435

is the same among firms,16 (c) the individual product-market condition is highlycorrelated with the firm’s overall market condition, and (d) the firm’s overall mar-ket condition is well approximated by the firm’s cash-flow-to-asset ratio. Theseassumptions imply the specification of markup behavior

logµikt = aik

0 + φkCFASit , (20)

whereaik0 is a parameter andCFASi

t is the cash-flow-to-asset ratio of each firmthat we used in NOA. We also utilize the approximation logµi

t ≈∑

k ωik logµik

tin our empirical analysis. Then we have (replacing≈ with =)

logµit = ai

0+∑

k

ωikφkCFASit , (21)

whereai0 ≡

∑k ω

ikaik0 .

Putting these specifications together, our base-line regression is of the form

FSit = −ai

0+∑

k

ωik(γ kNORCURi

t − δkKGROWTHit − φkCFASi

t

)+ wit , (22)

whereFSit is the log of the sum of factor and other input shares in periodt andωik

is firm i ’s time-average of the sales weight of productk. We use the 1994 value ofωik

t (reported in it Yano (1995)) as a stand-in for this parameter. The disturbancewi

t might be correlated with other variables in the right-hand side of (22). Thus,we use instrumental-variable estimation. In the following analysis, the instrumentsare lagged right-hand side variables (except for constant and industry dummies)and those of industry averages used in each of the regressions.

3.4. Base Results

The results for the base-line case is reported in Table III. In this table, figuresreported forγ , δ, andφ are the summary statistics for panel regression estimatesfor each of the four-digit manufacturing industries. Overall, we obtained fairlysharp estimates. The majority of the estimated coefficients have the correct signs(non-negative): Roughly 70, 60, and 75%, respectively, of the estimatedγ , δ, andφ are positive and significant at the 5% level. About 25% of the coefficients havethe correct signs (non-negative), and they are not significantly different from zero.Only a remaining 5% (20 out of 390= 130∗3) of the coefficients have wrong signs(negative), but only two of them are significant.

Across industries, we observe a clear pattern that the estimated coefficientsγ , δ,andφ are positively correlated. Namely, industries with a strong effect of short-run

16 In the homogeneous-product framework, NOA showed that this is in fact the case in the Japaneseindustries they considered.

Page 13: Determinants of Individual-Firm Markup in Japan: Market Concentration, Market Share, and FTC Regulations

436 ARIGA, OHKUSA, AND NISHIMURA

TABLE IIISummary Statistics of Base-Line Model Estimates

γ δ φ

All |t | > 2 |t | < 2 All |t | > 2 |t | < 2 All |t | > 2 |t | < 2

N 130 85(2)a 45(16)a 130 92 38(2)a 130 98 32(2)a

Mean 0.156 0.274 −0.066 0.142 0.198 0.0036 0.980 1.109 0.585Median 0.0097 0.164 0.0016 0.0153 0.0224 0.0031 0.584 0.601 0.538Min −1.807 −0.0322 −1.807 −0.086 0.0020−0.0860 −0.785 0.511 −0.785Max 6.564 6.564 0.165 5.563 5.563 0.0561 10.202 10.202 3.0871st Qrt. 0.0023 0.007 −0.0245 0.0047 0.0118 0.0014 0.541 0.559 0.5133rd Qrt. 0.0311 0.0523 0.0047 0.0347 0.0461 0.0084 0.667 0.685 0.594

a Number of negative estimates in parenthesis.

fixed costs, i.e., those with large value ofγ, tend also to have largeδ (significantPenrose effect) andφ (pro-cyclical markup).

When we classify industries in terms of relative magnitudes of estimated valuesof δ andγ,several interesting patterns emerge. They are shown in Table IV. In manyelectric machinery and industrial machinery industries, both of these coefficientsare statistically significant and large, suggesting a strongly U-shaped marginalcost curve. On the other hand, many processed foods (finished goods) industriesexhibit essentially flat marginal cost curves. Some of relatively low-tech metal

TABLE IVProduction Technology Characteristics

(1) U-shapedAC

1225 Fish paste products 1229 Misc. seafood products1231 Canned preserved fruits 1242 Soy sauce1295 Soy bean curd (incl. fried) 2213 Plastic pipe fittings2292 Plastic containers 2833 Heated air and hot water heating systems

(2) FlatAC

2038 Synthetic rubber 2221 Plastic film2092 Agricultural chemicals 2441 Leather footwear2811 Tin cans 2899 Fabricated metal products, n.e.c.2942 Metal working machinery 3231 Medical instruments

(3) Decreasing returns

1251 Sugar (exc. sugar refining) 1293 Noodles and pastas1799 Furniture and fixtures 2052 Soaps and synthetic detergents2212 Plastic pipes and tubes 2733 Aluminum rolling

(4) Increasing returns

3217 Analytic instruments

Notes.(1) U-shapedAC: γ andδ are significant and in the largest quartile. (2) FlatAC: γ andδ are in the smallest quartile. (3) Decreasing returns:γ is in the smallestquartile andδ is significant and in the largest quartile. (4) Increasing returns:δ is inthe smallest quartile andγ is significant and in the largest quartile.

Page 14: Determinants of Individual-Firm Markup in Japan: Market Concentration, Market Share, and FTC Regulations

INDIVIDUAL -FIRM MARKUP IN JAPAN 437

products industries also fall into this category. Compared to these two cases, onlya few industries have largeγ and smallδ, or vice versa. These patterns are broadlyin accordance with conventional wisdom on the cross-sectional characteristics ofindustry technology and scale effects on costs.

Table III also shows the estimate of the markup’s cyclical sensitivityφ. Themarkup shows statistically significant pro-cyclicality in most (98 out 130) of theindustries. Only two out 130 industries have negative cyclical sensitivity, but theircoefficient is not statistically significant. The cyclical effect is large in industrialmachinery, electric machinery, and chemical industries. In many consumer goodsindustries, most notably those in processed foods sectors, markups are largelyacyclic. The result is broadly consistent with previous studies examining the cycli-cality of Japanese markups. For example, Nishimura and Inoue (in press) couldnot reject the null hypothesis of no cyclicality in the case of the food processingindustry’s markup, whereas markups in other branches in manufacturing do exhibitcyclical fluctuations.

3.4.1. Sample-period average markup of individual firms and industries.Inorder to assess the magnitude of the market power that Japanese firms possess,we used the regression results and computed the individual firms’ sample-period-average of marginal-cost markup in a logarithm, logµi , such that

log µi = 1

Ti

Ti−1∑t=0

logµit , (23)

whereTi is the number of observations for firmi . Ti may differ from firm to firmsince some firms terminated and some others emerged within our sample period.From (21), we immediately get

log µi = ai0+

∑k

ωikφkCFASi , whereCFASi = 1

Ti

Ti−1∑t=0

CFASit . (24)

Table V and Fig. 1 report the result of the cross-sectional analysis based on (24).For about 50% of the firms the sample-period-average markupµi is significantly

TABLE VEstimated Individual-Firm Markupµi

Panel (1): DistributionFirm N Mean Median Min Max 1st Qrt. 3rd Qrt.

451 1.291 1.110 0.483 13.96 1.036 1.269

Panel (2):t-valuesSignificant at 5% Significant at 10% Not significant Total

µi > 1 229 83 96 408µi < 1 2 0 41 43

Total 231 (51.2%) 83 (18.4%) 137 (30.4%) 451

Page 15: Determinants of Individual-Firm Markup in Japan: Market Concentration, Market Share, and FTC Regulations

438 ARIGA, OHKUSA, AND NISHIMURA

FIG

.1.

Dis

trib

utio

nof

estim

ate

indi

vidu

alfir

mm

arku

p µi .

Not

eth

attw

oou

tlier

sla

rger

than

5ar

eex

clud

edfr

omth

efig

ure.

Page 16: Determinants of Individual-Firm Markup in Japan: Market Concentration, Market Share, and FTC Regulations

INDIVIDUAL -FIRM MARKUP IN JAPAN 439

greater than unity at the 5% level.17 In contrast, we find only two firms whoseµi

is significantly smaller than unity at the 5% level. If we adopt a 10% significancelevel, roughly three quarters of the sample firms have a markupµi significantlygreater than unity. This table thus reveals that the markup rate is significantlygreater than unity in most of the sample firms, and suggests the non-negligiblemarket power possessed by firms in this panel. The result is broadly in line withprevious studies (see, for example, Odagiri and Yamashita (1987) and Shinjo(1977)), although previous studies estimated the markup over the average costrather than the marginal cost.

3.4.2. Market shares and concentration.The estimated firm-level markup isthe weighted average of its markups in different product markets. We exploit thisstructure and investigate the possible effect of market structure on markups.

The logarithm of the product-specific markup logµikt in (20) is determined by

several market conditionsmkjt :

logµikt = β ik

0 +∑

j

β j mkjt .

Therefore, the sample-period-average of the firm-level markup, logµi in (23) canapproximately be written as (here we use an approximation logµi

t ≈∑

k ωik logµik

tas in (21))

log µi = β i0+

∑j

β j mkj , (25)

where

β i0 ≡

∑k

ωikβ ik0 and mk

j ≡1

Ti

Ti−1∑t=0

∑k

ωikmkjt .

Table VI reports cross-sectional regressions of the form (25), wherein as marketconditionsmk

j we use (a) the individual firm’s market share, (b) its advertizementand sales promotion expenditure as a stand-in for product differentiation, and(c) alternative measures of market concentration. The results indicate that the effectof theown-market shareis (either marginally or highly) statistically significant andlarge. The estimated coefficient of the own-market share suggests that the averagemarkup of a firm with a 20% market share is roughly 5 to 8% higher than that ofa firm whose average market share is 5%. The advertizement-expenditure effectis large though it is marginally significant. Concentration also increases markups,although its effect is only marginally significant and quantitatively modest.

17 Thet values reported in Table V are the ones for testing whether or not logµi differs significantlyfrom zero.

Page 17: Determinants of Individual-Firm Markup in Japan: Market Concentration, Market Share, and FTC Regulations

440 ARIGA, OHKUSA, AND NISHIMURA

TABLE VIMarket Structure and Markups (Cross Section Regressions)

Variable Impacts on the level of markup

Own market share 0.317 [1.74] 0.336 [1.75] 0.462 [3.65]Three-firm concentration ratio — 0.325 [1.60] 0.136 [1.58]Herfindahl index 0.440 [1.69] — —Advertizement/sales promotion 0.756 [1.79] 0.821 [1.77] —

Note.Thet values are in brackets. The left-hand side variable is the fixed effecttermai

0 for individual firms. (Number of samples= 451.)

4. THE IMPACT OF THE FTC’S REGULATORYMEASURES ON MARKUPS

4.1. The FTC’s Regulatory Actions

In this section we investigate the effectiveness of the FTC’s regulatory policies18

in curtailing monopolistic conduct in individual markets. The most important pol-icy instrument used by the FTC is administrative injunctions and punitive actionscalledshinketsu.

The process leading toshinketsuis in principle similar to court procedure, withthe FTC’s investigation body acting in the prosecutor’s role. However, the lawallows the skipping of the court-like procedure if the accused party concedesbefore the process begins. In this case, a formalshinketsuis issued immediatelyand suchshinketsuis calledkankoku shinketsu(kankoku(consent order) for short).In the past, very few cases have gone through the court-like process so that thekankokuis effectively a final decision and it constitutes the vast majority of theFTC’s injunctions and punitive actions. The FTC announceskankokutoward firms,organizations, and/or individuals suspected of conducts in violation of the Anti-monopoly Law.19 A kankokuconsists of specific corrective measures that the saidfirms, organizations, and/or individuals should take in order to avoid further actionsby the FTC. After the amendment of the Anti-Monopoly Law in 1977, the FTChas also been able to impose a compensatory surcharge.

This suggests thatkankoku, although being an administrative procedure and nota legal punitive action, is the strongest and largely the final action against anti-monopoly conduct. In recent years, the FTC has investigated more than 200 caseseach year and roughly 15 to 20% of them have resulted inkankoku.If the FTC

18 See Masuda (1997) for a thorough review and an analysis of the FTC’s activities and their impact.The FTC’s regulatory procedures and their major changes are discussed in Murakami (1999).

19 The violations are largely classified as (1) conduct to monopolize markets, (2) collusive conduct,or (3) unfair trade practices (mostly in the context of vertical restraints). The FTC specifies the natureof the violation in eachkankoku.In this paper, we do not divide the FTC’s actions according to theseclassifications. Instead, we selected cases according to the following criteria: (1) the accused are eitherprivate manufacturers or distributors and (2) the nature of violation is either collusive pricing or retailprice maintenance.

Page 18: Determinants of Individual-Firm Markup in Japan: Market Concentration, Market Share, and FTC Regulations

INDIVIDUAL -FIRM MARKUP IN JAPAN 441

TABLE VIIList of FTC Policy Variables

Variable name Number of casesa Number of industriesb Expected signs

Kankoku injunctions [inj kt ] 53 31 —

−Regional dummy [regdkt ] 31 19 − (on inj k

t )−Manufacturer dummy [mfgdk

t ] 43 28 + (on inj kt )

Monitoring1 [monit1kt ] 106 19 —

Monitoring2 [monit2kt ] 391 42 —

Compulsory reports [comprkt ] 56 18 —

a Total number of industry-year observations in which each of policy measures are taken.b Total number of industries to which each policy measure is taken at least once during 1975–1994.

considers a case not strong enough to warrantkankokubut is still suspicious ofimportant unlawful conducts, the FTC may issue akeikoku, or achui (both canbe translated as “warning”).

The FTC has also other policy measures to monitor market conduct and pre-empt collusive behavior. Many of them are conducted as covert investigations andinformal guidelines given to the suspected firms. Many of these activities are notmade public and thus are not available for our purpose. Instead, we use three overtmeasures as they are reported in the annual white paper of the FTC each year.The first is the list of industries which FTC considers as being potentially monop-olistic and conducive to anti-competitive conduct. The list is thus considered anannouncement that the FTC is closely monitoring the conduct in these markets.The second is the list of products for which the FTC closely monitors possiblecollusive price increases. The last is the list of products for which the FTC askedspecific producers to report on their decision to raise their product prices. In thelast case, the FTC’s action is prompted by what appears to be a concerted priceincrease by major producers. For a list of FTC policy variables, see Table VII.

4.2. Econometric Analysis of the FTC’s Regulatory Measures

In the analysis that follows, we would like to test two hypotheses on the regu-latory actions taken by FTC.

1. Hypothesis 1: The FTC’s variety of administrative punitive actions are effec-tive in curtailing anti-competitive conduct, and thus are able to reduce the markupof firms in the targeted markets.

2. Hypothesis 2: The FTC’s actions are correctly aimed at industries and prod-ucts wherein substantive departure from competitive pricing is suspected and thatalso are judged economically more important.

To proceed, we use a set of dummy variables, defined below.20

20 Block et al.(1980) employ a similar approach to test the deterrent effect of prosecutions for pricefixing by the Department of Justice.

Page 19: Determinants of Individual-Firm Markup in Japan: Market Concentration, Market Share, and FTC Regulations

442 ARIGA, OHKUSA, AND NISHIMURA

TABLE VIIIPanel Regression on the FTC’s Policy Variables

Policy variables Impacts on markup

Injunctions1 inj kt ∗m f gdk

t — −0.010 [.32] −0.010 [.32]Injunctions2 inj k

t ∗ regdkt — 0.061 [.74] 0.60 [.74]

Monitoring1 monit1kt −0.032 [1.18] — 0.014 [.018]

Monitoring2 monit2kt −0.043 [3.02] — −0.005 [1.61]

Compulsory reports comprkt −0.083 [2.41] — −0.046 [1.68]

DWHTest 2.160 1.079 3.877

Note.Thet-values are in brackets. Regressions also include the variables in the base-linemodel.

FTCmk,t

{= 1 if FTC’s mth policy varible is applied to marketk at timet= 0 otherwise

}The effects of policy variables are represented by

logµikt = ε ik

0 +∑

m

εmFTCmk,t ,

which is substituted for logµikt in the base model regression given in (21) and (22).

The equation is estimated as a fixed effect model.Let us start with the effect of thekankokuvariable. We include in the regressions

two additional dummy variables. First,regionis the dummy variable that indicatesthat the suspected firms operate in specific local markets only. In this case, the ef-fect of the FTC’skankokuinjunctions may be somewhat limited. Second,mfgis thedummy variable that identifies cases in which injunctions are issued to manufac-turing firms. The dummy variable is zero if injunctions are issued to distributors.21

In most of the cases wherein distributors are prosecuted, the suspected violationsconcern mostly retail price maintenance. We expectkankokuto generate a strongerimpact if it is issued directly to manufacturers.

Table VIII shows our regression results. The effect ofkankokuon markup isessentially zero. In this and many other possible specifications, we could not rejectthe null hypothesis that the estimated coefficient ofkankokuis zero.

Next we move on to the other policy variables. Here themonitoring1 dummycorresponds to the industry and the year in which the FTC announced that theysuspected possible anti-competitive conduct, whereas themonitoring2 dummycorresponds to the product and the year that the FTC specified an industry asbeing suspect of collusive pricing behavior. Thecomplusory reportdummy is theproduct and the year for which the FTC asked specific producers to report on theirprice decision. Formonitoring1, we obtained the expected signs but they were not

21 We exclude all cases in which the case does not involve manufacturing industries, most notablythe bid-rigging of local construction companies.

Page 20: Determinants of Individual-Firm Markup in Japan: Market Concentration, Market Share, and FTC Regulations

INDIVIDUAL -FIRM MARKUP IN JAPAN 443

statistically significant. For themonitoring2 andcompulsory reportvariables, theestimated coefficients are large negative values and both (highly or marginally)are statistically significant.

Our results suggest that the latter two are more effective since they are directedmore narrowly toward specific conduct in pricing. Unlike the other variables,compulsory reportis prompted by an actual price increase that the FTC suspectsis collusive. As such, the FTC ordered individual firms to submit reports on theirdecision to increase price. These variables have large impacts: in one regression theeffect ofmonitoring2 is to reduce markups by 4%, whereas the effect ofcompulsoryreport is roughly twice that size.

To sum up, the effects on markup of the FTC’s policy measures differ consider-ably according to whether formal injunction action,kankoku, or other less formalmeasures were used. The effects of the latter measures are of the expected signsand two of the three variables are statistically significant, whereas we basicallyfind no effect of the first variable on the markup.

4.2.1. Potential endogeneity and other econometric issues.To the extent thatthe FTC’s regulatory actions are prompted by suspected unlawful firms’ conductand/or oligopolistic market structure, policy variables may be endogenous, creatingestimation biases in the regressions results in Table VIII. To assess the possibleendogeneity of policy variables, we conductedDWH test.22 The figures shown inthe bottom of each column show the test statistics. Under the null hypothesis that thepolicy variables are not correlated with the error term in the regression, this statisticis distributed according to aχ2 distribution with a degree of freedom equal to thenumber of policy variables in each regression. As is evident in this table, we cannotreject the null hypothesis in any of the regressions with a conventional confidencelevel. To reject a null at the 5% confidence level, the statistic must be larger than5.99, 7.81, and 9.48, for two, three, and four policy variables, respectively. Basedupon the test results, we therefore rule out the possibility that policy variables areendogenous.

Another potential problem with the regressions in Table VIII is possible mis-specification, in particular of the timing of policy actions and their impacts. Forexample, For example, our model assumes that the effect of a policy does notdepend upon duration. it may be argued, however, that the effect of a policy actionis once-and-for-all so that the policy impact is concentrated in the first year that anaction is taken toward a particular industry. Alternatively, it is conceivable also thatthe same policy has different effects depending upon the duration of the policy.Moreover, it is possible that the precise timing of an action and its impact maybe misrepresented in the current model. For example, suppose a policy measure isenacted in December of a particular year. The impact, if any, of the policy is likelyto be recordewd in the observation for the next year.23

22 See, for example, Davidson and MacKinnon (1993) for detailed explanations of the procedure.23 It is also possible that the actual impact takes place earlier than the recorded timing if the FTC

informs the targetd firms or industries of the policy action before the announcement.

Page 21: Determinants of Individual-Firm Markup in Japan: Market Concentration, Market Share, and FTC Regulations

444 ARIGA, OHKUSA, AND NISHIMURA

TABLE IXCharacteristics of Targeted Industries

Policy measures Market sizea Cr3

All samples 504 0.756Kankoku> 0 528 0.737Monitoring1> 0 646 0.791Monitoring2> 0 700 0.812Comp. report> 0 1120 0.841

a 1994 billion yen.

To tackle these timing issues, we tested a variety of specifications, including(1) polynominally distributed patterns of policy impacts geometrically decliningas the policy is applied continuously over many years, (2) replacing current policydummies with those which equal unity only in the first year of policy action,(3) replacing current policy dummies with those which equal unity in the yearimmediately before the first year. We found no substantial differences between thebase model and any of these alternative specifications.24 In particular, we find thatthekankokupolicy variable is not significant (and often has the wrong sign) andalso thatmonitoring2 andcompulsory reportshave the correct signs and are oftensignificant at the 5% level.

4.2.2. Hitting the right targets? Our results so far show that the FTC haslargely done its job well at least in monitoring and guidance activities: they havesucceeded in substantially reducing the level of markup. However, our findingsmay substantially fall short of providing sufficient supporting evidence that theFTC has done its jobefficiently.Our foremost concern in evaluating the FTC’strack record is the fact that the FTC in Japan is seriously under-manned and itshuman, political, and administrative resources are severely limited.25 Given thesevere resource constraint, our next question is whether or not the FTC has chosenthe right (important) targets in order to best utilize its resources.

Table IX shows the characteristics of targeted industries, i.e., those for whichthe FTC’s policy actions have been taken. We show the industry size and thethree-firm concentration measure. Several observations follow immediately fromthe table. First of all,kankoku, or administrative injunctions, are not aimed ateither particularly important or highly concentrated industries. As a matter offact, injunctions have been issued repeatedly in the past against a relatively smallnumber of industries. We suspect that the FTC’s track record in injunctions reflectsprimarily their judgement regarding the seriousness of the suspected violation, andnot necessarily the magnitude of possible welfare or efficiency losses incurred by

24 These regression results are available from the authors upon request.25 The FTC is very small: its 1997 budget was 5.6 billion yen, and it employed roughly 550 people,

including 125 located in seven branch offices.

Page 22: Determinants of Individual-Firm Markup in Japan: Market Concentration, Market Share, and FTC Regulations

INDIVIDUAL -FIRM MARKUP IN JAPAN 445

these suspected violations.26 As a matter of fact, as a quasi-judiciary branch of thegovernment, it is unlikely that the FTC has much discretionary power to liberallyapply injunction measures toward selected industries. Instead, the FTC is boundto follow the due process of investigating and sanctioning violations of the law.

In contrast to the track record forkankoku, Table IX shows that the FTC hasconcentrated its activities in larger and highly concentrated industries for other, lessformal and weaker actions. This also makes sense because the FTC can apply thesepolicy measures toward industries and firms that they suspect as being oligopolisticor collusive. Naturally, the FTC focused their resources on monitoring and givingwarning signals to larger and more highly concentrated industries. These findingsare also consistent with our regression results on the impact of these measureson the markup. FTC policy has the strongest impact on industry markups whenthe FTC applies non-punitive actions that are designed to send warning messagestoward targeted industries. On the other hand, injunction measures do not seem tohave a significant impact upon the markup.

4.2.3. Discussion. Insofar as we evaluate FTC performance purely in terms ofefficiency gains, our estimation suggests that the FTC’s scarce resources should beallocated more to its monitoring/administrative guidance activities than to formalinvestigations leading tokankoku. There are many reasons to be cautious, however,in assessing the relative merits of the FTC’s activities. First of all, just like any otherlegal enforcement body, the FTC’s foremost objective is to deter law violations,not to maximize efficiency gain.

Punitive actions are more powerful measures than other activities such as mon-itoring or guidance. As a matter of fact, an argument can be made that non-compulsory measures such as guidance can be effective only under the imminentthreat of these actions. The deterrent effect of punitive measures may not be ad-equately estimated in our framework. Closer to home, we need also to remindourselves that our estimations rely exclusively on samples of large firms.Kankokuand other formal actions might well have had large impacts upon the pricing policyof smaller firms and non-firm entities. Last, but not least, there exist potential haz-ards in the FTC’s less formal activities as they are not reported officially, mainly tosecure the anonymity of suspected firms and individuals. Inevitably, such activitieslack transparency and may well invite a variety of moral hazards.27

In spite of these reservations, there are also arguments in favor of strengtheningthe FTC’s monitoring and guidance activities. First of all, we expect that evenif these activities are not punitive in nature, disclosing the nature of suspected

26 Recall thatkankokuis practically the strongest measure available to the FTC. The FTC must beable to substantiate its claim and justify its injunction decisions, first at the FTC’s quasi-court andpossibly also at a (real) court, as the accused can appeal to the court. (In the past, however, the FTChas had close to a 100% winning record in court).

27 One Diet member and a firm executive were prosecuted for bribing and receiving bribes in anattempt to influence an FTC investigation. The Diet member was later impeached by the Diet and losthis seat.

Page 23: Determinants of Individual-Firm Markup in Japan: Market Concentration, Market Share, and FTC Regulations

446 ARIGA, OHKUSA, AND NISHIMURA

conduct and the identity of suspected firms can be damaging enough for firms. TheFTC has been conservative, however, regarding information disclosure, in fear ofcomplaints from firms. Judging from the significant effect of thecompusory reportsvariable, we expect a substantial impact from fuller disclosure of information andjudgement on the market conduct of individual firms and markets.

Moreover, there exists concern over the effectiveness of punitive actions. Beforethe 1977 amendment, there was no surcharge penalty even if the prosecuted partyfailed to comply the injunctions. The 1977 amendment introduced a surchargedesigned basically to absorb estimated extra profits earned by collusive actions.Many suspect that not only that the surcharge is too small28 but also that the rulegoverning the amount of the surcharge is ill-fit. In particular, the rule does nottake into consideration the damage incurred by consumers and other parties in themarket. The FTC is given no discretionary power to adjust the surcharge accordingto these damages. Given the fact thatclass action suitsby private individuals orfirms are very difficult and, so far, close to non-existent in Japan, it seems possiblethat additional resources spent on policing law violations may not be very effective.

5. CONCLUDING REMARKS

In this paper, we employed an estimation procedure that relies only upon infor-mation on production technology to obtain an estimate of the firm-level markup.Our results indicate that most of major manufacturing firms in Japan are able toelevate prices substantially above their marginal costs. We also found substantialpositive impact on markup not only from market concentration and product dif-ferentiations but also by firms’ own market shares. Given the diversity of marketsand products covered in the sample, we believe that our analysis offers a ten-tative but reasonable bird’s eye view of market power and its consequences forpricing in the Japanese manufacturing industries. The view that we offer is closeto conventional wisdom. The relation between market concentration and markupis marginally significant and quantitatively small, suggesting clearly that thereexists no clear-cut strong causal relation linking concentration to markups. Thefinding is not new, but we should note that our results are based upon the esti-mate of individual firm-level markup, in contrast to most of the previous studiesusing industry-level cross-sectional data. We also find it significant that the own-market share does have a statistically significant and sizable positive impact onmarkup.

To our knowledge, ours is the first paper to systematically investigate the effec-tiveness of the FTC’s policy measures on industrial pricing. The result is mixed: we

28 According to Masuda (1996), the total surcharge for 1977 and 1991 was 23.5 billion yen for 1190firms and other entities, or 12 million yen per entity. On the other hand, according to rumors in theindustry circle,kankokuis still a powerful weapon because executives of firms receivingkankokuwillbe deleted from the list of candidates for imperial decorations.

Page 24: Determinants of Individual-Firm Markup in Japan: Market Concentration, Market Share, and FTC Regulations

INDIVIDUAL -FIRM MARKUP IN JAPAN 447

find significant impact by the non-punitive measures taken by the FTC, whereasits strongest bullet, injunctions, do not appreciably influence the markups of oursample firms.

Due primarily to data limitations, we leave many interesting and importantissues for future research. Among other concerns, we need to collect data onchanges in market size, concentrations, and shares in order to conduct full-fledgedpanel regressions. Expanding the scope of the research to smaller firms also hasobvious merits, particularly for estimating the impact of FTC injunctions in a moresatisfactory manner.29

APPENDIX A: DERIVATION OF (10)

Utilizing definitions (5) and (11), we have

r ikt K ik

t + wikt Lik

t + zikt Jik

t =(∑

k

pikt Qik

t

)ωik

t

([αik

K

]t+ [αik

L

]t+ [αik

J

]t

).

(A.1)Substituting (A.1) into (12) we have

µit

∑k

ωikt

([αik

K

]t+ [αik

L

]t+ [αik

J

]t

) =∑k

ωikt µ

ikt

([αik

K

]t+ [αik

L

]t+ [αik

J

]t

).

(A.2)From definition (13), we get

[αi

K

]t +

[αi

L

]t +

[αi

J

]t =

∑k

ωikt

([αik

K

]t +

[αik

L

]t +

[αik

J

]t

). (A.3)

Substituting (A.3) into (A.2) we have

µit

([αi

K

]t +

[αi

L

]t +

[αi

J

]t

) =∑k

ωikt µ

ikt

([αik

K

]t +

[αik

L

]t +

[αik

J

]t

). (A.4)

Substituting (9) into (A.4), we obtain

µit

([αi

K

]t+ [αi

L

]t+ [αi

J

]t

) ≈∑k

ωikt

[1+ γ ik

(Qi N

t

Qit

)− δik

(1Si

t

Sit

)2],

which immediately implies (10) since∑

k ωikt = 1.

29 A more satisfactory approach would be to estimate the “supply” schedule of the FTC’s policyactions jointly with its impact, as pointed out correctly by Bresnahan and Yanagawa in their comments.

Page 25: Determinants of Individual-Firm Markup in Japan: Market Concentration, Market Share, and FTC Regulations

448 ARIGA, OHKUSA, AND NISHIMURA

APPENDIX B: DATA CONSTRUCTION

This appendix briefly explains the procedure used to obtain the variables in theempirical analysis of the text.

Capital stock K. The firm’s capital stock includes equipment and structure,but excludes land holdings. We excluded land because of the huge discrepancy be-tween the market price and the book value of the land. In Japan, in filing the officialbalance sheet, public firms can use either the acquisition value or the replacementcost as the book value for most fixed assets. Virtually all firms use the original ac-quisition value, and there is no established way to correct for the discrepancy. Thisdiscrepancy also exists for equipment and structures. To minimize the problem,we employed the following procedure. First, the starting-year stock is constructedby deflating the book value of the capital stock in that year by the Wholesale PriceIndex for Capital Goods of the Bank of Japan, Keizai Toukei Nenpo. Starting fromthis year, the subsequent series of the capital stock is constructed by using theperpetual inventory method in which the Wholesale Price Index of capital goodsis used as a deflator of investment.

Current output Q and normal output Q∗. The real sales of the firm are used asa proxy for the firm’s output. They are obtained by deflating its total sales by theOverall Wholesale Price Index. The normal output is then the trend for real salesof the firm. It is simply a fitted value of the regression of real sales on time and aconstant, in which no higher order terms are included.

Rental price r of the capital stock.We use the following formula to correct forcorporate tax, depreciation allowance, and capital gains from inflation in capitalgoods:

r = (ρ + dep.− πeK

)× 1− k− τd

1− τ pK .

Here,r is the firm’s rental price of capital,ρ is the firm’s real cost of funds, dep.is the economic rate of depreciation,πe

K is the expected rate of inflation of capitalgoods,pK is the real price of capital goods,k is the effective rate of investmenttax credit (=0 in Japan),τ is the effective rate of corporate income tax, andd isthe present discounted value of tax deductions for depreciation.

The effective corporate tax rate is computed as

τ = (u+ v)(1+ ρ)/(1+ ρ + v) whereu = uc(1+ u`),

in which u is the overall corporate income tax rate,uc is the national corporateincome tax rate,u` is the local corporate tax surcharge rate, andv is the enterprisetax rate (=12%).

We used the following data for each of the variables above:ρ is the prime banklending rate;pK is the capital goods price index (a part of wholesale price indices);

Page 26: Determinants of Individual-Firm Markup in Japan: Market Concentration, Market Share, and FTC Regulations

INDIVIDUAL -FIRM MARKUP IN JAPAN 449

dep.,d, uc, andu` are data taken from an unpublished work by Nishimura andShirai. The last group of data are annual and are for dep. andd. Nishimura andShirai estimate these values for each industry. All of these data are available fromthe authors upon request.

Labor’s shareαL . The ratio of total employment compensation to the totalsales.

Capital’s shareαK . The ratio ofr K to the total sales.

Raw material’s share. The ratio is computed as the sum of total purchases inmaterials, intermediate goods, and utility expenses minus the increase in inventoryof materials and intermediate goods.

Cash flow/asset ratio CFAS.This ratio is the sum of the operating profits,the depreciation allowance adjusted for investment outlay on tangible fixed assetsexcept for land, divided by the value of the firm’s asset.

REFERENCES

Baba, N. (1995). On the determinants of home–abroad price differences,Kinyu Kenkyu14, 71–97. [InJapanese]

Bils, M. (1987). Pricing in a customer market,Quart. J. Econ.104, 699–718.

Block, M. K., Nold, K. F. C., and Sidak, J. G. (1980). The cost of antitrust enforcement,J. Polit. Econ.89(3), 429–445.

Bresnahan, T. (1989). Empirical studies of industries with market power,in “Handbook of Indus-trial Economics, Vol. 2” (R. Schmalensee and R. D. Willig, Eds.), pp. 1012–1057. North-Holland,Amsterdam.

Davidson, R., and MacKinnon, J. G. (1993). “Estimation and Inference in Econometrics,” OxfordUniversity Press, Oxford.

Fair Trade Commission (1974–1997).Annual Reports (1974–97).

Gottfries, N. (1991). Customer markets, credit market imperfection, and real price rigidity,Economica58, 317–323.

Green, E. J., and Porter, R. H. (1984). Non-cooperative collusion under imperfect price information,Econometrica52, 87–100.

Masuda, T. (1996). “Dokusen Kinshi Hou no Keizai Bunseki” (Economic Analysis of Anti-MonopolyLaw in Japan). Taga, Tokyo.

Murakami, M. (1999). Dokusen kinshi hou no sekou tetuzuki oyobi sikkou (Enforcement of anti-monopoly law),in “Nihon no Kyousou Seisaku” (Competition Policy in Japan) (Goto and Suzumura,Eds.). University of Tokyo Press, Tokyo.

Nishimura, K. G. (1989). Customer markets and price sensitivity,Economica56, 187–196.

Nishimura, K. G., Ohkusa, Y., and Ariga, K. (1999). Estimating markup over marginal cost and returnsto scale at the firm level: New approach applied to a panel of japanese firms,Int. J. Ind. Organ.,17(9), 1077–1111.

Nishimura, K. G., and Inoue, A. (in press). Labour’s share in Japanese manufacturing 1960–1990:“Dual structure” and imperfect competition,in “The Distribution of Income and Wealth in Japan”(T. Ishikawa, Ed.). Oxford University Press, Oxford.

Page 27: Determinants of Individual-Firm Markup in Japan: Market Concentration, Market Share, and FTC Regulations

450 ARIGA, OHKUSA, AND NISHIMURA

Odagiri, H., and Yamashita, T. (1987). “Price markups, market structure, and business fluctuations inJapanese manufacturing industries,J. Ind. Econ.35, 317–331.

Rotemberg, J. J., and Saloner, G. (1986). A supergame theoretic model of price wars during booms,Amer. Econ. Rev.76, 390–407.

Rotemberg, J. J., and Woodford, M. (1991). Markups and the business cycle,in “NBER Macroeco-nomics Annual,” pp. 63–129.

Snyder, E. A. (1990). The economic effects of antitrust laws,J. Law Econ.33(2), 439–462.

Shinjo, K. (1977). Business pricing policies and inflation: The Japanese case,Rev. Econ. Statist.59,447–455.

Roberts, M. J., and Supina, D. (1997). Output price and markup dispersion in micro data: The roles ofproducer heterogeneity and noise, NBER Working Paper 6075.

Yano Research Institute (1995). “Nihon Market Share Jiten” (Yearbook on Market Share in Japan).