Efficiency versus Market Structure: The Case of the Dutch Independent Automotive Industry

Efficiency versus Market Structure:

The Case of the Dutch Independent Automotive Industry

Freek Kamphuis

Master’s Thesis in Finance Date: 30-03-2013

Supervisor: prof. dr. C.L.M. Hermes

Author: Freek Kamphuis

2

Efficiency versus Market Structure:

The Case of the Dutch Independent Automotive Industry

Freek Kamphuis

ABSTRACT

This study provides insight in the mechanisms behind financial performance in the Dutch independent automotive industry. We moreover contribute to the debate whether market structure or firm level efficiencies drive this particular industry’s performance. We use a dataset covering 691 Rabobank financed Dutch independent automotive firms in the period of 2007 to 2011. These automotive firms compete on a local rather than a national basis. Therefore, we use market concentration proxies which are disaggregated to the level of the firm’s municipality. Moreover, we derive each firm’s technical efficiency from a stochastic Cobb-Douglass production function. Our findings provide support for the efficiency theory stating firm level efficiencies determine firm performance and observed market power effects are the result of efficiency discrepancies.

Keywords: Automotive industry, Efficiency, Market structure, SFA

JEL classification: L10, L80, D40

*Student at the University of Groningen, Faculty of Economics and Business, the Netherlands. I am grateful to

3

1. Introduction

The Dutch automotive industry faces difficult times. Automobile dealers experience a shift in consumer demand to small and environmentally friendly automobiles. Unfortunately for the dealer, the profit margins on these newly demanded automobiles are smaller than those previously enjoyed on larger vehicles. As a consequence, dealers now face pressure to keep their turnover and profits at healthy values. Moreover, this gap in income from sales is getting harder to fill with income from after sales, consisting of income from automobile repair and service. This is not only because of the ever increasing quality of automobiles, accounting for less automobile breakdowns and service needs, but also because of the modern consumer who drives less than before and attaches less priority to vehicle service in times of economic hardship.

The described phenomenon raises the important question what an automotive firm can do to protect itself from shrinking turnover and profits. Moreover, is it possible for an individual automotive firm to impact its performance in terms of profits in the first place? Or is its performance mainly exogenous determined by competition or the market’s structure? Answering this requires insight in the mechanisms behind financial performance of automotive firms. This is the topic of this thesis. More specific, we focus on the Dutch independent automotive industry and contribute to the classic debate whether market structure or firm level efficiencies drive this particular industry’s performance1.

We hereby contrast the traditional to the revisionist view (Amato and Wilder, 1990; Bain, 1951; Ekelund, Ford, e.a., 2000; Hirschey and Wichern, 1984; Mann, 1966; Peltzman, 1977; Schmalensee, 1985; Shepherd, 1972). The traditional view states market structure dictates the firm’s amount of market power, which in turn determines its performance in terms of profits. Contrary, the revisionist view rejects this reasoning and states any observed market power effects on the firm’s performance are the result of firm level efficiency discrepancies.

The benefit of this study lies in our detailed data regarding individual automotive companies and their competitive environment. This data allows us to use and compare the effect of multiple measures of market concentration on the firm’s performance. Previous studies experienced limited access to data and suggest further research to use richer data, preferably in a panel data set up (Amato and Wilder, 1990; Garza-Garcia, 2012). Amato and Wilder (1990) point at the weakness of commonly used

1 _{This thesis is part of a research internship at Rabobank Nederland in cooperation with BOVAG. Rabobank}

4

proxies for firm level efficiency. They argue the ideal measure for efficiency is the firm’s cost effectiveness in generating output. However, according to them, available data is usually not sufficiently detailed and disaggregated to adequately proxy for efficiency. Our data, however, contains inputs and outputs which are sufficiently disaggregated to the level of the firm to estimate the company’s technical efficiency using a stochastic Cobb-Douglas production function (Battese and Coelli, 1993; Bogetoft and Otto, 2011; Cobb and Douglas, 1928). Furthermore, we use a panel data methodology and consider both local competition measures and the Lerner index to proxy for the concentration variable. The use of these multiple proxies increases this study’s outcome validity.

The independent automotive industry differs from commonly researched industries in the sense that competition is on a local instead of a national basis. Independent automotive companies provide sales and repair services to consumers, whereby they are not formally restricted to specific automobile brands. Therefore, a single independent automotive firm may sell and repair the same type of cars as its competitors. This causes its products and services to be relatively homogeneous. Furthermore, the provided goods and services require the consumer to physically visit the firm. These two characteristics determine the important inclusion of firm location in measuring competition in this particular industry. This is because firms located in different areas of the Netherlands face different competitive environments since competition is more or less limited to an area. We use spatial proxy measures to provide interesting insights in the effect of the firm’s close market environment on its performance (Watson, 2009). Moreover, the local aspect allows us to include for instance the number of competitors in the firm’s environment as market structure component in a model. This variable varies per firm in our local setup, whereas it would be fixed for firms on a national basis. This variation is preferable, since it allows us to investigate these variables in a panel data setup using period fixed effects.

5

This thesis continues as follows. In section two we provide a literature overview about the topic. Thereafter, we formulate the research question and provide hypotheses in section three. Section four describes the methodology and section five the data. Furthermore, section six shows the results and section seven provides a summary and conclusion. Finally, we list limitations and suggestions for future research in section eight.

2. Literature Overview

We research firm performance in the Dutch independent automotive industry. The issue of firm performance in a specific industry is widely addressed in the literature and may be studied from either the traditional or revisionist view (Amato and Wilder, 1990; Bain, 1951; Ekelund, Ford, e.a., 2000; Hirschey and Wichern, 1984; Mann, 1966; Peltzman, 1977; Schmalensee, 1985; Shepherd, 1972). The first view focuses on industry concentration and tacit collusion and states profits within an industry do not differ. The later view acknowledges intra-industry profit variation and focuses on firm-level discrepancies in explaining this phenomenon. We describe the traditional view in section 2.1 and the revisionist view in section 2.2. Thereafter, we provide a common grouping of the main theories leading from both views in section 2.3 and list findings in previous empirical research in section 2.4.

2.1 Traditional view

The traditional view emphasizes the role of collusion within an industry in the determination of firm profit (Bain, 1951). This view states large market concentration within an industry may lead to tacit collusion. This collusion of firms brings the industry closer to monopoly causing prices and profits to be large and approximately indifferent between the industries’ firms. We describe the concept of collusion using the well-known models from Cournot (1838) and Bertrand (1883) in section 2.1.1. Thereafter, we introduce the structure conduct performance (SCP) theory, which leads from the traditional view and is the basis for Porter’s famous five forces framework, in section 2.1.2.

2.1.1 The concept of collusion: Cournot and Bertrand

6

point where marginal revenue equals marginal cost for its predicted residual demand. In Cournot equilibrium, each firm correctly predicts the amount its competitor produces and thus maximizes its own profit given what its competitor does (Nash, 1951). The concept of collusion describes the situation where these two firms are not competing but work together in setting quantities. Now they maximize total profit by determining the optimal total production quantity. They do this again by setting marginal revenue equal to marginal cost, but now for the total demand instead of the residual demand. When both firms equally share the production of this optimal total quantity, the produced quantity per firm will be less than in Cournot equilibrium while each firm’s individual profit will be higher.

The same reasoning holds in the Bertrand model. This model is similar to the Cournot model except for the fact that firms set prices instead of quantities. In this model, firms choose to set their price equal to marginal cost and earn zero profit. However, when firms collude they set a price equal to the monopolist’s price to maximize total profit. In this case they are better off, since this profit is of positive form.

Note however, collusion is generally forbidden by antitrust laws since it harms consumer’s utility through higher prices and less supply. Therefore, any collusion in the market is often tacit. Furthermore the described concept is only feasible and attractive for firms in high concentrated markets. This is because collusion in Cournot markets with many firms may induce individual firms to cheat on the production agreement and maximize its own profit instead of total profit. This cheating behavior overrides the total optimal production quantity and harms the other firms through lower prices making the collusion agreement or cartel instable. Similar, colluding firms in large Bertrand markets may cheat and maximize their own profit by setting a price slightly below the price agreed upon. This shifts total demand to the cheating firm, which causes price reactions from the other firms again making the collusion instable.

2.1.2 The structure conduct performance theory and Porter’s five forces framework

7

The SCP theory is an important basis of Porter’s (1980) famous five forces framework describing the industry’s attractiveness, a term which is a surrogate for the industry’s profitability (McGee, Thomas e.a., 2005). More specific, Porter’s framework describes the industry’s attractiveness by the relative strength of the forces supplier power, buyer power, threat of new entrants and threat of substitute products. Together, these forces determine the intensity of competition or the number of competitors in the market, which is also considered as a force. The supplier and buyer power are measured in terms of bargaining power. For instance, if the market has few suppliers, purchasing firms in the market face little bargaining power and thus have little opportunity to avoid premium purchase prices. The same reasoning holds for the force buyer power. That is, the greater the concentration of buyers, the greater their bargaining power in buying the selling firm’s products. Hence, large (small) supplier and buyer power decreases (increases) the industry’s intension of competition. Furthermore, the threat of new entrants depends on the presence of entry barriers. These barriers represent the costs for potential entrants to start up a competitive business. If the entrant’s per unit costs are higher than the incumbent’s, the entrant faces an entry barrier which may withhold him for setting up the new business. Hence, high (low) entry barriers decreases (increases) the industry’s intensity of competition. Furthermore, the threat of substitutes is the threat of the supplied product being replaced by a substitute, making it obsolete. Hence, a high (low) threat of substitutes decreases (increases) the industry’s intensity of competition. Remember, the relative strength of these described forces together determine the industry’s attractiveness or profitability. In terms of the SCP theory, the (tacit) collusion resulting from the market’s degree of concentration or competition, determines the firm’s price and quantity setting conduct. This conduct, in turn, is a key variable in determining profits.

2.2 Revisionist view

In contradiction to the traditional view described in the previous section, the revisionist view argues firms within the same industry may receive different profits because of firm-level efficiency discrepancies (Hirschey and Wichern, 1984; Shepherd, 1972; Peltzman, 1977). This view thus states performance is dependent on internal rather than external factors. This is known as the resource based view, which we describe in section 2.2.1.

8

firms may start producing more than competitors2. This causes these efficient firms to grow, implying higher industry concentration. This example shows that high firm concentration does not influence profits on its own, but may be merely the result of firm-level efficiencies. Furthermore, Peltzman mentions that efficiency is not free and needs investment. If the firm’s expansion costs are low, it is likely to invest in efficiency. Because of the low expansion costs, the efficiency may easily be exploited on a large scale to deliver the most benefit. This again indirectly increases the industry’s concentration through efficiency.

2.2.1 Resource based view

The main difference of the revisionist view compared to the traditional view is that firm performance stems from internal rather than external factors (McGee, Thomas e.a., 2005). This is also referred to as the resource based view (RBV) which describes an organization which has, develops or acquires a range of competences from resources. These competences enable the firm to gain a competitive advantage in the market and earn a larger profit than competitors.

This RBV is established by Prahalad and Hamel (1990) who define the company’s core competences as the collective learning in the organization about how to coordinate diverse production skills and integrate new technologies. They opt the organization’s strategy should be developed from an organizational perspective instead of an industry and market’s perspective. This furthermore involves an active internal focus on communication, involvement and commitment to identify and develop the company’s core competences. These core competences, in turn, are the key to the company’s competitive advantages. Note this active internal focus yields different firm strategies than the rather passive external approach.

Porter’s (1990) diamond of national competitive advantage provides a more hybrid view on the creation of competitive advantages. That is, he highlights both the internal and external factors yielding the firm’s competitive advantage. The diamond framework points at the interdependency of both national market conditions and the firms’ resources and skills in the development of (global) competitive advantages. Moreover, Porter proposes a national playing field in which the organization operates. This field consists of the four interdependent attributes factor conditions, demand conditions, related and supporting industries and firm strategy, structure and rivalry. Porter argues the factor conditions represent not only the available amount of capital, labor and infrastructure, but moreover the human skills and knowledge about the efficient use of these classic production factors. He furthermore argues these skills and knowledge may be created by investment and are most likely to develop in a competitive advantage when a firm continually improves upon them. The nature of home

9

market demand may furthermore encourage the firm to innovate. That is, the more sophisticated and demanding the home market buyers, the more pressure lies on the firm to meet these standards forcing them to become more innovative, efficient and competitive. Related internationally competitive industries may encourage the firm’s competitiveness through efficient input delivery and exchange of valuable information regarding innovation and efficiency. Furthermore, the attribute firm strategy, structure and rivalry refers to the firm’s national environment determining the nature of organization and management. This attribute also captures human motivation to learn and expand skills. This motivation yields knowledge, which may in turn creates a more efficient and competitive advantageous business. Porter furthermore emphasizes the importance of domestic rivalry in the creation of dynamic efficiency. He states domestic rivalry goes beyond competition and is moreover of personal form. This personal form is referred to as “bragging” or just showing competitors that improvement is possible.

2.3 Theories

The traditional- and revisionist view offer different theories about the determinants of firm performance. Berger (1995) provides a clear grouping of these into market power and efficiency theories.

10

As efficiency theories, Berger (1995) distinguishes scale efficiency (ESS) and X-efficiency (ESX). The first states abnormal profits may be received through large scale production being more efficient. These economies of scale are more efficient, since per unit cost of production decreases with the amount of output when constant costs are present (Pindyck and Rubinfeld, 2005). The X-efficiency theory states other efficiency sources such as superior management or technologies yield lower costs and therefore higher profits.

2.4 Empirical studies

Empirical studies testing the efficiency- versus market power theories mainly focus on the banking industry (Berger, 1995; Berger and Hannan, 1998; Garza-Garcia, 2012; Williams, 2012). This is because researchers are interested in the motives behind recent mergers and acquisitions in this industry. They try to define these motives by hypothesizing whether market power- or efficiency gains are the beneficial drivers behind these activities.

Berger (1995) provides one of the leading papers on this topic by testing the SCP, RMP, scale- and X-efficiency theories using thirty banking cross-sections over a ten year period. He finds X-X-efficiency to be associated with higher profits, although he finds the relationship between efficiency and market structure to be weak. That is, the sample supports the X-efficiency theory stating more efficiency yields higher profits, however this efficiency has only a small impact on the size of market shares or degree of industry concentration. Therefore, he finds efficiency only weakly accounting for any observed relationships between banking concentration and profitability. Furthermore, the inclusion of market shares in his model shows some support for the RMP theory stating large firms are able to exercise market power through pricing. He concludes, however, that due to limited found explanatory power neither the market power nor the efficiency theory is sufficient in explaining banking profits.

11

Recent research on mergers and acquisitions in the Mexican banking sector by Garza-Garcia (2012) finds, just like Berger (1995) significant support for the RMP theory and again no support for the SCP theory or efficiency theories. Moreover, Garza-Garcia finds banking profits to be sustainable over time, indicating the Mexican banking industry is not very competitive.

Existing research testing the concentration or firm-level efficiency debate mainly uses data on multiple industries. Most models contain accounting based firm efficiency variables to measure the impact of firm efficiency and proxy variables to capture the industry’s concentration. Different variables in previous studies proxy the firm’s efficiency, such as R&D expenditures, costs, management quality, firm growth and financial ratios. A more comprehensive measure of firm level efficiency comes from more recent estimation techniques such as data envelopment analysis (DEA) or stochastic frontier analysis (SFA) (Battese and Coelli, 1993; Bogetoft and Otto, 2011; Garza-Garcia, 2012). These two are input-output analyses determining firm efficiency by comparisons to best practice firms. Furthermore, different proxies in the literature capture the industry’s concentration level, for instance the Lerner index, Herfindahl-Hirschmann index, concentration ratio or industry specific dummy variables3 (Lerner, 1934; Sleuwaegen and Dehandschutter, 1986; Williams, 2012).

3. Research Question and Hypotheses

We aim to gain more insight in the mechanisms behind financial performance in the Dutch independent automotive industry by focusing on the market structure versus firm level efficiency paradigm. The research question is

Is performance in the Dutch independent automotive industry determined by firm efficiencies or market structure?

This question is answered by sub questions regarding the influence of firm level efficiencies and market structure on the firm’s performance. The independent automotive firm’s market is mainly limited to a relatively small local area. We expect variations in local market structures ranging from full competition to more monopolistic forms. We furthermore expect all automotive firms, regardless their faced market structure, to benefit from efficiency. That is, more efficiency benefits the firm through reduced costs or a larger sold quantity. However, we expect performance in more monopolistic local markets to be relatively independent on efficiency. These automotive firms having market power may afford themselves to conduct inefficient behavior and simply raise prices to receive their desired profit. However, we expect monopolistic power in this industry to be limited to some

3

12

extent. This is because absurdly high monopoly prices may induce the consumer to travel to another local market4. Therefore, based on this thinking, we expect efficiency effects to dominate this industry. The first sub question questions the impact of efficiency on performance

What is the influence of firm level efficiencies on firm financial performance in the Dutch independent automotive industry?

The traditional view states firm efficiencies have no positive effect on the firm’s profits, whereas the revisionist view states it has a positive effect. Therefore, we hypothesize

𝐻01: Firm efficiencies have no positive effect on firm financial performance

𝐻𝑎1: Firm efficiencies have a positive effect on firm financial performance

Next, we question the influence of the firm’s market structure by questioning

What is the impact of the Dutch independent automotive industry’s concentration on firm performance?

The traditional view states the industry’s concentration is positively related to the firm’s performance, whereas the revisionist view states concentration on its own has no impact on performance. Therefore, we hypothesize

𝐻02: The industry’s concentration has no positive effect on firm financial performance

𝐻𝑎2: The industry’s concentration has a positive impact on firm financial performance

4. Methodology

We research whether firm financial performance in the independent automotive industry stems from market or firm level efficiency factors. We use an econometric model in a panel data setup in which we determine the firm’s efficiency via a stochastic frontier analysis. The next sections describe our methodology and model with its appurtenant dependent and independent variables.

4.1 Model

Since the data comprises both cross-sectional and time series dimensions, we use panel data estimation techniques (Brooks, 2011). The benefit of panel data is we may reduce omitted variable bias through

13

use of fixed effects. We adopt a similar approach as Garza-Garcia (2012) and estimate a model where we regress the firm’s financial performance against its past financial performance, efficiency, local market concentration and control variables as in

𝜋𝑖,𝑡 = 𝛼 + 𝛽1𝜋𝑖,𝑡−1+ 𝛽2𝐶𝑜𝑛𝑐𝑒𝑛𝑡𝑟𝑎𝑡𝑖𝑜𝑛𝑖,𝑡+ 𝛽3𝐸𝑓𝑓𝑖𝑐𝑖𝑒𝑛𝑐𝑦𝑖,𝑡+ 𝛽4𝐶𝑜𝑛𝑡𝑟𝑜𝑙𝑠𝑖,𝑡+ 𝜇𝑖+ 𝜆𝑡+ 𝑣𝑖,𝑡 (1) where i refers to a specific company ranging from 1 to n = 691 and t to a specific year in the period 2007 to 2011. Term 𝜇𝑖 represents a firm fixed effect allowing the intercept, 𝛼, to vary cross-sectional.

This effect reduces omitted variable bias regarding firm specific variables. Similarly to this term, 𝜆𝑡

represents a period fixed effect, allowing the intercept to vary periodical. This effect reduces omitted variable bias regarding period specific variables. We perform a redundant fixed effects likelihood ratio test to check whether the use of these fixed effects is necessary. Furthermore, term 𝑣𝑖,𝑡 is an error

term. We describe the dependent variable, 𝜋𝑖,𝑡, the independent variables 𝐶𝑜𝑛𝑐𝑒𝑛𝑡𝑟𝑎𝑡𝑖𝑜𝑛𝑖,𝑡,

𝐸𝑓𝑓𝑖𝑐𝑖𝑒𝑛𝑐𝑦𝑖,𝑡 and control variables, 𝐶𝑜𝑛𝑡𝑟𝑜𝑙𝑠𝑖,𝑡, in the next sections.

4.2 Dependent variable: Financial performance

We use the automotive firm’s financial performance as dependent variable, represented by 𝜋𝑖,𝑡 in

model (1). We use the firm’s return on total assets, ROA, to measure this performance (Berger, 1995; Garza-Garcia, 2012). We furthermore use the lagged dependent variable, 𝜋𝑖,𝑡−1, at the right hand side

14

4.3 Independent variable: Concentration

The independent variable concentration, represented by 𝐶𝑜𝑛𝑐𝑒𝑛𝑡𝑟𝑎𝑡𝑖𝑜𝑛𝑖,𝑡 in model (1), measures the

firm’s market degree of concentration or competition. This variable tests the SCP hypothesis, stating that market structure affects firm performance. We use five different proxies for this variable, namely the Lerner index and four spatial competition proxies that take into account the important local market aspect of independent automotive companies. Competition in the independent automotive market is for the selling and repairing of non-specific manufacturer automobiles5. Therefore, products and services are relatively homogeneous and require the consumer to physically visit the firm. These two characteristics determine the important inclusion of firm location in measuring concentration or competition. We focus on the spatial distribution of automotive companies in the Netherlands through time using data sufficiently detailed to the level of the Netherland’s 418 different municipalities. This subchapter’s next sections describe the concentration or competition proxies in detail.

4.3.1 Concentration proxy: Lerner index

Most studies use the Lerner or Herfindahl-Hirschmann6 index to measure market concentration or competitiveness (Lerner, 1934; Sleuwaegen and Dehandschutter, 1986). The Lerner index proxies the amount of monopoly power by measuring the firm’s ability to set prices. It is the ratio of price minus marginal costs to price. Since marginal cost data are rare in empirical research, it is common practice to use the firm’s profit margin as a proxy for this index (Feinberg, 1980). We therefore proxy for this index by

𝐿𝑒𝑟𝑛𝑒𝑟 𝑖𝑛𝑑𝑒𝑥𝑖,𝑡 =

𝐸𝑎𝑟𝑛𝑖𝑛𝑔𝑠 𝑏𝑒𝑓𝑜𝑟𝑒 𝑖𝑛𝑡𝑒𝑟𝑒𝑠𝑡 𝑎𝑛𝑑 𝑡𝑎𝑥𝑒𝑠𝑖,𝑡

𝑇𝑜𝑡𝑎𝑙 𝑟𝑒𝑣𝑒𝑛𝑢𝑒𝑖,𝑡 (2)

Note a positive and significant coefficient on this concentration proxy, LERNER, supports the SCP theory, since this means firms in concentrated markets are able to set prices above marginal costs and therefore earn a larger ROA than firms in less concentrated markets. However, a positive coefficient on this market power measure may also be the result of firm level efficiency discrepancies (Peltzman, 1977; Berger, 1995). Therefore, we test the separate effects of market structure and efficiency on the firm’s performance. We use spatial competition proxies to measure market structure.

5

Note that the brand-independency of firms is important, since this represents products and services are homogeneous and competition between firms is likely. Contrary, competition is less likely for brand-dependent firms, because their products and services are more heterogeneous.

6 _{The Herfindahl-Hirschmann index is the sum of squares of all individual market shares. However, since we do}

15

4.3.2 Spatial competition proxies

In our first spatial proxy, we measure the number of rival automotive companies near the firm of interest to represent the market structure in terms of competitiveness (Davis, 2006; Watson, 2009). We simply measure this as the natural log of the number of automotive companies in the municipality,

AUTOMUN, of firm i. Hereby, large (small) amounts of automotive firms in a municipality means low

(high) concentration and high (low) competitiveness. Note we hereby implicitly assume a Hotelling (1929) location setting of firms, where in our case firms tend to be in the center of the municipality to reach the largest customer base7. Furthermore, we assume all municipalities to be equally sized, which may be considered unrealistic. Violations of this last assumption may cause an upward bias for the measure in relatively large municipalities. Therefore, this measure provides useful, although not exact, insight in the firm’s number of nearby competitors.

To solve the problem with unequally sized municipalities in reality, we provide a second proxy measuring the relative density of automotive firms in a municipality. This proxy scales the number of automotive companies by the total number of general firms in a municipality as in 𝐴𝑢𝑡𝑜𝑚𝑜𝑡𝑖𝑣𝑒 𝑑𝑒𝑛𝑠𝑖𝑡𝑦 𝑖𝑛 𝑚𝑢𝑛𝑖𝑐𝑖𝑝𝑎𝑙𝑖𝑡𝑦𝑡 =

𝑁𝑢𝑚𝑏𝑒𝑟 𝑜𝑓 𝑎𝑢𝑡𝑜𝑚𝑜𝑡𝑖𝑣𝑒 𝑓𝑖𝑟𝑚𝑠 𝑖𝑛 𝑚𝑢𝑛𝑖𝑐𝑖𝑝𝑎𝑙𝑖𝑡𝑦 _𝑡

𝑇𝑜𝑡𝑎𝑙 𝑓𝑖𝑟𝑚𝑠 𝑖𝑛 𝑚𝑢𝑛𝑖𝑐𝑖𝑝𝑎𝑙𝑖𝑡𝑦 _𝑡 (3)

This allows us to compare the density of automotive to general firms in unequally sized municipalities. Hereby, high (low) automotive density in a municipality, AUTODENSMUN, means low (high) concentration and high (low) competitiveness. Note we may again scale this variable to obtain a third proxy relating the municipality’s density to the country’s automotive firm density by

𝑀𝑢𝑛𝑖𝑐𝑖𝑝𝑎𝑙𝑖𝑡𝑦 𝑎𝑢𝑡𝑜𝑚𝑜𝑡𝑖𝑣𝑒 𝑑𝑒𝑛𝑠𝑖𝑡𝑦 𝑟𝑒𝑙𝑎𝑡𝑖𝑣𝑒 𝑡𝑜 𝑐𝑜𝑢𝑛𝑡𝑟𝑦𝑡 = 𝐴𝑢𝑡𝑜𝑚𝑜𝑡𝑖𝑣𝑒 𝑑𝑒𝑛𝑠𝑖𝑡𝑦 𝑖𝑛 𝑚𝑢𝑛𝑖𝑐𝑖𝑝𝑎𝑙𝑖𝑡𝑦 𝑡 𝐴𝑢𝑡𝑜𝑚𝑜𝑡𝑖𝑣𝑒 𝑑𝑒𝑛𝑠𝑖𝑡𝑦 𝑖𝑛 𝑡𝑕𝑒 𝑁𝑒𝑡 𝑕𝑒𝑟𝑙𝑎𝑛𝑑𝑠𝑡 where 𝐴𝑢𝑡𝑜𝑚𝑜𝑡𝑖𝑣𝑒 𝑑𝑒𝑛𝑠𝑖𝑡𝑦 𝑖𝑛 𝑡𝑕𝑒 𝑁𝑒𝑡𝑕𝑒𝑟𝑙𝑎𝑛𝑑𝑠𝑡 = 𝑁𝑢𝑚𝑏𝑒𝑟 𝑜𝑓 𝑎𝑢𝑡𝑜𝑚𝑜𝑡𝑖𝑣𝑒 𝑓𝑖𝑟𝑚𝑠 𝑖𝑛 𝑡𝑕𝑒 𝑁𝑒𝑡 𝑕𝑒𝑟𝑙𝑎𝑛𝑑𝑠_𝑡 𝑇𝑜𝑡𝑎𝑙 𝑛𝑢𝑚𝑏𝑒𝑟 𝑜𝑓 𝑓𝑖𝑟𝑚𝑠 𝑖𝑛 𝑡𝑕𝑒 𝑁𝑒𝑡 𝑕𝑒𝑟𝑙𝑎𝑛𝑑𝑠_𝑡 (4)

to state whether the firm’s municipality has relatively many or few automotive companies compared to the country’s average number of automotive companies. This proxy, AUTODENSCOUNT, is centered around the number one, where higher (lower) numbers indicate an above (below) country average number of automotive firms.

Next to these firm density effects, Davis (2006) points out the importance of population density and transport costs in a spatial competition study focusing on movie theater performance. He finds the number of people living within a specified distance from the theater to be positively related to sales,

7

16

whereby the effect is larger positive when the specified distance is shorter. In other words, sales go up when the population density close to the theater is high. This indirectly implies the consumer’s unwillingness to travel for the homogeneous product8. We take this effect into account by the number of inhabitants in each municipality through time. We use this variable, AUTOPERCAP, to scale the number of competitors to obtain our last proxy for spatial competition

𝐴𝑢𝑡𝑜𝑚𝑜𝑡𝑖𝑣𝑒 𝑓𝑖𝑟𝑚𝑠 𝑝𝑒𝑟 𝑖𝑛𝑕𝑎𝑏𝑖𝑡𝑎𝑛𝑡𝑡=

𝑁𝑢𝑚𝑏𝑒𝑟 𝑜𝑓 𝑎𝑢𝑡𝑜𝑚𝑜𝑡𝑖𝑣𝑒 𝑓𝑖𝑟𝑚𝑠 𝑖𝑛 𝑚𝑢𝑛𝑖𝑐𝑖𝑝𝑎𝑙𝑖𝑡𝑦 𝑡

𝑇𝑜𝑡𝑎𝑙 𝑖𝑛 𝑕𝑎𝑏𝑖𝑡𝑎𝑛𝑡𝑠 𝑖𝑛 𝑚𝑢𝑛𝑖𝑐𝑖𝑝𝑎𝑙𝑖𝑡𝑦 𝑡 (5)

where the number of competitors is scaled by the number of inhabitants. The smaller (larger) this measure, the smaller (larger) the degree of competition, since few (many) rivalry firms have many (few) potential customers nearby.

4.4 Independent variable: Efficiency

The independent variable efficiency represented by 𝐸𝑓𝑓𝑖𝑐𝑖𝑒𝑛𝑐𝑦𝑖,𝑡 in model (1) tests the efficiency

theory stating financial performance stems from firm-level efficiency discrepancies. We derive the firm’s technical efficiency via a stochastic frontier analysis or SFA, described in this subchapter’s next section.

4.4.1 Stochastic frontier analysis

We derive the firm’s technical efficiency, TECHEFF, via a stochastic frontier analysis or SFA (Battese and Coelli, 1993; Bogetoft and Otto, 2011). In short, the SFA uses an input-output oriented view to benchmark firms against an efficient frontier in order to generate firm-specific efficiency scores. Note this sounds similar to the data envelopment analysis or DEA. However, the main difference between these two is SFA is a parametric method, whereas DEA is a non-parametric method. This allows us to specify a production function in advance and estimate parameters that are not defined a priori. Furthermore, the SFA method is stochastic, whereas the DEA method is deterministic. The SFA method thus separates random noise from data in determining the efficient frontier. The stochastic approach also has interesting implications for the error term. Note the error term is, in both DEA and SFA, the distance from the firm to the efficient frontier. However, in DEA this error is considered as the firm’s inefficiency, whereas in SFA the error term constitutes of both a random and an inefficiency term. The SFA therefore yields a more precise estimation in the presence of noise.

8

17

A first question in determining firm-level efficiencies with SFA is whether to determine technical or cost efficiency. The firm’s technical efficiency is defined as the efficiency in which the firm uses its input x in generating its output y. In this case, we estimate a production function and the relative importance of production factors. Furthermore, inefficiency is dependent on the firm’s deviation from the optimal feasible production given its inputs. A stochastic cost function, on the other hand, uses total costs as dependent variable and outputs y as independent variable. In this case cost inefficiency is partly reflected by the distance the firm lies above the optimal cost frontier given its outputs.

Since our data comprises more input than output data, we choose to estimate the firm’s technical efficiency using a production function. Technical efficiency stems from the technology set T, which is the possibility set of all inputs and outputs. This set is not given, but generally approximated by plotting data points. The technical efficient line is the upper frontier of the technology set and describes the optimal output given input or the optimal input given output. Technical efficiency is therefore mathematically defined as the efficient subset of the technology set T and denoted T E as in

𝑇𝐸_{= (𝑥, 𝑦) ∈ 𝑇 𝑥, 𝑦 𝑖𝑠 𝑒𝑓𝑓𝑖𝑐𝑖𝑒𝑛𝑡 𝑖𝑛 𝑇 (6)}

where (x, y) is efficient in technology set T if and only if there exists no dominant input-output combination (x’, y’) as elements of T.

The production function may take many forms and the SFA requires us to specify it in advance9. We assume the independent automotive companies fit a standard Cobb-Douglas production function (Cobb and Douglas, 1928) as in

𝑦𝑖 = 𝛽0∗ 𝐾𝑖𝛽1∗ 𝐿𝑖𝛽2+ 𝜀𝑖 (7)

where the firm’s output 𝑦_𝑖 is dependent on its amount of capital 𝐾𝑖 and labor 𝐿𝑖. The beta coefficients

represent the relative importance of the production factors and need to be estimated. Furthermore, term 𝜀𝑖 is an error term.

Firstly, we specify the outputs for the independent automotive companies. These consist of the number of sold and repaired non-manufacturer specific automobiles. Since we assume these products and services to be more or less homogeneous and therefore equally priced, we use the firm’s turnover as dependent output variable 𝑦𝑖. Secondly, we specify the production factors capital 𝐾𝑖 and labor 𝐿𝑖. We

use three terms for the firm’s capital, namely the firm’s fixed assets, current assets and liquid

18

resources10. As labor we use the firm’s total amount of full time equivalents (fte). These factors allow us to specify our stochastic production function

𝑇𝑢𝑟𝑛𝑜𝑣𝑒𝑟𝑖,𝑡 = 𝛽0∗ 𝐹𝐴𝑖,𝑡𝛽1 ∗ 𝐶𝐴𝑖,𝑡𝛽2∗ 𝐿𝑖𝑞𝑖,𝑡𝛽3∗ 𝐿𝑎𝑏𝑜𝑟𝑖,𝑡𝛽4∗ exp 𝑣𝑖,𝑡 ∗ exp(−𝑢𝑖,𝑡) (8)

where 𝛽0 is a constant, 𝑇𝑢𝑟𝑛𝑜𝑣𝑒𝑟𝑖,𝑡 the natural logarithm of turnover, 𝐹𝐴𝑖,𝑡 the natural logarithm of

fixed assets, 𝐶𝐴𝑖,𝑡 the natural logarithm of current assets, 𝐿𝑖𝑞𝑖,𝑡 the natural logarithm of liquid

resources and 𝐿𝑎𝑏𝑜𝑟𝑖,𝑡 the number of full time equivalents of firm i at time t. Furthermore, we use

multiplicative stochastic error terms. The amount the firm deviates from the best practice frontier due to random noise is represented by 𝑣𝑖,𝑡. Note this random error may increase the output for positive

values and decrease the output for negative values. The amount the firm deviates from the frontier due to inefficiency is represented by 𝑢𝑖,𝑡. Note 𝑢𝑖,𝑡 may only take values ≥ 0, wherefore term exp(−𝑢𝑖,𝑡)

lowers the firms output due to inefficiencies11. We are however not interested in the firm’s inefficiency 𝑢𝑖,𝑡, but in the firm’s efficiency

𝐸𝑓𝑓𝑖𝑐𝑖𝑒𝑛𝑐𝑦𝑖,𝑡 = exp(−𝑢𝑖,𝑡) (9)

We use these firm specific technical efficiencies as dependent variable in model (1). In this model, a positive significant coefficient on this variable supports the efficiency theory since this means the return on assets is partly determined by the firm’s technical efficiency.

4.5 Control variables

Term 𝐶𝑜𝑛𝑡𝑟𝑜𝑙𝑠𝑖,𝑡 in model (1) represents the control variables. We control for firm size, SIZE, by the

natural logarithm of total assets. Furthermore, we control the spatial competition measures by the natural logarithm of employment, EMPLOYMENT, in the firm’s municipality (Watson, 2009). Recall we moreover reduce omitted variable bias through use of cross-sectional and period fixed effects.

10 _{Fixed assets is the sum of the firm’s property plant and equipment, intangible assets and other fixed assets.}

Furthermore, current assets is the sum of the firm’s inventory, accounts receivable and other current assets.

19

Table 1 Model Summary

This table describes the variables and proxies used in the model relating concentration and efficiency as key independent variables in explaining firm financial performance.

Dependent variable: Financial performance

ROA The ratio of the firm’s earnings before interest and taxes (EBIT) to its total assets. Independent variable: Concentration

LERNER The ratio of the firm’s earnings before interest and taxes (EBIT) to its total turnover. AUTOMUN The natural logarithm of the total number of automotive firms in a municipality.

AUTODENSMUN The ratio of the total number of automotive firms to the total number of general firms in a specific municipality.

AUTODENSCOUNT The AUTODENSMUN (see description above) divided by the ratio of the total number of automotive firms to the total number of general firms in the Netherlands.

AUTOPERCAP The ratio of the total number of automotive firms in a municipality to the total number of inhabitants in the municipality.

Independent variable: Efficiency

TECHEFF The firm’s technical efficiency derived from the error term in a stochastic

Cobb-Douglass production function.

Control variables

SIZE The natural logarithm of the firm’s total assets.

EMPLOYMENT The natural logarithm of the total number of full time jobs in a municipality.

5. Data

We use yearly data provided by Rabobank and BOVAG on a total number of 691 independent automotive companies12 for the period of 2007 to 2011. This dataset forms an unbalanced panel with a total of 2099 observations. The sample companies are all Rabobank financed BOVAG members13. Therefore, we first check whether the sample is free of selection bias and representative for all independent automotive BOVAG members. Thereafter we describe the data and provide summary statistics on the used variables.

5.1 Representativeness

Remember we use a sample of Rabobank financed BOVAG members to represent the independent automotive sector. We are aware the dataset may contain a selection bias regarding Rabobank financed firms. That is, firms in our sample may be selected on certain characteristics by the bank and may therefore be unrepresentative for the population of BOVAG independent automotive companies.

12 _{Note the initial dataset covers 906 different companies, but since some of them only have one year of data,}

they are unsuitable for panel data analysis and therefore removed. This results in the sample with 691 different companies.

13

20

Fortunately, we have data on the other, not Rabobank financed, BOVAG members regarding the firm’s number of full time equivalents, plant size and geographical location14

. This allows us to check whether our sample is representative for the population based on these factors. This subchapter’s next sections describe the sample representativeness for each factor in detail.

5.1.1 Sample representativeness for labor

Firstly, we discuss the amount of labor in the population versus sample. The labor variable plays an important role in the stochastic production function, wherefore sample representativeness for this variable is preferable. We measure labor by the total number of full time equivalents (fte). Firms in our sample have an average of 3.95 full time equivalents over the full period. This number is relatively constant over years, see Appendix Table A. Based upon yearly one-sample t-tests, we conclude the mean number of full time equivalents in the sample does not differ significantly from the population. Therefore, the sample is representative for the population on this factor.

5.1.2 Sample representativeness for plant size

Secondly, we discuss the sample representativeness for the firm’s plant size. This representativeness is again preferable, since the firm’s plant size plays an indirect role in the stochastic production function15. For the representativeness analysis, we measure this variable by the firm’s surface in square meters. Note we have only categorical data on this measure. Therefore, we derive the sample representativeness from Graph 1.

14 _{Due to data availability limitations, we are unfortunately unable to check representativeness in accounting}

terms.

15

21

Graph

1

Firm Plant Size Distribution (population versus sample)

This graph shows the distribution of firms over three plant size categories. The population refers to BOVAG independent automotive companies and the sample to Rabobank financed BOVAG independent automotive

companies.

Graph 1 shows the sample firms are similarly distributed over three categories of plant size. Hereby, the sample has slightly more firms in the middle category of 250 to 750 square meters compared to the population. Furthermore, it has slightly less firms in the category smaller than 250 square meters. This distribution is relatively constant over years, both in the population and sample, see Appendix Table B.

5.1.3 Sample representativeness for geographical location

Thirdly, we discuss the sample representativeness for the firm’s geographical location. It is important the sample firms’ distribution is similar like the population, since we use the firm’s location in spatial proxies for the concentration variable. For the representativeness analysis, we look at the geographical distribution of firms over the Netherland’s twelve provinces. Graph 2 summarizes the results.

0% 10% 20% 30% 40% 50% 60% 70% < 250 m² 250-750 m² > 750 m² P er ce nta g e

Plant size in square meters

22

Graph

2

Firm Geographical Distribution (population versus sample)

This graph shows the distribution of firms over the Netherland’s twelve provinces. The population refers to BOVAG independent automotive companies and the sample to Rabobank financed BOVAG independent

automotive companies.

Graph 2 shows the sample firms are similarly distributed over the Netherlands’ twelve provinces. Most firms are located in the province of Noord-Brabant and the least firms are located in the province of Flevoland. This holds for both the population and sample. Furthermore, this distribution is relatively constant over years, see Appendix Table C.

5.2 Descriptive statistics

Table 2 shows descriptive statistics for the individual sample.

23

Table 2 Descriptive Statistics

This table shows descriptive statistics for the individual sample. The upper part describes the model’s variables and proxies and the lower part the variables for the stochastic production function.

Model descriptives

Mean Median Minimum Maximum Standard deviation Number of observations ROA 19.7% 14.2% -56.7% 243.3% 23.0% 2099 LERNER 9.0% 7.1% -31.0% 109.5% 9.3% 2099 AUTOMUN 100.4 72.0 7.0 771.0 103.9 2099 AUTODENSMUN 3.1% 3.1% 0.8% 6.5% 0.9% 2099 AUTODENSCOUNT 118.1% 117.9% 31.4% 235.7% 34.5% 2099 AUTOPERCAP 0.22% 0.22% 0.06% 0.46% 0.07% 2099 TECHEFF 79.7% 82.7% 10.1% 93.7% 10.5% 1723 SIZE 778 526 24 31,540 1420 2099 EMPLOYMENT 26,493 13,458 1658 531,641 48,528 2099

Production function descriptives

Turnover 1,997 1,054 21 156,441 6,472 2099

Fixed assets 352 206 -11 13,374 632 2099

Current assets 394 216 0 18,109 917 2099

Liquid resources 32.2 6.0 -68.0 1656.0 90.2 2099

Labor 4.1 3.1 0.2 76.9 4.4 2099

Notes: ROA is the firm’s earnings before interest and taxes (EBIT) as percentage of its total assets; LERNER is the firm’s EBIT as percentage of its total turnover; AUTOMUN is the total number of automotive firms in a municipality; AUTODENSMUN is the total number of automotive firms as percentage from all firms in a municipality; AUTODENSCOUNT is the total number of automotive firms as percentage from all firms in a municipality, divided by this same percentage for the Netherlands; AUTOPERCAP is the percentage of automotive firms per capita in a municipality; TECHEFF is the firm’s technical efficiency derived from a stochastic production function; SIZE (times 1,000 euro) is the firm’s total assets; EMPLOYMENT is the total number of full time jobs in a municipality; The firm’s turnover, fixed assets, current assets and liquid resources times 1,000 euro represent actual values; Labor represents the firm’s amount of full time equivalents.

24

Note the technical efficiency is the only measure deviating in the number of observations. More specific, this measure has 1723 observations, whereas the others have 2099 observations. This deviation is because some firms have reported current assets or liquid resources equal to zero. Therefore, we are not able to determine the optimum feasible production for these firms, since in this case the production function yields zero16. In other words, we do not have a benchmark for these firms and are therefore not able to determine their relative efficiencies.

5.3 Correlations

Table 3 shows correlations between variables in the model.

Table 3 Correlation Matrix

This table shows correlations between the dependent, independent and control variables.

1. 2. 3. 4. 5. 6. 7. 8. 9. 1. ROA 1.00 2. LERNER 0.56 1.00 3. AUTOMUN -0.05 -0.07 1.00 4. AUTODENSMUN -0.10 -0.10 0.19 1.00 5. AUTODENSCOUNT -0.10 -0.10 0.20 0.99 1.00 6. AUTOPERCAP -0.06 -0.07 0.18 0.78 0.79 1.00 7. TECHEFF 0.14 -0.23 0.06 0.03 0.03 0.03 1.00 8. SIZE -0.56 -0.29 0.05 0.07 0.08 0.06 -0.01 1.00 9. EMPLOYMENT -0.02 -0.05 0.92 -0.10 -0.09 -0.15 0.06 0.05 1.00 Note: Table 1 describes the variable definitions.

Firstly, we discuss correlations of independent variables with the dependent variable, return on total assets. From Table 3, it is clear the firm’s technical efficiency is positively correlated to the firm’s ROA. This is in line with the efficiency theory, stating more efficient firms have better performance. Furthermore, the correlation signs of the concentration proxies with the dependent variable are in line with the SCP theory. This is clear from the positive correlation of the Lerner index with the firm’s ROA. Recall a large Lerner index, according to the SCP theory, means low concentration and high market power. This positively contributes to the firm’s return on total assets, which is in this table reflected by the positive correlation between these two variables. Furthermore, the spatial competition proxies are all negatively related to the dependent variable. This is also in line with the SCP theory. For instance, the number of automotive firm in a municipality is negatively correlated to the firm’s

16

25

return on assets. This suggests more competition leads to less return. This also holds for the density proxies. More automotive density in a municipality, or relative to the country, is negatively correlated to the independent automotive firm’s return on total assets. The same holds for the correlation of the number of automotive firms per capita with the dependent variable. Note however, these early conclusions based on correlations must be addressed with caution, since correlations do not provide insight in the simultaneous effects of multiple variables on the firm’s performance. We use panel data regressions in the results section to provide detailed insights in these relationships.

Secondly, we discuss the correlation between independent variables. High correlation between independent variables in the same model may lead to multicollinearity problems (Brooks, 2011). Note however, high correlation between different proxies for the same variable is harmless, since we not use them together in a model. For instance, from Table 3 it is clear the automotive density in a municipality has a high correlation of 0.99 with the municipality’s automotive density relative to the country’s density. This however does not cause multicollinearity problems, since we use both variables separately. Note the employment variable has high correlation of 0.92 with the variable measuring the number of automotive companies in a municipality. Both variables thus almost measure the same. Therefore, we prefer the other spatial proxies, since they have mostly 0.15 correlations with other variables in the model and are unlikely to cause multicollinearity issues.

6. Results

26

Table 4 Results

This table shows estimation results for various model specifications with return on assets as dependent variable. All models include period and cross sectional fixed effects. The corresponding t-values appear in parentheses

beneath the variables’ coefficients.

Model specification A B C D E CONSTANT 0.596 2.156 2.137 2.167 2.262 (0.47) (1.39) (1.35) (1.38) (1.47) ROA (-1) 0.045 0.066 0.067 0.067 0.066 (1.33) (1.60) (1.60) (1.60) (1.60) LERNER 1.385 (17.20)*** AUTOMUN 0.040 (0.51) AUTODENSMUN 0.908 (0.34) AUTODENSCOUNT 0.015 (0.22) AUTOPERCAP 17.983 (0.49) TECHEFF 0.439 0.139 0.140 0.140 0.139 (7.40)*** (2.00)** (2.01)** (2.01)** (1.99)** SIZE -0.189 -0.209 -0.209 -0.209 -0.209 (-13.42)*** (-12.07)*** (-12.07)*** (-12.06)** (-12.07)*** EMPLOYMENT 0.032 -0.099 -0.082 -0.085 -0.096 (0.25) (-0.61) (-0.50) (-0.52) (-0.59) R-squared 0.944 0.915 0.915 0.915 0.915 F-statistic 16.09*** 10.36*** 10.35*** 10.35*** 10.36*** Notes: t-values marked with */**/*** denote significance at the 0.10, 0.05 and 0.01 significance levels respectively; ROA (-1) is the firm’s one year lagged ROA; Table 1 describes the other variable definitions.

27

proxies lose their explanatory power and have a positive coefficient when firm level efficiencies are also included in the model. Based on these findings we may reject the structure conduct performance theory. Furthermore, we find no significant effect for profit sustainability in this market, represented by the positive, but insignificant coefficient of the one year lagged performance variable. In combination with the support for the efficiency theory, this means firm efficiencies are easily imitated by other firms resulting in loss of competitive advantages and unsustainable profits17.

6.1 Interaction between efficiency and the Lerner index

Looking at the efficiency proxy’s t-values in Table 4, we see similar values in model specification B to

E. However, model specification A, which includes the Lerner index as explanatory variable, shows a

much larger t-value of 7.40 compared to around 2.00 in other specifications. We therefore suppose an interaction effect between the Lerner index- and the efficiency proxy. The reasoning behind this interaction effect is performance in more monopolistic local markets may be relatively independent on efficiency. More specific, automotive firms having market power may afford themselves to conduct inefficient behavior and simply raise prices to receive their desired profit (Berger and Hannan, 1998). To test this we include an interaction term in model specification A. The results, described in Table E in the Appendix, show there is indeed interaction between these two variables. This may point at a different relative importance of the firm’s efficiency in explaining its return on assets when it has monopoly power. Note however, this observed interaction is between two proxies, where we proxy for the Lerner index by the firm’s profit margin. This raises the important question on how precise our profit margin proxy for the Lerner index truly measures monopoly power. To gain more insight in this possible interaction and the main effect of monopoly power, we provide a different measure of the Lerner index for a subsample in the next section.

6.2 Robustness check: Lerner index

Remember we may reject the SCP theory based on findings regarding the spatial competition proxies in Table 4. However, in one of the model specifications we find the Lerner index positive significant related to the firm’s return on assets. This finding, however, may be the result of an incoherent measurement of the Lerner index with respect to the dependent variable. We namely proxy for the Lerner index by the ratio of the firm’s EBIT to total turnover. Note the dependent variable shares the same numerator as this proxy, which may account for the found relationship between these two

17

28

variables. In this section we provide another measure for the Lerner index to gain more insight in the possible relationship between market power and return on assets. Moreover, we use this alternative Lerner index to reinvestigate the supposed interaction between monopoly power and efficiency. We estimate the alternative Lerner index from a firm specific cost function

𝑇𝐶𝑡 = 𝛼 + 𝛽1𝑇𝑢𝑟𝑛𝑜𝑣𝑒𝑟𝑡+ 𝛽2𝐿𝑎𝑏𝑜𝑟𝑡+ 𝛽3𝐴𝑠𝑠𝑒𝑡𝑠𝑡+ 𝜀𝑡 (10)

where 𝑇𝐶𝑡 represents the firm’s total cost18, 𝑇𝑢𝑟𝑛𝑜𝑣𝑒𝑟𝑡 the total turnover, 𝐿𝑎𝑏𝑜𝑟𝑡 the amount of full

time equivalents and 𝐴𝑠𝑠𝑒𝑡𝑠𝑡 the total assets at time t. Furthermore, term 𝛼 is a constant and 𝜀𝑡

represents an error term. Our alternative measure for the firm’s Lerner index, LERNERALT, is equal to 1 − 𝛽1, see Derivation A in the Appendix. We estimate this cost function for each firm i.

Unfortunately, some firms have too less variation in variables to estimate a decent cost function. In these cases we are unable to estimate coefficient 𝛽1. Therefore, we create a subsample of the 621 firms

for which it is possible to calculate the alternative Lerner index19. Table F in the Appendix shows this subsample’s descriptive statistics and Table 5 shows the estimation results.

18 _{Total cost is the sum of the firm’s wages, costs related to property and plant, other costs, depreciation and}

financial costs.

19

29

Table 5

Results for Subsample with Alternative Lerner Index Measure

This table shows estimation results for the subsample of firms having the Lerner index derived from their cost function. It shows various model specifications with return on assets as dependent variable. All models include

period and cross sectional fixed effects, except specification B, which uses only period fixed effects. The corresponding t-values appear in parentheses beneath the variables’ coefficients.

Model specification A B C D E F CONSTANT 1.551 0.381 2.157 2.138 2.177 2.263 (1.31) (7.11)*** (1.39) (1.35) (1.38) (1.47) ROA (-1) 0.027 0.586 0.066 0.067 0.067 0.066 (0.85) (26.95)*** (1.60) (1.60) (1.60) (1.60) LERNER -0.106 (-0.58) LERNERALT 0.023 (2.28)** AUTOMUN 0.040 (0.51) AUTODENSMUN 0.908 (0.34) AUTODENSCOUNT 0.015 (0.22) AUTOPERCAP 17.983 (0.49) TECHEFF 0.017 0.172 0.139 0.140 0.140 0.139 (0.23) (4.60)*** (2.00)** (2.01)** (2.01)** (1.99)** INTERACTION 2.320 -0.029 (8.98)*** (-2.47)** SIZE -0.196 -0.062 -0.209 -0.209 -0.209 -0.209 (-14.76)*** (-13.36)*** (-12.07)*** (-12.07)*** (-12.06)*** (-12.07)*** EMPLOYMENT -0.030 -0.005 -0.099 -0.082 -0.085 -0.096 (-0.24) (-1.28) (-0.61) (-0.50) (-0.52) (-0.59) R-squared 0.943 0.614 0.902 0.902 0.902 0.902 F-statistic 17.55*** 198.03*** 9.83*** 9.82*** 9.82*** 9.83*** Notes: t-values marked with */**/*** denote significance at the 0.10, 0.05 and 0.01 significance levels respectively; ROA (-1) is the firm’s one year lagged ROA; LERNERALT is the firm’s Lerner index derived from its cost function; INTERACTION is the interaction between the Lerner index and the firm’s efficiency;

INTERACTION takes two forms: in specification A it is LERNER times TECHEFF and in specification B it is LERNERALT times TECHEFF; Table 1 describes the other variable definitions.

30

significant interaction between these two variables. Furthermore, this model specification finds positive significant main effects for the alternative Lerner index and the firm’s efficiency. These findings are consistent with findings in Table 4 and may validate there is a positive effect of the Lerner index and efficiency on the firm’s return on assets. Furthermore, the interaction between the alternative Lerner index and the firm’s efficiency is of negative form. This implies the larger (smaller) the alternative Lerner index, the less (more) important the firm’s efficiency is in estimating its return on assets. In other words, it may be the case the more monopoly power the firm enjoys, the less important its efficiency is in earning a return on assets. This is similar to the quiet life hypothesis stating monopoly power reduces the importance of firm efficiency (Berger and Hannan, 1998) However, we again find the spatial competition proxies in model specification C to F insignificantly related to the firm’s performance. This is consistent with findings in Table 4. Therefore, we state it is unlikely the firm’s return stems from competition or concentration, wherefore the observed relationship between the Lerner index and efficiency may be merely the result of efficient firms having the ability to set price above marginal costs (Peltzman, 1977). Remember, this is an effect independent of market structure, wherefore our findings support the efficiency theory rather than the SCP theory. Furthermore, the observed relationship between monopoly power, efficiency and return on assets may also be the result of efficient firms developing large market shares enabling them to set price above marginal costs according to the RMP theory. Remember however, we are unable to test for this effect since market share data is unfortunately unavailable.

7. Summary and Conclusions

This study is a contribution to the debate whether firm performance is determined by market structure or firm level efficiencies. The traditional view states the market’s structure determines the firm’s amount of market power, which in turn determines its performance in terms of profits. Contrary, the revisionist view states any observed market power effects on the firm’s performance are the result of firm level efficiency discrepancies.

We test both views using a sample of 691 Rabobank financed Dutch independent automotive firms in the period of 2007 to 2011. We find this sample is representative for the population of Dutch independent automotive companies regarding the amount of labor, firm size and geographical location.

31

frontier analysis benchmarks each firm’s output against the best practices output given input. Deviations from this best practice output are split in random errors and inefficiency allowing us to separate technical efficiency from chance. We use the competition proxies, technical efficiency measures and one year lagged return on assets as main independent variables in a panel data model with return on assets as dependent variable. Furthermore, we control for firm size and employment in the firm’s municipality. Moreover, we use period and cross sectional fixed effects to control for omitted variable bias.

We conclude firm performance in the Dutch independent automotive industry stems from firm level efficiencies rather than market structure. This is based on the firm’s technical efficiency, which we find in all cases positive significant related to the firm’s return on assets. Hence, more efficient use of capital and labor in generating turnover strengthens financial performance in terms of the automotive company’s return on assets. Remember the efficiency theory furthermore states any observed market power effects are the result of firm level efficiency discrepancies rather than market structure effects. We find these market power effects reflected by a positive significant effect of the Lerner index on the firm’s performance. This holds for both the profit margin proxy of the Lerner index as well as the Lerner index measure derived from the firm’s cost function. We conclude this observed market power effect is unlikely to be the result of market structure, since we find none of the spatial competition proxies significantly related to the firm’s performance. Hence, we state the Dutch independent automotive industry supports the efficiency theory rather than the market power theory.

8. Limitations and Suggestions for Future Research

This study is subject to several limitations which may be resolved in future research. Firstly, our study is limited to data availability regarding market share. As market power theories, Berger (1995) distinguishes the structure conduct performance (SCP) and the relative market power (RMP) theory. We reject these market power theories by finding positive support for the contradicting efficiency theory and no support for market structure effects. However, our market structure proxies strictly only test the SCP theory. One needs market share data on the individual companies to test the RMP theory as well. Future research on independent automotive companies may test the impact of market share on performance and, moreover, the impact of efficiency on market shares in this industry.

32

should also consider privately financed firms and firms financed by another bank. Moreover, future research may expand the population by considering non-BOVAG members as well.

33

References

Amato, L.; Wilder R. (1990), “Firm and Industry Effects in Industrial Economics”, Southern

Economic Journal, Volume 57, Issue 1, p93-105

Bain, J.S. (1951), “Relation of Profit Rate to Industry Concentration: American Manufacturing, 1936-1940”, The Quarterly Journal of Economics, Volume 65, Issue 3, p293-324

Battese, G.E.; Coelli, T.J. (1993), “A Stochastic Frontier Production Function Incorporating a Model for Technical Inefficiency Effects”, Working Papers in Econometrics and Applied Statistics, Volume 69, p1-27

Berger, A.N. (1995), “The Profit-Structure Relationship in Banking: Tests of Market-Power and Efficient-Structure Hypotheses”, Journal of Money, Credit and Banking, Volume 27, Issue 2, p404-431

Berger, A.N.; Hannan, T.H. (1998), “The Efficiency Cost of Market Power in the Banking Industry: A Test of the “Quiet Life” and Related Hypotheses”, The Review of Economics and Statistics, Volume 80, Issue 3, p454-465

Bertrand, J. (1838), “Théorie Mathématique de la Richesse Sociale”, Journal des Savants, Volume 67, p499-508

Bogetoft, P.; Otto, L. (2011), “Benchmarking with DEA, SFA and R”, Springer

Brooks, C. (2011), "Introductory Econometrics for Finance", Cambridge University Press

Cobb, C.W.; Douglas, P.H. (1928), “A Theory of Production”, The American Economic Review, Volume 18, Issue 1, p139-165

Cournot, A.A. (1838), “Recherches sur les Pricipes Mathématiques de la Théorie des Richesses”,

Hachette

Davis, P. (2006), “Spatial Competition in Retail Markets: Movie Theaters”, The RAND Journal of