Firm-specific time preferences and postmerger firm performance

Accepted Manuscript

Firm-Specific Time Preferences and Postmerger Firm Performance Jeremiah Harris, Ralph Siebert PII: DOI: Reference:

S0167-7187(17)30255-2 10.1016/j.ijindorg.2017.04.001 INDOR 2362

To appear in:

International Journal of Industrial Organization

Received date: Revised date: Accepted date:

26 January 2016 19 March 2017 11 April 2017

Please cite this article as: Jeremiah Harris, Ralph Siebert, Firm-Specific Time Preferences and Postmerger Firm Performance, International Journal of Industrial Organization (2017), doi: 10.1016/j.ijindorg.2017.04.001

This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and review of the resulting proof before it is published in its final form. Please note that during the production process errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.

ACCEPTED MANUSCRIPT

Highlights • We estimate firm-specific discount factors for public/private semiconductor firms • Discount factors explain mergers and the impact on product market performance • Impatient firms are highly efficient and merge with efficient and innovative firms

CR IP T

• Patient firms are less efficient and merge with firms that are larger than they are

AC

CE

PT

ED

M

AN US

• Impatient firms achieve relatively more efficiency gains than patient firms

1

ACCEPTED MANUSCRIPT

Firm-Specific Time Preferences and Postmerger Firm Performance∗ †

Ralph Siebert‡

April 19, 2017

Abstract

CR IP T

Jeremiah Harris

ED

M

AN US

This study investigates the role of firm-level discount factors in evaluating the impact of mergers on market outcomes. Discount factors reflect time preferences for future cash flows and are used to determine the present value of investment projects such as mergers. Firmspecific discount factors imply that firms may attach different present values to mergers. We elicit firm-specific time preferences and identify firms’ discount factors using firm-specific production data while building on the existence of learning-by-doing in the semiconductor industry. Our estimation results show that firm-specific discount factors explain firms’ production decisions. We also find that firms’ discount factors and merger acquisition strategies explain heterogeneous merger outcomes. Our results show that acquiring firms characterized by low discount factors (impatient firms) are highly efficient and merge with highly efficient and innovative firms. Impatient acquirers achieve relatively higher efficiency gains in the short run than patient acquirers and adopt acquisition strategies that put more weight on achieving instant efficiency gains. In contrast, patient acquirers are least efficient and merge with firms that are larger than themselves. Patient acquirers place more value on achieving efficiency gains in the more distant future.

PT

JEL: D24, D43, G34, L13, L22.

AC

CE

Keywords: Discount Factor, Discount Rate, Dynamic Oligopoly Model, Market Performance, Mergers and Acquisitions, Semiconductor Industry.

∗ We thank seminar participants at Purdue University and Utah State University for their useful comments. We especially thank Mara Faccio, Christian Gollier, Christian Huse, Stephen Martin, John McConnell, and Richard Peter for their helpful discussions and valuable suggestions on this or an earlier draft. All errors are our own. † Kent State University, Department of Economics, College of Business Administration, Phone: 1-330-672-1097, Address: P.O. Box 5190, Kent, OH 44242, United States, E-mail: [email protected]. ‡ Purdue University, Department of Economics, Krannert School of Management, Phone: 1-765-494-3401, Address: 403 West State Street, West Lafayette, IN 47907-2056, United States, E-mail: [email protected].

ACCEPTED MANUSCRIPT

1

Introduction

Discount factors reflect the opportunity cost of spending limited resources today in exchange for expected payoffs tomorrow. Additionally, discount factors reflect time preferences for future

CR IP T

cash flows and are used to determine the present value of specific investment projects. Prominent studies argue that firms are characterized by different discount factors. Harrington (1989) reports two reasons why discount factors can vary among firms: imperfections in capital markets due to imperfect information and agency problems due to the inability of shareholders to perfectly monitor managers.1 Additionally, Andersson (2008) notes that different probabilities of going

AN US

bankrupt (Merton, 1974) explain different discount factors across firms. Firm-specific discount factors imply that firms may attach different present values to mergers (which determine different incentives for merger formation, lead to different merger objectives, and achieve different merger outcomes) (see also de Roos (2004)).

This study investigates the extent to which firm-specific time preferences (discount factors)

M

impact merger formation and merger outcomes. Our study examines whether firms with different discount factors select themselves into mergers for different purposes, i.e., mergers that achieve

ED

relatively higher efficiency gains compared to market power effects.2 Our study also evaluates the relationship between firm-level discount factors and the postmerger impact on the product

PT

market. To date, very little is known about these aspects, and more insight is desired. Recent studies have established a link between discount factors and merger formation. For

CE

example, it has been shown that acquiring firms are frequently characterized by a lower cost of capital than target firms (Erel, Jang, and Weisbach (2014), Fluck and Lynch (1999), and

AC

Khatami, Marchica, and Mura (2015)). Moreover, the value of investment projects or mergers is frequently assessed in a dynamic context in which future cash flows are discounted to determine a present value (see, e.g., Pesendorfer (2003), Gowrisankaran (1999), and Davis and Huse (2010)). Although firms may agree on the project’s or merger’s risk profile and the potential benefits, 1

See, e.g., Easley and O’Hara (2004) and Diamond and Verrecchia (1991). For more information on endogenous merger formation, see Tombak (2002), Compte, Jenny, and Rey (2002), Vasconcelos (2005), and Javonovic and Wey (2012). As an example of endogenous merger formation, Tombak (2002) shows that firm size (and firm efficiency) explains firms’ incentives to merge, i.e., the largest firm acquires the next largest firm, etc. 2

1

ACCEPTED MANUSCRIPT

differences in firm-specific discount factors may result in different present values attached to the merger. Therefore, depending on firms’ discount factors, firms might very well value mergers differently, leading to different merger objectives, merger incentives, and merger outcomes. More specifically, since discount factors reflect firms’ patience levels or willingness to wait for the

CR IP T

returns of investments, firms characterized by high discount factors place more weight on future cash flows. These firms are patient and more willing to wait for future returns on investments than firms characterized by low discount factors. Thus, firms with high discount factors assign comparatively greater value to merger projects that generate returns in the more distant future compared to firms with low discount factors.

AN US

In the merger literature, it is common to distinguish whether the value added from merging is dominated by efficiency or market power arguments (Williamson, 1968). Efficiency-dominated mergers are beneficial to producer and consumer surplus, as output units are transferred from less efficient to more efficient production facilities (Salant and Shaffer (1998; 1999)). In contrast,

M

market power-dominated mergers are associated with elevated prices, which are beneficial to producers but harmful to consumers.3

ED

It is also common to distinguish mergers by the timing of when they realize value added. While some mergers generate instant value added due to efficiency gains or market power gains, other mergers may generate value in the more distant future. For example, in the airline industry,

PT

it took up to two and a half years before the merger between Delta and Northwest and the merger between United and Continental realized efficiency gains (Kim and Singal, 1993).4

CE

It is reasonable to expect that acquirers with different discount factors merge with different firms depending on their merger objectives, which could differ by the extent and the timing

AC

of when postmerger value added or efficiency gains are realized. While some acquirers prefer to engage in mergers that realize efficiency gains more instantly, other acquirers would prefer investment and merger strategies that put more weight on long run returns. Our study concentrates on semiconductors, which are inputs for electronic devices. The 3 Common oligopoly models predict that if the cost synergies outweigh the market power benefits, then the price will decline and market share will increase. Likewise, if the market power benefits outweigh the cost synergies, then the price will increase and market share will decline (see also Farrell and Shapiro (1990), Stigler (1964), Perry and Porter (1985), Salant, Switzer, and Reynolds (1983), and Gugler and Siebert (2007)). 4 Mouawad, Jad, Big Problems for Biggest Airlines, The New York Times, November 29, 2012.

2

ACCEPTED MANUSCRIPT

semiconductor industry provides a natural setting for assessing the role of discount factors in the formation and impact of mergers for several reasons. First, numerous horizontal mergers are performed in the industry. Second, the industry is characterized by a large degree of heterogeneity in production and innovation across firms, which increases the relevance of evaluating

CR IP T

heterogeneous postmerger effects. Third, learning-by-doing, which is well documented in the semiconductor industry, leads firms to incorporate future cost savings into current production decision making. Against the background of learning-by-doing and the fact that our dataset contains detailed firm-level production data, we are able to identify firm-specific discount factors from firms’ intertemporal production decisions. Fourth, the semiconductor industry is char-

AN US

acterized by low concentration, and no antitrust merger interventions have taken place such that mergers characterized by different degrees of efficiency gains and market power effects are likely to be observed. Our dataset returns an average Herfindahl-Hirschman Index of 278 across years, which is classified as an unconcentrated market according to the Department of Justice

M

Horizontal Merger Guidelines (HHI < 1500).5 To summarize, the industry provides a natural setting for assessing the role of discount factors in the formation and impact of mergers.

ED

We use detailed firm-level production and innovation data on the semiconductor industry from 1989 to 2004. Our dataset contains publicly traded and privately held firms. The large majority of firms in our dataset is privately held and, therefore, lack sufficient information to

PT

estimate the discount factor using conventional methods from the finance literature. Using only the public firms would cause information loss and potential selection bias. Thus, one empirical

CE

challenge in our study is to obtain the firm-specific discount factors for both privately held and publicly traded firms. Building on a framework by Irwin and Klenow (1994), we elicit

AC

firm-specific time preferences and identify firms’ discount factors using firm-specific production data while building on the presence of learning-by-doing in the semiconductor industry (see also Berry and Pakes (2000) and Aguirregabiria and Magesan (2013)). Incorporating learning-bydoing into firms’ dynamic supply relations characterizes the intertemporal link between firms’ contemporaneous production decisions and their incentive to achieve future cost savings and enables us to identify firms’ discount factors. 5

https://www.justice.gov/atr/horizontal-merger-guidelines-08192010.

3

ACCEPTED MANUSCRIPT

Our estimation results provide evidence that firms differ in their discount factors. Against the background of learning-by-doing, we find that firms’ discount factors exert a significant impact on firms’ production strategies. More patient firms (firms with higher discount factors) consider current production as an investment to achieve future cost reductions via learning-

CR IP T

by-doing (see also Fudenberg and Tirole (1983) and Thompson (2010)). Next, we investigate the relationship between firms’ discount factors, merger incentives, and merger outcomes. In estimating postmerger outcomes, we put special emphasis on firm heterogeneities driven by firms’ different discount factors and efficiencies. We estimate a heterogeneous treatment effects model and control for two potential heterogeneity biases caused by: (1) the pretreatment heterogeneity

AN US

(endogenous selection); and (2) the post-treatment heterogeneity (heterogeneous impacts after merging).6

Our estimation returns a large degree of heterogeneity in the discount factors of merging firms. The results show that discount factors and merger acquisition strategies contribute to

M

explaining the impact of mergers. We find that acquiring firms with low discount factors as well as acquiring firms with high discount factors have incentives to merge with efficient, innovative,

ED

and large targets. Our results indicate that acquirers with low discount factors (impatient firms) were already more successful in generating efficiency gains before merging. They merge with highly efficient and innovative targets and achieve relatively higher efficiency gains than mergers

PT

with patient acquirers. We find that impatient acquirers achieve relatively larger efficiency gains (in the short run) than patient acquirers. Our results suggest that less patient acquirers

CE

adopt merger strategies that are characterized by shorter investment time horizons and place more value on instant efficiency gains. In contrast, acquiring firms with high discount factors

AC

appear to face premerger difficulties in generating efficiency gains. They are the least efficient firms among all merging firms. They merge with firms that are larger, more efficient, and more innovative than themselves. These acquirers adopt a long-term investment strategy that assigns higher value to achieving efficiency gains in the more distant future. The policy implications of our study suggest that firm-level discount factors can play a critical 6

See also Angrist and Krueger (1999), Morgan and Winship (2007), Dehejia and Wahba (2002), Brand and Xie (2010), Brand and Thomas (2013), and Pais (2011).

4

ACCEPTED MANUSCRIPT

role in explaining postmerger effects. The heterogeneity in firm-level discount factors can provide useful insights and help us develop more precise forecasts about the effects of a proposed merger on both producer and consumer surplus. For instance, our results would indicate that mergers performed by firms with low discount factors are more likely to achieve relatively larger efficiency

CR IP T

gains in the short run.

This study is structured as follows: Section 2 provides a description of the industry and the data. Section 3 introduces the empirical model and explains the estimation procedure. Section 4 discusses the estimation results for discount factors and the marginal costs. Here, we also present the estimates from the heterogeneous treatment model and assess the role of discount

AN US

factors on merger formation and the impact of mergers. Finally, Section 5 concludes.

2

Industry and Data Description

The semiconductor industry is one of the most important high-technology industries because it

M

affects many downstream industries. Semiconductors are used widely in the computer industry and consumer electronics, as well as in communication equipment. Semiconductors are usually

ED

distinguished between microprocessors, memory chips, and other related devices. Our firm-level revenue data are provided by the Gartner Group. The dataset contains annual

PT

firm-level revenues from 1989 to 2004 for the semiconductor market overall, as well as for several submarkets, i.e., static and dynamic random access memories and flash memories (see Gugler

CE

and Siebert (2007) for a more thorough description). It includes international firms that actively produce in the semiconductor industry and generate more than $1 million in annual revenues.

AC

In order to have a sufficiently large number of mergers in our study, we aggregate production to the semiconductor market level. Using the producer price index (PPI) as a proxy for the price of a semiconductor, we convert semiconductor revenues into quantities.7 Our merger data are taken from the Thomson Reuters SDC Platinum database for global

mergers for the 1989 to 2004 time period. Since our study focuses on the postmerger effect on the product market, we focus on horizontal mergers and select mergers where both the acquiring 7

The PPI for “Semiconductor and Other Electronic Component Manufacturing” is provided by the Bureau of Labor Statistics using 1988 as the base year, http://www.bls.gov/ppi/.

5

ACCEPTED MANUSCRIPT

and the target firms are active semiconductor chip producers according to the production data provided by the Gartner Group. This results in 133 mergers. We account for changes in ownership by assigning all revenue from the target firm to the acquiring firm starting the year the merger becomes effective. In the rare case where the same target is acquired multiple times

CR IP T

by different acquiring firms over time, we assign the target’s revenue to the first acquirer until the time of the subsequent merger. At this point, all revenue from the target firm is reassigned to the second (or next) acquiring firm.8 Table 1 shows the number of firms with revenue data, total industry revenue, and the number of mergers per year. The table shows that the number of mergers increased from the mid-1990s until peaking in the year 2000. On average, eight mergers

AN US

were performed each year. Moreover, the number of producing firms, total revenue, and average production increased until roughly 2000 and then slightly declined. The production pattern is highly correlated with the number of mergers in the industry. Within the industry, there is much heterogeneity among firms, especially with regard to production. Figure 1 shows a scatterplot of

M

annual quantity by firm.9 As shown in the figure, the industry consists of many small producers and a few large producers. Note that the order among firms (i.e., which firms are the largest and

ED

second largest producers) remains relatively stable over time. This observation indicates that time-invariant heterogeneity across firms is an important characteristic for which to control. Additionally, we use patent information from the United States Patent and Trademark Office

PT

(USPTO) available in the National Bureau of Economic Research (NBER) database.10 For descriptors of the data and methodological methods, see Hall, Jaffe, and Trajtenberg (2001). The

CE

patent database allows us to track firms’ patent applications over time. The dataset provides information by technology class for more than 109,000 patent applications submitted from 1975 to

AC

2004 in the semiconductor industry. We establish a patent stock for every firm by accumulating the annual firm-level patents over time, allowing for an annual depreciation rate of 5%. Figure 2 shows the accumulated number of patents for every firm and provides evidence that firms also 8

In unreported robustness tests, we relax this assumption and remove all acquisitions where the target is acquired multiple times. Our results provide evidence that the estimated discount factors remain significantly unchanged and that acquiring firms with low discount factors continue to achieve relatively greater efficiency benefits than acquiring firms with high discount factors, at least in the short run. 9 The following largest producers are illustrated in the figure: Intel, Toshiba, Hitachi, Texas Instruments, NEC Corporation, Fujitsu, and Vitesse. 10 The patent data is available at https://sites.google.com/site/patentdataproject/Home.

6

ACCEPTED MANUSCRIPT

exhibit heterogeneity in innovation, in addition to heterogeneity in production. Moreover, it is interesting to note that the firms with higher production levels also have larger patent stocks.11 The large degree of firm-level heterogeneity in production and innovation indicates that firm selection into mergers is an important fact to consider when evaluating the impact of mergers.

CR IP T

We also use several other controls. First, it is well documented that learning-by-doing based on own experience and based on other firms’ experience via spillovers are important phenomena in the semiconductor industry. To account for own learning-by-doing, we use past accumulated firm-level production as a proxy, and to account for spillover learning, we use past industry-level production by all other firms. Also, we include several industry-level variables (semiconductor

AN US

wage, number of firms, and the GDP in electronics) that will serve as supply and demand shifters.12 Table 2 provides the summary statistics for these variables. The average firm has a market share of 0.7%, which corresponds to an average annual production of 14.56 million semiconductor chips. The accumulated production at the firm level is, on average, 54.10 million

M

units, and the average accumulated production by others in the industry is much larger, at 11.188 billion units. A firm applies, on average, for 36.27 semiconductor patents per year and

ED

is described by a patent stock that consists of 227.04 semiconductor patents. The average semiconductor price over the period is $71.98.

The Empirical Model

PT

3

CE

Our ultimate goal is to investigate the role of firm-level discount factors in evaluating the impact of mergers on market performance. We are especially interested in the relationship between the

AC

discount factor and efficiency gains. In evaluating the competitive impact of mergers, we take advantage of our highly detailed production data and measure the impact of mergers on the postmerger output of merging firms. We follow Farrell and Shapiro (1990) and evaluate the change in market shares before and after the merger, which is helpful for drawing conclusions 11

IBM is the one notable exception, since it has the highest accumulated patents but little production volume. It is well known in the industry that IBM frequently licenses production to other firms. 12 Wage information is from the Yearbook of Labor Statistics (1988 to 2004), ISIC second revision 3832, which includes “semiconductors and related sensitive semiconductor devices” for U.S. manufacturers. The GDP information is from the Bureau of Economic Analysis for “electrical equipment, appliances, and components.”

7

ACCEPTED MANUSCRIPT

on postmerger prices. This seminal theoretical contribution has shown that if the efficiency effect dominates the market power effect, market shares will increase and prices will decline (see also Mueller (1985), Gugler and Siebert (2007), and Duso, R¨oller, and Seldeslachts (2014)). In

after merger formation and market shares decline.

CR IP T

contrast, if the market power effect dominates the efficiency effect in a merger, prices increase

Since a merger is described as two independent firms joining together to form one entity, we formulate firm-pairs and evaluate the change in market shares (before and after merging) and compare those changes between merging and nonmerging firms in firm-pairs. To formulate the outcome equation, we consider a set of semiconductor firms i ∈ I and form firm-pairs by

AN US

matching each firm with each other for every year t. Firm-pairs are denoted by a subindex j, k, where j ∈ I is specific to the acquiring firm in a merger and k ∈ I refers to the target firm. The main equation of interest evaluates the effect of a merger (indicated by the merger dummy Mj,k,t , which takes on a value of 1 if firms j and k merged in period t and 0 otherwise) on the

M

change in market shares M S from year t − 1 to year t + 1.13 In specifying the outcome equation, we follow Mueller (1985) and Gugler and Siebert (2007) and specify firms’ market shares as

ED

functions of the mergers and past market shares. We consider the sum of the market shares between firms j and k (M Sj,k,t = M Sj,t + M Sk,t ) in every period t. Hence, M Sj,k,t is the joint production if a firm-pair merged in period t, and it is the sum of the firm-pair market share if

PT

the firm did not merge in period t. The outcome equation is formulated as follows:

CE

M Sj,k,t+1 = ρ0 + ρ1 M Sj,k,t−1 + ρ2 Mj,k,t + ρ3 Mj,k,t (δj − δ) + ρ4 δj + ρX + j,k,t ,

(1)

AC

where ρ2 measures the average effect of a merger. Note that the interaction between the merger indicator and the acquirer’s discount factor (ρ3 ) allows for heterogeneous effects. Hence, we allow the impact on market shares to vary for acquiring firms with different discount factors (Mj,k,t (δj − δ)), where δ is the mean of the discount factors across all firms. Remember, we assume that the heterogeneity of the discount factors stems mostly from imperfections in capital markets, such that the discount factors enter our model exogenously.14 The parameter ρ4 13 14

We measure the impact on market shares in t + 1 since it is the first full year of postmerger production. Although this paper assumes the discount factor enters the outcome equation exogenously, it is possible that

8

ACCEPTED MANUSCRIPT

measures the relationship between a firm’s discount factor and market share. The matrix X contains additional controls we introduce later, ρ is a vector of parameters, and j,k,t represents the error term. In estimating the outcome equation, we face two important challenges: first, firm-level dis-

CR IP T

count factors (δj ) are unobserved, and second, mergers (Mj,k,t ) are endogenous events. To overcome the first problem, we estimate firm-specific discount factors from firms’ supply relationships, as will be explained in detail in Section 3.1. Next, in order to control for endogenous selection into mergers, we apply a heterogeneous treatment effects estimator. This endogenous

3.1

AN US

merger selection is explained in Section 3.2.

Supply Relationship

In following Irwin and Klenow (1994) and Siebert (2010), we consider an oligopolistic market and estimate firm-specific discount factors (δi ) from firms’ supply relations. We consider the

M

set of all semiconductor firms i ∈ I in this section. Note that we have firm-specific production information at the semiconductor market level, which is more disaggregate than overall firm-

ED

level information.15 Therefore, applying the same assumptions as Irwin and Klenow (1994), we assume that each semiconductor firm chooses its output (qi,t ) within a Cournot framework to maximize its discounted present value. The firm’s maximization problem is given by:

PT

max Πi = E0

∞ X t=0

δit (Pt

−

stat M Ci,t )qi,t

#

,

(2)

CE

qi,t

"

where E0 is the expectation operator conditional on information at time 0, Pt is the price at

AC

stat is the static marginal cost, and δ is the firm-level discount factor. Note that period t, M Ci,t i

the firm-level discount factor relates to the firm-level discount rate (ri ) as follows: δi =

1 1+ri .

The discount factor measures the value a firm places on future profits and is used to calculate

there is a reverse link between market shares and discount factors: for instance, a decreasing market share may imply that bankruptcy becomes more likely, leading to a higher discount factor. This research could be extended in the future to incorporate an endogenous discount factor. 15 Focusing on more disaggregate markets (such as the dynamic random access memory or static random access memory market) would leave us with very few merger cases. Another advantage with focusing on the semiconductor level is that most firms are specialized in semiconductors, and we are able to evaluate better an overall firm-level discount factor in the semiconductor industry.

9

ACCEPTED MANUSCRIPT

the present value. A lower discount factor indicates that a firm values future profits less. The first-order condition with respect to quantity becomes: "∞ #   stat X ∂M C M Si,0 i,t stat δit qi,t = M Ci,0 + E0 , P0 1 + α1 ∂qi,0

(3)

CR IP T

t=1

where α1 is the price elasticity of demand. The first-order condition (3) indicates that price, adjusted for a firm-specific markup, is equated to the dynamic marginal cost. The dynamic stat ) plus an adjustment term that marginal cost is composed of the static marginal cost (M Ci,0 stat P ∂M Ci,t t accounts for the discounted value of future cost reductions ( ∞ t=1 δi qi,t ∂qi,0 ) achieved from

AN US

learning. This incorporates firms’ intertemporal production strategies, as they increase current production to achieve future cost reductions (Wright, 1936). Forward-looking firms price according to dynamic marginal costs and increase production beyond the statically optimal production level. They “overproduce” in a static sense to benefit from future experience and cost savings. Hence, instead of determining their optimal output according to firms’ static marginal costs,

M

they produce along their dynamic marginal costs, which lie below the static marginal costs. Using a recursive formulation, the equation becomes:

where

stat ∂M Ci,t+1 ∂qi,t



"

#   stat ∂M Ci,t+1 M Si,t+1 stat + Pt+1 1 + qi,t+1 − M Ci,t+1 = 0, (4) ∂qi,t α1

ED

M Si,t 1+ α1

stat − M Ci,t − δi

PT

Pt



accounts for changes in the marginal costs in time t + 1, which result from

CE

production in time t.

Our identification argument builds on a well-established institutional feature of the semiconductor industry, i.e., the existence of learning-by-doing. Equation (4) forms the center of

AC

how we identify the discount factor.16 The discount factor (δi ) describes the intertemporal link

between current quantity and future savings in the next period through learning-by-doing. In the presence of learning-by-doing, firms account for the fact that higher contemporaneous production accumulates more experience in the future, which generates future cost savings. Firms 16

Although it is reasonable to believe that a firm’s discount factor will change over time, this identification process does not allow for estimation of firm- and time-specific discount factors. We will leave this extension to future research.

10

ACCEPTED MANUSCRIPT

characterized by high discount factors (patient firms) value future profit streams more than firms with low discount factors (impatient firms). Hence, firms with high discount factors realize a high present value from learning-by-doing in the future. In contrast, firms with low discount factors (impatient firms) place more weight on current profits and value future returns less.

CR IP T

They have higher incentives to invest in immediate cost reductions via learning-by-doing and lower incentives to invest in future cost reductions via learning-by-doing. In the extreme case, myopic firms value only current profits, adopt a statically optimal production plan, and produce according to their static marginal costs. To summarize, the interdependence between today’s optimal production decision and the incentive to invest in future cost reductions enables us to

AN US

identify firms’ discount factors.

Rearranging equation (4), we obtain the following estimation equation: " #     stat ∂M C M S M Si,t i,t+1 i,t+1 stat stat + Pt+1 1 + = M Ci,t +δi qi,t+1 − M Ci,t+1 Pt 1 + +νi,t , (5) α1 ∂qi,t α1

M

which includes a normally distributed error term, νi,t . We follow Irwin and Klenow (1994) and

ED

Liu, Siebert, and Zulehner (2013) and assume a semi-log marginal cost function: stat = λi + λ1 ln(Acqi,t ) + λ2 ln(Acq−i,t ) + λ3 ln(W aget ) + λ4 P ati,t . M Ci,t

(6)

PT

Learning-by-doing is incorporated at the firm level using a firm’s total past accumulated production (Acqi,t ) as a proxy for its experience. We also account for learning from others via

CE

spillovers from accumulated production of all other firms (Acq−i,t ). We expect both own learning and spillover learning to lower the marginal cost (i.e., λ1 < 0 and λ2 < 0). Also, the

AC

factor price of semiconductor wages is included to account for shifts in the marginal costs due to changes in input prices. Contemporaneous patent applications are included to control for innovations that could affect the cost of production. Finally, we allow for a firm intercept (λi )

to account for firm heterogeneity. The final estimation equation is obtained by inserting the corresponding marginal cost equation (6) into equation (5).

11

ACCEPTED MANUSCRIPT

3.2

Outcome Equation and Endogenous Selection

Our main equation of interest, also referred to as the outcome equation (1), assesses the evolution of firms’ market shares over time. Since different types of firms self-select into mergers, we are

CR IP T

concerned with potential biases arising from the fact that observed and/or unobserved firm-level attributes highly correlate between firms’ decisions to merge and their decisions to produce. To properly estimate the outcome equation, we must account for endogenous selection into a merger. Our solution to the selection problem is to use the heterogeneous treatment effects estimator, which is introduced later.

AN US

As detailed earlier, we regress the joint market shares of firms j and k in period t + 1 (M Sj,k,t+1 ) on their joint market shares in period t − 1 (M Sj,k,t−1 ). The outcome equation is specified as:

M Sj,k,t+1 = ρ0 + ρ1 M Sj,k,t−1 + ρ2 Mj,k,t + ρ3 Mj,k,t (δj − δ) + ρ4 δj + ρ5 N F irmst

+

25 X

M

+ ρ6 GDPelec,t + ρ7 M S j + ρ8 AcP atj + ρ9 HEC1j,k,t + ρ10 HEC0j,k,t ρy Y eart + j,k,t .

(7)

ED

y=11

We apply the heterogeneous treatment effects estimator suggested by Heckman, Urzua, and Vytlacil (2006), which allows us to control for a potential pretreatment bias (i.e., endogenous

PT

merger formation) as well as the post-treatment effect.17 In closely following the heterogeneous

CE

treatment effect literature, we account for heterogeneities in the effect of the merger for acquirers of different discount factors by including the interaction term Mj,k,t (δj − δ). The outcome equation also includes the number of firms (N F irms), to control for the degree of competition

AC

in the product market, and the electronic GDP, to control for downstream shifts in demand. We controlled for further unobservables that could affect the relationship between the dis-

count factor and the change in market shares. Since acquiring firms’ innovation and production capabilities likely affect postmerger market shares in period t+1, we include time-invariant factors for the acquiring firm (firm j). We follow Wooldridge (2002) and add firm-level average 17

Please note that equation (7) could be separately estimated on the treated and untreated groups. For efficiency reasons, we combine the two groups for estimation.

12

ACCEPTED MANUSCRIPT

information on market shares (M S j ) and accumulated patents (AcP atj ). We follow this method instead of including firm fixed effects because our estimate for the discount factor does not vary with time. Firm fixed effects would encompass the estimated discount factors and make it impossible to separately determine the effects of the discount factor on merger decisions.18

CR IP T

Finally, we include two Heckman correction terms (HEC1 and HEC0).19 The first correction φ(Zβ) term (HEC1 = Mj,k,t Φ(Zβ) ) explains firms’ endogenous selection into mergers, where Z and β

represent the regressors and parameter estimates from the selection equation (8) shown below, φ is the standard normal density function, and Φ is the standard normal cumulative distribution φ(Zβ) ) becomes active when firms function. The second correction term (HEC0 = (1 − Mj,k,t ) 1−Φ(Zβ)

AN US

do not form a merger and is calculated in a similar manner. Finally, j,k,t denotes the error term.

To account for endogenous merger formation and to derive the two Heckman correction terms, we use a selection model that formulates firms’ decisions to merge. Firms simultaneously

M

decide if and with whom they want to merge. We have to consider all feasible pairwise merger opportunities since we allow every individual firm to be a potential merger candidate. We specify

ED

the selection equation according to firms’ incentives and the value they generate from merging. Hence, the decision for two firms j and k to form a merger in period t is based on a comparison ∗ and V ∗ be the between firms’ values when they merge and when they do not merge. Let Vj,t k,t

PT

present values of firms j and k in period t. The merged pair realizes a postmerger value of M + V M , which is the summation of the individual payoffs to the acquirer and target, and Vj,t k,t

CE

the superscript M refers to a merger. If the firms do not merge, they earn profits denoted by ∗ M + V M − (V + V ) > 0, where M ∗ Vj,t + Vk,t . Hence, firms form a merger if Mj,k,t = Vj,t j,t k,t k,t j,k,t is

AC

the latent variable measuring the underlying propensity to merge. Our selection model is based on a probit model where Mj,k,t represents a dummy variable and takes on a value of 1 if firms

j and k merge in period t; otherwise it is 0. Hence, if firms engage in a merger in period t, ∗ ∗ Mj,k,t = 1 and Mj,k,t > 0, and if they do not merge, Mj,k,t = 0 and Mj,k,t ≤ 0. The specification 18

With regard to the innovation variable, we also used firm-level average patent flows. The main results with regard to the discount factor and its interaction with the merger dummy were not significantly different. 19 The structure of the outcome equation and the formation of the Heckman correction terms build on the work of Cerulli (2012) and Heckman et al. (2006).

13

ACCEPTED MANUSCRIPT

of the selection equation looks as follows: ∗ C δ Mj,k,t = β0 + β1 ∆M j,k,t−1 + β2 ∆j,k + β3 T Rj,k,t−1 + β4 Same Regionj,k

+ β5 M S j + β6 AcP atj +

21 X

βy Y eart + τj,k,t .

(8)

The independent variables are each described in detail below.

CR IP T

y=7

In order to properly estimate the outcome equation (7), the selection equation (8) must contain instruments that impact the formation of mergers but do not impact the combined market share. We use four instruments in the selection equation (8): two instruments account

AN US

for variation across time and firm-pairs, and two further instruments account for variation across firm-pairs.

C The first instrument, ∆M j,k,t−1 , represents the relative absolute difference between marginal C costs the year before a potential merger occurs, ∆M j,k,t−1 =

|M Cj,t−1 −M Ck,t−1 | max(M Cj,t−1 ,M Ck,t−1 ) ,

and is based

on arguments made in other theoretical studies.20 The studies by Bergstrom and Varian (1985)

M

and Salant and Shaffer (1998; 1999) have shown that equilibrium quantities and prices in the industry (or a firm-pair) depend on the average costs. However, they have also shown that

ED

industry output and prices are independent of the distribution of marginal costs in the industry or between firm-pairs. A mean-preserving spread in marginal costs (i.e., an increase in the

PT

differences in marginal costs) will leave the equilibrium quantities unchanged (see also R¨oller, Siebert, and Tombak (2007) for a related argument). Hence, production is dependent on the

CE

sum or the average of the firms’ marginal costs, but independent of the difference in firms’ marginal costs. Therefore, the difference in marginal costs between firms will not have an

AC

impact on the outcome or production equation. It is important to note that an increase in firms’ differences in marginal costs increases firms’ and industry profits since more efficient firms produce more output at a lower cost, which increases firms’ profits (see Bergstrom and Varian (1985) and Salant and Shaffer (1999)). Thus, an increase in firms’ differences in marginal costs will increase the merging firms’ profits and, therefore, determine firms’ decisions to merge. Therefore, asymmetries between firms directly affect the market share only through the merger 20

As discussed later, we will perform robustness checks with regard to the variable definition.

14

ACCEPTED MANUSCRIPT

channel, and the difference in marginal costs is an appropriate instrument for merging. The identification argument is also statistically tested (see further below). Next, we discuss ∆δj,k which represents the difference between the acquirer’s and the target’s discount factors, ∆δj,k =

|δj −δk | max(δj ,δk ) .

Recent literature has shown that potential merger benefits

CR IP T

can result from merging firms having differing access to capital markets (see Erel, Jang, and Weisbach (2014)). As firms make the decision to merge, they will consider the potential benefits of acquiring a target that has better access to capital markets. Applying the findings by Bergstrom and Varian (1985), i.e., industry output is independent of the distribution of costs (capital costs), the difference in discount factors between merging firms does not have an ef-

AN US

fect on combined market shares and is, therefore, an appropriate instrument for the selection equation.

For our third instrument, we relate to studies in industrial organization and innovation that emphasize the importance of absorptive capacity in explaining the formation of mergers

M

and R&D cooperations. Potential synergy effects between two firms are used as an additional instrument. Synergy effects and complementaries between firms are important arguments that

ED

attract the main attention in merger decisions and merger studies. This is because higher potential synergy gains between merging firms can increase profits (see Farrell and Shapiro (1990) and Gugler and Siebert (2007)). It is important to note that the synergy effects argument

PT

is a reasonable instrument since it determines the extent to which two firms are able to benefit from synergy effects resulting in postmerger profit gains, which eventually determine firms’

CE

incentives to merge.21 Prominent studies highlight the fact that synergy effects between firms are determined by their absorptive capacities or technological relatedness (see, e.g., Levin and

AC

Reiss (1988), Cohen and Levinthal (1989), Kamien and Zang (2000), Cassiman and Veugelers (2002), and Kaiser (2002), among others). In following Kaiser (2002), Siebert (2015), Branstetter and Sakakibara (1998), L´ opez (2008), and Duso et al. (2014), we consider that firms more closely

related in the technological market are able to benefit from higher synergy effects.22 We use 21 A firm’s marginal costs rather depend on the number of patents (which almost exclusively reflects process innovations), as shown in equation (6). Consequently, marginal costs and patents will eventually determine a firm’s market share, as shown in equations (2-5), rather than the potential synergy effects that would be achieved between two firms. 22 See also Bloom, Schankerman, and Van Reenen (2005) and Siebert and von Graevenitz (2010).

15

ACCEPTED MANUSCRIPT

the uncentered correlation coefficient to measure technological relatedness and use it as a proxy for potential synergy effects between two firms, as suggested by Jaffe (1986) and Kaiser (2002). Using the USPTO technological classification, we establish the relatedness of firms’ levels of activities in different technological (sub)markets. We categorize all firms’ semiconductor patents

CR IP T

into 10 different technological classes.23 For each pair of firms, j and k, in period t, and in technological classification (c), we define Acj,t and Ack,t as firm-level variables that count the number of patent applications in each technological class. This results in the following measure for technological relatedness (T Rj,k,t ): P10

c c c=1 Aj,t Ak,t

qP

.

AN US

T Rj,k,t = qP 10

c c=1 Aj,t

10 c c=1 Ak,t

This measure results in a value from 0 to 1, where a value of 0 refers to firm-pairs with completely unrelated technological research and a value of 1 refers to firm-pairs that are active in the exact

M

same technological areas and are characterized by higher absorptive capacity and higher synergy effects.

ED

Our last instrument follows Dafny (2009) and defines Same Region as an indicator that equals 1 if the two firms are located in the same region (USA, Europe, Japan, or Other Regions), otherwise Same Region takes a value of 0. The dummy variable is an appropriate instrument

PT

and serves as the exclusion restriction in our selection equation. It accounts for important unobserved factors that determine firms’ merger decisions, such as cultural similarities (Ahern,

CE

Daminelli, and Fracassi, 2012) and regional/financial regulatory similarities. Moreover, regional similarities in a firm-pair do not have an impact on a firm’s production decision since the

AC

semiconductor industry is characterized by firms competing in international markets. Hence, firms’ origins do not determine their production decisions, especially since the unit of observation is a firm-pair. We also control for firm-level heterogeneity in the selection equation and apply the same procedure as for the outcome equation (7). We include averages of the acquiring firms’ market 23

We recover patents belonging to the semiconductor industry using the following technological classes: 257, 326, 360, 365, 369, 438, 505, 711, 712, and 714 (see also the USPTO Webpage for further information).

16

ACCEPTED MANUSCRIPT

shares and accumulated patents over time, as suggested by Wooldridge (2002). This controls for time-invariant unobservable factors that may affect the propensity to merge. This procedure is preferred to accounting for firm fixed effects since it considers multiple variables. Lastly, τj,k,t

3.3

Estimation Algorithm

CR IP T

denotes the error term.

The complete estimation process incorporates several steps. The details of each of the following will be discussed later and are also illustrated in Figure 3:

AN US

1. Estimation of price elasticity of demand (α1 ): We estimate the price elasticity of demand using the aggregate demand function at the semiconductor market level. 2. Retrieval of marginal cost and discount factors: Using the estimated elasticity, we estimate the firm’s supply relation in combination with each firm’s marginal cost to obtain

M

firm-specific discount factors and firm-/time-specific marginal costs. 3. Impact on product market: We estimate the impact of mergers on the product market

ED

by applying a heterogeneous treatment effects estimator. stat d • Using the estimated discount factors (δbi ) and the constructed marginal costs (M C i,t ),

PT

we estimate the probit selection equation to investigate firms’ incentives to merge (equation (8)).

CE

• Accounting for endogenous selection, we finally estimate the heterogeneous impact of

AC

mergers on market outcome (outcome equation (7)).

4

Estimation Results

According to our outlined estimation algorithm, we begin with the estimation of the price elasticity of demand.

17

ACCEPTED MANUSCRIPT

4.1

Price Elasticity of Demand

In a manner similar to Zulehner (2003) and Siebert (2010), we estimate the following demand

ln(Qt ) = α0 + α1 ln(Pt ) + α2 ln(GDPelec,t−1 ) + et .

CR IP T

equation:

(9)

We instrument for price and apply a two-stage least squares (2SLS) method using a supply shifter (input price of silicon, ln(Silct−1 ), which is the main input for semiconductor production) and a proxy for competition (number of firms, ln(N F irmst−1 )) as instruments for price. Controlling

AN US

for supply shifters enables us to trace out the slope of demand. Because we abstract from estimating a long run equilibrium model that endogenizes entry and exit, the number of firms is used as an instrument for price due to shifts in supply. A demand shifter (electronics GDP, ln(GDPelec,t−1 )) is also included, and it controls for demand shifts originated by changes in the

M

electronics (downstream) markets.

The results of the 2SLS estimation are shown in Table 3. The first-stage regression returns

ED

an F-value of 40.62 with a p-value of < 0.001 and an adjusted R-squared of 0.86, which confirms a good fit for our regression. The weak identification test implies strong instruments.24 The estimated price elasticity of demand (α1 ) takes on a value of −2.236, which is similar to the

PT

elasticity estimates of −1.5 to −2.3 seen in previous studies.25 The estimate for the GDP in electronics is positive and significant, illustrating the fact that higher GDP in electronics shifts

Marginal Cost and Discount Factor

AC

4.2

CE

the demand function outward.

Using the estimated price elasticity of demand, we continue with the estimation of the marginal costs and discount factors. Ideally, we would like to simultaneously estimate equations (5) and (6). However, a simultaneous estimation procedure would require estimating a firm fixed effect 24

The Cragg-Donald Wald F-statistic of 44.02 is larger than the Stock-Yogo 10% critical value of 19.23. See the following studies for further references on estimates on price elasticities of demand for semiconductors: Irwin and Klenow (1994), Webbink (1977), Wilson, Ashton, and Egan (1980), Finan and Amundsen (1986), Flamm (1993), and Baldwin and Krugman (1988). 25

18

ACCEPTED MANUSCRIPT

in the marginal cost and another firm fixed effect as the discount factor. Instead, we proceed with two different methods to circumvent this complication. The first method accounts for firm and time heterogeneity by including average accumulated patent applications (AcP at) and by including a year effect in each firm’s marginal cost (equation

CR IP T

(6)). The year effect in the marginal cost counts the number of years until 2004, starting at 16 and counting down to 1. The firm average accumulated patent applications and the year effects are used in the place of a firm fixed effect. This allows us to treat the discount factor as the only firm fixed effect in the estimation of equation (5).

Second, we separately estimate the dynamic marginal cost with firm fixed effects and then

AN US

include the predicted static marginal cost (dynamic marginal cost excluding a dynamic adjustment) into equation (5). Following Irwin and Klenow (1994), we estimate the dynamic marginal   MS dyn . cost as being equal to the price, adjusted for a firm-specific markup, Pt 1 + α1i,t = M Ci,t

M

Since the dynamic marginal cost consists of the static marginal cost and a dynamic compo  MS nent, we combine equation (6) with a dynamic adjustment term and estimate Pt 1 + α1i,t =

stat + λ Dynamic Adj + u . We allow the dynamic adjustment term to refer to the time M Ci,t 5 t i,t

ED

period in the product life cycle (1, 2, · · · , 16), which proxies the difference between static and dynamic marginal costs. Firms operating at the early stages of the life cycle are able to benefit from higher learning-by-doing effects and further increase output, such that dynamic marginal

PT

costs are further below static marginal costs (see also Zulehner (2003) and Siebert (2010)). A negative coefficient on the dynamic adjustment term reflects that dynamic marginal costs lie

CE

below static marginal costs. The error term is denoted as ui,t . To calculate the static marginal stat d costs (M C i,t ), we remove the effect of the dynamic adjustment (set λ5 = 0) from the predicted

AC

dynamic marginal costs. Finally, we estimate equation (5) while treating the discount factor as a firm fixed effect. In order to ensure reasonable results for the discount factor, we use a constrained linear

regression (SAS’s Proc Model) such that δi ∈ (0.667, 1) (which corresponds to a discount rate of ri ∈ (0, 0.5)). We exclude all estimates where the constraint is binding.26 Upon execution, 26

We performed a robustness check and estimated the discount factor without constraints and eventually dropped observations where the discount factor was greater than 1 or less than 0.667. This method reduced the sample to 156 firms, with an average discount factor of 0.88.

19

ACCEPTED MANUSCRIPT

both methods provide estimates for firm-specific discount factors that are highly correlated (corr(δ1 , δ2 ) = 0.778), where the subscripts refer to the two methods. We chose the first method since it yields more estimates for the discount factor and ensures more mergers in the final sample.27

CR IP T

For the first method, the supply equation (5), in combination with the marginal cost equation (6), is estimated using firm-level average accumulated patents and a year effect to account for potential firm-level heterogeneity. Additionally, we set the term

∂M Ci,t+1 ∂qi,t ,

which accounts for

changes in marginal cost resulting from current production, to a value of -0.1 (similar to Zulehner (2003)). Note that the direct estimation of this term would cause two potential problems. First,

AN US

it could have caused endogeneity concerns due to the use of current quantity entering this equation as an endogenous regressor in equation (5). Second, we would have had problems in separately identifying the value of this partial derivative and the discount factor; the discount factor is multiplied by this term and not separately identifiable. Therefore, we use the estimates

M

from previous studies to avoid this problem. We apply robustness checks and can verify that our results are not dependent on this assumption. We apply a grid search and specify values

ED

different from those in the previous literature, from -0.1 to -0.3. The resulting estimates for the discount factors associated with the different values are highly correlated (i.e., > 0.99), and the estimated discount factor has an average that ranges from 0.930 to 0.934.

PT

The estimated discount factors (δi ) from equation (5) are summarized in Table 4, Panel B, which shows that the average estimated discount factor is 0.931 (equivalent to a discount

CE

rate of ri = 0.076). Figures 3a and 3b show the distribution of the estimated discount factors and the discount rates, respectively. The discount factors and discount rates are quasi-normally

AC

distributed around the mean, with a higher (lower) median than the mean for the discount factor (rate). Overall, the estimation results provide evidence that discount factors differ between firms. Turning to the estimated static marginal cost equation (6), Table 4, Panel A, shows that the

27 Corresponding estimation results and summary statistics from the second method can be found in Table 7. In both estimation routines, if the estimated discount factor of a firm is pushed to the boundary (0.667 or 1), then that specific firm is dropped and the discount factor and the corresponding marginal cost are not included in the summary statistics (one of 229 firms is dropped for the first method and 34 of 229 firms are dropped for the second method). This explains the difference in observations between Table 4, Panel B, and Table 7, Panel B.

20

ACCEPTED MANUSCRIPT

coefficient for own learning (λ1 ) is negative and significant and provides a learning elasticity of -0.605, which corresponds to a learning rate of 34.25% (i.e., doubling accumulated production reduces marginal costs by 34.25%). Likewise, the significant coefficient λ2 is negative, indicating that spillover learning lowers the marginal cost of production. For spillover learning, the learning

CR IP T

elasticity is −8.908 which, when scaled by the average number of firms, results in a learning rate of 3.85%.28 In comparison with previous literature, the own learning rate is low. However, components of own learning may be picked up in the coefficients on patent applications and mean accumulated patents (Irwin and Klenow, 1994; Siebert, 2010). Moreover, our estimates show that the wage is significant and positive and captures upward shifts in marginal costs. Finally,

AN US

the controls for firm heterogeneity (accumulated patents) enter significantly and negatively. Using the estimated coefficients, we calculate the static marginal costs for each firm based on equation (6). See Table 4, Panel B, which shows a summary statistic.

Merger Formation

M

4.3

Using the results from the estimation of the discount factors and the marginal costs, we now

ED

discuss firms’ incentives to merge. Selecting on estimates of firms’ discount factors, marginal costs, and production the year before and after merging leaves us with 49 mergers.29 For descriptive purposes, we separate acquiring firms into two groups: acquirers characterized

PT

by low discount factors (impatient acquirers) and acquirers characterized by high discount factors (patient acquirers).30 We use the median discount factor as a threshold to separate both groups

CE

of acquiring firms. Table 5 shows the descriptive statistics for firms prior to engaging in a merger. Table 5, Panel A, shows that acquiring firms with low discount factors (an average

AC

of 0.87) represent the largest, most efficient, and most innovative firms among all merging ∂M C

ˆ 1 = α = −0.605. The own learning elasticities are evaluated using the following relationship: ∂ln(Acqi,t =λ i,t ) Spillover learning is adjusted by the average number of firms in the market, and we use the following: ˆ2 ∂M Ci,t /∂ln(Acq−i,t ) −8.908 = N Fλirms = α = 157.278 = −0.057. Learning rates are calculated from 1 − 2α , where α N F irms represents the respective learning elasticity. See Siebert (2010) and Zulehner (2003) for a similar procedure. 29 The selection equation is estimated based on 44 mergers since we removed five mergers that took place in the year immediately following another merger by the same acquiring firm. 30 We categorize the groups according to acquirers’ discount factors since these firms decide with whom they want to merge. In fact, tests show that unobserved heterogeneity does indeed come from the acquiring firms. The results are reported in Section 4.4. 28

21

ACCEPTED MANUSCRIPT

firms.31 These acquiring firms have an average market share of 2.3% and a marginal cost of $164.13. This type of acquirer merges, on average, with the second largest, second most efficient, and second most innovative group of merging firms. Hence, efficient and innovative impatient acquirers merge with slightly less efficient and innovative target firms. This finding is consistent

CR IP T

with established theoretical merger models that show that two efficient or large firms have higher incentives to jointly engage in a merger since they impose higher (premerger) negative competitive externalities on each other that can be internalized (see, e.g., Farrell and Shapiro (1990)). The focus on efficient and innovative targets suggests that efficiency benefits may be a relevant objective of the merger. This type of acquiring firm seems to adopt an acquisition

AN US

strategy characterized by shorter investment horizons (allowing for more instant efficiency gains) rather than realizing gains only in the more distant future.

Table 5, Panel B, shows that patient acquiring firms have an average discount factor of 0.936, which is 0.066 higher than the average discount factor of impatient acquiring firms, and this

M

difference is significantly different from zero with a p-value of 0.001. The patient acquiring firms represent the smallest, least efficient, and least innovative firms among all merging firms.

ED

Their average market share is 0.5%, and their marginal cost is $197.62. The marginal cost is $33.48 higher than the marginal costs of the impatient acquirers (significantly different from zero with a p-value of 0.001). The fact that these acquirers are characterized by the lowest

PT

efficiency levels across all merging firms suggests that they already faced difficulties in achieving efficiency gains premerger. These difficulties could have been related to unobserved factors such

CE

as managerial ability, etc. Interestingly, the patient acquirers merge with targets that are almost twice as large, more efficient, and more innovative than themselves.32 The self-selected sample

AC

of underperforming patient acquirers adopts a merger strategy that concentrates on targets that 31 It should be kept in mind that the summary statistics condition on merging firms, and this does not represent the entire population of all low discount factor firms. For example, Table 5, Panel A, reports acquiring firms with low discount factors. 32 In comparing the targets across both groups of acquirers (Panels A and B), it should be recognized that the target sizes are not significantly different from each other. However, the targets of patient acquirers are more inefficient. The average marginal cost of the targets of patient acquiring firms is $13.21 higher (significantly different from zero with a p-value of 0.035) than the marginal cost of the targets of impatient acquiring firms. Furthermore, the targets of patient acquirers are less innovative. The targets of patient acquiring firms have a patent stock of 80.64 patents, while the targets of impatient acquiring firms have an average of 794.31 accumulated patents; the difference of 713.67 is significantly different from zero with a p-value of 0.0154.

22

ACCEPTED MANUSCRIPT

are larger and more efficient than themselves. The difference in marginal costs between acquirers and targets is much lower for the patient acquirers compared to the impatient acquirers. These facts strongly suggest that patient acquirers adopt an acquisition strategy characterized by realizing fewer efficiency gains in the short run and potentially more efficiency gains in the long

CR IP T

run.33

Following Farrell and Shapiro (1990), we evaluate the merging firms’ changes in market shares before and after merging. If postmerger market shares decrease, then efficiency gains are rather small and the postmerger outcome is dominated by market power effects. If postmerger market shares increase, then efficiency gains are rather large and the postmerger outcome is

AN US

dominated by efficiency gains. Summary statistics show that combined acquirer and target market shares decline (from one year before to one year after the merger) for mergers by both types of acquirers. The sum of the market share for impatient acquiring firms and their targets declines from 3.29% to 2.68% (a 19% reduction), and the market share for patient acquiring

M

firms declines from 1.28% to 0.79% (a 38% reduction). Based on this summary, the changes in market shares indicate that both types of acquirers achieved relatively low efficiency gains

ED

that were not sufficient to overcompensate for the market power effects. Moreover, it should be noted that the percentage change in the decline for the market shares is less for mergers with impatient acquirers, which is indicative of these mergers achieving relatively higher efficiency

PT

gains than mergers with patient acquiring firms, at least in the short run. To summarize, patient and impatient acquirers merge with different types of targets, which

CE

is indicative of adopting different acquisition strategies. Impatient acquirers seem to aim toward realizing more instant efficiency gains and focus on shorter investment time horizons. Patient

AC

acquirers seem to adopt acquisition strategies that focus less on generating immediate efficiency gains and more on potential efficiency gains in the longer time horizon.

4.4

Heterogeneous Impact of Mergers

To further elaborate on the relationship between discount factors and postmerger outcomes, we evaluate the change in market shares before and after a merger occurred and compare this effect 33

We would like to thank a referee for having pointed us in this direction.

23

ACCEPTED MANUSCRIPT

to the change in market shares of nonmerging firms, as shown in equation (7). We estimate a heterogeneous treatment effects estimator by Heckman et al. (2006) and first report the results from firms selecting into mergers (the selection equation (8)), and then we discuss the impact of mergers on the market outcome (the outcome equation (7)).

estimating ρ =

σc2 , 1+σc2

CR IP T

In preparation for the estimation, we test for the presence of unobserved heterogeneity by where σc2 is the panel-level variance. The estimated ρ is the proportion of

total variance contributed by the panel-level variance component. When ρ is zero, the panel-level variance is unimportant, and the panel estimator is not different from the pooled estimator. To test whether unobserved heterogeneity is related to the acquiring firms, we estimate equation

AN US

(8) (without M S j and AcP atj ), which results in an estimate of ρ =0.949 and corresponds to a p-value < 0.0001. The tests confirm that heterogeneity is originated by the acquiring firms. Thus, in our model, we control for unobserved firm-specific heterogeneity using acquiring firms’ average size and innovation activity, M S j and AcP atj .

M

We next discuss the results from estimating equations (7) and (8). Table 6, column (1), shows the results of estimating the first-stage selection equation (8).34 The probit selection

ED

equation returns a returns a likelihood ratio (LR) chi-square value of 92.27, which corresponds to a p-value < 0.0001 and a pseudo R-squared of 0.123. We applied a Wald test to examine the joint significance of the instruments. The test returns a chi-square value of 36.09 (p-value

PT

< 0.0001), which verifies that the instrumental variables are not jointly equal to zero. In this regard, we find that the measure of technological relatedness (T R) shows a significantly positive

CE

impact on merger formation. This result emphasizes that firms more related in the technological market achieve cost savings from merging, as they benefit from synergy effects and eliminating

AC

duplicative effort. Moreover, the dummy for same region has a positive impact on merger formation, which indicates that unobserved firm-level factors, such as organizational and cultural differences, play an important role in merger formation. The controls for firm heterogeneities, 34

One potential problem might arise due to the small number of mergers in comparison to the large number of potential mergers. This is commonly referred to as a rare event problem. A potential solution is to use the ReLogit by King and Zeng (2001). The problem with using this approach is that it will not allow for the use of the Heckman correction in the outcome equation. However, we estimated several different specifications, and the results appear to be robust. Hence, we are confident that the low probability of merger and the associated flat cumulative density function will not cause major problems for the first stage.

24

ACCEPTED MANUSCRIPT

i.e., the market size (M S), positively impact merger formation. From the first-stage estimation, we derive the Heckman correction terms (HEC1 and HEC0), which enter the outcome equation (7). Next, we discuss the results of the outcome equation (7) (see Table 6, column (2)). The

CR IP T

estimates show that almost 69% of the combined market share is explained by the lagged market share. This result indicates that the time series on market share is highly persistent over time. The positive parameter estimate on the discount factor shows that more patient firms further increase output. This result is reasonable since, against the background of learning-by-doing, patient firms have a higher incentive to achieve future cost reductions via experience. Hence, they

AN US

increase current production, which serves as an investment into future experience. To provide further support and insight into this result, we relate our findings to two theoretical studies. Based on the study by Fudenberg and Tirole (1983) and their output equations (15) and (16) on page 527, we performed comparative statistics with respect to the discount factor. The

M

comparative statistic results show that higher discount factors result in production increases due to achieving future cost reductions via learning-by-doing. Fudenberg and Tirole (1983)

ED

mention that more patient firms (firms with higher discount factors) further increase output in the context of learning-by-doing. Related, Fudenberg and Tirole (1983) mention on page 528 that it is “. . . less likely to induce decreasing output if firms are impatient. . . This is intuitive: if

PT

firms are impatient, they have less incentive to increase output. . . ” We also performed the same comparative static exercise in a theoretical learning model by Thompson (2010) and received the

CE

same results. The positive relationship between the discount factor and a firm’s output provides further evidence and support for our positive parameter estimate on the discount factor.35

AC

It is important to note that both Heckman correction terms turn out to be significant. This

result provides evidence that firms self-select into mergers, i.e., unobserved firm attributes drive a firm’s decision to merge. Regarding the average treatment effect of mergers, the negative coefficient on the merger dummy (Mj,k,t ) indicates that mergers reduce market shares. This result indicates that, on 35

We would like to thank a referee for suggesting to further clarify the parameter estimate on the discount factor.

25

ACCEPTED MANUSCRIPT

average, efficiency gains are dominated by market power arguments if we ignore the interaction effect between mergers and the discount factor and if we compare the effect across nonmerging and merging firm-pairs. Our estimation results return a significantly negative parameter estimate for the interaction between mergers and discount factors. The parameter estimate provides

CR IP T

evidence that acquiring firms with low discount factors (impatient firms) increase postmerger output relatively more than acquiring firms characterized by high discount factors (patient firms). In quantifying the heterogeneous postmerger outcomes between both groups of acquirers, we compare the postmerger changes in market shares for impatient acquiring firms with the patient acquiring firms. Recall that in equation (7), we follow Cerulli (2012) and demean the

AN US

acquiring firm’s discount factor by the sample average of 0.931 in the interaction term to obtain the heterogeneous treatment effect.

For impatient acquiring firms, we find that the combined market share increases by 0.00311 (i.e., -0.051*(0.870-0.931)), which represents a 44.51% increase from the sample average (i.e.,

M

0.00311/0.007) and a 9.44% increase from the average combined market share of the impatient acquiring firms and their targets (i.e., 0.00311/0.033). In contrast, for patient acquiring firms,

ED

we find that the combined market share decreases by 0.000264 (i.e., -0.051*(0.936-0.931)), which represents a 3.77% decrease from the sample average (i.e., -0.000264/0.007) and a 2.03% decrease from the average combined market share of a patient acquiring firms and their targets (i.e., -

PT

0.000264/0.013). When compared to the sample average, this results in an absolute difference of 48.29% (i.e., 44.52% - (-3.77%)) or, when compared with the average market shares of impatient

CE

and patient acquiring firms, it results in an absolute difference of 11.47% (i.e., 9.44% - (-2.03%)). Hence, impatient acquiring firms increase their market share, which provides evidence that they

AC

achieve higher postmerger efficiency gains than acquiring firms with high discount factors (at least in the limited time period). One explanation of this result is that merging firms with different discount factors are characterized by different abilities to achieve efficiency gains, which could be related to their organizational structure, managerial ability, etc. A further related explanation could be that acquirers with different discount factors adopt different acquisition strategies, as also mentioned in the discussion of Table 5.

26

ACCEPTED MANUSCRIPT

Remember, Table 5 has shown the summary statistics of acquirers with low and high discount factors, and we recognized that different acquisition strategies appear to play a major role in explaining the relationship between acquirers’ discount factors and the postmerger outcomes. We recognized that both types of acquirers seemed to focus on achieving postmerger efficiency gains,

CR IP T

but they differed in their efficiency levels and merged with targets of different efficiency levels. Acquiring firms with low discount factors are the largest, most efficient, and most innovative firms across all merging firms. These acquirers merge with the second largest, second most efficient, and second most innovative group of merging firms. We concluded earlier that the acquisition strategy of acquirers with low discount factors is characterized by a shorter time

AN US

horizon, i.e., these acquirers put more emphasis on efficiency gains that can be realized in the short run.

In contrast, acquirers with high discount factors seemed to face premerger difficulties in realizing efficiency gains that could have been related to unobserved factors. These acquirers seem to

M

adopt longer time horizon acquisition strategies and put more emphasis on returns and efficiency gains that are realized in the more distant future.36 To summarize, heterogeneous postmerger

ED

outcomes can be explained to some extent by firms’ discount factors in association with their acquisition strategies. Moreover, it is important to remember from Table 5 that mergers with impatient acquiring firms involve more accumulated patents than mergers with patient acquir-

PT

ers. Even though we control for patents, this observation is suggestive that innovation may play an important role in the relationship between the discount factor and merger outcomes; this

CE

deserves more attention in future research studies.37 To properly evaluate the merger effects between patient and impatient firms, we need to

AC

account for the average treatment effect on the treated (merging) firms (AT ET ) and incor-

porate the average treatment effect of merging, the heterogeneous effect of merging related to the discount factor, and the selection adjustments. We follow Cerulli (2012) and account for heterogeneity in acquirers’ discount factors (δj ) using the regression coefficients from equation 36

We would like to thank a referee for pointing us toward the above-mentioned arguments. In unreported tests, we applied several robustness checks in estimating equations (7) and (8) using alternative variable definitions, i.e., we used relative and absolute differences in firm-pairs. Most importantly, the estimation results show that the interaction effect of the discount factor and the merger dummy remained negative and significant. 37

27

ACCEPTED MANUSCRIPT

(7) in the subsequent equation:   ¯ ∗ ρ3 + (ρ9 + ρ10 ) ∗ φ(Xβ) AT ET (δj ) = ρ2 + (δj − δ) , Φ(Xβ) (Mj,k,t =1)

(10)

CR IP T

where ρ2 is the coefficient on the merger variable, (δj −δ) is the deviation of the firm-level discount factor from the mean, ρ3 is the coefficient on the interaction between the merger variable and the discount factor, and ρ9 and ρ10 are the coefficients on the two Heckman correction terms to control for selection.38 The ATET is less negative for impatient acquirers (-0.796%) than for patient acquirers (-0.898%). The difference between the two group’s AT ET is significantly

AN US

different from zero, with a p-value of 0.0004. This implies that efficiency benefits were relatively greater for impatient firms. These results are consistent with the summary statistics discussed in Section 4.3, showing an overall decline in market share for acquisitions by both types of acquiring firms, but a lesser decline for acquisitions by impatient acquiring firms.

Conclusion

M

5

ED

Our study assesses the relationship between firm-level discount factors, incentives to form mergers, and the resulting postmerger impact on the product market for firms with high and low

PT

discount factors. We estimate firm-specific discount factors from firms’ supply relations and estimate a heterogeneous treatment effects model.

CE

Our results show that acquiring firms characterized by low discount factors (impatient firms) merge with the second most efficient and innovative group of merging firms. We also find that

AC

impatient acquirers realize relatively higher postmerger efficiency gains, at least in the short run. The findings provide evidence that these acquirers apply merger strategies that put more weight on realized efficiency gains and value added in the short term. In contrast, acquiring firms characterized by high discount factors (patient firms) are rather small and inefficient before the merger, and they merge with firms that are larger than themselves and about as efficient. These targets are smaller than the targets acquired by firms with low discount factors. The estimation 38

Note, ρ2 , ρ3 , ρ9 , and ρ10 are all significantly different than zero.

28

ACCEPTED MANUSCRIPT

results show that mergers with patient acquiring firms generate relatively fewer efficiency gains in the short run. This finding supports the notion that patient acquiring firms apply acquisition strategies that put more weight on long run efficiency gains. In general, our results emphasize the importance of firms’ discount factors and firms’ merger strategies as an insightful tool for

CR IP T

forecasting the effect of mergers.

To conclude, this study provides evidence that the acquirers’ discount factors relate to specific types of mergers and provide useful information when evaluating the postmerger impact. From a policy point of view, our study suggests that firm-level discount factors may be useful in merger analyses for predicting the timing of when efficiency gains are realized and in evaluating the

AN US

dominance between market power and efficiency effects over time. Policy and antitrust authorities might want to consider using information on firm-level discount factors when determining

AC

CE

PT

ED

M

the impact of mergers.

29

ACCEPTED MANUSCRIPT

References Aguirregabiria, Victor, and Arvind Magesan, 2013, Structural Econometric Models (Advances in Econometrics) . chap. Euler Equations for the Estimation of Dynamic Discrete Choice

CR IP T

Structural Models, 3–44 (Emerald Group Publishing Limited). Ahern, Kenneth R., Daniele Daminelli, and Cesare Fracassi, 2012, Lost in translation? The effect of cultural values on mergers around the world, Journal of Financial Economics, 117, 165–189.

AN US

Andersson, Ola, 2008, On the role of patience in collusive bertrand duopolies, Economics Letters 100, 60 – 63.

Angrist, Joshua D., and Alan B. Krueger, 1999, Empirical strategies in labor economics, vol. 3 of Handbook of Labor Economics . chap. 23, 1277–1366 (Elsevier).

M

Baldwin, Richard E., and Paul R. Krugman, 1988, Empirical Methods for International Trade . chap. Market Access and International Competition: A Simulation Study of 16K Random

ED

Access Memories (Combridge Mass: MIT Press). Bergstrom, Theodore C., and Hal R. Varian, 1985, Two remarks on cournot equilibria, Eco-

PT

nomics Letters 19, 5–8.

Berry, Steve, and Ariel Pakes, 2000, Estimation from the optimality conditions for dynamic

CE

controls, Working Paper.

AC

Bloom, Nicholas, Mark Schankerman, and John Van Reenen, 2005, Identifying technology spillovers and product market rivalry, CEP Discussion Papers dp0675 Centre for Economic Performance, LSE.

Brand, Jennie E., and Yu Xie, 2010, Who Benefits Most from College? Evidence for Negative Selection in Heterogeneous Economic Returns to Higher Education, American Sociological Review 75, 273–302.

30

ACCEPTED MANUSCRIPT

Brand, Jennie E., and Juli Simon Thomas, 2013, Causal effect heterogeneity, in Stephen L. Morgan, ed.: Handbook of Causal Analysis for Social Research Handbooks of Sociology and Social Research . 189–213 (Springer Netherlands).

CR IP T

Branstetter, Lee, and Mariko Sakakibara, 1998, Japanese research consortia: A microeconometric analysis of industrial policy, The Journal of Industrial Economics 46, 207–233.

Cassiman, Bruno, and Reinhilde Veugelers, 2002, R&D Cooperation and Spillovers: Some Empirical Evidence from Belgium, American Economic Review 92, 1169–1184.

AN US

Cerulli, Giovanni, 2012, Ivtreatreg: a new STATA routine for estimating binary treatment models with heterogeneous response to treatment under observable and unobservable selection, CERIS Working Paper 201203 Institute for Economic Research on Firms and Growth - Moncalieri (TO).

Cohen, Wesley M., and Daniel A. Levinthal, 1989, Innovation and learning: The two faces of R

M

& D, The Economic Journal 99, pp. 569–596.

ED

Compte, Olivier, Frederic Jenny, and Patrick Rey, 2002, Capacity constraints, mergers and collusion, European Economic Review 46, 1–29.

PT

Dafny, Leemore, 2009, Estimation and identification of merger effects: An application to hospital mergers, Journal of Law & Economics 52, 523–550.

CE

Davis, Peter J., and Christian Huse, 2010, Estimating the ‘coordinated effects’ of mergers, Working Paper.

AC

de Roos, Nicolas, 2004, A model of collusion timing, International Journal of Industrial Organization 22, 351 – 387.

Dehejia, Rajeev H., and Sadek Wahba, 2002, Propensity Score-Matching Methods For Nonexperimental Causal Studies, The Review of Economics and Statistics 84, 151–161. Diamond, Douglas W., and Robert E. Verrecchia, 1991, Disclosure, liquidity, and the cost of capital, The Journal of Finance 46, 1325–1359. 31

ACCEPTED MANUSCRIPT

Duso, Tomaso, Lars-Hendrik R¨ oller, and Jo Seldeslachts, 2014, Collusion Through Joint R&D: An Empirical Assessment, The Review of Economics and Statistics 96, 349–370. Easley, David, and Maureen O’hara, 2004, Information and the cost of capital, The Journal of

CR IP T

Finance 59, 1553–1583. Erel, Isil, Yeejin Jang, and Michael S. Weisbach, 2014, Do acquisitions relieve target firms’ financial constraints?, The Journal of Finance, 70, 289–328.

Farrell, Joseph, and Carl Shapiro, 1990, Horizontal mergers: An equilibrium analysis, American

AN US

Economic Review 80, 107–26.

Finan, William F., and Chris B. Amundsen, 1986, Modeling U.S.–Japan competition in semiconductors, Journal of Policy Modeling 8, 305–326.

Flamm, Kenneth, 1993, Forward Pricing versus Fair Value: An Analytical Assessment of

M

‘Dumping’ in DRAMS (Chicago: Chicago University Press).

Fluck, Zsuzsanna, and Anthony W Lynch, 1999, Why do firms merge and then divest? A theory

ED

of financial synergy, The Journal of Business 72, 319–46. Fudenberg, Drew, and Jean Tirole, 1983, Learning-by-doing and market performance, The Bell

PT

Journal of Economics 14, 522–530.

CE

Gowrisankaran, Gautam, 1999, A dynamic model of endogenous horizontal mergers, The RAND Journal of Economics 30, pp. 56–83.

AC

Gugler, Klaus, and Ralph Siebert, 2007, Market power versus efficiency effects of mergers and research joint ventures: Evidence from the semiconductor industry, The Review of Economics and Statistics 89, 645–659.

Hall, Bronwyn H., Adam B. Jaffe, and Manuel Trajtenberg, 2001, The NBER patent citation data file: Lessons, insights and methodological tools, Working Paper 8498 National Bureau of Economic Research.

32

ACCEPTED MANUSCRIPT

Harrington, Joseph Jr., 1989, Collusion among asymmetric firms: The case of different discount factors, International Journal of Industrial Organization 7, 289–307. Heckman, James J., Sergio Urzua, and Edward Vytlacil, 2006, Understanding instrumental

389–432.

CR IP T

variables in models with essential heterogeneity, The Review of Economics and Statistics 88,

Irwin, Douglas A, and Peter J Klenow, 1994, Learning-by-doing spillovers in the semiconductor industry, Journal of Political Economy 102, 1200–1227.

AN US

Jaffe, Adam B, 1986, Technological opportunity and spillovers of R&D: Evidence from firms’ patents, profits, and market value, American Economic Review 76, 984–1001. Jovanovic, Dragan, and Christian Wey, 2012, An equilibrium analysis of efficiency gains from mergers, DICE Discussion Papers 64 Heinrich-Heine-Universit¨at D¨ usseldorf, D¨ usseldorf Insti-

M

tute for Competition Economics (DICE).

Kaiser, Ulrich, 2002, An empirical test of models explaining research expenditures and research

ED

cooperation: evidence for the German service sector, International Journal of Industrial Organization 20, 747–774.

PT

Kamien, Morton I., and Israel Zang, 2000, Meet me halfway: research joint ventures and absorptive capacity, International Journal of Industrial Organization 18, 995–1012.

CE

Khatami, Seyed Hossein, Maria-Teresa Marchica, and Roberto Mura, 2015, Corporate acquisitions and financial constraints, International Review of Financial Analysis 40, 107 – 121.

AC

King, Gary, and Langche Zeng, 2001, Logistic regression in rare events data, Political Analysis 9, 137–163.

Levin, Richard C., and Peter C. Reiss, 1988, Cost-reducing and demand-creating R&D with spillovers, The RAND Journal of Economics 19, pp. 538–556.

Liu, An-Hsiang, Ralph Siebert, and Christine Zulehner, 2013, Excessive entry and social inefficiencies: A policy experiment on dunamic efficiency gains, Working Paper. 33

ACCEPTED MANUSCRIPT

L´ opez, Alberto, 2008, Determinants of R&D cooperation: Evidence from Spanish manufacturing firms, International Journal of Industrial Organization 26, 113 – 136. Merton, Robert C., 1974, On the pricing of corporate debt: The risk structure of interest rates,

CR IP T

The Journal of Finance 29, 449–470. Morgan, Stephen L., and Christopher Winship, 2007, Counterfactuals and Causal Inference Methods and Principles for Social Research (Cambridge University Press) 1st edn.

Mueller, Dennis C, 1985, Mergers and Market Share, The Review of Economics and Statistics

AN US

67, 259–67.

Pais, Jeremy, 2011, Socioeconomic background and racial earnings inequality: A propensity score analysis, Social Science Research 40, 37 – 49.

Perry, Martin K, and Robert H Porter, 1985, Oligopoly and the incentive for horizontal merger,

M

American Economic Review 75, 219–27.

Pesendorfer, Martin, 2003, Horizontal mergers in the paper industry, RAND Journal of Eco-

ED

nomics 34, 495–515.

R¨ oller, Lars-Hendrik, Ralph Bernd Siebert, and Mihkel M. Tombak, 2007, Why Firms Form (or

PT

do not Form) RJVS, Economic Journal 117, 1122–1144. Salant, Stephen W., and Greg Shaffer, 1998, Optimal asymmetric strategies in research joint

CE

ventures, International Journal of Industrial Organization 16, 195–208. , 1999, Unequal treatment of identical agents in cournot equilibrium, American Economic

AC

Review 89, 585–604.

Salant, Stephen W, Sheldon Switzer, and Robert J Reynolds, 1983, Losses from horizontal merger: The effects of an exogenous change in industry structure on Cournot-Nash equilibrium, The Quarterly Journal of Economics 98, 185–99. Siebert, Ralph, 2015, What determines firms’ choices between ex ante and ex post licensing agreements?, Journal of Competition Law and Economics. 34

ACCEPTED MANUSCRIPT

, and Georg von Graevenitz, 2010, Jostling for advantage or not: Choosing between patent portfolio races and ex ante licensing, Journal of Economic Behavior & Organization 73, 225–245.

CR IP T

Siebert, Ralph B., 2010, Learning-by-doing and cannibalization effects at multi-vintage firms: Evidence from the semiconductor industry, The B.E. Journal of Economic Analysis & Policy 10, 1–32.

Stigler, George J., 1964, A theory of oligopoly, Journal of Political Economy 72, 44–61.

Thompson, Peter, 2010, Learning by doing, vol. 1 of Handbook of the Economics of Innovation

AN US

. chap. 10, 429–476 (Elsevier).

Tombak, Mihkel M., 2002, Mergers to monopoly, Journal of Economics & Management Strategy 11, 513–546.

Vasconcelos, Helder, 2005, Tacit collusion, cost asymmetries, and mergers, RAND Journal of

M

Economics 36, 39–62.

ED

Webbink, Douglas A., 1977, The Semiconductor Industry: A Survey of Structure, Conduct, and Performance (Washington: Federal Trade Commission).

PT

Williamson, O.E, 1968, Economies as an antitrust defense: The welfare tradeoffs, American Economic Review 58.

CE

Wilson, Robert W., Peter K. Ashton, and Thomas P. Egan, 1980, Innovation, Competition, and Government Policy in the Semiconductor Industry (Lexington, Mass: Lexington Books).

AC

Wooldridge, J.M., 2002, Econometric Analysis Cross Section Panel Data (Massachusetts Institute of Technology 2002).

Wright, Theodore P., 1936, Factors affeting the cost of airplanes, Journal of Aeronautical Sciences 3, 122–128. Zulehner, Christine, 2003, Testing dynamic oligopolistic interaction: evidence from the semiconductor industry, International Journal of Industrial Organization 21, 1527–1556.

35

ACCEPTED MANUSCRIPT

A

Appendix: Tables Table 1: Industry Description Firm Revenue

Mergers

N Firms

1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 Average

2 3 3 5 4 5 6 7 7 11 18 21 11 11 13 6 8.31

130 138 130 155 151 152 195 182 187 205 193 155 166 169 200 201 169.31

52,720 54,571 59,310 64,705 85,184 109,181 171,281 160,685 159,799 149,120 184,866 226,766 151,954 155,629 178,242 219,880 136,493.31

CR IP T

Year

Industry Revenue

M

AN US

405.538 395.442 456.231 417.452 564.132 718.296 878.364 882.885 854.54 727.415 957.855 1,463.01 915.386 920.882 891.21 1,093.93 783.91

ED

Table 1: Summary of the number of mergers, number of firms, industry revenue (million $), and revenue per firm (million

AC

CE

PT

$). The data are provided by SDC Platinum and the Gartner Group.

36

ACCEPTED MANUSCRIPT

Table 2: Variable Summary Statistics

Pt M Si,t qi,t Acqi,t

Semiconductor price Market share per firm Quantity produced Accumulated quantity produced Accumulated quantity produced by others Annual patent applications Accumulated patent applications, with 5% depreciation Semiconductor wage, PPI adjusted Number of semiconductor firms U.S. Electronics GDP

Acq−i,t P ati,t AcP ati,t

W aget N F irmst GDPelec,t

N

MEAN

STD

MIN

MAX

1,829 1,829 1,829 1,829

71.983 0.007 14.560 54.096

16.765 0.014 34.644 157.281

45.903 0.000 0.014 0.000

96.927 0.113 540.580 2,386.690

1,829

11,188.070

9,018.520

485.732

29,443.310

1,829

36.266

98.529

0.000

1,020.000

1,829

227.044

640.828

0.000

4,973.760

CR IP T

Label

AN US

Variables

1,829

13.162

0.975

11.476

14.521

1,829

157.278

19.400

124.000

189.000

1,829

41.696

2.357

38.100

45.600

M

Table 2: Summary statistics are shown for variables that are used for model estimation. Sources and methodologies are described in Section 2.

ED

Table 3: Elasticity Estimation (1) Dep. Var: ln(Qt )

PT

Variables

AC

CE

Constant

7.935 ( 5.935) -2.236*** (0.321) 2.440* (1.370)

ln(Pt ) ln(GDPelec,t−1 )

Observations Adjusted R-Squared

15 0.860

Table 3: Price elasticity of demand estimation as shown in equation (9) using 2SLS. Dependent variable is ln(Qt ). The

following instruments for price are used: number of firms and material price of silicon. Standards errors in parentheses, *** (**,*) denotes 1% (5%, 10%) level of significance.

37

ACCEPTED MANUSCRIPT

Table 4: Static Marginal Cost/Discount Factor Estimation Panel A: Static Marginal Cost Coefficients (1) Dep. Var: Price adj. for markup

Variables

204.194*** (13.198) -0.605*** (0.168) -8.908*** (0.696) 55.299*** (2.178) -0.008*** (0.002) -8.295*** (1.450)

CR IP T

Constant Own Learning Spillover Learning Wages

AN US

Patent Applications Mean Accumulated Patents

1,592 0.982 Yes Yes***

M

Observations Adjusted R-Squared Discount Factor FE Year Effect

δi ri MC

N 228 228 1,829

MEAN

STD

MIN

MAX

5th %

MED

95th %

0.931 0.076 187.679

0.034 0.044 23.277

0.709 0.038 124.757

0.963 0.410 236.573

0.846 0.043 145.281

0.942 0.061 190.227

0.959 0.182 223.700

PT

Variables

ED

Panel B: Summary Statistics for Estimated Factors

AC

CE

Table 4 Panel A: Estimation of marginal cost (M Ci,t ) and discount factor (δi ) from equation (5) and equation (6), substituting mean accumulated patents for the firm fixed effect. The price adjusted for firm markup is the dependent variable. The results are obtained using constrained OLS methods. The estimation includes a firm-specific discount factor and year effect. The summary statistics for the discount factor (δi ) are shown in Panel B and in Figure 4. Standards errors in parentheses, *** (**,*) denotes 1% (5%, 10%) level of significance. Panel B: Descriptive statistics are shown for estimated δi , ri , and marginal costs. The δi is estimated as a firm fixed effect according to equation (5). The estimation routine constrained δi such that δi ∈ (0.66, 1). All boundary estimates for δi were dropped (two firms). The interest rate, ri , is calculated as ri = δ1 − 1. The marginal cost is calculated based on i equation (6).

38

CR IP T

ACCEPTED MANUSCRIPT

Table 5: Merger Summary Statistics

0.870 0.154 0.023 55.721 164.127 223.674 181.800 1,038.580

0.050 0.073 0.018 47.590 16.895 270.359 202.242 1,085.110

0.909 0.103 0.010 23.280 180.268 116.356 131.080 794.307

0.044 0.056 0.013 30.411 24.123 166.325 229.451 1,363.100

M

δi ri M Si,ti −1 qi,ti −1 M Ci,ti −1 Acqi,ti −1 P ati,ti −1 AcP ati,ti −1

Panel B: Mergers with High δ Acquirers Acquirers Targets MEAN STD MEAN STD 0.936 0.068 0.005 12.634 197.612 30.943 7.500 35.990

0.009 0.011 0.006 15.152 12.299 48.264 9.027 53.690

AN US

Variable

Panel A: Mergers with Low δ Acquirers Acquirers Targets MEAN STD MEAN STD

0.926 0.082 0.008 21.777 193.479 90.645 31.833 80.640

0.028 0.034 0.009 25.998 18.211 133.495 58.968 135.817

Table 5: Summary statistics are shown for the acquiring firm and the target firm for completed mergers. Results are shown for mergers by acquirers with low and high discount factors, as determined by median acquirer discount factor. All time

AC

CE

PT

ED

variant measures are shown for the year before the merger.

39

ACCEPTED MANUSCRIPT

Table 6: Change in Market Share Following Merger

C ∆M j,k,t−1

∆δj,k Same Regionj,k T Rj,k,t−1

AcP atj Constant M Sj,k,ti −1 Mj,k,ti

δj

ED

N F irmst

M

Mj,k,ti ∗ (δj − δ)

GDPelec,t

PT

HEC1j,k,t HEC0j,k,t

CE

0.294*** (0.008) -2.26e-06*** (1.23e-07) -0.011*** (0.003) 0.685*** (0.003) -0.046*** (0.012) -0.051* (0.028) 0.006** (0.003) -3.68e-05*** (4.26e-06) 2.88e-04*** (4.37e-05) 0.012*** (0.004) -0.262*** (0.007)

AN US

M Sj

0.520 (0.879) -2.669 (2.855) 0.517*** (0.112) 0.521*** (0.136) 22.593*** (7.302) 2.71e-05 (1.51e-4) -4.055*** (0.244)

(2) M Sj,k,t+1

CR IP T

(1) Mj,k,t

VARIABLES

Observations Adjusted R-Squared Year FE

82,977 0.123 Yes

79,091 0.637 Yes

AC

Table 6: Estimation of a heterogeneous treatment effects selection model (as discussed in Section 4.4) for the change in market share from the year before the merger to the year after the merger. Column (1) provides results from estimating the probit selection equation (8) and column (2) provides results from estimating the outcome equation (7) using OLS for the year after the merger. Standards errors in parentheses, *** (**,*) denotes 1% (5%, 10%) level of significance.

40

ACCEPTED MANUSCRIPT

Table 7: Robustness: Fixed Effect Marginal Cost Estimation Panel A: MC Estimation (1) Dep. Var: Price adj. for markup

Variables

-100.936*** (3.584) -0.158** (0.066) -4.019*** (0.183) 85.807*** (1.461) -0.002*** (0.001) -1.417*** (0.071)

CR IP T

Constant Own Learning Spillover Learning Wages

Dynamic Adjustment

1,853 0.993 Yes

M

Observations Adjusted R-Squared Firm FE

AN US

Patent Applications

Panel B: Summary Statistics for Estimated Factors N

MEAN

STD

MIN

MAX

1,601 1,601 196 196

71.770 84.664 0.860 0.173

16.673 10.850 0.080 0.116

41.300 63.982 0.671 0.004

98.064 100.899 0.996 0.491

ED

Variable dyn

PT

MC stat MC δi ri

CE

Table 7: Estimation of the marginal cost from equation (6) with price adjusted for firm markup as the dependent variable. To incorporate the difference between the dynamic and static marginal cost, a dynamic adjustment term is included that counts from 1 to 16 for each year. The static marginal cost is calculated by removing the dynamic adjustment component. The predicted static marginal cost is included in equation (5), and the equation is estimated to obtain the discount factors

AC

at the firm level. Standards errors in parentheses, *** (**,*) denotes 1% (5%, 10%) level of significance. The summary statistics for the estimated marginal costs and discount factors are shown in Panel B.

41

ACCEPTED MANUSCRIPT

Appendix: Figures

AN US

CR IP T

B

AC

CE

PT

ED

M

Figure 1: Firm Quantity by Year from 1990-2004 Source: Gartner Group

Figure 2: Accumulated Patents by Year from 1990-2004 using a 5% depreciation rate Source: NBER Patent Database (Hall, Jaffe, and Trajtenberg, 2001)

42

ACCEPTED MANUSCRIPT

Demand estimation, equation (9)

Price elasticity of demand

Estimation of dynamic supply relationship, equations (5 and 6)

• Firm level discount factor • Firm/year marginal costs

Overall impact of merging on market shares, equation (7)

• Regression Coefficients • Heckman Correction Terms

Average effect of a merger for merging firms (ATET), equation (10)

Dominance of market power or efficiency arguments

(a) δi Histogram

M

AN US

CR IP T

Figure 3: Flowchart depicting the complete estimation process. The bold text describes the main steps and the subtext represents the outcome that is used in the next step.

(b) ri Histogram

AC

CE

PT

ED

Figure 4: Histogram: firm-level discount factor and discount rate estimates for 228 firms

43