Do the decisions of peer firms impact individual firm strategies? evidence from corporate investments in the United States

(1)

Amsterdam Business School

Master Thesis

Do the Decisions of Peer Firms Impact Individual Firm

Strategies? Evidence from Corporate Investments in the

United States

.

MSc Finance

Track: Quantitative Finance

Author: Ieva Augustinaite

Student number: 10554645

Thesis supervisor: Dr. Evgenia Zhivotova

1

st

_{of July, 2018}

(2)

Statement of originality

This document is written by Student Ieva Augustinaite who declares to take full responsibility for the contents of this document.

I declare that the text and the work presented in this document are original and that no sources other than those mentioned in the text and its references have been used in creating it.

The Faculty of Economics and Business is responsible solely for the supervision of completion of the work, not for the contents.

(3)

Abstract

Recent studies have provided increasing evidence that peer firms play an important role in determining corporate financial policy. Yet, empirical evidence about the magnitude of the imitation behavior is inconsistent and potential mechanisms through which peer effects operate have not been extensively discussed. The study investigates the influence of peer firms in corporate investment strategies for the sample of public U.S. firms for the period 1989 - 2017. Using 2SLS methodology, employing idiosyncratic equity returns to instrument for the average peer investments, I find consistent evidence supporting the presence of positive peer effects in capital investments. The results are shown to be robust across different peer identification criteria and the control groups. Less pronounced results are shown for the class of intangible investments, namely, R&D and acquisitions. Furthermore, the study provides new evidence for the role of macro-economic, micro-economic uncertainty and the recession on peer effects. The result suggest that equities market and economic policy uncertainty amplify the imitation behavior. Finally, I find empirical evidence, which enables to infer that firms, subject to higher industry, firm-specific uncertainty and information asymmetry are more sensitive to the peer firms’ behavior.

(4)

1. Introduction

Conventional corporate finance literature often assumes that decisions made by the firms’ authorities are independent of the ones made by their peers and are primarily determined by the firm-specific characteristics and circumstances (Graham & Harvey, 2001). However, recent literature has provided growing empirical evidence that CEO’s take into account the behavior other companies for own financial decision-making. Not only corporations are attentive to the behavior of their peers, but in some cases, this leads to firms startingmimicking each other for various strategic reasons. The lawsuit of Samsung copying the design features of iPhone involving over a billion dollars in damage claimed by Apple is only one of the numerous examples of the presence of imitation behavior in business domains (The New York Times, 2018). The latter phenomenon in finance is known as ‘peer effect’ or the ‘spillover effect’. While imitation, peer influence and learning behavior have been extensively studied in sociological, behavioral fields, marketing and advertising (Hoxby, 2000; Sacerdote, 2001; Falk & Ichino, 2006; Zimmerman, 2003; Bertrand, 1883), the empirical evidence in the financial sector remains scarce.

Leary and Roberts (2014) attempt to quantify the peer effects in corporate operational and capital structure decisions, concluding that not only peer firms have a substantial influence on the individual firm’s behavior, but the latter explains capital structure decisions better than many accounting measures recognized in previous literature. In the investments framework, the extent to which other firms influence own policy has been often ignored or assumed to indirectly operate through firm-specific determinants (Chen & Ma, 2017). Additionally, the investigation about the spillover effect has been often dismissed due to the complexity of the phenomenon (Manski, 1993). Most of the existing research has chosen to follow Tobin‘s Q or the theory of financial constraints to explain the way the corporations invest (Tobin, 1969). Despite the simplicity of the Tobin’s Q, it has been associated with poor empirical applicability and performance (Erickson & Whited, 2000; Fazzari et al., 1988; Leary & Roberts, 2014). The theory is restricted by several simplifying assumptions that are hardly applicable in real economies, such as no information asymmetry or perfect market conditions (Akerlof, 1970). Violation of these assumptions together with a large part of the variance in investment behavior remaining unexplained were the primary reasons for scholars to start questioning the reliability of the conventional theories and search for other potential determinants of corporate investments.

It has been shown that corporate investments are positively correlated within industries (Chen & Ma, 2017). One of the possible explanations for the latter is that firms, which face similar regulation, consumer demand or sell homogenous products are simply exposed to common underlying shocks,

(7)

leading to a similarity in the firms’ investment behavior. However, it can also be that it is the others’ influence, which leads to the commonality in behavior. In investment context, this would imply that firms are directly affected by the changes in peers’ investments. The direction of this relationship is dependent on the underlying economic channels driving the relationship. Learning or information-based models predict that firms respond positively to others’ behavior (Lieberman & Asaba, 2006). Imitation arises when firms that are not perfectly informed convey other’s behavior as a useful source of information, which can help to reduce uncertainty and provide insights about investment opportunities. The reasoning is built on the rational herding model, which emphasizes the importance of obtaining risk-free information from the surrounding environment (Devenow & Welch, 1996). The model relies on the assumption that, after a screening process, managers believe their chosen peers have more valuable information than the firm itself. Yet, this is not always the case. It is likely that peer firms all together face similar level of uncertainty and information asymmetry. The latter is especially applicable for the biopharmaceutical, R&D intense or technology companies, where the industry itself is subject to higher growth, success rate or even investment volatility and risk (DiMasi & Faden, 2011). Thus, not only it is important to asses how the firm-specific uncertainty and information asymmetry influence the investment behavior, but also investigate how firms behave if the industry or the overall fundamental uncertainty changes. While the learning models emphasize the importance of obtaining useful information, rivalry-based theories suggest that it is the competition that drives the commonality in investment behavior. As opposed to the above models, Fudenberg and Tirole (1984) argue that firms can respond negatively to the peers’ behavior if such actions are at the disadvantage of their competitors. Such a phenomenon is a consequence of the strategic interaction.

Depending on the underlying sources driving peer effects, the consequences can be radically different for companies and the society as a whole. On the one hand, imitation stimulates commonality among the firms which can lead to an increased competition (Devenow & Welch, 1996). On the other hand, such behavior can lead to collusion and strategic entry deterrence, which has the adverse effects on the overall welfare. Additionally, better understanding on the impact of a phenomenon is especially important, because the mimicking behavior, which arises purely due to the informational disadvantage/asymmetry or lack of managerial confidence and expertise, can lead firms to depart from what otherwise would be an optimal investment strategy (Lippman & Rumelt, 1982). The latter is a result of the investment-optimization process conducted by individual firms. Inevitably, this can lead to over- or under- investment. Departing from the optimal investment level can have adverse long-run influence on the financial performance. Yet, in order to understand the consequences of peer effects, it is first

(8)

necessary to identify the underlying mechanisms through which they operate and measure the extent of such behavior.

So far, the related literature for the U.S. market has focused on explaining the peer effects that arise due to sources such as rivalry, market competition or financial constraints (Park et al., 2017; Chen & Ma, 2017). However, no research has extensively linked the direct peer effects to the informational incentives, firm-specific and the overall fundamental uncertainty, which, according to the information-based theory, are assumed to be one of the main drivers of imitating behavior. Given the size of the U.S. market and arguing that a successful investment policy implementation can lead to sustaining growth, increased market valuation, overall performance and efficiency of allocating available resources by taking up on profitable investment opportunities, extending existing literature about the determinants of investments could provide crucial insights for managers as well as investors. Thus, the research question of this paper is, to what extent do the peer firms affect own corporate investment decision-making?

The study contributes to the existing literature in several ways. Firstly, it complements increasing literature about the positive peer influence in corporate investment policies. Secondly, the results presented in the paper are shown to be robust across different peer identification criteria. By testing the peer effects across various matching firm groups, I am able to minimize the potential risk of reflection and peer identification bias. While related literature has found support for the existence of positive peer effects, no paper has found consistent evidence tested on multiple peer firm samples. The latter finding is especially important for drawing statistical inference towards the causal relationship rather than merely providing evidence on the co-movement of the investment levels among the firms.

Thirdly, the positive peer influence is shown to be present by performing several methodological techniques. Yet, the magnitude of the focal firm’s sensitivity to its peers varies depending on the methodology employed. Thus, the study fails to provide enough evidence that enables to draw meaningful conclusions regarding the magnitude of peer influence. Finding support for the consistent magnitude of the peer influence would enable to furtherly investigate how the mimicking behavior impacts the long-term performance of the firms that engage in the latter activity or how this affects the overall investment efficiency.

Finally, by identifying and testing the effectiveness of different economic channels of uncertainty that potentially influence imitation behavior, I contribute by providing new empirical evidence on the effect of micro-economic, macro-economic uncertainty and recession on peer effects. While the existing

(9)

literature has provided evidence for the rivalry-based theory, the major focus of this research is to provide empirical support for the information-based theory.

The paper is structured as following. Theories, supporting the existence of peer influence and imitation are reviewed in Section 2. Following that, the most relevant empirical literature on the mimicking behavior in financial decision-making is presented. The main hypotheses derived from the literature review are developed in Section 3. Section 4 describes the empirical research design, sample selection, peer identification criteria and summary statistics. The results from the reduced form IV model are presented in Section 5. Section 6. provides evidence on the heterogeneity of peer effects. In section 7, I conduct several robustness tests. Finally, Sections 8 and 9 discuss the potential limitations, suggestions for the future research and provide the concluding remarks.

2. Literature review

2.1. Why do firms mimic each other?

The idea that peer firms play a significant role in shaping own corporate behavior has been long supported by economic theory (Bertrand, 1883). Before reviewing some empirical literature, first, it is crucial to identify economic channels supporting the behavioral interdependence among firms that can serve as a base for the empirical part of this research. The theories presented below originate from the rational herding mechanism, which I use as a core economic channel supporting the existence of peer effects (Devenow & Welch, 1996). In economic and finance framework, rational herding refers to a situation where agents react to the information coming from the behavior of other market participants rather than the fundamentals or the behavior of the market itself. Lieberman and Asaba (2006) distinguish two major economic sources for the imitation behavior driven by herding phenomenon: information and rivalry. I begin with introducing two major peer effect channels: information-based and rivalry-based theories. I end the section briefly introducing the contradicting view, suggesting that, in some cases, firms have incentives to behave opposite of their peers.

2.2. Information-based theory

Lieberman and Asaba (2006) suggest that ambiguity, uncertainty and imperfect information are the primary sources of imitation behavior. The information inquired from observing the actions of other firms can serve as a low-cost, low-risk guideline for an individual firm. Additionally, such strategy can significantly lower the cost of information which otherwise needs to be obtained from the

(10)

investment-optimization process (Conlisk, 1980). The author constructs a dynamic model that weights the costs and benefits arising from an agent choosing between two strategies: optimization and imitation. The model suggests that in the long-run, ‘cheap’ imitators perform no worse than the ‘costly’ optimizers, showing the clear benefits of the learning behavior from the peers. In financial corporate investments framework, not perfectly informed managers observe the activity of the peer firms and capture the most useful part of information, this way reducing the investing uncertainty and learning about new investment opportunities or the overall market prospects. Foucault and Fresard (2014) argue that uncertainty and information asymmetry play an important role in the process. In many cases, managers are unable to precisely predict the outcome of investment decisions as the likelihood of unpredictable and undesirable events increases. Furthermore, managers, who are subject to high information asymmetry, are able to benefit from directly observing the actions taken by other firms and its’ consequences. Intuitively, the investment decisions taken by other market participants reflect rationally formed expectations that can signal useful information. Such learning strategy not only enables to avoid high information acquisition costs but also provides information that signals about the growth, fluctuations or the success of a particular investment. For example, by observing the trends in peer stock prices and peer market valuation, managers can directly shape expectations about the industry growth prospects.

Although the above theoretical framework merely provides evidence for the existence of peer effects, it is likely, that these effects are rather heterogeneous (Foucault & Fresard, 2014). Therefore, the next theory links the imitation behavior to the channels that are assumed to amplify the effect, such as rivalry or market competition.

2.3. Rivalry-based theory

According to Lieberman and Asaba (2006), the sensitivity to peer behavior among firms is a response to the desire to minimize the downside consequences of rivalry. Such actions can be of great importance in order to maintain a relative competitive position among other firms. In competition-driven framework, the imitation process is characterized as the ‘reactive and passive response to risks and competition’ that enables to derive the trait in peer effects (Park et al., 2017). An execution of a successful corporate investment policy often involves acquisition of strategic assets, which play an important role in determining firms’ competitiveness and its’ relative position among the competitors. Therefore, rivalrybased theory can serve a ground for establishing the causality for peer effects in investment decision -making. The incentive to mimic is likely to magnify if the market is characterized by a large number of firms with comparable products, similar consumer demand and resource endowments. Intuitively, such

(11)

companies might benefit the most from observing the managements’ behavior and learn from each other about the launch, success of new products or adoption of new technology. For instance, firms are able to learn about the favorable timing of undertaking new investments. Therefore, firms operating in highly competitive and homogeneous environments have the most incentives to learn from each other, behave alike and this way ensure a status quo among its competitors. Several researches have supported the above view (Park et al., 2017; Leary & Roberts, 2014; Petersen & Rajan, 1994). The results are rather homogenous and suggest that the peer effects amplify with the level of industry competition.

2.4. Opposing view. Strategic investing

While the underlying peer effect mechanisms discussed above predict a positive relationship among the behavior of individual firms, Fudenberg and Tirole (1984) build a framework based on strategic interactions which, under certain circumstances, predicts the opposite. Based on their model, it is likely for the firms to respond negatively if such a strategy hurts the prospects of the other firms. For example, firms might purposely choose to over-invest compared to other firms if this action deters the entry of new competitors. In case a company has incentives to accommodate new entry, it chooses to under-invest. Yet, the model merely provides a theoretical framework based on strategic interactions among economic agents. While it is worth keeping in mind the possibility the peer effects that diverge from the classical imitation behavior predictions, so far, empirical literature has support for the information- and rivalry-based theories. Thus, in line with the trend, I expect to find support for the positive peer effects due to learning incentives discussed in Sections 2.2 and 2.3.

2.5. Empirical evidence

Recent empirical evidence has complemented economic theories predicting the existence of peer effects, suggesting that other firms significantly shape individual corporate behavior. John and Kadyrzhanova (2008) investigate the role of peer firms in corporate governance for a sample of the U.S. public corporations. Using the antitakeover provisions and firm headquarters location to construct the peer group sample, the authors provide robust evidence of positive peer effects. They conclude that firms are more likely to adopt appropriate governance policies if other peer firms do so. Furthermore, the magnitude of such a positive spillover effect increases with the quality of peers’ governance. Popadak (2012) finds that peers influence the dividend policies. The results suggest that one standard deviation increase in peer dividend payout is followed by approximately 16 % increase in own dividend payments.

(12)

The peer effects are shown to be significant for the dividend increase, however, no empirical support for peer influence is found for the decrease.

Leary and Roberts (2014) examine the impact of other firms on various corporate operational decisions, primarily focusing on a capital structure for the U.S. market. The authors suggest that firm i’s financing decisions respond significantly to the peer firms’ behavior and the magnitude of the effect depends on the characteristics of the peer firm sample. Furthermore, the authors show that peer effects explain the capital structure choices made by the corporations better than most of the determinants previously recognized in literature. By employing instrumental variable regression methodology using idiosyncratic stock returns as an exogenous instrument, the paper concludes that one standard deviation increase in the average debt ratio of the peer firms is accompanied by a 10% increase in firm i’s debt ratio. Furthermore, it is shown that less successful, smaller firms are sensitive to the behavior of industry incumbents, however, no support is found for the vice versa. Given these findings, one can expect that similar behavior could apply to corporate investment decisions.

One of the first attempts to link the commonality in corporate investments behavior among firms is presented in Fracassi’s (2016) paper. The author provides empirical evidence suggesting that investment decision externalities are widespread among corporations and arise primarily due to high costs involved in obtaining valuable information. The paper analyzes whether the presence of professional, educational and social ties among top managers of listed U.S. companies influences own financial decisions. By using a two-stage pairs model, the author provides evidence that the presence of social network connections among top managers and executives increases the homogeneity level in capital expenditure by 3 % for a median firm pair after controlling for common investment decision factors, cross-sectional endogeneity and multi-dimensional fixed effects. The homogeneity in investment decisions is shown to be stronger for firms located in the same region or operating in the same industry. However, the results provide no evidence of causal relationship regarding the social imitation among firms.

Foucault and Fresard (2014) investigate the impact of peers’ stock price and market valuation on firm i’s investments. The authors argue that stock prices in developed markets fully reflect the market valuation of companies, which can be useful in assessing industry growth opportunities and other important factors shaping financial policy. The authors provide evidence for the presence of a positive spillover effect on firms’ investment behavior. They find that an increase in the average peer value by 1 standard deviation increases own investment level by 5.9% for the sample of public U.S. companies for period 1996-2008. This effect is stronger when the informativeness of firm’s stock is lower and the

(13)

management is less informed. Furthermore, they find that the sensitivity of the investments to peer group valuation drops once the firm goes public. However, the paper only relies on the theory which predicts that managers learn from each other stock prices of the peers. They provide no evidence on the direct link between peer and firm i’s investments.

Chen and Ma (2017) conduct a research quantifying the extent to which companies try to directly mimic the investment decision of their industry peers for a sample of Chinese listed firms from 1999 to 2012. The authors argue that the stock market in China is rather underdeveloped and thus, not all relevant information is reflected in the companies’ stock price. Therefore, looking at the direct effect of peer investment decisions might provide more better insight in explain corporate investment behavior. After controlling for firm specific, industry and year fixed effects, authors show that 1 standard deviation increase in peer firms’ property, plant and equipment investment level is accompanied by a 4% rise in own investments. In line with authors’ predictions, it is shown that the effect is stronger if firms operate in highly competitive environment or when they are industry followers. Similar, however, less pronounced results are shown for R&D investments. In order to provide evidence for causality, the article addresses reflection problem and attempts to overcome endogeneity bias by employing two-staged least squares estimation procedure. Although the authors fail to overcome reflection problem, their instrumental variable methodology approach provides robust evidence for positive peer effects in corporate investments.

Regarding the U.S market, Park et al. (2017) measure the peer effects along the different levels of financial constraints and industry competition for period 1980-2010. Their instrumental variable (IV) results, using idiosyncratic equity returns as an instrument, suggest that more financially constrained firms and the ones facing more competition are more sensitive to peers’ investments. The major drawback of the study is that it solely looks at capital expenditures as investments, whereas the real investment decisions often include other types of investment decisions, such as R&D expenditure or acquisitions. Therefore, more evidence on whether the results about peer effects can be generalized to other types of investments, is needed. Furthermore, the paper, along with related research, finds support for rivalry-based theory. While uncertainty and ambiguity are other important channels supporting peer effects, no empirical evidence is present. Thus, investigating the real effect of various types of uncertainty can be very useful in identifying additional channels of the imitation behavior.

Finally, the choice of the peer group itself can have significant influence in determining the magnitude and the direction of the mimicking behavior. Intuitively, the more peer group firms resemble

(14)

the firm i, the less likely the imitation behavior will result in a failure. On the other hand, identifying a peer group that is too small can have adverse effects, such as produce the noisy estimates or make it difficult to infer any conclusions about the causality. While Park et al. (2017) define a peer group, based solely on the industry and year criteria, I suggest that it is crucial to understand whether the imitation behavior can be generalized across multiple peer groups identification methods. More detail on the peer group identification is presented in Section 4.4.

3. Hypotheses development and heterogeneity in peer effects

So far researchers have provided consistent evidence for the presence of peer effects in corporate decisions. What has not been agreed on is the heterogeneity in these effects. In this section, I develop several hypotheses that are tested in the empirical part of this research, based on the theoretical framework discussed in the literature review. According to the neoclassical competitive investments setup, companies invest until marginal benefit of investing intersects to the marginal cost (Abel, 1983). Based on Tobin’s (1969) framework, firms invest as long as the market value of a company is above its’ replacement costs. Both setups assume that the level of investments is primarily determined by firm-specific opportunities, resources and the internal optimization process. However, recent empirical evidence has shown that corporations often depart from the optimal investment strategy. Factors, such as information asymmetry, economic uncertainty, lack of information or rivalry can make it difficult for the management to perfectly predict the outcome of an investment strategy. Inevitably, directors pay attention to the investment behavior of peer firms to obtain useful information and reduce the uncertainty. In some cases, such behavior enables to ensure managerial reputation, assuming the peer firms decisions are desirable and well-respected from societal point of view. Based on the previous findings focusing on peer effects, I expect to find a positive relationship between own investment decisions in a current year and peer investment decisions in a prior year:

𝑯𝟏: Peer firms’ investment behavior has a positive influence on firm i’s investment decisions.

The role of uncertainty and ambiguity in peer effects is especially interesting, yet, lacking empirical support. Changes in economic, equities market or industry policy alter the fundamentals of the environment firms operate in. Intuitively, an increase in macroeconomic uncertainty level would make the volatility of investment returns to increase. Furthermore, uncertainty magnifies agency issues and makes information inquiry costlier. This can have adverse effects for financial performance of companies. Therefore, in periods of high uncertainty, managers are incentivized to be very attentive to the

(15)

surrounding environment and keep track of the strategic decisions, made by other, ideally, more informed, firms. On the other hand, high uncertainty increases the likelihood of a mimicking behavior failure, thus firms might be more conservative and rely on internalized knowledge in making financial decisions. Thus, further investigation is needed. Based on imitation incentives predicted by information-based theory, I hypothesize that macroeconomic uncertainty fosters the peer effects in investment decisions:

𝑯𝟐: Higher macro-economic uncertainty strengthens the peer effects in corporate investments.

In this paper, I refer to a macro-economic uncertainty arising from two different channels: equities market and economic policy. Literature has recognized multiple ways to measure macroeconomic uncertainty, such as the standard deviation of individual or aggregate analyst forecasts for GDP growth, unemployment rate or inflation. Yet, no consensus on the most appropriate proxy has been reached (Baker et al., 2016). Another well-known practice to proxy macro uncertainty is to use indexes, developed especially to track the level of particular type of ambiguity. For example, VIX index, tracking option-implied volatility for the S&P500 stocks, is often used to estimate equities market uncertainty. For the consistency of uncertainty measurement, in this paper, I use two U.S. market indexes developed by FRED that report uncertainty associated with different fundamentals. WLEMUINDXD index is used to track equities market uncertainty and USEPUINDXD index is used to measure the level of economic policy uncertainty.

Ambiguity arising from sources other than macro-economic factors can furtherly influence the way firms choose the optimal investment strategy. Arguing that a need for superior information about the industry opportunities and growth are additional reasons stimulating imitation behavior, it is interesting to investigate how firm-specific uncertainty and information asymmetry influence the peer effects. I refer to the latter two channels as micro-economic uncertainty. Chen and Ma (2017) focus on the role of information asymmetry in peer effects. The authors predict that firms, facing higher information asymmetry, have more incentives to mimic peer investment policies than the vice versa. In their research, information asymmetry is proxied by the distance from Beijing, suggesting that all most important investment decisions are made in the capital city. De Wet (2004) suggests that characteristics of the Chinese and the U.S markets are rather different, thus, one can expect the magnitude of market imperfections and information asymmetry in a well-developed market being less severe. As suggested in literature, the precision of informational signaling extracted from peer firms’ valuation or the stock prices is more accurate in developed markets. Even though market imperfections in the U.S. market are expected to be of smaller magnitude, I still expect to find significant consequences of information

(16)

asymmetry on the magnitude of peer effects. A commonly used practice to measure uncertainty and information asymmetry in developed markets is the forecast disagreement among the experts (McNichols, 1989). In this research, information asymmetry is proxied by the dispersion in analyst earnings per share (EPS) forecasts. While this method might not be applicable for emerging markets, it is assumed to be a reasonable measure for information asymmetry for public companies in developed stock markets, such as the United States.

Furthermore, I expect that a desire to reduce the individual investment policy uncertainty is another potential channel, supporting the existence of peer effects. Volatility in firm-specific processes has been a widely accepted measure for the micro-uncertainty (Henley et al., 2003). Therefore, I use the volatility of company investments as a proxy to represent firm-specific uncertainty. Although literature often employs GARCH models to estimate the time-varying conditional volatility, for simplicity, this paper use the standard deviation of a respective accounting variable to represent the level of uncertainty. This lead to a following hypothesis:

𝑯𝟑: Higher micro-economic uncertainty strengthens the peer effects in corporate investments.

4. Research methodology

4.1. Baseline investment model

The purpose of this thesis is to quantify whether investment decisions of the peers affect firm i’s investments and identify underlying mechanisms that determine the magnitude of the phenomenon. In order to do so, the first step is to establish the corporate investment model. The baseline empirical regression is a modified corporate investments model, which includes determinants, recognized by Richardson (2006), Cleary (1999) and Leary and Roberts (2014). Following previous literature, I include variables such ascash holding, financial leverage, dividend payout, cash flows, profitability and Tobin’s Q as explanatory variables that are known to influence financial investment policies. Firm- and peer-level control factors are included to increase precision of a model and mitigate omitted variable bias. Furthermore, to account for overall state of economy, I include macro-economic controls, such as the U.S. GDP growth rate, annual Treasury-bill rate and inflation. The baseline investment model is as following:

𝐼𝑛𝑣𝑡= 𝛽0+ 𝛽1𝐶𝑎𝑠ℎ𝑡−1+ 𝛽2𝐿𝑒𝑣𝑒𝑟𝑎𝑔𝑒𝑡−1+ 𝛽3𝐷𝑖𝑣𝑖𝑑𝑒𝑛𝑑𝑡−1+ 𝛽4𝐶𝑎𝑠ℎ 𝑓𝑙𝑜𝑤𝑡−1+

(17)

where Inv is a variable measuring corporate investment policy, expressed as a ratio of capital expenditure to the book value of assets in a respective year. Variable Cash represent firms’ cash holdings and is expressed as a sum of short-term activities and cash, scaled by total book assets. An increase in cash holdings comes at the expense of using the internal funds for new investments, therefore, I expect a negative impact of cash holdings on corporate investments (Almeida et al., 2014). Variable Leverage is calculated as a sum of short-term and long-term liabilities, divided by the total value of assets. Literature suggests that debt has a negative impact on investments level due to a debt overhang problem that forces forms to forgo profitable investment opportunities (Richardson, 2006). I expect a negative effect of firm’s leverage on the level of investments. Variable Dividend is defined as a common dividend divided by the book value of assets. As suggested in literature, dividend payout should negatively affect investments, because an increase in a payout to the shareholders reduces the free cash flows available for the managements’ disposition, which can be potentially retained and used to fund new projects (Fama, 1974).

Therefore, I expect a negative relationship between dividend payout and investments. Another variable closely related to investments is Profitability, measured in terms of firm’s returns on assets (ROA). Assuming a profitable performance brings in additional income for the company, the funds can be used for new investment opportunities. Thus, I expect a positive relationship among the two accounting measures. Furthermore, variable Cash flow is defined as a net cash flow from operating activities divided by total book value of assets. While the research has shown cash flows to be one of the determinants of the corporate investments, the direction of the relationship is subject to firm-specific characteristics (Almeida et. al., 2004). Finally, Tobin’s Q has been recognized as one of the primary determinants of

investment and growth opportunities (Tobin, 1969). In this paper, it is defined as a ratio of a market value of a company to the book value of assets. Year fixed effects is defined as a vector variable that enables to capture year-specific effects. Likewise, a vector variable industry fixed effects captures the industry-specific variance and partially mitigates the omitted variable bias problem arising due to the unobserved, industry-specific factors. Finally, 𝑢𝑡 is the heteroskedastic firm-specific error term. The detailed explanations and Compustat calculations for the variables used in the regressions are presented in Appendix A.

In order to empirically estimate the impact of peer effects in corporate investments, I specify the following regression:

𝐼𝑛𝑣𝑖𝑗𝑡= 𝛼 + 𝛽𝑃𝑒𝑒𝑟𝐼𝑛𝑣−𝑖𝑗𝑡−1+ 𝛾𝐹𝑖𝑟𝑚 𝑆𝑝𝑒𝑐𝑖𝑓𝑖𝑐 𝐶𝑜𝑛𝑡𝑟𝑜𝑙𝑠𝑖𝑗 𝑡−1+ 𝛿𝑃𝑒𝑒𝑟 𝑆𝑝𝑒𝑐𝑖𝑓𝑖𝑐 𝐶𝑜𝑛𝑡𝑟𝑜𝑙𝑠−𝑖𝑗 𝑡−1+ 𝜑𝑀𝑎𝑐𝑟𝑜𝐶𝑜𝑛𝑡𝑟𝑜𝑙𝑠𝑡+ 𝑦𝑒𝑎𝑟 𝑓𝑖𝑥𝑒𝑑 𝑒𝑓𝑓𝑒𝑐𝑡𝑠 + 𝑖𝑛𝑑𝑢𝑠𝑡𝑟𝑦 𝑓𝑖𝑥𝑒𝑑 𝑒𝑓𝑓𝑒𝑐𝑡𝑠 + 𝜀𝑖𝑗𝑡 (2)

(18)

where the subscripts i j and t indicate a firm, industry and year, respectively. The outcome variable 𝐼𝑛𝑣𝑖𝑗𝑡 measures corporate investment policy expressed in terms of capital expenditure. The independent variable 𝑃𝑒𝑒𝑟𝐼𝑛𝑣−𝑖𝑗𝑡 refers to the average peer capital investments excluding the firm i. As reasoned in the hypotheses section, I expect to find a positive relationship between focal firm’s and peer firms’ investments. Control variables are included to mitigate estimation error in peer effects. The controls are split into three categories: firm-specific, (average) peer firm and macro-economic factors. All independent variables in the model are lagged by one period. The lead-lag methodology strategy enables to mitigate a particular type of bias arising from trying to infer whether a group of peers influences own behavior or whether it is the individual entities in the group that drive the behavior. Intuitively, if corporate investment policies are affected by one another, two possibilities regarding the causality (if any) are present: either 𝑃𝑒𝑒𝑟𝐼𝑛𝑣−𝑖𝑗𝑡 causes 𝐼𝑛𝑣𝑖𝑗𝑡 or individual 𝐼𝑛𝑣𝑖𝑗𝑡 causes 𝑃𝑒𝑒𝑟𝐼𝑛𝑣−𝑖𝑗𝑡. In order to overcome the reverse causality problem, I use the lagged average peer investments in latter regressions.

4.2. Endogeneity bias

In order to choose the appropriate methodology that would enable to capture the peer effects, mitigate the estimation bias and enable to draw conclusions about the causality, it is first important to identify econometric issues common to the subject and recognized in previous literature. One of the main econometric issues arising in any corporate finance-related research is called endogeneity bias. In many cases, a failure to overcome the issue can lead to an inconsistent estimation of the parameters due to endogenous regressors (Roberts & Whited, 2013). The Ordinary Least Squares (OLS) estimation process is only able to capture the magnitude of association among the variables, but no inference regarding the magnitude of causality can be drawn. To some extent, instrumental variable regression methodology mitigates the issue and produces consistent estimation parameters. Following similar research about the peer effects (Leary & Roberts, 2014; Park et al., 2017; Chen & Ma, 2017), I use the lagged average peer idiosyncratic equity shocks to instrument for average peer corporate investment policy. The procedure of identifying the appropriate instrument for this research is discussed in the following section.

4.3. Choosing the instrumental variable

For an instrumental variable regression to produce consistent results, an instrument needs to meet several criteria: exclusion and relevance (Staiger & Stock, 1997). In this research context, the average peer firms’ investment is a variable that needs to be instrumented. The instrument should be related to the

(19)

average peer investments, but uncorrelated to firm-specific factors. The intuition behind using the idiosyncratic equity shocks is the following: stock prices reflect the market valuation of companies and are sensitive to corporate investment policies. Empirical evidence by Foucault and Fresard (2014) justifies the above reasoning. The authors find that stock prices react to the changes corporate investment policy, suggesting that equity shocks are correlated with corporate investments and thus, satisfy the relevance prerequisite.

Secondly, literature suggests that the above instrument also meets the exclusion criterion (Leary & Roberts, 2014; Liu, Whited & Zhang, 2009). The Idiosyncratic equity return captures the return component, occurring due to unique and firm-specific circumstances, contrary to the overall market. Due to this specific characteristic idiosyncratic equity shock, the instrument should not affect individual company’s investment policy or directly affect its disturbance. Separating the idiosyncratic shock from the overall stock return enables to isolate the component that is responsible for a substantial part of the stock price variation. Furthermore, by averaging the peer group variable, I am able to minimize the firm-specific disturbance and correlation among firm-specific factors. Although these characteristics do not imply full exogeneity, they assure that peer firm equity return shocks have little common variation.

Therefore, the next step is to construct the average peer firm idiosyncratic stock returns as a valid instrument and estimate the causal effect of peers’ investments on own investment behavior. The two-stage-least-squares (2SLS) process is as following. First, I estimate the fitted value of average corporate peer investments in the first-stage of the procedure by regressing average peer investments on idiosyncratic equity shocks of the peer companies and additional peer-specific controls (3). Then, I replace the average peer investments in the second-stage regression by the fitted value estimated in the first-stage (4):

𝐹𝑖𝑟𝑠𝑡 𝑆𝑡𝑎𝑔𝑒 2𝑆𝐿𝑆: 𝐼𝑛𝑣̂−𝑖𝑗𝑡= 𝛼 + 𝛽𝑃𝑒𝑒𝑟𝐼𝑑𝑖𝑜−𝑖𝑗𝑡−1+ 𝛿𝑃𝑒𝑒𝑟 𝑆𝑝𝑒𝑐𝑖𝑓𝑖𝑐 𝐶𝑜𝑛𝑡𝑟𝑜𝑙𝑠−𝑖𝑗𝑡−1+ 𝜖−𝑖𝑗𝑡 (3)

𝑆𝑒𝑐𝑜𝑛𝑑 𝑠𝑡𝑎𝑔𝑒 2𝑆𝐿𝑆: 𝐼𝑛𝑣𝑖𝑗𝑡= 𝛼 + 𝛽𝐼𝑛𝑣̂−𝑖𝑗𝑡−1+ 𝛾𝐹𝑖𝑟𝑚 𝑆𝑝𝑒𝑐𝑖𝑓𝑖𝑐 𝐶𝑜𝑛𝑡𝑟𝑜𝑙𝑠 𝑖𝑗 𝑡−1+

+ 𝛿𝑃𝑒𝑒𝑟 𝑆𝑝𝑒𝑐𝑖𝑓𝑖𝑐 𝐶𝑜𝑛𝑡𝑟𝑜𝑙𝑠 −𝑖𝑗𝑡−1+𝜑𝑀𝑎𝑐𝑟𝑜𝐶𝑜𝑛𝑡𝑟𝑜𝑙𝑠𝑡+ 𝑦𝑒𝑎𝑟 𝑓𝑖𝑥𝑒𝑑 𝑒𝑓𝑓𝑒𝑐𝑡𝑠 +

+ 𝑖𝑛𝑑𝑢𝑠𝑡𝑟𝑦 𝑓𝑖𝑥𝑒𝑑 𝑒𝑓𝑓𝑒𝑐𝑡𝑠 + 𝜖𝑖𝑗𝑡 (4)

In order to calculate the idiosyncratic component of equity returns, Fama-French four factor model is used. The first step in calculation is to obtain returns 𝑅𝑖𝑗𝑡 by using the market model regression:

(20)

where 𝑅𝑖𝑗𝑡 is the stock return and the subscripts i j and t indicate a firm, industry and year, respectively. The remaining four factor-model components are as following: excess market return (𝑅𝑒𝑡𝑚𝑘𝑡,𝑡− 𝑟𝑓,𝑡),

small minus big portfolio return 𝑆𝑀𝐵𝑡, high minus low portfolio return 𝐻𝑀𝐿𝑡 and the momentum term

𝑀𝑜𝑚𝑒𝑛𝑡𝑢𝑚 𝑡. Having estimated the necessary parameters for all the model factors, it is now possible to

calculate the expected returns and store the values:

𝐸(𝑅)𝑖𝑗𝑡 = 𝛼̂ + 𝛽̂(𝑅𝑒𝑡𝑚𝑘𝑡,𝑡− 𝑟𝑓,𝑡) + 𝛽̂ 𝑆𝑀𝐵𝑆𝑀𝐵 𝑡+ 𝛽̂ 𝐻𝑀𝐿𝐻𝑀𝐿 𝑡+ 𝛽̂ 𝑀𝑜𝑚𝑒𝑛𝑡𝑢𝑚 𝑀𝑂𝑀 𝑡 (6)

The idiosyncratic shock component is calculated by subtracting the expected return estimated above from the actual observed return for each firm:

𝐼𝑑𝑖𝑜𝑖𝑗𝑡= 𝑅𝑖𝑗𝑡− 𝐸(𝑅)𝑖𝑗𝑡 (7)

where 𝑅𝑖𝑗𝑡 is the actual monthly return and 𝐸(𝑅)𝑖𝑗𝑡 is the expected monthly return. In order to match the

frequency of stock returns to accounting data used in this research, monthly returns are annualized. Finally, using the annual values for 𝐼𝑑𝑖𝑜𝑖𝑗𝑡, I construct the instrument 𝑃𝑒𝑒𝑟𝐼𝑑𝑖𝑜−𝑖𝑗𝑡 which corresponds to

the average peer idiosyncratic return excluding the firm i.

4.4. Peer group identification

Another challenge in estimating how firms affect each others’ behavior is to determine the set of firms that can be used for an investment policy benchmarking. Intuitively, a peer group should consist of firms that share common characteristics, such as industry, market share, location, size or business complexity. The more commonality the firms have, the less likely is the imitation behavior to fail. Yet, incorporating many similar characteristics simultaneously might not be practical and result in having too few matching firms. Consequently, this can increase the risk of noisy statistical estimation results. Following Leary and Roberts (2014), I use two criteria to define a peer firm: industry and year. Additionally, I include a third criterion, namely, the state. I argue that firms operating in the same state might have more incentives to mimic each other due to similarity in the surrounding business/legal environment. I specify a peer group (excluding the firm i) as a set of companies that belong to the same industry-year and have their headquarters located in the same state.

4.5. Sample selection

In this research, Wharton Research Data Services database is used to obtain all necessary accounting data for the U.S. public companies from 1989 to 2017. The data for earlier years is excluded due to

(21)

discrepancy in its availability among the listed firms. All accounting data is retrieved from the Compustat database. Security prices data is obtained from Center for Research in Security Prices (CRSP). Information on the uncertainty in the U.S. market is retrieved from Federal Reserve Bank of St. Louis. (FRED) official database. Finally, the data regarding the GDP growth and inflation is retrieved from the World Data Bank database. Following methodology of Leary and Roberts (2014), I exclude all firm-year observations with missing values for the baseline peer effects model and firms that do not have a matching peer group. Additionally, financial, insurance (SIC codes 6000-6999) and utility companies (SIC codes 4000-4999) are excluded from the sample due to differences in investment behavior compared to other industries (Leary & Roberts, 2014). In order to mitigate the impact of outliers, all continuous variables are winsorized at 1% in both tails. Variables (Peer) Tobin’s Q and (Peer) Profitability are winsorized at 5% due to extreme values observed. All regression coefficients are robust in terms of heteroskedasticity and allow for intragroup correlation which enables to mitigate multiple statistical problems in regression analysis.

4.6. Summary statistics and correlation analysis

The final dataset consists of 53,925 firm-year observations for the period from 1989 to 2017. The unique number of firms for the overall sample is 7,178. The statistics reported resemble those of related research. Variables are classified into 3 categories: firm-specific, peer-firm specific and other. The variables presented in the category ‘other variables’ are used to test for the heterogeneity in peer effects. Table 1 presents descriptive statistics (See Appendix B).

The average rate of capital investments is 0.0614. The average peer investments are 0.0642. The means of firms’ PPE investments, R&D expenditure and acquisitions are 0.2662, 0.1382 and 0.0210, respectively. The peer average figures for PPE, R&D investments and acquisitions closely resemble those of individual firms (0.2593, 0.1597 and 0.0199). The means of firm specific variables, i.e. cash holdings, financial leverage, dividend payout, cash flows and profitability are 0.2285, 0.2234, 0.0838, 0.0207 and -0.1018, accordingly. The average peer firms’ control variables resemble those of individual firm average characteristics. This is in line with Chen and Ma (2017) findings. Interestingly, the peer average Tobin’s Q is twice as large as that of individual firm (9.309 compared to 22.694). It can be noted that the latter is also associated with large standard deviation compared to other accounting variables.

Additionally, I present summary statistics of variables used in the regression analysis that are common among the firms. The average peer idiosyncratic equity shock is -0.0095 with a standard deviation of 32.43%. The magnitude of equities shock observed in this dataset is smaller than the one

(22)

presented in Park et al. (2017) paper, where the authors use a comparable dataset. This could have occurred due to the differences in methodology in estimating the idiosyncratic equities returns. The minimum (maximum) numbers of firms per industry are 2 (521), accordingly. In the sample, on average, 110 firms belong to the same industry. The figures for annual earnings per share (EPS) range between -2.6658 and 7.6242 with a mean of 0.9228 and standard deviation of 1.6168. (Mean) figures for annual inflation, T-bill rate and GDP growth fluctuate around 0 with standard deviations of 1.1%, 2.2% and 1.6%, respectively. Thus, it can be argued, that the figures for the latter variables are quite stable over the period. The means for the Recession Index (possible values ranging from 0 to 100), equities market uncertainty (EMU) and economic policy market uncertainty (EPU) are 19.555, 79.735 and 97.686. Equities market uncertainty index is associated with the largest fluctuation over the period, producing a standard deviation of 38.381%. While no comparable research has included the latter indexes, the figures resemble the VIX index characteristics, often employed to measure the stock market uncertainty (Becker, et al., 2009). Finally, I present the statistics for total industry (company) sales. The averages for industry (company) sales are 102510 (1294) respectively, with a standard deviation of industry sales being approximately 40 times as large as that of individual company sales.

Table 2 (See Appendix C) presents the correlation matrix analysis of capital investments, firm-specific and peer-firm specific characteristics that are used in a baseline peer-effects regression model. The primary variable of interest is capital investment measure Inv. The correlation between firm capital investments (Inv) and average peer investments (Peer Inv) is 0.5158 and statistically significant at 1%. This suggests a large and significant co-movement between two variables. The correlation between the investments observed in the U.S. market are nearly twice as large as that of the Chinese market (Chen & Ma, 2017). Furthermore, it can be noted that the correlations between (peer) capital investments and (peer) firm-specific controls are of a small magnitude compared to Chen & Ma (2017) paper. Yet, it is worth noting that their research is conducted on the Chinese market, which is subject to different characteristics. Nevertheless, most of the correlation coefficients are significant at 1%. In line with expectations, the coefficients suggest a positive correlation between investments, cash flows and profitability. The correlation coefficient for cash holdings is -0.2409 and significant at 1%. This is again in line with the expectations. Interestingly, the correlations matrix suggests a positive correlation between investments, leverage and dividend payout and show a negative coefficient for the Tobin’s Q. Yet, the correlations merely present linear dependency for a set of variables. Thus, further econometric analysis is required.

(23)

5. Empirical results from a dynamic reduced form IV model

5.1. Results from the baseline peer effects model

The dynamic reduced form two-stage least squares (2SLS) regression results for the effect of peer firm investments at time t-1 on firm i’s investments at t are shown in Table 1.

Table 1. Reduced form 2SLS regression results for local state industry-year peers, by control groups.

Dep variable: Inv .(1) .(2) .(3)

Coefficient SE. Coefficient SE. Coefficient SE.

Peer Inv 0.4667*** (0.0575) 0.3889*** (0.0564) 0.3376*** (0.0493)

Firm -specific factors

Cash -0.0145*** (0.0023) -0.0145*** (0.0023) Leverage -0.0077*** (0.0022) -0.0077*** (0.0022) Dividend -0.0071*** (0.0012) -0.0072*** (0.0012) Cash flows 0.0161*** (0.0020) 0.0161*** (0.0020) Profitability 0.0054*** (0.0018) 0.0049*** (0.0018) Tobin's Q 0.0049*** (0.0003) 0.0048*** (0.0003)

Peer -specific factors (averages)

Profitability 0.0036*** (0.0007)

Tobin's Q 0.0003*** (0.0001)

Other variables

Inflation 0.6732** (0.2724)

Annual T-bill rate -0.2896 (0.2002)

GDP growth 1.0927*** (0.2174)

Constant 0.0402*** (0.0080) 0.0427*** (0.0077) -0.0039 (0.0096)

Industry Fixed Effects Yes Yes Yes

Year Fixed Effects Yes Yes Yes

N 53,925 53,925 53,925

Adj. R-sq. 0.335 0.364 0.372

Chi. sq. 65.91 586.62 688.45

The effect of peers' investments on firm i's investment policy for the public U.S. companies for 1989 – 2 017. The table presents coefficients for firm-specific (lagged by one period), peer firm-specific characteristics (lagged by one period) and macroeconomic variables related to corporate investments. All regressions include industry and year fixed effects. The main independent variable in models 1 - 3 is (average) peer capital investments. Standard errors, which are robust to heteroskedasticity and adjusted for clustering at company level and year, are reported in parentheses. *** denotes significance at the 1% level; ** 5% level, and * 10% level, respectively.

All the statistics are robust to heteroskedasticity and clustering on a company level and year. Each column refers to a different set of control variables. The primary figures of interest are reported in bold.

(24)

Column (1) presents the coefficients and the standard errors (in parentheses) for peer effects while controlling for industry and year fixed effects. The coefficient Peer Inv is 0.4667 and significant at 1% (Std. Error=0.0575). The result suggests presence of a significant linear relationship between peers’ and firm i’s capital investments. On average, 1 point increase in peer investments at time t-1 is accompanied by 0.4667 points increase in own capital investment level in a following year t.

Following related research, I estimate the peer effects by including firm-specific variables that are related to corporate investments for more accuracy. Column (2) reports the statistics for peer effects while controlling for firm cash holdings, leverage, dividend payout, cash flows, profitability, Tobin’s Q, industry and year fixed effects. The coefficient Peer Inv is 0.3889 and significant at 1%. Interestingly, the latter coefficient does not differ from the value reported in (1) column (calculation: 0.4667−0.3889

√0.05752_+0.05642=

1.2821) at 1 % level. The reported goodness-of-fit Adj. R-sq. for models (1) and (2) are 0.335 and 0.364, respectively. It suggests that the improvement in the model explanatory power by incorporating firm-specific characteristics is negligible. In line with predictions, Cash (coefficient= -0.0145), Leverage (coefficient= -0.0077) and Dividend (coefficient= -0.0071) suggest a significant negative relationship with corporate investments. The coefficients for Cash flows (coefficient= 0.0161), Profitability (coefficient= 0.0054) and Tobin‘s Q (coefficient= 0.0049) are positive and significant at 1%, implying a positive effect on capital investments. The findings are in line with related research on the determinants of corporate investments (Chen & Ma, 2017; Park et al., 2017; Almeida et al., 2014). Although firm-specific characteristics in model (2) suggest a significance relationship with corporate investment policy, it can be noted that the magnitude of the effect for most of the accounting measures is negligible. Finally, model (3) displays the result for peer effects while controlling for firm-specific characteristics, peer-firm (average) characteristics and macro-economic controls. The coefficient for peer investments is 0.3376, which is similar to that of models (1) and (2). This implies a rather stable significant and positive linear relationship between firm i‘s and peer firms‘ investments, regardless of the control variables included. The figures for firm-specific controls remain almost unchanged compared to model (2) and remain statistically significant at 1%.

For peer firm-specific characteristics, I include only Peer Tobin‘s Q and Profitability, arguing that the investment opportunities and profitability of peer firms can have an influence on individual firm investment strategies. The results show that the chosen peer firm controls indeed positively affect focal firm’s capital investments. Regarding the macro-economic controls, interestingly, the coefficient GDP

(25)

growth implies almost a 1-to-1 relationship with corporate capital investments (coefficient= 1.0927). This

suggests that the overall state of economy has major implications on the corporate investment decisions. The goodness-of-fit values for models (1) to (3) fluctuate around 0.350. The values for model appropriateness resemble that of similar research on peer effects (Chen & Ma, 2017; Foucault & Fresard, 2014). The same holds true for the standard errors reported. Similar to Park et al., (2017) findings, the standard errors do not exceed 10 % for any of the (peer) firm-specific control factors. This, however, does not hold true for the macro-economic (other) variables. Given that no past related peer effects research has employed macro-controls, no reasonable comparisons can be made. Additionally, The Wald Chi- squared statistics are large, suggesting that the joint significance of the variables included in each model. Overall, the results from Table 1. provide empirical support for the Hypothesis 1. Furthermore, the findings suggest that the influence of peer firms remains rather stable across different control groups.

5.2. First-stage 2SLS estimation statistical properties

Past research employs the idiosyncratic stock return as an instrument to analyze the effect of peer group characteristics on individual attributes due to its desirable properties, such as the independency of equity shocks across firms and over time. Nevertheless, it is important to verify the statistical appropriateness of the instrument for this sample. Table 2. reports post-estimation properties from the first-stage regression of the Table 1 across columns (1) to (3), respectively.

Table 2. First-stage 2SLS post-estimation statistical properties.

Dep var: Idiosyncratic return . (1) .(2) .(3)

Cragg-Donald Wald F-statistic 67.3377 67.0342 95.1691

𝐻0: weak instrument(s) Rejected Rejected Rejected

Adj. R-sq. 0.5893 0.5896 0.5932

Jansen J-statistic p-value 0.0152 0.3973 0.4137

𝐻0: under-identification Not rejected Not rejected Not rejected

Wooldridge statistic p-value 0.0000 0.0000 0.0000

𝐻0: exogeneity Rejected Rejected Rejected

Note: Hypotheses decisions are based on 1% significance level. Cragg-Donald Wald F-statistic and Jansen J-statistic p-value assess the appropriatness of the 2SLS instrument(s). Wooldridge J-statistic p-value reports the decision regarding the instrumented variable (Peer Inv) exogeneity (endogeneity). Additionaly, first stage goodness-of-fit Adj. R-sq. is presented.

(26)

In the first-stage, average peer investments are regressed on the lagged average peer idiosyncratic stock returns and peer-firm specific control variables defined in the baseline corporate investments model equation (1). Cragg-Donald Wald F-statistics suggest that the instrument used in this research passes the test of ‘weak instruments‘. The null-hypothesis is rejected based on the rule of thumb F-statistic > 10 (Stock et al., 2002). Furthermore, the first-stage estimation displays large goodness-of-fit values (0.5893, 0.5896 and 0.5932, respectively), suggesting that the variables included in this stage explain nearly 60 % of the variation in average corporate investments. Jansen-J statistic p-values for models (1) to (3) show that the instruments included do not contradict with each other. While this implies that there is no conflict among the first-stage variables, the statistic does not provide further insight regarding the appropriateness of the variables nor the model itself. Finally, Wooldridge statistic p-value implies that the instrumented variable (average peer investments) is endogenous. This suggests that 2SLS is an appropriate methodology that encounters and partially mitigates the issue of endogenous regressors. Overall, the post-estimation results suggest that the equities return shock contains several important desirable properties of a valid instrument.

5.3. Reflection bias

Manski (1993) argues that any research estimating the influence of the (average) peer group on individual attributes suffers from the reflection. Correlations observed between own investments and peers’ investments may be merely a consequence of a latter issue and tell nothing about the actual causality. According to the author, the bias arises due to the identification of a peer group. It is commonly assumed that peer group is chosen based on the level of similarity among the individuals/entities. These individuals/entities often interact in a similar environment, are exposed to the same fundamental shocks that drive the commonality in their behavior. In financial framework, companies operating in the same market and the same industry are subject to the same/similar regulatory environment, industry growth opportunities and mutual demand shocks that might drive the commonality in corporate behavior. The more similar the firms are, the more likely the reflection problem is present. Therefore, choosing the appropriate peer group might be crucial in determining the causal effect. Although the IV methodology partially mitigates the bias by employing idiosyncratic equity shock as an instrument that absorbs the exogenous variation, further investigation regarding the refection problem is needed. Chen and Ma (2017) argue that in order to overcome the reflection, it is important to test the peer effect robustness by choosing different identification of a peer group. Intuitively, broadening the peer group would make the reflection problem less severe. If the reflection is the primary reason for observing strong correlations

(27)

among individual firms and peer averages, choosing a peer group that shares less common characteristics would make those correlations disappear.

In the following section, I investigate the sensitivity of peer effects with respect to a different peer group identification. I remove the state criterion and identify a new peer group based on the same year and industry criteria. I argue that firms operating in the same state are subject to similar consumer demand and legal environment, which could be the primary drivers for the commonality in investment behavior. The influence of the industry-year peer group on firm i’s investments is reported in Table 3.

Table 3. Reduced form 2SLS regression results for industry-year peers, by control groups.

Dep variable: Inv .(1) .(2) .(3)

Coefficient SE. Coefficient SE. Coefficient SE.

Peer Inv 0.6637*** (0.0563) 0.5604*** (0.0550) 0.4738*** (0.0483) Firm -specific factors Cash _-0.0143*** (0.0018) _-0.0142*** (0.0018) Leverage _-0.0104*** (0.0017) _-0.0103*** (0.0017) Dividend _-0.0040*** (0.0006) _-0.0042*** (0.0006) Cash flows 0.0184*** (0.0018) 0.0185*** (0.0018) Profitability 0.0064*** (0.0016) 0.0061*** (0.0016) Tobin's Q 0.0054*** (0.0002) 0.0054*** (0.0002)

Peer -specific factors (averages)

Profitability 0.0015*** (0.0002)

Tobin's Q 0.0001*** (0.0000)

Other variables

Inflation 0.2741 (0.1825)

Annual T-bill rate -0.0802 (0.1352)

GDP growth 0.6979*** (0.1449)

Constant 0.0191*** (0.0065) 0.0233*** (0.0062) -0.0035 (0.0074) Industry Fixed

Effects Yes Yes Yes

Year Fixed Effects Yes Yes Yes

N 104,155 104,155 104,155

Adj. R-sq. 0.282 0.313 0.322

Chi. sq. 138.86 1000.04 1349.31

The effect of peers' investments on firm i's investment policy for the public U.S. companies for 1989 – 2 017. The table presents coefficients for firm-specific (lagged by one period), peer firm-specific characteristics (lagged by one period) and macroeconomic variables related to corporate investments. All regressions include industry and year fixed effects. The dependent variable of interest for models 1 - 3 is (average) peer capital investments. Standard errors,which are robust to heteroskedasticity and adjusted for clustering at company level and year, are reported in parentheses. *** denotes significance at the 1% level; ** 5% level, and * 10% level.

(28)

The results displayed are based on the identical control group choice methodology as of Table 1, models (1) to (3). The coefficient Peer Inv for in all three models is positive and significant at 1% level. In fact, the coefficients for models (1) to (3) are consistently larger than those reported in Table 1. In line with the findings from the local state, industry-year peers, the coefficient Peer Inv remains rather stable (0.6637, 0.5604 and 0.4738) after changing the control group. Furthermore, the direction of additional control variables remains unchanged. This suggest the consistency across two peer group identification strategies. The results suggest a significant presence of peer firms influence and support Hypothesis 1. Furthermore, the table implies that by identifying a broader peer group, the magnitude of the peer effect does not diminish. Contrary, the coefficients of average peer investments increase compared to that of local state industry-year peers. The latter suggests that the reflection issue is this research potentially is not the root cause of the significant correlations observed between peer investments and individual firm investments. Overall, I conclude that, by employing 2SLS with a preferable instrumental variable and testing the robustness of the peer influence by identifying a new peer group, I find enough consistent evidence to support the presence of peer effects and manage to provide empirical evidence that enables to partially solve the reflection problem.

5.4. The impact of macro-economic uncertainty

As presented in literature review, I argue that uncertainty is one of the main channels driving the imitation behavior. To test whether macro-economic uncertainty plays a role in determining the peer effects, I examine whether equities market and economic policy uncertainty magnify the peer effects. In order to capture the impact of two types of uncertainty on peer effects, I create two interaction variables:

Peer Inv x EMU to examine the effect of equities market uncertainty on peer effects in column (1) and Peer Inv x EPU to test for the influence of economic policy uncertainty in column (2).

As reported in column (1) of Table 4, the coefficient of the interaction term between the average peer investments and the equities market uncertainty index (WLEMUINDXD) Peer Inv x EMU is 0.0046, significant at 1% (SE= 0.0001). Although the magnitude of the effect is minor, this suggest that individual investments are more sensitive to the peer investments as the equities market uncertainty increases. The results in column (2) closely resemble those of column (1). As shown in Figure 1., both indexes move in the same direction during the entire sample period. This could be an explanation the similarity between the results of two models. The interaction term displayed in Column (2) Peer Inv x EPU is 0.0040 (SE.= -0.0001), significant at 1%. Interestingly, adding the interaction term in a respective column makes the

Do the decisions of peer firms impact individual firm strategies? evidence from corporate investments in the United States

Amsterdam Business School

Master Thesis