Political Ties and Predictable Returns

Transcription

1 Political Ties and Predictable Returns Siyi Shen Frist Version: March 2018 This Version: August 2018 Shen: Carroll School of Management, Boston College, Department of Finance; I would like to thank my dissertation committee members, Ronnie Sadka (Chair), Rui Albuquerque, Jeff Pontiff, and Jon Reuter for the valuable suggestions and support. For helpful comments and suggestions, I thank Pierluigi Balduzzi, Lenny Kostovetsky, Phil Strahan, and seminar participants at Boston College. All remaining errors are my own.

2 Political Ties and Predictable Returns Abstract This paper demonstrates the importance of inter-firm political links, measured based on common campaign contributions made by firm executives or corporate political action committees. A firm s price movement is predictable by the price movement of its politically connected firms, which is predominantly driven by exposures to common political risks. Using the 2010 Citizen United v. FEC as an exogenous shock to political activism, further evidence indicates that the effect is weaker for firms that are restricted from actively engaging in political campaigns. The crosspredictability effect is strongest among firms that are smaller, less owned by mutual funds, and operate in different states and sectors, suggesting that political links between firms are largely overlooked due to limited investor attention. Portfolios based on political ties yield risk-adjusted returns of 4-5% per annum. 1

3 Introduction Recent high-profile geopolitical events have renewed considerable interests among academics as well as practitioners in understanding the impact of politics on financial market. 1 The 2017 KPMG global CEO outlook survey reports that more than half of the interviewed CEOs believe that the uncertainty of current political landscape has a larger impact on their business than ever in the history and 70% of them have been taking steps to deal with this issue. 2 Indeed, during each congressional election cycle, firms and their executives donate sizable amount of money to the political candidates to the extent that political contribution is considered as one of the most widely used and effective tools for corporate political activism. 3 While the impact of politics on financial market is extensively studied, prior works focus mostly on the direct relation between firms and politicians. However, anecdotal evidence suggests that companies are more likely to work as a group rather than individually to engage in political agenda. For example, firms often form joint lobby groups to influence legislation, and their executives attend private events hosted by politicians they support in common together. 4 Thus in this paper, I explore a new channel of information flows between politically connected firms measured based on common campaign contributions made by firm executives. Firm executives 1 An unexhausted list of high-profile political events includes the Brexit vote, 2016 U.S. presidential election, and European populism threats. For more details, see In the Wake of a Tumultuous Year, the Global Elite Face a World of Uncertainty (Wall Street Journal, January 16, 2017). 2 For more details, see 3 For example, the total raised campaign contributions from corporate political action committees are more than $2.2 billion for the 2016 election cycle in contrast to $630 million in the 2000 election cycle. 4 As an example, see 2

4 donate to the same political candidate signal that they may attempt to achieve common purposes, either political rent extraction or certain political risk hedging, or both. If that is the case, I conjecture that politically-connected firms tend to exhibit correlated fundamentals. To test this postulation, I form inter-firm political links among the S&P500 firms by merging the individual contribution records from the U.S. Federal Election Commission (FEC) with BoardEx database to obtain firm executive (CEO) contributions from 2000 through The yearly measure of inter-firm political links is calculated by counting the number of political candidates their CEOs contribute in common over a four-year horizon. 5 On average, every year executives from approximately 364 firms have donated in political campaigns over a four-year measurement horizon, and 96% of them connect to other firms CEOs through common contributed political candidates. Evidenced by the return comovement test, I find that politically connected firms indeed have correlated fundamentals: controlling for other variables, a onestandard-deviation increase in returns of a firm s politically connected counterparts this month is associated with a 22 basis points increase in its return concurrently, with a t-statistic of Although publicly available, information on inter-firm political links can be overlooked by investors due to attention constraint binding. If so, the gradual information incorporation leads to incomplete price reactions, generating lead-lag effects. Indeed, I find that returns of politically connected firms are cross-predictable based on the Fama and MacBeth (1973) forecasting 5 The measure is similar to that in Cooper, Gulen, and Ovtchinnikov (2010), in which they use the number of candidates that each firm supports as a measure of firm political involvement. 3

5 regressions. Specifically, a one-standard-deviation increase in returns of a firm s politically connected counterparts this month leads to a 7 basis points increase in its return next month, with a t-statistic of Further analysis shows that the cross-predictability effect mostly pertains to firms that are more closely connected in the political network. To probe the economic magnitude, I construct a monthly rebalanced long-short portfolios based on this effect, and show it yields riskadjusted returns of 4-5% per annum. If limited investor attention is the driving force, I expect sluggish information dissemination among firm-pairs whose political links are intricate to uncover. Along this line of reasoning, the lead-lag effect should be more pronounced between inter-state or inter-sector politically connected firms since geographic domains and sector classifications are widely recognized as key attributes of political landscapes. 6 To probe this point, I partition politically connected firms based on whether their headquarters are in the same state or whether they operate in the same sector, and run the forecasting regressions separately for each specification. Consistent with the conjecture, the cross-predictability effect is in fact strongest among politically-connected firms operating in different states and sectors, which evidently demonstrates that information transmission is rapid within common identified groups, but less so outside them. The fact that the effect is most pronounced among inter-sector politically connected firms raises doubts whether those firms are truly fundamentally related. To examine this point, for each 6 Papers using geographic locations to classify political landscapes include Cohen, Coval, and Malloy (2011), Kim, Pantzalis, and Park (2012), and Kostovetsky (2015). Papers using sectors to classify political landscapes include Belo, Gala, and Li (2013) and Addoum and Kumar (2016). 4

6 industry I use annual Benchmark Input-Output survey from the Bureau of Economic Analysis to identify its primary supplier and customer industries and show that the lead-lag effect mainly pertains to politically connected firms within the same supply chain network. This finding is in line with Cohen and Frazzini (2008) and Menzly and Ozabas (2010), yet gives rise to concerns regarding a potential confounding problem that political links between firms may merely mimic their economic links. To rule out this possibility, I run a placebo test for a group of weakly politically-connected firms, and show that the lead-lag effect along the economic links is negligible. The evidence hence reveals that there is a unique and valuable channel of information flows along the political linkage that is not subsumed by the supply-chain network. I next explore whether the fundamental correlation between politically connected firms is driven by common political rent opportunities or common political risk exposures. To exploit it, I employ a firm-level political risk measure from Hassan et al. (2017). 7 The authors apply textual analysis of quarterly earnings conference-call transcripts to construct political risk matrices for each individual firm by assigning scores in eight major topics of political issues. Supportive of a shared political risk explanation, I find that the cross-predictability effect mostly exhibits among firms concerning common political issues. To further identify the underlying mechanisms, I employ the 2010 U.S. Supreme Court s landmark decision on Citizen United v. Federal Election Commision (CU) as a positive shock to 7 I thank the authors for making the firm-level political risk data available on their website: 5

7 political activism for firms in ban states. 8 Specifically, I study the differential patterns of the crosspredictability effect between politically connected firms in ban states (the treatment group) and those in no-ban states (the control group) in a span of five years surrounding CU. The differencein-difference analysis finds that a one-standard-deviation change of the lead-lag effect before and after CU is approximately 12% per annum larger for firms in ban states compared to those in noban states. The finding thus demonstrates that prior bans on state election independent expenditures before CU give rise to a negative effect on company political activism: the crosspredictability effect is significantly weaker among firms in ban states as they are restricted from actively engaging in political campaigns. The results hold up to a battery of robustness checks. Frist, I find consistent results when turning to measures of real activity, emphasizing that politically connected firms indeed exhibit correlated cash flows. Second, the cross-predictability effect is apparent when I form the inter-firm political links based on common contributions made by firm political action committees (PACs). Moreover, my findings barely change with various measurement horizons or using alternative political connectedness measures that account for firm political activeness or control for CEO tenure. Last, instead of using median number of common contributed candidates as a connectedness strength cutoff, I now assign top ranked firms as strongly connected, and continue to find that the main effect mostly pertains to strongly, but not weakly, politically connected firms. 8 In addition to lifting bans on independent political expenditures by corporations on the federal elections, the Supreme Court also overrules prior bans on independent political expenditures by corporations on state elections in 21 states (see Spencer and Woods (2014) for more details). 6

8 The remainder of the paper proceeds as follows. Section 1 provides an overview of the related literature. Section 2 describes the data and political network construction. Section 3 presents empirical findings of fundamental correlation and the lead-lag effect between politicallyconnected firms and exploits the underlying mechanisms. Section 4 examines the robustness of the main findings. Section 5 concludes. 1. Literature Review This paper speaks to several strands of literature. First, this study contributes to papers that investigate political impact on the financial market. Faccio (2006), Cooper, Gulen, and Ovtchinnikov (2010), Duchin and Sosyura (2012), and Kostovetsky (2015) all document that the value of political connections to firms as evidenced by stronger stock market performance, higher chance of granting government contracts, and better access to finance. In a different vein, Pástor and Veronesi (2013) develop a general equilibrium model and derive that aggregate political uncertainty commands a risk premium and makes stocks more volatile and correlated. Akey and Lewellen (2017) show that changes in political landscape affect policy-sensitive firms investment strategies and capital structures substantially, as a result those firms donate to political candidates to hedge their exposures. Hassan et al. (2017) construct a firm-level political risk measure and document that the majority of firm-level political risk is unexplained by the policy uncertainty at the aggregate- or sector-level. The contribution of my study is to identify firms political links 7

9 based on their common political contributions, and find that politically-connected firms are fundamentally related due to shared political exposures. This paper also adds to a vast literature on investor inattention. Hong and Stein (1999) lay out a theory framework and argue that inattentive investors produce information slowly, in turn generating return momentum. Cohen and Frazzini (2008) and Menzly and Ozbas (2010) document that returns of firms and industries in the same supply chain network are cross-predictable. Parsons, Sabbatucci, and Titman (2016) find a lead-lag effect between neighboring firms operating in different sectors due to gradual incorporation of price information by non-overlapping sets of investors. My paper adds to the literature by showing investor inattention on political links as evidenced by the return cross-predictability among politically-connected firms. 2. Data and Measure Construction In this section, I first describe the data used in the paper obtained through several sources, and then construct each firm s politically-connected counterparts based on their executives common contributed political candidates Data The Federal Election Committee (FEC) retains records of all federal election campaign hard money contributions in excess of $200 made by individuals since Although individual 8

10 contributions are limited to $10,000 per candidate per election cycle, prior works (e.g., Akey (2015)) have shown that the hard money channel is significantly correlated with other actions to develop political involvement that are more difficult to observe, such as soft money contributions and non-money favors. Therefore, several thousand dollars of hard money contributions can plausibly reveal important political links between firm executives and politicians. The biographic information of the S&P500 firms executives (CEOs) are obtained from BoardEx database along with complementary information from Bloomberg and LinkedIn. I then search for executives contributions by manually matching executives biographic information with donors records (home address, employer, and occupation) from 2000 through ,10 The stock-level data, such as stock returns and market capitalization, is obtained from the CRSP, and firms industry, geographic and accounting information is from the Compustat Common Candidate Contribution Index Following Cooper, Gluen, and Ovtchinnikov (2010), I use the number of political candidates that a firm s CEO donates as a proxy for political activeness. Specifically, I obtain information of political candidates (e.g., name, state, and party) from the candidate summary file, and match it 9 To match names in BoardEx and FEC databases, I follow the procedures by Hong and Kostovetsky (2012). For more details, see the Appendix A.1. of their paper. 10 The sample is limited to the post-2000 period as Fracassi and Tate (2012) and Engelberg, Gao, and Parsons (2013) point out that BoardEx s coverage of U.S. public companies is extremely limited prior to

11 with the recipient record in individual contribution file. For each firm at the end of October every year, I calculate the total number of candidates its executive donate over the past four years: 11 M Contribute i,t = Candidate i,m,t,t 3, (1) m=1 where Candidate i,m,t,t 3 is an indicator variable equal to one if firm i has contributed money to candidate m over the years t-3 to t. Having constructed the firm-level political activeness index, I next derive the key measure of this paper that is designed to form political networks between firms through their common contributed political candidates: I use the number of political candidates that two firms support in common as a measure of their political connectedness: M CommonContribute i,j,t = (Candidate i,m,t,t 3 Candidate j,m,t,t 3 ), (2) m=1 where Candidate i,m,t,t 3 Candidate j,m,t,t 3 is an indicator variable equal to one if firm i and firm j both have contributed money to candidate m over the years t-3 to t. 12 The political network is rolled over and updated at the end of October every year. The logic is that two companies contribute to same political candidates may signal that they have common political interest or political risk exposure. 11 Choosing October as the cutoff is to accommodate the fact that the U.S. federal elections are held on the first Tuesday of November every even-numbered year. 12 In the robustness checks, I also examine alternative measurement horizons such as 2-year or 5-year. 10

12 I also categorize all the connected firms into strongly or weakly connected counterparts to examine the importance of considering political connectedness strength: I rank firm i s connected firms based on their number of common contributed candidates. That is, if the number of common contributed candidates between firm i and one of its connected firms is above (below) the median, that connected firm is considered as strongly (weakly) connected counterpart to firm i. 13 The political linkage between two firms is non-directional, yet firm i can be ranked in firm j s strongly connected counterpart and firm j may be assigned to firm i s weakly connected counterpart Summary Statistics Table 1 presents the summary statistics of the S&P500 firm CEOs political contributions over the period from 2004 to Panel A shows that on average about 364 firms have given money to political candidates and 96% of them have shared contributed political candidates to other firms over the past four year. Panel B exhibits that on average a specific firm connects to about 61 companies with 1.13 candidates contributed in common over the past four years. Furthermore, among firm s politically-connected counterparts, only 14.37% and 18.29% of them are headquartered in the same state and operated in the same sector, respectively. 14 Thus, the effect 13 Note that if firm i only connects to one firm, mechanically this firm would be recognized as strongly connected counterpart. 14 I assign firms to 12 sectors based on the industry classifications from Ken French s website: 11

13 analyzed in this paper mostly pertains to inter-segment firms among which the friction of information dissemination is predominantly more severe. I also classify each firm s politically-connected counterparts based on strength in political connectedness and find approximately 17 (44) out of 61 firms can be categorized as strongly (weakly) connected with approximately 2 (1) candidates contributed in common. Strongly politically-connected firms are more likely to locate in the same state (30.57%) or operate in the same sector (17.09%), compared to only 12.62% and 13.16% of weakly connected counterparts in the same state and same sector, respectively. While in most of the paper, median is the cutoff for strongly or weakly politicallyconnected counterparts, I employ alternative cutoff for political strength as a robustness check to ensure that the main findings are not driven by the specific choices of cutoff methods. 3. Results In this section, I present evidence of contemporaneous correlations and cross-predictability in returns among politically-connected firms. I further investigate investor inattention as explanations for the return cross-predictability. Last, I demonstrate the underlying mechanisms Contemporaneous Correlations 12

14 Prior works have shown that stock prices of firms with similar characteristics tend to move together and naturally politics is considered as a common factor that significantly affects all companies. Thus, I reason that firms that make many political contributions in common are correlated in fundamentals. 15 I test this conjecture by running the contemporaneous monthly Fama and MacBeth (1973) regressions and adjust for heteroscedasticity and autocorrelation using Newey and West (1987) up to 12 lags as follows: Ret i,t = α t + β t Linkret i,t + Λ t Z i,t 1 + ε t, (3) where Z i,t 1 is a vector of control variables including firm i s own 1-month lagged return, its corresponding Fama-French 12-industry 1-month lagged return, same-state firm portfolio 1-month lagged return, as well as its own, industry s, and geographic momentum return cumulated from month t-12 to t-2. Firm size (log market capitalization) and log book-to-market ratio measured at the end of prior year, and total number of contributed candidates measured over the past four years are also included. The dependent variable is the monthly stock return denoted in percentage and the independent variable of interest is the contemporaneous monthly return of its politicallyconnected counterparts. Since many firms have multiple politically-linked counterparts, I weight returns of all connected firms by the number of common contributed candidates to derive a portfolio return. Therefore, more closely linked firms in the network are assigned by larger 15 For example, Pástor and Veronesi (2013) document that stock returns are more correlated when the aggregate policy uncertainty intensifies. 13

15 weights. 16 Hereafter, I refer to this return as the linked or connected firm return (Linkret in tables). For the ease of economic interpretation, all the return variables are standardized in the crosssection to yield unit standard deviation. The result in Table 2 confirms that politically-connected firms are indeed positively correlated in fundamentals. From Model (1), the monthly stock return increases at 45 basis points with a one-standard-deviation increase in the contemporaneous month return of its politicallyconnected firms. In column 11 accounting for various controls, the coefficient on Linkret i,t is still as large as 0.22%, with a significant t-statistic of Adding control variables to the regressions does not alter the overall finding, though the economic magnitude shrinks for about half. Thus, the presented evidence in the stock market suggests that politically-connected firms have correlated fundamentals Cross-Predictability Among Politically-connected Firms The above evidence reveals a previously undocumented but valuable channel of information flow among firms along the political network. Although firm executives political contributions are publicly available, investors may fail to link firms together due to limited attention or insufficient ability of information processing, leading to a substantive return predictability among politicallyconnected firms. 16 In Table A.5., I show qualitatively similar but weaker results using equal weighted scheme. 14

16 As a first step, I now investigate whether stock returns of politically-connected firms are cross-predictable. To do so, I conduct the same regression framework as Eq. (3), except that now the dependent variable is the next month s stock return denoted in percentage. Specifically, I run the Fama and MacBeth (1973) forecasting regressions and adjust for heteroscedasticity and autocorrelation using Newey and West (1987) up to 12 lags as follows: Ret i,t+1 = α t+1 + β t Linkret i,t + Λ t Z i,t + ε t+1, (4) where Z i,t is a vector of control variables including firm i s own 1-month lagged return, its corresponding Fama-French 12-industry 1-month lagged return, same-state firm portfolio 1-month lagged return, as well as its own, industry s, and geographic momentum return cumulated from month t-11 to t-1. Firm size (log market capitalization) and log book-to-market ratio measured at the end of prior year, and total number of contributions measured over the past four are also included. Table 3 shows the basic results of the paper. In Model (1), Politically-connected firm crosspredicts each other s return as evidenced by the statistically significant coefficient of Linkret i,t at 0.07%, with a t-statistic of Regression specifications that include different combinations of control variables all yields qualitatively similar results for Linkret i,t. In terms of the economic magnitude, a one-standard-deviation increase in returns of firm i s politically-connected firms this month leads to a 7 basis points increase in its return next month, suggesting that the subsequent price drift equals to approximately one third of the contemporaneous price association. 15

17 To probe the effect of political connectedness strength on the cross-predictability in return, I calculate Linkret i,t based on a subgroup of strongly or weakly politically-connected firms partitioned by the median number of common contributed candidates, and then run the forecasting regressions for each specification, respectively. Table 4 presents evidence suggesting that the only closely related firms in the political network exhibit cross-predictability in stock returns. In Panel A, coefficient of Linkret i,t constructed on returns of strongly-connected firms is at 0.08% (tstatistic: 3.30), while the effect is largely absent for Linkret i,t constructed on returns of weakly politically-connected firms. Panel B validates the above finding using firm i s excess return over its Fama-French 12- industry return as the dependent variable. Even after purging out the industry effect, the statistical significance and economic magnitude of Linkret i,t are essentially unchanged. An interesting note is that more executive political contributions do not seem to predict higher future stock returns; whereas Cooper, Gulen, and Ovtchinnikov (2010) document that contributions made by firm PACs are positively associated with future stock returns Geographic and Industry Effect Prior studies often view geographic domains as an important attribute of variations in political exposure. For example, Cohen, Coval, and Malloy (2011) show that congressional committee 17 In Table 11, I validate their finding using a sample based on firm PAC contributions. 16

18 chairman direct significantly more federal funding to their home states, and Kim, Pantzalis, and Park (2012) show that firms headquartered in states with higher federal political alignment have higher exposure to political risk. Therefore, stock market may react to information more promptly for intra-state politically-connected firms, but less so for geographical distant ones. To test this point, within each of the political strength subsample, I calculate Linkret i,t based on politically-connected firms that are headquartered in the same or different state(s), and then run the forecasting regressions for each specification, respectively. Consistent with the prior findings, Table 5 shows that the cross-predictability effect largely pertains to a group of strongly politically-connected firm-pairs. In addition, Panel A finds that the effect is strongest among closely connected firms whose headquarters are in different states, with a significant coefficient of Linkret i,t at 0.06% (t-statistic: 2.07), while the coefficient of Linkret i,t constructed on returns of same-state politically-connected firms is statistically indifferent from zero even though the difference is not significant from zero. The evidence alleviates the concern that the baseline finding is merely a reflection of common geographic factor and, furthermore, lends support to an investor inattention story since information incorporation is more sluggish for firms across states. Sector classification is another common proxy for political landscape. For example, Belo, Gala, and Li (2013) find that firms in industries with higher exposure to government spending have higher future stock returns due to investor underreactions. In contrast, Addoum and Kumar (2016) find that investors direct their capital flows to firms in sectors that are expected to benefit from the political election, thus pushing up the stock prices in these industries. To test industry 17

19 effect on the return cross-predictability, I calculate Linkret i,t based on returns of politicallyconnected firms that are operated in the same or different sector(s), and then run the forecasting regressions for each specification, respectively. Panel A of Table 6 displays results based on the Fama-French 12-industry classifications and find that the effect largely pertains to a group of strongly politically-connected firm-pairs that are operated in different sectors evidenced by a significant coefficient of Linkret i,t at 0.10% (tstatistic: 2.98). However, the coefficient of Linkret i,t constructed on returns of same-industry politically-connected firms is statistically indifferent from zero even though the difference between two coefficients is not significant from zero. Alternatively, I employ the 10-K text-based firm-level product similarity measure in Hoberg and Phillips (2010, 2016) as a means of sector classifications as the authors show that the measure captures a time-varying product market condition to better identify industry rivals. Specifically, I use their TNIC data which are calibrated to mimic the granularity of two-digit SIC codes: two firms are considered in the same sector if they obtain a non-zero pairwise TNIC score; otherwise they are defined as different-sector firms. 18 Displayed in Panel B of Table 6, using the TNIC measure yields similar finding as that using the Fama-French 12-industry classifications. That is, the cross-predictability in returns largely pertains to a group of strongly politicallyconnected firm-pairs that are operated in different product markets, with a significant coefficient 18 I thank Gerard Hoberg and Gordon Phillips for making 10-K Text-based Network Industry Classifications (TNIC) available on their website: 18

20 of Linkret i,t at 0.07% (t-statistic: 2.29). However, the coefficient of Linkret i,t constructed on returns of same-product-market politically-connected firms is statistically indifferent from zero even though the difference between two coefficients is not significant from zero. The overall findings provided in this section evidently demonstrate that common clienteles create a type of echo chamber, whereby information transmission is rapid among politicallyconnected firms within same geographic and sector boundary, but less so outside them. Sluggish information dissemination for inter-boundary politically-connected firms thus generates lead-lag effects Political and Economic Linkage The fact that the return cross-predictability is most pronounced among politically-connected firms operated in different industries raises concerns whether those firms are indeed fundamentally related. Cohen and Frazzini (2008) and Menzly and Ozabas (2010) document that although operated in different industries, firms along the supply chains are nevertheless economically related. To form the economic linkage, for each industry I use annual Benchmark Input-Output survey from the Bureau of Economic Analysis to identify primary supplier and customer industries. 19 Following Ahern and Harford (2014), for each industry I create the supplier and 19 To mimic the granularity of two-digit SIC codes, I use the summarized IO tables which classify all businesses into 71 industries and are updated every year. 19

21 customer matrices using information contained in the Use and Make Tables. 20 Each industry s supplier and customer industries are then ranked based on two industries relative trade flows. For robustness, for each industry I consider three scenarios based on different levels of economic linkage granularity: either top ranked, top 3 ranked, or top 10 ranked supplier and customer industries, as economically-linked industries, respectively. 21 Last, I map the industry economic linage to firms in the political network based on industries they belong to. Next, I calculate Linkret i,t based on returns of politically-connected firms that are operated in the economically related or unrelated industry(s), and then run the forecasting regressions for each specification, respectively. Table 7 shows that stock returns of politicallyconnected firms are cross-predictable only among those that are operated in economically related industries. For example, coefficient of Linkret i,t constructed based on returns of strongly politically-linked firms operated in top ranked supplier or customer industries is approximately 0.24% with a significant t-statistic of 3.60, while the coefficient is insignificant for politicallyconnected firm-pairs that are considered as economically unrelated. And the difference between the two coefficients is significantly different from zero, with a p-value of One may also concern that the return cross-predictability among politically-connected firms is subsumed by that along the economic network. In Table A.I, I run a placebo test using a group of weakly politically-linked firms to rule out this possibility. If economic linkage, rather 20 See Ahern and Harford (2014) for detailed process of supplier and customer matrices constructions. 21 Note, to a specific industry, it is likely that one industry can be both the top supplier and customer industries. For example, industry B can be industry A s most important supplier industry as well as its most important customer industry. 20

22 than political linkage, is the dominating force, the effect should continue to be pronounced even among weakly politically-connected firms. Contradicting to this view, I find that there is no significant return cross-predictability among weakly politically-connected firms even though they are economically related. The evidence hence indicates that there is a unique and valuable channel of information flow among politically-connected firms beyond their supply chain linkage Variation in Investor Attention The evidence presented thus far regarding industry and implies that the cross-predictability effect is due to investor inattention. To probe this proposition, I employ two stock-level measures which are widely recognized in the literature as investor (in)attention proxies: (1) firm size measured as the log of market capitalization at the end of prior calendar year; and (2) the percentage of quarterly mutual fund ownership at the end of prior quarter for each stock. In Panel A of Table 8, I examine the impact of investor inattention on the crosspredictability based on mutual fund ownerships. The mutual fund holding data is obtained from Thomson Reuters 12S mutual fund holding database. For each stock, I sum all actively managed domestic equity mutual fund holdings at a given quarter to derive the mutual fund ownership. 22 At the beginning of each month, I partition the stocks into two subsamples based on their mutual fund ownerships at prior quarter end. Consistent with the investor inattention argument, the return cross- 22 To control for firm size, the measure is divided by the number of outstanding shares. 21

23 predictability effect is most pronounced among low-mutual-fund-owned stocks with the Linkret i,t coefficient of 0.11% (t-statistic: 2.75), while the effect is minimal for stocks with high mutual fund ownership at a coefficient of 0.05 (t-statistic: 1.35). Panel B of Table 8 draws similar conclusion using firm size as a proxy for investor attention. Among large firms, A one-standard-deviation increase in returns of strongly politically-connected firms this month leads to a 10 basis points increase in next-month returns, while the effect is insignificant among a group of small firms. All in all, the results are supportive of the proposition that the lead-lag effect among politicallyconnected firms is driven by inattentive investors that overlook the political channel of information flow among firms Common Sources of Fundamentals along the Political Network In this section, I explore the underlying rationale that firms contribute to many political candidates in common, resulting in subsequent a subsequent return cross-predictability. I postulate that firms contribute to common political candidates because they share similar types of political risks and are eager to manage the political exposures by investing in political capital. Shared exposures regarding certain political issues imply that firms have correlated fundamentals, leading to the return comovement and a subsequent return cross-predictability documented in this paper. To exploit this line of reasoning, I employ a firm-level political risk measure from Hassan et al. (2017). For each firm, the authors apply textual analysis of quarterly earnings conferencecall transcripts to construct a political risk matrix. This measure is suitable for this analysis for two 22

24 reasons: first, it captures individual firm s political exposure perceived by executives and financial analysts, which is aligned with my political network constructed on CEO s campaign contributions; It addition, the authors construct individual firm s political risk matrices on eight different political topics, allowing me to identify whether two politically-connected firms face same type of political exposures. Specifically, for each firm every year, I calculate a simple average of four-quarter risk scores for each of the eight political types and identify the highest-scored political risk type as the firm s primary political concern. If shared political risk is the underlying explanation, I expect to observe that the return cross-predictability mostly exhibits among firms concerning same political issues. To do so, I calculate Linkret i,t based on returns of strongly politically-connected firms that have either the same or different top-ranked political risk type(s), and then run the forecasting regressions for each specification, respectively. Indeed, Table 9 shows that the coefficient of Linkret i,t is significant only for politically-connected firms that share the same type of political exposures, at 0.09% (t-statistic: 2.15), while the coefficient is statistically indifferent from zero for firms that care different political issues even though they are strongly politically-connected Citizen United v. FEC as Identification of Variation in Political Activism I now introduce a shock to the political activism for firms in some states to better identify the underlying mechanisms: the U.S. Supreme Court s landmark decision on Citizen United v. Federal Election Commision (CU) in January 2010 lifts prior bans on corporations to use their treasuries 23

25 to advocate in favor or against a political candidate on a federal election. The ruling of Supreme Court also overrules the prior bans on independent political expenditures by corporations on state elections in 21 states. 23 Although the political network in this paper focuses on federal elections, bans for the state election independent expenditures nonetheless give rise to a negative effect on political activism for firms in these states by signaling a conservative view of corporate political involvement and discouraging companies political activities. Specifically, I use politically-connected firms in ban states as the treatment group and those in no-ban states as the control group to study the differential patterns in return cross-predictability between these two groups surrounding CU using a window from 2008 to Albuguerque et al. (2017) apply the same identification strategy and show that firms in ban states relied on more traditional forms of political activism before CU, such as adding politicians as board members. To capture this cross-sectional difference, I interact the independent variable of interest, Linkret i,t, with ban states, which is a dummy variable that equals one if the headquarter of a firm locates in a state that had bans on independent expenditure on state elections and zero otherwise. I then conduct the monthly Fama and MacBeth (1973) forecasting regressions with the interaction term added every year from 2008 to In Table 10, the interaction term is negatively associated with next-month firm return in years prior to CU, suggesting that the crosspredictability is significantly weaker for firms located in ban states than those located in no-ban states. However, after the passage of CU when bans on corporation independent expenditure on 23 See Spencer and Woods (2014) for more details on 2010 Citizen United v. FEC. 24

26 state elections are lifted in these 21 states, the differential (negative) pattern between firms in ban states and those no-ban states becomes insignificant and dissipates. The difference-in-difference analysis of Table 10 further indicates that the change in magnitude of return cross-predictability before and after the CU is approximately 12% per annum larger for firms in ban states compared to those in non-ban states. The event study based on CU, an exogenous and positive shock to the activeness of political involvement for firms in some states, finds support that the effect of crosspredictability is significantly stronger among firms that are allowed to manage their political exposure more actively and freely through campaign contributions Political Network Using Firm PAC Contributions As an alternative source, in this section I construct the political network using firm political action committee (PAC) contributions. In contrast to executive contributions, Firm PAC strategically contributes to a significantly broader and more diversified pool of candidates, thus resulting in a larger political network. 24 Despite of the distinct contribution patterns, it is of interest to examine whether the return cross-predictability is also exhibited through the political network constructed on firm PAC contributions. Analogy to the executive political network construction, I calculate the number of common contributed political candidates between two firms based on PAC contributions over the past four years following Eq. (2). Table 11 finds similar result using firm 24 Table A.2. presents the summary statistics of S&P500 firm PAC contributions over the sample period from 1983 to

27 PAC contribution data. Specifically, a one-standard-deviation increase in returns of politicallyconnected firms this month leads to a 17 basis points increase in the firm i s return next-month, with a t-statistic of Politically-connected Portfolio Returns In this section, I construct the calendar-time portfolio based on returns of politically-connected firms. At the beginning of each month, firms are sorted into quartile portfolios based on returns of their strongly politically-connected firms in the previous month. I then construct a zero-cost portfolio that buys stocks whose strong politically-connected counterparts performed in the top quartile prior month and sells stocks whose strong politically-connected counterparts performed in the bottom quartile prior month. The portfolio is equal-weighted and rebalanced every month, and referred to as the political momentum portfolio henceforth. Results for political network constructed using contributions made by either firm executives or firm PACs are reported, respectively. Panel A of Table 12 shows that portfolio return increases nearly monotonically in lagged returns of politically-connected firms. For example, in the sample of firm PAC contributions, the long-short portfolio strategy yields a spread of 38 basis points per month (t-statistic: 3.41). Furthermore, the strategy profit is robust to traditional risk factors, such as the Fama and French (1993) three factors, Carhart s (1997) momentum factor, and Pástor and Stambaugh (2003) liquidity factor. Specifically, the five-factor political momentum portfolio alpha is still a 26

28 remarkable 33 basis points per month (t-statistic: 2.64). In Table A.3., I also expand the look-back and forecast horizons of the political momentum portfolio. Generally, expanding the horizon reduces forecasting ability, reflected in both smaller coefficients and, to a smaller extent, statistical significance. The fact that the return cross-predictability is strongest at 1-month forecast horizon suggests that the lead-lag effect in information dissemination among politically-connected firms is pervasive in the short-run. In Panel B, I examine the factor exposure on the return of political momentum portfolio to check if it loads on certain systematic risks substantially. Overall, it insignificantly correlates with the five risk factors in most specifications. The only exception is that the political momentum portfolio spread positively loads on the momentum and liquidity factors in the sample using firm PAC contributions. Nevertheless, the evidence is convincing: measured using either excess returns, industry-adjusted returns, CAPM alphas, three-factor alphas, or five-factor alphas, the political momentum strategy consistently earns a significant abnormal premium, ranging from about 1.8% (t-statistic: 3.28) per annum to 5.5% per annum (t-statistic: 4.01). 25 Figure 1 plots the cumulative returns of the political momentum portfolio strategy. The strategy, in general, is profitable over the entire period using either firm executive (Panel A) or firm PAC (Panel B) sample, while appears to be less so during election (even-numbered) years. This observation seems intuitive because during election years investors ought to be more attentive 25 In Table A.4., I also show that the political momentum portfolio return is insignificantly (significantly) associated with the geographic (economic) momentum portfolio return, which is in line with the finding in Table 5 (7). 27

29 to political activities. In Table 13, I formally test this idea by adding an election-year dummy to the time-series regression and confirm that the political momentum portfolio five-factor alpha is approximately 3.5% smaller per annum in election years in contrast to non-election years. Therefore, the finding lends additional support to an investor inattention story for the return crosspredictability among politically-connected firms. 4. Robustness Checks In this section, I conduct a battery of robustness checks to validate the baseline findings documented in the paper Cross-predictability in Operating Performance Throughout the paper, I show that politically-connected firms are cross-predictable in their stock returns. It is of importance to examine whether politically-connected firms fundamental performances are cross-predictable. Specifically, I measure fundamental performance using three different operating ratios: return on assets (ROA), gross profitability (Profit) from Nory-Marx (2013), and seasonal-adjusted earnings (EPS). Results in Table 14 indicate that politically-connected firms are cross-predictable in operating performance using either quarterly Fama and MacBeth (1973) regressions or panel regressions that controls for time fixed effects. For example, one unit increase in ROA of its 28

30 strongly politically-connected firms this quarter leads to about 2.33 unit increase in firm i sroa next quarter, with a t-statistic of 2.27 in panel regression after controlling for its own ROA, industry and geographic ROA this quarter, as well as other variables. One exception is that there is no cross-predictability in seasonal-adjusted earnings 4.2. Alternative Measure Formation Horizons and Connectedness Strength cutoff I now examine whether the baseline findings are sensitive to the choice of political network formation horizons. Panel A and B of the Table 15 show that the effect is apparent and robust to the two-year and five-year measurement horizons, respectively. For example, under the five-year measurement horizon, a one-standard-deviation increase in returns of strongly politically-linked counterparts this month leads to a 10 basis points increase in the firm return next month, with a t- statistic of Moreover, the main conclusion of the paper is that only strongly, but not weakly, politically-connected firms are fundamentally correlated, and cross predictable in returns. The strength of political connectedness is determined by whether politically-connected firms are ranked above or below the median within a political network based on the number of common contributed candidates. I now rank a firm s politically-connected counterparts based on their number of common contributed candidates, and assign the top 3, 5 or 10 ranked firms as strongly connected firms, while the rest are categorized as weakly-connected firms, respectively. 29

31 The alternative cutoff for connected strength delivers similar results, as Table 16 shows that politically-connected firms are cross-predictable in returns only for those closely linked ones. For example, a one-standard-deviation increase in averaged return of top 3 politically-connected counterparts this month leads to an 8 basis points increase in the firm return next month, with a t- statistic of Controlling for Political Activeness and CEO Tenure In all regression specifications, the total number of a firm s political candidate contributions is included to control for the political activeness. However, one might still be concerned that the fact that two firms have many common contributed candidates may just reflect that they are both active in political activism rather than truly connected. To alleviate this concern, I take the political activeness into account when constructing the political network: AlternateConnection i,j,t = CommonContribute i,j,t Min(Contribute i,t, Contribute j,t ), (5) where CommonContribute i,j,t is the baseline measure in Eq. (2) which counts the number of common contributed candidates between firm i and firm j over the years t-3 to t, while Min(Contribute i,t, Contribute j,t ) is the minimum value of firm i s and firm j s total number of political contributed candidates, which is the maximum number of common candidates that firm i and firm j are ever possible to achieve. 30

32 Panel A of Table 17 indicates that the finding in the paper is not driven by the political activeness. Using this alternative measure that controls for political activeness, I continue to find that strongly politically-connected firms can cross-predict each other s stock returns, with a significant coefficient at 0.06% (t-statistic: 1.85). So far, all contributions from a firm s CEO are included as long as they are made during the four-year formation period. However, one may argue that if the CEO is no longer incumbent at the time of political network construction, his/her contributions are no longer meaningful to the firm s political activism. To check this point, I consider executive contributions only from CEOs who are in office at the time of political network formation. Panel B of Table 17 suggests that controlling for CEO tenure has little impact on the overall finding. That is, I continue to find qualitatively similar result, with a significant coefficient of Linkret i,t at 0.07% (t-statistic: 3.08). 5. Conclusion Using data on firm executive political contributions from the U.S. Federal Election Commission over the period of 2000 through 2016, this paper explores a previously undocumented inter-firm political links measured based on common campaign contributions made by firm executives. Politically connected firms are fundamentally related as evidenced by the significant contemporaneous return comovment. 31

33 Further analysis shows that returns of politically connected firms are cross-predictable, and I reason that this incomplete price reaction results mainly from investors gradual information incorporation. Indeed, the fact that the cross-predictability effect mostly pertains to politically connected firms that are smaller, less owned by mutual funds, and operate in different states and sectors lends support to a limited investor attention explanation. To probe the economic magnitude, Portfolios based on political ties yield risk-adjusted returns of 4-5% per annum. I next explore the underlying sources and demonstrate that the fundamental correlation between politically connected firms is mainly due to their exposures to common political risks. Using the 2010 U.S. Supreme Court s landmark decision on Citizen United v. Federal Election Commision (CU) as an exogenous shock to the political activism for firms in ban states, further evidence indicates that the effect is weaker for those firms that are restricted from actively engaging in political campaigns. The findings in this paper thus highlight the importance of understanding inter-firm political links given the recent complexity of political landscapes faced by both firms and investors. The evidence presented herein also sheds light on research in better understanding the information dissemination as well as limited investor attention in the financial market. 32

34 References Addoum, Jawad M., and Alok Kumar, 2016, Political sentiment and predictable returns, Review of Financial Studies 29, Ahern, Kenneth R., and Jarrad Harford, 2014, The importance of industry links in merger waves, Journal of Finance 69, Akey, Pat, 2015, Valuing changes in political networks: Evidence from campaign contributions to close congressional elections, Review of Financial Studies 28, Akey, Pat, and Stefan Lewellen, 2017, Policy uncertainty, political capital, and firm risk-taking, Working Paper. Belo, Frederico, Vito D. Gala, and Jun Li, 2013, Government spending, political cycles, and the cross section of stock returns, Journal of Financial Economics 107, Cohen, Lauren, Joshua Coval, and Christopher Malloy, 2011, Do powerful politicians cause corporate downsizing?, Journal of Political Economy 119, Cohen, Lauren, Karl Diether, and Christopher Malloy, 2013, Legislating stock prices, Journal of Financial Economics 110, Cohen, Lauren, and Andrea Frazzini, 2008, Economic links and predictable returns, Journal of Finance 63, Duchin, Ran, and Denis Sosyura, 2012, The politics of government investment, Journal of Financial Economics 106, Engelberg, Joseph, Pengjie Gao, and Christopher A. Parsons, 2013, The Price of a CEO's Rolodex, Review of Financial Studies 26,

35 Faccio, Mara, 2006, Politically connected firms, American Economic Review 96, Fama, Eugene F., and James D. MacBeth, 1973, Risk, return, and equilibrium: Empirical tests, Journal of Political Economy 81, Fama, Eugene F., and Kenneth R. French, 1993, Common risk factors in the returns on stocks and bonds, Journal of Financial Economics 33, Fracassi, Cesare, and Geoffrey Tate, 2012, External networking and internal firm governance, Journal of Finance 67, Hassan, Tarek A., Stephan Hollander, Laurence van Lent, and Ahmed Tahoun, 2017, Firm-level political risk: Measurement and effects, Working Paper. Hoberg, Gerard, and Gordon Phillips, 2010, Product market synergies and competition in mergers and acquisitions: A text-based analysis, Review of Financial Studies 23, Hoberg, Gerard, and Gordon Phillips, 2016, Text-based network industries and endogenous product differentiation, Journal of Political Economy 124, Hong, Harrison, and Leonard Kostovetsky, 2012, Red and blue investing: Values and finance, Journal of Financial Economics 103, Hong, Harrison, and Jeremy C. Stein, 1999, A unified theory of underreaction, momentum trading, and overreaction in asset markets, Journal of finance 54, Kim, Chansog Francis, Christos Pantzalis, and Jung Chul Park, 2012, Political geography and stock returns: The value and risk implications of proximity to political power, Journal of Financial Economics 106, Kostovetsky, Leonard, 2015, Political capital and moral hazard, Journal of Financial Economics 116,

36 Menzly, Lior, and Oguzhan Ozbas, 2010, Market segmentation and cross-predictability of returns, Journal of Finance 65, Newey, Whitney K., and Kenneth D. West, 1987, A simple, positive semi-definite, heteroskedasticity and autocorrelation consistent covariance matrix, Economertrica 55, Parsons, Christopher A., Riccardo Sabbatucci, and Sheridan Titman, 2016, Geographic Momentum, Working paper. Pástor, Ľuboš, and Pietro Veronesi, 2013, Political uncertainty and risk premia, Journal of Financial Economics 110,

37 Table 1. Summary Statistics of the Firm Executive Political Contribution Data This table reports the summary statistics of the S&P500 firm executive (CEO) political campaign contribution variables as of October of each year (political year-end). Panel A reports the summary statistics of number of firms donating in political campaigns, number of candidates a firm donates to, and number of firms having political networks (connecting to at least one other firm through common contributed political candidates), measured over the past four years. Panel B illustrates the summary statistics of political network size (number of politically connected firms to a firm), proportion of connected firms that are in the same industry and whose headquarters are in the same state, and average number of common contributed political candidates per firm, measured over the past four years. At the bottom of Panel B, summary statistics of strongly and weakly politically connected subsamples are also displayed, respectively (a specific firm s connected firm is considered as a strongly (weakly) connected counterpart if the number of common contributed candidates between the two firms are above (below) the median of all its connected firms measured over the past four years). The sample is over the period from 2004:01 to 2016:12. Firm Executive Contribution Sample ( ) Contribution Variable Mean SD Min 25 th Pctl Median 75 th Pctl Max Panel A: Overall Sample Statistics Number of firms contributing to political campaign over years Number of contributed candidates per firm over years Number of firms with political networks (with other firms) over years Panel B: Time-Series Average of Cross-sectional Summary Statistics (Full Sample) Number of connected firms per firm % of connected counterparts in the same industry per firm % of connected counterparts in the same state per firm Average number of common candidates contributed per firm Strongly Connected Subsample Number of connected counterparts per firm % of connected counterparts in the same industry per firm % of connected counterparts in the same state per firm Average number of common candidates contributed per firm Weakly Connected Subsample Number of connected counterparts per firm % of connected counterparts in the same industry per firm % of connected counterparts in the same state per firm Average number of common candidates contributed per firm

38 Table 2. Contemporaneous Political Connected Return Correlations This table reports the monthly Fama and MacBeth (1973) regressions (Panel A) and panel regressions (Panel B) of the S&P500 firm return on contemporaneous one-month return of its politically connected firms with different model specifications. The dependent variable is the current one-month stock return (RET i,t ) in percentage, and the independent variable of interest is the return of its politically connected firms (LINKRET i,t, common-candidate-number-weighted). Control variables include firm s lagged 1-month return (RET i,t 1 ), its corresponding Fama-French 12- Industry return (INDRET i,t 1 ), same-state portfolio lagged 1-month return (GEORET i,t 1 ), as well as the firm s, industry s, and geographic momentum returns (past 11-month cumulative returns measured at [t-12, t-2]). Log of market capitalization (SIZE) and log book-to-market ratio (B/M) measured at prior year-end, as well as total number of contributed candidates over the past four years are also included. Cross-sectional regressions are run every month and the t-statistics reported in the square bracket are adjusted for heteroscedasticity and autocorrelation using Newey and West (1987) up to 12 lags. The panel regression includes the time fixed effects and the standard errors are clustered by firm. The sample is over the period from 2004:01 to 2016:12. Panel A: Fama and MacBeth (1973) Regressions Panel B: Panel Regressions Dep.Var.= RET t (%) (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) LINKRET i,t [6.98] [7.35] [6.23] [7.08] [7.90] [7.43] [7.08] [7.01] [6.76] [7.03] [5.87] [7.03] [7.03] [7.03] [7.04] [7.02] [7.04] [7.03] [7.00] [7.03] [7.03] [6.99] RET i,t [-0.86] [-1.69] [-1.31] [-1.92] INDRET i,t [0.38] [1.32] [1.10] [1.27] GEORET i,t [1.43] [1.46] [1.27] [1.22] RET i,t 11,t [0.84] [0.34] [2.33] [1.47] INDRET i,t 11,t [0.98] [0.63] [2.99] [2.29] GEORET i,t 11,t [0.31] [-1.24] [0.93] [-0.03] SIZE [-2.23] [-2.28] [-4.96] [-4.74] B/M [-0.13] [-0.58] [0.10] [-1.14] Contribute [-2.77][-0.48] [-2.39][-0.79] No. of Obs 54,911 54,911 54,911 54,911 54,911 54,911 54,911 54,911 54,911 54,911 54,911 54,911 54,911 54,911 54,911 54,911 54,911 54,911 54,911 54,911 54,911 54,911 37

39 Table 3. Return Cross-predictability Between Politically-connected firms This table reports the monthly Fama and MacBeth (1973) regressions of the S&P500 firm return on lagged one-month returns of its politically connected firms with different model specifications. The dependent variable is the future onemonth stock return (RET i,t+1 ) in percentage, and the independent variable of interest is the return of its politically connected firms (LINKRET i,t, common-candidate-number-weighted). Control variables include firm s lagged 1- month return (RET i,t ), its corresponding Fama-French 12-Industry return (INDRET i,t ), same-state portfolio lagged 1- month return (GEORET i,t ), as well as the firm s, industry s, and geographic momentum returns (past 11-month cumulative returns measured at [t-11, t-1]). Log of market capitalization (SIZE) and log book-to-market ratio (B/M) measured at prior year-end, as well as total number of contributed candidates over the past four years are also included. Cross-sectional regressions are run every month and the t-statistics reported in the square bracket are adjusted for heteroscedasticity and autocorrelation using Newey and West (1987) up to 12 lags. The sample is over the period from 2004:01 to 2016:12. Dep.Var.= RET t+1 (%) (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) LINKRET i,t [2.58] [2.65] [2.35] [2.63] [2.73] [2.51] [2.71] [2.44] [2.65] [2.42] [2.39] RET i,t [-0.69] [-1.55] INDRET i,t [0.35] [1.40] GEORET i,t [1.19] [1.12] RET i,t 11,t [0.82] [0.47] INDRET i,t 11,t [0.93] [0.49] GEORET i,t 11,t [0.32] [-0.93] SIZE [-2.15] [-2.12] B/M [-0.26] [-0.69] Contribute [-3.21] [-0.67] No. of Obs 54,770 54,770 54,770 54,770 54,770 54,770 54,770 54,770 54,770 54,770 54,770 38

40 Table 4. Industry-adjusted Return and Political Connection Strength This table examines the findings in Table 3 using industry-adjusted return and accounting for political connectedness strength. The dependent variable is the future one-month stock return (Panel A) or Fama-French 12-industry adjusted return (Panel B) in percentage. The independent variable of interest is the return of its politically connected firms (LINKRET i,t, common-candidate-number-weighted). From left to right in each panel, results using samples of all, strongly, and weakly politically connected firms are exhibited, respectively (a firm s politically connected firm is considered as strongly (weakly) connected if the number of common contributed candidates between them is above (below) the median of all its connected firms measured over the past four years). Control variables include firm s lagged 1-month return (RET i,t ), its corresponding Fama-French 12-Industry return (INDRET i,t ), same-state portfolio lagged 1-month return (GEORET i,t ), as well as the firm s, industry s, and geographic momentum returns (past 11- month cumulative returns measured at [t-11, t-1]). Log of market capitalization (SIZE) and log book-to-market ratio (B/M) measured at prior year-end, as well as total number of contributed candidates over the past four years are also included. Cross-sectional regressions are run every month and the t-statistics reported in the square bracket are adjusted for heteroscedasticity and autocorrelation using Newey and West (1987) up to 12 lags. The sample is over the period from 2004:01 to 2016:12. Panel A: Raw Return (%) Panel B: Industry-adjusted Return (%) (1) All Connected (2) Strongly Connected (3) Weakly Connected (4) All Connected (5) Strongly Connected (6) Weakly Connected LINKRET i,t [2.39] [3.30] [-0.75] [2.21] [3.61] [-1.12] RET i,t [-1.55] [-1.64] [-1.38] [-1.24] [-1.31] [-1.25] INDRET i,t [1.40] [1.42] [2.35] [1.28] [1.35] [2.39] GEORET i,t [1.12] [1.16] [-0.12] [1.58] [1.61] [0.07] RET i,t 11,t [0.47] [0.45] [0.89] [0.38] [0.37] [0.81] INDRET i,t 11,t [0.49] [0.55] [-0.18] [-0.81] [-0.76] [-1.72] GEORET i,t 11,t [-0.93] [-0.96] [-1.34] [-0.58] [-0.60] [-0.57] SIZE [-2.12] [-2.08] [-1.68] [-1.87] [-1.83] [-1.45] B/M [-0.69] [-0.69] [-0.66] [-0.88] [-0.88] [-0.74] Contribute [-0.67] [-0.56] [-0.00] [-0.89] [-0.77] [0.10] No. of Obs 54,770 54,770 36,737 54,770 54,770 36,737 39

41 Table 5. Return Cross-predictability and Geographic Proximity This table reports the effect of geographic proximity on the return cross-predictability along the inter-firm political links. The dependent variable is the future one-month stock return (RET i,t+1 ) in percentage, and the independent variable of interest is the return of its politically connected firms (LINKRET i,t, common-candidate-number-weighted). From left to right in each panel, the strongly and weakly politically connected firms are further partitioned based on whether their headquarters are in the same state (a firm s politically connected firm is considered as strongly (weakly) connected if the number of common contributed candidates between them is above (below) the median of all its connected firms measured over the past four years). Control variables include firm s lagged 1-month return (RET i,t ), its corresponding Fama-French 12-Industry return ( INDRET i,t ), same-state portfolio lagged 1-month return (GEORET i,t ), as well as the firm s, industry s, and geographic momentum returns (past 11-month cumulative returns measured at [t-11, t-1]). Log of market capitalization (SIZE) and log book-to-market ratio (B/M) measured at prior year-end, as well as total number of contributed candidates over the past four years are also included. The geographic information of firm headquarters is obtained from Compustat. Cross-sectional regressions are run every month and the t-statistics reported in the square bracket are adjusted for heteroscedasticity and autocorrelation using Newey and West (1987) up to 12 lags. The p-value of the regression coefficient (LINKRET i,t ) difference in two subsamples is reported in the bracket. The sample is over the period from 2004:01 to 2016:12. Panel A: Strongly Politically-connected Panel B: Weakly Politically-connected (1) Same State (2) Diff State (3) Same State (4) Diff State Diff (p-val) Headquartered Headquartered Headquartered Headquartered Diff (p-val) LINKRET i,t [1.28] [2.07] (0.71) [1.13] [-0.74] (0.23) RET i,t [-1.69] [-1.36] [-1.43] [-1.39] INDRET i,t [1.76] [1.31] [2.69] [2.37] GEORET i,t [-0.13] [0.70] [-1.38] [-0.26] RET i,t 11,t [0.53] [0.30] [0.59] [0.88] INDRET i,t 11,t [-0.02] [0.93] [-0.18] [-0.19] GEORET i,t 11,t [-0.27] [-0.17] [-1.94] [-1.31] SIZE [-1.76] [-2.05] [-1.70] [-1.75] B/M [-0.57] [-1.03] [-0.00] [-0.67] Contribute [0.41] [-0.80] [-0.14] [-0.12] No. of Obs 40,518 48,378 33,360 36,582 40

42 Table 6. Return Cross-predictability and Industry Effect This table reports the effect of industry classification on the return cross-predictability along the inter-firm political links. The dependent variable is the future one-month stock return (RET i,t+1 ) in percentage, and the independent variable of interest is the return of its politically connected firms (LINKRET i,t, common-candidate-number-weighted). From left to right in each panel, strongly (weakly) politically connected firms are partitioned into two subsamples based on whether they are in the same industry (a firm s politically connected firm is considered as strongly (weakly) connected if the number of common contributed candidates between them is above (below) the median of all its connected firms measured over the past four years). Control variables include firm s lagged 1-month return (RET i,t ), its corresponding Fama-French 12-Industry return (INDRET i,t ), same-state portfolio lagged 1-month return (GEORET i,t ), as well as the firm s, industry s, and geographic momentum returns (past 11-month cumulative returns measured at [t-11, t-1]). Log of market capitalization (SIZE) and log book-to-market ratio (B/M) measured at prior year-end, as well as total number of contributed candidates over the past four years are also included. Panel A uses Fama-French 12-Industry category and Panel B uses the text-based product similarity measure (Hoberg and Philips (2010)) to classify same/different industry pairs, respectively. Cross-sectional regressions are run every month and the t-statistics reported in the square bracket are adjusted for heteroscedasticity and autocorrelation using Newey and West (1987) up to 12 lags. The p-value of the regression coefficient (LINKRET i,t ) difference in two subsamples is reported in the bracket. The sample is over the period from 2004:01 to 2016:12. Panel A: Fama-French 12-Industry Panel B: Hoberg-Philips Product Similarity Strongly Connected Weakly Connected Strongly Connected Weakly Connected (1) Same Industry (2) Diff Industry Diff (p-val) (3) Same Industry (4) Diff Industry Diff (p-val) (5) Similar Product (6) Diff Product Diff (p-val) (7) Similar Product (8) Diff Product Diff (p-val) LINKRET i,t [0.79] [2.98] (0.45) [-0.56] [-0.41] (0.82) [1.04] [2.29] (0.64) [-0.02] [-0.14] (0.85) RET i,t [0.13] [-1.57] [-0.79] [-1.33] [0.16] [-1.50] [-0.78] [-1.37] INDRET i,t [0.85] [1.40] [2.05] [2.31] [0.91] [1.52] [2.22] [2.38] GEORET i,t [0.81] [1.11] [0.09] [-0.26] [1.23] [0.86] [0.73] [-0.19] RET i,t 11,t [0.41] [0.48] [0.67] [0.91] [0.54] [0.42] [0.73] [0.89] INDRET i,t 11,t [1.27] [0.59] [-0.09] [-0.27] [0.62] [0.57] [0.38] [-0.18] GEORET i,t 11,t [-0.10] [-0.82] [-1.33] [-1.39] [-0.25] [-0.90] [-1.85] [-1.29] SIZE [-1.92] [-2.08] [-1.69] [-1.65] [-1.49] [-2.06] [-1.56] [-1.66] B/M [-1.55] [-0.74] [-0.64] [-0.68] [-2.14] [-0.67] [-1.46] [-0.65] Contribute [0.35] [-0.46] [0.68] [-0.01] [0.96] [-0.53] [1.31] [-0.02] No. of Obs 34,599 53,123 33,905 36,673 26,126 53,686 28,933 36,699 41

43 Table 7. Return Cross-predictability and Economic Linkage This table reports the effect of economic links on the return cross-predictability along the inter-firm political links. The dependent variable is the future one-month stock return (RET i,t+1 ) in percentage, and the independent variable of interest is the return of its strongly politically connected firms (LINKRET i,t, common-candidate-number-weighted). From left to right in each panel, the strongly politically connected firms are partitioned into two subsamples based on whether they are in the same supply chain network (a firm s politically connected firm is considered as strongly connected if the number of common contributed candidates between them is above the median of all its connected firms measured over the past four years). Economically related industries are defined as top 1, 3 and 10 supplier and customer industries calculated using the annual Input-Output flows from the Bureau of Economic Analysis surveys following Ahern and Harford (2014). Control variables include firm s lagged 1-month return ( RET i,t ), its corresponding Fama-French 12-Industry return (INDRET i,t ), same-state portfolio lagged 1-month return (GEORET i,t ), as well as the firm s, industry s, and geographic momentum returns (past 11-month cumulative returns measured at [t-11, t-1]). Log of market capitalization (SIZE) and log book-to-market ratio (B/M) measured at prior year-end, as well as total number of contributed candidates over the past four years are also included. Cross-sectional regressions are run every month and the t-statistics reported in the square bracket are adjusted for heteroscedasticity and autocorrelation using Newey and West (1987) up to 12 lags. The p-value of the regression coefficient (LINKRET i,t ) difference in two subsamples is reported in the bracket. The sample is over the period from 2004:01 to 2016:12. Economically-linked Industry Classifications (Supplier/Customer Industries) (1) Top1 (2) N-top1 Diff (p-val) (1) Top3 (2) N-top3 Diff (p-val) (1) Top10 (2) N-top10 Diff (p-val) LINKRET i,t [3.60] [1.86] (0.04) [2.69] [1.35] (0.12) [2.89] [0.56] (0.13) RET i,t [0.09] [-0.75] [-0.11] [-0.70] [-0.30] [-0.54] INDRET i,t [-0.22] [1.45] [0.64] [1.34] [0.88] [1.63] GEORET i,t [0.50] [0.17] [0.36] [0.13] [-0.19] [-0.13] RET i,t 11,t [0.41] [0.17] [0.39] [0.10] [0.25] [0.17] INDRET i,t 11,t [0.33] [0.70] [-0.22] [0.69] [0.52] [0.50] GEORET i,t 11,t [-0.97] [-1.18] [-0.36] [-1.62] [-0.78] [-1.60] SIZE [-0.78] [-2.20] [-1.39] [-2.13] [-1.73] [-2.15] B/M [-2.51] [-0.78] [-1.86] [-0.68] [-0.70] [-0.92] Contribute [0.63] [-0.06] [-0.23] [-0.06] [0.20] [-0.14] No. of Obs 12,327 42,638 22,655 41,874 35,392 38,420 42

44 Table 8. Return Cross-predictability and Investor Attention This table reports the effect of investor attention on the return cross-predictability along the inter-firm political links. The dependent variable is the future one-month stock return (RET i,t+1 ) in percentage, and the independent variable of interest is the return of its politically connected firms (LINKRET i,t, common-candidate-number-weighted). From left to right in each panel, the strongly politically connected firms are partitioned into two subsamples based on the investor attention proxies (a firm s politically connected firm is considered as strongly connected if the number of common contributed candidates between them is above the median of all its connected firms measured over the past four years). Panel A uses last quarter s mutual fund ownership while Panel B uses firm size measured at the end of last year as investor attention proxies, respectively. Control variables include firm s lagged 1-month return ( RET i,t ), its corresponding Fama-French 12-Industry return (INDRET i,t ), same-state portfolio lagged 1-month return (GEORET i,t ), as well as the firm s, industry s, and geographic momentum returns (past 11-month cumulative returns measured at [t-11, t-1]). Log of market capitalization (SIZE) and log book-to-market ratio (B/M) measured at prior year-end, as well as total number of contributed candidates over the past four years are also included. Cross-sectional regressions are run every month and the t-statistics reported in the square bracket are adjusted for heteroscedasticity and autocorrelation using Newey and West (1987) up to 12 lags. The p-value of the regression coefficient (LINKRET i,t ) difference in two subsamples is reported in the bracket. The mutual fund ownership (size) sample is over the period from 2004:01 to 2015:12 (2016:12), respectively. Panel A: Mutual Fund Ownership Panel B: Firm Size (1) High (2) Low Diff (p-val) (3) Large (4) Small Diff (p-val) LINKRET i,t [1.35] [2.75] (0.26) [0.98] [2.36] (0.35) RET i,t [-1.30] [-1.59] [-2.08] [-0.92] INDRET i,t [1.14] [2.10] [0.80] [1.22] GEORET i,t [-0.51] [2.21] [0.74] [0.72] RET i,t 11,t [0.32] [1.23] [-0.15] [0.51] INDRET i,t 11,t [0.89] [0.46] [0.61] [0.81] GEORET i,t 11,t [-0.75] [0.10] [-1.37] [-0.17] SIZE [-2.95] [-1.78] [-1.23] [-1.45] B/M [-1.54] [-1.48] [-0.02] [-1.06] Contribute [-1.09] [0.90] [0.38] [-1.75] No. of Obs 25,849 25,764 27,431 27,339 43

45 Table 9. Sources of Return Cross-predictability: Same Type of Primary Political Risk This table reports the effect of political risk sharing on the return cross-predictability along the inter-firm political links. The dependent variable is the future one-month stock return (RET i,t+1 ) in percentage, and the independent variable of interest is the return of its strongly politically connected firms (LINKRET i,t, common-candidate-numberweighted). Strongly politically connected firms are partitioned into two subsamples based on based on whether they have common political risk exposures (a firm s politically connected firm is considered as strongly connected if the number of common contributed candidates between them is above the median of all its connected firms measured over the past four years). The annual political risk measures for each firm are from the quarterly eight topic-specific political risk scores in Hassan et al. (2017). Control variables include firm s lagged 1-month return (RET i,t ), its corresponding Fama-French 12-Industry return (INDRET i,t ), same-state portfolio lagged 1-month return (GEORET i,t ), as well as the firm s, industry s, and geographic momentum returns (past 11-month cumulative returns measured at [t-11, t-1]). Log of market capitalization (SIZE) and log book-to-market ratio (B/M) measured at prior year-end, as well as total number of contributed candidates over the past four years are also included. Cross-sectional regressions are run every month and the t-statistics reported in the square bracket are adjusted for heteroscedasticity and autocorrelation using Newey and West (1987) up to 12 lags. The p-value of the regression coefficient (LINKRET i,t ) difference in two subsamples is reported in the bracket. The sample is over the period from 2004:01 to 2016:12. Political Risk Sharing (1) Same Primary Political Risk (2) Diff Primary Political Risk Diff (p-val) LINKRET i,t [2.15] [0.77] (0.27) RET i,t [-1.39] [-1.29] INDRET i,t [1.39] [1.83] GEORET i,t [0.58] [1.01] RET i,t 11,t [0.62] [0.25] INDRET i,t 11,t [0.13] [0.50] GEORET i,t 11,t [0.19] [-0.37] SIZE [-1.54] [-2.13] B/M [-0.94] [-0.71] Contribute [0.25] [0.10] No. of Obs 36,558 44,057 44

46 Table 10. Cross-predictability and 2010 Citizen United v. FEC This table reports the effect of 2010 Citizen United v. FEC passage on the return cross-predictability along the interfirm political links from 2008 to 2012 based on panel regressions. The dependent variable is the future one-month stock return (RET i,t+1 ) in percentage. Ban States is a binary variable that equals one if the headquarter of a firm locates in the states that had bans on independent expenditures on state elections and zero otherwise. The independent variable of interest is the return of its strongly politically connected firms (LINKRET i,t, common-candidate-number-weighted) and the interacted variable of Ban States and LINKRET i,t. A firm s politically connected firm is considered as strongly connected if the number of common contributed candidates between them is above the median of all its connected firms measured over the past four years. Control variables include firm s lagged 1-month return ( RET i,t ), its corresponding Fama-French 12-Industry return (INDRET i,t ), same-state portfolio lagged 1-month return (GEORET i,t ), as well as the firm s, industry s, and geographic momentum returns (past 11-month cumulative returns measured at [t-11, t-1]). Log of market capitalization (SIZE) and log book-to-market ratio (B/M) measured at prior year-end, as well as total number of contributed candidates over the past four years are also included. Cross-sectional regressions are run every month and the t-statistics reported in the square bracket are adjusted for heteroscedasticity and autocorrelation using Newey and West (1987) up to 12 lags. The right panel of the table reports differences of the coefficients of the interacted variable (Ban States LINKRET i,t ) before and after the event year, as well as the associated t-statistics. The standard errors are clustered by firm. Dep. Var.= RET t+1 Difference-in-difference t-2 (2008) t-1 (2009) t (2010) t+1 (2011) t+2 (2012) (t+2)- (t) (t+2)- (t-1) (t+2)- (t-2) Ban States LINKRET i,t [-2.30] [-0.55] [-1.89] [0.57] [0.65] [1.88] [0.90] [2.32] Ban States [0.11] [0.23] [0.89] [0.94] [-0.99] LINKRET i,t [2.64] [1.16] [2.02] [-1.03] [-0.12] RET i,t [0.81] [-2.49] [-2.36] [-0.92] [-2.26] INDRET i,t [-0.90] [1.27] [-0.26] [4.56] [-0.05] GEORET i,t [-0.75] [1.01] [-0.30] [-0.70] [-0.26] RET i,t 11,t [3.47] [-7.73] [1.77] [-0.80] [1.67] INDRET i,t 11,t [4.66] [2.43] [1.48] [-4.23] [-1.27] GEORET i,t 11,t [0.95] [0.81] [0.52] [-1.23] [-4.03] SIZE [-0.44] [-3.31] [-3.13] [-0.62] [-2.08] B/M [0.62] [-3.06] [-2.56] [-2.87] [0.69] Contribute [0.97] [-0.05] [1.06] [-0.07] [-1.33] No. of Obs 4,428 4,390 4,383 4,341 4,075 45

47 Table 11. Return Cross-predictability Based on firm PAC Contributions This table examines the return cross-predictability along the inter-firm political links based on common contributions made by firm political action committees (PACs). The dependent variable is the future one-month stock return (Panel A) or Fama-French 12-industry adjusted return (Panel B) in percentage. The independent variable of interest is the return of its politically connected firms (LINKRET i,t, common-candidate-number-weighted). From left to right in each panel, results using samples of all, strongly, and weakly politically connected firms are exhibited, respectively (a firm s politically connected firm is considered as strongly (weakly) connected if the number of common contributed candidates between them is above (below) the median of all its connected firms measured over the past four years). Control variables include firm s lagged 1-month return (RET i,t ), its corresponding Fama-French 12-Industry return ( INDRET i,t ), same-state portfolio lagged 1-month return ( GEORET i,t ), as well as the firm s, industry s, and geographic momentum returns (past 11-month cumulative returns measured at [t-11, t-1]). Log of market capitalization (SIZE) and log book-to-market ratio (B/M) measured at prior year-end, as well as total number of contributed candidates over the past four years are also included. Cross-sectional regressions are run every month and the t-statistics reported in the square bracket are adjusted for heteroscedasticity and autocorrelation using Newey and West (1987) up to 12 lags. The sample is over the period from 1983:01 to 2016:12. Panel A: Raw Return (%) Panel B: Industry-adjusted Return (%) (1) All Connected (2) Strongly Connected (3) Weakly Connected (4) All Connected (5) Strongly Connected (6) Weakly Connected LINKRET i,t [4.88] [4.41] [-1.44] [4.17] [3.86] [-0.85] RET i,t [-3.95] [-3.97] [-3.88] [-3.82] [-3.84] [-3.82] INDRET i,t [2.08] [2.17] [2.61] [-0.03] [0.12] [0.72] GEORET i,t [1.60] [1.63] [2.32] [2.69] [2.68] [3.41] RET i,t 11,t [1.58] [1.57] [1.66] [1.47] [1.47] [1.52] INDRET i,t 11,t [-0.30] [-0.26] [-0.08] [-2.05] [-1.97] [-1.78] GEORET i,t 11,t [0.15] [0.18] [0.37] [0.70] [0.71] [0.97] SIZE [-4.13] [-4.05] [-4.11] [-4.50] [-4.41] [-4.39] B/M [-0.57] [-0.53] [-0.57] [-0.77] [-0.73] [-0.74] Contribute [3.15] [3.02] [2.73] [3.59] [3.38] [3.03] No. of Obs 134, , , , , ,631 46

48 Table month Calendar-time Portfolio and Factor Loadings This table shows 1-month calendar-time portfolio excess and industry-adjusted return (Panel A) as well as its factor loadings (Panel B). At the beginning of every month, firms are ranked into quartile portfolios (ascending order) based on last-month returns of their strongly politically connected firms last month (LINKRET i,t, common-candidatenumber-weighted). A firm s politically connected firm is considered as strongly connected if the number of common contributed candidates between them is above the median of all its connected firms measured over the past four years. L/S Spread of the zero-cost portfolio that buys firms with LINKRET at the top 25% and sells firms with LINKRET at the bottom 25% measured in the previous month. The equal-weighted portfolios are rebalanced every month. The equal-weighted portfolios are rebalanced every month. Alpha is the intercept on the time-series regressions of monthly long-short portfolio spread on traditional risk factors (i.e., Fama and French (1993) three factors, Carhart (1997) momentum factor, and Pástor and Stambaugh (2003) liquidity factor). All the coefficients are denoted in percentage. The t-statistics reported in the square bracket are adjusted for heteroscedasticity and autocorrelation using Newey and West (1987) up to 12 lags. The firm executives (PAC) sample is over the period from 2004:01 to 2016:12 (from 1983:01 to 2016:12). Firm Executive Firm PAC Firm Executive Firm PAC Panel A: Portfolio Return L/S 1-Factor 3-Factor 4-Factor 5-Factor 1 (L) (H) Spread Alpha Alpha Alpha Alpha (1) Excess ret [-0.86] [-1.68] [0.24] [2.00] [2.30] [2.89] [2.89] [2.74] [2.77] (2) Ind-adj. ret [-0.90] [-1.49] [0.00] [2.09] [2.65] [3.16] [3.24] [3.29] [3.39] (3) Excess ret [-3.39] [-2.44] [3.40] [2.59] [3.41] [3.71] [4.01] [3.71] [2.64] (4) Ind-adj. ret [-2.82] [-2.37] [2.68] [2.74] [3.35] [3.86] [4.06] [3.83] [3.28] Panel B: Factor Loading Intercept MKT SMB HML UMD LIQ (1) Excess ret [2.77] [-0.89] [-1.30] [0.51] [0.65] [0.15] (2) Ind-adj. ret [3.39] [-0.73] [-1.17] [0.88] [0.91] [-0.48] (3) Excess ret [2.64] [-0.09] [-0.50] [-0.86] [1.93] [1.62] (4) Ind-adj. ret [2.31] [0.84] [0.76] [-0.67] [2.62] [2.42] 47

49 Table month Calendar-time Portfolio and Factor Loadings, Election Years This table shows effect of election cycle on the politically connected portfolio alphas. At the beginning of every month, firms are ranked into quartile portfolios (ascending order) based on last-month returns of their strongly politically connected firms last month (LINKRET i,t, common-candidate-number-weighted). A firm s politically connected firm is considered as strongly connected if the number of common contributed candidates between them is above the median of all its connected firms measured over the past four years. L/S Spread of the zero-cost portfolio that buys firms with LINKRET at the top 25% and sells firms with LINKRET at the bottom 25% measured in the previous month. The equal-weighted portfolios are rebalanced every month. Alpha is the intercept on the time-series regressions of monthly long-short portfolio spread on traditional risk factors (i.e., Fama and French (1993) three factors, Carhart (1997) momentum factor, and Pástor and Stambaugh (2003) liquidity factor). Election-year dummy is added to the time-series regression which equals one in even-numbered years, and zero in odd years. The t-statistics reported in the square bracket are adjusted for heteroscedasticity and autocorrelation using Newey and West (1987) up to 12 lags. The sample is over the period from 2004:01 to 2016:12. Intercept Election-year Dummy MKT SMB HML MOM LIQ (1) 0.19 [2.30] (2) [2.50] [-1.52] (3) [2.91] [-1.77] [-1.01] [-1.30] [0.71] [0.78] [0.16] 48

50 Table 14. Robustness: Cross-predictability of Operating Performance This table reports the Fama and MacBeth (1973) regressions of the S&P500 stock firm s quarterly operating performance measures on lagged one-quarter of its strongly politically connected firms operating performance measures (OPM, common-candidate-number-weighted). The operating performance measures are return on asset (ROA), gross profitability (PROFIT) from Novy-Marx (2013), and seasonal-adjusted earnings (EPS). A firm s politically connected firm is considered as strongly connected if the number of common contributed candidates between them is above the median of all its connected firms measured over the past four years. Control variables include the firm s lagged 1-quarter operating performance measures, its corresponding Fama-French 12-Industry and same-state firms lagged 1-quarter operating performance measures (equal-weighted), firm s lagged 1-quarter returns (RET i,t ), its corresponding industry and same-state lagged 1-quarter portfolio returns, as well as the firm s, industry s, and geographic momentum returns (past 7-quarter cumulative returns measured at [t-7, t-1], skipping the most recent quarter), and log of market capitalization (SIZE) and log book-to-market ratio (B/M) measured at prior year-end. All coefficients are multiplied with 100. Cross-sectional regressions are run every quarter and the t-statistics reported in the square bracket are adjusted for heteroscedasticity and autocorrelation using Newey and West (1987) up to 4 lags. The panel regression includes the time fixed effects and the standard errors are clustered by firm. The sample is over the period from 2004:01 to 2016:12. Operating Measure (OP)= ROA t+1 PROFIT t+1 EPS t+1 (1) FMB (2) Panel (3) FMB (4) Panel (5) FMB (6) Panel LINKFIRM_OP i,t [3.03] [2.27] [1.99] [2.63] [0.06] [-2.31] FIRM_OP i,t [56.79] [25.98] [136.20] [81.46] [5.93] [10.29] IND_OP i,t [1.21] [0.43] [4.38] [3.80] [1.35] [2.80] GEO_OP i,t [1.99] [0.75] [0.32] [-2.94] [1.18] [1.29] RET i,t [11.72] [7.12] [5.11] [3.73] [5.03] [6.01] INDRET i,t [-0.08] [2.81] [0.59] [2.24] [-0.28] [2.18] GEORET i,t [-0.48] [0.33] [-0.71] [0.02] [0.44] [0.84] MOM i,t 7,t [5.70] [4.78] [0.77] [-0.19] [2.51] [2.41] INDMOM i,t 7,t [0.99] [3.01] [1.46] [1.58] [0.38] [0.93] GEOMOM i,t 7,t [1.73] [2.00] [0.67] [0.56] [-1.08] [-0.02] SIZE [2.59] [2.03] [-3.66] [-3.60] [-0.01] [-1.10] B/M [-12.48] [-6.26] [-12.13] [-4.85] [-0.35] [-1.16] No. of Obs 15,231 15,231 15,231 15,231 15,231 15,231 49

51 Table 15. Robustness: Alternative Measurement Horizon of Political Network This table examines the baseline results shown in Table III using an alternative political connection measurement horizon. The dependent variable is the future one-month stock return (RET i,t+1 ) in percentage and the independent variable of interest is the return of its politically connected firms (LINKRET i,t, common-candidate-number-weighted). From left to right in each panel, results using samples of all, strongly, and weakly politically connected firms are exhibited, respectively (a firm s politically connected firm is considered as strongly (weakly) connected if the number of common contributed candidates between them is above (below) the median of all its connected firms measured over the past two-year (Panel A) or five-year (Panel B)). Control variables include firm s lagged 1-month return (RET i,t ), its corresponding Fama-French 12-Industry return (INDRET i,t ), same-state portfolio lagged 1-month return (GEORET t ), as well as the firm s, industry s, and geographic momentum returns (past 11-month cumulative returns measured at [t-11, t-1]). Log of market capitalization (SIZE) and log book-to-market ratio (B/M) measured at prior year-end, as well as total number of contributed candidates over the past two (five) years are also included. Crosssectional regressions are run every month and the t-statistics reported in the square bracket are adjusted for heteroscedasticity and autocorrelation using Newey and West (1987) up to 12 lags. The sample for Panel A (B) is over the period from 2002:01 (2005:01) to 2016:12. Panel A: 2-Year Measurement Horizon Panel B: 5-Year Measurement Horizon (1) All Connected (2) Strongly Connected (3) Weakly Connected (4) All Connected (5) Strongly Connected (6) Weakly Connected LINKRET i,t [1.75] [1.95] [-0.47] [2.25] [4.01] [-0.83] RET i,t [-2.05] [-2.04] [-2.10] [-1.12] [-1.15] [-1.50] INDRET i,t [1.46] [1.43] [2.86] [1.26] [1.28] [2.00] GEORET i,t [0.76] [0.89] [-0.85] [0.55] [0.62] [-0.44] RET i,t 11,t [0.59] [0.59] [0.85] [0.59] [0.59] [0.75] INDRET i,t 11,t [1.08] [1.08] [0.12] [0.50] [0.54] [0.11] GEORET i,t 11,t [-0.41] [-0.30] [-0.45] [-0.54] [-0.56] [-0.63] SIZE [-2.93] [-2.88] [-2.32] [-2.16] [-2.12] [-1.47] B/M [-0.46] [-0.43] [-0.19] [-0.63] [-0.60] [-0.83] Contribute [0.25] [0.27] [0.19] [-0.79] [-0.65] [-0.85] No. of Obs 52,504 52,504 27,024 52,644 52,644 37,384 50

52 Table 16. Robustness: Absolute Cutoff of Political Connection Strength This table examines the baseline results shown in Table III using absolute cutoff to categorize strongly/weakly politically connected firms. The dependent variable is the future one-month stock return (RET i,t+1 ) in percentage and the independent variable of interest is the return of its politically connected firms (LINKRET i,t, common-candidatenumber-weighted). From left to right panel, results using samples of strongly and weakly politically connected firms are exhibited, respectively (a firm s politically connected firm is considered as strongly (weakly) connected if the number of common contributed candidates between the two firms is ranked in the (non-)top 3, 5 or 10 among all its connected firms measured over the past four years). Control variables include firm s lagged 1-month return (RET i,t ), its corresponding Fama-French 12-Industry return ( INDRET i,t ), same-state portfolio lagged 1-month return (GEORET i,t ), as well as the firm s, industry s, and geographic momentum returns (past 11-month cumulative returns measured at [t-11, t-1]). Log of market capitalization (SIZE) and log book-to-market ratio (B/M) measured at prior year-end, as well as total number of contributed candidates over the past two (five) years are also included. Crosssectional regressions are run every month and the t-statistics reported in the square bracket are adjusted for heteroscedasticity and autocorrelation using Newey and West (1987) up to 12 lags. The sample is over the period from 2004:01 to 2016:12. Panel A: Strongly Connected Cutoff Panel B: Weakly Connected Cutoff (1) Top 3 (2) Top 5 (3) Top 10 (4) N-Top 3 (5) N-Top 5 (6) N-Top 10 LINKRET i,t [2.75] [2.77] [2.12] [1.54] [1.12] [-1.65] RET i,t [-1.61] [-1.62] [-1.63] [-1.69] [-1.34] [-1.17] INDRET i,t [1.38] [1.42] [1.43] [1.53] [1.46] [1.63] GEORET i,t [1.19] [1.28] [1.23] [0.64] [0.95] [0.71] RET i,t 11,t [0.43] [0.44] [0.44] [0.64] [0.81] [1.04] INDRET i,t 11,t [0.55] [0.55] [0.52] [0.41] [0.27] [0.31] GEORET i,t 11,t [-1.03] [-0.98] [-0.94] [-0.84] [-1.17] [-1.25] SIZE [-2.10] [-2.10] [-2.08] [-2.14] [-2.13] [-1.87] B/M [-0.69] [-0.70] [-0.70] [-0.75] [-0.74] [-0.94] Contribute [-0.78] [-0.75] [-0.66] [-0.02] [0.05] [-0.03] No. of Obs 54,770 54,770 54,770 50,410 48,273 43,957 51

53 Table 17. Robustness: Controlling for Political Activeness and CEO Tenure This table examines the baseline results shown in Table III using an alternative political connection measure which controls for firm s political activeness or CEO tenure. Panel A calculates the measure that equals the baseline measure (the number of common contributed candidates) divided by the minimum of each of the two connected firms total number of contributed candidates measured over the past four-year (Eq. (5)). Panel B construct political network based on campaign contributions only from CEOs that in office over the formation period. The dependent variable is the future one-month stock return in percentage, and the independent variable of interest is the return of its politically connected firms (LINKRET i,t, common-candidate-number-weighted). From left to right in each panel, results using samples of strongly and weakly politically connected firms are exhibited, respectively (a firm s politically connected firm is considered as strongly (weakly) connected if the number of common contributed candidates between them is above (below) the median alternative political connection measure among all its connected firms measured over the past four years). Control variables include firm s lagged 1-month return (RET i,t ), its corresponding Fama-French 12- Industry return (INDRET i,t ), same-state portfolio lagged 1-month return (GEORET i,t ), as well as the firm s, industry s, and geographic momentum returns (past 11-month cumulative returns measured at [t-11, t-1]). Log of market capitalization (SIZE) and log book-to-market ratio (B/M) measured at prior year-end, as well as total number of contributed candidates over the past four years are also included. Cross-sectional regressions are run every month and the time-series standard errors reported in the square bracket are adjusted for heteroscedasticity and autocorrelation using Newey and West (1987) up to 12 lags. The sample is over the period from 2004:01 to 2016:12. Panel A: Controlling for Political Activeness Panel B: Controlling for CEO Tenure (1) Strongly Connected (2) Weakly Connected (1) Strongly Connected (2) Weakly Connected LINKRET i,t [1.85] [0.91] [3.08] [-0.59] RET i,t [-1.57] [-1.72] [-1.35] [-1.27] INDRET i,t [1.42] [1.89] [1.49] [2.50] GEORET i,t [1.08] [0.65] [1.12] [0.11] RET i,t 11,t [0.45] [0.73] [0.36] [0.78] INDRET i,t 11,t [0.55] [0.25] [0.56] [-0.06] GEORET i,t 11,t [-1.00] [-1.26] [-1.11] [-1.11] SIZE [-2.08] [-1.86] [-2.38] [-2.03] B/M [-0.67] [-0.71] [-0.69] [-0.63] Contribute [-0.56] [-0.42] [-0.48] [-0.14] No. of Obs 54,770 41,647 53,108 37,782 52

54 Table A.1. Placebo Test, Economic Linkage as a Potential Confounding Factor This table examines whether the return cross-predictability based on inter-firm political links is driven by the economic linkage. The dependent variable is the future one-month stock return (RET i,t+1 ). The independent variable of interest is the return of its politically connected firms (LINKRET i,t, common-candidate-number-weighted). From left to right in each panel, its weakly politically connected firms are partitioned into subsamples based on whether they are economically linked (a firm s politically connected firm is considered as weakly connected if the number of common contributed candidates between them is below the median of all its connected firms measured over the past four years). Economically related industries are defined as top 1, 3 and 10 supplier and customer industries calculated using the annual Input-Output flows from the Bureau of Economic Analysis surveys following Ahern and Harford (2014). Control variables include firm s lagged 1-month return (RET i,t ), its corresponding Fama-French 12-Industry return ( INDRET i,t ), same-state portfolio lagged 1-month return ( GEORET i,t ), as well as the firm s, industry s, and geographic momentum returns (past 11-month cumulative returns measured at [t-11, t-1]). Log of market capitalization (SIZE) and log book-to-market ratio (B/M) measured at prior year-end, as well as total number of contributed candidates over the past four years are also included. Cross-sectional regressions are run every month and the t-statistics reported in the square bracket are adjusted for heteroscedasticity and autocorrelation using Newey and West (1987) up to 12 lags. The p-value of the regression coefficient (LINKRET i,t ) difference in two subsamples is reported in the bracket. The sample is over the period from 2004:01 to 2016:12. Placebo Test: Economically-linked Industry Classifications (Weakly Connected Sample) (1) Top1 (2) N-top1 Diff (p-val) (1) Top3 (2) N-top3 Diff (p-val) (1) Top10 (2) N-top10 Diff (p-val) LINKRET i,t [0.27] [-0.32] (0.65) [0.11] [-0.41] (0.64) [-0.19] [-1.27] (0.35) RET i,t [-0.78] [-0.76] [-0.87] [-0.74] [-0.83] [-0.86] INDRET i,t [1.80] [1.93] [2.38] [1.95] [2.09] [1.91] GEORET i,t [0.12] [-1.14] [-1.11] [-1.13] [-0.93] [-0.80] RET i,t 11,t [0.39] [0.52] [0.56] [0.48] [0.54] [0.51] INDRET i,t 11,t [-0.21] [0.25] [0.19] [0.31] [0.13] [0.31] GEORET i,t 11,t [-0.20] [-0.76] [-0.52] [-0.86] [-0.72] [-1.10] SIZE [-0.26] [-1.42] [-0.58] [-1.44] [-1.30] [-1.49] B/M [-0.74] [-1.02] [-0.95] [-0.99] [-1.04] [-0.89] Contribute [-0.88] [-0.49] [-0.46] [-0.38] [-0.13] [-0.38] No. of Obs 17,879 26,719 23,753 26,683 26,514 26,389 53

55 Table A.2. Summary Statistics of the Firm PAC Political Contribution Data This table reports the summary statistics of the S&P500 firm PAC political campaign contribution variables as of October of each year (political year-end). Panel A reports the summary statistics of number of firms donating in political campaigns, number of candidates a firm donates to, and number of firms having political networks (connecting to at least one other firm through common contributed political candidates), measured over the past four years. Panel B illustrates the summary statistics of political network size (number of politically connected firms to a firm), proportion of connected firms that are in the same industry and whose headquarters are in the same state, and average number of common contributed political candidates per firm, measured over the past four years. At the bottom of Panel B, summary statistics of strongly and weakly politically connected subsamples are also displayed, respectively (a firm s politically connected firm is considered as strongly (weakly) connected if the number of common contributed candidates between them are above (below) the median of all its connected firms measured over the past four years). The sample is over the period from 1983:01 to 2016:12. Firm Executive Contribution Sample ( ) Contribution Variable Mean SD Min 25 th Pctl Median 75 th Pctl Max Panel A: Overall Sample Statistics Number of firms contributing to political campaign over years Number of contributed candidates per firm over years Number of firms with political networks (with other firms) over years Panel B: Time-Series Average of Cross-sectional Summary Statistics (Full Sample) Number of connected firms per firm % of connected counterparts in the same industry per firm % of connected counterparts in the same state per firm Average number of common candidates contributed per firm Strongly Connected Subsample Number of connected counterparts per firm % of connected counterparts in the same industry per firm % of connected counterparts in the same state per firm Average number of common candidates contributed per firm Weakly Connected Subsample Number of connected counterparts per firm % of connected counterparts in the same industry per firm % of connected counterparts in the same state per firm Average number of common candidates contributed per firm

56 Table A.3. Portfolio Alphas with Different Look-back and Holding Periods At the beginning of every month, firms are ranked into quartile portfolios (ascending order) based on last-month returns of their strongly politically connected firms last month (LINKRET i,t, common-candidate-number-weighted). A firm s politically connected firm is considered as strongly connected if the number of common contributed candidates between them is above the median of all its connected firms measured over the past four years. L/S Spread of the zero-cost portfolio that buys firms with LINKRET at the top 25% and sells firms with LINKRET at the bottom 25% measured in the previous 1, 3, 6, 9, or 12 month(s). Alpha is the intercept on the time-series regressions of monthly long-short portfolio spread on traditional risk factors (i.e., Fama and French (1993) three factors, Carhart (1997) momentum factor, and Pástor and Stambaugh (2003) liquidity factor). All the coefficients are denoted in percentage. The t-statistics reported in the square bracket are adjusted for heteroscedasticity and autocorrelation using Newey and West (1987) up to 12 lags. The firm executives (PAC) sample is over the period from 2004:01 to 2016:12 (from 1983:01 to 2016:12). Holding Period (Month) Firm Executive Sample (2004:01 to 2016:12) Firm PAC Sample (1983:01 to 2016:12) Look-back Period (Month) [2.77] [1.97] [0.58] [1.07] [2.15] [2.64] [2.47] [1.04] [1.47] [2.06] [1.58] [0.58] [-0.62] [1.33] [1.47] [1.89] [0.69] [0.22] [1.08] [1.13] [0.83] [0.28] [0.97] [1.38] [0.79] [0.67] [0.12] [0.70] [0.93] [0.48] [2.00] [1.93] [1.58] [1.41] [1.06] [1.73] [1.34] [1.22] [0.84] [0.12] [1.90] [1.39] [1.02] [1.13] [0.87] [1.57] [1.06] [0.58] [0.25] [-0.14] 55

57 Table A.4. Political, Geographic, and Economic Linkage Return Cross-predictability This table shows the effect of politically-connected portfolio returns on geographically linked (Panel A) and economically linked portfolio returns. At the beginning of every month, firms are ranked into quartile portfolios (ascending order) based on returns of their strongly politically connected firms, return of their geographically-linked firms (firms whose headquarters are in the same state but not in the same industry), and economically-linked firms (firms that are operated in the top 10 supplier/customer industries), respectively. The equal-weighted portfolios are rebalanced every month. The zero-cost portfolio buys firms with the top 25% strongly politically-linked (geographically-linked or economically-linked) firm returns and sells firms with the bottom 25% strongly politicallylinked (geographically-linked or economically-linked) returns in the past month. Alpha is the intercept on the timeseries regressions of monthly long-short geographically-linked (economic-linked portfolio returns) portfolio spread on traditional risk factors (i.e., Fama and French (1993) three factors, Carhart (1997) momentum factor, and Pástor and Stambaugh (2003) liquidity factor) and the politically connected portfolio returns. All the coefficients are denoted in percentage. The t-statistics reported in the square bracket are adjusted for heteroscedasticity and autocorrelation using Newey and West (1987) up to 12 lags. The sample is over the period from 2004:01 to 2016:12. Intercept H/L Politically connected Port Ret MKT SMB HML MOM LIQ Panel A: Geographically-linked Portfolio Returns (2004:01 to 2016:12) (1) 0.20 [1.80] (2) [1.67] [1.24] (3) [1.65] [0.93] [-1.56] [-0.33] [0.30] [-0.72] (4) [1.51] [1.14] [0.97] [-1.52] [-0.33] [0.23] [-0.73] Panel B: Economically-linked Portfolio Returns (2004:01 to 2016:12) (1) 0.31 [2.06] (2) [1.66] [1.67] (3) [1.49] [1.00] [-0.02] [1.39] [2.34] [0.80] (4) [1.13] [1.87] [1.24] [0.06] [1.44] [2.27] [0.81] 56

58 Table A.5. Different Weighting Schemes This table reports the impact of weight schemes on the return cross-predictability along the inter-firm political links. The dependent variable is the future one-month stock return (RET i,t+1 ). The independent variable of interest is the return of its politically connected firms (LINKRET i,t ) using common-candidate-number-weighted (Panel A) or equalweighted (Panel B). From left to right in each panel, results using samples of all and strongly politically connected firms are exhibited, respectively (a firm s politically connected firm is considered as strongly connected if the number of common contributed candidates between them is above the median of all its connected firms measured over the past four years). Control variables include firm s lagged 1-month return (RET i,t ), its corresponding Fama-French 12- Industry return (INDRET i,t ), same-state portfolio lagged 1-month return (GEORET i,t ), as well as the firm s, industry s, and geographic momentum returns (past 11-month cumulative returns measured at [t-11, t-1]). Log of market capitalization (SIZE) and log book-to-market ratio (B/M) measured at prior year-end, as well as total number of contributed candidates over the past four years are also included. Cross-sectional regressions are run every month and the t-statistics reported in the square bracket are adjusted for heteroscedasticity and autocorrelation using Newey and West (1987) up to 12 lags. The sample is over the period from 2004:01 to 2016:12. Panel A: Number of Common Candidates Weighted Panel B: Equal Weighted Firm Executive Sample Corporate PAC Sample Firm Executive Sample Corporate PAC Sample (1) All Connected (2) Strong Connected (3) All Connected (4) Strong Connected (1) All Connected (2) Strong Connected (3) All Connected (4) Strong Connected LINKRET i,t [2.39] [3.30] [4.88] [4.41] [2.12] [3.05] [0.93] [3.06] RET i,t [-1.55] [-1.64] [-3.95] [-3.97] [-1.54] [-1.63] [-3.91] [-3.92] INDRET i,t [1.40] [1.42] [2.08] [2.17] [1.39] [1.43] [2.56] [2.38] GEORET i,t [1.12] [1.16] [1.60] [1.63] [1.11] [1.16] [2.09] [1.81] RET i,t 11,t [0.47] [0.45] [1.58] [1.57] [0.46] [0.45] [1.68] [1.58] INDRET i,t 11,t [0.49] [0.55] [-0.30] [-0.26] [0.50] [0.55] [-0.18] [-0.19] GEORET i,t 11,t [-0.93] [-0.96] [0.15] [0.18] [-0.93] [-0.95] [0.20] [0.12] SIZE [-2.12] [-2.08] [-4.13] [-4.05] [-2.14] [-2.09] [-4.13] [-4.14] B/M [-0.69] [-0.69] [-0.57] [-0.53] [-0.68] [-0.68] [-0.55] [-0.59] Contribute [-0.67] [-0.56] [3.15] [3.02] [-0.66] [-0.55] [3.56] [3.27] No. of Obs 54,770 54, , ,170 54,770 54, , ,170 57

59 Panel A: Firm Executive Contribution Sample Panel B: Firm PAC Contribution Sample Figure 1. This figure shows the time-series of the long/short portfolio returns. At the beginning of every month, firms are ranked into quartile portfolios (ascending order) based on last-month returns of their strongly politically connected firms last month (LINKRET i,t, common-candidate-number-weighted) measured based on common campaign contributions made by firm executives (Panel A) or firm political action committees (Panel B) over the past four years (a firm s politically connected firm is considered as strongly connected if the number of common contributed candidates between them is above the median of all its connected firms measured over the past four years). The equal-weighted portfolios are rebalanced every month. The zero-cost portfolio that buys firms with LINKRET at the top 25% and sells firms with LINKRET at the bottom 25% measured in the previous month. The firm executives (PAC) sample is over the period from 2004:01 to 2016:12 (from 1983:01 to 2016:12). 58