Political Connections, Incentives and Innovation: Evidence from Contract-Level Data * Jonathan Brogaard, Matthew Denes and Ran Duchin April 2015 Abstract This paper studies the relation between corporate political connections and the allocation, design, and outcomes of government contracts. Using hand-collected data on Federal procurement contracts, we find that connected firms are 10% more likely to win a contract. Connected firms receive larger contracts, with longer durations and weaker incentive structures. Politically-connected firms are also more likely to increase contract amounts and extend deadlines through contract renegotiations. While government contracts enhance firm-level innovation on average, political connections and weak contractual incentives are associated with less innovation, as measured by patents and patent citations. Overall, we show that connections between firms and politicians are associated with distortions in contract allocation and design. * Contact: Jonathan Brogaard, Foster School of Business, University of Washington, e-mail: brogaard@uw.edu; Matthew Denes, Foster School of Business, University of Washington, e-mail: denes@uw.edu; Ran Duchin, Foster School of Business, University of Washington, e-mail: duchin@uw.edu. We thank seminar participants at the SIFR Conference on Innovation and Entrepreneurship for helpful comments.
I. Introduction The financial crisis of 2008-2009 resulted in a wave of government interventions around the world. Governments initiated tighter regulation, monetary policy interventions, and capital assistance to the private sector. Economists have debated the influence of government involvement on economic activity and the private sector since at least Keynes (1936). At the forefront of this debate are the questions of whether and when government involvement creates value. On the one hand, government involvement can enhance private sector activity through efficient allocation of capital. On the other hand, government involvement can crowd out private sector activity and introduce distortive political favoritism. In this paper, we provide evidence in this direction by studying the allocation and design of government procurement contracts. Our main innovation is to use contract-level data to study detailed contractual agreements awarded by the Federal government to the private sector. Our empirical analysis seeks to answer three questions. First, how does corporate political activism affect the allocation of government contracts? Second, what are the contractual features of awarded contracts, including their incentive structure, duration, and renegotiation terms? Third, do political activism and contract design impact private sector output? So far, empirical research has focused mostly on firms access to capital. Prior work finds that politically connected firms have better access to capital (Chiu and Joh (2004), Cull and Xu (2005), Johnson and Mitton (2003), and Khwaja and Mian (2005)), are more likely to be bailed out (Faccio, Masulis, and McConnell (2006) and Duchin and Sosyura (2012)), and consequently have higher values (Roberts (1990), Fisman (2001), and Faccio (2006)). Our contribution lies in identifying the direct contractual mechanisms that govern the efficacy of both the allocation of government capital and its subsequent use. 1
Our focus on procurement contracts is motivated by several factors. First, these contracts represent substantial government spending ($203.1 billion a year, which represent 19.3% of the total annual government expenditure). Second, this setting allows us to observe key information about the terms of each contract, including its incentive structure, duration, and subsequent renegotiation. Third, these contracts can be directly linked to individual firms over well-identified time intervals, generating both within-firm and across-firm variation in government spending. We collect detailed data on procurement contracts that cover over $1.6 trillion of government contracts between 2003 and 2010, and match each contract to the recipient firm. We identify 1,253 firms that received a total of $774 billion in government contracts during our sample period. After hand-matching these data to Compustat and patent data, our sample contains 299,789 contracts awarded to these firms. On average, our data cover 46.8% of total federal procurement spending in a year. The average firm receives $147.6 million in a given year, with an average duration of approximately 8 months. We measure corporate political activism using firms campaign contributions to politicians. This measure has two important advantages. First, it allows us to compare firms that contributed to a winning politician with firms that contributed to a losing politician, thus holding constant the firm s political activism through campaign contributions. Second, it allows us to separate between the formation of political connections and contract allocation, thus mitigating simultaneity and reverse causality concerns. Specifically, we study how contributions to a political campaign affect contract allocations after the campaign is over and the politician has either won or lost the elections. We start by showing that, consistent with prior evidence, politically-connected firms are more likely to receive government procurement contracts. In particular, our estimates suggest that firms contributing to winning politicians are 10.4% more likely to receive a contract. These effects 2
are highly statistically significant and hold after controlling for firm-level characteristics, as well as unobservable time and industry effects. It is possible that a firm s political connections are correlated with some unaccounted for or unobserved characteristics that increase a firm s likelihood of receiving government contracts but are unrelated to political influence. We address this possibility in two ways. First, we construct a placebo measure of campaign contributions to losing politicians. We find that firms contributing to a losing politician are not more likely to receive a contract. Second, we exploit cross-sectional variation in the influence of politicians. Firms contributing to House or Senate committee chairmen of the Committee on Appropriations or the Committee on the Budget, which play a key role in the allocation of procurement contracts, further increases the likelihood of receiving a government contract by 7.9%. The allocation of contracts to politically-connected firms is consistent with several hypotheses. One hypothesis is that firms with political connections receive favorable treatment in the allocation of government contracts. This view would be consistent with theories of the politics of government ownership and investment (e.g., Shleifer and Vishny, 1994), which suggest that federal capital is used to accommodate private interests of politicians, such as securing electorate votes and funding election campaigns. This hypothesis predicts that the value of contracts awarded to connected firms is likely to trail that of unconnected peers, as predicted in Stigler (1971), Shleifer and Vishny (1994), and Banerjee (1997). An alternative hypothesis posits that political connections can mitigate the information asymmetry between government officials and the firm and result in more informed federal investment decisions (Downs (1957)). This hypothesis predicts that contracts awarded to politically connected firms are likely to outperform those awarded to unconnected firms. 3
A third possibility is that firms political connections do not materially affect the efficacy of government contracts. Under this view, the terms of procurement contracts mitigate any concerns about the inefficient allocation or use of Federal capital. For example, these contracts may be structured in a standardized manner that monitors and incentives the firm through structured payoff schedules, pay-for-performance, and penalties for low quality or untimely execution. To investigate these possibilities, we provide novel evidence on the terms of procurement contracts and their association with political influence. First, we find that connected firms receive contracts with weaker incentive structures. Firms that contributed to a winning politician receive about 3.9% fewer incentive contracts. Second, we show that connected firms receive contracts with larger awards and later completion dates. In particular, politically connected firms receive contracts that are 15.6% larger at signing, on average. Moreover, they are 11.5% more likely to receive a longer deadline. Third, we find that, when contracts are renegotiated, politically connected firms are 10.9% to 11.7% more likely to receive increases in the contract award and 10.3% to 11.8% more likely to receive extensions in contract completion dates. Taken together, our findings are more consistent with the political favoritism view. They suggest that concerns about political favoritism are not mitigated, and in fact, exacerbated by the details of the contractual agreements that accompany government investment. Our final set of analyses investigates the ex-post outcomes of procurement contracts. Our research question is twofold. First, we ask whether government contracts spur private sector innovation. Second, we ask whether conditional on being awarded a contract, political connections and contract terms affect innovation. Our focus on innovation is motivated by the stated goal of procurement contracts and government spending to spur innovation (Bayh-Dole Act, 1980). We measure innovation using the adjusted number of patents and patent citations. These measures are motivated by Griliches (1990), who finds that patents are a better measure of innovation than 4
research and development expenditures, and by Hall, Jaffe, and Trajtenberg (2006), who show that patent citations are a measure of the value of innovation. We find that, on average, government spending fosters private sector innovation. Our estimates indicate that receiving a procurement contract is associated with an increase in the scale and novelty of innovation, as measured by the adjusted number of patents and patent citations, respectively. Moreover, based on the attributes of firms awarded Federal contracts, government investment may help alleviate financial constraints. In the cross-sectional analysis, we find that firms with lower cash balances, fewer tangible assets, and higher growth options are more likely to receive government contracts. For example, a decrease of one standard deviation in cash reserves and tangible assets is associated with an increase of 19.4% and 20.2% in the likelihood of receiving a government contract, respectively. However, we find that government investments in politically connected firms lead to less innovation than investments in unconnected firms. Specifically, firms contributing to winning politicians are less likely to innovate and these patents are less novel compared to unconnected firms that also received procurement contracts. We also investigate the link between innovation and contract design. We find that incentive contracts increase innovation. Winning a procurement contract that includes incentives increases both the number of patents and their novelty, as measured by citations, self citations and originality. Since political connections are associated with fewer incentive contracts, our evidence puts forth contract design as one potential channel through which political connections distort the allocation of government capital. Overall, the results in this article document a strong relation between a firm s political connections and the allocation, design, and outcomes of government contracts. Contracts won by politically connected firms are consistent with agency-type inefficiencies from political influence 5
predicted in Shleifer and Vishny (1994) and are inconsistent with the efficiency-improving role of political connections. Our paper contributes to prior research on government spending. Lerner (1999) shows that government investment in small firms results in superior growth. Conversely, Cohen, Coval, and Malloy (2011) use exogenous variation in Senate and House chairmanships and find that shocks to government spending tend to lower investment and employment. Our results shed new light on these findings. They suggest that while government spending fosters firms innovation on average, the terms of government contracts awarded to politically connected firms lead to less innovation. Our paper is also related to the growing literature that studies firm-level innovation. Our focus on innovation is driven by recent studies, such as, Kogan et al. (2012), which shows that innovation is an important source of long-term economic growth. The rest of the paper is organized as follows. Section II describes the data. Section III provides the empirical evidence on the allocation and design of procurement contracts. Section IV studies the link between political connections, contracts, and innovation. Section V concludes. II. Data The U.S. government commonly is a customer for firms. Contract-level data allows us to study the relation between political connections, contracting and innovation. This section details our novel dataset of contracts, which we hand-match to political contributions, patents and financial variables. A. Contracting with the U.S. Government The U.S. government often enters into contracts with firms and individuals. A contract is initiated when an agency of the federal government determines that it requires a good or service. A 6
contracting officer for the agency provides information about the contract on the Federal Business Opportunities website through a Request For Proposal. Firms have the opportunity to review the proposal and submit offers for the contract, which are then evaluated by agency employees. Contracting with the government has been increasingly unified, particularly with the creation of the Federal Acquisition Regulation in 1984. These regulations provide guidelines for many aspects of contracts, including bidding, competition and management (Feldman and Keyes, 2011). The Federal Procurement Data System (FPDS) tracks procurement contracts of the federal government of the United States. This comprehensive system provides detailed information on nearly all federal contracts from about 65 different branches, departments and agencies of the federal government. The U.S. government began providing detailed data on procurement contracts beginning in 1978, though reporting is often incomplete prior to 2000 (Liebman and Mahoney, 2013). The Federal Funding Accountability and Transparency Act of 2006 led to the creation of the USAspending.gov website, which provides data from the FPDS starting in 2000. Specifically, the system reports comprehensive details on any contract with a transaction value of at least $2,500 ($25,000 prior to 2004). There are mainly two types of procurement contracts: fixed price and cost plus. A fixed-price contract is a contract whose price does not vary by the cost of materials or labor. A cost-plus contract is a contract where the price is determined by the costs incurred by the vendor plus a certain amount. While the FPDS includes data on classified contracts, it does not contain records on the U.S. Postal Service and certain legislative and judicial branches. Firms often contract with the federal government. Table 1 summarizes the contracts observed in our sample, renegotiation of these contracts and the industrial composition of recipient firms. Insert Table 1 About Here 7
We restrict our sample to those contracts whose total award is at least $100,000. Panel A explores contract-level details at initiation. The sample comprises 299,789 contracts to Compustat firms from 2003-2010. The average initial award of a contract is $1.1 million, with a mean total award of $2.6 million throughout its duration. A contract typically lasts for over seven months and there is substantial variation in the length of a contract. Contracts with the government can vary in their type and about 11.7% of contracts include incentives. These contracts use monetary awards to incentivize firms to complete contracts timely. Appendix A details contracts with these features. After initiation, a firm can renegotiate a contract. Panel B of Table 1 details when and how renegotiation occurs. We observe changes to 49% of contracts and focus on modifications in the award and length of a contract. Just under 1 million contract level changes occur from 2003-2010 and the average contract has 3 modifications. About 41.4% of award changes are increases and the mean increase is $1.5 million. Only 13.6% of contracts receive a decrease and the average reduction is $150,000. Lastly, extensions of contract length are observed 38.7% of the time and, conditional on receiving a change to its completion date, are 2.2 years on average. The government contracts with many industries and Panel C highlights the count and size of contracts by industry. Business equipment, wholesale and manufacturing receive the most contracts both in terms of the number of contracts and their total value, summarized in the last column of the table. Business equipment collected $253.1 billion of government spending through 97,986 contracts, while manufacturing won $297.9 billion in 60,118 contracts. Overall, our dataset allows us to observe $774 billion in contracts awarded to 1,253 firms. 8
B. Political Connections Each election cycle provides firms with the opportunity to contribute to politicians. Firms allocate funding to candidates running for office in the U.S. Senate or House of Representatives using political action committees (PACs). In particular, a firm forms a PAC that contributes to a politician s election PAC, which finally distributes a firm s contribution to the politician s campaign. While we focus on election PACs, another form of contributions is through leadership PACs. 1 Firms can also donate to leadership PACs, which cannot use contributions on direct campaign expenses. The Federal Election Commission (FEC) provides detailed data on contributions and elections. We hand-match contributions from firms to our dataset. We additionally incorporate election data into our analysis. The FEC provides data on the outcomes of all U.S. Senate and House elections, including vote tallies by candidate. These data allow us to condition contributions on election outcomes. Lastly, we form political connections based on firm-level contributions to candidates running for election in the Senate or House. A firm is defined as politically connected if it contributed to a winning politician during the previous election cycle. This connection is specified as lasting for two years. For example, a firm contributing to a winning politician during the 2005-2006 election cycle is considered connected to this politician in the following two years in 2007 and 2008. We also explore the robustness of our results by looking at contributions to losing candidates and to winning politicians that are likely to have relatively greater influence in the allocation of contracts. A firm is considered as connected to a losing candidate if it contributes to a losing politician during the previous election cycle. We define those winning politicians likely influencing contracts as the chairmen of the Committee on Appropriations or the Committee on the Budget in 1 We do not include Super PACs in our data, since it is against the law for contributions to these PACs to be used for a politician s campaign. 9
the Senate or the House. Similar to above, firms are considered connected to committee chairmen if they contribute to a winning committee chair during the previous election cycle. C. Measuring Innovation Innovation is considered an important driver of long-term economic growth (Kogan et al., 2012). The main proxy for firm-level innovation is patents. While research and development (R&D) expenditure is a firm s allocation of capital towards innovative activity, it does not capture the productive output of its investment. Griliches (1990) demonstrates that patenting activity is a better measure of research productivity than R&D spending. Further, Hall, Jaffe and Trajtenberg (2005) highlight that patents alone do not indicate technological breakthroughs. Patent citations are a proxy of the value of a firm s innovations. The United States Patent and Trademark Office (USPTO) issues patents and trademarks, in addition to providing comprehensive data on these forms of intellectual property. The NBER dataset, expanded by Kogan et al. (2012), is the source of data on firm-level patent activity in our sample. The count of patents and patent citations are subject to truncation bias. For the count of patents, Seru (2014) reports that the average time from application to granting of a patent is two years. Patent citations are prone to a similar effect, since patents are often not cited until several years after being granted. To correct for these biases in our sample, both the number of patents and patent citations are divided by their annual average for a particular patent s technology class. Technology class is a grouping of patents that is analogous to an industry classification. These variables are referred to as adjusted number of patents and adjusted patent citations. 10
D. Sample Summary The sample includes all firms in Compustat between 2003 and 2010, excluding financial firms (SIC 6000-6999) and regulated utilities (SIC 4900-4999). Table 2 summarizes the firm characteristics, contracts, political contributions and patent activity of the firms included in our paper. Insert Table 2 About Here We include the following firm characteristics as control variables in the analysis. Size is the natural log of firm assets. R&D/Assets is research and development expense scaled by assets. Profitability is measured as earnings before interest, taxes and depreciation over total assets of the firm. Tangibility is the ratio of net property, plant and equipment to total assets. Book leverage is the book value of debt over total assets. Cash/Assets is measured as cash and short-term investment divided by total assets. Market-to-book is the market value of the firm s equity and its book value of debt relative to the firm s assets. HHI is the Herfindahl-Hirschman Index of sales for the industry (at the SIC level). Profitability, Book leverage and Market-to-book are winsorized at the 1% level in each tail. The sample consists of 33,091 firm-years, of which 6,121 observations are matched to contracts. The sample accounts for a yearly average of 45.2% of contracts reported by the federal government. In the analysis, we focus on the full sample (labeled All firms ) when examining whether firms receive contracts and its relation to innovation. We focus on the subsample of firms receiving contracts (labeled Firms receiving contracts ), when the outcome is a contract-level variable, such as its length. This subsample allows us to test how contract-level attributes vary with political connections. Panel A details firm characteristics. The average R&D investment is 9.2% of 11
assets and about 51% of firms invest in R&D. Firms are 32.7% levered and hold about 23.9% of their assets in cash and short-term investments. Panel B summarizes contract data by firm-year. Contract indicator equals one if a firm receives at least one contract during a given year. Award amount is the total amount of awards to a firm in a particular year. Incentives is the percent of contracts with incentives, as defined in Appendix A. Length is the average contract length (in years). Award increase is a binary variable, equaling one if a firm receives an increase in a contract s award and Extension is an indicator variable, equaling one if a firm receives an extension in a contract s completion date. Contracts are awarded to 18.5% of firmyears in our sample and the average size of contracts awarded in a year is $147.6 million. Firms receive incentives in about 9.9% of contracts and the average length for a firm s contracts is over eight months. Contracts receive award increases 68.5% of the time and extensions in 60% of the sample of firms receiving contracts. Political connections are detailed in Panel C. We use contributions from firm PACs to candidates PACs to proxy for political connections, as described in Section B above. Contributions to winner is a binary variable, equaling one if a firm contributes to a winning politician in the previous election cycle. Contributions to loser is a binary variable, equaling one if a firm contributes to a losing politician in the previous election cycle, and Contributions to winning chairman is a binary variable, equaling one if a firm contributes to a winning candidate who is the chairman of the Committee on Appropriations or the Committee on the Budget in the Senate or the House. About 8.1% of firms contribute to winning politicians and 6.2% of firms donate to losing politicians. Firms contribute to a winning chairman just under 1% of the time. Lastly, Panel D summarizes innovation. We measure innovative activity, using number of patents and patent citations. Additionally, we incorporate self citations and patent originality as proxies for a patent s importance. Self citations are defined as a firm s citations to its own patents 12
and proxy for a firm s internal spillovers. Lastly, a patent s originality is defined by its citations to a different technology classes. It is measured as one minus the Herfindahl-Hirschman Index of citations to technology classes. Specifically, Number of patents is the number of patents awarded to the firm in a year and Patent citations is the average citations per patent awarded in a year. Self citations is the average citations to a firm s own patents per patent awarded in a year and Originality measures the diversity of citations made by a patent. These measures of innovation are adjusted by dividing by their annual-technology class average. The mean number of patents in our sample is 6.8 patents and the adjusted number of patents is 0.19. The distribution of patent counts is skewed right, as the median firm produces no patents in a year. Patent citations are a measure of a patent s innovative impact. The average number of patent citations per year is 0.59 and the adjusted number of patents is 0.27, with a distribution that is also skewed right. The average firm has 0.19 adjusted self citations and an adjusted originality of 1.154. III. Political Connections, Contracts and Renegotiation Sections 3 and 4 present the main results of the paper. Section 3 studies the allocation of contracts to firms and its relation to political connections. This section further explores how connections influence contract-level characteristics, such as incentives and renegotiation. Section 4 considers the relation between contracts and firm-level innovation, highlighting the role of incentives in these contracts. A. Contracts In this section, we examine the firm-level determinants of receiving a contract from the U.S. federal government, including political connections. We find that firms contributing to politicians and having weaker balance sheets are more likely to win contracts. Specifically, firms donating to 13
winning candidates and those with lower cash balances and fewer tangible assets tend to receive government spending. We additionally find support that these contracts are awarded to firms with higher growth opportunities and lower leverage. Table 3 reports on how receiving a contract is related to political connections and firm-level characteristics. Insert Table 3 About Here The first column details the marginal effects (at the mean) from a probit model, where the dependent variable equals one if a firm receives a contract from the U.S. federal government. The model includes all firm-years from 2003-2010 and is specified as: Contract Indicator!" = α + β Contributions to winner!" + γx!" + μ! + η!"# + ε!", (1) where Contributions to winner equals one if a firm contributes to a politician and X is a vector of firm characteristics including size, profitability, tangibility, book leverage, cash-to-assets, market-to-book and the Herfindahl-Hirschman Index. Models 1 controls for unobserved, time-invariant industry heterogeneity (η!"# ), in addition to year fixed effects (μ! ), and standard errors are clustered by firm. Firms donating to winning politicians are 10.4% more likely to win contracts from the government. Further, firms with lower cash balances (internal finance) and fewer tangible assets (external finance) are significantly more likely to receive government funds. We also find that large firms with lower leverage and those with higher growth opportunities are significantly more likely to be awarded contracts. 14
B. Initial Contract Characteristics Contract-level observations allow us to explore how characteristics, such as contract type, are related to political connections and firm attributes. Specifically, we study the contract award at signing, in addition to incentives and length. We restrict the analysis to those firms who receive a contract. Columns 2 uses the natural log of contracts awarded in a particular year as the dependent variable. The model uses ordinary least squares (OLS) and firm fixed effects, rather than industry fixed effects, and is otherwise similar to Equation (1) in that it includes the same firm-level controls and year fixed effects (μ! ). The results show that connected firms win contracts that are 15.6% larger than firms that do not make political contributions. One might be concerned that the previous results are driven by small contracts and that the increase in contract award is not economically meaningful. To address this, we estimate a probit model as in Equation (1), where the dependent variable is an indicator for above median awards. The sample is similarly restricted to firms receiving contracts. We find results similar to those for the continuous measure of awards. Firms donating to winning candidates are 18.3% more likely to win large awards than firms that do not contribute to winning candidates. Firms with weaker balance sheets, as proxied by lower cash balances and fewer tangible assets, tend to be awarded large government contracts. Also, larger firms are more likely to win large contracts. Each contract with the government has a specified form of pricing. This pricing scheme can induce firms to complete contracts timely and with high quality by providing monetary rewards to the recipient firm. In the Appendix, we detail the types of contracts classified as having incentives. We study the percent of contracts each year with incentives, equaling weighting each contract. Column 4 of Table 3 details the results on contract-level incentives using a similar specification to Column 2 (OLS with firm fixed effects). The results show that contracts awarded to firms 15
contributing to winning candidates have 3.9% fewer incentives. We find nearly identical results when we weight our measure of incentives by contract value. Another important lever that can be analyzed at the contract level is the length of the contract. Many contracts are completed quickly and small in magnitude. To capture economically meaningful change to the terms of a contract, we create an indicator equaling one if the average contract length is above median and if the award is large (defined as above median in Column 3). All else equal, a longer contract is more favorable for the receiving firm. Using a probit specification as in Equation (1), we find that politically-connected firms are 11.5% more likely to receive contracts with a longer time to completion. Taken together, Table 3 provides evidence on the role of being connected to winning political candidates. Connected firms are more likely to receive a contract. Beyond influencing the allocation of contracts, firms use their political connections to receive contracts that tend to have larger awards, with less incentive-based payoffs and longer completion horizons. C. Contract Renegotiations After a contract is signed, it can be renegotiated or altered. Table 1, Panel B, highlights that renegotiation is frequently observed, with just over 49% of all contracts being changed. We focus on two prevailing forms of renegotiation between a federal agency and a firm. First, we explore changes in award size. A contract obligation can be either increased or decreased once it has been awarded. Second, we study the time to complete a contract. Table 4 provides results about how renegotiated contracts relate to political connections and financial variables. Insert Table 4 About Here 16
This table focuses on those characteristics after a contract has been awarded. The probit models in Columns 1 through 4 are of the form: Renegotiation!" = α + β Contributions to winner!" + γx!" + μ! + η!"# + ε!", (2) where Contributions to winner equals one if a firm contributes to a winning politician and X is a vector of firm characteristics including size, profitability, market-to-book and the Herfindahl-Hirschman Index. These models control for unobserved, time-invariant industry heterogeneity (η!"# ), in addition to year fixed effects (μ! ). Standard errors are clustered at the by firm. Each model is estimated on the sample of firms receiving a contract in a given year. Marginal effects at the mean are reported. We consider four different renegotiation variables. Award Increase is an indicator variable equaling one when a firm receives an increase in awarded contracts. Large Award Increase is a binary variable equaling one if award increase is above median for firms receiving contracts. Extension is an indicator variable equaling one when a firm receives an extension in the completion date in awarded contracts and Long Extension is a binary variable equaling one if the extension is above median for firms receiving contracts. Column (1) evaluates Award Increase. The results show that firms contributing to winning politicians are 11.7% more likely to receive increases in an award. Column 2 considers Large Award Increase and shows that connected firms are 10.9% more likely to receive large increases in their average award. Columns 3 and 4 focus on contract length. Column 3 has Extension as the dependent variable. The marginal effects show that firms contributing to winning candidates are 10.3% more likely to receive contract extensions. Focusing on Long Extension, Column 4 reports 17
that politically-connected firms are 11.8% more likely to receive above average extensions. These results suggest that, in addition to increasing the chance of receiving a contract, firms contributing to winning politicians are more likely to receive increases in contracts already awarded and obtain more time to complete contracts. D. Robustness Tests So far we have taken a straightforward view of political connections. Either a firm contributes to a winning candidate or it does not. If the political connection measure is able to capture preferential treatment for firms, then we may be able to observe additional cross-sectional heterogeneity. We examine two possible situations. First, we examine whether there is a different effect between contributing to winning candidates versus contributing to losing candidates. Second, we evaluate whether contributions to more powerful candidates, as proxied by the politician being a chairman on the Committee for Appropriations or the Committee on the Budget, has a differential impact from contributing to other winning candidates. We find evidence that contributing to a losing candidate is the equivalent of not contributing at all. The results also show that contributing to a committee chairman typically doubles the effect of being politically connected. Instead of repeating the entire analysis in Tables 3 and 4, we select the three indicator variables from Table 3, Contract Indicator, Large Award, and Long Contract, and the two main variables in Table 4, Award Increase and Extension. Panel A reports the analysis for contributing to winning candidates versus contributing to losing candidates. Contributions to winner is a binary variable, equaling one if a firm contributes to a winning candidate during the previous election cycle, and Contributions to loser is a binary variable, equaling one if a firm contributes to a losing candidate in the previous election cycle. In Tables 3 and 4 the benchmark was firms not contributing to the winning candidate, whereas now the benchmark is firms not contributing at all. 18
The results are documented in Table 5. Insert Table 5 About Here Each column in Table 5 Panel A reports the marginal effects at the mean from a probit model, where the dependent variable equals one of the five indicator variables previously mentioned. The model is: Indicator Variable!" = α + β! Contributions to winner!" + β! Contributions to loser!" +γx!" + μ! + η!"# + ε!", (3) where X is a vector of firm characteristics including size, profitability, tangibility, book leverage, cash-to-assets, market-to-book and the Herfindahl-Hirschman Index. The regression controls for unobserved, time-invariant industry heterogeneity (η!"# ), in addition to year fixed effects (μ! ), and standard errors are clustered by firm. For all five dependent variables, the Contributions to winner coefficient is positive and statistically significant at the 5% level or higher. At the same time, the Contributions to loser coefficient is never statistically significant. The effect is not driven by contributing to a candidate, but rather it is driven by contributions to politicians who win. We find that firms contributing to losing candidates receive contracts that look no different (along the dimensions we analyze) than firms that make no political contributions at all. This test also helps to address a possible endogeneity issue. One concern might be that firms contributing to politicians might be somehow different than firms that do not donate to candidates. This test provides evidence against this concern. Next we evaluate whether, among winning candidates, there is heterogeneity in the effect of contributions, specifically whether contributing to a committee chairman has a differential impact 19
than contributions to a less senior candidate. The dependent variables and regression specification is identical to Equation (3) except that the independent variable Contributions to loser is substituted with Contributions to winning chairman, a binary variable equaling one if a firm contributes to a winning politician who is a chairman on the Appropriations or Budget committees of the Senate or the House of Representatives. In each regression, the coefficient on Contributions to winning chairman is nearly the same size as the coefficient on Contributions to winner. For instance, in the Large Award regression (Column 2), firms contributing to a winning candidate are 16.1% more likely to receive a large award. In the same specification, we find that firms contributing to a winning chairman are 20.7% more likely receive a large award. This effect on contributing to a winning chairman is in addition to the effect of contributing to a winner. Therefore, contributing to a winner who is a committee chairman has the combined effect of increasing a firm s likelihood of receiving a large award by 36.8%. The results in the other columns are similarly large in economic magnitude. These results further suggest that political connections is a channel through which contracts can be influenced. The measure of connectivity is reasonably strong as to capture interesting cross-sectional variation. E. Contract-Level Analysis The analysis in the sections above has studied contracts at the firm-year level. The unit of observation in the data is a contract, which allows us to additionally estimate models at the contract level. A concern might be that contract characteristics differ by the Federal agency awarding the contract. By estimating our specification at the contract-level, we can directly address this by removing time-invariant agency heterogeneity. We focus on the main variables of interest: contract award, length and incentive at signing, in addition to award change and extension through renegotiation. Award is the natural log of the initial 20
contract award (in thousands of dollars). Length is the contract length at signing (in years). Incentives is a binary variable equaling one if a contract has incentives. Award Change is the percent change of an award after its initial signing, relative to the total award of the contract. Extension is the change in a contract s completion date after its initial signing (in years). We now include similar controls across specifications, using size, profitability, market-to-book and the Herfindahl-Hirschman Index, as well as year and industry fixed effects. The results are reported in Table 6. Insert Table 6 About Here We largely find that our results are robust to including agency-level fixed effects. Column 1 reports that firms contributing to winning politicians receive contracts that are 24.1% larger and 18.8% longer (relative to mean contract length), though we no longer find a significant effect for contracts with incentives. Through renegotiation, we report that firms contributing to winning candidates receive 2.2% larger increases in the average award and 8.3% longer extensions (relative to the mean extension). The conclusion of these findings is that, even controlling for agency-level variation, the political-connected firms continue to receive preferential contract allocation. IV. Innovation, Contracts and Incentives This section studies innovation and firms receiving contracts from the federal government. A measureable outcome and stated goal of federal contracting is to spur innovation. We find that firms receiving government contracts tend to experience increases in the scale and novelty of innovation, as measured by number of patents and citations. One mechanism through which contracts can stimulate innovation is with incentives, which award firms for the quality and 21
timeliness of performing a contract. We find evidence that these contract-level incentives are associated with increases in patent production and their citations, consistent with this channel. Further, firms contributing to winning politicians that receive contracts tend to experience a decrease in the scale and novelty of innovation. Since politically-connected firms tend to have less incentive contracts, this provides contract design as a channel through which political connections distort the allocation of government capital. Before linking incentives and political connections with innovation, we first evaluate the relationship in general between government contracts and firm-level innovation. Table 7 highlights the relation between future patenting activity and government contracts. Insert Table 7 About Here The dependent variable the number of patents a firm is awarded in a year divided by its annualtechnology class mean. We analyze patents in 1,2, 3 and 4 years after a firm receives a contract. The analysis includes all firm-years. For the 1-year patent analysis, this results in 33,091 observations. The number of observations decreases for longer patent periods as the later years cannot be included due to their shorter horizon. For the 4-year patent analysis the sample is reduced to 17,083 observations. The model is: Patents!" = α + β! Contracts Indicator!" + γx!" + +μ! + η! + ε!", (4) where Contracts Indicator is an indicator variable equaling one if a firm received a contract and X is a vector of firm characteristics including size, R&D, profitability, tangibility, book leverage, cash holdings, market-to-book, and the Herfindahl-Hirschman Index. All of the models control for 22
unobserved, time-invariant firm heterogeneity (η! ), in addition to year fixed effects (μ! ), and clusters standard errors by firm. The main coefficient of interest is β!, which captures the association between receiving a government contract and patent productivity. Controlling for firm characteristics, we find that firms receiving government contract tend to have significantly higher levels of patent production. One year after receiving a government contract, firms tend to experience a 0.149 increase in the adjusted number of patents. This increase lasts through the following four years and slightly decreases to 0.108 in the fourth year. Not only do firms receiving government contracts immediately become more innovative, they continue to increase their patent production into the foreseeable future. While patent production is a straightforward measure of innovation, patent citations may be a better proxy for the value of innovation. Table 8 studies the relation between patent citations and government contracts. The model is the same as in Equation (4) except the dependent variable now captures patent citation. Specifically, the dependent variable is defined as the average citations per patent granted, divided by its annual-technology class mean. The results are reported in Table 8. Insert Table 8 About Here Controlling for firm characteristics, we find that firms receiving government contracts have more novel patents, as measured by citations. Similar to the number of patents, we find that evidence that innovation novelty increases for the four years after receiving a contract. Tables 9 and 10 study the relation between innovation, political connections and incentives. The model is: Innovation Variable!" = α + β! Contracts Indicator!" + β! Contributions to winner!" +β! Contracts X Contributions to winner!" +β! Contracts X Incentives!" + γx!" + μ! + η! + ε!", (5) 23
As in tables 7 and 8, we measure innovation using the number of patents or citations. We also incorporate measures of the importance of citations using self citations and originality. Self citations is the average number of a citations that a patents receives from patents in the same firm. Originality is a Herfindahl-Hirschman Index of citations to technology classes and proxies for the diversity of citations made by a patent. Contributions to winner is an indicator variable equaling one if a firm contributed to a winning politician in the previous election cycle and Incentives is a binary variable equaling one if a firm receives contracts with incentives during the current year. The coefficients β! captures the association between receiving a contract and innovative activity, and β!, which is the association between contributing to a winning candidate and patent measures. Further, the coefficient β! is the association between receiving a contract, being politically-connected and innovation, while β! captures the relation between contract-level incentives and firm-level patent activity. Table 9 presents results for patent production. We find a positive and significant level effect for receiving a contract, while we find a negative and significant association between contributing to a winning candidate and firm-level patenting. Interestingly, we report that firms receiving contracts with incentives have significantly higher levels of innovation, while politically-connected firms receiving contracts tend to produce less patents. These results point to a channel through which political connections lead to declines in innovative activity. Firms with connections tend to have lower incentives in their contracts, as shown in Table 3. Consistent with this channel, these firms have much lower levels of patenting activity. Table 10 details the association between three measures of citations, political connects and incentives. The relationship with patent citations is similar but weaker. We find that politically connected firms tend to produce less novel innovation, as measures by citations, self citations and originality, while we do not find a level effect for receiving a contract. We report that politically- 24
connected firms receiving contracts tend to have less self citations and lower originality. Additionally, we find that contract-level incentives are related to increases in all measures of citations. V. Conclusion Using hand-collected data on government contracts won by public firms from 2003 to 2010, we examine political connections, contract-level characteristics and firm-level innovation. We find that politically-connected firms are more likely to win contracts and these contracts tend to be larger with later completion dates. We additionally find that contracts awarded to connected firms have lower incentives and are favorably renegotiated. In robustness tests, we show that firms connected to losing candidates do not receive any benefits in the allocation of contracts, while firms connected to committee chairs tend to receive nearly twice the average effect of being politically connected. Further, we provide evidence that firms winning contracts innovate less and incentive-based contracts foster innovation. This suggests that lower contractual incentives is one channel through which politically-connected firms innovate less. 25
Appendix Section A of this appendix describes the variables examined in the paper. Section B details the matching procedure for linking contract data to Compustat firms. A. Variable definitions This section defines the main variables of the paper and their construction, providing the Compustat definition where applicable. U.S federal contract data is from the Federal Procurement Data System (FPDS) and retrieved from USAspending.gov. Patent data is provided by Kogan et al. (2012), which builds on the NBER patent data matched to Compustat firms. The underlying patent data is provided by the United States Patent and Trademark Office (USPTO). Campaign contributions and election data is from the Federal Election Commission. Variable Name Description Compustat Definition Contract indicator A binary variable equaling one if a firm receives contracts from the U.S. federal government in a particular year Award amount Natural log of contract awards (in millions of dollars) in a given year Incentive Percent of contracts awarded with incentives, which are defined as those contract whose type is Fixed Price Incentive, Cost Plus Incentive, Fixed Price Award Fee, Cost Plus Award Fee, Fixed Price Level of Effort, Fixed Price with Economic Price Adjustment or Cost Plus Fixed Fee Length Time to complete a contract in years Award Increase Extension An indicator variable equaling one if a firm receives an award increase after receiving a contract A binary variable equaling one if a firm receives an extension to complete a contract 26