Investment incentives attract foreign direct investment: evidence from the great recession

Abstract

Do investment incentives influence private firms’ location decisions? Whereas prior research emphasizes tax incentives, we focus on incentives that require real-time government spending including job training and infrastructure. Real incentives influence where firms invest by resolving costly information asymmetries, and are subject to budget constraints that give rise to political targeting. This paper evaluates how real incentives shape the location decisions of foreign firms, investors who suffer from acute information asymmetries. We leverage features of the Great Recession and the 2009 American Recovery and Reinvestment Act stimulus, which temporarily increased states’ fiscal capacity to fund real incentives. During the narrow stimulus spending window, states that received more federal Medicaid stimulus—instrumented with the exogenous component of the Act’s funding formula—attracted more foreign direct investment (FDI) and increased state spending on real incentives. The stimulus window approximately coincides with FDI’s temporary geographic expansion into US counties that lacked a history of these investments. On average, these counties had narrow vote margins in the prior gubernatorial election and garnered more state real incentive spending. These correlates are pronounced in counties with idle industrial capacity and in states whose governors sought re-election. Tax incentives had no effect on FDI. These findings have important implications for the efficacy of investment incentives and the political economy of industrial policy.

1 Introduction

State and local governments in the United States spend an estimated $50 billion annually on investment incentives, a threefold increase since 1990 (Bartik, 2019b, 2020). These subsidies to attract new investment by private firms are controversial. Evidence is mixed as to whether incentives influence firms’ location decisions or contribute to economic development enough to justify their cost to taxpayers (Austin et al., 2018; Greenstone et al., 2010; Kline & Moretti, 2014; Slattery & Zidar, 2020). Sharper critiques cast incentives as little more than veiled corporate subsidies that fuel destructive bidding wars among states (Burstein & Rolnick, 1995). As US policymakers grapple with global supply chain vulnerabilities and rising income inequality, incentives garner renewed interest as a key tool of industrial policy (National Economic Council, 2023).

Current research focuses on tax incentives, which reduce firms’ tax liability for undertaking activities such as job creation and research (Bartik, 2019a). Conventional wisdom holds that tax incentives rarely drive where firms locate because firms first select locations that meet their basic production needs and then seek incentives (Jensen, 2018). Politicians nonetheless offer tax incentives to opportunistically claim credit from voters for job-creating investments (Jensen & Malesky, 2018). Credit claiming is immediate while the fiscal costs of lower tax revenues are deferred. Unconstrained by current budgets, tax incentives give rise to bidding wars (Sobel et al., 2022).

By contrast, we argue that certain types of incentives can influence firms’ location decisions by addressing information asymmetries that raise firms’ costs. These incentives typically entail real-time government spending including cash grants, specialized business inputs, and land development. Although real incentives account for a minority of incentives that US states offer, they contribute to local economic development more generally and their value to firms can exceed their dollar cost to taxpayers (Bartik, 2020). They are especially valuable to firms who are unfamiliar with the market and in circumstances that magnify costs of information asymmetries.

Spending on real incentives is subject to current budget constraints whereas tax incentives forgo future income. From a welfare perspective, constraints effectively preclude bidding wars of the magnitude that tax incentives enable. Constraints also require politicians to offer real incentives selectively. We argue that real incentives are a form of targeted spending (Ansolabehere & Snyder, 2006; Cain et al., 1987; Grimmer et al., 2012). Politicians strategically offer real incentives to help attract investment to jurisdictions with key voters. Empirical tests of these claims must overcome the inference challenge that states with the fiscal capacity to offer real incentives may be systematically different in other ways relevant to firms’ location decisions. Further, these tests must disentangle politicians’ electoral motives to target incentives from other location characteristics salient to investing firms.

We meet these challenges by leveraging features of the Great Recession and the 2009 American Recovery and Reinvestment Act. The Act dispersed $200 billion to state governments, an exogenous positive budget shock that temporarily increased states’ fiscal capacity to fund real incentives. Nearly half was distributed across states using the existing federal funding formula for Medicaid, the means-tested health insurance program jointly funded by federal and state governments. States, however, had full discretion over how to spend these funds, which they had to spend before October 2010. Our instrumental variable framework centers around an exogenous source of variation in the Medicaid formula, states’ pre-recession Medicaid spending (Chodorow-Reich et al., 2012). Using this framework, we analyze the effects of Medicaid stimulus on new manufacturing foreign direct investment (FDI) into US states during 2009–2011. Foreign firms are sought-after investors who create new jobs and positive productivity spillovers, and can suffer from especially acute information asymmetries.

We find that Medicaid stimulus corresponded to higher state FDI inflows during 2009–2011, controlling for several other state-level drivers of FDI. FDI concentrated in counties that had not received new FDI in the preceding 5 years, suggesting distinctive drivers of firms’ location decisions during this period. Our findings are robust to different model specifications and alternative definitions of the stimulus spending window. Using the same empirical framework, we find that stimulus corresponded to higher state spending on real incentives. On average, states spent $170,000 more on real incentives for every $1 million in Medicaid stimulus received. A subset of real incentives drives this pattern, incentives that resolve information asymmetries related to site selection and adapting production processes. Stimulus had no direct bearing on state’s capacity to offer tax incentives and, accordingly, we find no effect on tax incentives granted during the stimulus spending window.

Next, we investigate the mechanisms through which real incentives influence FDI location decisions. Figure 1 plots the geographic distribution of new manufacturing FDI in the US by quarter during 2003–2016. The black line indicates the number of US counties that had at least one new announced investment in that quarter. The gray line depicts what we call “new” counties, the subset of counties in that quarter that received FDI for the first time since 2003, the first year for which county-level FDI data are available. During 2003–2008, a total of 208 US counties received FDI. During 2009–2011, 396 counties received FDI, of which 229 (fifty-eight percent) were new counties. The trend is short lived, with the expansion returning to pre-recession levels after the stimulus spending window closed.

Fig. 1

County Distribution of US Greenfield Manufacturing FDI Inflows, 2003–2016 (Quarterly). The figure plots the number of US counties that received at least one new (greenfield) manufacturing FDI investment by quarter. Black line plots the total number of counties that received investment. Gray line is the subset that received investment for the first time since 2003. Vertical lines demarcate the Great Recession. Data Source: fDi Markets database

We analyze the correlates of new county status to explore how real incentives influence firms’ location choices. Although we cannot directly observe if incentives were decisive, our empirical setting of a temporary expansion holds constant many of the fundamental location characteristics thought to attract FDI. We estimate a multinomial logit model of county FDI status during 2009–2011: “new” (received FDI during but not before), “old” (received FDI before and during) and “none” (no FDI during). Controlling for county-level drivers of FDI, the sum of real incentives correlates with new county status. Tax incentives, however, do not. A positive interaction of mass layoffs, a proxy for idle industrial capacity, and narrow vote margin in the prior gubernatorial election suggests that state politicians directed real incentives to counties that both met firms’ needs and in which politicians could claim credit for investment from swing voters. Narrow margin counties were more likely to be new in states with incumbent governors who sought re-election, consistent with these governors’ stronger motives to claim credit from voters. We address several alternative explanations for new county status including local governments’ incentive spending, stimulus-funded infrastructure improvements, and changed industry composition of FDI. We validate our classification of county FDI status with historical proxies, and also show no geographic expansion of domestic investment.

Our study contributes to scholarship on industrial policy by demonstrating that real incentives attract FDI by alleviating information asymmetries. We offer an important qualification to the conventional wisdom, based on tax incentives, that incentives are ineffective (Slattery & Zidar, 2020). Our findings extend prior work on how information asymmetries influence investment (Criscuolo et al., 2019; Crescenzi et al., 2021). A clear policy implication of our findings - offer real incentives when information asymmetries are acute - is especially relevant for contexts in which firms must quickly reconfigure supply chains such as the 2018 US–China trade war and the Covid-19 pandemic.

Our findings also contribute to scholarship on politicians’ motives to offer investment incentives. Prior work emphasizes that politicians offer tax incentives because their successors bear little cost of forgone tax revenue (Jensen & Malesky, 2018). Our findings suggest that politicians are willing to prioritize incentive spending despite real-time trade offs with other categories of public spending. Whereas existing research focuses on electoral drivers of local government incentives (Jensen et al., 2020, 2015), our analysis of state incentives allows us to analyze potential geographic targeting of incentives. Our county-level findings are consistent with state politicians using incentives to claim credit from a specific subset of the electorate, swing voters.

More broadly, we contribute to research on the political economy of targeted spending (Ansolabehere & Snyder, 2006; Cain et al., 1987; Dixit & Londregan, 1996; Grimmer et al., 2012). We show that, despite federal efforts to allocate stimulus in a transparent manner, state politicians’ electoral motives guided both the programmatic (re)direction and geographic distribution of federal transfers (Young & Sobel, 2013). Additionally, our focus on foreign-owned firms holds constant to some degree formal lobbying for spending policies, demonstrating that institutions can influence public spending purely through shaping politicians’ electoral incentives (Aidt & Shvets, 2012).

2 Theoretical framework

The goal of investment incentives is to influence where private firms locate economically productive activities. Tax incentives reduce firms’ tax liabilities and may be contingent on the number of jobs created, or linked to specific activities such as research and development. Real incentives, defined by the real-time government expenditures that they typically entail, defray investing firms’ costs or enhance productivity. They take several forms including cash grants, loan guarantees, and specialized inputs such as job training, industrial land, and infrastructure. In practice, firms receive incentive packages that bundle multiple types of incentives. Some incentives are automatically available to firms who meet published criteria whereas others are subject to policymakers’ discretion.

In the United States, state governments are the single largest source of incentives (Slattery & Zidar, 2020). Figure 2 plots annual state incentive expenditures during 1977–2016. Tax incentives grew sharply just prior to the Great Recession, driven by exceptionally large tax incentives for specific projects (Sobel et al., 2022). Real incentives exhibited modest growth. During 2009–2011, the average tax incentive was almost $3 million whereas the average real incentive was about $500,000.

Fig. 2

Data Source: Good Jobs First Subsidy Tracker

US State incentive expenditures, 1977–2016. Annual spending measured in constant 2010 dollars. Vertical lines demarcate the Great Recession

Conventional wisdom, focused on tax incentives, holds that incentives do not influence where firms locate because, the wisdom holds, firms select locations according to basic production needs and then seek incentives (Jensen, 2018). Thus, even if incentives correlate with new investment, they do not necessarily drive location choices. Politicians may recognize this but nonetheless offer tax incentives to give voters the impression that the politician’s efforts were decisive in attracting the investment (Jensen & Malesky, 2018). Tax incentives allow politicians to claim credit for new investment immediately while deferring the consequences of reduced tax revenue.

We argue that this conventional wisdom does not extend to real incentives. Real incentives can influence where firms invest by alleviating costly information asymmetries. We develop this argument in the context of FDI, investments by foreign firms, which, in the US, are generally to produce and sell within the US.¹ Firms undertake FDI to capture global scale economies from highly productive intangible assets - technology, managerial practices - while maintaining control over the assets (Alfaro, 2017). Given high costs of multinational production, foreign firms are among the world’s most productive (Helpman et al., 2004). Therein lies politicians’ intense interest in attracting FDI. Not only does FDI create new jobs, FDI spillovers increased American firms’ productivity by 14% (Keller & Yeaple, 2009) and American workers’ wages an average of seven percent (Setzler & Tintelnot, 2021). Like most firms, foreign firms seek locations that meet their basic production needs, including skilled labor, reliable infrastructure, and low regulatory burden, among other needs.

To varying degrees, foreign firms confront costs arising from their lack of familiarity with the host market. We characterize these costs as information asymmetries, using the term to encompass cultural/linguistic frictions and the absence of relationships with key partners (Eden & Miller, 2004; Zaheer, 1995). Foreign firms have higher initial production costs associated with adapting products and production practices to a new market (Chor & Manova, 2012), and higher labor and input search costs (Javorcik, 2015). Their borrowing costs are higher due to the relative lack of relationships with host country lenders (Antràs & Yeaple, 2014; Desbordes & Wei, 2017),² Underlying many of these costs are cultural and linguistic differences that further complicate the already complex communication required to produce and sell goods made with sophisticated production technologies (Oldenski, 2012).

2.1 Real incentives influence FDI location decisions

Real incentives address costly information asymmetries.³ Cash grants and loan guarantees offset higher capital requirements, and often with less scrutiny than a private lender would exercise (Buch et al., 2009). Publicly funded, firm-specific worker training helps foreign firms overcome cultural frictions to realize the productivity potential of advanced technologies (Fosfuri et al., 2001). These programs reduce labor turnover and increase productivity, especially in collaboration and communication-oriented production tasks (Hollenbeck, 2008). All else equal, foreign firms would have a harder time establishing such programs themselves. They are less familiar with local educational and training resources and require local expertise to adapt production processes to the local market. European and Japanese firms indicate a particular interest in training programs because they anticipate that American workers lack skills essential to their production techniques (Schneider, 2010). Incentives can reduce costs in other ways. Manufacturing firms require factories and other kinds of industrial land (Munson & Schultz, 2013). Incentives that subsidize these inputs reduce both firms’ real financial costs and search costs. Depending on the industry, firms may require upgrades to local infrastructure such as roads or power grids, improvements for which private alternatives are few. Manufacturing extension programs exist to facilitate connections with suppliers and the adaption of products to the local market (Brandt et al., 2018; Lowe et al., 2023).

Real incentives can influence where foreign firms locate by providing a location-specific solution to information asymmetries.⁴ When foreign firms evaluate the internal rate of return for potential investment sites, they seek to minimize infrastructure and start-up costs (Woessner-Collins, 2010). Many types of real incentives, though not all, have an inherent location-specific component. For example, provision of land and factory space necessarily dictate firms’ location decisions. Bartik (2020, p. 117) relays the example:

One CEO told me that his decision about where to locate a particular new facility had been determined by the availability in that city of an empty factory. The empty factory allowed the new facility to get into production quickly.

Incentives related to infrastructure and specialized inputs often build upon existing, place-based capacity. Job training programs often require collaboration with educational institutions with the relevant capacity to facilitate training (Miller, 2014).

2.2 Electoral motives influence distribution of real incentives

A crucial difference between tax and real incentives is that latter are subject to budget constraints. Whereas politicians can offer tax incentives widely, their ability to offer real incentives is limited by both budget constraints and, for some incentives, the local characteristics upon which the incentives build. Within these parameters, politicians may still prioritize areas where incentives yield larger electoral rewards. For example, analyses of governors’ designation of areas eligible for a large federal tax incentive program find that designated census tracts met eligibility criteria and were also more likely to have a state representative from the governor’s party (Alm et al., 2021; Frank et al., 2022) or have firms connected to the governor (Eldar & Garber, 2023). Thus, any account of how real incentives influence FDI location decisions has to consider how politicians’ motives fuel geographic variation in the availability of incentives.

FDI can have electoral benefits for incumbent politicians (Jensen & Malesky, 2018; Owen, 2019).⁵ New plant announcement are high profile local media events that connect a politician to job creation in voters’ minds (Bao & Chen, 2018). Despite a resurgence of economic nationalism (Andrews et al., 2018), nearly 75% of Americans surveyed report a favorable view of new FDI (Pew Research Center, 2014). Politicians do not discriminate against foreign firms when offering incentives (Jensen et al., 2020).

We propose that, to the extent that politicians can favor certain areas, they offer more incentives to projects in counties with swing voters. Investment incentives are well suited to persuade voters who lack strong party allegiances (Cox, 2009). Such voters are often more responsive to targeted spending (Dixit & Londregan, 1996; Grimmer et al., 2012), economic policies of ideologically distant candidates, and politicians’ skill and experience (Fowler et al., 2022). In general, voters hold incumbent governors accountable for state economic conditions regardless of whether they have control over the state’s economy (Atkeson & Partin, 1995).

3 Empirical context: the great recession

The Great Recession is an insightful setting to examine how real incentives influence where in the US foreign firms locate. From foreign firms’ perspective, borrowing from US banks became even more costly (Chodorow-Reich, 2014). Foreign parent companies, themselves credit constrained, had less capacity to provide working capital (Biermann & Huber, 2023; Buch et al., 2009). Additionally, the US dollar appreciated, which increased foreign investors’ real costs (Froot & Stein, 1991). Firms reported greater interest in incentives (Johnson & Toledano, 2022), citing incentives as more salient to their location decision than in previous years (Gambale, 2011). Tax incentives, however, may have been less attractive because secondary markets for tax credit monetization contracted (Aldy, 2013).

Federal stimulus partially relieved states’ budget constraints in offering real incentives. In general, states operate under tight fiscal constraints that require trade-offs across spending categories and between spending and taxation (Poterba, 1994). Most states are legally required to balance their budgets and have limited scope to borrow (Jonas, 2012). The recession strained state budgets to a degree not seen since the Great Depression (National Association of State Budget Officers, 2009). On average, federal stimulus replaced approximately one-fourth of lost state revenue (Leachman & Williams, 2021).

Despite these constraints, states increased incentive spending (Bosman, 2009; Stringer, 2010), especially cash incentives (McIntosh, 2012). Ohio’s governor claimed credit for a $650 million investment by French steel manufacturer Vallourec, explicitly stating that federal stimulus was redirected towards incentives (Akron Beacon Journal, 2010). Many states expanded the geographic scope of previously localized incentive programs (Goodman & Wakefield, 2021; The Pew Charitable Trusts, 2021). States redoubled investment promotion efforts, which inform potential investors about specific counties in the state suited to their needs. Anecdotal evidence is consistent with governors’ electoral motives to privilege certain counties for incentive spending. Wisconsin Governor Scott Walker designated Milwaukee as a city-non-grata because of its pro-union orientation and the city failed to receive a single major subsidy package during 2010–2016 (Hinkley & Weber, 2021; McCarthy, 2015, p. 835).

3.1 Empirical strategy

Our empirical strategy leverages distinctive features of the 2009 Recovery Act, which transferred $200 billion to state and local governments. Health care, education, and transportation accounted for over ninety percent of transfers. The act relied heavily on existing statutory funding formulas to distribute funds quickly and in a transparent manner and to limit pork barrel-style targeting. Consistent with this goal, congressional district-level stimulus expenditures do not correlate with partisanship (Boone et al., 2014; Gimpel et al., 2012).

Our state-level analysis focuses on the single largest stimulus transfer, $88 billion towards Medicaid. Typically, the federal government funds 50–83% of a state’s Medicaid expenditures. The precise federal contribution is based on the Federal Medical Assistance Percentages (FMAP) formula. FMAP incorporates a 3-year rolling average of state unemployment and other state economic characteristics, such that states with worse economic performance receive more funding. The Recovery Act temporarily increased the federal government’s share of Medicaid expenses by 6.2 percentage points across the board, with additional increases indexed to current state unemployment. The Act retroactively applied this modified formula from October 2008. Medicaid stimulus accounted for 75% of Recovery Act transfers distributed in the first quarter of 2009.

Our state-level empirical strategy rests on three features of this stimulus. First, Medicaid stimulus was a positive shock to state budgets. Federal lawmakers “intended to boost the level of discretionary funds available to states and not simply to relieve Medicaid burdens” (White House Council of Economic Advisors, 2009). States also replaced some of their own planned spending with stimulus, freeing up state funds for other uses (Conley & Dupor, 2013; Dupor, 2013).⁶ Peter Orzag, then-director of the federal Office of Management and Budget, blamed this practice for the act’s modest effects on economic growth (Boone et al., 2014).

Second, the Recovery Act relied on the FMAP formula to distribute stimulus across states. Following research on the employment effects of stimulus (Chodorow-Reich et al., 2012), we deploy the plausibly exogenous portion of the formula, state Medicaid spending in 2007, in an instrumental variable framework. Third, states forfeited any unspent Medicaid stimulus left at the end of the 2010 federal fiscal year (September 30, 2010). States were prohibited from depositing funds into reserves or narrowing Medicaid eligibility. This requirement provides us a discrete window to evaluate the effects of stimulus on incentive spending.

Our county-level analysis considers the potential effects of non-Medicaid stimulus spending on which counties received FDI. Of particular interest is the $54 billion in education stimulus. While we discuss this spending in greater detail below, we note here two important features relevant to our research design. The act relied on funding formulas to distribute these funds across states so, similar to Medicaid stimulus, their distribution across states was apolitical. Unlike Medicaid stimulus, governors had little flexibility to use these funds. The act required states to distribute education stimulus to local education agencies according to existing education funding formulas and required that states maintain their education spending at specified levels. The law also granted states an additional year to spend non-Medicaid stimulus and subsequently extended that deadline one to two years for many programs.

Finally, two additional stimulus provisions are relevant to our argument. Federal grants and loan guarantees sought to expand private investment in renewable energy. Both foreign and domestic firms were eligible for this support. Consistent with our theoretical argument, grants under this program were considered successful in attracting renewable energy investment whereas tax provisions were not (Aldy, 2013). Second, the act’s “Buy American” provision required all building materials used in stimulus-funded construction of public buildings be sourced domestically. Both provisions plausibly motivated foreign firms in relevant industries to invest in the US quickly, suggesting greater sensitivity to information symmetries.

3.2 FDI during great recession

During 2009–2011, new manufacturing FDI was more than triple than during 2005–2007. Despite the recession, the US remained an attractive market for foreign firms. A 2009 United Nations survey revealed that most multinational companies anticipated that US market demand would rebound by 2012 (UNCTAD, 2010).

FDI’s post-recession geographic expansion defies simple explanation.⁷ Ostensibly, the basic logic of firms’ location decisions did not change,⁸ FDI tends to spatially agglomerate, reflecting firms’ common location-specific needs including labor and infrastructure, and positive externalities from proximity to other foreign firms and firms in their industry (Bobonis & Shatz, 2007; Head et al., 1995). Location decisions for manufacturing FDI, our focus, are relatively flexible as compared other industries in which proximity to natural resources or customers dictate location. For example, Golden Dragon, a Chinese manufacturer of copper pipes and tubes reported considering 62 sites across the US before selecting Thomasville, Alabama for its first US plant in 2011 (Amy, 2011).

Appendix A compares FDI across new and old counties. One notable difference is that among new counties, two industries feature prominently, metals and renewable energy, consistent with Recovery Act provisions that motivated foreign firms in these industries to establish themselves in the US market quickly. Renewable energy firms display classic features of information asymmetries including high initial capital requirement and the need for specialized labor (Woessner-Collins, 2010). Both industries have arguably weaker motives to locate in old counties. Renewable energy, as a relatively new industry, had fewer opportunities for agglomeration externalities. As a relatively low value added industry, steel has inherently lower externalities. The average project value and distribution of FDI across source countries and industries were broadly unchanged after the recession, with most FDI flowing from advanced industrialized source countries into manufacturing industries.⁹

Some might associate FDI with very large investments by sophisticated multinational companies with experience and internal capacity to overcome information asymmetries. New county investments tended to create fewer jobs, suggesting that new county investments do not fit this description (Table A1). For example, Alpla, an Austrian manufacturer of plastic packaging, invested in Hoke County, North Carolina, creating 40 jobs. The state’s One North Carolina Fund awarded Alpla a $120,000 grant (McCleary, 2009). To the extent that smaller investments imply larger information asymmetries, real incentives may have had a larger role in these firms’ location decisions. As compared to new counties, old counties have higher mean project values and higher variance, indicating that investment by firms less swayed by real incentives concentrated in old counties.

4 State-level empirical analysis

4.1 Data

Our dependent variable in the baseline state-level analysis is the sum of new state manufacturing FDI during 2009–2011 (inflation-adjusted millions of US dollars).¹⁰ We measure investment using project-level FDI data from the Financial Times’s fDi Markets database.¹¹ The database reports salient project characteristics including industry, investors’ country of origin, production activities, and the plant’s US county location. Our sample is restricted to new manufacturing plants.¹² We model FDI’s geographic expansion by disaggregating state FDI into “new” FDI, counties that had not received FDI during 2003–2008 but did during 2009–2011, and “old”, which had received FDI both before and during. “None” counties received no FDI during 2009–2011. The sample consists of 229 new, 56 old, and 2827 none counties.¹³

Our independent variable is total Medicaid stimulus transfers to the state. In an ordinary least square framework, FDI may correlate with the regression error term through the potential effect of contemporaneous state economic conditions on location decisions.¹⁴ Following Chodorow-Reich et al. (2012), our instrument is the exogenous component of the FMAP formula, 2007 state Medicaid spending.

4.2 Model

We use a two-stage least squares regression to estimate the causal effect of Medicaid stimulus on state FDI inflows during 2009–2011, $Y_s$. Our sample ends in 2011 to allow for potential lags between the end of stimulus (September 2010) and announcement of foreign investment. In the first stage, we regress Medicaid stimulus on our 2007 Medicaid spending instrument $Z_s$. In the second stage, we regress state FDI inflows on instrumented Medicaid stimulus, ${\hat{S}}_s$. The coefficient of interest is $\beta$, which captures the causal effect of Medicaid stimulus on FDI. A $\beta$ that is positive and significant would be consistent with stimulus attracting new FDI.

First stage:

$$\begin{aligned} \text {1(a)}\;{\hat{S}}_s = \gamma + \lambda \mathbf {Z_s} + \zeta \mathbf {X_s} + \delta v_s + \psi _{s}, \end{aligned}$$

Second stage:

$$\begin{aligned} \text {1(b)}\;Y_s = \alpha + \beta {\hat{S}}_s + \eta \mathbf {X_s} + \kappa v_s + \epsilon _{s}, \end{aligned}$$

We include a vector of controls, $X_s$, for state characteristics that may have influenced a state’s propensity to receive FDI independent of stimulus. Given FDI’s tendency to geographic agglomeration, we control for stock of state FDI with the sum of announced projects during 2003–2008. We account for multiple state labor market characteristics including union membership in 2007, lagged employment growth (from May to December 2008), 2008 state unemployment rate, and 2008 manufacturing share of state employment. We also control for 2008 state gross domestic product (GDP) per worker capita and adult population, which may correlate with levels of Medicaid stimulus received. John Kerry’s 2004 presidential vote share controls for a state’s political and regulatory investment climate. Finally, we use region-fixed effects, $v_s$, for nine census divisions to account for unobserved differences that may influence FDI location decisions.

4.3 Results

Table 1 presents our first-stage estimates. Model 1 reveals a significant and positive bivariate correlation between the instrument and Medicaid stimulus. The estimated coefficient of the instrument in Model 2 with full covariates is 0.16 and is statistically significant at the 95% confidence level, which reflects the importance of pre-recession Medicaid transfers in determining stimulus payments to states. With F-statistics well above 10, we can reject the null hypothesis that our instrument is weak.

Table 1 First stage regression results

Table 2 presents our second stage results. The dependent variable in Model 1 is total state FDI inflows during 2009–2011. Model 1 shows that Medicaid stimulus had a positive and significant effect on FDI. Models 2 and 3 restrict the sample to FDI in new and old counties, respectively. This positive effect is driven exclusively by new counties. The coefficient estimate in Model 2 shows that, on average and accounting for other relevant factors, each additional $1 million in Medicaid stimulus corresponds to $300,000 in FDI in new counties. We find no effect for old counties (Model 3).

Table 2 Two-stage least squares regression of greenfield FDI on medicaid stimulus

Given our cross-sectional model, we cannot control for time-invariant state characteristics. High Medicaid spending states may be systematically different in their propensity to receive FDI. We address this possibility by estimating a model of change in FDI in new counties between 2006–2008 and 2009–2011, which provides a rough approximation of growth in new investment in 2009–2011 as compared to the preceding three years (Model 4). The effect of Medicaid stimulus remains positive and significant. Results are not sensitive to changing the sample period to 2009–2010 (Table A3). Additionally, we regress FDI inflows during 2006–2008 on post-recession Medicaid stimulus. Our null result indicates that pre-recession FDI did not correlate with the Medicaid stimulus (Table A4).

4.4 Incentives drive FDI growth

Next, we evaluate the role of incentives in driving state FDI inflows during 2009–2011. Incentives data are from Good Jobs First, a non-governmental watchdog group. Their Subsidy Tracker reports data collected from media and direct government inquiries, among other sources, and include incentive type, cost, and source. These data are the most comprehensive accounting of incentives in the U.S. We note some shortcomings. Consistent with the widely criticized lack of transparency surrounding incentives, the sample may be biased towards large incentives, which receive more media coverage. Jurisdictions vary in the methodologies used to calculate incentives’ reported value. One possible implication is that real incentives are underreported because they tend to be smaller. We have no reason to believe that any measurement error is correlated with other variables of interest.

First, we use our instrumental variable framework to estimate the causal effect of Medicaid stimulus on state government incentive spending during 2009–2011 as our second stage outcome. We also disaggregate incentives into real and tax incentives based on Subsidy Tracker descriptions.¹⁵ An observable implication of our empirical strategy is that real incentive spending grew because states had a limited time frame to spend Medicaid stimulus. Although Fig. 2 indicates high growth of tax incentives, their use should not directly correlate with stimulus. Given our cross-sectional analysis, we cannot control for time-invariant state characteristics including the legal capacity to offer certain types of incentives. We proceed on the plausible assumption that this capacity does not correlate with 2007 state Medicaid spending.

Table 3 summarizes our results. The positive and significant coefficient estimate in Model 1 implies that, on average, an additional $1 million in Medicaid stimulus corresponds to $170,000 more in real incentives. We also estimate the effect on change in incentive spending between 2006–2008 and 2009–2011, which addresses the possibility that high Medicaid spending states are systematically different in their incentive spending. The coefficient on Medicaid stimulus remains positive and significant for real incentives (Model 2).¹⁶ We find no effect of stimulus on tax incentives (Model 3) or change in tax incentives (Model 4).

Table 3 Two-stage least squares regression of investment incentives on medicaid stimulus

We next focus on two type of real incentives that address costly information asymmetries: cash incentives - grants and training costs - and land incentives include firm-specific infrastructure spending and improvements to industrial land and buildings.¹⁷ We find positive and significant effects of both incentive types (Table A6), lending further support to our theory that these subsidies may help foreign firms overcome information asymmetries. Additionally, we regress FDI in old and new counties on investment incentives to demonstrate the positive correlation between FDI in new counties and real incentives (Table A7).

Appendix B reports additional evidence that supports our findings. Content analysis of local newspapers confirms that the FDI projects in our sample received state investment incentives. We also estimate our baseline model with new domestic manufacturing investment, which suffers less from information asymmetries, in place of FDI to show no analogous pattern for domestic firms (Table A8).

5 County-level empirical analysis

In this section, we analyze correlates of new county status. Our empirical setting of a temporary expansion holds constant many standard drivers of firms’ location decisions including infrastructure, labor force, and regulatory climate, which are relatively fixed during the narrow stimulus spending window. Firms’ decision calculus cannot be directly observed but anecdotes suggest that real incentives were important during this period. For example, Mobile County, Alabama lost pipe manufacturers Lakeside Steel (Canada) or Golden Dragon (China) to neighboring counties. Despite boasting multiple foreign plants that would offer potential agglomeration externalities, the county was unable to match cash and land grants offered by the other counties (Amy, 2010).

We estimate a multinomial logit regression of county FDI status with state fixed effects.¹⁸ Counties are classified as new, old, or none corresponding to if and when they received FDI during 2003–2011.¹⁹ Our sample includes 217 new counties and 47 old counties.²⁰ The baseline category is none counties. Relative to old counties, new counties had lower educational attainment and less racial diversity, and were more rural and politically conservative.²¹ County FDI data begin in 2003 so we verify our county classification using 1991–2002 local foreign firm employment data.²² The average of annual median employment in foreign-owned manufacturing plants during 1991–2002 is consistent with our classification.²³

Our main variable of interest is the sum of state spending on incentives in the county during 2009–2011 (inflation-adjusted millions of US dollars). We also control for federal education stimulus received by a county, which could have indirectly funded local government incentives. Recall that Recovery Act education stimulus first went to state governments who further distributed the funds according to existing educational funding formulas. We cannot rule out the possibility that formulas correlate with a county’s propensity to receive FDI or state-funded investment incentives. We capture geographic variation in governors’ motives to fund incentives using county vote margins in the prior gubernatorial election. Governors should be more likely to offer incentives in narrow vote margin countries, defined as a less than ten percentage point difference in the two-party vote share. Additionally, we account for overall political context by controlling for the partisanship of the county’s US Congressional representative and John McCain’s 2008 presidential vote share, reasonable proxies for local partisanship (Gerber & Huber, 2010). Controls for other county-level traits that may affect foreign firms’ location decisions: unemployment rate, working-age population, lagged FDI, lagged domestic investment, and lagged foreign mergers and acquisitions. The latter two variables capture unobserved county characteristics that influence the overall environment for new business. We also control for patents issued in a county, an observable and time-varying proxy for innovation, which may attract FDI. Although our cross-sectional research design precludes controls for time-invariant county characteristics, the short-lived geographic expansion of FDI makes our focus on time-varying county characteristics salient to foreign firms.

Figure 3 illustrates the average marginal effect of each predictor in our county-level model. Real incentives correlate with new county status only. An additional $22,000 in real incentives raises the average probability of new county status by two percentage points. Consistent with our state-level findings, real incentives that address informational asymmetries drive this correlation (Table A12) and tax incentives do not correlate with county status (Table A13).

Fig. 3

The marginal effect of economic and political predictors on the county categorization probabilities. Marginal effect of covariates on the probability of none, old, or new county status with respect to FDI during 2009–2011. Horizontal lines indicate 95% confidence intervals. Coefficients are reported in Table A12

We find that counties with a narrow vote margin in the prior gubernatorial election had a three-percentage point higher average probability of new county status.²⁴ Vote margin does not correlate with none or old county status. This finding is consistent with politicians’ electoral motives to channel investment to gain approval from swing voters. Given that our model already includes real incentives, this variable likely captures politicians’ broader unobserved efforts to attract investment.

One implication of our theoretical framework is that political effort to influence firms’ location decisions is conditional on counties’ underlying capacity to support investment. During our sample period, many foreign firms cited the importance of “shovel-ready” conditions that would allow them to establish production quickly. We capture these conditions with a proxy for idle industrial capacity, the sum of county workers affected by extended mass layoffs during 2000–2006.²⁵ These layoffs often coincide with plant closures, indicating idle capacity such as factories and machinery as well as specialized labor. We add this proxy to our baseline specification. Figure 4 illustrates our findings.²⁶ Mass layoffs increase the probability of new county status only in competitive counties. One interpretation of this finding is that incentives support adapting idle capacity to meet foreign firms’ needs.

Fig. 4

Mass layoffs increase the probability of new county categorization in competitive elections. This figure plots the predicted probability of being categorized as a none, old, and new county depending on mass layoffs and competitiveness of prior gubernatorial election

An additional implication of our framework is that governors seeking re-election have stronger motives to claim credit from swing voters for new investment. Term limits create a partially exogenous source of variation in governors’ motives.²⁷ Figure A9 visualizes the interaction between governors seeking re-election and county vote margin, and the predicted probability of county status. In states whose governors subsequently ran for re-election, narrow vote margin counties had a three percentage point higher probability of being a new FDI recipient relative to states whose incumbent governors did not seek re-election. Within states with governors seeking re-election, competitive counties were 5.5 percentage points more likely to be new as compared to non-competitive counties.

6 Alternative mechanisms

A plausible alternative mechanism for FDI expansion is that local governments in new counties offered real incentives that attracted investment. Typically, local governments have little to offer investors beyond property tax incentives. In our setting, however, local governments were the ultimate recipient of most Recovery Act education stimulus. Figure 3 illustrates a positive correlation between education stimulus and new county status. Local governments may have spent this stimulus in ways that subsidized new investment. Alternately, stimulus may have freed up other local funds to use for incentives.

Two tests help to verify that local government incentives do not drive new county status. We analyze the subset of education stimulus designated for K-12 students with disabilities, which the Recovery Act required states to distribute according to the pre-recession share of county students eligible for federal disability benefits under the 1990 Individuals with Disabilities Education Act. Proceeding on the identifying assumption that county share of these students is otherwise uncorrelated with the propensity to be a new country, we estimate a two-stage least squares regression using the pre-recession share of benefits-eligible students as our instrument. We find no evidence consistent with local governments diverting this stimulus into investment subsidies and the correlation of state-sponsored real incentives remain significant.²⁸ A separate analysis finds that education stimulus spent for community colleges, which often coordinate job training programs, does not correlate with new county status (Table A18).

We address multiple alternative mechanisms through which stimulus expenditures could have made counties more attractive for FDI.²⁹ Infrastructure quality is a high priority in manufacturing firms’ location decisions. Though most stimulus-funded infrastructure projects were not yet operational during the sample period (US Government Accountability Office, 2011), firms may have invested in anticipation of improved infrastructure. We evaluate this mechanism within our county-level empirical framework by controlling for three types of infrastructure stimulus relevant to FDI: highway and bridge improvements; renewable energy; and job training and broadband Internet. For each type, we control for logged stimulus spent in the county during 2009–2011. Our baseline findings remain unchanged (Table A19). Renewable energy FDI grew during the sample, the most notable shift in FDI’s industrial composition. Our results are unchanged if we exclude renewable energy FDI projects (Table A20).

Finally, we address other mechanisms through which the Great Recession may have contributed to FDI’s geographic expansion. Using our county-level framework, we show that foreign firms did not strategically invest in key congressional districts to head off trade restrictions (Table A21), and that corporate inversions, the nominal movement of American companies’ headquarters overseas for tax avoidance, did not produce the illusion of new FDI (Table A22).³⁰ The statistical significance of real incentives changes only modestly if we control for county GDP in 2008 (Table A23). The federal government did not increase FDI promotion during the sample period and generally has limited capacity to directly offer incentives.³¹ The US offers immigration visas in exchange for investment in economically distressed areas, but these investments are not considered FDI.

7 Conclusion

Investment incentives are a controversial policy tool because they redistribute public resources to private firms despite mixed evidence on tangible economic benefits. We leverage unique circumstances of the Great Recession and 2009 Recovery Act stimulus to shed new light on these controversies. Highlighting the distinction between tax and real incentives, and the latter’s potential to resolve information asymmetries, we find that stimulus corresponded to more FDI, much of which went to counties that with little history of FDI. Our findings are consistent with states using stimulus to increase real incentive spending. New FDI recipient counties were more likely to have idle industrial capacity and a narrow vote margin in the prior gubernatorial election, suggesting governors’ electoral motives to offer incentives in counties with higher electoral rewards for attracting new FDI.

Future research can build on these findings by further unpacking how information asymmetries influence firms’ location decisions. Some types of asymmetries may be more costly and circumstances can create new asymmetries. Policymakers could use these insights to deploy incentives more efficiently. Our findings also introduce a new dimension to the long standing puzzle of why politicians offer incentives. We confirm the conventional wisdom that tax incentives are ineffective and show that real incentives can, under certain circumstances, help to attract investment in a relatively cost effective manner. Future research might compare the electoral return to offering large yet ineffective tax incentives versus modest real incentives that actually yield investment.