The Voice of Foreign Direct Investment Foreign Investor Policy Preferences and Experiences in Developing Countries

This paper provides insights to inform government efforts to attract and retain foreign direct investment, by analyzing the results of a survey of more than 2,400 affiliates of multinational enterprises across 10 middle-income countries. The paper explores corporate perspectives and decision-making on countriesâ€™ legal and regulatory environments, political risk, and investment promotion activities. The survey finds that a business-friendly policy environment is critical to multinational enterprisesâ€™ investment decisions, confirming the importance of removing regulatory barriers to foreign direct investment (particularly approval processes), lowering political risks, and having investment promotion agencies. The survey results also show that investors are heterogeneous, with affiliatesâ€™ sectors, trading behaviors, sizes, ages, source countries, and foreign ownership levels affecting their perceptions of and sensitivity to various policy factors. Thus, policy makers should tailor their policy efforts to the needs of priority investor segments. Notably, the analysis consistently finds variation based on the extent to which affiliates import their inputs, suggesting that this relatively understudied topic deserves increased research and policy attention.


Policy Research Working Paper 9425
This paper provides insights to inform government efforts to attract and retain foreign direct investment, by analyzing the results of a survey of more than 2,400 affiliates of multinational enterprises across 10 middle-income countries. The paper explores corporate perspectives and decision-making on countries' legal and regulatory environments, political risk, and investment promotion activities. The survey finds that a business-friendly policy environment is critical to multinational enterprises' investment decisions, confirming the importance of removing regulatory barriers to foreign direct investment (particularly approval processes), lowering political risks, and having investment promotion agencies. The survey results also show that investors are heterogeneous, with affiliates' sectors, trading behaviors, sizes, ages, source countries, and foreign ownership levels affecting their perceptions of and sensitivity to various policy factors. Thus, policy makers should tailor their policy efforts to the needs of priority investor segments. Notably, the analysis consistently finds variation based on the extent to which affiliates import their inputs, suggesting that this relatively understudied topic deserves increased research and policy attention.
This paper is a product of the Finance, Competitiveness and Innovation Global Practice. It is part of a larger effort by the World Bank to provide open access to its research and make a contribution to development policy discussions around the world. Policy Research Working Papers are also posted on the Web at http://www.worldbank.org/prwp. The authors may be contacted at asaurav@worldbank.org and rkuo@worldbank.org.
However, FDI inflows to developing countries have been decreasing since the global financial crisis in 2008-09. From a pre-crisis average of 3 percent of GDP per year, FDI inflows have contracted to less than 2 percent of GDP in recent years. 2 This trend reflects a mix of economic factors, including declining rates of return on FDI, changes in U.S. tax policy, increasingly asset-light forms of international production, and rising trade and investment policy uncertainty. On top of these preexisting trends, the COVID-19 pandemic will further decrease FDI flows.
To drive economic growth and resilience, developing countries are prioritizing efforts to strengthen investment competitiveness to better attract and retain FDI. To that end, policy makers need to identify the drivers of FDI and specific market entry and operational constraints that may deter foreign investors. Via a new survey covering over 2,400 foreign investors across 10 major middle-income countries (MICs), this paper responds to this need by assessing the importance of various policy factors-including FDIrelated regulations, political risk, and investment promotion-as well as studying heterogeneity across different types of MNE affiliates.
The remainder of the paper is structured as follows: Section 2 reviews the literature on policy factors that affect FDI. Section 3 describes the survey data and presents descriptive analysis of variables. Section 4 presents the descriptive statistics. Section 5 details the analytical methodology. Section 6 discusses the results. Finally, Section 7 concludes and discusses policy implications. their companies' business strategies, policy barriers, operational obstacles, and investments in the host economy. Information was collected on general characteristics and investments of the company, the importance and effect of global megatrends on the company's business operations, FDI's contribution to the host economy, and the importance of investment policy factors and operational obstacles faced by foreign-owned affiliates.
The survey was designed to generate results that are representative at the country level and comparable across countries. It targeted a statistically representative sample of existing foreign-owned affiliates across the 10 surveyed MICs. The target was to reach 125 interviews per sector (manufacturing and services). Each country sample comprises roughly 250 affiliates with at least some level of foreign equity ownership and at least five employees. The only exception is Nigeria, where, due to sampling frame limitations, the sample comprises 164 respondents (55 manufacturing and 109 services). Thus, across the 10 target countries, more than 2,400 responses were collected.

DESCRIPTIVE STATISTICS
The GIC 2019 Survey attempted to draw a representative sample of foreign investors across the 10 surveyed emerging markets. This section outlines the profiles of the 2,424 survey respondents.
Sector and subsector: Survey respondents operate across a wide range of sectors. By design, about half of surveyed affiliates were in the manufacturing sector, and about half were in services. Within each sector, the sample covers many subsectors: Machinery and Equipment comprise the largest subsector within manufacturing, and Wholesale and Retail Trade is the largest subsector within services in our sample. However, no one subsector accounts for more than 11 percent of the overall sample (see Table  1).  Employment: About one-quarter of surveyed businesses have more than 250 employees. The remainder are small and medium enterprises (SMEs) with 250 or fewer employees, and roughly half of the SMEs have 100 or fewer employees (Table 4). Investment stock: Roughly one-quarter have invested more than US$10 million in host countries. More than one-tenth have invested more than US$50 million (Table 5). Age: Respondents are generally fairly established in their respective markets. The average respondent has over 17 years of experience operating in its respective host market (Table 6). On average, respondents have over 10 years of experience across all 10 countries.
Trading behavior: Respondents are active in global trade, both in terms of importing into and exporting from host countries. On average, respondents import 46 percent of their inputs, and exports account for 34 percent of their sales (Table 6). There is wide variation in average trading behavior across host countries, with average export share of sales ranging from 15 percent (Brazil) to 53 percent (Vietnam) and average import share of inputs ranging from 33 percent (China) to 58 percent (Mexico and Vietnam).
Ownership. On average, foreign investors hold large shares in affiliates in our sample. The average stake held by foreign investors is 87 percent across the 10 MICs, and within-country averages are above 80 percent in every country except for Nigeria (Table 6). Roughly two-thirds of respondents are 100 percent owned by foreign investors.

METHODOLOGY
We employ simple summary statistics to assess different factors' relative importance to foreign investment decisions. We examine and compare the distribution of responses across questions in a given thematic area (location decision-making criteria, legal and regulatory obstacles, investment protection, or investment promotion). Within each thematic area, questions are phrased in the exact same way, with the same response options; only the factor in question varies. This standardization allows for comparability across factors and questions.
The data also allow us to employ regression specifications to study heterogeneity across investors with respect to individual factors. We explore variation in self-reported data on investment decision-making factors and obstacles with respect to overall sector (i.e., manufacturing versus services), exports' share of sales, imports' share of inputs, source country income group, size of the workforce, investment size, number of years in country, and level of foreign ownership.
As data for several questions are ordinal, we employ ordered logistic regressions for these analyses of heterogeneity. These specifications account for how question responses follow an established order. A factor that is 'critically important' is of greater importance than one that is that is merely 'important', which in turn is more important than a factor considered 'not important at all'. At the same time, unlike OLS regressions, ordered logistic regressions do not assume that the data are cardinal. For example, it might be possible that the distance between 'critically important' and 'important' is different than the distance between 'important' and 'somewhat important' despite these jumps each being one gradation in the question scale.
In particular, we estimate the following equation: for respondent i in subsector j and country c, where * is a latent continuous variable representing a given factor's importance, is a matrix of covariates corresponding to the affiliate-level dimensions of interest (overall sector, exports' share of sales, imports' share of inputs, etc.), is a matrix of coefficients corresponding to , and are subsector and host country fixed effects, respectively, and is an error term with a logistic distribution. 4 We do not observe * directly in the data. Rather, we observe the ordered categorical variable , which takes values according to the following equation: where Min is the minimum value that can take, 1 ,…, −1 are a series of unobserved thresholds, and is the maximum value that can take. The regressions thus estimate 1 ,…, −1 and the parameters in Equation 1.
Except when assessing the effect of being in the broad services sector, our preferred specification is an ordered logistic specification with all affiliate-level covariates, country fixed effects, and subsector fixed effects included (model 8 in all of our regression tables). We also employ OLS regressions using Equation 1 where y* is the observed value of the variable to check the robustness of the results (i.e., treating the dependent variable as cardinal and continuous), although the ordered logistic specification is our preferred specification for the reasons outlined earlier.
To ensure representativeness, analyses contained in the paper incorporate weights to account for different sample sizes across countries, different probabilities of sampling, and bias due to non-response. Design weights are included to ensure that the different strata (country-sector intersections) are given equal weight. Sampling weights are included to account for different probabilities of being sampled, weighting each observation by the inverse probability of selection. Finally, non-response weights are applied to maintain consistency between the distribution of MNE affiliates in the sampling frame and results from the sample along observable characteristics. 5 A key limitation of these approaches is that they are not suitable for robust causal inference. As our data are cross-sectional, self-reported, and cannot be linked to other measures of affiliate behavior and performance (e.g., affiliate-level financial statements), we cannot definitively rule out omitted variables, simultaneity, or reverse causality or derive quantitative estimates of these factors' impact on FDI. Nevertheless, by identifying novel patterns and associations that occur systematically, we can compare our findings with predictions made in the existing literature as well as suggest directions for future research.

Overall determinants of foreign investment decisions
Overall results While survey respondents generally consider political and macroeconomic stability to be the most important factors informing their investment decisions, survey data also confirm that host countries' policy stance towards foreign investors and the private sector is critical. Forty-two percent of respondents consider host countries' legal and regulatory environments to be "critically important" to their decisions to invest in host countries. By this measure, legal and regulatory environments rank ahead of factors such as physical infrastructure and low labor and input costs and behind only political and macroeconomic stability. This finding is in line with the general consensus in the literature that the legal and regulatory environment in host countries plays a critical role in attracting FDI (e.g., Kusek andSilva 2018, Mistura andRoulet 2019). Similarly, 35 percent of respondents consider investor protections to be "critically important". Again, this finds support in the broader literature regarding the importance of political risk in informing foreign investors' decisions (Biglaiser & DeRouen 20106;Krifa-Schneider & Matei 2010).

Figure 1. Overall determinants of investment location decisions (% of respondents)
Source: Computation based on the 2019 GIC Survey.

Heterogeneity across investor types
More export-intensive respondents place lower importance on host countries' legal and regulatory environments. In the preferred ordered logistic specification, the coefficient on export percentage of sales is negative and statistically significant at the p<0.05 level. This finding is robust to the exclusion or inclusion of country and subsector fixed effects as well as alternative specifications using OLS (Table 11, Column 8). A 10-percentage point change in the portion of sales from export is associated with a 1percentage point increase in the likelihood of considering legal and regulatory environments to be 'critically important' (Table 12). Although further research is required to explain this pattern, this coefficient may reflect exporters being less subject to regulations governing sales within their countries of operations (e.g., local consumer protection laws) relative to market-seeking investors. Affiliates with larger workforces place greater importance on host countries' legal and regulatory environments. The regression coefficient on the dummy variable for having more than 250 employees is positive and statistically significant at the p<0.01 level in our preferred specification (Table 11, Column 8). This finding is robust to the inclusion or exclusion of country and subsector fixed effects. In terms of magnitude, analyzing average marginal effects in our ordered logistic regressions reveals that having more than 250 employees is associated with a 10-percentage point increase in the likelihood of finding the legal and regulatory environment to be 'critically important'. Thus, we find support for findings by Aterido, Hallward-Driemeier, and Pages (2007) that regulatory enforcement increases with affiliate size, whereas smaller affiliates generally experience less consistent and laxer implementation of regulations on average.
Having foreign owners from high-income countries may also translate to placing greater importance on host countries' legal and regulatory environments. The coefficient on the dummy for having a foreign owner from a high-income country is positive and marginally significant (p<0.10) in our preferred specification with the importance of host countries' legal and regulatory environment as the dependent variable (Table 11, Column 8). This finding may reflect how investors from developing countries are better able to navigate challenging institutional environments given their firsthand experience with similar environments in their home economies, supporting conclusions by researchers such as Demir and Hu (2016). However, the coefficient in our analysis is only significant in specifications with both country and subsector fixed effects, so further research is required to draw more robust conclusions.
Finally, the survey results suggest that foreign affiliates which import a greater share of their inputs place greater importance on investor protection guarantees. In our preferred specification with the importance of investor protections as the dependent variable, the coefficient on share of sourcing via imports is positive and statistically significant at the p<0.01 level (Table 13, Column 8). This finding is robust to alternative specifications and the inclusion and exclusion of fixed effects. From a marginal effects perspective, a 10-percentage point increase in the portion of inputs sourced via imports is associated with a 1-percentage point increase in considering investor protection guarantees to be 'critically important' (Table 14). Importers may be more sensitive to political risk due to how political risks can impact imports: Relative to companies that source locally, importers are more exposed to sudden, adverse changes in import laws and regulations as well as currency restrictions which may affect their ability to engage in trade. However, definitively explaining this pattern requires further research as the existing literature does not explore the relationship between importing and political risk in great depth.

Overall results
To provide more detail on legal and regulatory barriers to investment, the GIC Survey 2019 asked respondents about the degree to which various legal and regulatory issues presented obstacles to their operations. Such issues covered issues ranging from cumbersome investment approval processes to restrictions on expatriate staff.
Cumbersome investment approval processes and operational restrictions are the most often cited regulatory barriers for FDI in surveyed MICs ( Figure 2). Fifty-six percent of respondents list cumbersome investment approval processes as moderate or major obstacles to operations, and 44 percent cite price, technology, or product restrictions as moderate or major obstacles. The salience of these top two concerns holds across most host countries and sectors. In addition, these findings are consistent with prior work that also finds approval processes and restrictions on prices, technology, or products to be a significant obstacle for foreign affiliates (Mistura and Roulet 2019; UNCTAD 2019).

Heterogeneity across investor types
Compared with manufacturing affiliates, services affiliates perceive joint venture requirements to be more significant obstacles on average, as evidenced by the positive and statistically significant (at the p<0.10 level) coefficient on the dummy of services in our preferred specification (Table 16, Column 6). Affiliates in the services sector also perceive expatriate restrictions to be more significant obstacles on average (see the positive and significant-at the p<0.10 level-coefficient in Table 18, Column 6). Taken together, these findings support explanations by Tatoglu and Glaister (1998) and Erramilli and Rao (1993) that the higher skill content of services impacts FDI decision-making criteria. Affiliates in services may rely more on foreign expertise and brand equity. If foreign parents of affiliates are unable to exercise sufficient control over affiliates, they may be constrained in translating their expertise into operational changes at affiliates, affecting performance as well as brand perceptions. Similarly, being unable to import sufficient foreign talent may limit the degree to which affiliates may tap global expertise and skills. MNCs commonly rely on expatriate staff in affiliate operations for liaison, management capacity and control, and cross-cultural communications (Geng 2003;Seak and Enderwick, 2008;Selmer, 2005), and the role of such expatriates may be especially important in skill-intensive services. However, as the existing literature does not cover this topic in depth, further research is required to definitively explain this pattern in the survey data. Positive signs denote that the row variable is associated with encountering more significant obstacles with respect to the column variable. Signs and significance levels are shown from ordered logistic regressions with all affiliate-level covariates (i.e., row variables) and country and ISIC A38 subsector fixed effects. Blanks denote coefficients that are not statistically significant (i.e., only signs of statistically significant coefficients are shown).

Figure 2. Salience of legal and regulatory obstacles (% of respondents)
Finally, sourcing a higher share of inputs via imports is associated with experiencing more legal and regulatory obstacles on average. Higher import shares of input are associated with experiencing more serious obstacles with respect to joint venture requirements (Table 16, Column 8; significant at the p<0.10 level) as well as foreign investment limits (Table 17, significant at the p<0.05 level). These findings potentially relate to prior research that suggests that imports are associated with higher technological content (Javorcik and Spatareanu 2008). MNEs with more advanced intellectual property prefer full ownership of affiliates to prevent leakages to local partners (Saggi 2002).

Investment protection
Overall results Survey data confirm the importance of investor protections and other measures to minimize political risk in attracting and retaining FDI. To explore the effects of political risk on investments, the GIC Survey 2019 asked respondents how they would respond to hypothetical political risk scenarios. Two in three existing investors would consider withdrawing investments or cancelling planned investment in the face of political risk exposure in host countries ( Figure 3). These findings are consistent with the broader literature on political risk which consistently identifies political risk and regulatory uncertainty as major concerns for foreign businesses. They are also in line with other surveys on the topic. 6 Risks of expropriation and government breach of contract are likely to elicit particularly negative investment reactions, in line with findings by Kusek and Silva (2018). About 50 percent and 40 percent, respectively, of investors would consider withdrawing existing or cancelling planned investments when faced with these risks (Figure 3). Sudden legal changes, currency restrictions, and delays in obtaining permits and approvals elicit somewhat less severe reactions. Such risks are more likely to cause investors to delay investments rather than cancel or withdraw investments completely.

Heterogeneity across investor types
Affiliates with more years of experience in a given host country are likely to be less sensitive to political risk in general. For each respondent, we recorded the most negative reaction across all hypothetical political risks (i.e., we took the maximum value of responses across all questions dealing with individual political risks). In our preferred specification with this composite variable as the dependent variable, the coefficient on number of years in the country is negative and significant at the p<0.10 level and robust to alternative specifications (Table 29). This finding supports Barry and DiGiuseppe's (2019) conclusion that affiliates with better information (i.e., those with more experience within a country) are better able to navigate political risks through local connections and superior knowledge of legal systems.
In contrast, affiliates which import a higher share of inputs are likely to be more sensitive to political risk. The coefficient for import share of sourcing is positive and significant at the p<0.10 level in our preferred specification, although it is not significant in all alternative specifications (Table 29). As discussed earlier, while further research on the topic is needed, this pattern may reflect importers' higher exposure to political risk concerning import and currency restrictions: A greater assortment of policies may impact their operations, meaning higher political risk.

Sudden, adverse law change
Services -* -* Export % of sales Share of sourcing via imports +* High-income source country -* >250 employees -** >10M investment Years in country -* -* -** -* Level foreign ownership +** *** p<0.01, ** p<0.05, * p<0.1 This table displays signs of coefficients for the variables in the rows for regressions with the column variables as the dependent variables. Positive signs denote that the row variable is associated with more negative reactions if the event in the column were to occur. Signs and significance levels are shown from ordered logistic regressions with all affiliate-level covariates (i.e., row variables) and country and ISIC A38 subsector fixed effects. Blanks denote coefficients that are not statistically significant (i.e., only signs of statistically significant coefficients are shown).
Respondents also exhibit heterogeneity in their sensitivity to individual political risks. Affiliates with more years of experience in host countries would react less negatively to expropriation, government breach of contract, and currency restrictions, as evidenced by the negative and statistically significant coefficient (at least at the p<0.10 level) on years in country in our preferred specification (Table 33, Column 8). As discussed previously, this pattern may reflect more experienced affiliates' greater knowledge and connections within host countries (Barry and DiGiuseppe 2019).
Similarly, affiliates in services would react less negatively to expropriation on average, as evidenced by the negative and statistically significant (at the p<0.10 level) in our preferred specification (Table 33, Column 6). Our findings could be driven by selection issues related to how expropriation is somewhat more common in services (i.e., the investors that choose to invest are those that are less sensitive to expropriation risks) (Hajzler 2012). Nevertheless, more research is required to come to a more robust explanation as this topic has received little research attention to date.
In contrast, affiliates with higher foreign ownership shares would react more negatively to expropriation on average, as evidenced by the positive and significant (p<0.05) coefficient on foreign ownership share in our preferred specification (Table 33, Column 8). This finding may reflect high foreign ownership shares being associated with more advanced technologies, which may in turn make affiliates more sensitive to expropriation of their intellectual property (Saggi 2002). It may also reflect how domestic owners of affiliates which are joint ventures (i.e., only part foreign-owned) may be less likely to leave their home markets, even in the face of expropriation.
If faced with delays in permits and approvals, affiliates with foreign owners in high-income countries are likely to react more negatively on average. In our preferred specification, the coefficients on the dummy for high-income source country is negative and statistically significant at the p<0.10 level, and the results are robust to alternative specifications (Table 30). Given that such delays are generally more common in developing country contexts, this pattern may reflect how investors from developing countries are better able to navigate routine regulatory challenges in other developing economies given their experiences with their home markets (Demir and Hu 2016). That this pattern does not exist for the other political risks likely reflects the relative seriousness of risks; even developing country investors' prior experience may be insufficient to ameliorate the impacts of more serious risks like expropriation.
Finally, affiliates in the services sector are more likely to react negatively to delays in permits and approvals. The coefficient on the dummy for services is negative and statistically significant at least at the p<0.10 level across all specifications, including our preferred ordered logistic specification with fixed effects (Table 30). This pattern may reflect the greater prevalence of subjective screening and approval requirements related to services FDI relative to manufacturing FDI, as documented by the OECD's FDI Regulatory Restrictiveness Index and Golub (2009). In this context of greater bureaucratic discretion, delays for FDI in services may increase uncertainty regarding whether approvals will ever be granted, whereas delays for manufacturing investments may be more likely to reflect simple processing delays.

Overall results
Our results confirm findings in the literature that IPAs play an important role in attracting and retaining FDI. Almost 90 percent of existing investors value at least one IPA service, and most IPA services are considered important or critically important by at least two-thirds of MNE affiliates in surveyed emerging markets ( Figure 4). Indeed, respondents widely value both economy-wide advocacy efforts as well as project-specific support: Three quarters of investors value IPA efforts toward improving the overall business environment in the country. About seven in ten investors value IPA provision of preinvestment information such as location guides, sector and project profiles, and regulatory procedures. A similar share of investors value IPA assistance in setting up operations to fulfill registration requirements and obtain entry permits, among others. These findings are broadly consistent with findings in Kusek and Silva (2018), where advocacy for improved business environment and business setup assistance ranked among the most valued IPA services. 7

Figure 4. Importance of IPA services
Source: Computation based on the 2019 GIC Survey.

Heterogeneity across investor types
On average, importing a higher share of inputs is associated with placing greater importance on IPA services. When the importance of affiliates' most-valued IPA service is used as the dependent variable, the coefficient on import share of inputs is positive and significant at the p<0.10 level in our preferred ordered logistic specification with country and subsector fixed effects (Table 41). This result is also robust to the exclusion of fixed effects and alternate specifications using OLS. Looking across specific services, we find similar results with respect to IPA assistance with business setup (Table 46, p<0.05) and with operational issues and grievances (Table 47, p<0.10). Both of these results are also robust to alternative specifications. Importers may value IPA services more insofar as IPAs help them navigate issues related to obtaining import licenses and addressing grievances concerning import processes, although further research is necessary to fully explain these results. 7 Assistance with resolving operational issues and grievances was the most valued IPA service in the 2017 GIC Survey, compared to the fourth-ranked service in the 2019 survey. Nevertheless, the 2019 survey continues to show a sizable majority (67 percent  Positive signs denote that the row variable is associated with placing greater importance on the column variable. Signs and significance levels are shown from ordered logistic regressions with all affiliate-level covariates (i.e., row variables) and country and ISIC A38 subsector fixed effects. Blanks denote coefficients that are not statistically significant (i.e., only signs of statistically significant coefficients are shown).
Having fewer years of experience in host countries is also associated with placing greater importance on IPA services. In our preferred specification, with importance of the most important IPA service as the dependent variable, the coefficient on number of years in country is negative and significant at the p<0.05 level, although this result is only significant in ordered logistic specifications with country fixed effects. In terms of specific IPA services, having fewer years of experience is also associated with placing greater importance on personalized contact and response (Table 43, p<0.05), pre-investment assistance such as site visits (Table 45, p<0.10), and efforts to improve the host country's business environment (Table 48, p<0.10). This finding likely reflects more experienced affiliates having more in-country knowledge and connections and therefore less need for IPA services.
Having foreign parents in high-income countries is associated with placing less importance on select IPA services. The coefficient on the dummy variable for having a foreign owner from a high-income country is not significant in regressions with the importance of the most important IPA service as the dependent variable. However, the coefficient on the dummy for high-income source country is negative and significant in regressions where the dependent variable is the importance of pre-investment assistance (Table 45) or assistance with operational issues and grievances (Table 47). This may reflect how investors from high-income source countries may have access to greater private (e.g., consultancies) or diplomatic resources (e.g., embassy pressure) to deal with such issues.
In contrast, large investment sizes are associated with placing greater importance on select IPA services. The coefficient on the dummy variable for having over USD 10 million in total investment is positive and significant in regressions with the importance of promotion of investment opportunities (Table 42), personalized contact and responsiveness (Table 43), and pre-investment assistance (Table 45) as dependent variables. Large, systemically important affiliates may value personalized assistance tailored towards their specific needs, whereas smaller companies may not expect similar support from IPAs.

CONCLUSION: POLICY IMPLICATIONS AND DIRECTIONS FOR FUTURE RESEARCH
The stark decline in FDI resulting from the COVID-19 pandemic means that policy makers must redouble their efforts to attract FDI in order to safeguard FDI's role in driving jobs and economic transformation. This paper's findings contribute to the literature informing such efforts by reaffirming the importance of various policy dimensions in creating a conducive environment for FDI and by providing suggestive evidence to inform efforts to tailor policy prescriptions to target investors.
Our survey findings confirm that countries' policy environments are critical drivers of foreign investment: • Legal and regulatory barriers to FDI: Regulatory barriers, especially investment approval processes, commonly obstruct MNE affiliates' operations. Policy makers should thus renew their focus on making business-friendly reforms by simplifying approval processes and removing limits on foreign investment, among other areas. • Political risk and investor protections: Investors are likely to delay, cancel, or even withdraw investments when faced with political risks such as expropriation and sudden changes in laws and regulations. These findings reaffirm the importance of government efforts to strengthen institutions and property rights, and financial products to manage against such risks (e.g., MIGA Breach of Contract coverage). • IPAs: We find that foreign investors value a broad range IPA services spanning the entire investment lifecycle. Governments seeking to boost FDI should thus ensure that IPA services are tailored to foreign investors' needs, including post-entry issues such as business environment advocacy and support with grievances.
We also consistently find evidence that foreign investors' characteristics affect the degree to which they encounter policy-related obstacles and value IPA services. These differences between types of investors suggest that governments should prioritize and tailor reforms to most closely match their target segments for FDI. Notably, in addition to findings about the importance of affiliates' export behavior and sector-which have been covered in the literature to some degree-we consistently find evidence that affiliates' import intensity is an important factor shaping their policy experiences and preferences. To the extent that such affiliates are important (e.g., for GVC participation and introducing new technologies to host countries), governments should investigate and address why import-intensive affiliates commonly experience greater policy-related obstacles (e.g., eliminating policy obstacles related to import licensing and processing).
Beyond specific policy areas, we find that both the content and implementation of policies matter. Due to the divergence between regulatory provisions and their implementation, foreign investors must navigate the bureaucratic landscape in host economies. Thus, 'on-paper' policy reforms to ease entry, strengthen property rights, and ensure non-discriminatory treatment of foreign investors may not be adequate. Governments should also invest in agency staff capacity and install transparent, efficient, and simple processes to break down siloes and reduce ambiguities that may give rise to bureaucratic discretion.
Finally, our findings suggest new directions for future research on the relationship between FDI and government policy levers. While this paper provides suggestive evidence regarding sources of heterogeneity in foreign investors' policy experiences and preferences, it is insufficient to draw causal conclusions. More robust research leveraging more granular administrative data sets of FDI would thus be valuable for expanding our understanding of such heterogeneity. As such data sets are rare and more common in high-income countries, efforts to collect and publish granular FDI data (i.e., comprehensive data on MNE affiliate activities in a country) are vital. Finally, as the implications of COVID-19 become clearer and new data are released, future research to understand whether investor preferences have changed post-pandemic will also be critical. Table 11. Importance of legal and regulatory environment The dependent variable in all models is an ordered categorical variable corresponding to responses to the following question: "How important were the following factors in your foreign owner company's decision to invest in this country? [Business-friendly legal and regulatory environment]". Higher numbers denote greater importance. Subsector fixed effects are at the ISIC A38 level. Coefficients for constants (for OLS), cut points (for ordered logit), and fixed effects not shown in table. The numbers in the table represent average marginal effects (i.e., marginal change in probability) on the dependent variable taking the value of 'critically important' associated with a one-unit change in the row variable. The dependent variable in all models is an ordered categorical variable corresponding to responses to the following question: "How important were the following factors in your parent company's decision to invest in this country? [Businessfriendly legal and regulatory environment]". Higher numbers denote greater importance. Average marginal effects are not estimable in models with subsector fixed effects due to the small number of observations for select subsectors. Marginal effects fixed effects are not shown in the table. The dependent variable in all models is an ordered categorical variable corresponding to responses to the following question: "How important were the following factors in your foreign owner company's decision to invest in this country? [Investment protection guarantees against expropriation and other risks (e.g. repatriating profits, currency transfers, discrimination, and breach of contract by the government)]". Higher numbers denote greater importance. Subsector fixed effects are at the ISIC A38 level. Coefficients for constants (for OLS), cut points (for ordered logit), and fixed effects not shown in table. The numbers in the table represent average marginal effects (i.e., marginal change in probability) on the dependent variable taking the value of 'critically important' associated with a one-unit change in the row variable. The dependent variable in all models is an ordered categorical variable corresponding to responses to the following question: "How important were the following factors in your parent company's decision to invest in this country? [Investment protection guarantees against expropriation and other risks (e.g. repatriating profits, currency transfers, discrimination, and breach of contract by the government)]". Higher numbers denote greater importance. Average marginal effects are not estimable in models with subsector fixed effects due to the small number of observations for select subsectors.         The dependent variable in all models is an ordered categorical variable corresponding to responses to the following question: "In general, how much of an obstacle were the following factors in this country? [Complexity of administrative procedures]". Higher numbers denote more significant obstacles. Subsector fixed effects are at the ISIC A38 level. Coefficients for constants (for OLS), cut points (for ordered logit), and fixed effects not shown in table.  The dependent variable in all models is an ordered categorical variable corresponding to responses to the following question: "In general, how much of an obstacle were the following factors in this country? [Coordination between public agencies]". Higher numbers denote more significant obstacles. Subsector fixed effects are at the ISIC A38 level. Coefficients for constants (for OLS), cut points (for ordered logit), and fixed effects not shown in table. The dependent variable in all models is an ordered categorical variable corresponding to responses to the following question: "In general, how much of an obstacle were the following factors in this country? [Discretion of the bureaucracy]". Higher numbers denote more significant obstacles. Subsector fixed effects are at the ISIC A38 level. Coefficients for constants (for OLS), cut points (for ordered logit), and fixed effects not shown in table. The dependent variable in all models is an ordered categorical variable corresponding to responses to the following question: "In general, how much of an obstacle were the following factors in this country? [Quality of laws and regulations]". Higher numbers denote more significant obstacles. Subsector fixed effects are at the ISIC A38 level. Coefficients for constants (for OLS), cut points (for ordered logit), and fixed effects not shown in table. The dependent variable in all models is an ordered categorical variable corresponding to responses to the following question: "In general, how much of an obstacle were the following factors in this country? [Accessibility of laws and regulations]". Higher numbers denote more significant obstacles. Subsector fixed effects are at the ISIC A38 level. Coefficients for constants (for OLS), cut points (for ordered logit), and fixed effects not shown in table. The dependent variable in all models is a composite ordered categorical variable constructed from responses to the following series of questions: "How would the following situations affect your investments in this? Please tell us whether they would cause you to significantly delay an investment, to cancel a planned investment, to withdraw an existing investment, or to consider a delay or cancellation." Higher numbers denote more adverse reactions. Subsector fixed effects are at the ISIC A38 level. Coefficients for constants (for OLS), cut points (for ordered logit), and fixed effects not shown in table. The dependent variable in all models is an ordered categorical variable corresponding to responses to the following question: "How would the following situations affect your investments in this? Please tell us whether they would cause you to significantly delay an investment, to cancel a planned investment, to withdraw an existing investment, or to consider a delay or cancellation. [Delays in obtaining necessary permits and approvals to start or operate a business]". Higher numbers denote more adverse reactions. Subsector fixed effects are at the ISIC A38 level. Coefficients for constants (for OLS), cut points (for ordered logit), and fixed effects not shown in table. The dependent variable in all models is an ordered categorical variable corresponding to responses to the following question: "How would the following situations affect your investments in this? Please tell us whether they would cause you to significantly delay an investment, to cancel a planned investment, to withdraw an existing investment, or to consider a delay or cancellation. [Restrictions on your ability to transfer and convert currency]". Higher numbers denote more adverse reactions. Subsector fixed effects are at the ISIC A38 level. Coefficients for constants (for OLS), cut points (for ordered logit), and fixed effects not shown in table. The dependent variable in all models is an ordered categorical variable corresponding to responses to the following question: "How would the following situations affect your investments in this? Please tell us whether they would cause you to significantly delay an investment, to cancel a planned investment, to withdraw an existing investment, or to consider a delay or cancellation. [Breach of contract by the government]". Higher numbers denote more adverse reactions. Subsector fixed effects are at the ISIC A38 level. Coefficients for constants (for OLS), cut points (for ordered logit), and fixed effects not shown in table. The dependent variable in all models is an ordered categorical variable corresponding to responses to the following question: "How would the following situations affect your investments in this? Please tell us whether they would cause you to significantly delay an investment, to cancel a planned investment, to withdraw an existing investment, or to consider a delay or cancellation. [Expropriation or taking of your property or assets by the government]". Higher numbers denote more adverse reactions. Subsector fixed effects are at the ISIC A38 level. Coefficients for constants (for OLS), cut points (for ordered logit), and fixed effects not shown in table. The dependent variable in all models is an ordered categorical variable corresponding to responses to the following question: "How would the following situations affect your investments in this? Please tell us whether they would cause you to significantly delay an investment, to cancel a planned investment, to withdraw an existing investment, or to consider a delay or cancellation. [Sudden change in the laws and regulations with a negative impact on your company]". Higher numbers denote more adverse reactions. Subsector fixed effects are at the ISIC A38 level. Coefficients for constants (for OLS), cut points (for ordered logit), and fixed effects not shown in table. The dependent variable in all models is an ordered categorical variable corresponding to responses to the following question: "In general, how much of an obstacle were the following factors in this country? [Complexity of administrative procedures]". Higher numbers denote more significant obstacles. Subsector fixed effects are at the ISIC A38 level. Coefficients for constants (for OLS), cut points (for ordered logit), and fixed effects not shown in table. The dependent variable in all models is an ordered categorical variable corresponding to responses to the following question: "In general, how much of an obstacle were the following factors in this country? [Capacity of public agencies]". Higher numbers denote more significant obstacles. Subsector fixed effects are at the ISIC A38 level. Coefficients for constants (for OLS), cut points (for ordered logit), and fixed effects not shown in table. The dependent variable in all models is an ordered categorical variable corresponding to responses to the following question: "In general, how much of an obstacle were the following factors in this country? [Coordination between public agencies]". Higher numbers denote more significant obstacles. Subsector fixed effects are at the ISIC A38 level. Coefficients for constants (for OLS), cut points (for ordered logit), and fixed effects not shown in table. The dependent variable in all models is an ordered categorical variable corresponding to responses to the following question: "In general, how much of an obstacle were the following factors in this country? [Discretion of the bureaucracy]". Higher numbers denote more significant obstacles. Subsector fixed effects are at the ISIC A38 level. Coefficients for constants (for OLS), cut points (for ordered logit), and fixed effects not shown in table.  The dependent variable in all models is an ordered categorical variable corresponding to responses to the following question: "In general, how much of an obstacle were the following factors in this country? [Accessibility of laws and regulations]". Higher numbers denote more significant obstacles. Subsector fixed effects are at the ISIC A38 level. Coefficients for constants (for OLS), cut points (for ordered logit), and fixed effects not shown in table.    The dependent variable in all models is an ordered categorical variable corresponding to responses to the following question: "How important are the following services offered by Investment Promotion Agencies to your company in this country? Please rate each on a scale from 1 to 4 where 1 is not at all important and 4 is critically important. [Pre-investment Information (e.g. location's guide, sector and project profiles, regulatory procedures)]". Higher numbers denote more greater importance. Subsector fixed effects are at the ISIC A38 level. Coefficients for constants (for OLS), cut points (for ordered logit), and fixed effects not shown in table. The dependent variable in all models is an ordered categorical variable corresponding to responses to the following question: "How important are the following services offered by Investment Promotion Agencies to your company in this country? Please rate each on a scale from 1 to 4 where 1 is not at all important and 4 is critically important. [Pre-investment assistance (e.g. site visits, briefings, meetings with stakeholders)]". Higher numbers denote more greater importance. Subsector fixed effects are at the ISIC A38 level. Coefficients for constants (for OLS), cut points (for ordered logit), and fixed effects not shown in table. The dependent variable in all models is an ordered categorical variable corresponding to responses to the following question: "How important are the following services offered by Investment Promotion Agencies to your company in this country? Please rate each on a scale from 1 to 4 where 1 is not at all important and 4 is critically important. [Assistance in setting up of business (e.g. registration requirements and entry permits)]". Higher numbers denote more greater importance. Subsector fixed effects are at the ISIC A38 level. Coefficients for constants (for OLS), cut points (for ordered logit), and fixed effects not shown in table. The dependent variable in all models is an ordered categorical variable corresponding to responses to the following question: "How important are the following services offered by Investment Promotion Agencies to your company in this country? Please rate each on a scale from 1 to 4 where 1 is not at all important and 4 is critically important. [Assistance with operational issues and grievances]". Higher numbers denote more greater importance. Subsector fixed effects are at the ISIC A38 level. Coefficients for constants (for OLS), cut points (for ordered logit), and fixed effects not shown in table.