Is ‘no news’ really ‘good news’? Country visibility and FDI location choice

Abstract

In choosing where to invest, firms seek out information on a set of possible locations. Information asymmetries may make country visibility particularly important in decisions to locate investment abroad. We develop a country visibility index based on international news stories in The Economist, and show that broad country visibility is at least as important in attracting foreign direct investment (FDI) as other specific investment promotion activities or proxies for information frictions. Controlling for standard gravity model determinants of FDI, we find that greater visibility of developing countries, in particular lower middle- and low-income countries, increases the investment that they receive from US multinational corporations.

Keywords:

• Foreign direct investment
• information asymmetries
• international news coverage

JEL Classification:

• F21
• F23
• D83

The good news is that transparency can work. When information is relevant, standardised and public, it fosters intelligent decision-making.

(The Economist, Economics Focus, 19 February 2009)

1. Introduction

International trade and capital flows have become increasingly important in today’s integrated world economy. As one particularly vital type of capital flow, foreign direct investment (FDI) has been seen as both a response to, and a cause of, globalization and growth. In choosing where to invest, firms take into account a wide variety of factors, including the expected costs and benefits of investing in a particular location. To assess the expected costs and benefits, investors seek out information on a set of possible locations. Information asymmetries may exist, however, that make such assessments difficult. These information asymmetries are thought to be more extreme at the international level than at the domestic level. Thus, publicly available information is likely to play a particularly important role in decisions to assess markets for locating investment abroad. News stories, in particular, may raise awareness of a given country, thus leading managers to allocate more market research to that country, ultimately increasing that country’s chances of receiving FDI.

To examine this question, whether country visibility in the news increases FDI, we employ a country visibility index based on international news articles in The Economist. This visibility index provides a measure of the frequency with which countries appear in the news. We estimate the impact of country visibility on FDI flows from the US using a gravity model approach that accounts for a wide range of standard determinants of FDI flows. We assume that firm managers in the US choose where to locate foreign investment abroad and gather information on the expected costs and benefits of investing in particular countries. This information gathering, however, is costly and is typically limited to a short list of countries. Thus, the question remains: What gets a country onto the short list? Greater awareness of countries can certainly place them in the running for the short list. The visibility index that we employ attempts to measure this awareness. While not a direct measure of information, this index may provide an adequate proxy for country visibility to decision-makers in US multinational corporations (MNCs) given the wide readership of The Economist among managers and professionals in North America.

Taking the visibility measure to the data, we explore the impact of greater visibility on FDI flows. Ideally, we would like to measure directly the investment decision by managers in order to test whether the visibility of a particular country increases the probability of being selected as an investment location for FDI. Since we do not have data on this direct investment decision, we infer this choice using the amount of investment flows at the aggregate level. Thus, the specific hypothesis tested is whether greater country visibility, as captured by a higher count of news stories over the year, has a positive impact on international capital flows through increased awareness about potential FDI location sites.

Conditioning on a variety of widely used explanatory variables for FDI flows, we find that country visibility plays a significant role as an additional determinant of FDI flows. In particular, developing countries significantly benefit from greater visibility. Conditional on a set of covariates shown to be important in previous literature, we find that those developing countries that receive greater news coverage in The Economist receive greater FDI from US MNCs. Rather than ‘no news is good news’, therefore, our findings suggest that ‘more news is good news’ for developing countries.

To ensure that our findings are not contaminated by endogeneity, we also employ a set of instrumental variables (IVs). The IV results are strongly supportive of our baseline findings. Overall, our baseline and IV results suggest that country visibility is at least as important as that found in previous studies for investment promotion agencies (IPAs) (see Harding and Javorcik 2011) or using distance or other proxies of information frictions (see Daude and Fratzscher 2008).

Our paper contributes to the broad literature that examines the determinants of FDI capital flows across countries. It is most closely related to recent work that highlights the importance of information and information asymmetries in international capital flows. Among these studies, Harding and Javorcik (2011) examine the role that IPAs play in attracting FDI to a country. Using survey data on investment promotion activities at the sectoral level, they show that investment promotion activities aimed at particular sectors are associated with higher FDI inflows from the US. Their findings suggest that IPAs increase the amount of FDI in particular to developing countries, where information is usually harder to obtain. Focusing on another information channel, Javorcik et al. (2011) examine how migrant networks in the US affect the stock of US FDI in their country of origin. They employ a large set of IVs in this paper to address the potential for endogeneity. Their estimation results suggest that diasporas can significantly contribute to the FDI in their country of origin by serving as information channels about investment opportunities abroad.

Recent literature has augmented the standard set of FDI determinants to highlight the effects of information asymmetries on both FDI and other forms of capital flows. Portes, Rey, and Oh (2001) and Portes and Rey (2005) show that a gravity model (which relies on size and distance as explanatory variables) explains international trade in financial assets at least as well as trade in goods. Applied to trade in goods, distance has been used in the gravity model as a proxy for transportation costs. Applied to trade in assets, however, distance may be a proxy for informational frictions. Portes, Rey, and Oh (2001) concentrate on portfolio equity trade, corporate bonds, and treasury bonds while Portes and Rey (2005) concentrate on portfolio equity flows, with both papers finding a significant negative relationship between distance and trade in financial assets. Also, Daude and Fratzscher (2008) use information and host country institutional variables to explain the composition of capital flows across borders, focusing on bilateral stocks of FDI, portfolio equity and debt, and loans for 77 countries. They use geographic distance, the amount of telephone traffic between countries, and trade in newspapers as alternative proxies for information frictions in their study of the composition of cross-border flows. Their results indicate a pecking order of capital such that information frictions play a larger role for FDI and loans than for portfolio holdings. In particular, they find that distance (their primary proxy for information) has a larger impact on FDI stocks than on portfolio equity or debt securities.

Our specification focuses explicitly on FDI flows, rather than on portfolio equity or debt, and we utilize a different proxy for information than those used by the above studies. The count of news stories allows us to generate a visibility index that is unrelated to geographic distance. While geographic distance may proxy some forms of informational frictions, these frictions may be overcome somewhat by media attention. We measure media attention directly and assess its impact on FDI flows while also controlling for the distance between countries.

Utilizing the idea that news conveys information pertinent to investment decisions, Veldkamp (2006, 22) argues that ‘when countries appear in the news more often, they should have higher asset prices, on average, because the information conveyed in the news stories decreases the investment risk’. Similarly, Barber and Odean (2008) focus on the stock purchase decision of individual investors and show that attention determines the choice set while preferences determine the choice. Here, attention determines the choice set of where to locate FDI while fundamentals determine the choice actually taken. Thus, in our regression equation, we take into account the more traditional trade determinants of FDI, including variables of size and distance as indicated by standard gravity models, along with institutional variables as indicated by the most recent approaches, while focusing on the impact of news counts on FDI flows.

In Section 2, we describe the data, with particular attention on how we form the country visibility index. We also set up our basic econometric approach, describing the underlying regression function. In Section 3, we present the baseline estimation results along with a range of robustness tests, including the incorporation of IVs to address issues of endogeneity. Section 4 concludes.

2. Data and empirical approach

In this section, we first describe our approach in measuring country visibility, which is the independent variable of interest in this paper. Then, we describe our measure for FDI flows, which is the dependent variable of interest. We also provide our basic econometric approach and justify our model selection by describing the control covariates, which include an augmented set of controls in addition to those from the standard gravity approach applied to FDI flow. We then conclude this section by describing our identification strategy.

2.1. The country visibility index

The country visibility index is based on a count of all stories in the US print editions of The Economist during a given year. We have collected these data for the years 2002 and 2003. We create this index by aggregating the number of articles on a specific country for each year.

The Economist provides comprehensive business and financial analysis, along with political commentary. This magazine is a useful media outlet for the purpose of this research. First, it has wide readership in the US. The total average paid circulation of the North American edition of The Economist, most of which is read in the US, reached about 403,000 copies during the six months ended in June 2002 and about 437,000 copies during the six months ended in December 2003 (Audit Bureau of Circulation 2002, 2003). These figures make up approximately half of The Economist’s total audience in 2002 and 2003.¹

Second, The Economist is likely to be read by the decision-makers in US MNCs. In 2002 and 2003, the median household income of The Economist’s audience is estimated to be about $110,000. About 60% of its readers identify themselves as Managers or Professionals.² In particular, about 20% of its readers hold Top Managerial Positions in 2002 and 2003.³ As the president of The Economist Group Media Businesses, Paul Rossi, indicates:

[The Economist has] reached CEOs and politicians and financiers around the world. And that is, in some sense, an aspiration that we promote in our advertising. So, one of our tag lines is: It’s lonely at the top, but at least there’s something to read. (Langfitt 2006)

Although this magazine is widely read by decision-makers in the US, we do not assume that the management body of all US MNCs is reading The Economist. Rather, the news reported by The Economist indicates the type and focus of stories that are available to US managers, who may be utilizing a wide range of news sources. Relative to other countries, the number of news stories about a given country in The Economist is an indicator of the general media focus on that country in each year.

Third, The Economist is a weekly magazine, which reduces the chance of double counting the same news event. A daily newspaper, such as the Wall Street Journal and the Financial Times, may provide an alternative visibility index but would be subject to a greater chance of double counting since a single news event may be reported for subsequent days in a row. Using a composite site, such as LexisNexis, would also entail multiple counts of the same news story as multiple news sources may report the same story at the same time.

We create a quantitative measure of country visibility by counting the number of articles on a specific country in each issue. We then add these up over the year in order to match the country visibility data to annual flows of direct investment abroad. We determine the primary country focus (and secondary if the article content appears to be evenly split) by reading the title and the body of each article. We count stories from all sections, except the US section. If the article does not have a clear country focus, we count the story but do not assign it to a particular country. Counting the number of articles on a specific country in this fashion enables us to precisely identify the country focus of each article, which minimizes the possibility of measurement error in the independent variable of interest.

We also categorize each story according to its main focus: economic, political, or social. Economic stories are categorized based on providing concrete economic data, descriptions of economic systems, clear discussion of economic policies, or the effects of economics on particular industries or firms. Political stories are categorized based on descriptions of political parties, elections, or politicians. Social stories are any that are not categorized as economic or political, but largely consist of stories on society at large.

This approach in data collection may greatly improve our measurement of country visibility. However, it has its own shortcoming. For this study, we survey two years of publications thoroughly, which results in a well-measured set of data but limits the time dimension of the collected data-set. We find it costly to increase the time dimension while keeping the quality of measurement intact. Given this trade-off, we emphasize better measurement to avoid any bias caused by measurement error, but the resulting data-set has a relatively small time dimension.

We use the count of stories to capture the broad visibility of countries available in a public forum. While it may seem useful to categorize stories by whether there is a positive or negative emphasis for a country, we view such judgment as purely subjective. Further, such judgment may be firm or industry specific. For example, an article highlighting that a country’s GDP has increased may be read as positive as it indicates faster economic growth, leading to higher firm profits from FDI as consumers have greater incomes to spend. Such an increase in GDP, however, could be read as negative as the cost of inputs, such as wages, may also rise. Further, the assessment of news may depend on the expectations previously held by investors. A reported increase in GDP may be below the expected increase and thus be characterized by an investor as negative while we would erroneously categorize it as a positive news story. Political changes in a country make this judgment particularly difficult. We include examples of stories and their categorization in Appendix 1 to highlight the difficulty in judging the type of news presented.

Consequently, we retain a focus on the count of stories, where the visibility index is a broad measure used to indicate the availability of information on a country. The way we handle the FDI capital flows may alleviate the concern over being unable to capture the positive/negative content of a story. We focus on direct investment abroad by US firms, utilizing Bureau of Economic Analysis (BEA) data on FDI. In particular, we use the values the BEA reports as capital flows.⁴ Measured FDI flows may be positive (indicating investment abroad) or negative (indicating the reversal of previous flows). In the regression analysis, we use the absolute value of capital flows as the dependent variable. As independent variables, along with the lagged number of news counts, we include a dummy variable for negative capital flows and an interaction term between the negative flow dummy and the log of the lagged number of news counts. Although the dependent variable is measured in absolute value, we are able to test whether the negative capital flows respond differently to the country visibility index, potentially capturing the effect from more favorable vs. less favorable news on a country.

We view the count of stories as an indicator for awareness of a country, rather than providing fundamental business statistics or cost comparisons across countries. Thus, it makes sense to think of the visibility index as helping place a country on the short-list for consideration for investment. Once on the short-list, the firm would then gather additional data on the costs and benefits of locating production in the country. Our approach is similar to that in Veldkamp (2006, 19), who uses the number of Financial Times stories on 23 emerging markets as ‘a proxy for the extent to which information about a market is easily accessible from any number of high-demand, low-cost information sources’. She finds that high news levels are associated with high asset price levels. While Veldkamp associates the number of news stories with a country’s asset prices, we associate our measure with FDI capital flows. Further, our news variable plays a similar role to that in Barber and Odean (2008), who examine stock purchasing behavior, by focusing attention on a particular country. Greater attention means that a country is more likely to appear in the investor’s choice set and thus is more likely to be chosen for FDI than a country that does not appear in the choice set.

Tables 1a and 1b provides descriptive statistics of the country visibility index both with and without the UK included in the sample. For this table we only include those countries that are used in our baseline regressions. Thus, the US is excluded. The UK is over-sampled since each story in the Britain regional section is counted as an individual story about the UK. Thus, all regressions have been estimated both with and without the UK included in the sample. After the UK, the countries receiving the next highest values for count for 2002 are Japan (146 stories), Germany (131 stories), France (105 stories), China (97 stories), India (59 stories), Israel (59 stories), and Italy (59 stories), and for 2003 are Iraq (169 stories), Japan (131 stories), Germany (111 stories), China (105 stories), France (87 stories), India (73 stories), and Canada (66 stories).⁵

Table 1a. Country visibility index, summary statistics.

	Variable	Mean	SD	Median	Min	Max
2002	News Count	14.65	42.5	3	0	455
153 countries	Economic News Count	6.65	19.52	0	0	172
	Political News Count	4.08	10.81	1	0	113
	Social News Count	3.86	14.25	1	0	168
2003	News Count	14.56	49.78	2	0	535
138 countries	Economic News Count	4.51	14.14	0	0	120
	Political News Count	2.27	6.02	0	0	58
	Social News Count	7.74	31.47	1	0	356

Note:

The countries that are used in the baseline regressions for 2003 (153 countries) and 2004 (138 countries) are included. The US is excluded.

Table 1b. Country visibility index, summary statistics – excluding the UK.

	Variable	Mean	SD	Median	Min	Max
2002	News Count	11.76	22.92	3	0	146
152 countries	Economic News Count	5.56	14.18	0	0	112
	Political News Count	3.37	6.21	0	0	41
	Social News Count	2.78	5	1	0	32
2003	News Count	10.77	22.15	2	0	131
137 countries	Economic News Count	3.66	10.13	0	0	82
	Political News Count	1.86	3.68	0	0	18
	Social News Count	5.20	9.97	1	0	57

Note:

Excluding the UK, other countries that are used in the baseline regressions for 2003 (152 countries) and 2004 (137 countries) are included. The US is excluded.

The variable News Count indicates the overall count of articles on a particular country during 2002 and 2003. We employ log(News Count + 1) for each country as our country visibility index, where we have added one to all observations before taking logs. If a country does not appear in The Economist, it is given a count of zero, indicating zero visibility via this source, allowing us the greatest country coverage.

Using news stories to characterize awareness is certainly not new. What is new, however, is the application specifically to FDI flows. For example, the previous literature has linked news stories to investors’ trading behavior (Nofsinger 2001; Barber and Odean 2008), stock market trading volume (Onder and Simga-Mugan 2006), volatility in financial markets (Janssen 2004; Onder and Simga-Mugan 2006; Mondria and Quintana-Domeque 2007), country fund asset values (Klibanoff, Lamont, and Wizman 1998), and sovereign bond spreads (Mauro, Sussman, and Yafeh 2006). We use these studies as a guide in calculating our country visibility index. Our focus is unique and contributes to this literature by exploring the impact of news on FDI flows, which combine a financial variable with a real decision of where to locate production activity.

We expect that greater coverage in The Economist increases awareness of a country, which in turn leads to greater FDI flows to that country. This approach does not discount the additional information that investment managers use in choosing where to locate FDI. The final decision is based on expected profitability of such investment. Greater awareness of a country, however, may help a country make it on a short list of countries on which the firm would conduct further market analysis. While we do not observe the short list of countries, we observe the flow of news that may make it more likely for a country to appear on a manager’s short list. Those countries that are short listed more frequently are more likely to receive FDI flows relative to those that are not short listed.

2.2. FDI data

Since the visibility index is designed to measure public visibility of countries to US managers deciding where to locate foreign subsidiaries, we focus on FDI outflows from the US. The Bureau of Economic Analysis (BEA) reports all countries for which there are FDI flows in a given year.⁶ Specifically, we use the BEA direct investment capital flows, defined as the ‘funds that US parent companies provide to their foreign affiliates (outflows) net of funds that foreign affiliates provide to their US parents (inflows)’. We focus on direct investment from the US (denoted as ‘without current-cost adjustment’), which ‘arise from transactions that increase US assets or decrease US liabilities’. Note that direct investment capital flows consist of equity capital, intercompany debt, and reinvested earnings.⁷

Based on the above definitions of capital flows, reported capital flows from the US may be positive or negative. It has been standard to truncate the negative values.⁸ Recent work, however, has accounted for negative capital outflows by counting them as inflows.⁹ This approach implies that the decisions governing inflows and outflows are the same. Since we focus on the decisions of US managers, we only examine capital flows from the US (and not FDI flows into the US from other countries). Consequently, we take a different approach here and first focus on the absolute value of flows so that outflows and the reversal of such flows are handled identically. We then explicitly ask whether the negative values respond differently by including a dummy variable equal to one for all negative values to test whether the regression line intercept differs. Further, we allow this dummy variable to interact with our country visibility index to test whether the slope of the regression line differs.¹⁰

The country visibility index covers all editions of The Economist in 2002 and 2003. The FDI flows that we focus on are the 2003 and 2004 flows. Since FDI decisions may take time to implement, we examine whether country visibility in a given year (e.g. 2002) is likely to affect future FDI location decisions (e.g. FDI flows for 2003). In the benchmark regression, we have 153 observations for the regression on 2003 capital flows and 138 observations for the regression on 2004 capital flows (see Appendix 1 for the list of included countries). To increase the number of countries and observations, we present results for the data pooled together so that we have at most 291 observations.¹¹ Given that others have shown that the impacts of various variables on FDI flows may differ for developing countries, we split the full sample into two sub-samples based on income levels: the low- and lower middle-income countries are grouped together while the upper middle- and high-income countries are grouped together. For the observations in the pooled data estimation, Figure 1 plots the FDI flows from the US to other countries against the variations in their visibility index. There appears to be a positive correlation between the two variables in both sub-samples of countries. We examine this unconditional correlation more fully by controlling for a wide range of covariates in our regression analysis.

Figure 1. FDI and country visibility.

2.3. The regression function

We use the pooled data to examine the influence of the constructed country visibility index on the flow of FDI from the US to other countries by employing the following regression:(1)

where FDI_i,t is the log of the absolute value of FDI flows from the US to country i at time t, and NewsCount_i,t−1 is the log of the lagged number of news counts for country i. In order to perform the log transformations, we add 1 to the value of capital flows and to the value of news counts to include those countries with zero capital flows or zero news counts.

The baseline empirical specification includes the country visibility index and a standard set of explanatory variables thought to influence the location of FDI. These explanatory covariates (X_i,t) include a set of conventional gravity model variables. In our most parsimonious specification, we control for GDP, distance from the US, GDP growth rate, and GDP per capita to capture the basic gravity model variables. Except for the growth rate, all gravity variables are in log form. To take into account the differences between the positive and negative capital flows at time t, we also include a dummy variable for negative flows along with an interaction term between the negative flow dummy and the log of lagged number of news counts for country i. As sensitivity analysis, we add additional control covariates to X_i,t including a dummy variable for all countries with the same legal origin as the US (i.e. countries that have a British legal origin), a dummy variable for all countries that have common language with the US (i.e. countries that speak English), a NAFTA dummy variable (for Canada and Mexico), and three dummy variables for the European Union (i.e. a dummy for 15 members of the EU in 2003, a dummy for 10 countries that became members in 2004, and a dummy for 3 EU candidate countries at the time). X_i,t also includes a dummy variable for 2004 and a vector of ones for the constant term. Ɛ_i,t is the error term under the usual assumptions.

Global FDI patterns are often examined from trade and/or financial perspectives. There is a rich tradition examining the determinants of FDI stocks from a trade perspective (e.g. Wheeler and Mody 1992; Markusen and Maskus 2002; Aizenman and Spiegel 2006). Among the determinants are the cost of locating and producing in a particular country, the size of markets, the distance between the parent country and the host country, as well as other trade barriers.¹² From a financial perspective, other influences on FDI flows may include barriers to international capital flows (capital controls) and institutional variables, such as the stability of governments and the enforcement of private property rights (e.g. Harding and Javorcik 2011). Other covariates that capture common language, legal origin, and regional characteristics are also used in these studies.

Recently, a number of model selection studies have examined a large set of covariates that have been used in previous empirical analyses of FDI. Notably, Chakrabarti (2001), Blonigen and Piger (2014), Eicher, Helfman, and Lenkoski (2012), and Jordan and Lenkoski (2012) employ different model selection techniques to identify the key determinants of FDI. Chakrabarti (2001) employs Extreme Bound Analysis to identify the robust determinants of FDI. Blonigen and Piger (2014) and Eicher, Helfman, and Lenkoski (2012) employ Bayesian Model Averaging to account for model uncertainty. Given the large share of missing observations in FDI flows, Eicher, Helfman, and Lenkoski (2012) also incorporate the selection bias correction proposed by Heckman (1979) into their model selection procedure.¹³ Relying on their findings, we choose a set of control covariates that have been shown to be among the most important determinants of FDI flows while considering the possibility of selection bias.

The results of the above model selection studies illustrate the importance of conventional gravity model variables, which reinforce some of the findings in the studies that examine the determinants of FDI from a trade perspective. In particular, they motivate those studies in which a gravity equation is employed to explain the variations in FDI (e.g. Head and Ries 2008). In our most parsimonious regression function, therefore, the gravity model variables, including host country GDP, distance from the US, GDP growth rate, and GDP per capita, are included in X_i,t. In subsequent estimations, we also add an augmented set of covariates to X_i,t, which are described above.

2.4. Identification

Controlling for a large set of covariates, our working hypothesis is that greater country visibility, which raises awareness about a given country, is likely to lead to greater investment by US MNCs in that particular country. Thus, we expect the primary coefficient of interest, α, to be positive. The parameter of interest can be identified due to the timing of the underlying regression function. We assume that country visibility contributes to FDI flows with a lag. In our regression function, the error term captures the unexplained variations in FDI at time t, which are expected to be orthogonal to the variations in our country visibility index at time t-1. Therefore, a shock to the error term at time t is expected to affect FDI contemporaneously, while it has no effect on the lagged visibility measures.

An important consideration in our identification strategy is whether or not the visibility index is influenced by FDI flows. This is a broader question as to country focus in The Economist. We do not have first-hand knowledge of how story topics are chosen for this magazine. Nonetheless, we utilize The Economist because it appears to have broad country coverage each week, with specific regional sections always present in each issue (the regional sections are United States, The Americas, Asia, Middle East and Africa, Europe, and Britain). While the specific countries covered in the regional sections may change each week, the existence of the different sections indicates that a broad focus of countries is taken in each issue. Anecdotal evidence suggests that the editors at The Economist plan what they cover, acting as a filter for their readers. It also appears that the ongoing events, new data, and trendspotting topics are likely to affect the editors’ plan for each issue (The Economist 2013). Hence, one can imagine a situation in which the variations in the country visibility index are endogenous to the variations in FDI. For example, consider a hypothetical country that introduces some business-friendly regulations at time t. This shock is expected to contribute to the FDI from the US to that country at time t, without having any impact on the visibility of that country at time t-1. However, the new regulations might have already been discussed in the media at time t-1. If such a story appears in The Economist, it is reflected in the variations of the lagged country visibility index. The news coverage about that particular story at time t-1 may have some predictability power for the new regulation at time t. This correlation may go either way: greater media attention may encourage or discourage the policy-makers in the host countries to put the new set of regulations in place. In this case, the contemporaneous error term may depend on the lagged visibility measure. Under these circumstances, we are unable to identify the parameter of interest by solely relying on the timing structure of our regression function, and the resulting estimations may be biased. Consequently, we employ the exogenous variations in a set of IVs as we explore an alternative identification strategy as a robustness check in Section 3.

It should be noted that our estimations of the main models and the IV tests do not fully capture a causal relationship. We neither have access to a randomized trial, nor do we have access to any natural experiments. Also, to test whether greater visibility puts a country on a short list for investment, we need to have access to detailed firm level observations. We do not have access to such data, which limits our ability to test for causality directly. Nevertheless, our main identification strategy and IV tests enable us to estimate the effect of lagged country visibility on FDI without any bias caused by endogeneities.¹⁴

3. Estimation results

In this section, we first describe the results of our baseline estimations. Then, we divide our sample by country income groups to examine the effect of country visibility on the FDI outflow from the US to developing countries. We also examine the importance of economic news, compared to political and social news stories. We conclude this section by reporting the results of our IV estimations, which address additional endogeneity concerns and generally confirm the baseline results.¹⁵

3.1. Baseline estimation results

In the most parsimonious empirical specification (Table 2, columns I-1 and I-2), we estimate the effect of the number of news counts for country i at time t-1 on the FDI flows between the US and country i at time t. We use the count data pooled for the years 2002 and 2003 along with the FDI data pooled for the years 2003 and 2004.¹⁶ For this estimation we control for GDP, GDP growth rate, and GDP per capita of country i at time t as well as the distance between the US and country i.

Table 2. Baseline estimation results.

Dependent variable: FDI (log absolute value)	All countries	Without the UK	All countries	Without the UK
	I-1	I-2	II-1	II-2
Lagged News Count (in log)	0.402***	0.377***	0.347***	0.328**
	(0.126)	(0.136)	(0.133)	(0.140)
GDP (in log)	0.600***	0.610***	0.624***	0.630***
	(0.093)	(0.096)	(0.098)	(0.099)
Distance (in log)	−0.714***	−0.711***	−0.692***	−0.683***
	(0.163)	(0.163)	(0.202)	(0.203)
GDP Growth Rate	−0.029	−0.029	−0.024	−0.025
	(0.028)	(0.028)	(0.029)	(0.029)
GDP Per Capita (in log)	0.445***	0.438***	0.424***	0.423***
	(0.091)	(0.092)	(0.106)	(0.106)
Negative Flow Dummy	0.652**	0.627*	0.713**	0.692**
	(0.329)	(0.334)	(0.327)	(0.331)
Negative Flow Dummy * Lagged News Count (in log)	−0.442***	−0.425**	−0.469***	−0.455***
	(0.162)	(0.166)	(0.160)	(0.164)
Time Dummy (for 2004)	0.329*	0.327*	0.350*	0.349*
	(0.184)	(0.186)	(0.186)	(0.187)
Legal Origin Dummy			0.255	0.247
			(0.240)	(0.241)
Common Language Dummy			−0.331	−0.344
			(0.239)	(0.242)
NAFTA Dummy			0.468	0.511
			(0.447)	(0.454)
EU1 (for 15 original members)			0.430	0.395
			(0.419)	(0.428)
EU2 (for 10 additional members)			−0.846**	−0.848**
			(0.386)	(0.388)
EU3 (for candidate countries)			−0.306	−0.314
			(0.483)	(0.479)
Constant	−7.986***	−8.156***	−8.552***	−8.724***
	(2.274)	(2.299)	(2.506)	(2.536)
Number of observations	291	289	283	281
R^2	0.7	0.691	0.716	0.705

Notes:

Huber/White standard errors are reported in parenthesis. ***, **, * denotes significance at the 1, 5, and 10%-level, respectively. The dependent variable is the log of the absolute value of the flow of FDI from the US to country i in 2003 and 2004. Lagged News Count is the log of the number (plus one) of The Economist’s stories in 2002 and 2003 with primary focus on country i. See the text for other variable definitions.

The estimation results reported in Table 2 suggest that the absolute value of FDI flows from the US to other countries is greater for those countries that appear more frequently in The Economist, implying that greater visibility of a country may put it on the short list for the FDI conducted by US MNCs. This effect is evident even after we control for a large set of covariates, including the gravity model controls. To ensure that the baseline results, reported in column I-1 of Table 2, are not driven by the UK, we exclude the UK from the sample. The same pattern emerges, as can be seen in column I-2 of Table 2.

Since we pool data for the years 2003 and 2004, we also include a time dummy variable, which takes one for 2004, and zero for 2003. We incorporate further normalizations by controlling for more detailed country characteristics to ensure that the baseline results are not sensitive. The results, with the UK in column II-1or without the UK in column II-2, are similar to the above. This set includes a dummy for British legal origin, a dummy for English as the common language, a dummy for NAFTA countries, and three different dummies for different groups of EU countries. For the purpose of our estimations, the FDI flows, lagged news counts, GDP, distance, and GDP per capita are transformed into logs following the usual convention. Appendix 1 contains specific variable definitions and sources.

Table 2 shows that the flow of direct investment to a country is significantly and positively related to visibility of that country. The coefficient on Lagged News Count, which is our measure of country visibility, is significantly different from zero and its magnitude varies in a narrow band between 0.33 and 0.40. To fix intuition, consider the regression in column I-1. The estimated coefficient on Lagged News Count is 0.40, indicating a 4% increase in FDI flows for a 10% change in the visibility index when other covariates are kept at their mean. From the 291 observations used in the regression, 225 observations have positive FDI. Among them, the average observation has a count of stories of approximately 17, thus a 10% change in the information variable is associated with less than two stories.¹⁷ The average FDI flows for these observations is approximately $1514 million, while the median is $48 million.¹⁸ Thus, an increase of 4% is an increase in FDI of about $60.5 million for the average observation and an increase of about $1.9 million for the median observation.

In order to test if country visibility has different effects on the negative flows, we control for a negative flow dummy as well as the interaction between the number of news counts and the negative flow dummy.¹⁹ At time t, this dummy is equal to one when we observe a negative capital flow for country i (i.e. when FDI reversals occur). For negative flows, the estimated intercept would be equal to the original estimated intercept in the model (denoted by Constant in Table 2) plus the estimated coefficient for the negative flow dummy. Also, the estimated effect of visibility on negative FDI flows would be equal to the original estimated coefficient in the model (denoted by Lagged News Count in Table 2) plus the estimated coefficient for the interaction between negative flow dummy and the news counts.

The contribution of country visibility to FDI flows appears to be significant only for the positive flows. To illustrate this, consider (a hypothetical average) country k receiving FDI flows from the US at time t. The Negative Flow Dummy is equal to zero in this case, so that the estimated parameter associated with Lagged News Count shows the contribution of country k’s visibility to the FDI flow from the US to country k. The results in Table 2 show that, even after controlling for gravity variations, this estimated parameter is positive and statistically significant. Now consider (another hypothetical average) country l with negative values of FDI flows to the US at time t (i.e. a reversal of previous FDI flows from the US). Since the Negative Flow Dummy is equal to one in this case, the sum of the estimated parameters associated with Lagged News Count and Negative Flow Dummy * Lagged News Count shows the contribution of country l’s visibility to the FDI flow from country l to the US. We are unable to reject the null hypothesis that the sum of those estimated parameters is equal to zero.²⁰ This suggests that greater visibility matters for the FDI outflows from the US to other countries but not for the reversal of such flows. Excluding the dummy variable for negative flows and its interaction with the news count measure does not affect the sign and the significance of the parameter of interest.²¹ However, in order to be able to distinguish between the effect of country visibility on capital outflows and inflows, we take into account the binary variations in negative flows (i.e. reversals) and their interaction with the variations in the news counts. The results indicate that greater visibility of country i at time t-1 increases the chance of greater capital outflow from the US to country i at time t, but has no significant effect on the reallocation of capital from country i to the US at time t. The interpretation here is that US MNCs that have already invested in a country have intimate knowledge of the operating conditions of that country. Thus, greater visibility has no significant effect on the reallocation of capital from country i to the US.²²

Among the gravity variables, GDP and GDP per capita are positively correlated with the absolute value of capital flows while distance is negatively correlated with the absolute value of capital flows. As for the other control covariates, we find that on average the absolute value of FDI flows are greater for the year 2004. Incorporating the time dummy, therefore, we are able to control for the trend. We also find that the estimated parameter for the negative flow dummy is significant. This indicates that the absolute value of the constant term may be of lower magnitude for the negative flows.

Excluding the UK has little impact on the coefficient on Lagged News Count. In the most parsimonious regression (columns I-1 and I-2), excluding the UK reduces the coefficient somewhat on Lagged News Count. In the other specifications, excluding the UK has little effect; we show this with columns II-1 and II-2.

Along with the covariates included in the most parsimonious model, we further explore whether common legal origin impacts FDI by including a dummy variable taking the value of one for all countries with the same legal origin as the US (i.e. those countries also having a UK legal origin).²³ Common language may also be important in determining where FDI locates. Thus, we include an English language dummy for those countries that use English (either as the official language or as a primary language as reported in the CIA world factbook). We also explore regional considerations by including a NAFTA dummy variable and three possible European Union dummy variables: EU1 for the 15 members of the EU in 2003, EU2 for the countries that became members in 2004, and EU3 for the European Union candidate countries at the time.²⁴

The estimated parameter of interest, shown in columns II-1 and II-2 in Table 2, is quite robust to this additional set of explanatory variables. Controlling for the standard gravity covariates, however, the additional control variables have little explanatory power. The only exception is the EU2 dummy. Compared to other countries that are not controlled for by NAFTA, EU1, and EU3 dummies, the FDI activities are estimated to be lower in Czech Republic, Cyprus, Estonia, Latvia, Lithuania, Hungary, Malta, Poland, Slovenia, and Slovakia. This does not change the coefficient on Lagged News Count.

To make sure that the above results are not driven by relatively large flows, we conduct another test in which we only include observations where the absolute value of FDI inflows or outflows are less than or equal to the value of the 90th percentile. Thus, along with the UK, countries such as Australia, Belgium, Brazil, Canada, China, France, Germany, Ireland, Italy, Japan, Korea, Luxembourg, Mexico, the Netherlands, Singapore, Spain, Sweden, and Switzerland are excluded from the sample. The results of the estimations using the smaller sample support the baseline results: conditional upon a wide range of covariates, greater country visibility contributes to FDI flows from the US to the countries in the smaller sample, but has no significant effect on the reversal of such flows from those countries to the US.²⁵

3.2. Income groups

Harding and Javorcik (2011) show that IPAs significantly contribute to the FDI in developing countries but are not effective for industrialized countries. Thus, we consider subsamples of our countries to examine whether news stories contribute to FDI flows only in the less developed countries. This hypothesis implies that greater news is providing visibility to countries for which information is less available. We examine the baseline estimation results across country groupings based on income, using the World Bank classifications of countries into high income, upper middle income, lower middle income, and low income. We present results for the developing countries compared to the high-income countries. We also present results for the high- and upper middle-income countries grouped together and the lower middle- and low-income countries grouped together. While the income levels may be a somewhat arbitrary delimiter, these divisions appear to be important for the impact of news on FDI flows.

Table 3 reports the results. The coefficient on Lagged News Count for the developing countries, reported in column I-1, is similar to that for all countries. The results reported in column I-1 and column I-2 indicate a significant positive association between country visibility and FDI flows from the US to developing countries but not for high- income countries. The positive association is also statistically significant for lower middle- and low-income countries (column II-1), and remains insignificant for high- and upper middle-income countries (column II-2).²⁶ As in the baseline results, country visibility is not a significant determinant of the FDI reversals, as the sum of the coefficients on Lagged News Count and Negative Flow Dummy * Lagged News Count is statistically not different from zero for all country income groups.

Table 3. Country income groups.

Dependent variable: FDI (log absolute value)	Developing countries	High-income countries	Lower middle- and low-income countries	High- and upper middle-income countries
	I	I-2	II-1	II-2
Lagged News Count (in log)	0.419**	0.091	0.450**	0.101
	(0.169)	(0.275)	(0.192)	(0.218)
GDP (in log)	0.591***	0.642**	0.521***	0.884***
	(0.110)	(0.322)	(0.117)	(0.242)
Distance (in log)	−0.861***	0.547	−0.913***	−0.706*
	(0.237)	(0.994)	(0.297)	(0.429)
GDP Growth Rate	−0.020	−0.277**	−0.016	−0.041
	(0.030)	(0.112)	(0.033)	(0.045)
GDP Per Capita (in log)	0.379***	1.753***	0.459***	0.343
	(0.125)	(0.606)	(0.173)	(0.417)
Negative Flow Dummy	0.700**	2.550	0.906**	0.411
	(0.354)	(1.948)	(0.398)	(0.716)
Negative Flow Dummy * Lagged News Count (in log)	−0.489***	−1.464*	−0.578**	−0.300
	(0.182)	(0.826)	(0.235)	(0.275)
Time Dummy (for 2004)	0.391*	0.638	0.348	0.354
	(0.204)	(0.457)	(0.233)	(0.323)
Legal Origin Dummy	0.130	2.122**	0.065	1.116**
	(0.256)	(0.940)	(0.282)	(0.529)
Common Language Dummy	−0.397	−0.487	−0.156	−0.802
	(0.255)	(0.656)	(0.274)	(0.547)
NAFTA Dummy	0.544	2.352		0.271
	(0.378)	(2.613)		(0.977)
EU1 (for 15 original members)		1.073		0.416
		(0.709)		(0.586)
EU2 (for 10 additional members)	−0.645	−1.647**		−0.669
	(0.447)	(0.712)		(0.446)
EU3 (for candidate countries)	−0.306		−0.231
	(0.504)		(0.546)
Constant	−6.032*	−32.971***	−4.598	−13.758***
	(3.133)	(9.394)	(3.877)	(4.032)
Number of observations	225	58	177	106
R^2	0.602	0.719	0.551	0.722

Note:

See notes to Table 2. Country income groups are from the World Bank (see Appendix 1 for details).

One interpretation of this finding is that investors already may be aware of the investment potential in developed countries. News stories on developing countries may have a larger impact in raising awareness so that articles in The Economist provide a higher marginal value. Thus, for the developing countries as a whole and for the low-income and lower middle-income countries in particular, rather than ‘no news is good news’, we find that ‘more news is good news’!

To examine the magnitude of this effect, let us consider the estimation for lower middle- and low-income countries (column II-1). The estimated coefficient on Lagged News Count is 0.45, indicating a 4.5% increase in FDI flows from the US to lower middle- and low-income countries for a 10% change in their visibility index when other covariates are kept at their mean. From 177 observations in the sample, 130 observations have positive FDI flows from the US. Among them, the average observation has a count of stories of approximately 8, thus a 10% change in the information variable is associated with less than one story. The average FDI flows for these observations are approximately $181 million, while the median flows are $9.5 million. Within this group, therefore, an increase of 4.5% is an increase in FDI of about $8.14 million for the average observation and an increase of about $0.42 million for the median observation.

To put this number in perspective, we compare our results with those in Harding and Javorcik (2011), who examine the impact of IPAs on FDI flows. They ask how the existence of an IPA affects the FDI in the targeted sectors. Their estimations suggest that in developing countries targeted sectors receive about 155% higher inflows relative to non-targeted sectors. For the median sector-country observation, an increase of 155% translates into $17 million dollars of additional FDI. Given the estimated magnitude reported in Harding and Javorcik (2011), our estimation appears to be reasonable.

As for the gravity control covariates, GDP continues to be a significant determinant of FDI flows. Except for the group of high-income countries, distance is also a significant determinant, with greater distance associated with less FDI. Further, we find that among developing countries GDP per capita is also a significant determinant of FDI flows. Legal origin appears to have a positive impact on FDI flows for the high- and upper middle-income countries.

To test the robustness of the above results, we exclude countries with relatively large flows from each subsample, including only observations for which the absolute value of FDI inflows or outflows are less than or equal to the value of the 90th percentile. Excluding the observations with relatively large flows from the subsample of lower middle- and low-income countries, the results suggest that, conditional upon a wide range of covariates, greater country visibility contributes to FDI flows from the US to developing countries. Greater country visibility has no significant effect on the reversal of such flows from those countries to the US. As for developed nations, we find no significant effects. The findings of this robustness check are in line with the results reported above.²⁷

3.3. Type of stories

While we find a consistently positive effect of country visibility on FDI flows from the US to developing countries, our results thus far say little about the type of news that is important. Therefore, we modify our country visibility index by categorizing news stories as predominantly Economic, Political, or Social.²⁸

Table 4 shows that stories with a predominantly economic focus tend to have a positive impact on the FDI flows. This effect is particularly evident among lower middle- and low-income countries.²⁹ When we employ a full vector of control covariates (column II-2), for instance, we find that economic news counts are positively and significantly correlated with FDI outflows from the US to lower middle- and low-income countries.³⁰ Further, based on the negative flow interaction term, more news about the economic condition of developing countries matters for the FDI outflows from the US to those countries but not for the reversal of previous flows. Although the estimated parameter for Neg. Flow Dummy * Lagged Economic News Count in column II-2 is statistically insignificant, it is again of negative sign. More importantly, the sum of the estimated parameters for Lagged Economic News Count and Neg. Flow Dummy * Lagged Economic News Count is statistically not different from zero.³¹

Table 4. Stories categorized as economic, political, and social.

Dependent variable: FDI (log absolute value)	All countries	Lower middle- and low-income countries	All countries	Lower middle- and low-income countries
	I-1	I-2	II-1	II-2
Lagged Economic News Count (in log)	0.565***	0.600**	0.562***	0.597**
	(0.176)	(0.262)	(0.175)	(0.267)
Lagged Political News Count (in log)	0.030	0.189	0.066	0.210
	(0.207)	(0.291)	(0.209)	(0.293)
Lagged Social News Count (in log)	−0.077	0.055	−0.201	−0.007
	(0.172)	(0.252)	(0.179)	(0.253)
GDP (in log)	0.607***	0.440***	0.645***	0.488***
	(0.088)	(0.116)	(0.092)	(0.114)
Distance (in log)	−0.732***	−1.005***	−0.760***	−1.077***
	(0.162)	(0.253)	(0.200)	(0.299)
GDP Growth Rate	−0.024	−0.014	−0.018	−0.010
	(0.028)	(0.030)	(0.029)	(0.032)
GDP Per Capita (in log)	0.371***	0.410**	0.330***	0.335*
	(0.092)	(0.173)	(0.106)	(0.186)
Negative Flow Dummy	0.343	0.716**	0.364	0.668*
	(0.286)	(0.357)	(0.288)	(0.367)
Neg. Flow Dummy * Lagged Economic News Count (in log)	−0.708**	−0.323	−0.663**	−0.338
	(0.314)	(0.503)	(0.308)	(0.500)
Neg. Flow Dummy * Lagged Political News Count (in log)	0.524	0.012	0.419	0.044
	(0.423)	(0.609)	(0.431)	(0.631)
Neg. Flow Dummy * Lagged Social News Count (in log)	−0.256	−0.449	−0.234	−0.439
	(0.321)	(0.396)	(0.336)	(0.413)
Time Dummy (for 2004)	0.451**	0.464*	0.520**	0.534**
	(0.203)	(0.258)	(0.209)	(0.264)
Legal Origin Dummy			0.391	0.100
			(0.240)	(0.271)
Common Language Dummy			−0.361	−0.164
			(0.235)	(0.260)
NAFTA Dummy			0.174
			(0.456)
EU1 (for 15 original members)			0.430
			(0.418)
EU2 (for 10 additional members)			−0.884**
			(0.386)
EU3 (for candidate countries)			−0.309	−0.201
			(0.528)	(0.641)
Constant	−7.289***	−1.538	−7.610***	−1.490
	(2.271)	(3.626)	(2.550)	(4.127)
Number of observations	291	183	283	177
R^2	0.711	0.558	0.725	0.568

Notes:

See notes to Table 2. Economic, Political, or Social News Counts are defined as the log of the number (plus one) of The Economist’s stories in 2002 and 2003 with primary focus on country i and categorized as predominantly Economic, Political, or Social respectively.

The gravity control covariates (notably, GDP, distance, and GDP per capita) are significant and of expected sign. Conditional upon the conventional gravity variations, therefore, cross-section differences in the visibility of economic conditions in developing countries, as measured by the Lagged Economic News Count, is a good predictor for the amount of FDI outflows from the US to those countries. Greater visibility of a developing country’s economic condition may put it on the short list for the FDI made by US MNCs.

3.4. The IV estimations

The results reported in Table 4 suggest that stories with a predominantly economic focus have a positive impact on the FDI outflows from the US to lower middle- and low-income countries. One may worry that there exist some endogeneities in our model. We minimize the potential bias caused by endogeneity since in our baseline regression (Equation 1(1) ) we examine the effect of the lagged country visibility (i.e. NewsCount_i,t-1) on current investments (i.e. FDI_i,t). Given the underlying regression function, a shock to the error term (i.e. Ɛ_i,t) is expected to affect the FDI contemporaneously, and it may have no effect on the lagged visibility measures. This timing structure preserves the underlying identification restriction. Nonetheless, one may imagine a case in which the error term is affected by the lagged country visibility. For instance, the economic news stories about a given country that appear in The Economist may encourage/discourage some regulations in that country. With a one year lag, those changes may appear as shocks to the error term, affecting the FDI flows. In that case, the identification restriction is not binding.

To address the potential for endogeneity, we make use of the exogenous variations in two IVs to predict the variations in country visibility and estimate its effect on the FDI flows. We consider the total medal counts for the winter Olympics in 2002 (Salt Lake City, US) and the number of natural disaster incidents in 2002 and 2003.³² Before using these variables as our IVs, we add one to each count and take the log.

The two-stage least square estimations (2SLS) are reported in Table 5. For these estimations we use either the medal counts (in log) or the number of natural disaster incidents (in log) to predict the country visibility. The first stage results show that our instruments are positively correlated with country visibility, and the resulting parameters of interest in the second stage are supportive of our baseline findings.

Table 5. The IV estimations (2SLS).

Dependent Variable	All countries		Lower middle- and low-income countries		All countries		Lower middle- and low-income countries
	Lagged News Count (in log)	FDI (log absolute value)	Lagged News Count (in log)	FDI (log absolute value)	Lagged News Count (in log)	FDI (log absolute value)	Lagged News Count (in log)	FDI (log absolute value)
	I-1	I-2	II-1	II-2	III-1	III-2	IV-1	IV-2
Lagged News Count (in log)		−0.276		1.138**		−0.015		0.365
		(0.989)		(0.581)		(0.891)		(0.720)
GDP (in log)	0.548***	0.977*	0.460***	0.188	0.514***	0.829	0.398***	0.563
	(0.031)	(0.561)	(0.033)	(0.310)	(0.045)	(0.507)	(0.038)	(0.354)
Distance (in log)	0.058	−0.655***	0.270**	−1.114***	0.080	−0.67***	0.304**	−0.888**
	(0.088)	(0.205)	(0.113)	(0.360)	(0.091)	(0.207)	(0.118)	(0.359)
GDP Growth Rate	−0.015*	−0.034	−0.019**	−0.003	−0.015*	−0.030	−0.016*	−0.018
	(0.008)	(0.034)	(0.009)	(0.031)	(0.008)	(0.034)	(0.009)	(0.036)
GDP Per Capita (in log)	−0.251***	0.273	−0.150**	0.560***	−0.183***	0.336	−0.067	0.446**
	(0.045)	(0.256)	(0.067)	(0.172)	(0.055)	(0.238)	(0.070)	(0.201)
Negative Flow Dummy	−0.827***	0.199	−0.918***	1.543**	−0.803***	0.414	−0.887***	0.827
	(0.137)	(0.874)	(0.135)	(0.645)	(0.137)	(0.804)	(0.145)	(0.774)
Negative Flow Dummy * Lagged News Count (in log)	0.491***	−0.179	0.577***	−0.959***	0.460***	−0.300	0.541***	−0.531
	(0.073)	(0.475)	(0.086)	(0.362)	(0.072)	(0.434)	(0.093)	(0.447)
Time Dummy (for 2004)	−0.101	0.287	−0.059	0.390	−0.082	0.314	−0.020	0.342
	(0.088)	(0.228)	(0.102)	(0.241)	(0.088)	(0.201)	(0.105)	(0.226)
Legal Origin Dummy	0.321**	0.439	0.107	0.007	0.285**	0.362	0.031	0.072
	(0.137)	(0.378)	(0.138)	(0.285)	(0.139)	(0.346)	(0.136)	(0.277)
Common Language Dummy	0.165	−0.220	0.180	−0.278	0.191	−0.266	0.200	−0.141
	(0.144)	(0.306)	(0.142)	(0.301)	(0.143)	(0.288)	(0.138)	(0.302)
NAFTA Dummy	0.506**	0.834			0.518**	0.681
	(0.235)	(0.719)			(0.214)	(0.688)
EU1 (for 15 original members)	0.259	0.630			0.323	0.546
	(0.224)	(0.517)			(0.217)	(0.483)
EU2 (for 10 additional members)	0.179	−0.719			0.233	−0.772*
	(0.160)	(0.449)			(0.147)	(0.445)
EU3 (for candidate countries)	−0.279	−0.482	−0.367	−0.025	−0.277	−0.408	−0.274	−0.257
	(0.429)	(0.525)	(0.480)	(0.702)	(0.408)	(0.480)	(0.435)	(0.543)
IV: Total Medal Counts in 2002 Winter Olympics (in log)	0.143*		0.347***
	(0.086)		(0.129)
IV: Lagged Number of Natural Disaster Incidents (in log)					0.177*		0.275***
					(0.093)		(0.086)
Constant	−9.989***	−15.061	−10.528***	3.156	−10.06***	−12.334	−10.25***	−5.563
	(0.933)	(10.457)	(1.438)	(8.136)	(0.977)	(9.628)	(1.421)	(8.697)
Number of observations	283	283	177	177	283	283	177	177
R^2	0.745	0.692	0.7	0.511	0.746	0.708	0.706	0.551

Note:

See notes to Table 2.

As shown in columns I-1 and II-1, the total medal counts for the winter Olympics in 2002 is positively correlated with our measure of country visibility in the full sample as well as the subsample of lower middle- and low-income countries. In the second stage, however, country visibility is positively correlated with US FDI outflows only in the subsample of lower middle- and low-income countries.³³ As reported in column II-2, the estimated coefficient for Lagged News Count is positive and statistically different from zero, while the sum of the estimated coefficients for Lagged News Count and Negative Flow Dummy*Lagged News Count is statistically not different from zero. These estimations imply that, similar to our baseline findings, greater visibility contributes to the US FDI in lower middle- and low-income countries, but it has no effect on the FDI reversals.³⁴

We repeat the same exercise, employing the exogenous variations in natural disasters in 2002 and 2003 to predict the effect of country visibility in those years on US FDI in 2003 and 2004. As shown in columns III-1 and IV-1, natural disaster incidents are significant predictors of country visibility. The second stage results for the full sample, reported in column III-2, is comparable to what we find when we employ winter Olympics medal counts as our IV (column I-2). However, the results for the subsample of lower middle- and low-income countries, reported in column IV-2, is not as significant as our previous IV estimations (column II-2).³⁵

Given our first set of IV estimations, reported in Table 5, we employ both of our instruments in an IV vector to improve the precision of our IV estimations. Since we use more than one instrument to identify the parameter of interest in Equation 1(1) , we employ the Generalized Method of Moments (GMM) estimation technique. Using GMM, we are able to estimate the parameters of interest of our overidentified model, and test for the exogeneity of such instruments using the overidentifying condition.³⁶

To obtain an optimal two-step GMM estimation, in which variances are the smallest, we employ 2SLS estimations to predict the residuals first. Using the residuals, we then form an optimal robust weighting matrix for our GMM estimation.³⁷ Thus, along with the final GMM results, we also report the first stage correlations.

Conditional upon a wide range of covariates, the results reported in column II-2 of Table 6 suggest that the lagged country visibility, predicted by the exogenous variations in winter Olympics medals in 2002 (in log) and the number of natural disasters in 2002 and 2003 (in log), contributes to the FDI made by US MNCs in lower middle- and low-income countries. The estimated parameter for the Lagged News Count variable is insignificant when we consider the full sample (column I-2). Yet, it is positive and statistically significant when we only consider lower middle- and low-income countries. The sum of the estimated parameters for visibility (Lagged News Count) and its interaction with the negative flow dummy (Negative Flow Dummy * Lagged News Count), as reported in column II-2, is statistically not different from zero, which again indicates that greater visibility only matters for the FDI flow from the US to those countries and not for the FDI reversals.³⁸

Table 6. The IV estimations (optimal two-step GMM).

Dependent variable	All countries		Lower middle- and low-income countries
	Lagged News Count (in log)	FDI (log absolute value)	Lagged News Count (in log)	FDI (log absolute value)
	I-1	I-2	II-1	II-2
Lagged News Count (in log)		−0.138		0.846**
		(0.649)		(0.428)
GDP (in log)	0.498***	0.899**	0.383***	0.337
	(0.046)	(0.370)	(0.039)	(0.232)
Distance (in log)	0.078	−0.660***	0.283**	−1.014***
	(0.090)	(0.202)	(0.117)	(0.324)
GDP Growth Rate	−0.013	−0.033	−0.016*	−0.007
	(0.008)	(0.032)	(0.009)	(0.032)
GDP Per Capita (in log)	−0.193***	0.306*	−0.076	0.539***
	(0.054)	(0.182)	(0.070)	(0.167)
Negative Flow Dummy	−0.806***	0.312	−0.882***	1.294**
	(0.136)	(0.625)	(0.142)	(0.541)
Negative Flow Dummy * Lagged News Count (in log)	0.485***	−0.243	0.561***	−0.812***
	(0.073)	(0.326)	(0.091)	(0.300)
Time Dummy (for 2004)	−0.083	0.306	−0.020	0.352
	(0.088)	(0.201)	(0.104)	(0.228)
Legal Origin Dummy	0.312**	0.398	0.055	0.013
	(0.138)	(0.299)	(0.136)	(0.273)
Common Language Dummy	0.178	−0.244	0.202	−0.236
	(0.143)	(0.271)	(0.137)	(0.284)
NAFTA Dummy	0.440**	0.756
	(0.216)	(0.575)
EU1 (for 15 original members)	0.263	0.585
	(0.226)	(0.448)
EU2 (for 10 additional members)	0.208	−0.739*
	(0.157)	(0.423)
EU3 (for candidate countries)	−0.274	−0.435	−0.335	−0.136
	(0.426)	(0.456)	(0.470)	(0.629)
IV: Total Medal Counts in 2002 Winter Olympics (in log)	0.140*		0.305**
	(0.083)		(0.125)
IV: Lagged Number of Natural Disaster Incidents (in log)	0.174*		0.255***
	(0.093)		(0.086)
Constant	−9.620***	−13.641*	−9.663***	−0.583
	(0.984)	(7.124)	(1.463)	(6.264)
Number of observations	283	283	177	177
R^2	0.75	0.702	0.714	0.538
The p-value for Hansen’s (1982) J-statistic		84.5%		41.5%

Notes:

See notes to Table 2. The first stage results, reported in columns I-1 and II-1, are used to form the weighting matrix for the optimal two-step GMM estimations, reported in columns I-2 and II-2.

As for the gravity control covariates, the IV estimations reported in column II-2 suggest that, among the lower middle- and low-income countries, the GDP of a host country is positively correlated with the US FDI flows to that country, although the estimated parameter is significant only at the 15% level (p-value = 14.7%). Also, greater distance negatively affects the FDI flow between the US and other countries. Lastly, the GDP per capita of a host country is positively correlated with the US FDI flows to that country.

To make sure that our model is identified, we test the overidentifying restrictions using Hansen’s (1982) J-statistic. The relatively high p-values, reported in column I-2 and II-2, imply that we are unable to reject the underlying orthogonality conditions. Thus, it is safe to assume that the instruments in use are exogenous to the variations in FDI, and our model is identified. Employing the exogenous variations in our IVs, therefore, we are able to verify that greater country visibility contributes to FDI by US MNCs in lower middle- and low-income countries, while it has no significant effect on the FDI reversals.

4. Conclusion and some extensions

Is ‘no news’ really ‘good news’? Perhaps not when it comes to FDI flows to developing countries, particularly to the lower middle- and low-income countries. The findings of our research imply that for those countries ‘more news is good news’. To examine the impact of news stories on the amount of direct investment flows from the US to its investment partners, we develop a basic index of news stories that counts the number of articles on a particular country in The Economist over a year. We then use this country visibility index as an additional explanatory variable in a standard econometric analysis of the FDI flows from the US to other countries, utilizing a number of other explanatory variables to isolate the impacts of the visibility variable. Overall, we find strong support for a significant, positive impact of the number of stories on a country and FDI flows to that country.

The estimated impact of the number of news stories on the amount of direct investment flows is in particular significant for the lower middle- and low-income countries. For the upper middle- and high-income countries, the effect appears to be insignificant. This suggests that greater country visibility contributes to the direct investments made by US multinationals in developing countries. We also find that among different types of stories, economic news is of great importance. Political and social news stories do not significantly contribute to the direct investment flows. This suggests that greater visibility of economic conditions contributes to the FDI flows from the US. Since the economic news are shown to be important, we employ two instrumental variables that are exogenous to the variations in the FDI to make sure that the results are not biased due to some endogeneities. The same pattern emerges: greater country visibility of lower middle- and low-income countries contributes to the FDI outflows from the US to those countries.

Focusing on the visibility of countries, based on the count of stories, provides evidence that FDI flows from the US to other countries are significantly affected by the attention paid to these countries by media outlets. One criticism of the approach used is that it may be useful to collect the visibility index over multiple years to more fully utilize the event study methodology. This would allow a specification based on the change in visibility (rather than the raw count of articles). A change in visibility may provide additional information about how changes in visibility affect changes in capital flows. This would be asking a somewhat more nuanced question about whether increases in visibility are associated with increases in capital flows. In this case, the type of information included in each article may become of particular importance. Given a baseline level of information, whether additional information on a country conforms to expectations (e.g. providing more favorable or less favorable information relative to expected information) is likely to matter. Expected information vs. realized information, however, may be particularly difficult to measure. Thus, we focus on broad visibility, as captured by the count of stories. This allows less normative judgment on our part.

FDI entails some management control of foreign operations of a subsidiary. As such, it is typically seen as much more stable than other forms of capital flows. This drives our choice of an annual index and the utilization of annual data in the initial empirical analysis. As a financial flow, however, FDI may react more like a portfolio flow, as firms move funds between operations at home and abroad, leading to greater volatility within a given year. Comparing longer term FDI and information flows to shorter term FDI and information flows may provide a compelling new way of looking at the question of volatile capital flows. Expanding the coverage by examining the level of visibility available on a daily basis using the Wall Street Journal or the Financial Times may provide additional insight on the timing of FDI decisions.

Disclosure statement

No potential conflict of interest was reported by the authors.

Acknowledgements

We are grateful to two anonymous referees. We would also like to thank seminar participants at the University of Colorado – Boulder, the University of Colorado – Denver, University of Wisconsin – Milwaukee, and Marquette University for helpful comments and suggestions. Lastly, we would like to thank Adam Hyland, Juan C. Veloz, and Shyh Yoa Chea for helpful research assistance.

Notes

1. In email correspondence, the staff members at The Economist Group provided an estimate of the total audience of The Economist, which is about 852,000 for 2002 and 2003.

2. Information obtained via email exchange with the staff members at The Economist Group.

3. These positions include C-Suite, President, Vice President, Treasurer, Chairman of the Board, Member of the Board, Comptroller, General Manager, Owner-Partner, or Manager, with individual earnings of at least $35,000. Information obtained via email exchange with the staff members at The Economist Group.

4. Using the FDI stock positions to calculate flows provides similar conclusions.

5. Given the lack of FDI data, Iraq is not included in our regressions.

6. The BEA uses three specific codes for the non-availability of FDI data for certain countries: an ‘n.s.’ indicates the FDI data are zero or included elsewhere, a ‘star’ indicates the FDI data are small (non-zero between −$500,000 and +$500,000), a ‘D’ indicates disclosure would violate the privacy of individual investing firms. To achieve the greatest country coverage, we include the ‘n.s.’ values as zero and ‘star’ values as $500,000. However, the ‘D’ values remain somewhat problematic since there is no indication of size or value of these variables. There are more ‘D’ values using the 2004 capital flows compared to the 2003 capital flows; some of these values may by arguably large, since those with a ‘D’ for 2004 include Australia, Hong Kong, and Singapore, among others. Following Harding and Javorcik (2011), we set the ‘D’ values as missing. For the 2003 capital flows, there are 13 ‘n.s.’ values, 26 ‘star’ values, and 23 ‘D’ values. For the 2004 capital flows, there are 7 ‘n.s.’ values, 18 ‘star’ values, and 40 ‘D’ values.

7. We employ aggregate variations reported by the BEA since we do not have access to data from a well-stratified sample of multinational firms in the US, which would allow us to capture the effect of country visibility on changes in firms’ external assets or liabilities.

8. For instance, Harding and Javorcik (2011) follow Blonigen and Wang (2004) and Eichengreen and Tong (2005) and truncate all negative values of FDI flows to US$0.1.

9. See Binici, Hutchison, and Schindler (2010), in particular.

10. Using the log of the absolute value of FDI flows where the negative flows are given a negative sign after taking the log provides a similar coefficient on the news count variable but does not allow us to consider the differing impacts on the reversal of previous FDI flows.

11. Employing a larger vector of control covariates, the number of observations in use reduces slightly.

12. In a paper that examines the impact of tax treaties on direct investment positions, Blonigen and Davies (2004) provide a nice description of the empirical specifications of the application of gravity models versus the Markusen and Maskus (2002) model of vertical vs. horizontal FDI.

13. Jordan and Lenkoski (2012) use the same set of data, proposing an alternative method to improve the precision of the estimations that are reported in Eicher, Helfman, and Lenkoski (2012).

14. We are grateful to an anonymous referee for raising this point.

15. In Appendix 1, we report another set of results in which institutional variations are used as additional control variables. Adding these institutional controls does not change the results reported here.

16. The same pattern emerges when we estimate the above regression function for each year 2003 and 2004 separately, although there is some loss in efficiency due to smaller sample sizes. Detailed estimation results available upon request.

17. Using the pooled data, we observe 153 countries in 2003 and 138 countries in 2004. The average news count for the 153 countries in 2003 is 14.65. For the 138 countries in 2004, this average is equal to 14.56.

18. The average FDI flow for the 153 countries in 2003 is about $840 million, while the median is about $4 million. For the 138 countries in 2004, this average is about $1460 million, while the median is about $15 million.

19. Motivated by a point raised by an anonymous referee, we also separate the outflows and reversals of flows (i.e. negative flows) and estimate a modified model using only the outflows or the reversals. Though in some cases the precision of the estimated parameter of interest is affected, the same pattern emerges. Detailed estimation results available upon request.

20. For instance, in column I-1, this value is 0.402–0.442 = −0.04, which is insignificantly different from zero.

21. Detailed estimation results available upon request.

22. Though it is evident that greater country visibility contributes to FDI flows from the US to other countries, we are unable to test whether this effect is through the extensive margin (i.e. entry of new US MNCs to a given country) or the intensive margin (i.e. increase in existing investment position made by a US MNC in a given country). Such a test requires detailed firm-level information that are not available in the FDI data used in this study.

23. See La Porta, Lopez-de-Silanes, and Shleifer (2008) for more information on the legal origin variable.

24. A full list of countries that are considered for each dummy variable is included in Appendix 1. The countries that are included in the EU2 vector are likely to have appeared more often in the news during the time leading up to their membership. Further, they may have received more FDI flows based on their expected EU membership, which may also be the case for the countries that are included in the EU3 vector.

25. We are grateful to an anonymous referee who made this suggestion. Detailed estimation results are available in Appendix 1.

26. The same pattern emerges when we estimate our model using the FDI data for 2003 and 2004 separately, although there is some loss in efficiency. Detailed estimation results available upon request.

27. From the subsample of lower middle and low-income countries, Algeria, Azerbaijan, Brazil, China, Colombia, Ecuador, Egypt, Equatorial Guinea, India, Nigeria, Peru, Philippines, Russia, South Africa, and Thailand are excluded. Also, from the subsample of high- and upper middle-income countries, Canada, Germany, Ireland, Japan, Mexico, the Netherlands, Switzerland, and the UK are excluded. Detailed estimation results are available in Appendix 1.

28. As stated before, we have decided not to categorize news stories based on their tone to avoid the potential for measurement error. However, we acknowledge that negative news or a negative tone to a story may play a role in the response of FDI. To capture this idea, we have explored results in which we include a dummy for lagged armed conflict and interact this dummy with lagged news count. The results are supportive of the main findings such that news stories remain a significant positive predictor of FDI flows for the lower middle- and low-income countries where there is no armed conflict. Where there is armed conflict, the combined coefficient (i.e. the sum of lagged visibility measure and its interaction with lagged conflict dummy) is not significant, indicating no response of FDI to news stories in this case. It should be noted, however, that the combined coefficient does not turn negative. The results are reported in Appendix 1. Further exploration of the tone of news coverage may be useful in future research.

29. Detailed estimation results for the sub-sample of high income and the full set of developing countries available upon request.

30. Similar results for the types of stories are found when we repeat our estimation for the FDI data in 2003 and 2004 separately. Although there is a loss of efficiency in some cases, we generally find that greater visibility of the economic conditions contribute to the FDI flows from the US to other countries. Detailed estimation results available upon request.

31. This is the same for the parameters reported in column I-2.

32. The data for Olympics medal counts are provided by the International Olympics Committee, and the data for the number of natural disaster incidents are provided by the Center for Research on the Epidemiology of Disasters (School of Public Health at Université catholique de Louvain).

33. The mean of the total medal counts for the winter Olympics in 2002 is equal to 0.95 in the full sample and 0.21 in the subsample of lower middle- and low-income countries. The standard deviations are 3.9 and 1.43, respectively.

34. We use the exogenous variations in winter Olympics medals in 2002 to estimate the effect of country visibility in 2002 on US FDI flows in 2003. In terms of sign and magnitude, the results are comparable to what we find in the pooled data, as reported in Table 5 (columns I-1 to II-2), but they are not as significant.

35. The same pattern emerges when we employ the number of natural disaster incidents to predict the effect of country visibility in 2002 and 2003 on FDI flows in 2003 and 2004 separately.

36. Refer to Hayashi (2000, 198–208) for the general formulation and definition of the GMM estimator. The GMM estimation is used when there are more than one identification restrictions; i.e. when the model is overidentified. Since we only assume one identification restriction in our baseline regression, we use OLS estimation for the baseline. Also, since we only employ one instrument in each of the IV estimations that are reported in Table 5, we use 2SLS in our first set of IV estimations. However, now that we employ more than one instrument to predict the variations in our country visibility index (i.e. we have an overidentified model), we use GMM for the IV estimations that are reported in Table 6.

37. Refer to StataCorp (2013, 7–10) for a detailed description of the computational procedure of an optimal two-step GMM estimator.

38. Although there is a loss of efficiency, the same pattern emerges when we repeat the optimal two-step GMM estimation procedure for the FDI data in 2003 and 2004 separately. Detailed estimation results available upon request.

Appendix 1

In this appendix, we first report the baseline estimation results (as given by Equation 1(1) ) controlling for a wide range of institutional variables. Then, we report the results of a robustness check where we only include observations for which the absolute value of FDI inflows or outflows are less than or equal to the value of the 90th percentile. We also report the result of another test where we include variations in armed conflicts to examine if events that may lead to news stories with potentially negative tones have any negative impact on the FDI flows.

Further, we include a full list of variable definitions and data sources. We also include a full list of countries that are used in this study. Lastly, we provide example stories from The Economist, along with their classification for the visibility index.

Institutional considerations

A variety of institutional variables that may capture the costs of investing abroad, the role of political systems, and the quality of governance are likely to be among the important determinants of firms’ decisions to invest abroad. However, the empirical evidence on the effect of institutional variables on FDI is mixed. For example, Aizenman and Spiegel (2006) focus on a measure of institutional inefficiency (using an index of bureaucracy) and find that higher levels of institutional inefficiency have a significant negative impact on the share of FDI in gross fixed capital formation. Yet, Daude and Fratzscher (2008) suggest that, although portfolio investments are relatively sensitive, FDI appears to be insensitive to the quality of institutions in the host country.

We explore the inclusion of a number of institutional variables. While these variables may be important additional controls, they may not show up as significant in our regressions due to the large vector of controls, which include the standard gravity covariates. We incorporate 5 types of institutional variables that have been considered in a wide range of previous literature, including measures of economic freedom (from the Heritage Foundation), civil liberties (from Freedom House), political risk (from the International Country Risk Guide), trade openness (measured as the sum of exports and imports relative to real GDP per capita), and governance (from the Worldwide Governance Indicators).

Table A1 shows the results of including a variety of institutional variables, as well as trade openness, which is often used in the trade literature on FDI to indicate the amount of trade flows as a proxy for trade barriers. In each case, we add the institutional variable along with the explanatory variables included in the baseline regression. The results reported in Table A1 suggest that in most cases the institutional variables included are not significant on their own. Greater trade openness is associated with greater FDI activities as shown in column V. The estimated effect is, in fact, highly significant, but does not change the relationship between news and FDI activity.

Incorporating the cross-section institutional differences does not change the pattern that we find in the data: conditional upon a wide range of covariates, which include institutional variables, greater country visibility is associated with greater FDI outflow from the US to other countries. In fact, institutional variables appear to have little impact on the sign, size, and significance of the coefficient on our country visibility index, with the coefficient on Lagged News Count remaining in the 0.245–0.399 range. The effect of greater country visibility on the FDI reversals, as estimated by the sum of coefficients on Lagged News Count and Negative Flow Dummy * Lagged News Count, continues to be insignificant. (Including the institutional variables in the vector of control covariates, we estimate the model for the FDI data for 2003 and 2004 separately. Trade openness is again the only significant institutional determinant. Though in some cases the estimated coefficient of interest is not as significant, in most cases we find it to be statistically significant. It also remains in the same range as the estimated parameter of interest as reported in Table A1. Detailed estimation results available upon request.)

Table A1. Including institutional control covariates.

Table

Dependent variable: FDI (log absolute value)	I	II	III	IV	V	VI
Lagged News Count (in log)	0.347***	0.245*	0.333**	0.305**	0.399***	0.307**
	(0.133)	(0.142)	(0.135)	(0.155)	(0.132)	(0.136)
GDP (in log)	0.624***	0.692***	0.637***	0.662***	0.759***	0.646***
	(0.098)	(0.125)	(0.101)	(0.140)	(0.120)	(0.097)
Distance (in log)	−0.692***	−0.849***	−0.698***	−0.943***	−0.984***	−0.641***
	(0.202)	(0.212)	(0.203)	(0.236)	(0.237)	(0.209)
GDP Growth Rate	−0.024	0.011	−0.023	−0.036	−0.072**	−0.025
	(0.029)	(0.030)	(0.029)	(0.030)	(0.029)	(0.028)
GDP Per Capita (in log)	0.424***	0.408***	0.408***	0.344**	0.152	0.499***
	(0.106)	(0.146)	(0.120)	(0.140)	(0.124)	(0.157)
Negative Flow Dummy	0.713**	0.479	0.623*	0.161	0.950***	0.613*
	(0.327)	(0.330)	(0.329)	(0.342)	(0.354)	(0.322)
Negative Flow Dummy * Lagged News Count (in log)	−0.469***	−0.388**	−0.444***	−0.267^a	−0.466***	−0.443***
	(0.160)	(0.154)	(0.162)	(0.167)	(0.153)	(0.163)
Time Dummy (for 2004)	0.350*	0.339*	0.370**	0.411**	0.815***	0.354*
	(0.186)	(0.195)	(0.185)	(0.208)	(0.217)	(0.188)
Legal Origin Dummy	0.255	0.346	0.275	0.546*	0.401*	0.293
	(0.240)	(0.242)	(0.240)	(0.279)	(0.239)	(0.238)
Common Language Dummy	−0.331	−0.234	−0.329	−0.293	−0.405	−0.284
	(0.239)	(0.268)	(0.245)	(0.311)	(0.255)	(0.247)
NAFTA Dummy	0.468	0.045	0.532	−0.025	−0.223	0.652
	(0.447)	(0.535)	(0.454)	(0.582)	(0.512)	(0.475)
EU1 (for 15 original members)	0.430	0.268	0.534	0.458	0.129	0.650
	(0.419)	(0.406)	(0.443)	(0.474)	(0.371)	(0.488)
EU2 (for 10 additional members)	−0.846**	−0.985**	−0.760*	−0.792*	−1.388***	−0.669
	(0.386)	(0.421)	(0.404)	(0.431)	(0.424)	(0.418)
EU3 (for candidate countries)	−0.306	−0.289	−0.275	−0.265	−0.518	−0.279
	(0.483)	(0.476)	(0.497)	(0.467)	(0.426)	(0.493)
Economic Freedom		0.017
		(0.016)
Civil Liberties			0.027
			(0.092)
ICRG				0.002
				(0.016)
Trade Openness					0.013***
					(0.003)
WGI Score						−0.237
						(0.290)
Constant	−8.552***	−9.621***	−8.771***	−6.643**	−8.005***	−10.090***
	(2.506)	(2.709)	(2.534)	(2.985)	(2.975)	(2.872)
Number of observations	283	253	281	229	204	282
R^2	0.716	0.725	0.718	0.707	0.777	0.719

Notes: See notes to Table 2 and variable definitions in text. We use ICRG and WGI Score for the year 2003 assuming that they are unchanged in 2004. Other institutional variables are observed for 2003 and 2004.

^aP-value = 0.11.

Excluding large FDI flows

Motivated by a point raised by an anonymous referee, we conduct a robustness check on the baseline estimations. In this robustness check, we only include observations for which the absolute value of FDI inflows or outflows are less than or equal to the value of the 90th percentile. The results remain intact.

The countries with FDI inflows or outflows greater than the 90th percentile value, which are excluded from the full sample in our robustness checks, are the usual suspects. The list includes Australia, Belgium, Brazil, Canada, China, France, Germany, Ireland, Italy, Japan, Korea, Luxembourg, Mexico, the Netherlands, Singapore, Spain, Sweden, Switzerland, and the UK. The result of the estimation using the smaller sample, reported in Table A2 (column I-1), are in line with the baseline results using the full sample: conditional upon a wide range of covariates, greater country visibility contributes to FDI flow from the US to other countries in the smaller sample, but it has no significant effect on the reversal of such flows from those countries to the US.

We then turn to the sub-sample of lower middle- and low-income countries. Excluding the destinations with large flows, country visibility remains a significant determinant of direct investment from the US to developing countries. From the sub-sample of lower middle and low-income countries, we only include observations for which the absolute value of FDI inflows or outflows are less than or equal to the value of the 90th percentile. The countries that are dropped from this sub-sample include: Algeria, Azerbaijan, Brazil, China, Colombia, Ecuador, Egypt, Equatorial Guinea, India, Nigeria, Peru, Philippines, Russia, South Africa, and Thailand. The result of the estimation using the smaller sub-sample, reported in column I-2, are in line with the baseline results using all developing countries. While the estimated parameter for Lagged News Counts is statistically greater than zero, the sum of the estimated parameter of Lagged News Counts and Negative Flow Dummy * Lagged News Count is statistically not different from zero (p-value = 42.28%). GDP, GDP per capita, and distance are also among the significant determinants of FDI flows. This result implies that, conditional upon a wide range of covariates, greater country visibility contributes to FDI flow from the US to developing countries in the smaller sub-sample, but it has no significant effect on the reversal of such flows from those countries to the US.

We finally turn to the sub-sample of high- and upper middle-income countries. Excluding the destinations with large flows, country visibility has no significant effect on direct investment from the US to developed countries. From the sub-sample of high- and upper middle-income countries, we only include observations for which the absolute value of FDI inflows or outflows are less than or equal to the value of the 90th percentile. The countries that are dropped from this sub-sample include: Canada, Germany, Ireland, Japan, Mexico, the Netherlands, Switzerland, and the UK. The results, reported column I-3, imply that greater country visibility has no significant effect on FDI flows between the US and the remaining countries in the sub-sample of high- and upper middle-income countries.

The findings of this robustness test are in line with our baseline findings, suggesting that the baseline results are not driven by outliers with relatively large FDI flows.

Table A2. Robustness test for baseline estimations – excluding large FDI flows^a.

Table

Dependent variable: FDI (log absolute value)	All countries	Lower middle- and low-income countries	High- and upper middle-income countries
	I-1	I-2	I-3
Lagged News Count (in log)	0.324^**	0.363^*	0.0411
	(0.144)	(0.209)	(0.235)
GDP (in log)	0.632^***	0.401^***	0.869^***
	(0.1000)	(0.121)	(0.253)
Distance (in log)	−0.821^***	−1.202^***	−0.626
	(0.214)	(0.288)	(0.506)
GDP Growth Rate	−0.0206	−0.0258	−0.0218
	(0.0300)	(0.0317)	(0.0447)
GDP Per Capita (in log)	0.312^***	0.240^b	0.184
	(0.109)	(0.167)	(0.444)
Negative Flow Dummy	0.718^**	1.093^**	0.394
	(0.336)	(0.434)	(0.758)
Negative Flow Dummy * Lagged News Count (in log)	−0.399^**	−0.535^**	−0.241
	(0.172)	(0.263)	(0.307)
Time Dummy (for 2004)	0.324^*	0.194	0.225
	(0.194)	(0.224)	(0.348)
Legal Origin Dummy	0.253	0.170	1.271^**
	(0.253)	(0.266)	(0.565)
Common Language Dummy	−0.426^*	−0.211	−1.084^*
	(0.242)	(0.259)	(0.644)
NAFTA Dummy			0.679^c
			(0.836)
EU1 (for 15 original members)	−0.264		0.556
	(0.456)		(0.676)
EU2 (for 10 additional members)	−0.626		−0.629
	(0.397)		(0.478)
EU3 (for candidate countries)	−0.178	0.723
	(0.471)	(0.516)
Constant	−6.813^***	2.051	−12.73^***
	(2.573)	(3.995)	(4.227)
Number of observations	253	158	94
R^2	0.593	0.451	0.659

^aFrom each reference sample, the observations for which the absolute value of FDI inflows or outflows are less than or equal to the value of the 90th percentile are included in the above regressions.

^bIn this case, the p-value is 0.153.

^cIn the sub-sample of high- and upper middle-income countries, the FDI flow to Mexico in 2003 is included. This is done because its value in that year is less or equal to the 90th percentile. In 2004, however, FDI flow to Mexico is excluded, as its value in that year is greater than the 90th percentile. That is why NAFTA dummy is included in the regression function.

Standard errors in parentheses: ^*p < 0.10; ^**p < 0.05; ^***p < 0.01.

Armed conflicts

As indicated in the manuscript, we find it a subjective matter to assign negative or positive tone to a given news story. In practice, a given news story may be considered negative for some firms, while the same story may be considered positive for some other firms. Plus, there are often news stories that explore different aspects of a given event, which makes it again impossible for us to assign a negative, positive, or neutral tone to the story. To avoid any measurement error, we focused on the counts (i.e. how many times a given country appears in The Economist) and the type of news stories (i.e. economic, political, or social).

Motivated by the point raised by an anonymous referee, we conduct another test to explore whether the presence of armed conflicts in a given country, which may potentially lead to some news stories with negative tone, may affect the way that country visibility contributes to FDI flows. Conditional upon a wide range of covariates, we find no evidence of such effect.

We employ UCDP/PRIO Armed Conflict Data-set, collected by the Conflict Data Program at Uppsala University and Peace Research Institute in Oslo, to identify the countries that have had armed conflict in 2002 and 2003. This data-set is an extension of the data collection by Gleditsch et al. (2002).

We estimate the parameters of a linear regression function in which absolute value of FDI flows (in log) is used as dependent variable and lagged number of news stories (in log) is used as independent variable along with the other covariates (GDP, distance, GDP growth rate, GDP per capita) and a vector of dummy variables that we use in previous regression functions. We add a dummy variable for lagged armed conflict and an interaction term between lagged armed conflict and lagged number of news stories. For countries with no lagged armed conflict, the parameter associated with Lagged News Counts (in log) captures the effect of country visibility on FDI flows. For countries with lagged armed conflict, the sum of the parameters associated with Lagged News Counts (in log) and Conflict Dummy * Lagged News Count (in log) captures the effect of country visibility on FDI flows.

The results, reported in Table A3, suggest that in countries where there were no lagged conflicts, the effect of country visibility on FDI flows remains essentially the same: greater country visibility contributes to FDI flows. In the results reported in column I-2, where observations from all countries are employed, the estimated parameter for Lagged News Count (in log) is positive and statistically different from zero. We observe almost the same pattern for countries where there were some lagged conflicts reported. The sum of the parameters associated with Lagged News Counts (in log) and Conflict Dummy * Lagged News Count (in log) is statistically different from zero, though the p-value is slightly above the conventional level (p-value = 11.08%). In the results reported in column II-2, where observations from developing countries are employed, the parameter associated with Lagged News Counts (in log) is statistically different from zero, though the p-value is slightly above the conventional level (p-value = 15.4%). However, the sum of the parameters associated with Lagged News Counts (in log) and Conflict Dummy * Lagged News Count (in log) is statistically not different from zero (p-value = 20.26%).

It should be noted that the results reported in column I-1 and II-1 of Table A3 are a bit different from the baseline results reported in Table 2. The reason is that Negative Flow Dummy and Negative Flow Dummy * Lagged News Count (in log), which are included in baseline regressions to capture the effect of country visibility on the reversal of flows, are excluded from the regression function for this test. That is why the precision of the parameter associated with Lagged News Counts (in log) is affected a bit. That said, an interesting pattern emerges, which suggest that events that may potentially lead to news stories with negative tones may have no effect on FDI flows for developing countries. It is also important to note that we find no reversal of sign for this result. Thus, we can only conclude that for developing countries there is not a significant positive effect on FDI flows for these types of news stories.

Table A3. Baseline estimation results, including a dummy for armed conflict and an interaction term.

Table

	All countries^a	All countries	Lower middle- and low-income countries^a	Lower middle- and low-income countries
	I-1	I-2	II-1	II-2
Lagged News Count (in log)	0.243^**	0.224^*	0.280^*	0.246^b
	(0.120)	(0.126)	(0.163)	(0.172)
GDP (in log)	0.649^***	0.648^***	0.573^***	0.569^***
	(0.0967)	(0.0974)	(0.109)	(0.110)
Distance (in log)	−0.673^***	−0.670^***	−0.917^***	−0.912^***
	(0.203)	(0.203)	(0.297)	(0.297)
GDP Growth Rate	−0.0245	−0.0243	−0.0142	−0.0141
	(0.0307)	(0.0301)	(0.0356)	(0.0349)
GDP Per Capita (in log)	0.402^***	0.413^***	0.406^**	0.428^**
	(0.107)	(0.112)	(0.174)	(0.183)
Time Dummy (for 2004)	0.365^*	0.359^*	0.367	0.365
	(0.185)	(0.186)	(0.233)	(0.236)
Legal Origin Dummy	0.207	0.189	0.0163	0.00328
	(0.247)	(0.252)	(0.291)	(0.293)
Common Language Dummy	−0.285	−0.288	−0.103	−0.113
	(0.248)	(0.250)	(0.289)	(0.292)
NAFTA Dummy	0.670	0.732
	(0.431)	(0.451)
EU1 (for 15 original members)	0.525	0.536
	(0.407)	(0.410)
EU2 (for 10 additional members)	−0.814^**	−0.803^**
	(0.391)	(0.397)
EU3 (for candidate countries)	−0.325	−0.386	−0.256	−0.339
	(0.444)	(0.481)	(0.491)	(0.536)
Lagged Armed Conflict Dummy		0.0486		0.0664
		(0.526)		(0.586)
Conflict Dummy * Lagged News Count (in log)		0.0513		0.0739
		(0.165)		(0.236)
Constant	−8.967^***	−9.047^***	−5.132	−5.217
	(2.493)	(2.479)	(3.826)	(3.825)
P-value for the sum of parameters associated with Lagged News Count and Conflict Dummy * Lagged News Count		0.1108		0.2026
Number of observations	283	283	177	177
R^2	0.710	0.710	0.537	0.538

^aThe estimated parameter reported in this column follow the same pattern as the estimated parameter reported in baseline regressions, reported in Table 2 of the manuscript. Since Negative Flow Dummy and Negative Flow Dummy * Lagged News Count (in log), which are included in baseline regressions, are excluded from the regression function for this test, the magnitude of the estimated parameters may be a bit different.

^bIn this case, the p-value is equal to 0.154.

Standard errors in parentheses: ^*p < 0.10; ^**p < 0.05; ^***p < 0.01.

Variable definitions, data sources, and country list

Tables A4, A5, A6, and A7 provide the definition of variables that are used in our baseline estimations, IV estimations, institutional considerations, and the test using armed conflict, respectively. These tables also include the data sources. Table A8 includes a full list of countries used in this study.

Table A4. Variables, definitions, and sources.

Table

Variable name	Variable definition	Source
News Count (2002 and 2003)	Count of stories with primary focus on a particular country; enters as log(Count + 1)	The Economist www.economist.com
FDI (2003 and 2004)	FDI outflow measure provided by the BEA	Bureau of Economic Analysis (BEA) www.bea.gov
	Equity capital flows are net flows, where ‘equity capital outflows occur when a U.S. parent increases its equity investment in one of its existing foreign affiliates or makes a new equity investment in a foreign business enterprise, either by acquiring an existing foreign business or by establishing a new one. Equity capital inflows occur when a U.S. parent reduces its equity interest in an existing affiliate. Intercompany debt flows are of two types: U.S.-parent receivables and U.S.-parent payables. U.S.-parent receivables represent loans that a U.S. parent extends to its foreign affiliate. An outflow on U.S.-parent receivables occurs when the parent extends a new loan to its affiliate; an inflow occurs when an affiliate repays part or all of a loan from its U.S. parent. U.S.-parent payables represent loans that a foreign affiliate extends to its U.S. parent. An outflow on U.S.-parent payables occurs when the parent repays part or all of a loan from its foreign affiliate; an inflow occurs when an affiliate extends a new loan to its U.S. parent. Reinvested earnings are the U.S. parents’ claim on the undistributed after-tax earnings of its foreign affiliate’.	See the BEA definitions: https://www.bea.gov/scb/account_articles/international/0395iid/maintext.htm
GDP per capita	GDP per capita (constant 2000 US$)	World Bank World Development Indicators (WDI) https://go.worldbank.org/U0FSM7AQ40
GDPgrowth rate	GDP growth (annual %)	World Bank World Development Indicators (WDI) https://go.worldbank.org/U0FSM7AQ40
Population	Population, total	World Bank World Development Indicators (WDI) https://go.worldbank.org/U0FSM7AQ40
Income groups	Economies are divided according to 2002 GNI per capita, calculated using the World Bank Atlas method. The groups are: low income, $735 or less; lower middle income, $736–$2,935; upper middle income, $2,936–$9,075; and high income, $9,076 or more	World Bank https://siteresources.worldbank.org/INTRGEP2004/Resources/classification.pdf
Distance	‘Geodesic distances are calculated following the great circle formula, which uses latitudes and longitudes of the most important cities/agglomerations’.	https://www.cepii.fr/anglaisgraph/bdd/distances.htm
Common Language (English)	English listed as the official language or as a primary language	CIA world factbook https://www.cia.gov/library/publications/the-world-factbook/geos/al.html
Legal origin (UK)	UK legal origin	https://mba.tuck.dartmouth.edu/pages/faculty/rafael.laporta/publications.html
Nafta	Membership in Nafta	US, Canada, Mexico
EU1	EU15 Member States: Germany, France, Italy, the Netherlands, Belgium, Luxembourg, Denmark, Ireland, United Kingdom, Greece, Spain, Portugal, Austria, Finland and Sweden	European Union Membership based on date of entry. https://europa.eu/abc/history/2000_today/index_en.html
EU2	New Member States (May 2004): Czech Republic, Cyprus, Estonia, Latvia, Lithuania, Hungary, Malta, Poland, Slovenia and Slovakia
EU3	Candidate Countries (as of May 2004): Bulgaria, Romania and Turkey

Table A5. Instrumental variables: definitions, and sources.

Table

Olympic Medals	Medal counts for Winter Olympics (Salt Lake City, 2002)	International Olympics Committee
Natural Disasters	Number of natural disaster incidents in a country in 2002 and 2003	Center for Research on the Epidemiology of Disasters, School of Public Health at Universitecatholique de Louvain. https://www.cred.be/

Table A6. Institutional variables: definitions, and sources.

Table

Economic Freedom (EF)	We use the simple average of 9 individual freedoms: business freedom, trade freedom, monetary freedom, freedom from government, fiscal freedom, property rights, investment freedom, financial freedom, and freedom from corruption (a 10th component, labor freedom was added in 2005).A higher score indicates greater economic freedom	Heritage Foundation Index of Economic Freedom www.heritage.org
ICRG	We use the overall measure, which is a weighted combination of 12 components, with higher numbers indicating lower risk: Government Stability (12 pts), Socioeconomic Stability (12 pts), Investment Profile (12 pts), Internal Conflict (12 pts), External Conflict (12 pts), Corruption (6 pts), Military in Politics (6 pts), Religion in Politics (6 pts), Law and Order (6 pts), Ethnic Tensions (6 pts), Democratic Accountability (6 pts), Bureaucracy Quality (4 pts)	International Country Risk Guide (ICRG) www.icrgonline.com
Civil Liberties	Political Rights (‘PR’) and Civil Liberties (‘CL’) are measured on a one-to-seven scale, with one representing the highest degree of Freedom and seven the lowest	Freedom House, Freedom in the World Country Ratings www.freedomhouse.org
Trade Openness	(Exports + Imports)/(real GDP per capita)	Penn World Tables (Heston et al. 2006).
WGI Score	The six dimensions of governance are constructed using an unobserved components methodology described in Kaufmann, Kraay, and Mastruzzi (2008). The six governance indicators are measured in units ranging from about −2.5 to 2.5, with higher values corresponding to better governance outcomes. The six dimensions are: Voice and Accountability, Political Stability & Absence of Violence/Terrorism, Government Effectiveness, Regulatory Quality, Rule of Law, and Control of Corruption.	Worldwide Governance Indicators www.govindicators.org

Table A7. Armed conflict: definition, and source.

Table

Conflict Dummy

This dummy variable is based on armed conflict information. For 2002 and 2003, we assign value one to a given country provided that at least one armed conflict is reported for that country. Otherwise, we assign zero

UCDP/PRIO Armed Conflict Data-set https://www.prio.org/Data/Armed-Conflict/UCDP-PRIO/

Table A8. Country list.

Table

Albania	Cyprus	Kyrgyzstan	Russia
Algeria	Czech Republic	Laos	Rwanda
Angola	Denmark	Latvia	Samoa
Antigua	Djibouti	Lebanon	Saudi Arabia
Argentina	Dominican Republic	Lesotho	Senegal
Armenia	Ecuador	Liberia	Serbia Montenegro
Aruba	Egypt	Libya	Sierra Leone
Australia	El Salvador	Luxembourg	Singapore
Austria	Equatorial Guinea	Macedonia	Slovak Republic
Azerbaijan	Eritrea	Madagascar	Slovenia
Bahrain	Estonia	Malawi	Solomon Islands
Bangladesh	Ethiopia	Malaysia	South Africa
Belarus	Finland	Maldives	Spain
Belgium	Fiji	Marshall Islands	Sri Lanka
Belize	France	Mauritania	St. Kitts Nevis
Benin	Gabon	Mauritius	St. Vincent Grenadines
Bhutan	Gambia	Mexico	Sudan
Bolivia	Georgia	Moldova	Suriname
Bosnia Herzegovina	Germany	Mongolia	Sweden
Botswana	Ghana	Morocco	Switzerland
Brazil	Greece	Mozambique	Tajikistan
Brunei	Grenada	Namibia	Tanzania
Bulgaria	Guatemala	Nepal	Thailand
Burkina Faso	Guyana	Netherlands	Togo
Burundi	Haiti	New Zealand	Tonga
Cambodia	Honduras	Nicaragua	Trinidad Tobago
Cameroon	Hong Kong	Niger	Tunisia
Canada	Hungary	Nigeria	Turkey
Central African Republic	Iceland	Norway	Uganda
Chad	India	Oman	Ukraine
Chile	Iran	Pakistan	United Arab Emirates
China	Ireland	Panama	United Kingdom
Colombia	Israel	Papua New Guinea	Uruguay
Comoros	Italy	Paraguay	Uzbekistan
Congo, Dem. Rep.	Jamaica	Peru	Venezuela
Congo, Republic of	Japan	Philippines	Vietnam
Costa Rica	Kazakhstan	Poland	Yemen
Cote d`Ivoire	Kenya	Portugal	Zambia
Croatia	Korea, Republic of	Romania	Zimbabwe

Note: 156 countries with both the count variable (2002 and 2003) and FDI flows (either 2003 or 2004 or both).

Example stories and classification within the visibility index

Norway’s far north

What about rights for reindeer?

The Sami people are unhappy

Oct 3rd 2002 | KARASJOK, ARCTIC NORWAY | From the print edition

•	OVER the past decade, Norwegians have earned a reputation as peacemakers: in Guatemala, for a while in the Middle East, now in Sri Lanka. Yet they are still to sort out an ethnic squabble of their own, between Norway’s fair-haired 4.5 m majority and its 45,000-strong Sami minority up in the Arctic.
•	Categorized as Norway – Political

AIDS and South African business

Strategic caring

Firms strategise about AIDS

Oct 3rd 2002 | johannesburg | From the print edition

•

GREATER transparency for American firms may mean little more than describing what outlandish perks the boss gets. In South Africa, the disclosures will make much grimmer reading. Johannesburg’s stock exchange is currently ‘considering what form of accounting standard is needed from listed companies’. Translated: in six months, firms listed on the exchange will probably be required to publish anti-AIDS policies. Before long, they will have to spell out how AIDS affects their business, markets and workers, and show how they are fighting it.

•

Categorized as South Africa – Economic; Social

Bank reform in Japan

The Takenaka challenge

A new regulator, but will he wield a new broom?

Oct 3rd 2002 | Tokyo | From the print edition

•

IS THIS the start of genuine reform, or a bad-dream repeat of 1998–1999, when hopes were also raised only to be dashed? The prime minister’s dismissal on September 30th of Hakuo Yanagisawa, the minister who was supposed to be sorting out Japan’s rotten banks, surprised many. The move, the main feature of Junichiro Koizumi’s first cabinet reshuffle since he took office last year, seems to offer a chance to speed up bank reform.

•

Categorized as Japan – Economic

Mexico’s film industry

The Mexicans are coming!

Americans take to Mexico’s movies

Oct 3rd 2002 | mexico city | From the print edition

•	AMERICANS have developed a taste for many things Mexican, including tacos, salsa and maids. But the latest successful Mexican export is more surprising: arty films. Over the past few years, Mexico’s film industry has enjoyed a small renaissance. Americans have started to take note.
•	Categorized as Mexico – Social