Dean Shaun Curry April 2012 Registration No:

Transcription

1 University Of Essex Undergraduate Final Year Project How Do Exchange Rate Levels And Volatility Affect Levels Of Foreign Direct Investment? What Would The Effects Be On FDI Flows Should A Recipient Adopt A Fixed Currency Policy? Evidence From US FDI Outflows. Abstract Dean Shaun Curry April 2012 Registration No: I explore disaggregated US FDI outflows by industry components, extending from previous literature such as Kiyota and Urata (2002) and Froot and Stein (1999). Empirical analysis of US FDI flows across industries show heterogeneous Foreign Direct Investment (FDI) decisions across industries, when facing exchange rate volatility. I find an appreciation of a parent countries currency encourages FDI flows, however no significant relationship is observed. Empirical testing of annual US FDI stocks support that higher levels of exchange rate volatility suppressed incentives to invest overseas between 1999 and The magnitude of this affect is greatest in the manufacturing industry. I extend my model to determine whether this divergence of reactions to exchange rate volatility, are a result of sunk costs enhancing risk averse behaviour from Multinational Corporations (MNC s). I find that the theoretical prediction outlined of sunk costs effects on FDI, are not supported directly in the data; However economic intuition of the results draws some interesting conclusions. 1. Introduction One of the fastest growing forms of investment in the world economy is Foreign Direct Investment (FDI). In 2008 alone, developed countries experienced an inflow of FDI of more than 1 trillion U.S. dollars (OECD 2010) 1. FDI is defined as the acquisition of foreign capital, structures and securities by a domestic firm, known as a multinational corporation (MNC). Foreign direct investment has been under much scrutiny over the years by international economists, due to the benefits associated with FDI both for the MNC and the recipient economy. It is claimed that FDI triggers technology spill overs, assists human capital formation, contributes to international trade integration, helps create a more competitive business environment and enhances enterprise development (OECD 1 Data Source: World Bank. As taken from Jeanerette (2011)

2 report, 2002). There are many factors which are believed to attract FDI into a country such as taxes, market size, political stability and the openness of the economy (Walsh and Yu 2010). One causal factor which is attracting a growing coverage of literature is the effects of exchange rate volatility, as an increasing number of countries adopt floating exchange rates. Exchange rate volatility increases uncertainty surrounding overseas investments and raises the variance of expected costs and profits faced by the MNC. This rise in uncertainty has been thought to suppress FDI inflows as MNC s are faced with an opportunity cost of not waiting, before committing large financial capital flows overseas (Campa 1993). The full extent of the current debt crisis surrounding the Euro zone is yet to become clear, the World Bank has cut global growth forecasts for 2012 from 2.7% to 1.4%. A global crisis as such generates economic uncertainty of many types; including exchange rate uncertainty, potentially distorting FDI inflows. As the global economy continues to integrate and FDI flows become increasingly influential on economic growth and development, the importance of policy makers to be able to accurately predict the effects of exchange rate volatility on FDI inflows has become increasingly essential. However past literature has drawn conflicting conclusions regarding the effects of exchange rate volatility. Cushman (1998) found a positive relationship between exchange rate volatility and FDI, whereas Campa (1993) witnessed a negative link between the two. Kiyota and Urata (2002) suggested that national level data is not adequate to analyse the exchange rate relationship as, such analysis masks impacts of exchange rate volatility among industries by possibly offsetting the impacts across industries (Kiyota and Urata, 2002, p.1). I will explore whether exchange rate levels and their volatilities have independent effects on FDI flows across 8 US industries. This will enable to me to shed light on the effects of adopting a pegged exchange rate, dependant on the industry components a country receives. Using a proxy for sunk costs, I will extend my model to assess whether unrecoverable costs play a significant role in MNC s reactions to exchange rate volatility. One of the largest distributors of FDI flows is the United States (US). The US is home to some of the largest multinationals in the global economy, as a result US FDI outflows account for around 16% of total world flows. Therefore in order to shed light on the relationship between exchange rate volatility and FDI flows, I will be analysing disaggregated US FDI outflows by industry, between 2000 and Over the last 10 years there has been a clear relative increase in US overseas investments of finance, insurance and non-bank holding MNC s, as the US continues to develop its financial sector. US FDI outflows from these industries, particularly since 2005 onwards, have been considerably large. FDI from non-bank holding, finance and insurance MNC s have grown by 117% and 76% respectively, in 2 P a g e

3 comparison to the total of remaining US FDI outflows which have only grown by 44% across the five years. Cumulative US FDI stocks by industry between 2000 and 2010 are presented in Figure 1 2 Industry Key MAN: Total Manufacturi ng SER: Professional, scientific, and technical services FIN: Finance and insurance MIN: Mining WHO: Wholesale Trade UTI: Utilities INF: Information DEP: Depository Institutions Regarding aggregate FDI flows in figure 2, there appears to be two characteristics to note of FDI leaving the US. Levels of FDI leaving the US have been trending upwards across the dataset. Flows in 2010 are almost 7 times that of what they were in Furthermore from 2003 onwards there appears to be a cyclical pattern to FDI flows, as suggested by Yeyati Panizza and Stein (2002). US FDI flows fell sharply in 2005 and I believe the latter to be cyclically linked to the recent global recession in 2008; resulting in heightened uncertainty surrounding investment decisions. 2 Data Source for Figure 1 and Figure 2: Bureau of Economic Analysis; U.S. Direct Investment Position Abroad on a Historical-Cost Basis 3 P a g e

4 Another interesting property observed in the data, is that the manufacturing component of FDI appeared the most suppressed during these periods of decline. In contrast, FDI from other industries mentioned earlier such as finance, insurance and nonbank holding companies continued to grow; suggesting that characteristics of FDI are far from homogenous across industries. This paper will take the following format; Section [2] discusses previous literature analysing exchange rates effects on FDI, summarising conclusions and the data employed to do so. Section [2.2] discusses similarly how exchange rate volatility effects FDI, A summary of this literature is presented in table format at the end of the section (Section [2.3]). Assumptions, hypothesis and theoretical background are outlined within Section [3]. Section [4] discusses potential serial correlation, heteroskedacity and trends, along with how they were treated prior to empirical testing. Section [5] discusses preparation of the empirical model. Section [6] presents the empirical results along with analysis. Section [7] extends my model to test whether sunk costs effects reactions to exchange rate volatility. Finally Sections [9] and [10] critically asses the successfulness of empirical work and finishes with some concluding remarks respectively. 2. Exchange Rates and FDI Literature Early criticism of claimed causality between exchange rates and FDI, under the assumption of perfect capital mobility, was that if a German has an advantage purchasing a particular U. S. asset with marks, why can t an American with access to global capital markets borrow marks (at the same opportunity cost as the German) and avail himself of the same advantage? (Froot and Stein, 1991, p.1). However with the reality of perfect capital markets far from accurate and most MNC s operate in various currencies across the globe, literature of exchange rate movements have been fast growing. The two exchange rate components heavily discussed in previous literature are exchange rate levels and their volatility. Exchange rate movements actively distort relative wealth and costs for MNC s, higher levels of exchange rate volatility poses additional risks to investors. I will firstly discuss the standings of the former, and then move onto literature of the latter. Klein and Rosengren (1992) observed inflows to the US arriving from seven industrial countries including Japan, Canada and various industrial economies from Europe. During appreciation of the parent currency, there were observed increases FDI outflows. Of the two mechanisms suggested to link exchange rates to FDI, Klein and Rosengren (1992) only observed a causal relationship through the relative wealth mechanism and that there was no evidence of relative labour costs playing a significant role. Klien and Rosengren (1992) introduced dummy variables for each country in order to deal with serial correlation. Similarly, Goldberg and Kolstad (1994) found that a depreciation of the domestic currency will suppress outwards FDI flows. These effects were generally neither large nor for the most part statistically significant and that any relationship was highly dependent on a high 4 P a g e

5 elasticity of demand for investment assets (Goldberg and Kolstad 1994). Blonigen (1997) touched upon the idea of sunk costs adjusting MNC s investment decisions. Blonigen (1997) assessed FDI flows across the acquisition of different forms of assets, particularly firm specific assets. Empirical testing supported Froot and Stein (1991) and Klein and Rosengren (1992), an appreciation of the parent currency increased FD outflows as relative prices of foreign assets were reduced. Using industry level data of investment between Japanese and American firms, Blonigen (1997) concluded that currency depreciations ease foreign firms' purchases of U.S. host-country technological assets, the majority of these changes were observed from industries with higher levels of firm specific assets. It is likely that levels of firm-specific assets are similar to those of sunk costs. Froot and Stein (1991) analysed effects of exchange rate movements on quarterly FDI inflows across the window 1973 to Froot and Stein (1991) disaggregated FDI flows by industry and found that an appreciation of the dollar increased a MNC's propensity to directly invest abroad. Significant negative relationships were observed across the chemical, machinery and real estate industries; all of which I would expect to engage in more permanent investments. 2.2 Exchange Rate Volatility and FDI Literature Cushman (1985) kicked off the volatility debate investigating the effects of exchange rate risk. Cushman (1985) assessed the effects across four schemes; depending upon where inputs were purchased, output was produced, financial capital was acquired and finally where output was sold. Cushman (1985) witnessed that higher exchange rate volatility increased FDI from the US entering Canada, Japan and Europe. Goldberg and Kolstad (1994) supported Cushman ( ) distinguishing between the three different risk characteristics of firms; risk averse, neutral and loving. Goldberg and Kolstad (1994) employed quarterly bilateral data of US FDI entering the UK, Japan and Canada. Exchange rate uncertainty was found to increase levels foreign direct investment from risk averse multinationals when uncertainty is correlated with export demand shocks within the foreign market (Goldberg and Kolstad 1994). Campa (1993) observed trends in the wholesale markets which conflicted with Cushman ( ) and Goldberg and Kolstad (1994). Campa (1993) found that exchange rate volatility reduced FDI flows, as an increase in exchange rate volatility suppresses a firms desire to invest abroad. It was suggested that the option value of waiting rises when facing exchange rate uncertainty (and therefore the opportunity cost of investing rises). Industries in which firms obtain large amounts of physical and intangible assets were suggested to have possessed the most volatile responses to volatility. Benassy, Fontagne and Lahreche (2001) found a similar trend in FDI stocks from 17 developed countries in developing economies between 1984 and Benassy et al (2001) obtained 5 P a g e

6 quarterly nominal exchange rate variation as a measurement of volatility and employed real FDI stocks data from OECD. More recent literature has attempted to explain the contradicting results obtained by earlier papers. Heterogeneity theories have been employed in explaining ambiguous results from past empirical analysis guilty of using national aggregated data. Barrel, Gottschalk and Hall (2007) employed Tobin s q theory of investment to explain the observed US FDI outflows into Europe. US firms appeared risk averse with volatility suppressing outwards FDI flows. Barrel et al (2007) found that market structure, a proxy of market power held by the MNC, had no significant impact on responsiveness to exchange rate volatility. Jeanneret (2011) used heterogeneity of productivity to explain the aggregation issues, building on work by Melitz (2003) and Bernard et al (2003). Jeanneret (2011) assessed the trade off between direct investment and exporting as methods of supplying the foreign market, using a large panel data set of Outwards FDI for 27 OECD countries through a 20 year period ( ). Jeanneret (2011) concluded that less productive firms are more inclined to relocate production (therefore increase FDI) when exchange rate volatility is low, whereas more efficient firms tend to invest abroad when the level of uncertainty is higher (Jeanneret 2011, pg.11). A U-shaped relationship between FDI and uncertainty was observed. Heterogeneity of conditions across industries has also been tested to effect investment decisions. Kiyota and Urata (2002) followed the framework of Froot and Stein (1991) and Benassy et al (2001). By disaggregating FDI down to its industry components, Kiyota and Urata (2002) tested for individual characteristics across industries which influence firms reactions to exchange rate volatility. Kiyota and Urata (2002) observed FDI outflows leaving the US and Japan across eleven different industries from A negative significant relationship with volatility was observed across aggregated FDI, with 10 /11 industries providing significant coefficients on volatility for Japanese FDI and 8/11 for the US. The largest magnitude was measured on heavy machinery and transport equipment, finance and food industries failed to provide significant results in Japan and the US respectively. 6 P a g e

7 2.3 Summary Table Of Previous Studies Paper Dependant Variable Period Window Effect Of Recipient Exchange Rate Appreciation Exchange Rate Volatility Effects Other Factors Considered In Empirical Testing Cushman (1985) US outflow to Japan, Canada and Europe Ambiguous Positive Host market demand Cushman (1988) US inflow, arriving from Japan Canada and Europe Ambiguous Positive Host market demand Froot and Stein (1991) Quarterly US FDI inflows Negative, lowering of domestic wealth N/A Disaggregated FDI Campa (1993) Goldberg and Kolstad (1994) Benassy, Fontagne and Lahreche (2001) Barrel et al. (2007) FDI inflow into 61 US wholesale industries Bilateral US outflow into UK, Japan and Canada Developed countries Stocks in developing countries Outwards FDI at historical cost N/A Negative Negative, yet mainly small insignificant Strongly Negative Positive For Oil Producing Countries (OPEC) Positive Negative N/A Negative Option Value Of Waiting Assets Required Risk Aversity Competitiveness, Openness, Distance Market power. Tobin s q model. Blonigen (1997) Japanese acquisitions in the United States Negative (Particularly to firm specific Assets ) N/A Firm Specific Assets Jeanerett (2011) Log of FDI flows Negative Klein and Rosengren (1992) Kiyota and Urata (2002) Outwards from seven industrial countries US and Japanese outflow across 11 industries Negative under low levels of uncertainty and Positive under higher levels Negative N/A Negative Largely negative across capital intensive industries Notes Table is based on Table 1 Major Regression Results of the Previous Empirical Studies from Kiyota and Urata(2002) Firm productivity, Free Trade Agreements, U-shaped volatility effects Relative wealth. Relative Wages. Cumulative FDI. Distance. Disaggregated FDI. 7 P a g e

8 3. Assumptions And Hypothesis Throughout my analysis I will be under the assumption that all agents in the model are risk averse. A risk averse investor is defined as an investor who, when faced with two investments with a similar expected return but different risks, will always prefer the investment with the lower risk (Goldberg and Kolstad 1994). Multinationals make investment decisions at the end of each year with access to perfect information of all currency market values and their corresponding volatilities. Similarly to the model used in Cushman ( ) firms have two production possibilities, either domestically or to invest into production abroad however assume that no firm is yet to invest abroad and each MNC can only make one investment. Finally I will assume perfect competition across all countries resulting in uniform prices across all international markets, so that there are no incentives generated through market conditions. Multinational corporations will be encouraged into FDI during an appreciation of the parent currency. The value of foreign currency would therefore be negatively correlated with MNC s propensity to invest. Previous literature has outlined the two main channels in which this occurs. Firstly the relative wage effect, a depreciation of a countries exchange rate reduces the relative costs of labour and therefore would attract FDI through lower costs for the multinational and therefore higher profits. Additionally the relative wealth effects which are created through the existence imperfect capital markets (Froot and Stein 1991). Imperfect capital markets result in a relative depreciation of the recipient country altering the relative wealth of multinationals, which as a result influences FDI when the purchase of an asset requires the provision of some internally generated funds (Klein and Rosengren 2002). FDI flows would therefore be encouraged through enabling the multinational to have an advantage over the native competition through the decreasing of local investor s relative wealth. With all firms assumed to be rational, when deciding to invest I would expect firms utility to increase along with expected profits and to diminish as the variance of expected profits rise by definition of risk aversity. The higher the exchange rate volatility, the larger the opportunity cost to the capital intensive MNC s as the option value of waiting grows (Campa 1993) due to acquisition of less liquid assets. Take a less capital intensive firm, uncertainty becomes less of an opportunity cost as upon exit of the market only redundancy payments would be accrued upon exit. This could potentially allow for durations of exchange rate uncertainty to be viewed more as an opportunity for higher profits for MNC s, Similarly to Cushman ( ) and Goldberg and Kolstad (1994). Figure 3 below displays my prediction of FDI as a function of industry capital required within their respective industries, the slopes therefore represent responsiveness of direct investment to volatility in 8 P a g e

9 exchange rates, ceteris paribus. Assume that Capital Intensity Industry C > Capital Intensity Industry B > Capital Intensity Industry A. Figure 3: Predicted FDI Flows As A Function Of Volatility As figure 3 illustrates, I would expect an increased volatility of a recipient currency (V 1 V 2 ) to have a variety of effects on FDI flows across different industries. I would predict to see the relationship labelled Industry C to represent the Mining industry due to the huge fixed structures set up of a mine; therefore the increase in volatility would have a huge suppressing effect (FDI c -FDI* C ). For similar reasons Industry B represent a similar relationship of manufacturing FDI, a large suppressing volatility through firm specific technology and capital required for production. Finally I would assign industry A to represent that of finance multinationals, with the increase in volatility having a minute effect upon MNC s investment decisions due to the lack of physical capital required. 4. Trend, Serial Correlation And Heteroskedacity Issues FDI outflows across the globe have a positive trend across time, the same is observed in US FDI flows (see Figures 1 and 2). It is therefore clear that there are other factors across time which significantly distort MNC's propensity to invest abroad. Possible trending factors are development of technology, stronger economic relations or improved information available. Additionally recall the cyclical behaviour of FDI flows presented in figure 2. Time trends of data can create spurious issues 9 P a g e

10 when conducting empirical analysis, in which two trending variables can be mistaken for a causal relationship. One method employed to solve spurious regression problems throughout previous literature, such as Klein and Rosengren (1994) and Kiyota and Urata (2002), was to introduce a time trend variable to account for the growth of unobserved factors across time such as productivity of labour, communications and technology. A variable (here T ) which represents increases in time between observations (e.g. T{1,2,3 }), formally presented in equation [1] below. Yt = β0 + β1xt + β2t + ut [1] Introducing a time trend is the equivalent to shifting the intercept of the relationship between the dependant and independent variables each period. This method assumes a uniform annual intercept shift,; which when considering the cyclical patterns observed, would not be adequate. Due to my dataset containing sufficient degrees of freedom, I was able to employ time fixed effects to my model. Time fixed effects are accounted for with the use of a dummy variable for each year. This again represents a shift in the intercept, however now enabling each shift to be independent. Time fixed effects in my regressions will be employed formally as in regression [2]. Yt = β0 + β1xt + [β2(1999) + β3(2000) + β4(2001) + + β12(2009)] + ut [2] Where (1999), (2000),, (2009) represent dummy variables for each respective year in the sample. These time fixed effects in my regression will absorb the variation caused by exogenous trends in FDI across time such as technological shocks and effects of the business cycle in the US, preventing any spurious trends creating biases throughout. In order for the coefficients and standard errors to be unbiased, we rely on the OLS residuals to be independently identically distributed (IID), that is that we expect that on average, the values of our error term is zero, and also are independent of the independent variables, formally presented in equation [3]. E(u X1, X2,, Xn) =0 [3] Panel data observes multiple agents across a period of time. The data set I have constructed contains three dimensions; country, industry and year (see table 1.1 in appendix). One issue when using panel data is that In large panel data sets such as these, it is likely that the residuals could be correlated within these dimensions, possibly across industry or across countries, resulting in generated OLS standard errors to be potentially biased (Peterson, 2008). Peterson (2008) reported that OLS residuals were downwards biased within a panel data set when regular standard errors are employed. There is a high probability that my data set suffers serial correlation across recipient countries due to fixed factors 10 P a g e

11 across countries. Therefore to prevent my standard errors being subject to bias, I will be computing OLS regressions using clustered robust standard errors. Clustered robust standard errors allow the model to account for serial correlation and heteroskedacity across selected a selected dimensions Peterson (2008). I will be adjusting my OLS regressions for serial correlation and hetroskedacity within countries. Standard errors presented in results will be adjusted for 35 clusters, where the 35 clusters represent the 35 recipient countries in my sample. 5. Econometric Methodology Throughout previous papers there has been a variety of methods used to measure the volatility of exchange rates. Chowdhury (2008) used the conditional variance of the log of each country s real effective exchange using a GARCH model. However similarly to Goldberg and Kolstad s (1995) and Kiyota and Urata (2002), annual standard deviation will serve as a proxy measurement of exchange rate volatility. That is, the exchange rate volatility variable included my model is given by the square root of each yearly variance. A selection of average recipient currencies volatilities are presented in table 1 below 3 : (note that values in table 1 are not the values used in my regressions, they are the averages of the 10 years in the dataset). Table 1: Average Exchange Rate Volatility Currency Average Annual Volatility (between ) Canadian Dollar Euro Japanese Yen Chinese Yuan Danish Krone Pound Sterling Russian Ruble Mexican Peso Singapore Dollar Indian Rupee Korean Won Brazilian Real US Dollar 0 South African Rand Empirical analysis of FDI and exchange rate volatility was conducted using Ordinary Least Squares (OLS) regressions on Stata (software version 12.1). As mentioned in section 4, clustered robust standard errors were employed meaning all standard errors presented are adjusted for clusters of 35 countries. Distance, recipient GDP and FDI in the regressions were logarithms taken of the value collected. Logarithms reduce the effects of any anomalies in the data, smooth out the time 3 Note: Not all currencies included in empirical analysis are presented in Table 1 11 P a g e

12 series variables (especially GDP and FDI) and remove any scaling issues from arising. Additionally I do not believe these variables to have a level effect on FDI outflows, but rather a proportional effect. Using a log-log model enables elasticises with FDI to be easily extracted (i.e. a 1% change in GDP results in a β 1 % change in FDI outflow ). The full regression which I will be running individually across each of the 8 industries therefore is as shown in [4]. LogFDI t = β 0 + β 1 (XRATE t XRATE t-1 )+β 2 XRATEVOL t +β 3 LogDISTANCE t +β 4 Log(Real)GDP t + ε [4] I introduced distance and real GDP into empirical analysis due to the gravity model approach of FDI suggested by Blonigen and Piger (2011) who suggest distance, parent GDP, recipient GDP and other gravity variables are the base causal variables of FDI. In contrast only include real recipient GDP and distance are introduced as my FDI flows are from the US. Therefore only variation will be present across years, which are already mopped up by time fixed effects. As suggested by the gravity model employed in previous literature such as Blonigen and Piger (2011), I will expect to see distance to be negatively correlated with FDI flows from the US and GDP to be positively correlated with outflows (I will explain these later on). The FDI stocks data collected represented foreign direct investment position by US multinationals, it is a cumulative measure across years of FDI stock. In order to measure FDI flows, I took the logarithm of the difference between years (Log[FDI t - FDI t-1 ]). This gives me the log of FDI outflows and prevents past investment decisions creating biased results. Similarly, I employed changes in exchange rates as a representative of altering of relative costs of production, as the nominal exchange rate level provides no indication of incentives. Using the difference of the logarithms of the exchange rate presents percentage change in exchange rates from the previous. That is delta XRATE will be given by (LogXRATE t - LogXRATE t-1 ), the percentage change in exchange rate from the previous year Full Data Description And Descriptive Statistics Available In Data Appendix I would expect to see higher responsiveness to exchange rate volatility and exchange rates from capital intensive industries. Industries such as manufacturing and mining I would expect to see a larger magnitude of β1 and β 2 from equation 4, suggesting a smaller change in exchange rates and exchange rate volatility are highly influential on FDI outflows. I would expect small insignificant coefficients for the finance and insurance industry. As mentioned regression [4] will be run individually for each of the 7 industries considered in figure 1, clustered standard errors are employed for each regression run in order to prevent serial correlation and hetroskedacity from creating any bias of my standard errors. Full lists of coefficients for each industry are presented in table 2 below. 12 P a g e

13 6. Empirical Results Table 2: Regressions By Industry. [DEPENDANT VARIABLE: Logarithm Of Change In FDI Stocks] Industry Independent Variables _CONS ΔXRATE ALL MAN SER FIN MIN WHO UTI INF HOL (4.0523) (0.5939) (4.7155) (0.6856) ** (4.553) (1.2839) (6.6503) (1.4712) (8.4833) (1.422) (5.4638) (1.2029) ( ) ** (4.7214) (7.7433) (0.8259) (9.4543) (3.7056) XRATEVOL t ** (0.0005) *** (0.0025) *** (0.0006) (0.0009) * (0.0055) *** (0.0004) (2.3361) (0.0008) (0.001) lndistance *** (.2075) *** (0.2241) *** (0.2356) *** (0.334) (0.6703) *** (0.188) ** (0.5194) ** (0.3644) *** (0.2872) lngdp t Time Effects Fixed Number Of Observations (N) 0.405*** (0.1318) *** (0.1293) *** (0.1462) (0.2373) (0.2757) (0.1648) (0.5945) 0.497** (0.2338) Yes Yes Yes Yes Yes Yes Yes Yes Yes R (0.3321) Notes 1. Robust standard errors are presented in parenthesis (adjusted for country clusters ). 2. *, ** and *** denote p<0.10, p<0.05 and p<0.01 respectively. 3. Time Fixed Effects denotes a dummy variable for each year (barring 2010) that is [2000,2001,,2009]. 4. lngdp denotes the logarithm of the nominal GDP of recipient nation 5. ΔXRATE denotes the differences in logarithms of exchange rates between the current period (t) and the previous period (t-1) Industry Key ALL: All industries aggregated MAN: Total Manufacturing SER: Professional, scientific, and technical services FIN: Finance and insurance MIN: Mining WHO: Wholesale Trade UTI: Utilities INF: Information HOL: Non-Bank holding companies Appreciating foreign exchange rates discouraged FDI flows as found by many previous papers. All industries possessed a positive coefficient for appreciations of the dollar (barring mining, which will be discussed later in this section). Considering all industries aggregated from table 2 (ALL), if the dollar was to appreciate by 10% then I would predict this to increase US FDI flows from the US by approximately 8%. In contrast to past literature, there was no evidence that fluctuations in exchange rates significantly alter MNC s FDI decisions. Insignificant results were observed throughout all industries except for utilities 4, another strange result to note is from the mining industry. Overall the relationship observed of exchange rates do loosely support the relative wealth and labour cost effects proposed by Klein and Rosengren (1992). As the US Dollar appreciates, foreign labour costs relatively fall, attracting multinationals seeking productive efficiencies. Additionally an appreciating dollar raises relative wealth, allowing the multinational with a cost advantage over local 2 Note the utilities industry (UTI) contains 15 observations, therefore results should be generally disregarded. 13 P a g e

14 competition. Suppression was observed most predominantly in the services industry and least in the wholesale industry. The insignificance of nominal exchange rates are suggestion firms are only concerned with real exchange rate movements. Appreciations of the US Dollar could potentially be from inflation in the foreign economy, removing relative wealth advantages held by the US MNC. As suggested by the gravity model of FDI, distance between recipient and the US played a significant result in MNC s international investment decisions. The distance between recipient and host unsurprisingly appears to be negatively related to FDI flows. The model predicts a strong relationship between the two, ceterus paribus, I would expect a 10% variation in distance accompanied by 8.1% difference in flows of FDI from the US. Benassy et al (2001) also found a strong negative relationship and suggested distance to represent primarily a proxy for transportation costs. However in contrast, I believe that the negative relationship is also accounting for higher uncertainty linked with long distant investments. Generally I would assume that the levels of information regarding economic conditions are higher for more local neighbour economies. Knowledge of wages, industry laws, levels of competition etc all of which are vital for MNC s, will be relatively uncertain within distant countries. Taking the idea of risk averse investors, higher levels of uncertainty would suppress incentives to invest as multinationals would seek more stable returns from an investment. It is no surprise that the finance and insurance FDI are most discouraged regarding distance; with premiums and loans decisions relying upon information available on agents. It would be estimated that the same 10% distance difference would actually suppress finance and insurance FDI flows by 11.1% (compared to the 8.1% above when considering all industries aggregated). One major technological breakthrough which potentially will diminish this negative relationship is the internet. Communication costs between headquarters and foreign plants are significantly lowered as the internet continues to develop, additionally information on agents becomes increasingly cheap and simple to obtain. The real gross domestic product (GDP) of the recipient country appeared strongly influential in FDI decisions by US multinationals. My results suggest that GDP significantly influenced US MNC s investment decisions. The effect of recipient real GDP had a consistent effect on FDI inflows to the respective country. β 4 from [4] was positive throughout and possessed a p-value smaller than 0.10 across all industries. Higher levels of GDP suggest a stable efficient environment for investments. Higher GDP would signal to MNC s that the economy possesses high labour productivity, low interest rates and efficient financial markets; all of which lower production costs for MNC s. In addition, higher levels of real GDP represent greater purchasing power of the population. If the MNC intends to supply the foreign market, assuming they produce a normal good, would expect a higher demand due to the income effects. A real GDP growth of 10% would increase of FDI flows from the US by roughly 4.1%. However the industry most influenced by GDP levels was the services industry. My model predicts 14 P a g e

15 that, should a country manage to increase GDP by 10%, then I would expect to see FDI inflows from the US services industry increase by approximately 7.8%. Results from [4] support my hypothesis that there are a significant variety of reactions to exchange rate volatility across industries. However as mentioned earlier, one surprising relationship observed was the affects of exchange rate volatility on the US mining industry. The results from table 2 suggest that higher levels of exchange rate volatility increased US multinationals FDI stocks from the mining industry. The opposite of what I would expect the mining industry due to high levels of required capital upon start up. The model regressed in table 2 suggests that a one point increase in volatility increases FDI flows by 1%. I believe this to be due to the nature of the mining industry, MNC s would only mine where there are valuable materials to be mined. MNC s will base investments predominantly upon where is believed to withhold the most valuable materials to be mined. This also explains the insignificance of other variables observed when considering the mining industry. It is also generally the case that most productive mining locations, are contained within less economically developed countries. Take for example one of the largest mining economies, South Africa. South Africa s average inflation and unemployment rates are approximately 10% and 26.38% respectively. When we consider a more developed economy like the US, average inflation and unemployment rates are a lower 3.38% and 5.7% respectively 5. It is common for economies with unstable macroeconomic factors to also suffer from highly volatile currency. As a result, US mining MNC s would appear to invest into countries of higher exchange rate volatility. However I do not believe there to be any causation between exchange rate volatility and FDI in the mining industry. Small yet highly significant coefficients were observed across the services and wholesale industries. Both share the same relationship of a one point rise in volatility results in a suppression of FDI in both industries by 0.22%. An even smaller relationship was present across finance, insurance and non-bank holding MNC s. A 1 point rise in exchange rate volatility is only predicted to finance and insurance FDI inflows by a tiny 0.01% and non-bank holding FDI by 0.15%. Note that the coefficient in the finance industry was also highly insignificant. This result is as predicted, the finance industry holds of firm specific capital, with most insurance companies based upon levels of human capital and decision making. As mentioned earlier across all industries FDI from the finance and insurance industries were most suppressed by distance and therefore uncertainty. These results are consistent with the previous literature which disaggregate FDI flows, with Kiyota and Urata (2002) failing to observe significant coefficients across less capital industries. I believe finance and insurance firms to be most concerned with quality of human capital (tertiary education levels) and levels of information 5 Average inflation and unemployment rates taken from Trading Economics website, ( 15 P a g e

16 available, i.e. MNC s will be discouraged from entering foreign markets for similar arguments to those behind missing markets. To summarise my findings it appears that higher capital intensive industries appear to be most responsive to exchange rate volatility. however there are some industries which do not adhere to this general trend, take mining for instance where I discussed other factors which I believe to overwrite any exchange rate factors. The two variables which appeared most significant throughout were the variables taken fro the gravity model of FDI used in Blonigen and Piger (2011). My hypothesis that exchange rate volatility affects different industries to different degrees was observed in the data. Section 7 aims to discover, what causes this? 7.Sunk Cost Extension One extension which to my knowledge has not been directly considered by previous literature, is whether of sunk costs play a significant role in the responsiveness of MNE s to exchange rate volatility. Using the sunk cost proxy presented in figure 4, I will be assessing whether sunk costs create the heterogeneous reactions to exchange rate volatility across industries; and therefore create the variety of results obtained in Empirical Results. Sunk costs are generally funds, time or other resources which are spent which can never be retrieved regardless of leaving or continuing production in the market. High-tech US firm IBM require large quantities of expensive specialist capital, should they decide to open a production plant abroad. This huge investment into necessary capital represents larger level of irretrievable costs, relative to a labor intensive firm entering into the same country. A lack of sunk costs incurred by the labor intensive firm enables the MNC to be free to leave the foreign market. In contrast capital intensive firms would almost face a barrier to exit due to the high levels of sunk costs incurred, as they act as an opportunity cost of leaving the market. I predict that higher levels of sunk costs will raise firm s responsiveness to exchange rate volatility; as irreversible sunk investments are therefore riskier. Figure 4 shows estimates for each industries sunk costs by year (See Data Appendix, Sunk Costs for full definitions). As mentioned earlier, a sunk cost is a cost which can never be recollected by a firm, we would expect to see high levels of these costs from the mining, utilities and manufacturing industries. 16 P a g e

17 Figure 4: Annual Net Sunk Costs By Industry Industry Key MAN: Total Manufacturin g SER: Professional, scientific, and technical services FIN: Finance and insurance MIN: Mining WHO: Wholesale Trade UTI: Utilities INF: Information The effects of sunk costs will be tested through an interaction term between sunk costs and exchange rate volatility. I have chosen to use the mean of fixed assets and fixed structures for each industry, from 2003 to 2010, as a proxy for sunk costs across the period These averages will represent start up sunk costs incurred by an investing MNC. The full regression run therefore will be equation [5] below. LogFDI t = β 0 + β1(xrate t XRATE t-1 ) + β 2 XRATEVOL t + [5] β 3 LogDISTANCE t +β 4 Log(REAL)GDP t + β 5 Log(SunkCosts) + β 6 [Log(SunkCosts)*XRATEVOL] + β 7 (MAN) + β 8 (FIN) + β 9 (MIN) + β 10 (HOL) + β 11 (UTI) + β 12 (SER) + β 13 (WHO)+ ε Note that I will be including dummies for every industry in order to mop up fixed factors across industries unrelated to sunk costs (i.e. productivity, market demand, etc ), which otherwise would be picked up by β 6 and resulting in a biased estimation. The coefficient of the interaction term represents; when holding constant exchange rate volatility, how do higher levels of sunk costs affect the levels of FDI flows. The full effect of an increase in exchange rate volatility can be interpreted as an alteration of the slope of the relationship between exchange rate volatility and FDI dependant on sunk costs, similar to figure 3 from section 3. The overall gradient for each firms sunk cost and therefore reaction of the firm to volatility can be extracted through taking the partial derivative of regression 4 with respect to XRATEVOL. The results are presented below in table P a g e

18 Table 3: Do Sunk Costs Explain Variation Of Exchange Rate Volatility Reactions Across Industries? [Dependant Variable: lnfdi ] Independent Variables Coefficients _CONS ( ) XRATE (0.5259) XRATEVOL * (0.0022) lngdp *** (0.1234) lndistance *** (0.1924) lnsunkcost (0.387) lnsunkcost*xratevol (0.0003) Industry Fixed Effects Yes Time Fixed Effects Yes R N 1350 Notes Robust standard errors are presented in parenthesis (adjusted for 35 clusters ) *, ** and *** denote p<0.10, p<0.05 and p<0.01 respectively. The results of the data presented above conflicts with what I would expect to see, based upon my model and hypothesis. I would have expected sunk costs to exaggerate the risk averse characteristics of investors due to sunk costs definition, Sunk costs, on the other hand, are those costs that cannot be eliminated, even by total cessation of production. As such, once committed, sunk costs are no longer portion of the opportunity cost of production (Baumol and Willig 1981). Therefore I would have expected a negative value of β 6. However the data above suggests a higher level of sunk costs actually reduces MNC s sensitivity to changes in exchange rate volatility. As mentioned above taking the partial derivative of [5] with respect to exchange rate volatility allows for the full effects of exchange rate volatility to be revealed. [lnfdi] / [XRATEVOL] = [lnSUNKCOST] [6] Partial derivative [6] suggests MNC s from industries which incur higher levels of sunk costs, are less responsive to exchange rate volatility than industries which incur relatively less. Take for example two firms from two industries; one who incurs no sunk costs, and another who incurs a (log) sunk cost of 10. Both of which are investing into an economy of which the exchange rate volatility increases by 1 point. The model suggests the firm with no sunk costs will reduce investments by 0.04%, whereas the firm which incurs a sunk cost of 10 will only cut back FDI flows entering the country by 0. 01%. 18 P a g e

19 Although this seems like an unfeasible result to conclude, I believe these results are a consequence of the contrast between my model outlined and what would be witnessed in reality. As Froot and Stein (1991) suggested, it is likely that MNC s already own foreign production plants, usually many. If I take this into consideration, then the idea that sunk costs actually reduce a firm s responsiveness to exchange rate volatility becomes a reasonable concept. If a firm has already incurred sunk costs acquiring the necessary firm specific structures and capital when entering a country; then this creates almost a barrier to exit, due to the levels of sunk costs which have been incurred. Expanding on from this; the higher the levels of sunk cost incurred upon entering the market, the less multinationals will respond to changes in exchange rate volatility. When higher levels of sunk costs are incurred, then the opportunity cost of relocating and having to incur these sunk costs again will be greater. This is a cost which the firm knows would not be incurred; should the multinational continue to proceed operating at its current location. It is important to note the relationship was both small and insignificant between sunk costs and exchange rate volatility reactions. I believe this could be due to the inadequacy of the data collected as a proxy of sunk costs. Although it was the best available, it is likely that a firm is able to recover at least a fraction of the costs incurred in fixed assets and fixed structures. This therefore would not give a fully accurate proxy variable for sunk costs. 8. Are Pegged Currencies An Effective Tool To Attract FDI Flows? Kiyota and Urata (2002) and Benassy et al (2001) have previously empirically predicted the effects of a pegged currency. I will be adopting a different approach, this section will conduct a thought experiment through the effects of the potential policy. Governments in search of attracting FDI inflows could possibly adopt a fixed pegged currency, similar to polices employed during the Bretton Woods era. Fixing the currency is achieved through the buying and selling of the domestic currency, through the same channels as that of monetary policy. Fixing an exchange rate to the dollar would have two direct primary effects on FDI flows from the US. Firstly policy makers could set an exchange rate relative to the US dollar of their choice, providing potential cost advantages to MNC s. Secondly exchange rate volatility relative to US multinationals would fall to virtually zero, depending on how successfully monetary policy was implemented by the central bank. The effects which this policy would have upon FDI flows entering the economy would depend on two things; the current composition of FDI inflows and the current level of volatility (as this would distinguish the change in volatility as volatility against the dollar becomes zero). 19 P a g e

20 This paper has proposed that exchange rate volatility has different affects across industries to different extents, sometimes even in different directions. Depending upon the industry components of the FDI capital inflows to the economy, the effects of a pegged exchange rate would vary. Throughout empirical testing I have found that higher levels of exchange rate volatility suppresses the level of FDI inflows from the US the greatest in the manufacturing industry and the smallest measured effect within finance and insurance. Take two recipients of different compositions of FDI from the US; Germany and the United Kingdom (UK). For simplicity I shall only be considering two industries, manufacturing and finance and insurance. UK and Germany s current stock of US FDI in 2010, across the two industries are presented below in figure s 4 and 5. As expected the UK attracts an abundance of financial FDI, due to the UK s impressive infrastructure and human capital in this sector. Again as expected Germany, attracts a significant inflow of FDI from the manufacturing industry. Although highly unlikely, take an adopted US dollar pegged exchange rate for Germany and the UK. The effects of this policy would not be homogenous across the two countries. Take an adoption of a pegged (at the current rate 6 ) currency policy in 2010 for both countries. The UK s Pound Sterling and the Euro s volatility to the US Dollar were and 0.06 standard deviations respectively. Therefore the change in standard deviation into the next year will be and (as when pegged, volatility will be 0). From the manufacturing industry results from table 3; ceteris paribus, a fall in volatility of will attract an increase of 0.038% 7 of manufacturing FDI inflows to the UK. In contrast, within finance industry we would predict to attract a minute % 8. With the UK specialising in the finance sector I would not expect to see much of 6 Effects of changing exchange rate levels were not considered due to lack of significance observed in empirical results. 7 [(-0.008*-0.047)*100]% change in volatility subbed into XRATEVOL from regression MAN in table X 8 [( *-0.047)*100]% change in volatility subbed into XRATEVOL from regression FIN in table X 20 P a g e