Department of Economics Trinity College Hartford, CT USA.

Transcription

1 Department of Economics Trinity College Hartford, CT USA TRINITY COLLEGE DEPARTMENT OF ECONOMICS WORKING PAPER FDI, Exchange Rate, and Economic Growth in Hungary, : Causality and Cointegration Analysis Zsofia Komuves and Miguel D. Ramirez* June 2013 Revised October 2013 Abstract This paper investigates the causal relationship between FDI, GDP and the Euro/Hungarian Forint exchange rate in Hungary during the period. Although the question has great significance from an economic policy standpoint, there has been little to no empirical analysis undertaken so far in the case of transition economies such as Hungary. Utilizing unit root and cointegration analysis, this study finds a stable long-run relationship among the included variables, thus an error correction model is developed to capture the short-and long-run behavior of the variables. In the long run, changes in rea lgdp are positively associated with changes in the stock of FDI, while changes in the real effective exchange have a negative effect. In the short run, a 1 % deviation of FDI from its long-run relationship will be corrected by 0.48 % the following year. The VEC model leads to the general conclusion that all the variables in the system have a short-run adjustment mechanism. Finally, the Block Granger Causality tests generate mixed results with regard to the direction of causality among the variables, thus leadingto the conclusion that they are all endogenous. J.E.L. Codes: C22, F21, O52 Keywords: Granger causality test, error correction model (ECM), foreign direct investment (FDI), Johansen cointegration test, KPSS unit root test, vector error correction model (VECM), Zivot-Andrews single-break unit root test.

2 1. Introduction The role of foreign direct investment (FDI)in a developing country s economic growth and development is an important, controversial, and widely discussed topic in economics. Although the literature analyzing the determinants and the effects of FDI flows on domestic variables is rich, further studyof this question is needed because the strength of the relationship and the direction of causality tend to vary from country to country and over time. Local economic and political factors strongly influence the macro variables in question, which makes country-specific analysis particularly important. As with most developing and transition economies, sufficiently longtime-series have only recently become readily available; the FDI dynamics of transition economies such as Hungary or Romania, important as it is, is not yet a wellresearched topic in the literature. In particular, there is no extensive time-series analysis of the relationship between FDI and other macro-variables in the Hungarian case, which makes studying this question all the more vital. During the period examined by this study ( ), Hungary s economy changed from a posttransition state-led economy to a fast growing and outward-oriented economy. For example, the country s exports of goods and services as a proportion of its GDP rose from 45.2 percent in 1995 to 71.2 percent in 2001, and a high of 81.7 percent in 2008 before it was hit by the global crisis of (World Bank, 2012). More importantly, from the standpoint of economic development, manufactured exports as a percentage of overall exports rose dramatically from 62.8 percent in 1994 to 81.2 percent in In this connection, the role of foreign companies in promoting Hungarian exports has increased rapidly. According to the Hungarian Central Statistical office, percent of Hungarian exports and close to 73 percent of its imports were directly related to FDI during the period. Exports consist mostly of machinery and cars, other vehicles, and food products. The structure of imports is similar, but raw materials and energy sources also have a significant share (Balatoni and Pitz, 2012). Gross fixed capital formation--the country s source of future growth and employment creation also experienced a robust increase, particularly during the , when it grew at an average annual rate of 4 percent, reaching 1

3 23.5 percent of GDP in Finally, value added in industry and services grew, respectively, at an impressive average annual rate of 4.1 percent and 3 percent during the period and, as a percentage of GDP, industry value added rose from 28 percent in 1994 to almost 31 percent in 2008, while that of services increased from 62.5 percent to 66.1 percent (World Bank, 2012). In fact, Hungary led the way among the Eastern Central European countries in terms of socioeconomic and political reforms during the transition period and became highly integrated with the EU countries, so much so that over 70 percent of its trade is now undertaken with European countries. The country s rapid integration with the West culminated in its accession to the European Union in 2004, but not the adoption of the euro as it tries to meet the Maastricht Convergence criteria. However, during the period in question Hungary slowly lost its initial regional advantage. The transition created multi-layered and long-lasting social and economic problems. The economy was shaken after the post-accession crisis and the market opening of the new EU member states and, as discussed below, was hard hit by the global economic crisis of due to high levels of private and public borrowing (Hungary 2012, BTI). The opening and integration of the Hungarian economy has also led to a surge in inward FDI flows, with FDI inflows as a percentage of GDP rising from 2.7 percent in 1994 to 7.5 percent in 2001, and an incredible 48.6 percent in 2008.Table 1 below shows that, from the standpoint of contributing to the financing of gross fixed capital formation, FDI inflows as a percentage of gross fixed capital formation rose impressively from 24.1 percent in 2000 to 33.7 percent in 2005, and an incredible 75.5 percent in 2007! (computed from World Bank Development Indicators, 2012). As can be surmised by Table 1, FDI flows as a percentage of gross fixed capital formation averaged an impressive 30.3 percent for the period, excluding the very high figures for 2007 and 2008 which appear to be anomalous and may have been influenced by the crisis. Insofar as the stock of FDI is concerned, it more than quadrupled between 2000 and 2009, from $22.8bn to $99.1bn the latter figure representing 78.3 percent of the country s gross domestic product (UNCTAD, 2012). In 2009 the country was severely hit by the global crisis of and its GDP plummeted by 6.8 percent only to recover slowly since then, 2

4 growing by 1.3 percent in 2010 and 1.7 percent in The unemployment rate also jumped from 7.4 percent in 2007 to 10 percent in 2010 and 10.9 percent in 2011 and, more worrisome, the percentage of long-term unemployed relative to total unemployed rose from 47.5 percent in 2007 to 50.6 percent in Not surprisingly, FDI inflows to the country were negatively affected as they plummeted from an average of $6.3bn in to an average of only $2.9bn over the period (UNCTAD, 2012). Finally, gross fixed capital formation which had grown so impressively up to the eve of the crisis, dropped precipitously in 2009 by 29.3 percent and its ratio to GDP fell to 17.9 percent in 2010 and 16.7 percent in 2011 (World Bank, 2012). Table 1. Hungary: FDI flows as a percentage of Gross Fixed Capital Formation, * Country: Hungary *Source: Computed from World Bank (2012), World Bank Indicators. The most detailed time-series analysis of the economic role of FDI in Hungary is by Balatoni and Pitz (2012) who report that the FDI/GDP ratio in Hungary, already the highest among Eastern European countries in 1995, rose from 24.5 percent in 1995 to 71.0 percent in They also indicate that 68 percent of FDI inflows are invested by other European countries, mostly EU member states, and that 82 percent of these flows are channeled to the services sector. According to the Central Statistical Office of Hungary foreign subsidiaries of TNCs operated more efficiently and paid higher wages than domestic firms and accounted for between 60 and 70 percent of Hungarian exports in However, there are few, if anytime series studies which investigate the relationship between Hungarian FDI and other key macro variables. For example, Mileva (2008) using a pooled analysis for 22 transition economies (including Hungary) found that FDI has a positive effect on total factor productivity via spillover effects and enhanced domestic investments. In their study, Balatoni and Pitz (2012) undertook a cointegration analysis to examine the relationship between FDI inflows (as a proportion of GDP), GNI, and Total Factor Productivity (TFP). Their findings suggest that FDI inflows not only accounted for percent 3

5 of the average annual GDP growth over the period, but contributed to enhanced TFP growth more so than other sources of financing (loans and portfolio investments). To the best of our knowledge, there are no rigorous time series studies which examine the direction of causality and/or strength of the potential feedback between FDI inflows, the real exchange rate, and real GDP for Hungary. For example, a larger than expected rate of GDP growth could generate greater inflows of FDI, thereby appreciating the real exchange rate. The latter, in turn, could lead to even greater inflows if it enhances the foreign currency value of the remittances of profits and dividends back to the parent company (see De Mello, Jr., 1997).Although it would be interesting to investigate the relationship between FDI flows and other important macro variables, the lack of data over a sufficiently long period constrains the scope of the present paper. In view of this, we investigate the economic relationship between Hungarian real GDP, the Hungarian Forint/Euro real effective exchange rate, and Foreign Direct Investment during the period using quarterly data. The paper is organized as follows. First, it develops a conceptual framework of analysis that seeks to identify some of the major economic and institutional determinants of FDI. Second, the paper briefly discusses the nature and sources of the data used in this study. Third, unit root and multivariate Johansen cointegration analysis is undertaken, respectively, to investigate the time series properties of the variables and whether a unique long-run (equilibrium) relationship exists among them. It is determined that a unique contegrating relationship exists which makes it possible to generate an error correction model that reconciles the short-and long-run properties of the estimated model. That is, it enables the investigator to determine the structural relationship among the variables in the long run, as well as their short-run adjustment back to the equilibrium relationship. Fourth, a more general VECM framework is estimated to investigate the causal relationships and the adjustment process of the variables in the system as a whole; it is determined that a negative adjustment process exists for each of three equations and that, with the possible exception of real GDP, none of the variables are weakly exogenous. The concluding section summarizes the main results 4

6 and points to some shortcomings in the estimated model that can be addressed by future research and data availability. 2. Conceptual Framework Based on the theoretical framework developed by Dunning (1981;1988), TNCs undertake cross-border investments to increase their profitability through three main strategic channels. First, via the use of exclusive ownership rights, control over the production process, and related patents, the parent firm can obtain a competitive advantage vis-à-vis foreign and domestic firms, thereby effectively denying them access to the targeted market. Second, TNCs, via their cross-border investments, generate significant locational advantages in the form of direct access to cheap(unskilled and skilled) labor, key raw materials, reduced transportation and communication costs, the avoidance of tariffs, and access to growing markets for their goods and services. Third, by establishing a subsidiary abroad the parent firm can internalize and reduce contractual and administrative costs associated with leasing licenses and securing patents to undertake production(seekatona 2006).In this connection, Markusen (1995) argues that firms choose direct investment rather than licensing primarily because of the non-excludability property of new knowledge capital; viz., it is too costly for TNCs to prevent licensees from defecting and copying the new technology at little cost and setting up their own domestic firms in direct competition with the TNCs. From the standpoint of the host country, FDI is usually considered to be a source of external financing that is relatively more stable and involves a longer term commitment than portfolio flows such as bond and equity investments. For capital scarce economies such as Hungary s, FDI is essential for financing investments and promoting economic growth. FDI inflows also bring modern technology and managerial know-how which promotes economic growth and encourages the development of financial markets (Ramirez, 2006; and Kumar, 2007). This technological and managerial transfer, particularly if it involves the establishment of new enterprises in leadings sectors so-called Greenfield investments --generally 5

7 has a positive spillover effect on the country s economy as a whole. 1 The majority of the literaturehas found a strong connection between the economic growth of the host country and the inflow of FDI(see Chakraborty and Basu 2002, Liu, Burridge & Sinclair 2002, Huang, 2004, Ram and Zhang, 2002,Ramirez, 2006). The nature and scope of government policies are also a highly important factor in determining whether FDI flows to developing and transition economies such as Hungary. For example, FDI is likely to be attracted to countries where governments ensure an adequate provision of economic and social infrastructure in the form of paved roads, ports, airfields, relatively cheap energy supplies, and a welleducated and disciplined work force. In this connection, several investigators have found that the availability of skilled workers and adequate physical infrastructure are important determinants of FDI flows because it enables TNCs to strengthen both their ownership and locational advantages, thus allowing them to expand their market not only in the host country but the region as well (see Ramasamy and Young, 2004). In addition, FDI flows are likely to be encouraged by government policies that lead to the establishment of a legal-institutional framework that is conducive to business activity; viz., one that significantly reduces the transactions costs associated with negotiating contracts, improves information about the quality of goods and services, and make sure that the parties to a formal agreement honor their commitments (see Yeager, 1998). Another important factor in determining FDI flows is the real exchange rate,given that it is the most important link between economic policy and international competitiveness. However, there is no consensus in the literature as to whether a depreciation or appreciation of the exchange rate stimulates FDI flows. On the one hand, if the currency of the domestic country depreciates in real terms, it lowers the production costs (i.e. wages, costs of raw materials and transportation) for the parent company, thus 1 However, the short-and long-run effects of FDI can differ from each other and might even be negative for the country s economy if it leads to the elimination of domestic firms and/or substantial reverse flows of profits and dividends to the parent company that divert resources away from financing domestic capital formation (see Ram and Zhang, 2002). 6

8 attracting FDI. On the other hand, it could discourage FDI flows because it lowers the remittances of profits and dividends measured in the investors currency. Therefore, the effect of the depreciation of the host country s currency depends on which of the above mentioned factors is stronger. Finally, it is likely that there is bidirectional causality between FDI flows and other macro variables (including the real exchange rate). For example several studies have found that higher inflows of FDI have a positive effect on GDP as it leads to greater investment, exports, and production (see De Mello, 1997; Zhang, 2001). On the other hand, other studies suggest that the line of causation runs the other way, viz., GDP growth acts as a proxy for a larger market size which, in turn, attracts market-seeking FDI (see Mortimore, 2003). Clearly, modeling the macro determinants (and impact) of FDI in a developing or transition economy such as Hungary s is fraught with conceptual challenges and empirical difficulties that, in the best case scenario, can only lead investigators to formulate tentative conclusions Data The data used in this study were obtained from the databases of the National Bank of Hungary (NBH), the Hungarian Central Statistical Office and Haver Analytics. The data is quarterly for the period 1995 Q Q4 (72 observations for each variable).the Stock FDI variable is measured in millions of Hungarian Forints (HUF) at constant prices, using the 1995 Q1 price levels. 3 The same applies to the Real GDP variable. Both series are seasonally adjusted. The exchange rate data used in this paper is the real effective exchange rate index; one which takes into account the average relative purchasing power of 2 It would interesting to analyze the relationship between FDI and domestic investments to learn more about the effects of FDI inflows on the GDP. Domestic savings are often can be used as a proxy for domestic investment, but in Hungary there are no quarterly data available for it during the given period (see Arvai and Menczel (2000)). 3 The FDI series used in this paper includes capital in transit flows and the operation of special purpose entities (SPE). Capital in transit (as it is defined in the database of the National Bank of Hungary (NBH)) includes large capital inflows and outflows during a short period within companies or group of companies. The capital transactions flow through special purpose entities not affecting the external financing of the resident economy. Special purpose entities (SPE) are playing a role in the intermediation of substantial financial resources within groups of companies, due to channeling funds (the direction and size of which are controlled by their parents), rather than being a target for direct investment (net flows on various financial instruments are close to zero taking a longer period).balatoni and Pitz (2012) note that during the privatization and acquisition of formerly state-owned firms the capital in transit had a major part in inducing FDI inflows which might cause some bias in the results. 7

9 the domestic currency relative to the currencies of its main trading partners. An increase in the real effective exchange rate denotes a real average effective appreciation of the Hungarian Forint relative to the currencies of its trading partners. 4. Unit Root and Cointegration Analysis 4.1.Stationarity Before including time series in regression analysis, it is critical to test for unit roots or non-stationarity in order to avoid misspecified or spurious regressions (see Engle and Granger, 1987).Given the relatively low power of units root tests, this study used a variety of tests, including the well-known Augmented Dickey Fuller (ADF) and non-parametric Phillips-Perron test (PP) unit root tests, as well as the less well known (confirmatory) Kwiatkowski-Phillips-Schmidt-Shin (KPSS) stationary (no unit root) test to investigate the order of integration of the series. The Doldado-Sosvilla-Rivero (1990) procedure was used to determine the best specification of the test equation (in terms of whether to include an intercept and/or a deterministic trend while testing). The appropriate lags were determined via the Eviews 7.2 default procedure, using the more robust SBC as a criterion. Finally, Zivot-Andrews single-break unit root tests were also undertaken because the power of unit root tests is reduced significantly when the stationary alternative is true and a possible structural break is ignored. By and large, the results reported in Table 2 below yield the same general conclusion, viz., the variables in level form are non-stationary while in first difference they are stationary or I(0). 8

10 Table 2.Stationarity Tests* Logarithmic Level Data First Log-Difference ADF PP KPSS ZA ADF PP KPSS ZA REER *** *** *** *** RGDP *** ** ** ** 0.126* ** SFDI *** *** *** *** *Table 2 summarizes the results of the stationarity tests. The table contains the calculated t-statistics. * denotes a significance of 10%, ** stands for 5% and *** is for 1%. The ADF and PP test critical values are 1%: -4.09, 5%: -3.48, 10%: The KPSS critical values are 1%: 0.216, 5%: 0.146, 10%: 0.119The Zivot Andrews critical values are 1%: -5.57, 5%: -5.08, 10 %: More detailed results and the plots of the series can be found in Appendix 1-2. More specifically, the ADF and PP results are entirely consistent and indicate that for the level data we cannot reject the null hypothesis, while for first differences it can be rejected mostly at the 1% level. The confirmatory KPSS LM test reverses the null hypothesis of a unit root and tests for stationarity in the variables. The KPSS test rejects stationarity for the level data and, with the exception of GDP at the 10 percent level, concludes that the variables are stationarity in their first differences. The results for the Zivot-Andrews test also suggests the same general conclusions, again, with the exception of the GDP variable in level form where it rejects the null of a unit root in the presence of a single structural break. This might be an indication of more structural breaks in the series. It is even visible in the graph of GDP (included in Appendix 1). Before 1996 Q3 and after 2006 Q4 the series has a completely different trend than in between. Overall most of the results suggest that all the series used in this study are integrated of the first order, I(1) series. 4.2Cointegration Analysis Having shown that the variables are integrated of order one, I(1), it is necessary to determine whether there is at least one linear combination of these variables that is I(0). In other words, is there a stable and non-spurious (cointegrated) relationship among the regressors in each of the relevant specifications? This was done by using the Johansen and Juselius(1990) cointegration method because it is capable of determining the number of cointegrating vectors for any given number of non-stationary series (of the same order) and, unlike the ADF tests, the likelihood ratio tests used in this procedure have well-defined 9

11 limiting distributions. The Johansen test includes the log of the stock of FDI, real GDP, and the effective exchange rate, and it is estimated in three different (relevant) model specifications. The number of lags in the test equation is 4, which is the usual choice when using quarterly data (other lag specifications were tried but the results were not altered).the Pantula Principle was used to select the appropriate specification of the model regarding the deterministic components, trend, and intercept of the equation. The procedure compares the trace and the Max-Eigen statistics of Models 2, 3 and 4 (in this order) to their critical values starting from the most restrictive model (no cointegration) to the least restrictive one (2 cointegrating vectors), and stops when the null hypothesis of no cointegration cannot be rejected for the first time. Table 3. Johansen Test Results* Trace statistics Max - Eigen Statistics R Model 2 Model 3 Model 4 Model 2 Model 3 Model * (35.193) (29.798) ( ) (22.230) ( ) ( ) (20.262) (15.495) (25.872) (15.892) (14.265) (19.387) (9.165) ( 3.841) ( ) (9.165) (3.841) ( ) *Table 3 reports the Trace and Max-Eigen value statistics of the Johansen cointegration method for the three relevant models.the test is conducted on the logarithmic series. R stands for the number of cointegrating vectors and the5 % critical values of the tests are in parenthesis. Model 2 includes a constant but no trend in the cointegration equation (CE) and does not allow a trend in the VAR part of the test equation. Model 3 allows an intercept, but no trend in both the CE and the VAR parts of the test equation. Model 4 includes an intercept in both the CE and VAR, a trend in the CE, but no trend in the VAR. It should be noted that the Johansen test does not include additional exogenous variables. Although this would make the model more complete, the conclusions regarding the existence of cointegrating relationships do not change. (Appendix3). Based on the Pantula selection procedure, Model 2 is the appropriate specification. Model 2 includes a constant but no trend in the cointegration equation (CE), and does not allow for a trend in the VAR part of the test equation. This means that there is no linear trend in the data and the first differenced series has a zero mean (which is consistent with the unit root tests for the level and differenced series suggested by the Doldado et al. procedure). Model 2 indicates that one unique cointegration vector exists; that is, there 10

12 exists a unique linear combination of the I(1) variables that links them in a stable and long-run relationship. The cointegrating equation (normalized on the log of the stock of FDI) is reported below and it shows that the long- run estimate for GDP is positive, while the long-run estimate for the real effective exchange rate is negative (signs are reversed because of the normalization process); the t-ratio for real GDP is significant at the 1 percent level for a two-tailed test (p-value:.0044), while that for the real exchange rate is only significant at the 5 percent level for a one-tailed test (p-value:.041). In other words, a ceteris paribus increase in real GDP (a proxy for market size) induces FDI flows to Hungary, while a ceteris paribus increase in the real effective exchange rate (appreciation) reduces inflows to the country. Thus, in the long run, it appears that investors tend to focus less on the rate of return measured in their own currency. The presence of one cointegrating equation from which residuals (EC terms) can be obtained also makes it possible to investigate whether there is a short-run adjustment back to the long-term relationship after a shock, using the Engle-Granger two-step procedure. Table 4. Normalized Cointegration Coefficients LSFDI LRGDP LREER C cointegration coefficients t-ratios (-2.85) (-1.74) (-2.74) 4.2. Short-Run Error Correction Model Given that there is a stable long-run relationship among the relevant variables, it is possible to estimate an error correction (EC) model that captures both the short-and long-run behavior of the FDI relationship (Engle and Granger, 1987). The changes in the relevant variables represent short-run elasticities, while the coefficient on the EC term represents the speed of adjustment back to the long-run relationship among 11

13 the variables. The estimated an EC model is reported below in terms of logarithmic differences of the variables. DLSFDI t = β 0 + β 1 DLRGDP t + β 2 DLREER t + β 3 ECT t 1 + u DLSFDI t = DLRGDP t DLREER t ECT t 1 + D IMF + u [3.000***] [1.346] [2.504**] [-2.000**] R-Squared: Adjusted R-Squared: Prob(F): SBC: AIC: DW: (Appendix 4.2) The short-run estimate for real GDP suggests that a 1 percent increase in this variable generates a 1.2 percent increase in FDI flows, but, as opposed to the long-run relationship reported above, it is not statistically significant at the 5 percent level. The short-run impact of the real effective exchange rate, on the other hand, is statistically significant at the 5 percent level but, in contrast to its long-run effect, it suggests that a 1 percent real appreciation generates a 0.62 percent increase in FDI flows. These results suggest that, in the short run, foreign investors focus more on the real rate of return measured in the investors currency, while in the long run they are, perhaps, more concerned with the real domestic costs of labor and raw materials (which a ceteris paribus real appreciation of the domestic currency would raise in terms of the investors currency). As the theory predicts, the EC term is negative and statistically significant, suggesting that a 10 percent deviation from the long-run FDI relationship during the current quarter is corrected by about 1 percent in the next quarter, and by 4.1 percent on an annual basis. Although the adjusted R-squared is not very high, the p-value for the F-stat. suggests that the EC model as a whole is statistically significant at about the 1 percent level. The Durbin Watson statistics is close to 2 which suggests that there is no first-order autocorrelation in the model. To ensure that the model does not suffer from higher order serial correlation, a Breusch-Godfrey test was performed. Again, the results indicate that there is no serial correlation in the model. The non-significant p-values of the Ramsey Reset Test with 1 to 4 powers also suggest that the model is not misspecified. 12

14 Finally, the EC model was estimated with the inclusion of several dummy variables to determine whether the inclusion of important qualitative factors or events would improve the relative fit of the EC model, but they did not increase the adjusted R-squared and, in fact, actually decreased the SBC and AIC values significantly, thus rendering the error correction term insignificant 4. The best model was the one which controlled for the two IMF loans that Hungary received during the period (1996 Q1, 2008 Q3) to handle a local currency crisis and the severe consequences of the global economic crisis of Although the dummies for this model were not significant, they slightly increased the significance of the other explanatory variables and the adjusted R-squared. On the other hand, their inclusion decreased the AIC and SBC statistics. (see Appendix 6) Vector Error Correction Model The VECM framework allows investigators to model the short-run correction mechanism of a system of variables to their long-run equilibrium without deciding, apriori, about the endogeneity or exogeneity of the included variables. The vector autoregressive framework treats all variables as endogenous and determines the direction of causality between them based on econometric tests instead of assuming exogeneity based on economic theory. The VAR/VECM modeling has faced severe criticism because of its atheoretic, empirically-based methodology, though it often generates better forecasts than the complex economic theory based-models. In our case there are advantages from not predetermining the direction of causality. First, as indicated in Section 2, there is no consensus in the literature about the direction of causality between the exchange rate, GDP, and FDI flows. Second, Hungary is not only a developing 4 Different specifications were estimated utilizing dummy variables which could have explanatory power in the model. However, most dummies proved to be insignificant except for the first two:d_cr(financial crisis of ), d_imf (two IMF packages Hungary received during the period (1996 Q1, 2008 Q3)), d_econp (for the two restrictive and very powerful economic reforms (1995 Q1: Bokros-Package, 2006 Q: Gyurcsany-Package) which addressed local economic problems, d_eu (steps involved in joining the EU (1997 Q3: member states decide that Hungary can join, 2003 Q2: successful referendum about joining the EU in Hungary, 2004 Q2: joining the EU), d_noimf(controls for the unsuccessful negotiations with IMF starting from the end of 2011 which eroded foreign investors trust in the credible and secure economic environment of Hungary), and D_junk(controls for a similar event). In 2011 Q3 two major credit rationing agency downgraded the Hungarian government s debt to the junk category which diminished the trust and the level of foreign investments to Hungary. 13

15 country, but a relatively small economy that has experienced significant structural and institutional change and opened itself to the West in a very short period of time as attested by its accession to the EU in 2004 and the negative aftermath it experienced in the wake of the great recession of This could cause any predictions from economic theory about the behavior of some key macro variables to deviate significantly from their actual trajectory. Thus, the flexibility of this specification represents a distinct advantage in modeling the system-wide error correction mechanisms of the relevant variables. VECM specifications can only be estimated if there is a cointegrating relationship among the variables which, according to the Johansen test, is indeed the case for our study. First, an unrestricted VECM is estimated using LSFDI, LRGDP and LREER variables to examine whether a system-wide error correction mechanism exists for these variables. The VECM is estimated based on Model 2 (using the Pantula Principle) with 4 lags. The abovementioned and significant d_imf dummy (which controls for the two loans provided by the IMF to Hungary) is included in the equation as an exogenous parameter. The unrestricted VECM finds negative and significant adjustment coefficients for the three equations. This means that our variables work as an endogenous system in which there is a short-run adjustment mechanism to the equilibrium relationship after shocks. The best specification, based on the R-squared values and the AIC/SBC criteria, is the equation with D(LRGDP) as the dependent variable. In this specification (with the dummy variable), the long-run cointegration equation does not have a significant coefficient on the exchange rate variable, but real GDP remains positive and statistically in terms of its effect on the stock of FDI 5.(Appendix 8A). 5 VECM/VAR modeling is especially useful for building strong forecasting models on the basis of R-squared and SBC/AIC criteria. In this study we focus on the relationship among the three variables and a forecast model is not estimated. 14

16 Table 5. Error Correction D(LSFDI) D(LRGDP) D(LREER) Error Correction Coefficient Standard Error ( ) ( ) ( ) t-statistics [ ] [ ] [ ] R-squared Adj. R-squared Akaike AIC Schwarz SBC Second, a very useful property of the VECM framework is that it enables the investigator to impose zero restrictions on the adjustment coefficients of each equation, thus determining which variable can be treated as weakly exogenous in the system, thereby omitting it from the interdependent system of variables. Based on this weak exogeneity test, none of the variables can be omitted from the system (treated as weakly exogenous) because the null hypothesis of a zero restriction is rejected for each variable at least at the 5 percent level. In other words, in this simple three equation system everything causes everything else and all variables must be assumed to be endogenous.(appendix 8B). Table 6.Exogeneity Test H0: weakly exogenous variable Chi-Square statistics (65 obs) Probability LSDFI, A(1,1)= LRGDP, A(2,1)= LREER, A(3,1)=

17 Third, in order to investigate further the causal relationship among these variables, we performed a Granger Block Causality test. This test examines all three equations and tries to determine whether the presumed exogenous variables can be omitted from each equation. This test only finds causality or precedence in the DLREER(DLSFDI, DLRGDP) equation at the 5% level, from FDI to the real effective exchange rate (p-value DLSFDI :0.0069, p-value DLRGDP : ). (Appendix 9).Pairwise versions of this test were also undertaken which allows for different specifications using different number of lags for the variables. This test also suggests that FDI Granger Causes the REER at the 5 % level (using 1,2,3 or 4 lags).(appendix 10). Overall, based on the analysis of the unrestricted VECM specifications, we can conclude that short-run deviations of the included variables from their long-run cointegrated relationship are corrected in subsequent quarters between 1.2 and 11.3 percent. The restricted VEC models suggest that there are no weakly exogenous variables in the system, that is, everything causes everything else. On the other hand, the Granger causality tests find a causal relationship from FDI flows to the real effective exchange rate. In view of these mixed results, it seems best to proceed cautiously and assume that the interdependence of the variables is relatively more important than the pairwise causality between them. 5. Conclusion To the best of our knowledge, this is the first time series analysis of the relationship between FDI flows to Hungary, real GDP, and the Hungarian Forint / Euro real effective exchange rate during the period. The unit root and stationarity tests indicated that the economic variables are, by and large, integrated of order one. The multivariate Johansen cointegration test showed that there is a stable long-run relationship among the included variables, which means that non-spurious EC models can be estimated on the basis of the series. Using the Engle-Granger approach, an error correction model was generated that reconciles both the short-and long-run properties of the variables. In the long run a higher real exchange rate (real appreciation of the domestic currency) decreases FDI inflows, while the level of real GDP has a 16

18 positive and statistically significant effect on the stock of FDI. In contrast to the long-run results, the real exchange rate has a positive effect on FDI flows in the short run, while real GDP retains its positive and statistically significant effect. The EC term is negative and significant, suggesting that a 10 percent deviation during the current quarter from the equilibrium relationship is corrected in the next quarter by 4.8 percent or 19.2 percent on an annual basis. The estimates from the unrestricted VECM model indicate thatthe variables as a system also have a statistically significant adjustment mechanism. Restricted versions of the VECM model and Granger Block Causality tests generate mixed results on the direction of causality among the included variables, and rather suggest that is best to take a conservative stance and regard all the variables as endogenous in nature. Despite the important findings reported in this study, further research and data is needed to come to more general and robust conclusions regarding the determinants of FDI flows to Hungary. First, the lack of data did not allow for the inclusion of other macro variables that are important in terms of explaining the variation in FDI flows to Hungary, such as public investment in economic infrastructure, unit labor costs, or proxies for human capital such as secondary or tertiary enrollment ratios. Second, the long-run cointegration equation part of the VECM specification (with the dummy variable) has an insignificant exchange rate variable. This contradicts the previous Johansen cointegration test without the dummy variable where both the real exchange rate and real GDP are statistically significant. Third, the VECM results regarding the direction of causality are rather mixed, which again might be a sign of some problem with the specification of the model. Finally, none of the dummy variables improved the estimated models significantly which is puzzling given that the crisis of had a highly negative effect on real GDP, gross fixed capital formation, and FDI flows. 17

19 References Balatoni, A. &Pitz, M. (2012). Aműködőtőkehatása a bruttónemzetijövedelemremagyarországon (The Effect of Foreign Direct Investment on the Gross National Income of Hungary).KözgazdaságiSzemle (Hungarian Economic Review), 59(1),pp Chakraborty, C. &Basu, P (2002). Foreign direct investment and growth in India: a cointegration approach. Applied Economics, 34(9), pp Csáki, G. (2013). IMF-hitelekMagyarországnak, (IMF loans to Hungary, ).PénzügyiSzemle (Hungarian Financial Review), (1), p De Mello, L.R. Jr. (1997). Foreign direct investment in developing countries and growth: A selective survey. Journal of Development Studies, 34, pp., Dunning, J.H. (1981). International Production and the Multinational Enterprise. Boston: Allen and Unwin. Dunning, J.H. (1988). Explaining International Production. London: Harper and Collins. Economic Survey of Hungary.(2010). OECDCountry Report. Retrieved from: Hungary. Country Report (2012).BTI. Retrieved from: Hungary.pdf Engle, R.F., and Granger, C.W.J. (1987).Cointegration and Error Correction: Representation, Estimation, and Testing. Econometrica, 55, pp

20 Huang, T. Jr. (2004).Spillovers from FDI in China. Contemporary Economic Policy, 22, pp Johansen, S. and K. Juselius.(1990). Maximum likelihood estimation and inference on cointegration with applications to the demand for money, Oxford Bulletin of Economics and Statistics, 52, pp Katona K. (2006). A magyarorszagitokeimportotbefolyasolotenyezokujraertelmezese.(reinterpretatingof the determinants of FDI inflows to Hungary).KozgazdasagiSzemle,(Hungarian Economic Review), 53(11), p Kumar, A. (2007). Does foreign direct investment help emerging economies? Economic Letter, 2, pp.1-8. Liu, X., Burridge, P., & Sinclair, P. J. (2002). Relationships between economic growth, foreign direct investment and trade: evidence from China. Applied Economics, 34(11), Markusen, J.R. (1995). Incorporating the multinational enterprise into the theory of international trade, Journal of Economic Perspectives, 9, pp Mileva, E. (2008). The Impact of Capital Flows on Domestic Investment in Transition Economies.ECB Working Paper Series, No Mortimore, M. (2003).The impact of TNC strategies on development in Latin America and the Caribbean.Published for OverseasDevelopment Institute (UK), pp Ram, R. and Zhang, K.H. (2002). Foreign Direct Investment and Economic Growth: Evidence from Cross Country Data for the 1990s. Economic Development and Cultural Change, 51, pp Ramirez, M. D. (2010). Economic and Institutional Determinants of FDI Flows to Latin America: A Panel Study, Working Papers 1003, Trinity College, Department of Economics. Ramirez, M.D. (2006). Is Foreign Direct Investment Beneficial for Mexico? An Empirical Analysis. World Development, 34, pp

21 UNCTAD (2012). World Investment Report 2012: Towards a new generation of investment policies. New York. World Bank (2012). World Development Indicators : Yeager, T.J. (1998). Institutions, Transition Economies, and Economic Development. Westview Press. Zhang, K.H. (2001). Does foreign direct investment promote economic growth? Evidence from East Asia and Latin America. Contemporary Economic Policy, 19, pp

22 6. Appendix 1. Graphs 11.5 LSFDI.4 DLSTOCKFDI LREER DLRER LRGDP DLRGDP

23 2. Unit Root Tests 2. A. Stock Foreign Direct Investment- Level ADF Null Hypothesis: LSFDI has a unit root Exogenous: Constant Lag Length: 0 (Automatic - based on SIC, maxlag=11) t-statistic Prob.* Augmented Dickey-Fuller test statistic Test critical values: 1% level % level % level *MacKinnon (1996) one-sided p-values. Augmented Dickey-Fuller Test Equation Dependent Variable: D(LSFDI) Date: 05/01/13 Time: 04:26 Sample (adjusted): 1995Q2 2012Q4 Included observations: 71 after adjustments LSFDI(-1) C R-squared Mean dependent var Adjusted R-squared S.D. dependent var S.E. of regression Akaike info criterion Sum squared resid Schwarz criterion Log likelihood Hannan-Quinn criter F-statistic Durbin-Watson stat Prob(F-statistic) PP Null Hypothesis: LSFDI has a unit root Exogenous: Constant Bandwidth: 57 (Newey-West automatic) using Bartlett kernel Adj. t-stat Prob.* Phillips-Perron test statistic Test critical values: 1% level % level % level *MacKinnon (1996) one-sided p-values. 22

24 Residual variance (no correction) HAC corrected variance (Bartlett kernel) Phillips-Perron Test Equation Dependent Variable: D(LSFDI) Date: 05/01/13 Time: 04:26 Sample (adjusted): 1995Q2 2012Q4 Included observations: 71 after adjustments LSFDI(-1) C R-squared Mean dependent var Adjusted R-squared S.D. dependent var S.E. of regression Akaike info criterion Sum squared resid Schwarz criterion Log likelihood Hannan-Quinn criter F-statistic Durbin-Watson stat Prob(F-statistic) KPSS Null Hypothesis: LSFDI is stationary Exogenous: Constant, Linear Trend Bandwidth: 6 (Newey-West automatic) using Bartlett kernel LM-Stat. Kwiatkowski-Phillips-Schmidt-Shin test statistic Asymptotic critical values*: 1% level % level % level *Kwiatkowski-Phillips-Schmidt-Shin (1992, Table 1) Residual variance (no correction) HAC corrected variance (Bartlett kernel) KPSS Test Equation Dependent Variable: LSFDI Date: 05/01/13 Time: 04:26 Sample: 1995Q1 2012Q4 Included observations: 72 23

25 C R-squared Mean dependent var Adjusted R-squared S.D. dependent var S.E. of regression Akaike info criterion Sum squared resid Schwarz criterion Log likelihood Hannan-Quinn criter F-statistic Durbin-Watson stat Prob(F-statistic) ZA Zivot-Andrews Unit Root Test Date: 05/01/13 Time: 04:05 Sample: 1995Q1 2012Q4 Included observations: 72 Null Hypothesis: LSFDI has a unit root with a structural break in both the intercept and trend Chosen lag length: 3 (maximum lags: 4) Chosen break point: 2005Q2 t-statistic Prob. * Zivot-Andrews test statistic % critical value: % critical value: % critical value: * Probability values are calculated from a standard t-distribution and do not take into account the breakpoint selection process Stock Foreign Direct Investment- Log-difference ADF Null Hypothesis: DLSTOCKFDI has a unit root Exogenous: Constant, Linear Trend Lag Length: 0 (Automatic - based on SIC, maxlag=11) t-statistic Prob.* Augmented Dickey-Fuller test statistic Test critical values: 1% level % level % level *MacKinnon (1996) one-sided p-values. Augmented Dickey-Fuller Test Equation Dependent Variable: D(DLSTOCKFDI) Date: 04/26/13 Time: 13:45 24

26 Sample (adjusted): 1995Q3 2012Q4 Included observations: 70 after adjustments DLSTOCKFDI(-1) C R-squared Mean dependent var Adjusted R-squared S.D. dependent var S.E. of regression Akaike info criterion Sum squared resid Schwarz criterion Log likelihood Hannan-Quinn criter F-statistic Durbin-Watson stat Prob(F-statistic) PP Null Hypothesis: DLSTOCKFDI has a unit root Exogenous: Constant, Linear Trend Bandwidth: 13 (Newey-West automatic) using Bartlett kernel Adj. t-stat Prob.* Phillips-Perron test statistic Test critical values: 1% level % level % level *MacKinnon (1996) one-sided p-values. Residual variance (no correction) HAC corrected variance (Bartlett kernel) Phillips-Perron Test Equation Dependent Variable: D(DLSTOCKFDI) Date: 04/26/13 Time: 13:47 Sample (adjusted): 1995Q3 2012Q4 Included observations: 70 after adjustments DLSTOCKFDI(-1) C R-squared Mean dependent var Adjusted R-squared S.D. dependent var S.E. of regression Akaike info criterion Sum squared resid Schwarz criterion