First supervisor: Naomi Leefmans
Second reader: Dr. K. Mavromatis
Name: Madalena Pinheiro
Student number: 11376406
ECONOMIC
GROWTH, FDI
AND TRADE
Statement of Originality
This document is written by Student Madalena Pinheiro who declares to take full responsibility for the contents of this document.
I declare that the text and the work presented in this document is original and that no sources other than those mentioned in the text and its references have been used in creating it. The Faculty of Economics and Business is responsible solely for the supervision of completion of the work, not for the contents.
Table of Contents
I. Introduction ... 4
II. Liberalization of trade and FDI in India ... 7
III. Literature Review ... 12
IV. Methodology ... 16
a) Unit Root test ... 17
b) Johansen’s Co-integration Test ... 19
c) Granger causality test ... 22
V. Data ... 24
VI. Empirical Results ... 27
a) Unit Root Test ... 27
b) Johansen Co-integration test ... 28
c) Granger causality test ... 30
VII. Limitations ... 34
VIII. Conclusions ... 35
IX. References ... 36
Annex ... 39
I. Introduction
India is one of the largest economies in the world, ranked 7th in 2015 (World Bank GDP ranking). Along with its economic size, international trade and Foreign Direct Investment (FDI) inflows in India have seen a massive increase since the 1990s. Before 1991, India was quite closed to foreign investment and trade, with the exception of some industries. This economic self-sufficiency ideal came with India’s aversion to its colonial past and dependence on Britain. However, following a big economic crisis in 1990, caused by balance of payments problems, a liberalisation policy was set in place in 1991, where the goal was to turn India into a market-oriented economy and expand the role of private and foreign investment. With this in mind, controls over foreign trade and investment were considerably relaxed, including the removal of ceilings on equity ownership by foreign firms (V.N. Balasubramanyam and Vidya Mahambare, 2003).
These reforms and efforts to open up the economy of India to the rest of the world resulted in increasing inflows of FDI into India during that decade and this trend continued until today. India’s FDI stock went from US$1,7 billion in 1991 to US$282,3 billion in 2015 (UNCTADstat), which, according to the World Bank Group, represented only 0,027% of GDP in 1991 and in 2015, the share was already 2,1%. In 2015, India replaced China as the largest recipient of FDI in the Asia-Pacific region and, of the global top 10 destination countries for FDI, India featured 5th (The FDI Report, 2016).
At the same time, as India opened its doors to the world, international trade gained a significantly larger role in the economy. According to the World Bank Group statistics, exports accounted for 19,9% of GDP in 2015, while in 1991 this share was only 8,3%. Likewise, the imports share of GDP grew from 8,3% to 22,5%.
The impact of FDI on the growth of host economies, both in the short and long run, has been widely studied in the recent literature. Theoretically, FDI has a positive impact on economic growth, either through increased investment and potentially its efficiency, or through technological diffusion from the developed country to the host one (Borensztein, Gregorio, & Lee, 1998). However, it appears that among the existing empirical literature, there is still no consensus on the sign and size of this effect.
Additionally, recent research has been done with the goal of understanding the causal connections between FDI and economic growth and the possible existence of a two-way
causality. This arises because, as discussed, FDI promotes economic growth in the host country; and simultaneously, higher growth and prospects of future growth increase FDI flows to the host economy. However, the majority of the existing research is focused on investigating the causality running from FDI to economic growth.
The relationship between FDI and trade volumes has been hardly studied, although some papers suggest a bi-directional link between them. One possible channel for this feedback is if we assume that multinational enterprises (MNE) will fragment their production process optimally and locate labour intensive stages of production in labour abundant countries. Developing countries with high productivity/wage ratios will experience high vertical FDI1 inflows from these MNEs, seeking to benefit from the cost advantage associated. As a result, trade relations between the two countries will deepen, due to a two-way trade: on one hand, increased imports of primary and intermediate goods by the investing firm from the developing economy, and on the other, higher exports of final goods from the investor to the developing country. The demand for skilled labour in the developing country will increase as this is the type of workers employed by the MNE, raising the return to human capital. The supply of skilled labour will consequently increase, which, in turn, raises the attractiveness of the developing country to future FDI inflows (Joshua Aizenman and Ilan Noy, 2005). India could be a real example of this channel, as increased trade in services due to lower communications costs led to higher demand for skilled labour, which in turn increased the return for education. Consequently, the supply of human capital increased, making India more attractive for FDI.
The aim of this thesis is to study the long-run and short-run dynamics between FDI, GDP growth and trade, measured by import and export volume for India during the period after liberalisation, starting in 1991 until 2015. It is clear that the linkages between these variables are beyond simple, and addressing them is of great importance, especially for a growing economy like India. The methodology used in this work is based on existing literature on co-integration analysis, specifically the paper by Xiaohui Liu, Peter Burridge and P. J. N. Sinclair (2010) which investigates the relation between Chinese economic growth, FDI inflows, exports, and imports. First, they test the integration properties of the data, using the Augmented Dickey-Fuller test for a unit root. Second, using the Johansen
1
Vertical FDI takes place when the MNE fragments the production process internationally, locating each stage of production in the country where it can be done at the least cost.
(1988) procedure, they test the presence of co-integration among the variables and then detect the number of co-integrating vectors. Co-integration can be regarded as a long-run equilibrium relationship among the variables. Finally, the authors study the short run causality between all variables in order to distinguish between unidirectional and bi-directional causality among them.
My contribution to the literature consists of testing the presence of co-integration between Indian GDP, FDI and export and import volume, as a measure of international trade, using data from the period from 1991 until 2015, which the literature, until now, has failed to do. These four variables have never been studied in a co-integration framework using data from India, more specifically, most of the studies have failed to include exports and imports in the study of this long-term relation. Understanding these causal linkages can help policy makers to improve the country environment in order to attract more foreign investment, and also ensure that the effect that each variable has on the other is optimally maximised, in order to boost their impact in the economy.
The remainder of this thesis is structured as follows. Section II describes the historical background regarding the liberalization process of the Indian economy, which constitutes the motivation of this thesis. Section III presents an overview of the existing literature on the different linkages between FDI, growth and international trade, with the intention of providing the context in which this thesis is included and review the results already reached by existing works. Section IV explains the methodology behind the study in a theoretical framework. Section V describes the selected variables and their source. Section VI shows the empirical results. Finally, Section VII enumerates some limitations of this thesis and suggestions for future studies in this area, and Section VIII presents the main conclusions drawn from the previous analysis.
II. Liberalization of trade and FDI in India
India was part of the British colonial imperium until its independence in 1947. The British presence in India contributed close to nothing to the country’s development since all investments were made with the aim of promoting colonial interests, instead of India’s economic development and prosperity. After independence, the Indian economy was influenced by the colonial experience and the government’s ideals were shaped by an aversion to the country’s colonial past and dependence on Britain.
Trade policy was developed around the principle of import substitution, where the government would deny the allocation of foreign exchange for importing a product if domestically produced substitutes were available in sufficient quantity. During this period, some failed attempts of liberalization arose. However, as these coincided with a major crop failure and a consequent recession, the result was tighter import controls and further isolationism of the economy, supported by the affected population. By the mid-1970s India’s trade regime had become so repressive that the share of nonoil and noncereal imports of GDP fell from an already low 7% in 1957-58 to 3% in 1975-76 (Arvind Panagariya, 2004).
Foreign capital was banned from specified industries and all kinds of collaboration agreements between Indian owned and foreign firms were preferred to FDI. The introduction of the Foreign Exchange Regulation Act (FERA) in 1973 restricted foreign ownership of shares in firms operating in India. On top of that, foreign firms’ operations were severely limited by “local content” and “foreign exchange balancing” rules imposed by Indian authorities (K.S. Chalapati and Rao Biswajit Dhar, 2011).
In the late 1970s, two main factors were crucial as starting points for India’s way towards its liberalization. First, industrialists started to realize how the controls over imports were having adverse effects on their business profits, as raw materials and machinery were not produced domestically, which forced them to import these at high tariff rates. Second, as a result of improved export performance and increased remittances from overseas workers in the Middle East, the authorities stock of foreign exchange reserves increased, relaxing the effect of liberalization on the balance of payments, feared before by the authorities.
The new phase of liberalization started in 1976 with the introduction of the Open General Licensing list, which consisted of a list of items that did not need a license from the Ministry of Commerce to be imported. The government also introduced numerous export subsidies,
neutralizing the antitrade bias effect that resulted from import controls. Additionally, in 1985-90 the rupee was devalued in nominal effective terms by 45%, that resulted in a real depreciation of 30%, improving the economy’s terms of trade and making Indian exports more attractive to the rest of the World.
The liberalization process and the implemented expansionary fiscal policy together raised India’s growth rate from an average of 3.5% during the period between 1950 and 1980 to 5.6% in the period between 1981 and 1991. However, the external and internal borrowing that financed the fiscal expansion became unsustainable and later culminated in a balance of payments crisis in June 1991. This time, instead of reversing the course of liberalization, the Indian government took the opportunity to launch a reform program to be gradually implemented.
Tariffs were reduced in nonagricultural and industrial goods. The process consisted of gradual compression of the top tariff rates, that for several products would exceed 100%, and fell to 85% in 1993-94 and further to 50% in 1995-96. Before the beginning of liberalization, India also restricted exports of several commodities, which as part of the liberalization policy were to be relaxed and eventually eliminated. The March 1992 Export-Import Policy reduced the number of items subject to import and export control from 439 to 296, and only 16 were absolutely prohibited, compared to the previous 185. This process continued until recently so that nowadays export prohibitions apply only to a reduced group of items based on health, environmental or moral grounds.
Foreign investment before 1991 was highly regulated and suppressed by ceilings on equity ownership by foreign firms in the majority of the sectors in the economy. The 1991 economic reforms came to change this by severely relaxing the ceilings imposed before by FERA, so as to restructure the country’s foreign trade policy. Foreign brand names were accepted in the domestic market and the restrictions imposed before on foreign firms were removed, allowing their entry in the Indian market and further expansion.
Two routes by which India receives FDI were created. An automatic route, where no approval from Indian authorities is required, that applies to the majority of the industries, or at least until a threshold of ownership by the foreign firm. Included in this group are industries like the private banking sector and Internet services provision. However, certain activities are not covered by the automatic route and require prior approval by the government for foreign investment. In March 2005, further attention was given to the
potential of FDI in the Indian economy, as its policy was again revised and the decision was made that FDI was to be allowed up to 100% foreign equity, through the automatic route in any firm operating in townships, housing, built-up infrastructure and construction-development projects. More recently, the prime minister Narenda Modi launched the initiative “Make in India”, with the purpose of encouraging multinational enterprises to manufacture their products in India and attract more FDI to the country.
These reforms resulted in continuously increasing FDI and trade volumes in India, in the following years and also recently. Table 1 plots a summary of the evolution of international trade in India. Both import and export of goods and services show a very significant increase since the first attempts of liberalization, through a strengthening of the liberalization process in recent years.
Table 1 Export and import of goods and services evolution (in billions of US$)
Billions of current US$
1985
1995
2005
2015
Export of good and services
19.8
61.4
224.6
479.3
Import of goods and services
25.2
76
241.5
502.4
Total trade
45
137.4
466.1
981.7
Table 2 Share of World Export/Imports (%)
Share of World Exports/Imports (%)
1985
1995
2005
2015
Exports
0.53
0.61
1.24
1.96
Imports
0.75
0.69
1.44
2.24
India’s share in world exports of goods and services, which had declined from 2% at independence (Arvind Panagariya, 2004) to 0.53% in 1985 (Table 2), bounced back to 0.61% in 1995 and further to 1.24% in 2005. Since the mid-1980s, India’s exports of goods and services have grown faster than world exports (World Bank). India’s share of world imports more than tripled since the implementation of the liberalization policy, as a result of the
reduction of tariffs imposed before by the authorities. Again, Indian imports have been growing faster than world imports ever since the mid-1980s. The numbers leave no doubt that India’s external sector grew significantly and gained an important role in the economy since the liberalizations of the 1990s.
FDI flows have also been playing an increasingly prominent role in the economy. Figure 1 shows the evolution of FDI as a percentage of GDP, where we can clearly identify an upward trend starting with the implementation of the liberalization policy in the country. FDI went from a small fraction of 0,027% in 1991 to 2,07% in 2015.
In 2008, a peak was reached as a result of the big reform implemented in 2005, where several industries were completely opened for FDI. With the new rules regarding foreign investment, India became a more attractive destination for investment in the eyes of the world, raising the inflows of FDI. The global crisis that arose in the subsequent years led to a clear decline in the inflows of foreign investment in India. While other Emerging Market Economies (EMEs) recovered by 2010-11, India continued to suffer from a significant reduction of inflows of FDI. A survey of empirical literature and analysis developed by the Reserve Bank of India (RBI) suggests that these divergences were a result of procedural delays, complex rules, and regulations regarding the investment process that pushed investors away. Furthermore, policymakers’ actions created uncertainty regarding the stability of the country, creating an unfriendly business environment in India.
The campaign launched in 2014 had these problems in mind and its results show the potential impact that FDI has in this economy. The boost in FDI in the following years resulted in a major increase in FDI job creation, from 116000 new jobs in 2013 to 225000 in 2015, the highest number in the world (The FDI Report 2016).
0 0.5 1 1.5 2 2.5 3 3.5 4
FDI (% of GDP)
Figure 1 FDI as a % of GDPFinally, we analyse Indian Gross Domestic Product (GDP) and its evolution by looking at Figure 2. The Indian economy has been continuously growing ever since the first attempts of liberalization came up in the country. However, the pace of growth has visibly been speeding up as time goes by and as India liberalization was intensified. The average annual growth rate went from 5,5% in the period between 1986 until 1995 to 7,6% in the period between 2006 and 2015.
The above analysis has shown that international trade, FDI and economic growth have increased ever since India has started to become a market oriented economy. The liberalization process initiated back in 1970s and strengthened in the 1990s resulted in deepened international relations, both in trade and investment, between India and the rest of world, that consequently led to an exponential growth of the economy. It is therefore interesting to study the interrelation between these four variables in the period between the implementation of the liberalization policies and recent years and understand if and how they are connected. 0 250 500 750 1000 1250 1500 1750 2000 2250 2500 1985 1986 1987 1988 1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014 2015
GDP (billions of 2010 constant US$)
III. Literature Review
The amount of literature relating FDI, economic growth and international trade is still scarce. However, a greater deal of attention has been given to the interrelations between these variables. In the next paragraphs, I will describe some works about these links that somehow motivated my thesis.
Despite the large amount of literature regarding the linkage between FDI and growth, there still remains large controversy among researchers about the sign and size of this relation. As already mentioned earlier in this paper, Borensztein, Gregorio and Lee (1998) in a cross-country regression framework, with data on FDI flows from industrial countries to 69 developing countries, found that knowledge diffusion as a consequence of FDI can lead to improvements in productivity and efficiency in local firms. Dees (1998), using a constant elasticity of substitution (CES) production function to assess the effects of FDI on growth in China, finds that FDI positively affects Chinese growth through the diffusion of ideas and its influence on technical change. Also, Lucyna Kornecki and Vedapri Raghavan (2011) using a regression growth model based on the production function, conclude that FDI stock, in the past two decades in the Central and Eastern European countries, constitutes an essential factor of economic growth, when compared with labour, capital, and exports.
On the other hand, Peter Nunnenkamp and Julius Spatz (2003) in a cross-country analysis, concluded that positive growth effects of FDI in developing countries are anything but guaranteed, and depend largely on country characteristics, such as GDP per capita, schooling, institutional development, and trade openness, and the type of FDI receiving industry. Charkovic and Levine (2002) used Generalized-Method-of-Moments (GMM) estimators and found that while FDI flows may go hand-in-hand with economic success, they do not tend to exert an independent growth effect.
Other studies have been trying to understand whether the causality could also go the other way around, that is if growing economies tend to attract market-seeking FDI. Based on the Granger Causality Test approach presented by Toda and Yamamoto (1995) and Yamada and Toda (1998), which consists on a modified Wald test statistic that can be applied to any time series regardless of their order of integration, Ericsson and Irandoust (2001) examine the effect of FDI on output and total factor productivity growth in the host country for Denmark, Finland, Norway and Sweden. Their results show that FDI and output in Norway and Sweden
are co-integrated, implying that they are causally connected in the long-run. The Granger-causality is bi-directional in Sweden, proving the possibility that economic growth is also causing current values of FDI flows. However, in Norway, they find only unidirectional causality, running from FDI to economic growth.
Also in a co-integration framework, Parantap Basu, Chandana Chakraborty and Derrick Reagle (2003), study the two-way link between FDI inflows and economic growth for 23 developing countries, including India, in the time period between 1978 and 1996, while exploring the role of trade liberalization in this link. The co-integrating vectors reveal a bi-directional relation between FDI and GDP growth for more open economies. However, for relatively more closed economies, the relation is unidirectional and runs from GDP growth to FDI implying that, in these economies, FDI and economic growth are not reinforcing each other emphasizing the role of international trade in the FDI and growth relation.
Focusing on works regarding the Indian economy, Sarbapriya Ray (2012) and Chakraborty and Basu (2002), following the Johansen procedure to test the presence of co-integration and performing the Granger causality test in order to find short run causality, using data from, respectively, 1991 to 2011 and 1974 to 1996, both found unidirectional Granger causality, running from real GDP growth to FDI inflows, which goes along the results found by Parantap Basu, Chandana Chakraborty and Derrick Reagle (2003). Moreover, Sarbapriya Ray concludes that FDI has not contributed to India economic growth and Chakraborty and Basu find that FDI and the share of import duty in the tax revenue are negatively related, leading to the conclusion that further trade openness and FDI inflows into India move hand in hand.
That said, there has also been an ongoing effort to study the causality relation between FDI and international trade, whether FDI reinforces trade or if trade relations have an effect on the direction of FDI, or if they are actually reinforcing each other. Kevin Honglin Zhang and Shunfeng Song (2000) construct a dynamic model with panel data from 1986 to 1997 and show that increased levels of FDI in China positively affect the provincial manufacturing export performance. James R. Markusen and Anthony J. Venables (1996), through the construction of a two country oligopoly model, where multinational enterprises arise endogenously, find that FDI becomes more relevant than trade as countries become more similar regarding size, relative endowments and as world income grows. In other words, FDI and trade coexist in the case where countries involved apart in their size and endowment
characteristics, otherwise FDI will substitute trade. Joshua Aizenman and Ilan Noy (2005), through the estimation of linear regressions between a trade openness index and a financial openness index, using data from 81 countries from 1982 to 1998, find that that greater trade openness in goods has a positive effect on FDI net inflows and vice versa.
Deborah L. Swenson (2004), using a simple model of production in the United States of America (USA) and data for the time period from 1974 to 1994, analyses the possibility of a substitution relation between FDI and USA imports, as FDI might induce the production of previously imported goods and services. The author distinguishes three possible effects of FDI on imports, calling them Product, Industry, and Overall manufacturing effect. The Product effect measures the effect of product j FDI on product j imports, through, on one hand, potential import demand expansion, on the other, substitution effect between FDI and imports. The Industry effect is the more ambiguous as is a mix between competing effects. If the foreign production technique and input mix for product j are different from the home one, then imports of intermediary inputs needed for that production will increase. However, as foreign producers become familiar with the home market, the need for importing intermediate goods will drop. Finally, the last effect captures any network effects or information externalities brought by foreign firms know-how and is expected to have a positive impact on imports. Swenson finds that FDI substitutes for trade at the product and industry level, yet it stimulates imports at the overall manufacturing level.
In an attempt to assess the contribution of FDI inflows into India to its rising exports, Kishor Sharma (2000) uses annual data for the period between 1970 and 1998 and in a simultaneous equation framework investigates the determinants of export performance in India. He finds that, though not statistically significant, the coefficient for FDI has a positive sign.
Finally, the causal links existing simultaneously between FDI, economic growth, and trade have been given little attention. G. Jayachandran and A. Seilan (2010) analysed the long-run and causal relationship between economic growth, measured by GDP, FDI, and trade, measured only by exports of goods and services, in India using yearly data from the period between 1970 and 2007. Based on Johansen co-integration test they reject the null hypothesis of no co-integration between the variables, leading to the conclusion that these are governed by a long-run relation. Moreover, using the Granger causality test, they find an unidirectional causality running from FDI and exports to economic growth.
Xiaohui Liu, Peter Burridge and P. J. N. Sinclair in 2010 introduced the literature with the only study that investigates these links extensively. The authors used quarterly data for Chinese exports, imports, GDP growth and FDI inflows for the time period from 1981 to 1997 and base their study in the Johansen co-integration test to explore the connections between these variables. The Johansen test results in the existence of two co-integrating vectors, which are also described in the author’s paper, after the imposition of normalizing and exclusion restrictions. They find bi-directional Granger causality between GDP, FDI, and exports, but only unidirectional running from these three variables to imports.
The interest in the causal connection between economic growth, FDI, exports and imports arises in a context of development strategies, of special importance in a growing economy like India. Failing to address the possible existence of these links might produce spurious results in the analysis of the relation of these four variables, and mislead the countries into policies that do not take advantage of the real causal connections between them. It is clear from this section that these links are yet to be fully understood by the literature, and with that in mind, this thesis will attempt to make them more clear, based on data from India.
IV. Methodology
The idea that historical relationships can be generalized to the future is formalized by the concept of stationarity. A time series is said to be stationary when its joint probability distribution does not change when shifted in time. In other words, the mean and variance of the series are constant and independent from time. However, if the future differs fundamentally from the past, then these historical relationships might not be reliable predictors for the future. In this case, times series are said to be non-stationary.
Non-stationary time series that follow a common stochastic trend might move together in an equilibrium relationship characterized by a stationary stable process. Non-stationary variables with this property are called co-integrated. Co-integration means that, although many developments can cause individual changes in each variable, moving them apart from each other in the short-run, there is some long-run relation tying the components together.
In a stationary scenario, in order to understand the interplay of the four variables, a standard statistical significance test would be carried out via a Vector Autoregressive Model (VAR) involving them. The corresponding VAR to this thesis is represented by the four equations below. The null hypothesis that FDI does not cause GDP, given Exports and Imports would be tested via a standard F-test to all the coefficients in Equation (1). The F-test is a test statistic that follows a F-distribution, usually used to compare statistical models and to find which one is the best representation of the population model. The hypothesis is rejected in the case where all coefficients are jointly significant, therefore a causality relation between FDI and GDP, running from FDI to GDP is confirmed. Likewise, if in Equation (2) all coefficients are jointly significantly different from zero, the null hypothesis that GDP does not cause FDI, given Exports and Imports is rejected, implying that GDP is causing FDI, given Exports and Imports. This test would be carried out for all possible relations among the four variables. !"#$ = &' + &)*!"#$+* , *-) + &./0"1$+/ , /-) + &2345$+3 , 3-) + &6718$+7 , 7-) + 9)$ (<) 0"1$= >'+ >)*0"1$+* , *-) + >./!"#$+/ , /-) + >2345$+3 , 3-) + >6718$+7 , 7-) + 9.$ (?)
45$ = @'+ @)*45$+* , *-) + @./!"#$+/ , /-) + @230"1$+3 , 3-) + @6718$+7 , 7-) + 92$ (A) 18$ = B'+ B)*18$+* , *-) + B./!"#$+/ , /-) + B230"1$+3 , 3-) + B6745$+7 , 7-) + 96$ (C)
However, if the series under study are not stationary, usual test statistics no longer follow a normal distribution so that the tests described before are not valid. Under non-stationarity, the assumption that Ordinary Least Squares (OLS) estimators are consistent and asymptotically normally distributed does not apply. This implies that statistical inference based on standard tests, like the F-test, cannot be made, since these tests don not have a normal distribution anymore. Therefore, the analysis described in the previous paragraph does not apply to non-stationary data. Using this procedure to find statistical relations between the variables, if they are non-stationary, might result in misleading results, like spurious correlation, where a significant statistical causality is found due to either coincidence or some unobservable factor.
Moreover, if we are in the presence of co-integration, the equations relating the variables will be different from the described above, as will be their interpretation. The VAR model is not the appropriate model to represent the interplay of the four variables and should be substituted by a Vector Error Correction Model (VECM). In a VECM, changes in a variable depend on the deviations from some equilibrium relation that connects the group of variables. A VECM describes how the variables interconnect both in short and long-run framework, where the long-run represents the steady state, and the short-run the path to this equilibrium. Thus, it is of great importance to considerer the possibility that non-stationary variables are co-integrated.
a) Unit Root test
In order to proceed with our co-integration analysis, we first have to find the integration properties of our time series. The existence of a co-integration relation between variables implies that each variable is non-stationary and that all series share the same degree of integration. The degree of integration of a time series is a summary statistic which reports the minimum number of differences required to obtain a stationary series. Differencing a time series consists of calculating the difference from one period to the other, for all periods.
The first difference of a time series is the series of changes from one period to the next, the second difference is instead the change from one period to two periods ahead, and so on. Therefore, if a time series is integrated of order one, this means that its first difference is integrated of order zero, or that is stationary. A time series integrated of order one is also known as unit root process. Unit root processes have specific characteristics, namely that shocks to this process will have permanent effects, since lagged values of the dependent variable will always have an impact on the current value, unlike stationary processes, where temporary shocks decay from one period to another.
This analysis will be done by performing the Augmented Dickey-Fuller test (ADF) for unit roots. The ADF test is an extension of the Dickey-Fuller test, named after its inventors David Dickey and Wayne Fuller, and is used to test for the existence of a unit root in an autoregressive process of order p (AR(p)). Each variable will be individually tested, both in levels, 5$, and first differences, ∆5$+), to ensure that the time series are all integrated of
order one (I(1)) and, therefore, the four series share a common stochastic trend. As already mentioned, this study works throughout with the natural logarithms of the variables, so that the first differences correspond to growth rates. If the null hypothesis of a unit root is not rejected for the variables in levels and is rejected for the variable first differences we can conclude that the variable is integrated of order one, I(1).
Depending on the existence of a drift or a time trend in the series, the ADF tests for a unit autoregressive root are based on the following different regression equations
1. Teste for no constant (drift) and no trend in the data
∆5$ = &5$+)+ B)∆5$+)+ B.∆5$+.+ ⋯ + BF∆5 + 9$ (G) 2. Test for constant (drift) but no time trend
∆5$ = H'+ &5$+)+ B)∆5$+)+ B.∆5$+.+ ⋯ + BF∆5 + 9$ (I) 3. Test for constant (drift) and linear time trend
∆5$ = H'+ H)J + &5$+)+ B)∆5$+)+ B.∆5$+.+ ⋯ + BF∆5 + 9$ (K) where the null hypothesis is L': & = 0 against the alternative L): & < 0.
Under the null hypothesis, X has a stochastic trend; under the alternative hypothesis, X is stationary. The ADF statistic is the OLS t-statistic testing & = 0 in Equations (5), (6) and (7).
is possible to infer that each variable in levels would be accurately modelled by including a constant and a linear time trend. Therefore, when performing the ADF test, both will be included, corresponding to Equation (7). However, the graphs that describe each series first differences, Figure 6 to 9, suggest that the inclusion of a time trend no longer makes sense as the variables show no movement towards a general direction. Hence, in order to test the integration level of each series first differences, only a constant is included in the model, represented by Equation (5).
The ADF statistic does not have a normal distribution, even for large samples. Thus, critical values applicable have been identified by the creators of the test through a Monte Carlo simulation. These critical values depend on whether a constant or a time trend have been included in the regression and the value p of lags included.
The appropriate lag length will be chosen by minimizing the information criterion proposed by Gideon E. Schwarz (1978), also known as the Schwarz information criterion (SIC). The SIC trades off the advantages and disadvantages of including more lags in the model, so that the number of lags that minimizes the SIC is a consistent estimator of the true lag length. A. Hall (1994) examined the impact of data-based lag-length estimation on the behaviour of the ADF test for a unit root, more specifically derived the conditions under which the ADF test converges to the appropriate Dickey-Fuller distribution. They found that the SIC yields on average more parsimonious models compared to other lag length selection methods, which results in more powerful ADF test. For this reason, SIC is the method chosen to be used in this thesis.
b) Johansen’s Co-integration Test
The Engle-Granger residual based test is one of the most commonly used co-integration test. The test consists of two simple steps: first, the estimation of a co-integrating regression between the variables in consideration by applying OLS, and second use the augmented Dickey-Fuller (ADF) test to find the order of integration of the residuals. If the residuals are stationary we can conclude that the variables are co-integrated and that the regression previously estimated is a true estimation of this relationship.
Despite its frequent use in the literature, this method has some crucial limitations. If we are dealing with more than two variables in the model, there can be more than one co-integration vector describing the equilibrium relationships featured by the variables in
analysis and their joint evolution. Assuming that there is only one co-integration vector, when in fact there are more, leads to inefficient results in the sense that only a complex linear combination of all possible vectors can be obtained when estimating a single equation model (Chandana Chakraborty and Parantap Basu, 2002). Even if there is only one co-integration vector, estimating a single equation can lead to inefficiency due to the possible loss of information by only allowing one of the variables to be potentially endogenous and imposing exogeneity on all the others. Given this approach limitations, the results of the Engle-Granger test when there are more than two variables in the model might be misleading.
Johansen (1988, 1991) developed an approach, further extended later by Johansen and Juselius (1990), which solves the main limitations of Engle-Granger approach and therefore is considered superior. This approach strategy, based on maximum likelihood estimates, makes it possible to estimate all co-integrating vectors when there more than two variables.
If the variables are indeed co-integrated, then their relation can be represented by the following Vector Error Correction Model (VECM)
∆P$ = QP$+)+ 3+)R*∆P$+* *-) + S$ (8) where T = 4, P$= !"#$ 0"1$ 45$ 18$
and S$ is white noise, which may be contemporaneously
correlated, and we assume to be stationary.
In a vector error correction model, changes in a variable depend on the deviations from some equilibrium relation. Already we can see how these type of model, introduced before the concept of co-integration, closely relates to the it.
The VECM is a transformation of the VAR described by Equations from (1) to (4). If our series are indeed integrated of order one, I(1), than in Equation (8), the matrix ∆P$ is
constituted by stationary series, since these are each series first differences, which are integrated of order zero, I(0). A stationary series can only be equal to something that is also characterized by stationarity. Therefore, the right hand side of Equation (8) must also be stationary. However, recall that if P$ is non-stationary, then so is P$+). That said, in order to
have a valid equality in Equation (8), there must be a linear combination of the variables in P$+) so that QP$+) is stationary. This combination is represented by the matrix Q.
Testing for co-integration in the Johansen’s framework amounts to finding the number of linearly independent vectors that describe the relations between the variables. A linearly independent vector is a vector that can not be defined as a linear combination of other vector in the set of vector in analysis. In our case, this set of vectors is described by the matrix Q in Equation (8). These vectors are represented by coefficients corresponding to each of our non-stationary variables, that combined result in a non-stationary series. Q is called the long-run matrix.
To find the number of linearly independent vectors in Q we must calculate the matrix rank. A matrix rank is the maximum number of linearly independent row vectors in the matrix. Q can be full rank if all vectors are linearly independent form each other. In this case the assumed stationarity of error term, S$, requires that the levels of the P$ process
themselves are stationary, implying the absence of any stochastic trends in the data, and therefore, of any possible co-integration relation among the variables. Q may have rank zero, and in this case Equation (8) reduces to a standard VAR, in first differences, similar to the one represented by Equations from (1) to (4), and there are no stationary long-run relations among the elements of P$. If 0 < V < n, there exist V co-integrating vectors relating the variables included in the model.
The matrix Q can be factorized in two matrices, the matrix of error-correction parameters W and the matrix of co-integrating vectors X as
Q = WX′
The coefficients in the co-integrating matrix X multiply the variables to deliver the linear combination of variables that does not have a unit root, therefore are stationary. The coefficients in W are the adjustment coefficients, which multiply the co-integrating relationship to deliver the response of variables to deviations of the co-integrating relationship from zero, thus the speed of adjustment to the equilibrium.
For the Johansen method, two tests for the presence of co-integration are proposed. The first test is called trace statistic; and the second is called maximum eigenvalue statistic. In the former, the null hypothesis is that the number of co-integrated vectors is less than or equal to V, where V = 0, 1, 2, …, and is tested against the relevant hypothesis for the study. In the latter test, the alternative for V = 0 is V = 1, and then V = 1 is tested against V = 2 and so on. H.
Lütkepohl, P. Saikkonen and C. Trenkler (2001) compare the two tests in their study and come to the conclusion that, for small samples, the trace test is superior to the maximum eigenvalue test, and recommend the usage of the former if one only wants to perform one of the tests. Therefore, in case of contradicting conclusions regarding the two methods, results from the trace test will be considered.
c) Granger causality test
Causality in the Granger framework contains a different connotation from the one usually given to the word. Granger causality means that if X Granger-causes Y, then X is a useful predictor of Y (Stock and Watson, 2011). According to Granger (1969), if some series Yt contains information in the past terms that helps in the prediction of Xt, and if this
information is contained in no other series used in the predictor, then Yt is said to cause Xt.
Granger causality refers to short-run causality since only a small sample of the past values of each variable is included to test if there is indeed causality running from one variable to the other. Therefore, by performing this test, we are analyzing the short-run dynamics between these variables to complement the long-run results from the previous section.
This test is developed under the assumption that the tested series are stationary. In the non-stationary case, the existence of causality may vary over time, leading us to values that are not testable or to ambiguous conclusions. Therefore, the test will be applied to each variable first differences, instead of levels. Since the stationary of the first differences is a requirement for the performance of Johansen co-integration test, this property is already ensured when we perform the Granger causality test.
The Granger Causality statistic is basically the F-statistic testing the hypothesis that all coefficients of one of the variables of the VAR model estimated with each variable first differences are equal to zero. Thus, the null hypothesis inherent to this test is that variable Y does not Granger-cause variable X, implying that Y and its lagged values are not useful in predicting values of X, beyond the predictive power of the other variables.
∆!"#$= &'+ ^)∆0"1$+ ^.∆45$+ ^2∆18$+ &)*∆!"#$+* , *-) + &./∆0"1$+/ , /-) + &23∆45$+3 , 3-) + &67∆18$+7 , 7-) + 9)$ (G) ∆0"1$ = >'+ _)∆!"#$+ _.∆45$+ _2∆18$+ >)*∆0"1$+* , *-) + >./∆!"#$+/ , /-) + >23∆45$+3 , 3-) + >67∆18$+7 , 7-) + 9.$ (I) ∆45$= @' + `)∆!"#$+ `.∆0"1$+ `2∆18$+ @)*∆45$+* , *-) + @./∆!"#$+/ , /-) + @23∆0"1$+3 , 3-) + @67∆18$+7 , 7-) + 92$ (K) ∆18$ = B'+ a)∆!"#$+ a.∆0"1$+ a2∆45$+ B)*∆18$+* , *-) + B./∆!"#$+/ , /-) + B23∆0"1$+3 , 3-) + B67∆45$+7 , 7-) + 96$ (b)
To decide whether, for example, FDI Granger-causes GDP, an F-test is carried out to examine the null hypothesis of non-causality, L': &.)= &..= ⋯ = 0. For the F-test, the
unrestricted model will include lagged values of the FDI variable, which will look like Equation (5), whereas the restricted model will only include lags of GDP and the other independent variables, excluding lags of FDI.
The rejection of the null hypothesis means that FDI Granger-causes GDP, thus past values of FDI are useful when predicting values of GDP. The same test will be performed of each variable first differences.
V. Data
The data used in this thesis includes 25 years, and the study period spans from 1991 through 2015. All data was collected from World Development Indicators database made available by World Bank. Xiaohui Liu, Peter Burridge and P. J. N. Sinclair (2010) used quarterly data for their study about China, which guarantees their analysis a wider range of observations and, consequently, more precise results. However, due to lack of complete and consistent quarterly data for Indian variables, yearly data will be used in this study.
The choice of the time period for the sample data set is explained by two factors. First, 1991 was chosen as the first observation since is the year when India’s most pronounced liberalization process began. From this year on, India’s government has done a continuous effort by reforming India’s policy regarding the rest of the world. Therefore, analysing the variables from this year on will result in meaningful relations among the variables and reflect the connections aimed to be found. Second, 2015 is the last year for which data was available for all four variables.
Gross Domestic Product (GDP) includes the gross value added by all resident producers in the economy plus any product taxes and excludes any subsidies not included in the value of the products. The dollar figures for this series have been converted from the Indian rupee, the domestic currency, using each year exchange rate. GDP is the chosen variable to measure economic growth, following the existing literature.
FDI inflows are defined by the World Bank as the sum of equity capital, reinvestment of earnings and other capital. It is a category of cross-border investment, in which a resident in one economy has control, or a significant degree of influence, on the management of an enterprise resident in another country. This series is also measured in US dollars.
Exports of goods and services represent the value of all goods and other market services to the rest of the world, while imports of goods and services represent the value of all goods and market services received by the country from the rest of the world. Both series include merchandise, freight, insurance, transport, travel, royalties, license fees, and services like communication, construction, financial information and government services. They exclude compensation of employees, investment income, and transfer payments. The two series are measured in US dollars.
Exports and imports are treated separately in order to allow for the possibility that their impact is asymmetric (Xiaohui Liu, Peter Burridge and P. J. N. Sinclair, 2010). This will enable the two variables to have different, even opposite influences in the interrelations we will study.
GDP, exports, and imports were calculated in constant 2010 US dollars in order to avoid the effect of inflation in our study. Since FDI was only available in current US dollars, following Chandana Chakraborty and Parantap Basu (2002), this series was deflated using the GDP deflator available in World Bank database, which base year is 2010. Therefore, all four series are presented in constant 2010 US dollars.
For the propose of our study, each variable will be converted into its natural logarithm so that the first differences correspond to growth rates. Figures 3, 4, 5 and 6 plot the evolution of each variable, over the period under analysis. First, looking at each graph individually, we can see how each series shows an upward stochastic trend that could be described by a unit root process. Second, putting the four graphs together, we can easily see how the four variables share a common trend, which can result in a co-integration relation.
Figure 5 Natural Logarithm of GDP (LnGDP) Figure 5 Natural Logarithm of FDI (LnFDI)
Finally, each variable first differences series is represented in Figures 7, 8, 9 and10, where, just by looking and analysing these figures, we can already make some suppositions regarding their stationarity properties. The four figures show no time trend since there is no movement towards a general direction. However, they appear to be developing around a constant mean over the time period under study with similar variations when departing from this constant mean. These observations lead us to suspect that these processes are characterized by stationarity.
Figure 7 First Differences of lnGDP Figure 8 First Differences of LnFDI
VI. Empirical Results a) Unit Root Test
In order to proceed with the Johansen co-integration test, the variables in study must be non-stationary and share the same order of integration. With this in mind, the Augmented Dickey Fuller test is applied for each variable, first in levels and second in first differences.
Before performing the unit root test, the optimal lag length used is chosen by calculating Model Selection criteria measures. According with the SIC, all Autoregressive models involving the variables in levels should be estimated with only one lag of the variable itself, and zero when involving variables’ first differences. Given this, we continue the performance of the ADF using one lag for each variable in levels and zero for each variable first differences.
The results from the first test are plotted in Table 3. The null hypothesis of a unit root is not rejected for all of the four variables in levels since the value of the test statistic is smaller then the critical values for all three significance levels, 1%, 5% and 10%. This means that we fail to reject the null hypothesis that the series are non-stationary, implying that there is no statistical evidence to prove that the four series under study are stationary processes.
However, when performing the same test for each variable in their first differences, the null hypothesis of non-stationary is rejected for all four variables, since this time the test statistic of each test is higher then the critical values correspondent to the three significance levels. The results from the performance of this test are presented in Table 4. We strongly reject the null hypothesis of non-stationarity for GDP, exports and FDI at a significance level of 1%. In the case of imports, the null hypothesis is only rejected at a significance level of 5%. Nevertheless, this implies that there is enough statistical evidence to prove the presence of stationarity in these processes. Therefore, the conclusion of the ADF test performed for each first differences series is that these are characterized by stationarity.
These two results combined imply that the four series are unit root processes, integrated of order one, confirming that these are not stationary variables. However, when differentiated, they become stationary.
Table 3 ADF test for unit root in variable levels
Variables
ADF (including trend)
Level
Critical Values
1% 5% 10% lnGDP -1.895-4.380 -3.600 -3.240
lnEX -0.735 lnIM -1.076 lnFDI -2.628Table 4 ADF test for unit root in variable First Differences
Variables
ADF (intercept only)
First Differences
Critical Values
1% 5% 10% lnGDP -3.846 ***-3.750 -3.000 -2.630
lnEX -3.897*** lnIM -3.037** lnFDI -4.347***Given the results from the ADF tests applied to the four series we can now continue our analysis and test for the presence of co-integration making use of the Johansen procedure.
b) Johansen Co-integration test
Since we have already confirmed the presence of a unit root in all our variables, we can now proceed to the Johansen integration test in order to test the presence of a co-integration relation among the variables, and find the number of co-integrating vectors that describe the long-run relation between the variables.
The first step towards the test is to find the optimal lag length for the unrestricted VAR described by Equations (1) to (4), since this is the model in which the procedure bases its analysis to test the presence of co-integration between the variables. This step will be done by minimizing the information criterion proposed by Gideon E. Schwarz (1978), also known as Schwarz information criterion (SIC). The results are shown in Table 5, where the Akaike Information Criterion (AIC) was also included as a measure of comparison. The maximum lag length chosen to calculate AIC and SIC is four due to fact that we are dealing with a small Note: ***, ** and * denote significance at the 1%, 5% and 10% levels, respectively.
sample and higher number of lags will imply lower number of observations. The optimal lag length that minimizes the two criterions is four, and therefore will be the one used in the performance of the Johansen co-integration test.
Table 5 AIC and SIC for Optimum lag length in Unrestricted VAR
Lag
Length
0
1
2
3
4
AIC -1.37772 -8.70478 -8.66277 -10.91 -17.4194*
SIC -1.17876 -7.71 -6.87216 -8.32352 -14.0371*
Table 6 lays out the results from the co-integration test, where both trace and maximum eigenvalue statistics are presented. The several tested hypotheses are presented in the first column, followed by the correspondent alternative hypothesis. The eigenvalue and trace statistic are presented, respectively, in the third and fifth column and the correspondent critical values, for a 5% significance level, are presented in the fourth and sixth column.
Starting with the first row, both trace and maximum eigenvalue statistics strongly reject the null hypothesis of no co-integration, r = 0. Similarly, the null hypothesis that there is at most one co-integrating vector, corresponding to r ≤ 1, is rejected by the trace statistic, since its value, 71.7279 is above the 5% critical value of 29.68. The maximum eigenvalue test also rejects the null hypothesis of the presence of only one co-integrating vector r =1, against the alternative of r = 2.
The trace test fails to reject the null hypothesis that r is at most 2 at a 5% significance level. The trace statistic, 15.01215, is below the 5% critical value, 15.41. The maximum eigenvalue statistics confirms this result, as the statistic for the null hypothesis that the rank is 2 is 11.5281, below the 5% critical value of 14.07. Thus we accept the null hypothesis that there are two co-integrating vectors in the multivariate model selected for the purpose.
Table 6 Results from Johansen Co-integration test
Null
Hypothesis
Alternative
Hypothesis
Max-Eigenvalue
Statistic
Critical
Value
(5%)
Trace
Statistic
Critical
Value
(5%)
r = 0 r ≥ 1 126.6855** 27.07 198.4133** 47.21 r ≤ 1 r ≥ 2 56.6063** 20.97 71.7279** 29.68 r ≤ 2 r ≥ 3 11.5281 14.07 15.1215 15.41 r ≤ 3 r = 4 3.5935 3.76 3.5935 3.76This outcome implies that these time series can be described by two different and linearly independent long run relationships. These linearly independent combinations of GDP, FDI, exports and imports, which we found previously that are non-stationary series, are stationary processes, with constant mean and variance. Unlike non-stationary times series, a stationary process implies that historical relationships can be generalized to the future, which can be useful when trying to forecast future value of these variables, making use of these equilibrium relation.
The conclusion we take from this test is that there are two steady-state equilibriums that define the interactions between economic growth, FDI and international trade in India. What is meant by equilibrium relationship is that this would be the relationship these four variables would have without any short run shocks, or the relationship from which variables might deviate temporarily, but always return to.
The co-integrating vectors that describe this long run relationship should be seen as two identities, since they should always be true independently from the values that each variable represented in them takes. Therefore, despite the possibility of temporary deviations from this equilibrium, which can happen in the short run, the presence of co-integration among Indian GDP, FDI, exports and imports imply that there are two co-integrating equations relating them that will always be true in the long run, when the variables have return to their steady-state value.
c) Granger causality test
Finally, we look at short-run dynamics between economic growth, FDI, exports, and imports in India by performing the Granger causality test described before. The results from Note: ***, ** and * denote significance at the 1%, 5% and 10% levels, respectively.
the test are plotted in Table 7. The Wald test reveals that there is a unidirectional Granger causality running from exports, imports and FDI to GDP. According to the definition of causality in this framework, this result implies that past values of volumes of international trade and FDI are useful when predicting future values of India’s GDP. This result goes along with what one would suspect, and also follows some of the results from previous literature (Ericsson Irandoust (2001) and G. Jayachandran and A. Seilan (2010)). Nevertheless, is an important outcome as it implies that there is statistical evidence that proves that future economic growth in India is, at least, partly caused by past value of FDI inflows, exports and imports. Therefore, the positive and growing economic growth we witnessed in India for the past two decades is certainly caused by the growing international trade and investment that have gained a significant role in the economy ever since the liberalization policy was set in the country. India’s change of strategy towards the rest of the world, and all the ongoing efforts by the local government to keep improving and innovating its international policy have resulted in higher and more consistent growth of the economy.
Moreover, the test shows a unidirectional causality relation running from volume of imports to FDI inflows. Thus, past values of import volumes into India are useful information when predicting value of FDI inflows into he country. A plausible explanation for this outcome is that stronger commercial relationships between two economies make investment between the two more propitious to occur. Thus, lagged values of import from one country into India could be causing future value of FDI into India from this country, and consequently from others.
However, the test fails to reject all the other hypothesis of non-causality, the null hypothesis that all higher-order coefficients in the VAR, on each variable in turn, are zero. Thus, there is not sufficient statistical information to confirm any causality between the remaining variables.
Table 7. Results from Granger causality test
Dependent
Variables
Wald test statistics
Causality
Inference
∆lnGDP ∆lnEX ∆lnIM ∆lnFDI
∆lnGDP - 9.1071* 9.8289** 10.793** EX > GDP IM > GDP FDI > GDP ∆lnEX 2.0805 - 1.125 0.29103 ∆lnIM 1.8812 3.2363 - 0.23691 ∆lnFDI 4.8651 2.6386 14.227** - IM > FDI
The failure to find a causality also running from GDP to FDI is rather surprise since it is in disagreement with a several studies mentioned in the literature review chapter, therefore it is a result worth to considerer in more detail. Sarbapriya (2012) and Chakraborty and Basu (2002), using data from, respectively, 1991 to 2011 and 1974 to 1996 in India, both found unidirectional Granger causality, running from real GDP to FDI inflows, contrary to our findings. The first probable reason for these contradicting results is related to the difference in the time periods used as basis for each. Particularly in this thesis, that makes use of the most recent data available until today, adds new information that might have changed the direction of the causality between the two variables. Thus, it might be that the contemporary relation between real GDP and FDI inflows in India is now running from FDI to GDP, and no longer the other way around. The change in the direction of the causality might be explained by India’s historical relation with foreign investment, which used to be banned from most industries before the liberalization of economy took place. As FDI started to gain a prominent role in the Indian economy, economic growth was a vital information for foreign investors decision on whether or not to proceed with investment in India. In this case, we would expect that the causality would run from GDP to FDI. However, as FDI inflows in India keep expanding and gaining an important role in India’s economy, the causality changed direction from FDI to economic growth. FDI inflows into India have made a great contribution to the country economic develop.
Second, we take into consideration the results reached by Parantap Basu, Chandana Chakraborty and Derrick Reagle (2003) in their study regarding developing countries, including India, during the time period of 1978 and 1996, where they find a bi-directional causality relation between FDI and GDP for more open economies, and a unidirectional Note: ***, ** and * denote significance at the 1%, 5% and 10% levels, respectively.
causality otherwise. Even though India liberalization process has undoubtedly opened India’s economy to the world, it is also certain that there is still a lot of improvements to be done. Or, it is possible that, even after Indian efforts in liberalizing its economy, when compared to other economies, India is still a closed economy, which would justify the absence of a two-way causality between FDI and GDP according to the authors.
VII. Limitations
After the empirical analysis we did in the previous chapter, is important to note that some limitations are to be considered when withdrawing conclusions from the results generated by the methodology chosen.
The first limitation of this thesis is the small sample in which the analysis in based on, which might threaten the quality of the results. Moreover, since we are dealing with time series data, and the usage of lagged values of the variables is important and recurrent, several observations are lost, further reducing the size of the sample. Therefore, subsequent research of this topic should take this fact in consideration. The study would be optimized by using quarterly data as Xiaohui Liu, Peter Burridge and P. J. N. Sinclair did in their analysis about China, ensuring a much larger number of observations during the same period. However, as mentioned before, this kind of data is not currently available in a reliable source or in a usable form, which led to the decision of proceeding the thesis with the yearly data that was found from consistent sources.
Nevertheless, the results obtained with this analysis are still a valued contribution to the literature regarding the interconnections existent between economic growth, FDI and international trade in India, using the most recent data available which guarantees results that match the reality from the present.
The second limitation is to related the methodology used in this study. Though the Johansen co-integration test is considered to be superior to other procedures, still has its share of flaws. The first and most prominent one is the fact that the test results are very sensitive to the lag length chosen for the test performance, which risks the validity of the results in case the lag length used in not the appropriate. Second, this method assumes that the co-integrated vectors remain unchanged during the period of study, which might not be the case. Technological progress, economic crisis, policy or regime changes might change the long-run relation described by those equations. This problem is more pronounced in the case of large samples, however it should not be forgotten when withdrawing results from this thesis results.
