Financial Market Integration of South Asian Countries: Panel data Analysis

: According to Frankel (1992) in order to find financial integration from Feldstein Horoika (FH, 1980) model, the real interest parity must hold. This paper estimates the degree of financial market integration of South Asian countries i.e. Pakistan, India, Bangladesh, Sri Lanka and Nepal with both the techniques. The study finds some degree of integration with FH model has which increased after 1990s, post liberalization period. Furthermore, Panel Unit Root techniques i.e. LLC, IPS and Hadri has been used to estimate the real interest rate differentials (RIDs) of South Asian countries are found to be stationary with USA, Canada, UK, Germany, Sweden, Netherland, Australia, Malaysia, Indonesia, South Korea, Singapore, China and Japan. The empirical evidence of integration with both the techniques in my study is unique in the literature. Even though, the RIDS technique provides strong evidence of integration, correlation between savings and investment is still significant.


INTRODUCTION
In the era of Globalization and information technology, countries have come closer to each other.
The volume of merchandized trade and mobility of capital flows have been enhanced. Investors are able to diversify their portfolios by investing their capital almost anywhere in the world. The emerging markets are eliminating capital controls and introducing market friendly policies to attract foreign capital flows in the form of foreign direct investment or the equity flows. Free and perfect capital mobility refers to highly integrated financial markets.
The degree of capital mobility or financial integration is vital to be known for macroeconomic models. The degree of market integration can be estimated with interest parity conditions, saving-investment correlations of Feldstein and Horoika (1980) and degree of monetary autonomy 1 . Frankel (1992) mentions that if there is low correlation between savings and domestic investment, real interest parity must hold. My study contributes to the literature by applying both these techniques to the panel of South Asian countries. This is also the first study on South Asian markets to the best of my knowledge. Furthermore I applied liberalization dummies and estimated country slope dummies to find whether there is asymmetry in the correlation of savings and investments. FH (1980) used panel data of 21 OECD countries from 1960-1974 and found controversial result that the domestic savings coefficient in investment is almost one implying capital immobility. Feldstein (1983) added post OPEC years in regression and found same results. Penati and Dooley (1984) estimated same results and argued that since incremental savings remain in home country so capital is not very mobile. Dooley et al. (1987) used data from 64 industrialized and developing countries 1960-1984 and found higher savings coefficient. Bayoumi (1990) also confirmed the results of FH (1980) but held government policy responsible for this correlation. Haque and Montiel (1991) estimated degree of financial openness in developing countries and found higher integration. Yamori (1995) found higher savings coefficient but argues that it's due to non-zero currency premium similar to Frankel (1991Frankel ( , 1992. Jansen (1996) found stationary current account to be the reason of FH (1980) findings. Coakley et al. (2001) used panel unit roots and cointegration techniques using quarterly data on 12 OECD countries 1980-2001 found that savings and investment are I(1) and generally do not cointegrate. Chakrabarti Avik (2006) used annual data of 126 countries and found positive and significant association between savings and investment. But interestingly found lower coefficient for non OECD countries than OECD. Cooray and Sinha (2007) used data for 20 African countries and found high correlation using Johansen and fractional cointegration tests. Adedeji and Thornton (2008) used pooled data for 50 developed and developing countries for the period 1970-2000 and found that savings and investment are non stationary and cointegrated but also found differences in savings retention ratios.
The real interest rate parity hypothesis (RIPH) states that if the agents are rational and arbitrage forces are free to act in goods and assets markets, then real interest rates between countries will equalize. According to Ferreita et al. (2007) there are few studies which have tested RIPH through Unit Root analysis on RIDs. 2 But the literature does not offer conclusive answer. This is obvious from literature on both the techniques provided different results. The results remained mixed as shown by literature. I noticed there is hardly any work which has provided estimates from more than one technique as suggested by Frankel (1992). My study intends to fill this gap.
. The South Asian countries i.e. Pakistan, Bangladesh, India, Sri Lanka and Nepal started financial liberalization process in the early 1990s. 3 The region is important since the aggregate net flows to this region increased since mid 1980s. 4 It reached $9.3 Billion in 1989. According to Global Development Finance (2006) private capital flows to South Asia more than doubled since 2000. They reached $23.6 billion in 2005 as compared to $9.7 billion in 2000. The FDI increased to $8.4 billion in South Asia, an increase of $1.2 billion since last year. The report mentions that India received major share of capital flows to South Asia. The Liberalization efforts in the 1990s and the subsequent surge in the capital flows to South Asian countries make them a special candidate to study.
The study intends to use panel of South Asian countries to examine integration of financial markets in these countries. Furthermore, it also measures the impact of liberalization on integration whether it increased or not?
The present study also adjusts some of the econometric criticism levied against FH (1980) and observes whether the estimates remain the same when the model is adjusted accordingly.
Evaluating the overall results from all the techniques will make the final conclusion.

ORGANIZATION OF STUDY
The study has been analyzed and arranged as follows: In section 1 introduction and main objectives of the study are provided. In section 2, the estimates of descriptive statistics are discussed. The average savings and investments pertaining to individual South Asian countries are calculated. Section 3 provides the detailed methodology pertaining to Feldstein Horoika, and real interest rate differential test. The variable wise data sources and discussion of the methodologies to overcome the shortcomings have been provided.
The original form of equations and the expected signs are briefly discussed. Section 4 pertains to the interpretation of empirical findings and the comparison of panel data results with various techniques. Last chapter contains concluding remarks and possible policy implications for South Asian economies. Feldstein and Horioka (1980) model is primarily based on domestic saving and Investment

PATTERN OF SAVING AND INVESTMENT IN SOUTH ASIA
Relationship. Therefore it is necessary to observe average saving and investment. The decadewise changes in the saving-investment to GDP ratios are discussed below.
The average saving and investment to GDP ratio in Bangladesh are 10% and 18%, respectively, for the period 1970-99. In 2000 the S/Y and I/Y increased and their gap decreased to -5.7. In each decade, the ratio of savings and investment to GDP increased and indicate stable pattern overtime.
In India, savings and investment to GDP ratios were highest in the whole region. Both the ratios showed a stable pattern and a slight increase over decades. The saving-investment gap for India is also lowest in the region estimated as 1.67% of the GDP showing very low dependence on Foreign Capital. It clearly shows that domestic savings could finance most of the Domestic Investment in the case of India.
In the case of Nepal, saving investment gap is 12.74 of the GDP in 2000s, highest in region.
Although savings to GDP ratio increased over time except in 2000s, the ratio of investment to GDP increased more than that. In the 2000s investment to GDP ratio was at ever-highest level of 23.6%.
Average saving and Investment to GDP ratio in Pakistan is estimated as 16 and 17.8 percent to GDP ratio in 2000s. The S/Y increased but I/Y decreased in 2000s as compared to 1990s. But the savings investment gap decreased sharply to 1.81 in 2000s from 11 in the 1990s. But this gap in 2000s remained more volatile than 1990s since CV is very high in 2000.
In Sri Lanka the average I/Y and S/Y are estimated to be 24.8 and 16.26 respectively. The I/Y is 2 nd highest after India in the region in 2000s. The savings-investment to GDP gap was zero in 1990s but increased to 8.51 which show that in the 20000s Sri Lanka is depending more on other sources to finance this gap and the relationship between savings and investment is weak.
India is the only country where saving-investment gap remained lowest around -2 in the 1980s and the 1990s. It decreased to 1.6 in 2000s. It means that the domestic saving is financing most of the domestic investment and dependence on foreign capital is relatively low in the case of

The Saving-Investment Approach
Feldstein and Horoika (1980) estimated the following equation for panel of OECD countries: e GGDI is ratio of gross domestic investment to GDP and GDS is the ratio of gross domestic saving to GDP. The null hypothesis of perfect capital immobility is failed to reject if β is not significantly different from one and rejected if β is not different from zero.
The equation (1)  In this study, an effort has been made to remove some of the econometric issues raised in literature by incorporating remedial measures to make this approach more applicable for my sample.

Criticism to F-H Approach and Possible Remedial Measures
Dooley (1987), Bayoumi (1990), Feldstein (1983), and Feldstein and Horoika (1980) explained the problem that saving and investment both are strongly procyclical in nature even when they take the form of ratio to GDP. If both rise due to an exogenous shock, the correlation cannot be attributed to low capital mobility.
That's why I have used growth rate of GDP as an explanatory variable. It can take care of possible specification bias due to single variable equation. 5 But its inclusion may reduce the correlation of savings coefficient.
The other common issue is Endogeniety problem. It is said that the government reacts to a trade deficit induced by an increase in investment by slashing down government expenditure or raising 5 For details see Summer (1985) and Dooley et al. (1987)   taxes. In this scenario, saving and investment will be correlated for the reasons other than capital mobility. This is how government policy creates endogeniety. Dooley et al. (1987) and Bayoumi (1990) in order to dismantle endogeniety problem used instrumental variables which affect saving but irrelevant for investment. I intend to use savings lag as an instrument and report the results.
Given the above evidence, the following equation will be estimated after the inclusion of growth Whereas GG is the growth rate of panel countries; The other variable, which is used and suggested by Kim (1993), is openness which is proxied by Imports to GDP ratio. After the inclusion of openness, the regression equation will be of the following form.
The selected South Asian countries have introduced liberalization policies and opened their economies in 1990s. First, the fundamental F-H equation shall be estimated. After that time country dummies for intercept and slope will be introduced and incorporated in the regression separately. Another dummy for post liberalization period has been used which is one after 1993 and 0 before 1993.
All the variables are in shape of panel data, pooling cross section and time series of 5 countries.
GGDI= Ratio of gross domestic investment to GDP; GDS= Ratio of gross domestic Savings to GDP; 93 2 D  is slope dummy interacting with GDS, its value is 1 after 1992 and 0 otherwise.

Real Interest rates Differentials Hypothesis (RIPH):
The RIPH states that if the agents make their forecasts using rational expectations and arbitrage forces are free to act in the goods and asset markets, the real interest rates among countries will equalize. However, the empirical literature does not offer a conclusive answer regarding the existence of real interest rate differentials (RIDs). Ferreira et.al (2007)  Theoretically if agents make their forecasts rationally and arbitrage forces in goods and assets markets are working, real interest parity holds. 6 The arbitrage forces are formalized by uncovered interest parity (UIRP) and relative purchasing power parity (PPP) conditions mentioned in the following equations: If PPP holds, one can substitute Equation 9 in to 10 and after manipulation, get the following resultant Equation; The rid it may follow the following stochastic process: The purpose is to check the stationarity of RID series by applying Panel Unit Root tests. If the RIDs series is estimated to be stationary that implies real interest rates differentials are converging, hence financial markets are integrated. This is the first study to estimate RIDs with panel unit root techniques for South Asian countries which has higher power of the test.
The LLC test assumes that the persistence parameters are same across cross sections. It means that ψ i =ψ for all i. Alternatively, IPS allows ψ to vary across all cross sections.
The LLC model allow for fixed effects and unit specific time trend along with common time effects. The structure of their model is the following: The unit specific fixed effect is important to capture heterogeneity since the coefficient of lagged

Data Source
My main data source for this study is IMF's International Finance Statistics (

The variable of Gross Domestic Savings and Investments are divided with Gross Domestic
Product. The data pertaining to imports of goods and services is also divided with GDP. In case of paneling cross section the data of all the countries is taken in million of US dollars. The data is taken from the same source for consistency.
The interest rates and Consumer Price Index (CPI) data has been taken from IMF (2009)

FH Model and its Extension:
The main results of FH (1980) and the extended model are presented in Table 3. Note: In most of these regressions we have used fixed effect model.   The GG variable becomes significant with a positive sign. The IMP coefficient increased but its t value decreased but still very high. When the same model is estimated in ordinary form, the savings coefficient estimated to be 0.6. The size of GG coefficient and its t value increased.
The size of IMP coefficient and its t value decreased but still highly significant. The savings coefficient ranges in between 0.8 to 0.4 depending on the model specification. In the original FH model, the savings coefficient is estimated to be close to 0.8 but with the addition of GG and IMP, the size significantly reduced. The savings coefficient remains high, 0.8 even when Panel -2SLS has applied in single variable regression. But in the presence of GG and IMP the savings coefficient decreases to 0.4 with Panel 2SLS model. Higher growth rate and openness are estimated to be positively related to GDI.
In order to capture the effect of liberalization and openness policies, i incorporate dummy variable which is 1 after 1993 and 0 otherwise. According to Bekaert, Harvey and Lundblad (2000) most of the South Asian countries started liberalization in early 1990s. 89 . The results are shown in row 10 of Table 1. I notice that the dummy variable is not significant but all the other variables are significant. The coefficient of GDS is 0.6 which implies the fact that almost 60 percent of GDI has been estimated to be financed by GDS. The overall results remained almost same when I estimated the model with Panel 2SLS (Model 11). Note: *, **, *** denote significance at 1, 5 and 10 percent respectively.
It is also important to find change in the slope of GDS in the post liberalization period. I have incorporated a slope dummy in model 12 (row 12). The model shows the sign of slope dummy is negative and it's significant which may imply that the relationship between domestic savings and investment weakened in the post liberalization period. It further provides evidence in favor of increased integration after 1990s. The overall intercept and intercept dummy after 1993 both are insignificant.
The R 2 is estimated at 0.91 which is very high. The overall results support the moderate degree of integration which increased in the post liberalization as shown by Model 12. But according to model 11 no significant structural shift has been observed after 1993 period.

Cross section Dummies and FH model:
It is important to estimate the cross section intercept and slope dummies for this model. It contains important information about difference in cross section behavior. The results are reported in  *, **, *** denote significance at 1, 5 and 10 percent respectively.
The dummy variable for the post liberalization period implies that over the time South Asia integrated with the world although the degree of integration may vary across countries.

Conclusion:
This study estimates the degree of financial integration in panel of 5 South Asian countries by applying 2 Econometric techniques i.e. Savings Investment relation and Real interest rate differential condition. The overall GDS coefficient is estimated to be in between 0.8 and 0.4 which may imply some degree of integration. The post liberalization dummy has shown a reduction in the size of GDS which can be interpreted as increased integration with the initiation of liberalization process in South Asia in the 1990s. The degree of integration may vary across countries. The real interest differential model when applied provides evidence in favor of high degree of financial integration in the overall South Asia. The result stands consistent with all the panel unit root methods Hadri, IPS and LLC.
The interesting aspect of my study is to find evidence of financial integration with saving investment technique which is considered to be a method to estimating low integration and capital mobility. I found the case of some integration after adding Growth rate of GDP and Imports to GDP ratio as explanatory variables in to FH model. Furthermore, I also estimated the model with Panel 2SLS using the lag of GDS as an instrument and the result remained consistent. Since the saving Investment relationship technique requires real interest parity to hold suggested by Frankel (1992), my estimates strongly support that the real interest parity hold for Panel of major South Asian countries with 13 major economies of the world. The empirical evidence with real interest rate parity provides stronger evidence of integration as compared to savings investment technique which provides moderate evidence. Hence Feldstein Horoika savings investment model remains a puzzle for South Asian countries.