BUS613: Advanced International Business

The frequently used methods for dynamic heterogeneous panels are the mean group (MG) estimator and fixed effect and random effect estimator. MG estimator, on one hand, estimates equation for each group separately and inspect the mean of the estimate which, according to Pesaran and Smith, are consistent estimates of the average of parameters. However, MG estimator is not capable of considering similarities of certain parameters across groups. On the other hand, fixed and random effect estimator allows intercept to vary group wise while all other coefficients and variances in error are restricted to be the same.

This study considers Pooled Mean Group (PMG) estimator because of the advantage that it takes into account both pooling and averaging. The intercept, short-run coefficients and variances in error, in (PMG) estimator, varies across groups while the long-run coefficients are restricted to be the same. Their reason behind similar relationships between variables in the long run across groups is that common arbitrage conditions, technologies and other common factors influence all groups in the same pattern. Besides, it seems less compelling to assume short-run variation and variances to be the same across groups.

a. Model specification

The latest work on Panel data analysis involving time span (T) and number of cross section (N) is presented under two headings, i.e., (Pooled Mean Group (PMG) and Mean Group (MG) panel ARDL models. In PMG, cross sections are pooled and intercept terms are permitted to vary across cross sections while in MG, the model may be built individually for each cross section with possible difference in intercepts, slope coefficients, and error variances. PMG and MG permits short-run parameters, intercepts terms and error variance to vary across groups, however, the two approaches differ in the long run. Contrary to MG, PMG restrains the long-run coefficients to be homogenous. The homogeneity of the long-run slope coefficient is useful when there are reasons to expect the long-run equilibrium relationship between the variables are similar across countries. MG model imposes no restrictions on coefficient, both in the long as well as in the short run; however, the necessary condition for the validity of MG approach is to have a sufficiently large time-series dimension of the data. Pesaran consider that the MG approach is quite sensitive to outlier and small model permutation. Keeping in view the small number of countries (N) and sufficiently large time-series data (T), in this study we opt for PMG. The Hausman test will confirm that the PMG or MG approach was used in this case. The general form of the empirical specification of the PMG model can be written as:

$Y_{it} = \mathop \sum \limits_{j = i}^{p} \gamma_{ij} Y_{i, t - 1} + \sum \emptyset_{ij} Z_{i, t - j} + \mu_{t} + \varepsilon_{it} ,$                                                        (1)

where number of cross sections i = 1, 2, …. N and time t = 1, 2, 3 …. T. Z_it is a vector of K × 1 regressors, γ_ij is a scalar, μi is a group-specific effect. The disturbance term is an I(0) process if the variables are I(1) and co-integrated then a major characteristic of co-integrated variables is their rejoinder to any deviance from long-run equilibrium. This characteristic infers error correction dynamics of the variables in the system are swayed by the deviance from equilibrium. So it is common to re-parameterize above equation into the error correction equation as:

$\Delta Y_{it} = \theta_{i,} y_{i,t - j} - \beta_{i} Z_{i,t - j} \mathop \sum \limits_{j = i}^{p - 1} \gamma_{ij} \Delta y_{i, t - j} + \sum \emptyset_{ij} \Delta Z_{i, t - j} + \mu_{t} + \varepsilon_{it} .$              (2)

The error correction parameter θ_i indicates the speed of adjustment. If θ_i = 0, then there is no evidence that variables have long-run association. It is expected that θ_i is negative and statistically significant under the prior supposition that variables indicate a convergence to long-run equilibrium in case of any disturbance.

With increase in time period of analysis, dynamic panels; non-stationarity is a very important issue and in present study this issue has been taken into consideration by applying Levin, Lin and Chu (LLC) and Im, Pesaran and Shin (IPS) unit root tests. The condition is: when all the chosen variables in the model are stationary at I(1), I(0) or a mixture of I(0) and I(1). PMG being an ARDL-model is sensitive to the selection of lag length and hence, we utilize the Akaike Information Criteria (AIC) to obtain our optimal lag length. On the basis of the above model, we consider the following hypothesis.

H0

There is no impact of infrastructure on exports in South Asia.

H1

There is a significant impact of infrastructure on exports in South Asia.

H01

There is no impact of infrastructure on the trade deficit in South Asia.

H2

There is a  significant impact of infrastructure on the trade deficit in South Asia.

b. Panel cointegration tests

In order to examine the presence of a long-run convergence among our variables of interest, we carry out a panel cointegration test. The objective of the panel cointegration test is to combine information on similar long runs across the various panel members. Pedroni suggested seven cointegration tests for panel data on the basis of the cointegrating residuals of εit, three of which are considered to be group mean panel cointegration tests and are based on the between-dimension. They are devised by dividing the numerator by the denominator before adding over the N-dimension. The other four, referred to as panel cointegration tests, are based on the within-dimension and are formulated by adding both the numerator and the denominator over the N dimension. Moreover, the Kao test is also being used for cointegration between dependent and independent variables, on the foundation of Eq. (1) with the test for the null hypothesis of no cointegration being considered.

H0

There is no cointegration between infrastructure and exports in South Asia.

H1

There is significant cointegration between infrastructure and exports in South Asia.

H01

There is no cointegration between infrastructure and trade deficit in South Asia.

H2

There is significant cointegration between infrastructure and trade deficit in South Asia.