3. Methodology and empirical results

When investigating the exchange rate uncertainty on the domestic investment under the panel data model expressed as in Eq. (1), the panel data analysis is carried out by following the steps in Aktas et al. [19]. The panel data consist of countries which may involve individual effects of

<sup>3</sup> The annual volatility for each country is derived by multiplying σdaily and ffiffiffi T <sup>p</sup> since the volatility escalates with the square root of time [18].

countries (denoted as δi). Therefore, F-test is implemented so as to determine whether the model is fixed effect model or pooled least square model [20]. The null hypothesis and test result of F-test having degrees of freedom as F (n-1, nT-n-k)<sup>4</sup> are given in Table 2. F-test statistic is statistically significant at 1% significance level, which indicates the model can be fixed effect model.

The model can also include random effect. In order to test whether the model involves random individual effects, Breusch and Pagan (1980) Lagrange Multiplier (LM) test having Chi-square distribution with a degree of freedom of 1 is employed [20]. The null hypothesis and test statistics of the Breusch Pagan LM are given in Table 3. The test result, statistically significant at 1% significance level, points out that the model can include random individual effects.

Since the model could involve either fixed effect or random effect, a well-known test Hausman (1978) is conducted. The Hausman test, having a null hypothesis of no correlation between unobservable individual effects and regressors (i.e., Random effect model), has a chi-square distribution with degrees of freedom of k [21]. The null and alternative hypotheses and test statistics of Hausman specification test are suggested in Table 4. The Hausman test indicates that the model is a fixed effect model, since the test statistic is significant at 5% significance level.

The fixed effect model is found to be appropriate to estimate the parameters in the main model. After constructing fixed effect model, the Wald test for groupwise heteroskedasticity is implemented in order to detect heteroskedasticity of the residual of fixed effect model [22]. The test has a chi-square distribution with a degree of freedom of n. The chi-square test statistics (25) is found to be 1833.61 with a prob. value of 0.000, which indicates the existence of groupwise heteroskedasticity in the residuals of the fixed effect model. It is also necessary to check the serial correlation in the panel data model, since serial correlation may offer biased


Table 2. Null hypothesis and test result of F-test.

On the other hand, the exchange rate volatilities of Colombia, India, Indonesia, Mexico, Mongolia, Poland, Serbia, Seychelles, and Turkey are most properly modeled with GJR-GARCH(1, 1) model as offered in Table A4 and Table A5. The ω, α , and β coefficients are found to be statistically significant. The asymmetry coefficient (γ) is found positive and statistically significant for Colombia, India, Indonesia, Mexico, Mongolia, Poland, Serbia, and Turkey, which points out leverage effect and is a sign that negative shocks on the exchange rate returns have more impact on the volatility when compared to the positive shock. On the other hand, for only Seychelles, the asymmetry coefficient (γ) is found negative and statistically significant, which suggests that positive news has more impact on volatility than the negative shocks. The

relation between the residuals for these GJR-GARCH(1, 1) models. Additionally, no ARCH

As a summary, the exchange rate uncertainties of the countries, which are estimated by selecting

Country Exchange rate uncertainty model Country Exchange rate uncertainty model

When investigating the exchange rate uncertainty on the domestic investment under the panel data model expressed as in Eq. (1), the panel data analysis is carried out by following the steps in Aktas et al. [19]. The panel data consist of countries which may involve individual effects of

T

<sup>p</sup> since the volatility escalates with the

effect exists in the residuals of GJR-GARCH(1, 1) model of each country.

Brazil EGARCH(1, 1) Paraguay EGARCH(1, 1) Chile GARCH(1, 1) Peru EGARCH(1, 1) Colombia GJR-GARCH(1, 1) Philippines GARCH(1, 1) Georgia GARCH(1, 1) Poland GJR-GARCH(1, 1) Hungary EGARCH(1, 1) Serbia GJR-GARCH(1, 1) India GJR-GARCH(1, 1) Seychelles GJR-GARCH(1, 1) Indonesia GJR-GARCH(1, 1) South Africa EGARCH(1, 1) Kenya GARCH(1, 1) Tanzania EGARCH(1, 1) Madagascar EGARCH(1, 1) Thailand GARCH(1, 1) Mexico GJR-GARCH(1, 1) Turkey GJR-GARCH(1, 1) Moldova EGARCH(1, 1) Uganda GARCH(1, 1) Mongolia GJR-GARCH(1, 1) Uruguay GARCH(1, 1)

the most appropriate volatility models, are offered in Table 1<sup>3</sup>

3. Methodology and empirical results

The annual volatility for each country is derived by multiplying σdaily and ffiffiffi

Table 1. Countries and their exchange rate uncertainty models.

Papua New Guinea EGARCH(1, 1)

3

square root of time [18].

) of squared standardized residuals imply that no autocor-

:

acquired Ljung-Box-Q statistics (Q<sup>2</sup>

226 Financial Management from an Emerging Market Perspective


Table 3. Null hypothesis and test result of Breusch Pagan LM test.


Table 4. Null and alternative hypotheses and test result of Hausman test.

<sup>4</sup> n, T and k are number of groups (countries), number of years and number of regressors in the model, respectively.

standard errors, hence indicating less efficient parameter estimations. Thus, the serial correlation test developed by Wooldridge (2002) is utilized under the null hypothesis of no serial correlation [23]. The Wooldridge test for autocorrelation in panel data has a test statistic of F (1, 24) that equals to 35.434 with a prob. value of 0.000, which is found to be statistically significant at 1% significance level, thereby denoting existence of autocorrelation in the panel model.

Due to the existence of heteroskedasticity and autocorrelation problems in the fixed effect panel model, the acquired fixed effect model results may offer biased results. Therefore, the feasible generalized least square (GLS), which allows the estimations of panel data model under heteroskedasticity across panels and autocorrelation presence, is employed so as to conclude the results of the model [21, 24].5 The feasible GSL estimators are obtained as in Eq. (7).

$$
\widehat{\beta\_{FGLS}} = \left(X^{\prime}\widehat{\boldsymbol{\Omega}}^{-1}X^{-1}\right)^{-1}X^{\prime}\widehat{\boldsymbol{\Omega}}^{-1}\boldsymbol{y} \tag{7}
$$

where Ω = ∑<sup>n</sup> � <sup>n</sup> ⦻ I, which is the error variance matrix and obtained as in Eq. (8).

$$
\widehat{\sum}\_{i,j} = \frac{\widehat{\epsilon}\_i^{\prime\prime} \widehat{\epsilon}\_j}{T} \tag{8}
$$

The estimated test results from the Feasible GLS for both two models are suggested in Table 5.

As observed in the estimation results of model 1, the impact of economic growth on the domestic investment is positive and significant at 1% significance level. This result is anticipated, since growing economy such as emerging markets and developing economies may offer valuable prospects for private investors to obtain profitable returns, when they invest in these countries. Similarly, the studies of Bahmani-Oskooee and Hajile [12] and Safradi and


Notes: Robust standard errors are given in square parentheses.

\*,\*\*,\*\*\* denote the significance level at 1%, 5% and 10% respectively.

Table 5. The feasible GLS estimation results.

<sup>5</sup> See also http://www.stata.com/manuals13/xtxtgls.pdf.

Soleymani [11] also prove positive association between GDP and domestic investment. As for real interest rate, the impact of real interest rate on the domestic investment is found to be negative; however, this impact is statistically insignificant. When considering the real interest rate and investment linkage, it is inevitable to observe that increases in real interest rates lead to declines in domestic investment due to the increasing cost of capital stock. Finally, it is observed that an increase in the exchange rate uncertainty leads to an increase in domestic investment in these EMDEs. The result is found to be statistically significant at 1% significance level. In general, it is expected that heightened uncertainty in exchange rates may constrain the investors from involving in domestic investments, if the investors hold the position of "wait and see." But, if the investors are risk-neutral or risk appetent, they may perceive the volatile environments in terms of exchange rates as lucrative opportunities to engage in investments. Likewise, Bahmani-Oskooee and Hajile [12] find the impact of exchange rate uncertainty on the domestic investment as positive for Colombia, Italy, Singapore, Sweden, and US in the long run. For the positive linkage, they suggest that some investors may tend to invest more in order not to be exposed to the future price volatility arising from exchange rate uncertainty. When considering model 2, the effect of exchange rate uncertainty, economic growth, and real interest rate on domestic investment is found similar to the results of model 1. The impact of GFC on domestic investment of these countries is negative and statistically significant at 10% level.
