Desktop version

Home arrow Business & Finance arrow Financial Liberalisation: Past, Present and Future

Empirical Methodology and Data

This section develops an empirical methodology to test the main implications of our model, namely that the reaction of the current account to net output shocks is larger and more persistent in financially more liberalized countries.

General Methodology

In practice, all the variables required to test the model described in previous sections are likely to be endogenous with respect to each other. Furthermore, taken together with the driving process for net output shocks, the data-generating process for the current account implied by equation (8.9) will be a reduced form VAR model with cat and A ln NOt as the regressors. Finally, the contemporaneous correlation between cat and A ln NOt is positive (negative) in case of a log-level (difference) shock. If both shocks are equally likely, fitting a reduced-form single-equation model may therefore lead to a false rejection of a relationship between these variables. To test the proposed hypothesis that the reaction of the current account to net output shocks is larger and more persistent in financially more liberalized countries rigorously, it is therefore necessary to formally identify both of the shocks. For all of these reasons, we adopt a panel VAR approach where we explicitly identify net output shocks while addressing endogeneity concerns. An alternative approach would, of course, be the instrumental variable approach, but for annual data Chinn and Prasad (2003) have shown that it is difficult to obtain reliable instruments for the variables driving the current account. We use annual data given that the index of financial repression we use, as well as the corresponding macroeconomic data for many countries, is not available at a quarterly frequency. The panel VAR includes both of the variables that drive the data-generating process for the current account according to our model: the current account to net output ratio and the net output growth rate. The proposed panel VAR model is:

where Yc, t is an 2 x 1 vector consisting of the current account to net output ratio and the per-capita net output growth rate in country c at time t. Since the theory provides predictions for variables expressed in deviations from their steady-state values, we express these two variables in deviations from their country-specific means (which are our proxies for their steady state values). This removes the country fixed effect for these variables. Ft is an 2 x 1 vector consisting of unobserved common factors, a separate one for each equation, and Dc is a diagonal matrix. The strategy of introducing unobserved common factors as additional explanatory variables allows us to control for global factors, such as world oil price and interest rate shocks, but also cross-sectional dependence (spillovers across countries). As in previous work on interacted panel VARs, we estimate the impact matrix of the VAR, A0 , c , t, as oppose to obtaining it from the variance matrix of the shocks post estimation. Ak , c , t is the matrix of country- specific autoregressive coefficients. Uc , t is an 2 x 1 vector of residuals, assumed to be uncorrelated across equations and normally distributed with a covariance matrix Z.

Since we want to understand to which extent the impact of a net output shock is amplified or dampened by economic conditions in a given country, we allow the impact matrix coefficient al,ct to vary with the country’s economic structure. Similarly, in the case of external habits, our theory predicts that the autoregressive coefficients ajlc t, where j is the row and l the column, also depend on the country’s economic structure. In contrast to work that models time variation in the VAR coefficients as a random walk (Cogley and Sargent 2005 and Primiceri 2005), we therefore adopt the approach presented in Towbin and Weber (2013) and allow the coefficients to vary with observable economic structure variables. The time-varying coefficients, ajct in A0 , c , t and Ak, c , t are therefore given by j = j + PH,2 * FRc t + ? Ц3 * KAc t + PHA * FXRct + j * TRcJ,

where FRc , t is the proposed index of financial liberalization at time t in country c. Previous work by Lewis (1997) has documented that restrictions on international transactions may lead consumers to act as if they are liquidity constrained. Similarly, Towbin (2008) has shown that the trade balance, an important component of the current account, becomes more persistent with greater capital account openness. To avoid omitted variable bias, capital account openness in country c at time t, KAC, t, is therefore included as an independent determinant of the VAR coefficients. One would expect that the speed of adjustment of the current account under a fixed and floating exchange rate regime would differ (Friedman 1953), which is why we also include the exchange rate regime, FXRc , t, as a control variable. Finally, an alternative important determinant of external adjustment could be a country’s trade openness, TRc , t, which we include as an additional variable to minimize chances of omitted variable bias. Substituting the definitions of the time-varying coefficients in equation (8.11) leads to the following interacted panel VAR model:

where Zc, t = [1 FRc, t KAc , t FXRc, t TRc , t] and w is a superscript indicating the number of the given column of Zc , t, with W as the total number of variables in Zc, t. Hw is a 2 x 1 vector with the coefficients corresponding to ZWt. In other words, to avoid omitted variable bias, all of the economic structure variables also enter the model in levels.

In summary, we are allowing the coefficients in both the impact and lagged dependent variable matrix to vary with observable deterministic variables. The advantage of this approach is that we can assess to which extent the impulse responses to a given shock differ with the degree of financial repression (liquidity constraints). Since our theory predicts that the effect of a net output shock on the current account to net output ratio depends on the degree of liquidity constraints, this method is better suited to test our theory than standard time-varying coefficient VARs, which typically do not provide information on the source of time variation.

 
Source
< Prev   CONTENTS   Source   Next >

Related topics