**4. Empirical results**

(CESEE) economic governances, in terms of the pace of convergence, share standard features such as, among other things, a sharp improvement in institutional efficiency and human capital, more outward-oriented economic policies, favorable demographic-economic developments, and the quick reallocation of labor from agriculture into other sectors [38]. Forward-looking, speeding-up, and sustaining convergence in the WB 6 requires in-depth efforts to improve institutional quality and innovation, reinvigorate foreign investment, and address the adverse impact of

*Linear and Non-Linear Financial Econometrics - Theory and Practice*

Stabilizing policies and implementation of reforms are the vital drivers of WB 6 growth, in the meantime declining the impact of initial conditions of the 1990s. Since the government efficiency indicator reflects perceptions of the quality of public services, the quality of the civil service and the degree of its independence from political pressures, the quality of policy formulation and implementation, and the credibility of the government's commitment to such policies, we give special attention to this indicator in explaining the growth differences among the WB 6

Even though government efficiency has been studied to some extent, we reveal a significantly wider knowledge gap. First, conceptual specification, based on which empirical examinations of government efficiency is analyzed, is not prevailing combining theory and empirical analysis. Secondly, we identify six structural VAR models. To our knowledge, it has not been applied to WB 6 data. VARs turn out to be one of the key empirical tools in modern macroeconomics, and they allow one to

The range of the data is from January 2006 to December 2018. In order to control for time trends in our analysis, we include dummy variables. The expression

Here, we present parameter estimates and the main characteristics of the

where *yt* is the gross domestic product for each of the WB 6, *GovEfft* is the government efficiency indicator, and *π<sup>t</sup>* represents inflation (a proxy for macroeconomic stability). This specification contains independently identically distrib-

observe how economic governance shocks and macroeconomic stability impact GDP growth and vice versa*.* Of particular interest for this paper is to examine the role of economic integrations and macroeconomic stabilization in determining the growth of GDP in Albania, Bosnia and Hercegovina, Kosovo, Montenegro, North Macedonia, and Serbia. Thus, government efficiency and inflation are considered as important explanatory factors. For North Macedonia Model, we added purposely the corruption indicator variable, in order to observe the potential shocks to growth.

As we will see, the indicator to this specific case shows no impact.

of the main objectives of our VAR model is forecasting, and it has common

*u*

How well the models describe the dynamic behavior of economic variables? We will proceed with our VAR models for structural inference and policy analysis. One

*Yt* ¼ *a*<sup>0</sup> þ *βXt* þ *ut* (1)

*:* The above model will allow us to

*yt* ¼ *at* þ *β*1*GovEfft* þ *β*2*π<sup>t</sup>* þ *ut* (2)

(through GDP) and macroeconomic stability (through inflation).

model macroeconomic data informatively [39].

referring to an SVAR model is used as follows:

uted stochastic disturbance term *ut IID* 0; *σ*<sup>2</sup>

models. The identified recursive SVAR model is as follows:

population aging seriously.

**3. Methodology**

**222**

All variables are stationary based on unit root tests of ADF, PP, and KPSS stationary test. Visual inspection and statistical correlograms portray and confirm stationarity. Test results of *t-*statistics and *p* values reject the null hypothesis of a unit root.

We proceed with empirical construction and testing for potential structural breaks, which are crucial to identify for forecasting purposes as well as confidence bounds. Stability diagnostics, under recursive estimates, show that all coefficients have a lot of instability, indicating structural breaks. Chow breakpoint test confirms the above indication, having *F*-statistics and *p* values smaller than 5%, meaning to reject the null hypothesis of no breakpoints at 5% significance level. The Quandt-Andrews test indicates for rejecting the null hypothesis of no break. It reveals breaks for all cases Albania, Bosnia and Hercegovina, Kosovo, Montenegro, North Macedonia, and Serbia. Testing for multiple breaks in intercept and coefficients using Bai-Perron to sequentially test the hypothesis of L + 1 vs. L sequentially determined breaks. The Bai-Perron test recommends that there are breaks. We can conclude that all four tests indicate that there is a switch of parameters at 5% significance level, and we are dealing with multiple breaks in parameters. We will add the following dichotomous variables (**Table 1**):

### *Linear and Non-Linear Financial Econometrics - Theory and Practice*


#### **Table 1.**

*Dichotomous variables.*


*VAR satisfies the stability condition. Source: Authors' calculation. \**

*No root lies outside the unit circle in Table 2.*

#### **Table 2.**

*Root of characteristic polynomial.*

Including 12 lags in the lag exclusion test or lag length criteria about deciding the maximum number of lags to be used in our VARs, we get an estimated fitting lag length denoted by an asterisk. We select 2, 11, 3, 2, 10, and 2 lags, respectively, as the appropriate lag length for our VAR models.

introduced, which is called the baseline. Deterministic simulation ignores the fact that relationships do not hold exactly, because of random disturbances and estimated coefficients, which are not known or predetermined values. We should account for these sources of uncertainty by using stochastic simulations. **Figure 1** performs gdp\_gap and inflation stochastic simulations for static solution model

**Lags AL B&H KS MNE NM SRB**

*Governance and Growth in the Western Balkans: A SVAR Approach*

*DOI: http://dx.doi.org/10.5772/intechopen.91731*

 **0.2790 0.2574 0.3468 0.9348 0.1017 0.6712 0.4415 0.1356 0.6129 0.9167 0.0639 0.9669** 0.0103 **0.7364 0.8614** 0.7818 **0.0545** 0.4213 0.1171 **0.1947** 0.8915 0.1140 **0.2147** 0.1376 0.4000 **0.2323** 0.9923 0.0169 **0.2487** 0.3582 0.9380 **0.1057** 0.0000 0.9868 **0.6963** 0.0611 0.6340 **0.4410** 0.5016 0.0019 **0.9956** 0.4222 0.8501 **0.1188** 0.0757 0.6773 **0.3241** 0.3912 0.3897 **0.4393** 0.2606 0.9880 **0.2077** 0.4678 0.9490 **0.5177** 0.3401 0.7466 **0.7631** 0.9671 0.8103 **0.3337** 0.9346 0.3383 0.5245 0.2678 0.0000 0.0000 0.0000 0.0000 0.0001 0.0000

**Prob Prob Prob Prob Prob Prob**

Forecasting performance of the static solution performs very well, both in terms of the fit and small standard error bounds, coming as a result of de facto one period

It uses actual instead of forecasted lagged values over the forecast period. The blue lines portray the actual data for both gdp\_gap and inflation, while the green lines represent the forecasting performance of the stochastic-static model. As seen in **Figure 1**, both predictions are very close to the real data and within the confidence bands, except in the 5th month for the North Macedonia GDP. The red lines show the upper and lower bounds of the stochastic-static solution model simulator. The comovement is noticeable for both variables. Including bootstrapped errors and

Analyzing **Table 4**, the first thing we notice is low RMSE for all WB 6 countries. The RMSEs for gdp\_gap and inflation are 0.1979 and 0.3768, respectively. Theil's coefficient U1, which measures the forecast accuracy, is acceptable for all variables

The impulse response function will tell us the change in endogenous variables for each structural shock at t, t + 1, and so on. Our goal is to trace out the effects of internal shocks to the WB 6 economies. First, we employ Sims' (1980) orthogonalized impulse response functions [42]. We will trace out the responses of the depen-

simulators for period 2017m01 till 2017m12.

dent variables in the SVAR models to shocks.

coefficient uncertainty, we get forecast measures (**Table 4**).

ahead forecast.

*Source: Authors' calculation. Bold values represent lag values.*

*VAR residual serial correlation LM tests.*

**Table 3.**

of the VAR models.

**225**

**4.2 Impulse responses**

All inverse roots of the characteristic polynomial are <1, as seen in **Table 2**, confirming the stationarity of the VARs.

We have reached significant results, and based on the stationarity assessed so far, we can infer that impulse response standard errors are valid (**Table 3**). The largest inverse root of the AR characteristic polynomial is 0.987044. The correlograms of short-term error correlations of the estimated VARs suggest no autocorrelation. The entire lines lie within the 2 standard error bounds, showing at first lags another backup to the suggestion of missing autocorrelation in nonnoticeable continual wave sinusoidal. Based on 95% significance level, the null hypothesis, which states there is no autocorrelation of residuals in our estimated VARs, cannot be rejected. It has *p* values of 44.15, 33.37, 86.14, 91.67, 76.31, and 96.69%, respectively, for lag orders up to 2, 11, 3, 2, 10, and 2 lags. Therefore, there is no indication, based on the LM tests, that there is the autocorrelation of errors.

#### **4.1 Forecasting models**

To generate a forecast, we can use known values or forecasted values. Using the known values for forecasting is static forecasting. In case we proceed using the predicted values from regression, then it is dynamic forecasting. There are two types of simulation processes. One is a deterministic simulation, where we get only one value for the solution, which does not respond to innovations. It calculates under the current set of assumptions or known facts without any shocks

**Lags AL B&H KS MNE NM SRB Prob Prob Prob Prob Prob Prob 0.2790 0.2574 0.3468 0.9348 0.1017 0.6712 0.4415 0.1356 0.6129 0.9167 0.0639 0.9669** 0.0103 **0.7364 0.8614** 0.7818 **0.0545** 0.4213 0.1171 **0.1947** 0.8915 0.1140 **0.2147** 0.1376 0.4000 **0.2323** 0.9923 0.0169 **0.2487** 0.3582 0.9380 **0.1057** 0.0000 0.9868 **0.6963** 0.0611 0.6340 **0.4410** 0.5016 0.0019 **0.9956** 0.4222 0.8501 **0.1188** 0.0757 0.6773 **0.3241** 0.3912 0.3897 **0.4393** 0.2606 0.9880 **0.2077** 0.4678 0.9490 **0.5177** 0.3401 0.7466 **0.7631** 0.9671 0.8103 **0.3337** 0.9346 0.3383 0.5245 0.2678 0.0000 0.0000 0.0000 0.0000 0.0001 0.0000

#### *Governance and Growth in the Western Balkans: A SVAR Approach DOI: http://dx.doi.org/10.5772/intechopen.91731*

*Source: Authors' calculation.*

*Bold values represent lag values.*
