Assuming that rural tourism is an element that determines growth and that, in turn, economic growth contributes to the development of tourism, between the two elements there is a positive feedback loop; the objectives of the research were, on the one hand, the analysis of tourist traffic (as evidenced by the number of arrivals and overnight stays in agrotouristic boarding houses) and, on the other hand, to highlight how the fluctuations of the GDP (Gross Domestic Product) affect tourist traffic developments, with implications in the sustainable development of rural areas.
The choice of methodology for the analysis of this interdependence between rural tourism quantification indicators and economic growth through the panel-type econometric model was based on the latter’s capacity for flexibility in modeling the differences registered in the individual components of the influence factors, as well as on the degree of trust in the results obtained, which leads to an increase in the efficiency and consistency of econometric estimates.
The research methodology has been contoured in such a way as to lead to a better description of results that would reflect the achievement of the objective set.
Similarities and Regional Particularities in the Development of Rural Tourism
The modeling of the process that manifests between the variable of rural tourism and the one that reflects economic growth (GDP) has in view the establishment of a causality relation between regional structures between 2000 and 2012; it starts with an analysis of the statistic features of the data series, to the aim of applying the Granger Test [
27,
28].
The testing of the stationarity of the data series “overnight stays” (OS) and GDP (Null Hypothesis: Unit root), for the chosen significance threshold (
Table 1), revealed the non-stationarity of both the OS (Probabilities for Fisher test = 0.9994 > 0.05) and GDP (Probabilities for Fisher test = 0.1710 > 0.05) series, which required transforming them through differentiation.
The results for the differentiated series d(OS) and d(GDP) are shown in
Table 2.
Table 1.
Stationarity test results for overnight stays (OS) and economic growth (GDP) series.
Table 1.
Stationarity test results for overnight stays (OS) and economic growth (GDP) series.
Series: OS | | | Series: GDP | | |
---|
Method | Statistic | Probability ** | Method | Statistic | Probability ** |
PP—Fisher Chi-square | 3.65317 | 0.9994 | PP—Fisher Chi-square | 21.1995 | 0.1710 |
PP—Choi Z-stat | 5.94868 | 1.0000 | PP—Choi Z-stat | −1.76443 | 0.0388 |
Table 2.
Test results for the existence of a unit root in the d(GDP) and d(OS) differentiated series.
Table 2.
Test results for the existence of a unit root in the d(GDP) and d(OS) differentiated series.
Series: d(OS) | | | Series: d(GDP) | | |
---|
Method | Statistic | Probability ** | Method | Statistic | Probability ** |
PP—Fisher Chi-square | 35.3455 | 0.0036 | PP—Fisher Chi-square | 112.384 | 0.0000 |
PP—Choi Z-stat | −3.02592 | 0.0012 | PP—Choi Z-stat | −8.83166 | 0.0000 |
The results indicate the fact that the error associated with the rejection of the hypothesis according to which the series d(OS) has a unit root is 0.36%, lower than the 5% standard threshold (in fact lower than that of 1%). In this context, we accept the hypothesis that the d(OS) series is stationary. Also, testing the existence of a unit root in the d(GDP) differentiated series leads to an error associated with the rejection of the hypothesis according to which the d(GDP) series has a unit root (it does not have the first significant decimal digits). We accept the hypothesis that the d(GDP) series is stationary.
Since the GDP and OS series, calculated in regional structures between 2000 and 2012, are not stationary, the potential causality relation between the two series cannot be determined directly through the Granger test. Hence, we use the Toda-Yamamoto version of the respective test.
The applicability of the test with the Toda-Yamamoto version requires the observation of the following procedure, which presupposes these stages:
Determine the order of a VAR (p)-type process between the series in the level;
Estimate the VAR model of a p + q type, where d is the maximum order of integration for the OS and GDP series. As we have shown above, both series are I(1), so d = 1;
Calculate the Toda-Yamamoto (Wald type) statistics and compare them with the theoretical value of the distribution χ2 with (p + d − 1) degrees of freedom.
The first results are presented in
Table 3.
With the exception of the Akaike information criterion, which indicates a lag 8 order, all the other criteria suggest an order of the VAR process q = 7. Therefore, taking into account that the stationarity of the d(OS) and d(GDP) series was obtained by a 1-order differentiation of the OS and GDP series, it is clear that the chosen process will be VAR (7 + 1).
Table 3.
Results regarding the establishment of the order in the VAR process.
Table 3.
Results regarding the establishment of the order in the VAR process.
Lag | LogL | LR | FPE | AIC | SC | HQ |
---|
6 | −709.3583 | 44.47138 | 6.68 × 1014 | 39.74910 | 40.88110 | 40.14818 |
7 | −699.0731 | 12.23103 * | 4.93 × 1014 * | 39.40936 | 40.71551 * | 39.86984 * |
8 | −694.3501 | 5.105994 | 4.99 × 1014 | 39.37028 * | 40.85058 | 39.89215 |
Finally, we test the existence of the influence of the GDP on OS (Null hypothesis: GDP influence on OS is insignificant) using Toda-Yamamoto statistics, the Wald-type (
Table 4), which is compared with the theoretical χ
2 distribution with 7 degrees of freedom.
Table 4.
Results of the Granger causality tests s in Toda-Yamamoto.
Table 4.
Results of the Granger causality tests s in Toda-Yamamoto.
Excluded | Chi-sq | df (Degree of freedom ) | Probability |
---|
GDP | 23.22066 | 7 | 0.0016 |
All | 23.22066 | 7 | 0.0016 |
The probability estimated for the hypothesis according to which the GDP does not influence OS is 0.16%, inferior to the 5% standard threshold (and to that of 1%), which means that the respective hypothesis cannot be accepted at a reasonable threshold. Consequently, we accept the alternative hypothesis, according to which GDP significantly influences the OS variable (GDP is the cause, in a Granger-Toda-Yamamoto sense, for OS). In the established context, we continue with the methodology for the elaboration of an econometric model, starting from this result, through the test of two hypotheses:
H1: GDP is an explanatory variable for OS;
H2: On short term, the relation between the GDP and OS variables is distorted by circumstantial factors, but there is a stable relationship between them long-term (econometrically, the GDP and OVERNIGHTS variables are co-integrated).
To test the H1 hypothesis, a VAR model was built between the stationary variables d(OS) and d(GDP). Due to the small number of observations over time (T = 13), the order of the VAR process could not be greatly increased. The applied information criteria led, however, to a 9-order VAR process, which suggests the existence of long-term links between the analyzed variables.
Regarding the H
2 hypothesis, the test results are presented in
Table 5.
Table 5.
Results of the application of the test to build a VAR model between the stationary of differentiated series for overnight stays d(OS) and also differentiated series for Gross Domestic Product d(GDP) variables.
Table 5.
Results of the application of the test to build a VAR model between the stationary of differentiated series for overnight stays d(OS) and also differentiated series for Gross Domestic Product d(GDP) variables.
Cointegrating Equation: | CointEq1 | Standard. Error | t-Statistic | | | |
---|
OS(−1) | 1.000000 | | | | | |
GDP(−1) | −3.074897 | 0.31233 | −9.84502 | | | |
Error Correction: | d(OS) | Standard. error | t-Statistic | d(GDP) | Standard. error | t-Statistic |
CointEq1 | −0.027281 | 0.01293 | −2.10969 | 0.173205 | 0.01827 | 9.48280 |
D(OS (−1)) | 0.511978 | 0.12130 | 4.22083 | −0.265240 | 0.17133 | −1.54814 |
D(OS (−2)) | −0.475473 | 0.14447 | −3.29115 | −0.027695 | 0.20406 | −0.13572 |
D(OS (−3)) | 0.206023 | 0.14414 | 1.42930 | −0.138526 | 0.20360 | −0.68040 |
D(GDP(−1)) | −0.043463 | 0.05006 | −0.86816 | −0.236349 | 0.07071 | −3.34240 |
D(GDP(−2)) | −0.096805 | 0.04702 | −2.05885 | −0.192904 | 0.06641 | −2.90465 |
D(GDP(−3)) | −0.042931 | 0.03914 | −1.09695 | −0.170733 | 0.05528 | −3.08860 |
d(OS) | d(GDP) |
R-squared | 0.188150 | R-squared | 0.724480 |
Adjusted R-squared | 0.109584 | Adj. R-squared | 0.697817 |
Sum squared resids | 2.46 × 1010 | Sum sq. resids | 4.91 × 1010 |
F-statistic | 2.394794 | F-statistic | 27.17149 |
Log likelihood | −777.2895 | Log likelihood | −801.1171 |
Akaike AIC | 22.73303 | Akaike AIC | 23.42368 |
Schwarz SC | 22.95968 | Schwarz SC | 23.65033 |
Mean dependent | 9825.725 | Mean dependent | −14544.88 |
The negative value of the coefficient of co-integration (−0.027281), which is also significantly different from zero (
t-Statistic = −2.10969), indicates the presence of a stable long-term relationships between the variables OS and GDP which allows the development of a panel-type analysis. The test is of the type:
where:
a0—is the common effect;
a1—estimates the inertial effect;
a2—measures the influence of the GDP modification on the endogenous variable (OS);
αt—is the specific effect in time (t = 2000, …, 2012);
γi—represents the individual effect (the specificity of every region);
eit—is the idiosyncratic error.
If at least one of the values a
0, a
1 and a
2 is significant, then the panel is homogenous, in the sense that, beyond the elements that are specific to every region, or any moment in time, in the connection between the two variables (OS and GDP) elements that are common to all the regions and years under analysis can be found. The results of model testing are presented in
Table 6.
Table 6.
Test results of the model.
Table 6.
Test results of the model.
Variable | Coefficient | Standard Error | t-Statistic | Probability |
---|
C | 22411.05 | 12366.56 | 1.812230 | 0.0755 |
OS(−1) | 0.819036 | 0.191551 | 4.275814 | 0.0001 |
d(GDP(−1)) | 0.094438 | 0.032669 | 2.890741 | 0.0055 |
Fixed Effects (Cross) | | | | |
_NV--C | 9120.251 | | | |
_CE--C | 37498.26 | | | |
_NE--C | 15831.54 | | | |
_SE--C | −11543.02 | | | |
_SM--C | −7787.306 | | | |
_BI--C | −20115.99 | | | |
_SV--C | −11875.66 | | | |
_VE--C | −11128.09 | | | |
Fixed Effects (Period) | | | | |
2004--C | −10047.28 | | | |
2005--C | 7200.238 | | | |
2006--C | −2974.132 | | | |
2007--C | 3955.534 | | | |
2008--C | 9231.927 | | | |
2009--C | −15320.37 | | | |
2010--C | −16160.52 | | | |
2011--C | 8814.043 | | | |
2012--C | 15300.56 | | | |
| Effects Specification | | |
Cross-section fixed (dummy variables) | |
Period fixed (dummy variables) | |
R-squared | 0.955780 | Mean dependent var | 75105.10 |
Adjusted R-squared | 0.941859 | S.D. dependent var | 73379.15 |
Sum squared resid | 1.69 × 1010 | Akaike info criterion | 22.61211 |
Log likelihood | −796.0359 | Schwarz criterion | 23.18127 |
F-statistic | 68.65650 | Hannan-Quinn criter. | 22.83870 |
Prob(F-statistic) | 0.000000 | Durbin-Watson stat | 1.645208 |
| | | | |
The values of the determination ratio, R-squared = 0.955780, and of the F-statistic = 0.955780, which corresponds to an insignificant probability of error, Probability(F-statistic) = 0.00, lead to the conclusion that the model is valid, and the dependencies between the variables OS and GDP in the eight regions developed in Romania during 2004–2012 are well approximated. Consequently, the model can be used to estimate the evolution of overnight stays by GDP evolutions.
Coefficients
a0(22,411.05),
a1(0.819036) and
a2(0.094438) are significantly different from zero, as shown in
Table 6, and the corresponding variables OS(−1) and d(GDP(−1)) are statistically significant for a probability of error of less than 1% even for a bilateral t test. Regarding the coefficient C (constant term
a0), it is statistically significant for a bilateral t test for a probability of 90%.
However, given that the test used to test the validity of model coefficients is a t-Student unilateral test, where t-statistic = 1.812230 corresponds to a probability of error of 3.77%, it can be said, with a 95% probability, that the panel is homogeneous, which means that in all development regions there are common elements in time and space, which explains the dynamics of the OS variable, over which regional specific effects overlap.
At a general level, considering the coefficients a1 and a2 what results is, on the one hand, that the evolution of changes in the volume of the GDP, expressed in the variable factor d(GDP) directly influences, in the same sense, evolutions of overnight stays in the agrotouristic boarding houses, in a proportion of 9.44% (a2 = 0.094438); on the other hand, it forms a positive feedback loop between overnight stays (OS) in two consecutive years; the value of the dependent variable OS in the current year depends on its value in the previous year, the influence being significant (a1 = 0.819036).
Above this general trend, the individual effects of each region overlap. The data from
Table 6 and
Figure 1 show that, in the North-West, North-East and Central regions, the level of overnight stays in agrotouristic boarding houses exceeds the average, the higher correction corresponding to the Central region (37,498.26), which means that the pressure of this kind of tourism on sustainability in its counties is much higher.
Figure 1.
The effects of the specificity of regions on OS.
Figure 1.
The effects of the specificity of regions on OS.
In the other five development regions, specific effects manifest in reverse. The evolutions of the number of overnight stays in agrotouristic boarding houses are lower compared to the average. A significant decrease was recorded in South-West Oltenia, the South-East and the West, which means that the pressure of rural tourism on sustainability in this region is much lower.
Regarding the Bucharest-Ilfov development region, although the specific correction is −20,115.99, here the pollution and environmental degradation are caused mainly by the urban agglomeration, by the very large number of motor vehicles and by some of the existing commercial and industrial complexes.
Analyzing the temporal effects, we see that corrections are generally positive. However, the corresponding values of 2009 (−15,320.37) and 2010 (−16,160.52) are worth mentioning, as they highlight the effects of the economic crisis on rural tourism in Romania. Since 2011, values which highlight favorable circumstances have been recorded.