The Impact of Tourism on Economic Performance: The Case of Turkey

RESULTS AND DISCUSSION

Variables subject to empirical analysis in this study determined based on the previous empirical work on the relationship between economic growth and tourism. The data is annual data on real Gross Domestic Product (GDP), Real Effective Exchange Rate (REER) and real total tourism receipts (TY) covering the period 1963-2006. All data has been taken from the Central Bank of the Republic of Turkey (CBRT) electronic data delivery system and Turkish Statistical Institute (TSI) (Table 1). Real tourism figures are obtained by deflating nominal tourism receipts by Consumer Price Index (CPI). All variables used in the empirical analysis are transformed variables by the use of natural logarithms.

Table 1:	Definitions of variables
*: Increases in LREER show the revaluation of the domestic currency because exchange rate is defined as foreign currency per unit of domestic currency

Table 2:	Augmented Dickey-Fuller and Phillips-Perron statistics for unit root tests
The selection of lag lengths for the ADF tests and the truncation lags for the PP test is determined based on the Schwartz Information Criterion (SIC) and the Newey-West bandwidth criterion, respectively, *: Indicates that the null hypothesis is rejected at the 5% level

Recent research on time series analysis shows that many macroeconomic time series contain unit roots and non-stationary regressors may invalidate most of the standard empirical results (Engle and Granger, 1987; Enders, 1995). Therefore, it is important to determine the stochastic properties of the series (whether a particular series is stationary at levels or difference stationary) before undertaking a modelling exercise in order to avoid spurious results. Such an analysis was undertaken for each of the variables of interest considered at levels using the Augmented Dickey-Fuller (ADF). The ADF statistics were calculated for the LGDP, LTY and LREER series including intercept and trend in the underlying Dickey-Fuller regressions. Considering the fact that in the presence of a structural break in a stationary series ADF test may reject the null of a unit root process where as in fact it is stationary, we also tested the level of integration of series using the Phillips-Perron (PP) test. In both cases, the selection of the number of lags is carried out using the Akaike Information Criteria (AIC). The results of the ADF and PP tests computed over the sample period for the levels and first differences of variables are shown in Table 2.

Inspection of the results table shows that the null hypothesis of unit root cannot be rejected in levels of variables but it is rejected in their first differences indicating that the variables considered are integrated of the same order 1, I(1).

Having established that our variables are integrated of the same order 1, we proceed with the multivariate Vector Autoregression (VAR) approach developed by Johansen (1988) and Johansen and Juselius (1990) to examine whether there is a long-run relationship among LGDP, LTY and LREER series. The VAR methodology is a better approach to investigate the long-run relationship because it approximates well the unknown model of the true economic structure by taking the dynamic interactions into account among the variables in the system. The selection of the optimum lag length in the underlying Vector Autoregression (VAR) model is carried out using the Schwarz ’s Information Criterion (SIC), Akaike Information Criterion (AIC) and Hannan-Quinn (HQ) criteria. The SIC, AIC and HQ criterions yield a VAR (1).

The inspection of Table 3 indicates that the null hypothesis of no cointegration (r = 0) among the variables is rejected in favour of the general alternative r = 1 since the trace statistic is 30.06, which is above the 95% critical value of 29.80. In addition, the null of r = 1 can not be rejected at a 5% level of significance. The maximum eigenvalue tests also show that the null hypothesis of no cointegration (r = 0) is rejected at 10% and the null of r = 1 is rejected at 5% level of significance the presence of only one cointegrating vector, r = 1. Taken the results provided by the trace and the maximum eigenvalue tests together, it can be concluded that there is only one cointegrating relationship among the three variables subject to empirical analysis.

Table 3:	Johansen-Juselius maximum likelihood cointegration tests
In the determination of lag length in the VAR model, the Schwarz ’s Information Criterion (SIC), Akaike Information Criterion (AIC) and Hannan-Quinn criteria (HQ) are employed, *: indicates that the null hypothesis is rejected at 5% significance level and **: denotes statistical significance at 10%. r stands for the number of cointegrating vectors

Table 4:	Granger causality tests, 1963-2006
*: Indicates that the null hypothesis is rejected a the 1% level

The estimates of the long run cointegrating vector and corresponding t-values are presented in Eq. 1. Examination of the Eq. 1 shows that the tourism receipts have a positive and statistically significant long-run effect on output level in Turkey. Interestingly, the positive and statistically significant coefficient on the effective exchange rate implies that the revaluation of the domestic currency increases the level of output in the long-run.

(1)

Although this finding seems counter-intuitive, there is well documented literature which shows the revaluation of domestic currency has a positive impact on output level. This relationship between economic growth and real appreciation stems from a differential productivity growth between the manufacture of traded goods and non-traded goods and services internationally and known as the Balassa-Samuelson effect. Such an appreciation over the long term has been evident among fast growing countries, such as Japan, Korea, Taiwan Province of China, Singapore and Hong Kong (Caramazza and Aziz, 1998). Bahmani-Oskooee (1992), Bahmani-Oskooee and Rhee (1996), Kamin and Rogers (2000), Faria and Leon-Ledesma (2003) and Lommatzsch and Tober (2004) provides evidence on this hypothesis.

Considering the discussions about the direction of causality among tourism and long-run growth, we carried out the Granger Causality test to determine the direction of causality among LGDP, LTY and LREER variables in the case of Turkey (Table 4).

As evident from the inspection of the table, the null hypotheses that both the LTY and LREER do not cause the LGDP are rejected at 1% level of significance implying that there is one-way causality going through the tourism receipts and the real effective exchange rate to the level of output. In addition, we found that the null hypothesis that LTY (LREER) does not cause LREER (LTY) could not be rejected at 5% level of significance indicating the lack of causality among these variables. These results imply that Turkey has exceeded the threshold level in tourism, which is measured by the share of tourism in GDP and therefore it is expected that the contribution of tourism to the economic performance of Turkey will be higher in the long-run.

CONCLUSION

The main goal of this study was to assess the relationship between tourism expansion and economic performance and to test the direction of causality among these variables using Turkish data over the period of 1963-2006. The estimated VAR model showed that there is only one cointegrating vector among real output, real tourism spending and real effective exchange rate and tourism has a long-run effect on output. In addition, causality test results indicated the presence of one-directional causality going through tourism and exchange rate to output. Obviously, these findings indicate that the threshold level in tourism has been exceeded and the tourism industry contributes substantially to the economic performance in Turkey. In this sense, it can be argued that the tourism receipts can be regarded as a stable source of foreign currency in Turkey and the payoff of economic policies directed to develop tourism will be higher in the long-run.