A SVECM Analysis of the Relationship between International Tourism Arrivals,GDP and Trade in Italy

Abstract

This article provides an SVECM investigation of long-run, short-run and contemporaneous relationships across per capita international tourism arrivals (ar), real GDP (y), and total international commercial transactions (tr) for the Italian economy. We find that variables span a bidimensional cointegrating space, which we normalize as long-run relationships between y and ar and between ar and tr. Signs and magnitudes of the estimated elasticities are as expected and compare well with the literature. The causation mechanism shows that none of the variables are weakly exogenous. What we find is that, whereas there appears to be unidirectional long-run causality running from y to tr and from tr to ar, bidirectional causality is detected between y and ar. Structural estimation and a study of the Impulse-Response functions of “meaningful” shocks hitting the economy are used to provide valuable insights for policy and business strategy design.

Keywords

tourism arrivals real GDP trade short-run and long-run causality SVECM

Introduction

The worldwide expansion in tourism demand that occurred over recent decades is generally acknowledged to have had a significant impact on the economies of recipient countries. For a specific destination area, in addition to generating foreign exchange earnings, international tourism is believed to have increased production, income, and employment. At the same time, it is argued that overall economic growth sustains the expansion of the tourism industry. Higher quality of tourism infrastructure and better management of local resources are only two of the many advantages that general growth can bring to tourism activity.

In consideration of the relevance of these issues, empirical research has devoted much effort in studying the complex dynamics linking tourism and GDP. The recurrently tested hypothesis is the so-called “tourism-led growth hypothesis”, which can inform policy making in countries seeking to improve their economic performance through tourism (cf. Balaguer and Cantavella-Jordá 2002). Not surprisingly, this kind of investigations relate primarily to either small tourism-based economies or to low- to medium-income countries where it can be safely assumed that tourism activity can push the overall growth process. Conversely, besides some notable exceptions (cf. Lee and Chang 2008; Nowak, Sahli, and Cortés-Jiménez 2007), the case of high-income diversified economies is largely neglected. Yet, the analysis of the tourism–growth nexus in this typology of countries would be a welcome addition to the tourism literature. On the one side, there is an increasing interest of many rich countries to approach inbound tourism as a specific growth mechanism to help boost overall economic performance and recover regional disparities. On the other, the tourism-growth nexus in these countries may well present specific characteristics with regard, for instance, to causality links and shock transmission mechanisms which are worth isolating.

This paper aims to contribute along the lines traced above. In particular, we propose, for the case of Italy, a Structural Vector Error Correction (SVECM) investigation of the relationship between real GDP and international per capita tourism arrivals where, as a further covariate, the share of total international commercial transactions in GDP is considered. The time profile of the reactions of the variables to (meaningful) shocks hitting the system is also obtained.

The choice of the country is motivated by the fact that Italy is a “representative” or “typical case” of a high-income diversified country interested in stimulating a competitive reorientation of its tourism industry as a way to boost growth performance. The country, although remaining one of the leading tourist destinations worldwide in absolute terms, seems to be losing some of its attractiveness with regard to its main international competitors. In 2010, according to the UNWTO (2011), it ranked 5th in the world, with respect to both international tourism receipts and arrivals, but it is expected to fall to 7th position in 2020 (cf. WTTC 2010). Moreover, in terms of the growth rate of visitors exports, Italy was at a surprisingly low 144th position in 2010, with forecasts placing the country even lower, at 161st, in 2020. Furthermore, although the Italian economy is large and broadly diversified, it was surprising to find that, in 2010, it only stood in 98th position for the weight of visitors exports in total exports (WTTC 2010), well below some of its main competitors such as Greece and Spain (respectively occupying the 43th and 59th position). This declining relative position combined with low shares and mounting economic woes for the country as a whole, have amplified pressures for public action to stimulate tourism competitiveness. Public debate mainly focuses on the need to profoundly reorganize the contradictory body of regulations that presently governs the sector and of recentralizing to a national level many key functions previously delegated to local authorities (e.g., attraction of foreign capital; plan and fund of “big events” of international relevance).

As for the empirical strategy, there are important novelties in the type of investigation we propose. To our knowledge it is the first time that a SVECM analysis is suggested in the study of the interplay between tourism activity and overall economic growth. Yet, the method appears particularly suitable for the scope, since it allows to clearly separate structural exogenous shocks from movements in the variables due to endogenous responses to current development of the economy. Moreover, policy implications are more easily derived in contexts where short and long-run restrictions consistent with theory can be imposed. Admittedly, the standard methodology in the field is the reduced-form VAR approach where mutual interactions within the period cannot be devised, and where disturbances arise from combinations of unobserved structural shocks. A VAR can be used for forecasting but not for structural analysis and policy evaluation. Therefore vital questions in tourism analysis remain to be appropriately addressed. Are there adverse feedback effects that might weaken long-run policy effectiveness? Is it possible, from the perspective of increased sustainability, that supply-side services and infrastructure investment can anticipate response hikes in the tourism market? Answering such questions requires a method with the ability to mimic the complex dynamic mechanism through which a shock in one variable diffuses through the system. This, in turn, urges investigators to go beyond the standard (at least in tourism literature) reduced-form VAR approach and to set up a structural analysis.

A further crucial element of our study that deserves attention concerns the use of a trade variable in the definition of the empirical setting. In this regard, we have made the choice of aligning our contribution to a promising new strand of the literature where a real-side variable is preferred to the more traditional real exchange rate in taking into account the external linkages of the destination country (cf. Katircioglu 2009a). According to this literature, important interactions and feedback effects among the three variables can be established. International trade can encourage tourism, for instance by means of more business travels occurring between countries and business people returning to these countries for pleasure (cf. Khan, Toh, and Chua 2005). At the same time, trade can stimulate economic growth through enhanced efficiency and promotion of technical progress.¹ Of course, the emergence of inverse causality links, with international trade responding to both growth and tourism developments, can also be taken into consideration in our multivariate setting.

The rest of the paper is organized as follows. The next section reviews the relevant literature. The third section presents the details of the empirical model. The fourth section discusses the results. In particular, it presents the characteristics of the cointegrating space and discusses the causality relationships between variables. Then, it reports contemporaneous coefficient estimates together with the corresponding Impulse-Response analysis. The final section reassesses the main findings of the study and draws together some implications for the tourism community.

Related Literature

There are two literatures on tourism that are closely related to our empirical analysis. The most prominent one investigates the relationship between tourism activity and growth. The theoretical rationales justifying a role for tourism in driving the growth process root in the classical international trade literature. A first mechanism refers to the possibility of using foreign exchange earnings to finance capital goods imports (McKinnon 1964). Other recurring arguments postulate that tourism can enhance the potential growth rate of the economy by increasing the overall efficiency of the economic system via increased competition among firms (Krueger 1980) and/or exploitation of economies of scale (Helpman and Krugman 1985).

Empirical investigations of the links between tourism activity and growth have recently undergone a qualitative jump thanks to the work of Balaguer and Cantavella-Jordá (2002). Following in this seminal contribution, researchers focus on the existence of a long-run relationship between tourism and growth variables and on the study of the direction of causality links. These issues are of strategic importance for the policy making: a complementarity long-run relationship, with unidirectional causality running from tourism to growth, for instance, would justify policies to promote tourism as a driver of regional or national economic growth. Causality in the opposite direction would indicate that tourism expansion results from general growth and so a focus on purely tourism-based intervention might be misplaced. Signs of bidirectional causality would imply feedback effects and a reciprocally reinforcing mechanism between the two phenomena suggesting that the policy maker should work in both directions. Empirical results remain controversial. Whereas evidence of a clear positive long-run equilibrium relationship between tourism activity and growth is almost always detected (notable exceptions are Oh 2005; Katircioglu 2009b), results on causality links differ widely across contributions, and sometimes for the same country, when different methodologies and samples are used. Limiting the presentation to long-run links, a one-way causal relationship running from tourism to income is found for Turkey (Gunduz and Hatemi 2005), Latin America (Eugenio-Martin and Morales 2004), Italy and Spain (Cortés-Jiménez and Pulina 2006) and the OECDs (Lee and Chang 2008). Inverse causality is found for Fiji islands (Narayan 2004), where long-run causality runs from income to tourism. Finally, bidirectional causality between tourism and income is detected for Greece (Dritsakis 2004), Taiwan (Kim, Chen, and Jang 2006), and Turkey (Ongan and Demiroz 2005).

The second literature relevant to our study scrutinizes the relationship between tourism and international trade, either in aggregate or in terms of product category. Theoretical arguments justifying in the field a link from tourism to trade assume, for instance, that tourism drives trade because of import needs of the tourism industry and that tourism activates exports because of consumption shifts from origin to destination country (cf. Khan, Toh, and Chua 2005; Santana-Gallego, Ledesma-Rodríguez, and Pérez-Rodríguez 2011b). Other recurrent elements are based on the possibility that international visitors may identify (and subsequently seize) business opportunities that could lead to either export sales or import purchases. Convincing arguments for supporting inverse causality links are also developed in the field. What is suggested is that international trade encourages tourism, for instance by means of more business travels occurring between countries and business people return to these countries for pleasure (Khan, Toh, and Chua 2005). Again, there are important policy implications depending on the detected direction of causality links. For instance, a causality running from tourism to trade would justify tourism policy as a means for addressing structural deterioration in the trade balance. An inverse causality direction would motivate government interventions on exploiting the international trade channel as a means for increasing tourist arrivals into a country.

Here again, empirical evidence on the nexus between tourism and trade is controversial. While a long-run complementary relationship between the two variables is almost always detected, evidence on the causality links is rather mixed. In this regard, works appear more rooted in the short-run and only few authors provide details of long-run Granger causality between tourism and trade variables. A notable exception is Fry, Saayman, and Saayman (2010), who study the relationship between tourist arrivals and total trade in South Africa. Panel results favor a bidirectional causality between the two variables. Santana-Gallego, Ledesma-Rodríguez, and Pérez-Rodríguez (2011a, 2011b) propose a detailed study of the causal links between trade and tourism for the case of the Canary Islands. At aggregate level, it is interesting to find that a bidirectional link exists between exports and arrivals, and a link in the sense that tourist arrivals lead to imports and total trade. In a preceding study, Santana-Gallego, Ledesma-Rodríguez, and Pérez-Rodríguez (2010) found the same type of link between arrivals and total trade for United Kingdom. Kulendran and Wilson (2000) also look at long-run Granger causality. The authors use data for Australia and some of its most important travel and trading partners and find that tourism Granger-causes total trade to Japan but nowhere else. Finally, tourism is found either unrelated or weakly related to trade for Singapore (Khan, Toh, and Chua 2005).²

As anticipated in the Introduction, the two discussed literatures have recently converged in an all-encompassing framework of analysis where the complex dynamics linking real GDP, tourism, and trade are jointly studied. Contributions are scant in number, but provide interesting insights. Among them, the work of Katircioglu (2009a) is noteworthy. The author tests for the existence of cointegration and causal links between international tourism arrivals, international trade (also disaggregated into exports and imports) and economic growth in Cyprus. The investigation concerns the existence of a long-run relationship between each pair of variables by means of the bounds test for cointegration in an ARDL-ECM model. To study the causality links, given the evidence of cointegration, an ECM framework is established. He finds that a triangular causal nexus exists among the three variables with tourism and trade (considered in the three specifications of total trade, exports, and imports) responding to real GDP, and tourism responding to trade. Following in the same methodological footsteps, Sarmidi and Salleh (2010) propose a cointegration analysis for Malaysia and its bilateral relationships with its major ASEAN tourism partners (Singapore, Thailand, Indonesia, and Brunei-Darussalam). Evidence roughly confirms the Katircioglu (2009a) results for the trade–GDP and the tourism–GDP causal links, but on the trade/tourism side, cointegration is barely found. When detected (as in the case of bilateral Malaysia/Indonesia and Malaysia/Brunei-Darussalam flows), causality links either present inverted directions with respect to Katircioglu (2009a) or evidence of feedback effects.

Related contributions also come from Durbarry (2004), Nowak, Sahli, and Cortés-Jiménez (2007), and Akinboade and Braimoh (2010). Durbarry (2004) shows, by means of Johansen’s methodology, that for Mauritius, tourism receipts cointegrate with GDP, exports of sugar, manufactured product, and human and physical capital in a production function setting. The author does not provide a comprehensive study of the causal relationship. Nowak, Sahli, and Cortés-Jiménez (2007) test the hypothesis that tourism exports, as a source of foreign currency, finance imports of capital goods and, eventually, growth. The authors use the Johansen’s cointegration approach (Johansen 1988, 1991) and apply multivariate Granger tests to find that there exists one stationary linear combination among the variables and that bidirectional long-run causality links exist between each pair of variables. Akinboade and Braimoh (2010) find for South Africa that tourism receipts cointegrate in a four-variate VECM with GDP, real exports, and real exchange rate. In terms of long-run causality effects, they find that both tourism receipts and real exports Granger-cause GDP. The causality links between real exports and tourism receipts, however, are not investigated by the authors. The chosen methodology again relies on Johansen’s (1988, 1991) ML procedure.

Data and Empirical Model

As discussed in the Introduction, this article develops an SVECM analysis for the Italian economy of the relationships linking real GDP, per capita international tourism arrivals and the share of total commercial transactions (exports plus imports) with respect to GDP. We consider quarterly observations for the time span 1987q1-2009q4 from the National Institute for Statistics (ISTAT).³ GDP and trade flows are expressed in real terms (euros at 2000). All data are seasonally adjusted. The normalization of arrivals with respect to population, and trade with respect to real GDP, allows the results to be read as a relationship between an income variable and the degrees of tourism specialization and openness of the economy. The choice of arrivals to proxy for inbound tourism demand arises from the fact that, in Italy, this variable is better documented than others (for instance expenditure) and less likely to report measurement errors. Moreover, longer homogeneous time series are available for arrivals.

Consider a Reduced Finite-Order VAR of the Form

X_{t} = \sum_{k = 1}^{m} Ψ_{k} X_{t - k} + Ξ D_{t} + e_{t},

where m is the appropriate order of lag polynomial, Ψk is the matrix containing the parameters of interest, et is a vector of observable residuals, and Ξ is the coefficient matrix associated with deterministic terms, such as a constant, trend and observation-specific dummies. In the above model, X_t = [Y_t, Ar_t, Tr_t]' represents the vector of our variables, where Y_t is national real GDP, Art is per capita international tourism arrivals, and Tr_t is the share of imports plus exports over GDP.

When the study of the dynamic properties suggests that the variables are nonstationary and cointegrated, equation (1) can be expressed in a VECM representation by subtracting X _t−1 from both sides. The model can be rewritten as follows:

Δ X_{t} = α β^{'} X_{t - 1} + \sum_{k = 1}^{m} Γ_{k} Δ X_{t - k} + e_{t}

where Γ _k contains the coefficients of the variables in differences (with k indicating the lag order, with a maximum lag of m), β′ is the (3 × r) matrix of long-run coefficients, with r the cointegration rank of the system. For the sake of simplicity, as often in practice, equation (2) omits the deterministic components.

We have three possible outcomes relating to the dimension of the αβ' impact matrix. The least interesting cases occur when the rank is zero or full. When the rank is full, all elements of the vector X_t are stationary; when the rank is zero, all variables are I(1) but not cointegrated. In this case, a VAR in first differences is appropriate. However, when the impact matrix is estimated to be of intermediate rank, r, there exist r cointegrating vectors, that is, linear combinations linking the variables in the long run. Since these vectors are not uniquely determined, it is necessary to impose some restrictions in order to identify the cointegrating space. This requires one normalization and (r − 1) restrictions on each cointegrating vector. Once an identification scheme is chosen, we are in the position to evaluate the type of causal relationships linking the variables of the system. However, when structural exogenous shocks have to be separated from movements in the variables due to endogenous responses to current developments in the economy, it is difficult to give an economic interpretation to reduced-form VAR equations. This problem is solved by using an SVECM approach. The first step to getting a structural representation of equation (2) requires identification of the structural innovations that induce informative responses of the system variables X_t. In particular, effects of the fundamental shocks ε_t on the system variables X _t can be described by the following structural specification:

A Δ X_{t} = A α β^{'} X_{t - 1} + A \sum_{k = 1}^{m} Γ_{k} Δ X_{t - k} + ε_{t},

The effects of fundamental shocks on the system variables X _t are expressed as:

e_{t} = ε_{t} A^{- 1} = B ε_{t},

where the ( K × 1) vector of unobservable structural disturbances ε _t has zero mean and covariance matrix Σε. In order to compute the responses to structural shocks ε_t, we have to recover the K² elements of the matrix B (cf. Amisano and Giannini 1997). For this purpose, we need a set of identifying restrictions. Notice that the process of imposing contemporaneous restrictions is always somewhat arbitrary. When sound theoretical models behind the phenomena under study are available, some guidance can be provided by economic considerations. Otherwise, a more empirically based approach can be used. For instance, viable restrictions can be searched for by means of a Granger causality analysis. Contemporaneous relationships can also be discerned by relying on statistical tests regarding the significance of the coefficients. By assuming that structural shocks are uncorrelated and have unit variance Σ_ε = I _K, using equation (4) we get:

\sum_{e} = E [e_{t} {e^{'}}_{t}] = B E [ε_{t} {ε^{'}}_{t}] B^{'} = B \sum_{t} B^{'} = B B^{'} .

Now in order to uniquely identify the K² matrix B , the K(K + 1)/2 independent restrictions due to the covariance matrix Σ _e = BB' are not enough. At this level, we need to impose additional K(K − 1)/2 linearly independent restrictions. Once the B matrix is identified, the impulse response analysis can be applied. This analysis depicts the time profile of the effect of structural shocks on each variable of the system and turns out to be very useful for policy simulation exercises.

Results

This section outlines the main results obtained by our empirical analysis. As a first step, we study the long-run properties of the variables. Then, we examine short-run and long-run co-movements and study the type of causal relationships linking the three variables within the system. Finally, we implement the SVECM analysis to estimate the contemporaneous relationships and to compute Impulse-Response functions.

Unit-Root Analysis

The first step of the analysis is to study the order of integration of the (log-transformed) variables. To minimize the danger of erroneous inference, we perform an array of different tests. On one hand, besides the baseline Augmented Dickey-Fuller (ADF), we perform the Elliott, Rothenberg, and Stock (1996) (ADF-GLS) and the Phillips and Perron (1988) (PP) unit-root tests, which are all various generalizations of the ADF principle, where the null of a unit root is tested against the alternative of a stationary process.⁴ On the other hand, we apply the KPSS (Kwiatkowski et al. 1992) procedure where the hypotheses are interchanged.⁵

Results are presented in Table 1, where lowercase initials denote log-scaled variables. The lag structure for ADF and ADF-GLS is selected by means of the Akaike information criterion. PP and KPSS statistics are generated using Bartlett Kernel with Newey-West’s Bandwith. For variables in levels only, models with a time trend are tested (trend stationarity). For first differences, models with only a constant, and models with a constant and a time trend are run. As shown in the table, the ADF test establishes clear evidence of nonstationarity for y and ar variables. Conversely, it leads to the undesirable result of trend stationarity for tr. The ADF-GLS and PP tests produce more reassuring results, since nonstationarity is recovered for all variables. The KPSS tests mirror these results, rejecting the nulls of trend stationarity for all variables.

Table 1.

Unit Root Testing

Variables in levels: trend and intercept included in test equations
	y	(lags/bandwidth)	ar	(lags/bandwidth)	tr	(lags/bandwidth)
ADF	−0.95	(1)	−1.38	(1)	−4.66***	(2)
ADF-GLS	−0.95	(1)	−1.58	(1)	−2.50	(0)
PP	−0.73	(5)	−1.66	(1)	−2.75	(2)
KPSS	0.19**	(6)	0.17**	(7)	0.19**	(0)

Variables in differences: only intercept included in test equations
	Δy	(lags/bandwidth)	Δar	(lags/bandwidth)	Δtr	(lags/bandwidth)

ADF	−5.13^***	(0)	−11.84^***	(0)	−5.53^***	(5)
ADF-GLS	−0.83	(4)	−2.30^**	(1)	−9.92^***	(0)
PP	−5.18^***	(3)	−11.84^***	(0)	−10.22^***	(2)
KPSS	0.49^**	(5)	0.20	(4)	0.08	(1)

Variables in differences: trend and intercept included in test equations
	Δy	(lags/bandwidth)	Δar	(lags/bandwidth)	Δtr	(lags/bandwidth)

ADF	−5.54^***	(0)	−11.82^***	(0)	−5.46^***	(5)
ADF-GLS	−5.08^***	(0)	−4.21^***	(1)	−5.51^***	(5)
PP	−5.58^***	(2)	−11.82^***	(0)	−10.21^***	(2)
KPSS	0.09	(5)	0.08	(4)	0.05	(1)

Note: The number of lags for the ADF and the ADF-GLS tests is selected using AIC. PP and KPSS use Bartlett Kernel with Newey-West’s Bandwith.

***, **, and * denote rejection at a 1, 5, and 10 percent critical value, respectively.

As for statistics of first differences, stationarity is straightforwardly established by the ADF, ADF-GLS, and PP tests for Δar and Δtr, both in models where the deterministic component is only represented by a constant and when the model also contains a time trend. This does not happen for Δy. For this variable, we observe that when the model contains only a constant, the ADF-GLS test is not able to reject the null of non-stationarity. The result is mirrored by the KPSS test, which also establishes the presence of a unit-root in the first difference of y. However, noncontradictory results on the stationarity of Δy are established when a time trend is taken into consideration in the model.⁶

Therefore, since there is enough evidence of a stochastic trending behavior of all the variables, we can proceed to the estimation of the parameters of the VECM.

The Cointegrating Space and the Causal Relationships

We now turn our attention to the analysis of the cointegrating properties of the above set of variables according to Johansen’s procedure (Johansen 1988, 1991). Recalling the results of the unit-root testing analysis, the choice of the deterministic structure requires, in our case, some caution. Therefore, to avoid mis-specification errors, we sequentially LR test for the presence of the deterministic components in the cointegrating space and in the short-run equations. According to the likelihood ratio test, the preferred model includes a deterministic time trend also in the first difference equations. Some observation-specific dummy variables are also included. The preferred lag order of 1 in the short-run equations is Akaike-selected.

Overall, as it is reported in Table 2, the Johansen’s procedure does not reject the null that the rank of the stochastic matrix is 2. The rejection of H ₀: r = 1 is at a 5 percent critical value in the case of the Trace Stat.⁷ It is slightly less robust in the case of the Max-Eigenvalue Stat.

Table 2.

Cointegration Test Results

	Max-Eigenvalue		Trace
Rank	Statistics	p-Value	Statistics	p-Value
r = 0	23.60	0.06	42.09	0.01
r = 1	16.16	0.07	18.48	0.04
r = 2	2.33	0.23	2.33	0.23

Having determined the dimension of the cointegrating space, we are in the position to perform our error-correction analysis. Results are reported in Table 3. In general terms, the performance of the model is satisfactory, with regard to both normality and serial correlation of the residuals (statistics not reported). To facilitate the reading of the results, we have normalized the cointegrating space in such a way the first error correction mechanism (Ecm1) links y and ar in the long-run, while the second (Ecm2) expresses a long-run relation between ar and tr. Estimated long-run coefficients are statistically significant at a 1 percent level and present expected signs and reliable magnitudes (cf. section A in Table 3). More specifically, the long-run elasticity of y with respect to ar is estimated to be 0.45, suggesting that a 1 percent point increase in ar goes with a 0.45 percent increase in y. This result perfectly matches with some part of the extant empirical evidence (cf. the elasticities of 0.24 and 0.61 reported by Lee and Chang 2008, for OECDs and non-OECDs respectively), while it appears much larger than the value of 0.08 estimated for Italy by Cortés-Jiménez and Pulina (2006). Notice however that the empirical setting developed by the latter authors is different than ours in many relevant aspects.

Table 3.

Estimation Results

A: Long-Run Analysis
Ecm1	y = 13.51 + 0.45ar – 0.001t
	[5.85***]
Ecm2	ar = 3.48 + 4.14tr – 0.013t
	[4.63***]
B: Causality Tests
	Δy	Δar	Δtr
Long run
Ecm1(–1)		5.00**		4.04**	1.35
		(0.03)		(0.04)	(0.24)
Ecm2(–1)		0.14		8.11***	7.18***
		(0.70)		(0.00)	(0.01)
Ecm1(–1); Ecm2(–1)		6.57**		14.28***	13.22***
		(0.04)		(0.00)	(0.00)
Short run
Δy		−		11.87***	0.06
				(0.00)	(0.81)
Δar		0.05		−	0.73
		(0.83)			(0.39)
Δtr		0.06		0.01
		(0.81)		(0.94)

The p-values are given in parentheses. Square brackets give t-values. χ² statistic is used for LR and Wald tests.

***, **, and * denote, respectively, rejection of the null of non-stationarity at a 1, 5 and 10 percent critical value.

As far as the second long-run equilibrium relationship is concerned, the estimated elasticity implies that a 1 percent point increase in tr is associated with a 4.14 percent increase in ar, suggesting that a complementary relationship also exists between these two variables. The sign of this relationship is aligned with the rest of the relevant empirical literature. Point estimates, however, appear somewhat higher than those reported by other authors. For instance, Santana-Gallego, Ledesma-Rodríguez, and Pérez-Rodríguez (2011a) report elasticities ranging from 0.76 to 1.73 depending on the estimation method for a panel of OECD countries.

We now turn our attention to the causal relationships linking each pair of variables. The error correction model allows for two sources of causation: short-run causality, detected through the lagged differences, and long-run causality, detected through the error correction term. With respect to short-run causality, Wald testing highlights the presence of only one direction of causality running from Δy to Δar (i.e., the null of noncausality is rejected only in the Δar equation for the coefficient reported by Δy). Conversely, multiple causality relationships can be detected in the long-run. In such a case (cf. again Table 3), we can see that the lagged value of the Ecm1 is significant for y and ar, whereas the Ecm2 is significant for ar and tr. Accordingly, there appears to be unidirectional long-run causality from GDP to tr and from tr to ar. Bidirectional long-run causality is obtained only once for the relationship between GDP and tourism arrivals. All in all, from the long-run perspective, none of the variables included in the analysis can be considered as being weakly exogenous.

Comparing our results with earlier literature is not an easy task, since, as highlighted in the Related Literature section, extant empirical evidence is rather mixed. Restricting attention to three-variate analyses, it is interesting to confront our results with the triangular long-run causal nexus found by Katircioglu (2009a) for Cyprus (cf. Figure 1). The author finds that real GDP causes both tourism and trade. Tourism is also led by trade in the long run. In contrast to these findings, our results also show feedback effects running from tourism to real GDP.

Figure 1.

Trade, tourism and, growth triangle

Contemporaneous Coefficients

In this section, we estimate the contemporaneous coefficients of the structural model. As explained above, a set of K(K−1)/2 linearly independent overidentifying restrictions on the coefficients of matrix B is required. In setting to zero some entries of the matrix B , we could make the choice of drawing information from theory. For instance, the identification scheme could be derived from the tourism-led or export-led growth literatures, which can be extended to imply a contemporaneous causal ordering going from tr and ar towards y. However, given the empirical uncertainty regarding causal links in the trade–growth–tourism nexus, we have preferred to proceed through a purely statistical criterion, searching for significant coefficients. Therefore, zeros are assigned in matrix B wherever we fail to reject the null hypothesis indicating the absence of contemporaneous impact. Interestingly, the significant coefficient gives rise to the following quasi lower triangular B matrix

B = [\begin{matrix} 0.005 & 0 & 0 \\ 0.006 & 0.028 & 0 \\ - 0.006 & 0 & 0.022 \end{matrix}],

implying an almost complete recursive structure that results in a particular causal ordering of the variables in the system. More specifically, the contemporaneous causal ordering corresponds to the following system of equations:

e_{y} = b_{11} ε_{y} = 0.005 ε_{y},

e_{a r} = b_{21} ε_{y} + b_{22} ε_{a r} = 0.006 ε_{y} + 0.028 ε_{a r},

e_{t r} = b_{31} ε_{y} + b_{33} ε_{t r} = - 0.006 ε_{y} + 0.022 ε_{t r} .

Accordingly, GDP is contemporaneously exogenous to any other variable in the system and only responds to its own innovations. Moreover, there is no significant within-quarter relationship between tr and ar. The signs of the estimated coefficients, furthermore, imply that a shock in y has a positive influence on the tourism variable whereas it has a negative impact on tr.

For a clearer reading of this contemporaneous analysis, we compare the causal ordering emerging in equations from (7) to (9) with the short-run (cf. Table 3) and long-run causality links (cf. Table 3 and Figure 1). We note that the link between y and ar is confirmed in both the short and the long-run. In this regard, however, the mechanisms at work in these different time perspectives need not be the same. In the case of the contemporaneous causal ordering, within-quarter complementarity might rely on advertising and promotion facilities, which attracts more international tourists. In the longer term, positive real GDP shocks tend to stimulate investment opportunities and capacities in the tourism sector, increasing the attractiveness of the country as a destination. Notice, however, that in the long-run the two variables show bidirectional causality. This implies that external resources provided by tourism expansion also create a feedback effect that further reinforces the growth process.

In interpreting the contemporaneous effect of GDP shocks on the trade variable, the negative sign does not contradict the long-run causality link (cf. Table 3). In fact, by construction, the competitiveness variable is normalized with respect to y and, therefore, in the very short-run we only observe a “denominator” effect. Once the income effect on real imports and exports starts working, then our trade variable increases.

The only long-run link missing in the contemporaneous analysis is the one running from tr to ar. This implies that growth in trade sectors is able to pull tourism activities through channels that are only effective in the long-run.

Impulse-Response Analysis

We now turn our attention to the Impulse-Response analysis that provides a dynamic description of the effects of structural shocks to the economy. There are several possible types of shock: on the one side, the shock may depend on external (international) events impacting on any one of the variables in the system; on the other, internal choices can provide the basis for innovation in the variables. Confining the discussion to the tourism variable, events of the first type include, among others, changes in tastes and attitudes that affect any of the origin countries, or international episodes with a more worldwide relevance such as terrorist attacks, or diseases. Events of the second type derive from internal choices, both private and public, aimed at encouraging the expansion of tourist activities within the country. Consider, for instance, the possibility that, from the private investor’s side, the business community agrees on more aggressive price or marketing strategies. On the public side, we can think of any public action aimed at favoring the competitiveness of the tourism sector, such as appropriate fiscal policies, and/or increasing the attractiveness of the country as a tourism destination, for example, by organizing “big events” or by promoting cultural and natural endowments.

Figures from 2 to 4 depict estimated responses to a one standard deviation shock hitting the system variables. Figures also report bootstrap percentile 95 percent confidence intervals constructed using the method proposed in Hall (1992). For simplicity, comments will be limited to central point estimates.

Figure 2.

Impulse response to a tourism arrivals shock

Figure 3.

Impulse response to an GDP shock

Figure 4.

Impulse response to a trade shock

Let us first comment on the simulated effects of a shock hitting tourist arrivals on real GDP and trade. As reported in Figure 2, we find that a positive shock on ar increases real GDP steadily up to the 13th quarter, when the response stabilizes at around 0.35 percent. This time pattern suggests that any kind of event that contributes to raise tourism activity has a positive impact on real GDP. In parallel, the shock on ar has an initial positive (although weaker) influence also on tr. The response of the trade variable is not monotonic: as seen in Figure 2, the positive initial impact is soon reabsorbed, to become slightly negative between the 2nd and 6th quarters. After that, the tr variable starts rising again, showing a maximum response (within the reported 20-quarters) of 0.19 percent.

Turning now to the response of ar to a one standard deviation shock in real GDP, we observe a modest instantaneous effect that accumulates over time (cf. Figure 3). The maximum reaction is reached very fast; then the response coefficient remains steady between 1.3 and 1.4 percent. Notice that this profile is consistent with the causality links discussed in the previous subsections; that is, both short- and long-run factors are effective in leading to an increase in inbound tourism.

As far as the trade variable is concerned, the contemporaneous impact of a one standard deviation shock in y is negative (through the denominator effect). However, tr takes only three-quarters to regain its initial value. After that it rises until the 15th quarter, reaching a value of 0.4 percent that persists over the entire time horizon.

To conclude, Figure 4 shows that consistent with the theory, a shock hitting tr positively affects both real GDP and tourism arrivals over the entire 20-periods horizon.

Concluding Remarks and Discussion

In this study, we investigate long-run, short-run, and contemporaneous relationships across international per capita tourism arrivals, real GDP, and the share of total commercial transactions on GDP for the Italian economy. This trivariate context aligns this study to promising recent literature, which encompasses, in a unifying framework, studies of the relationship between growth and incoming tourism with those investigating the nexuses between tourism activity and international trade. The SVECM approach is particularly fitting, since, by allowing for a clear separation of exogenous shocks from movements in the variables due to endogenous responses, it provides the correct basis for analyzing the overall effects of policy actions.

The cointegration analysis gives evidence of two independent long-run relationships across the variables according to which real GDP and international arrivals on the one side, and total trade and international arrivals on the other, emerge as complementary activities. As far as long-run causality is concerned, we find that none of the variables is the exclusive bearer of the stochastic trend driving the system. On the contrary, we find that real GDP, while causing trade, is characterized by a reciprocal, reinforcing mechanism from tourism expansion. At the same time, a unidirectional causality running from trade to tourism flows is uncovered. This causality network across variables simplifies the short-run perspective, since Granger causality tests only give evidence of real GDP preceding tourism arrivals.

From the structural coefficients perspective, our statistically inferred identification scheme reveals that the most exogenous variable is the real GDP, which only reacts to its own shocks and instantaneously affects both tourism arrivals (positively) and trade share (negatively). The investigation of the dynamic interactions of the variables by means of the Impulse-Response Analysis provides insights into the reaction of the three variables to shocks. A particularly interesting general result is that tourism and trade variables react very fast to real GDP shocks but a much longer time lapse is required for real GDP to stabilize after a shock in the other two variables.

Several aspects of our analysis can be useful to policy makers and business actors. The policy maker, for instance, is encouraged to devise sectoral policies able to support national destinations in the world market since there are significant returns for the economy in terms of higher GDP growth rates. This effect also appears self-reinforcing in two distinct ways; one direct, through the GDP-tourism feedback, and one indirect, through increased trade opportunities. This latter channel is made possible by the causal links running from real GDP to trade and from trade to international arrivals. In more detail, we find that the increase in real GDP made possible by a sudden injection of international arrivals impacts on the trade sector, creating the conditions for a further positive effect on inbound tourism demand, namely, a virtuous circular mechanism. In addition, policy makers should be aware that this latter mechanism can be further fed by direct policy measures aimed at bettering Italy’s international standing in trade partnerships. Establishing trade negotiations with emerging trading nations, and liberalizing trade through free commerce agreements can also be important drivers for tourism and real GDP.

Below the policy-making level, other stakeholders in Italian tourism may find our analysis useful. For instance, a robust complementarity relationship between tourism and trade is of importance for managers (among others) in airlines, tour operators, and hotel sectors of targeting countries that develop high trade growth rates with Italy. Also, the complementary relationship between real GDP growth and tourism provides valuable information on how to program tourism activity from private investors. In particular, given the simultaneous links between growth and tourism, forecasts of general growth in the economy are signals that can help in the timing of supply-side services in tourism organizations.

Further implications for the policy maker and the tourism community are difficult to isolate in the present study. Limitations derive from the aggregate nature of the analysis which, although providing an appropriate framework for a macroeconomic investigation into the effects of tourism expansion, do not enable stakeholders to devise specific actions to assist Italy in its quest for increasing tourism demand. Demand segmentations and/or regional perspectives are necessary to provide more locally tailored or business-tailored details in approaching Italian tourism expansion.

Footnotes

Declaration of Conflicting Interests

The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Funding

The author(s) received no financial support for the research, authorship, and/or publication of this article.

Notes

Bios

Carla Massidda is assistant professor at Cagliari University (Italy), where she teaches tourism and environmental economics. Her main research interests concern the development of regional economies with particular attention to the role of tourism.

Paolo Mattana is full professor at Cagliari University (Italy). He holds courses of econometrics and applied economics. His research activity mainly focus on growth theory (indeterminacy issues and chaotic solutions) and on time series analysis in contexts of regional and tourism economy.

References

Akinboade

O. A.

Braimoh

L. A.

(2010). “International Tourism and Economic Development in South Africa: A Granger Causality Test.” International Journal of Tourism Research, 12: 146-63.

Ahmed

Kwan

A. C. C.

(1991). “Causality between Exports and Economic Growth.” Economic Letters, 37: 243-48.

Amisano

Giannini

(1997). Topics in Structural VAR Econometrics (2nd ed.). Berlin: Springer-Verlag.

Aradhyula

Tronstad

(2003). “Does Tourism Promote Cross-Border Trade?” American Journal of Agricultural Economics, 85: 569-79.

Asikainen

(2003). “Long Memory and Structural Breaks in Finnish and Swedish Party Popularity Series.” University of Helsinki, Department of Economics, Discussion Papers 586.

Assaf

A. G.

Pestana-Barros

Gil-Alana

(2011). “Persistence in the Short- and Long-term Tourist Arrivals to Australia.” Journal of Travel Research, 50: 213-29.

Baillie

R. T.

Chung

Tieslau

M. A.

(1996). “Analyzing Inflation by the Fractionally Integrated Arfima-Garch Model.” Journal of Applied Econometrics, 1: 23-40.

Balaguer

Cantavella-Jordá

(2002). “Tourism as a Long-Run Economic Growth Factor: The Spanish Case.” Applied Economics, 34: 877-84.

Çetintaş

Barişik

(2009). “Export, Import and Economic Growth: The Case of Transition Economies.” Transition Studies Review, 15: 636-49.

10.

Cortés-Jiménez

Pulina

(2006). “A Further Step into the ELGH and TLGH for Spain and Italy.” Fondazione Eni Enrico Mattei. Nota di lavoro 118.

11.

Cuñado

Gil-Alana

L. A.

Perez de Gracia

(2008). “Fractional Integration and Structural Breaks: Evidence from the International Monthly Arrivals in the US.” Tourism Economics, 14: 13-23.

12.

Deme

(2002). “An Examination of the Trade-Led Growth Hypothesis in Nigeria: A Co-integration Causality, and Impulse Response Analysis.” The Journal of Developing Areas, 36: 1-15.

13.

Dritsakis

(2004). “Tourism as a Long-Run Economic Growth Factor: An Empirical Investigation for Greece Using Causality Analysis.” Tourism Economics, 10: 305-16.

14.

Durbarry

(2004). “Tourism and Economic Growth: The Case of Mauritius.” Tourism Economics, 10: 389-401.

15.

Elliott

Rothenberg

Stock

J. H.

(1996). “Efficient Tests for an Autoregressive Unit Root.” Econometrica, 64: 813-39.

16.

Eugenio-Martin

J. L.

Morales

N. M.

(2004). Tourism and Economic Growth in Latin American Countries: A Panel Data Approach. Social Science Research Network Electronic Paper.

17.

Fischer

(2004). “The Influence of Immigration and International Tourism on the Demand for Imported Food Products.” Food Economics, 1: 21-33.

18.

Fischer

Gil-Alana

L. A.

(2009). “The Nature of the Relationship between International Tourism and International Trade: The Case of German Import of Spanish Wine.” Applied Economics, 41: 1345-59.

19.

Fry

Saayman

(2010). “The Relationship between Tourism and Trade in South Africa.” South African Journal of Economics, 78: 287-306.

20.

Gil-Alana

L. A.

Cuñado

Perez de Gracia

(2004). “Seasonal Fractional Integration in the Spanish Tourism Quarterly Time Series.” Journal of Travel Research, 52: 408-14.

21.

Gonzalo

Lee

T. H.

(1998). “Pitfalls in Testing for Long-Run relationships.” Journal of Econometrics, 86: 129-54.

22.

Gonzalo

Lee

T. H.

(2000). “On the Robustness of Cointegration Tests When Series Are Fractionally Integrated.” Journal of Applied Statistics, 27: 821-827.

23.

Gunduz

Hatemi

J. A.

(2005). “Is the Tourism-Led Growth Hypothesis Valid for Turkey?” Applied Economics Letters, 12: 499-504.

24.

Hall

(1992). The Bootstrap and Edgeworth Expansion. New York: Springer.

25.

Hatemi

J. A.

Irandoust

(2001). “Productivity Performance and Export Performance: A Time-Series Perspective.” Eastern Economic Journal, Eastern Economic Association, 27: 149-64.

26.

Helpman

Krugman

(1985). Market Structure and Foreign Trade. Cambridge MA: MIT Press.

27.

Jin

C. J.

(1995). “Export-Led Growth and the Four Little Dragons.” The Journal of International Trade and Economic Development, 4: 203-15.

28.

Johansen

(1988). “Statistical Analysis of Cointegration Vectors.” Journal of Economics Dynamics and Control, 12: 231-54.

29.

Johansen

(1991). “Estimation and Hypothesis Testing of Cointegration Vectors in Gaussian Vector Autoregressive Models.” Econometrica, 59: 1551-80.

30.

Katircioglu

S. T.

(2009a). “Tourism Trade and Growth: The Case of Cyprus.” Applied Economics, 41: 2741-50.

31.

Katircioglu

S. T.

(2009b). “Revisiting the Tourism-Led-Growth Hypothesis for Turkey Using the Bounds Test and Johansen Approach for Cointegration.” Tourism Management, 30: 17-20.

32.

Khan

Toh

R. S.

Chua

(2005). “Tourism and Trade: Cointegration and Granger Causality Tests.” Journal of Travel Research, 44: 171-76.

33.

Kim

J. H.

Chen

M. H.

Jang

(2006). “Tourism Expansion and Economic Development: The Case of Taiwan.” Tourism Management, 27: 925-33.

34.

Krueger

(1980). “Trade Policy as an Input to Development.” American Economic Review, 70: 188-292.

35.

Kulendran

Wilson

(2000). “Is There a Relationship between International Trade and International Travel.” Applied Economics, 32: 1001-9.

36.

Kwan

A. C. C.

Cotsomotis

(1991). “Economic Growth and the Expanding Export Sector: China 1952-1985.” International Economic Review, 5: 105-17.

37.

Kwiatkowski

Phillips

P. C. B.

Schmidt

Shin

(1992). “Testing the Null Hypothesis of Stationarity against the Alternative of a Unit Root.” Journal of Econometrics, 54: 159-78.

38.

Lee

C. C.

Chang

C. P.

(2008). “Tourism Development and Economic Growth: A Closer Look at Panels.” Tourism Management, 29: 180-92.

39.

Lee

Schmidt

(1996). “On the Power of the KPSS Test of Stationarity against Fractionally Integrated Alternatives.” Journal of Econometrics, 73: 285-302.

40.

Marin

(1992). “Is the Export-Led Growth Hypothesis Valid for Industrialized Countries.” Review of Economics and Statistics, 74: 678-88.

41.

McKinnon

(1964). “Foreign Exchange Constrain in Economic Development and Efficient Aid Allocation.” International Economic Review, 5: 105-17.

42.

Narayan

P. K.

(2004). “Economic Impact of Tourism on Fiji’s Economy: Empirical Evidence from the Computable General Equilibrium Model.” Tourism Economics, 10: 419-33.

43.

Nowak

J. J.

Sahli

Cortés-Jiménez

(2007). “Tourism, Capital Good Imports and Economic Growth: Theory and Evidence for Spain.” Tourism Economics, 13: 515-36.

44.

C. K.

(2005). “The Contribution of Tourism Development to Economic Growth in the Korean Economy.” Tourism Management, 26: 39-44

45.

Ongan

Demiroz

D. M.

(2005). “The Contribution of Tourism to the Long-run Turkish Economic Growth.” Ekonomicky Casopis [Journal of Economics], 53: 880-94.

46.

Phillips

P. C. B.

Perron

(1988). “Testing for a Unit Root in Time Series Regression.” Biometrika, 75: 335-46.

47.

Santana-Gallego

Ledesma-Rodríguez

F. J.

Pérez-Rodríguez

(2010). “On the Relationship between Trade and Tourism.” In: Economics of Sustainable Tourism, edited by Cerina

Markandya

McAleer

New York: Routledge, pp. 92-107.

48.

Santana-Gallego

Ledesma-Rodríguez

F. J.

Pérez-Rodríguez

(2011a). “Tourism and Trade in OECD Countries. A Dynamic Heterogeneous Panel Data Analysis.” Empirical Economics, 41: 1-22.

49.

Santana-Gallego

Ledesma-Rodríguez

F. J.

Pérez-Rodríguez

(2011b). “Tourism and Trade in Small Island Regions: The Case of the Canary Islands.” Tourism Economics, 17: 1-22.

50.

Sarmidi

Salleh

N. H.

(2010). “Dynamic Inter-relationship between Trade, Economic Growth and Tourism in Malaysia.” Munich Personal RePEc Archive: Working Paper 21056.

51.

Thornton

(1997). “Exports and Economic Growth: Evidence from 19th Century Europe.” Economic Letters, 55: 235-40.

52.

UNWTO (World Tourism Organization). (2011). Tourism Highlights.

53.

WTTC (World Travel & Tourism Council). (2010). Travel & Tourism Economic Impact (Italy).