Causal Dynamics Between Tourism Receipts, Economic Growth, and Real Exchange Rate in Algeria’s Economy Between 1990 and 2019

Previous studies on tourism receipts and economic growth and exchange rate have had different areas of interest and different focal points. Some studies investigated the relationship between tourism and other variables in a single country or within a cross-section of countries.

Ohlan (2017) studied the relationship between tourism, financial development, and economic growth in India during the period from 1960 to 2014. The results indicated the relationships between the variables of the study in the short and long term. One-way causal relationships were found between the study variables; however, the study neglected some variables that affect economic growth, which provided better results for the study.

Kumar and Stauvermann (2016) examined linear and nonlinear relationships between tourism and growth. A 1% increase in tourism receipts was found to increase output by 0.10% in the long term. A long-term U-shape was detected, with minimum tourism receipts consisting 1.26% of the gross domestic product (GDP). The causality results indicated that a higher level of tourism receipts causes growth. This study neglected some crucial variables that could have produced more wide-ranging results.

Liu et al. (2016) studied the dynamic relationship between international tourism, economic growth, and energy consumption in Taiwan during the period between 1965 and 2010. They used Granger causality¹ analysis to examine the causal relationships among international tourism development, economic processes, and energy consumption in Taiwan during that period. The results indicated a two-way causal relationship between economic growth and energy consumption, and between the development of international tourism and energy consumption. No reciprocal causal relationship was found between economic growth and international tourism development. This study separated the study variables from each other, as it was better to estimate the model with three variables; however, the results might have been better without separating the variables.

Katircioǧlu (2011) studied tourism and growth in Singapore between 1960 and 2007. The results confirmed a long-term balance of the relationship between international tourism and economic growth in the case of Singapore, and real income and growth converged. The level of long-term equilibrium significantly increased, by 51.4%, in the tourism-led growth (TLG) hypothesis model. The main conclusion of this study was that the TLG hypothesis is specific for the long-term economy of Singapore, as was determined by conditional causal tests. However, this study did not include control variables that are important for research.

The study of Belloumi (2010) on the relationship between tourism revenues, real exchange rate, and economic growth in Tunisia during the period from 1970 to 2007 concluded that there was a correlation between tourism and economic growth. Additionally, the results of the Granger causality test indicated that tourism had an indirect positive impact on GDP growth. The results of this study were focused on tourism revenues and economic growth and did not explain the outcome of the real exchange rate.

Kreishan (2010) studied tourism and economic growth in Jordan from 1970 to 2009. The results of the Granger causality test revealed a one-way causality from tourism revenues to economic growth. The results of this study indicated that the government should focus on economic policies to promote international tourism as a potential source of economic growth in Jordan.

Balaguer and Cantavella-Jordá (2002) studied tourism as a long-term economic growth factor in Spain and concluded that economic growth in Spain, for at least the past three decades, was reasonable for the continued expansion of international tourism. This increased activity has doubled tourism’s effects over time. External competitiveness was also found to be a crucial variable for Spanish economic growth.

Arslanturk et al. (2011), studying the Turkish economy during the period between 1963 and 2006, concluded that there was no causal link between tourism revenues and economic growth. Over time, economic growth did not have a predictive capacity for tourism revenues.

Boğa and Erkişi (2019) studied the relationship between reception of international tourism and economic growth in Asia-Pacific countries during the period from 1995 to 2017. Their results indicated that there was a bilateral causal relationship between tourism receipts and economic growth in the short term. The authors hypothesised that there was a two-way relationship between international tourism receipts and economic growth. This study neglected the control variables that affect the model.

Wu and Wu (2018) studied the interrelationship between world tourism and the growth of the Chinese economy in 12 western regions in China from 1995 to 2015 and found a causal correlation between the two variables in seven of those regions, whereas the remaining five regions affected only one variable. The heterogeneity of the study areas affected those results.

Tugcu (2014) analysed the state of the Mediterranean region between 1998–2011 and found a causal relationship between tourism and economic growth. European countries were able to generate growth better than tourism in that region. The study’s results differed between Europe, Asia, and Africa.

Lee and Brahmasrene (2013) found a long-term equilibrium relationship between study variables. Tourism, carbon dioxide emissions, and foreign direct investment (FDI) had a positive and robust impact on economic growth. The latter, in turn, showed a positive and robust impact on CO₂ emissions during tourism, and FDI had a robust negative effect on CO₂ emissions.

The survey of previous studies revealed that the topic of the relationship between tourism receipts, economic growth, and the real exchange rate in the period from 1990 to 2019 was not examined previously. Especially in light of the COVID-19 pandemic, examining this gap can provide valuable insights to decision-makers in the Algerian economy and academic researchers in this field.

To answer the research problem and achieve the desired objectives, I proposed the following set of hypotheses:

Before constructing the model, I collected data regarding the study variables from the World Bank (n.d.) database (WDI) and the Algerian National Bureau of Statistics (ONS). I identified the dependent variables, the explanatory variables, and the expected impact of each variable; the variables are summarised in Table 1.

Table 2 shows the most important statistical indicators for the variables used in the model for the period from 1990 to 2019; that is, the indicators found to be highly acceptable to the nature of this study within 30 observations.

In Table 3, which represents the correlation matrix between study variables, there is a correlation between those variables. The accuracy of the model is increased, using the best unbiased linear capabilities.

The ARDL model methodology was used, as one of its characteristics is that it can be applied regardless of whether the chain is stable at the level or integrated from the first order. However, it cannot be integrated from the second order.

Based on economic theory and empirical models used in previous studies on the same topic, the equation was estimated to measure the impact of tourism revenue, economic growth, and the real exchange rate in Algeria during the period of 1990 to 2019.

After testing for the unit roots, the subsequent step consisted of investigating the long-term relationships between the variables. Using the ARDL bounds testing approach, the ARDL representation between the competing variables can be described as: 1 LNTOURt=α0+α1LNGDPGt+α2LNREERt+εt 2 LNGDPGt=α0+α1LNTOURt+α2LNREERt+εt 3 LNREERt=α0+α1LNTOURt+α2LNGDPGt+εt

Where LNTOUR_t is the logarithm of international tourism receipts in Algeria at time t, LNGDPG_t is the logarithm of GDP growth in Algeria at time t, LNREER_t is the logarithm of the real effective exchange rate in Algeria at time t α_i is the parameters, and ɛ_t is the error term.

The ARDL bounds testing approach to cointegration was used as an estimation technique in the current empirical research. The advantages of this technique, developed by Pesaran et al. (2001), can be divided into three significant aspects. First, it enables analysis of both short- and long-term associations between dependent and independent variables without regard to the order of integration of the regressors, provided it does not exceed the order of one I(1) (Pesaran, 1997; Pesaran et al., 2001). Second, this approach is well-suited to small sample sizes because it provides a reliable outcome (Pesaran & Shin, 1998). Third, under this technique, variables can have different optimal lagged periods (Pesaran et al., 2001). Fourth, the long-term coefficients are unbiased even in the case of the endogeneity of the regressors.

Hence, according to Pesaran et al. (2001), Equations 1, 2, and 3 were tested by estimating the unrestricted error correction model (UECM) as follows: 1.a ΔLNTOURt=ϑ0+ϑ1LNTOURt−1+ϑ2LNGDPGt−1+ϑ3LNREERt−1+∑i=0pρ1iLNTOURt−1+∑j=0qρ2jLNGDPGt−1+∑j=0qρ3jLNREERt−1+εt 2.a ΔLNGDPGt=ϑ0+ϑ1LNGDPGt−1+ϑ2LNTOURt−1+ϑ3LNREER t−1+∑​i=0pρ1iLNGDPGt−1+∑​j=0qρ2jLNTOURt−1+∑​j=0qρ3jLNREERt−1+εt 3.a ΔLNREERt=ϑ0+ϑ1LNREERt−1+ϑ2LNGDPGt−1+ϑ3LNTOURt−1+∑​i=0pρ1iLNREERt−1+∑​j=0qρ2jLNGDPGt−1+∑​j=0qρ3jLNTOURt-1+εt

Where Δ is the first difference operator, ϑ₀ is the constant intercept, μ_t is Gaussian white noise, the parameters ϑ₁ − ϑ₃ are the long-term coefficients, the parameters ρ₁ − ρ₃ represent the short-term parameters, and (p, q) represent lag order on the regression variables.

Even though the ARDL bounds testing approach to cointegration requires no pretesting for unit roots, we conducted augmented Dickey–Fuller (ADF; Dickey & Fuller, 1981; Said & Dickey, 1984) and Phillips–Perron (PP; Phillips & Perron, 1988) unit root tests to determine the level of integration of variables, ensure that none of the variables is I(2) or beyond and, therefore, provides justification for the suitability of the model estimation technique used here.

The second step involved examining the existence of a long-term association among the variables stated in our models based on the F-bound test procedure. To this end, the null hypothesis of no cointegration to be tested in Equations 1.a, 2.a, and 3.a was H0: ϑ₂ = ϑ₃ = 0, and the alternate hypothesis of the presence of cointegration was H1: ϑ₂ ≠ ϑ₃ ≠ 0. The computed value of F statistics was compared with the upper and lower critical values proposed by Pesaran et al. (2001).

In the third stage, and after establishing the existence of long-term cointegration, long-term coefficients were estimated by applying the long-term ARDL models. The error correction model (ECM) was estimated to determine the short-term coefficients and parameters of the short-term speed of adjustments to the long-term equilibrium.

Granger causality was used to determine the direction of the relationship between the study variables, as there were variables that were affected by each other and there were one-directional relationships. This varied from one economy to another (Engle & Granger, 1987; Granger, 1988).

The existence of cointegration between series confirmed that there should be at least one causal relationship, but it failed to give the relationship’s direction. Hence, we followed the well-known procedure developed by Engle and Granger (1987) to examine the short- and long-term causal dynamics between the competing variables. Following Engle and Granger, a vector error correction model (VECM) was used for testing the Granger causality among tourism receipts, economic growth, and real exchange rate. The method can be written as follows: 1.b ΔLNTOURt=b10+∑i=0ρ1ρ1iLNTOURt−1+∑i=0q1ϑ1iLNGDPGt-1+∑i=0r1ϕ1iLNREERt−1+ω1+ε1t 2.b ΔLNGDPGt=b20+∑​​i=0p2ρ2iLNTOURt−1+∑​​i=0q2ϑ2iLNGDPGt−1+∑​​i=0r2ϕ2iLNREER t−1+ω2+ε2t 3.b LNREERt=b30+∑​i=0p3ρ3iLNTOURt−1+∑​i=0q3ϑ3iLNGDPGt−1+∑​i=0r3ϕ3iLNREERt−1+ω3+ε3t

Where b_j0, ρ_j, ϑ_j, and φ_j (j = 1, 2, 3) are the parameters to be estimated; ɛ_jt (j = 1, 2, 3) is the white noise error term; ECT is derived from the corresponding long-term equilibrium relationship; and the coefficients ω_j (j = 1, 2, 3) of the ECTs represent the deviation of the dependent variables from the long-term equilibrium.

The ECM allows testing for the existence of Granger causality in three possible ways (Sebri & Abid, 2012). First, the short-term Granger causality was investigated by testing the significance of the sum of lagged differences of explanatory variables by using the partial F statistic. Second, the long-term causality was checked by examining the coefficients of the ECT_t-1 based on t statistics. A long-term Granger causality exists if this coefficient is negative and statistically significant. Finally, strong Granger causality, which means that the two sources of causality are jointly significant, can be exposed by testing the joint hypothesis through the joint F test on both ECT_t-1 and the sum of lagged differences of the explanatory variables.

Based on Table 8, there was a causal relationship between the variables of the study: Both the tourism receipts and real exchange rate affected the economic growth one-directionally. Economic growth and the real exchange rate did not affect the tourism receipts in Algeria. That is, there was no causal link in this direction. Tourism receipts and economic growth did not affect the real exchange rate.

In ARDL models, it is advisable to use structural stability testing and diagnostic tests.

4.3.2

Structural Stability Test for the Estimated ARDL Model

The stability of the estimated coefficients in the model was also proven using CUSMUS and CUSMUSQ stability tests, as detailed by Brown et al. (1975). Figure 1 shows that the cumulative residual starvation test (CUSUM) expressed a linear medium within the boundaries of the critical region, indicating a pattern of stability in the model at 5% in both the long and short terms.

Figure 1:

The same is true for the cumulative sum test for the residual follow-up boxes (CUSUMSQ), as shown in Figure 2.

Figure 2: