4.1. Data and variable measurement

To achieve the main research objective, yearly data from 1996 to 2021 were used. However, given the short data span for corruption and insecurity, Gandolfo’s interpolation procedure (1981) was applied to convert yearly data to quarterly data. This resulted in a larger number of observations (i.e., 1997:1-2020:4, n=96). The starting year (1996) and ending year (2021) were removed during interpolation. This procedure has been applied in other research involving short-span data (Abu, Obi, et al., 2019; Abu & Karim, 2021; Abu, David, et al., 2022; Abu, Karim, et al., 2022; Baharumshah & Rashid, 1999; Baharumshah et al., 2006). Data generated in this way do not produce biased estimates based on the cointegration approach (Smith, 1998; Tang, 2008). All data (except those for insecurity and corruption) were collected from the World Bank’s Development Indicators database. The data on insecurity and corruption were gathered from Nigeria’s Central Bank and the World Bank’s Governance Indicators databases, respectively.

In related research (Aktaş et al., 2014; Algieri, 2006; Alola et al., 2019; Altindag, 2014; Jimoh & Gold, 2019; Joshi et al., 2017; Lim, 1997; Saha & Yap, 2013; Xu et al., 2023; Yap & Saha, 2013), tourism receipts served as a proxy for tourism development. Tourism receipts refer to expenditures made by international tourists, including payments to national carriers for international transport. These may also comprise pre-payments on goods and/or services received in the host nation, and receipts from same-day tourists in current US$.

Corruption is captured by the World Bank’s ‘Control of Corruption’ governance indicators. Values closer to -2.5 imply less control of corruption (high corruption), and those nearer to +2.5 signify greater control of corruption (less corruption). For interpreting results, a positive corruption coefficient implies that greater control of corruption (or reducing corruption) has a positive effect on tourism development, while a negative sign would indicate a negative effect from increased corruption. Control of corruption has been used as an indicator in similar studies (Xu et al., 2023; Yap & Saha, 2013). To capture insecurity, government expenditure on internal security was used as a proxy, which is consistent with existing literature (Abu, David, et al., 2022; Adenike, 2021; Riti et al., 2017; Udoh et al., 2019). Higher government expenditure on internal security suggests greater insecurity, and in Nigeria increased security allocations are indeed commensurate with increased insecurity (and terrorism) (Riti et al., 2017). For interpreting the results, a negative coefficient of the insecurity variable indicates that greater internal insecurity (less internal security) impacts tourism development adversely, and a positive coefficient would indicate a positive impact from greater internal security.

Income was measured by GDP per capita in this study, as has been adopted in the literature (Adeola & Evans, 2019; Adeola et al., 2017; Das & DiRienzo, 2010; Luo et al., 2017; Raza & Shah, 2017; Xu et al., 2023). In addition, toeing the path of Asiedu (2002, 2004), Adeola et al. (2017), and Idowu and Bello (2010), infrastructure development was measured by access to telecommunications (proxied by the number of mobile phone subscriptions per 100 persons). Finally, the exchange rate variable expresses the exchange rate of the local currency (Nigerian Naira) to the US dollar, as in Algieri (2006), Saha and Yap (2013), Yap and Saha (2013), Aktaş et al. (2014), and Saha et al. (2017).

4.2. Stationarity and multicollinearity tests

The data employed were subjected to the unit root test to determine if they were stationary, because series must be stationary or devoid of unit root to be included in a regression exercise. Hence, the Augmented Dickey–Fuller (ADF) unit root test (Dickey & Fuller, 1979), and the Philips–Perron (PP) unit root test (Philips & Perron, 1988) were applied to check the stationarity of the variables. These two commonly used tests were chosen for complementarity (Hossain & Hasanuzzaman, 2013). The non-linear ARDL used in this research requires the dependent variable to be integrated to order one (1), while the order of integration of the explanatory variables should be zero (0) and/or 1, but not 2 (Cheah et al. 2017; Ibrahim, 2015). A multicollinearity test was also carried out on the variables via the Variance Inflation Factors (VIF) approach. A VIF value of 10 or above suggests a problem of multicollinearity.

4.3. Non-linear ARDL

While most empirical research uses linear estimation techniques, the notion of non-linearity between time series and/or economic variables has been gaining ground and with it the use of econometrics such as non-linear ARDL (Abu, Karim, et al., 2022; Shin et al., 2014). Cointegrated positive and negative components of time series indicate hidden cointegration, which implies a case of non-linear cointegration (Granger & Yoon, 2002; Katrakilidis & Trachanas, 2012). Non-linear ARDL (NARDL) is well suited to this research because it recognizes hidden cointegration and can estimate both short- and long-run asymmetries (Meo et al., 2018; Zangina & Hassan, 2020).

Therefore, to explore the asymmetric influence of corruption and insecurity on tourism development, the NARDL cointegration approach (Shin et al., 2014) which is the asymmetric or non-linear extension of the linear ARDL method (Pesaran & Shin, 1999; Pesaran et al., 2001), was adopted in this research. Compared to other cointegration methods such as the residual-based technique (Engle & Granger, 1987) and the maximum likelihood method (Johansen, 1988; 1991; Johansen & Juselius, 1990), NARDL has several advantages. It can detect the presence of cointegration between series integrated to order one [I(1)], or a mixture of I(1) and order zero [I(0)]; it is more appropriate for analyzing small samples; it allows variables to possess different lags; and it uses a single-equation model to estimate short- and long-run parameters simultaneously. The Equation capturing the non-linear (asymmetric) cointegrating or long-term relationship among corruption (COR), insecurity (LINS) and tourism development (LTOUR) including the control variables such as income (GDPCG), infrastructure (INFR ) and exchance rate (LEXC) is expressed as:

where L denotes the logarithm introduced to reduce skewness and convert the absolute values of the variables to identical scale. All variables are as defined earlier. β⁺, β^-, σ⁺, σ^-, α₀, α₁, α₂, and α₃, denote long-term coefficients. COR_t and LINS_t refer k x 1 vectors of regressors, decomposed as:

And

$\text{CO}{\text{R}}_t^{+}$ and $\text{CO}{\text{R}}_t^{-}$ refer to the partial sums which correspond to both positive and negative changes (shocks) in COR_t, and are generated by calculating:

Similarly, $\text{LIN}{\text{S}}_t^{+}$ and $\text{LIN}{\text{S}}_t^{-}$ denote partial sums that correspond to both positive and negative changes (shocks) in COR_t, and are generated by calculating:

In line with Shin et al. (2014), the non-linear form of Model 1, NARDL (p,q) that portrays asymmetric relation amongst tourism development, corruption and insecurity (including the control variables such as income, infrastrucuture and exchange rate) is expressed as:

where ${\theta }^{+}=-p{\beta }^{+}$ and ${\theta }^{-}=-p{\beta }^{-};{\gamma }^{+}=-p{\sigma }^{+}$ and ${\gamma }^{-}=-p{\sigma }^{-}.p$ and q represent lag orders.

The values of p and q were selected using the Akaike information criterion (AIC) and the value of q can be different for each variable. Certain steps were taken to implement the non-linear ARDL method. First, NARDL(p,q) Model 9 was estimated via conventional Ordinary Least Squares (OLS). Next, a non-linear (asymmetric) cointegration test was conducted to check for long-term relations among $\text{LTOU}{\text{R}}_t$, $\text{CO}{\text{R}}_t^{+}$, $\text{CO}{\text{R}}_t^{-}$, $\text{LIN}{\text{S}}_t^{+}$, $\text{LIN}{\text{S}}_t^{-}$, GDPCG_t, INFR_t and LEXC_t. The null hypothesis of no cointegration among the variables, captured by: ρ = θ⁺ = θ^- = Υ⁺ = Υ^- = α₂ = α₃ = α₄ = 0 in Model 9, was tested by applying the Pesaran et al. (2001) bounds test based on the modified F-test. The F-statistic generated was compared to upper [I(1)] and lower [I(0)] critical bounds values. Once cointegration was ascertained, long-term and short-term asymmetric relation tests were explored using the standard Wald test. The long-run asymmetry test was validated by testing the following null hypotheses:

and

Similarly, the short-run asymmetry test was conducted by testing these null hypotheses:

and

If the short-term (or long-term) coefficient of the positive partial sum of corruption ( $\text{CO}{\text{R}}_t^{+}$) varies from the negative partial sum of corruption ( $\text{CO}{\text{R}}_t^{-}$), then, it is established that corruption (COR) has an asymmetric short-run (or long-run) influence on tourism development (LTOUR). Likewise, if the short-term (or long-term) coefficient of the positive partial sum of insecurity ( $\text{LIN}{\text{S}}_t^{+}$) varies from the negative partial sum of insecurity ( $\text{LIN}{\text{S}}_t^{-}$) then an asymmetric short-run (or long-run) impact of insecurity (LINS) on LTOUR is confirmed.

4.4. Diagnostic and stability tests

After model estimation, the results were subjected to diagnostic and stability tests to check the reliability or validity of the estimates. We used the Breusch-Godfrey (BG) serial correlation test to determine if the residuals were correlated, and the Breusch-Pagan-Godfrey (BPG) heteroscedasticity test to ascertain whether the residuals were homoscedastic. Tests for the cumulative sum of recursive residuals (CUSUM) and the cumulative sum of squares of recursive residuals (CUSUMQ) were also applied to check the stability of the model and parameters.

5.1. Results of unit root and multicollinearity tests

The unit root test results (Table 2) signify that all the variables had a unit root at level but became stationary after the first differencing. Thus, they were integrated to I(1), thereby justifying application of the NARDL estimation method. A multicollinearity test was also conducted on the variables using the variance inflation factors (VIF) approach, and the results are reported in Table 2. The centered VIF values for the variables ranged from 1.4910 to 5.1676. Individual VIF value was less than 10, indicating that multicollinearity was not a problem.

3.2. Results of cointegration and asymmetry tests

The outcome of the bounds test provides evidence that tourism, insecurity and corruption (along with income, exchange rate and infrastructure) have a cointegrating relation. The derived value of the computed F-statistic (7.0569) was larger than the upper critical bound value of 3.90 (Table 3, Panel A), indicating a long-term relationship among the variables.

Also, the outcome of the asymmetric test showed long-term asymmetry between the corruption-tourism and insecurity-tourism pairs (Table 3, Panel B). The calculated F-statistic of 16.5746 and a probability of 0.0000 imply asymmetry between corruption and tourism, and the computed F-statistic of 8.2783 with a probability of 0.0006 indicate asymmetry between insecurity and tourism.

Moreover, the long-run coefficient of a positive shock to control of corruption (COR⁺) was positive and significant, while the coefficient of a negative shock to control of corruption (COR^-) was positive and insignificant. Finally, the long-run coefficient of a positive shock to insecurity (LINS⁺) was negative and significant, but the coefficient of a negative shock to insecurity (LINS^-) was positive and significant. These results supply evidence of long-run asymmetry between corruption and tourism development, as well as between insecurity and tourism development.

5.3. Results of NARDL model estimation

According to the NARDL results (Table 4), the optimal lag length selected by the AIC was: 3,0,0,1,0,4,0,1. The results revealed that a positive shock to control of corruption (reducing corruption) promoted long-term tourism development, but a negative shock (increasing corruption) did not influence long-term tourism development significantly. Increasing control of corruption by one unit enhanced tourism development by 0.11% at the 1% level over the long-term, keeping other variables constant. In addition, a negative shock to expenditure on internal security (lower internal insecurity or greater internal security) raised long-term tourism development, while a positive shock to expenditure on internal security (greater internal insecurity) reduced long-term tourism development. Raising internal insecurity by 1% lowered long-term tourism development by 1.51% at the 5% level, but lowering insecurity by 1% raised long-term tourism development by 4.88% at the 1% level, holding other variables constant. Furthermore, the exchange rate was positively connected to long-term tourism development and the influence was significant. Depreciation of the domestic currency by 1% raised long-term tourism development by 0.57% at the 1% level, keeping other variables constant. However, income and infrastructure seemed insignificant in dictating the long-term behaviour of tourism.

The short-term results (though not the focus of this research) signify that a 1% decrease in insecurity raised short-term tourism development by 0.35% at the 1% level, holding other variables constant. Also, raising income by 1% led to a 0.16% increase in short-term tourism development at the 1% level, keeping other variables constant. The error correction coefficient was negative and significant at 0.20, suggesting that 20% of the deviation from equilibrium was corrected in the fourth quarter of the year.

Some implications can be drawn from these findings. First, the supportive role of less (or higher control of) corruption on long-term tourism development lends credence to the research of Osinubi et al. (2021) on Nigeria, Jimoh and Gold (2019) for the ECOWAS region, and others such as Das and DiRienzo (2010), Yap and Saha (2013), Poprawe (2015), Chulaphan and Barahona (2021), and Xu et al. (2023). Thus, reducing corruption implies, for example, that incoming or potential tourists would not have to pay bribes to obtain visitor’s permit or facilitate visa processing to visit Nigeria. The lower costs to prospective visitors to Nigeria will lead to higher inflow of tourists and tourism receipts, and development of the sector.

The negative influence of rising insecurity on long-term tourism development lends credence to the research of Riti et al. (2017), Alola et al. (2019) and Awode (2022), who found an adverse impact of insecurity on tourism development in Nigeria, and that of Jimoh and Gold (2019) for the ECOWAS region. Other studies that have uncovered a negative impact of insecurity and related activities on tourism include Pizam and Fleischer (2002), Altindag (2014), Buigut and Amendah (2016), Parida et al. (2016), Saha et al. (2017), Habibi (2017), Joshi et al. (2017), Mariyono (2017), Fourie et al. (2020), and Santana-Gallego and Fourie (2020). Thus, higher uncertainty about individual safety as a result of increased insecurity from terrorism, violent attacks, bombings, killings, kidnappings and similar activities discourages prospective tourists from travelling to (or visiting) Nigeria, leading to lower receipts and a decline in the sector’s development.

The positive relation between long-term tourism development and the exchange rate coincides with findings by Zhang et al. (2009), Yang et al. (2010), Yap and Saha (2013), Saha and Yap (2013), Altindag (2014), and Saha et al. (2017). Thus, as Nigeria’s domestic currency depreciates, it lowers travel costs for tourists willing to visit the country. This would lead to higher inflows of international travellers or tourists, more tourism receipts, and greater long-term tourism development.

The insignificant influence of infrastructure on long-term tourism development supports the findings of Phakdisoth and Kim’s (2007) in Laos, and Awode (2022) in Nigeria. Similarly, Riti et al. (2017) reported an insignificant and negative long-term impact of infrastructure on tourism in Nigeria. This may be connected to poor government funding of infrastructures (roads, communications, electricity, etc.) over time. The already inadequate funds have also been misappropriated or embezzled by corrupt government officials (Riti et al., 2017). The insignificant impact of income on long-term tourism development corroborates the research of Hanon and Wang (2020), who found income to be insignificant and negatively related to tourism in Syria, and similar outcomes elsewhere (Leitão, 2009; Yang et al., 2010). These findings demonstrate Nigeria’s poor economic performance and the incapacity of its government to invest in and build tourism infrastructures or maintain existing ones.

5.4. Results of diagnostic and stability tests

The results were subjected to conventional post-estimation (diagnostic) tests such as the Breusch-Godfrey (BG) serial correlation test and the Breusch-Pagan-Godfrey (BPG) heteroscedasticity test, to ascertain their validity. Table 5 shows that the chi-square ( ${\chi }^2$) and probability value (ρ-value) for the BG serial correlation test were 6.2953 and 0.1782, respectively. Also, the ${\chi }^2$ and ρ-value for the BPG heteroscedasticity test were 13.6916 and 0.6217, respectively. Therefore, the results are devoid of serial correlation and heteroscedasticity problems and can be considered reliable.

The stability of the estimated coefficients and model was evaluated by the cumulative sum of recursive residuals (CUSUM) and the cumulative sum of squares of recursive residuals (CUSUMQ) tests. The results of these tests, shown in Figure 2 and Figure 3, imply that both the model and the coefficients estimated are stable because the plots remain within the lower and upper boundaries. Hence, the findings can guide policy formulation.

Figure 2Plots of CUSUM 

Figure 3Plots of CUSUMQ