Financial Resilience of Tourism-Dependent Economies under Global Uncertainty: A Panel Explainable Machine Learning Approach

Özet

Purpose – Tourism-dependent economies are particularly exposed to persistent global uncertainty, yet tourism resilience and macro-financial resilience have largely been examined as separate phenomena. This study develops a Tourism Financial Resilience Index (TFRI) and investigates how tourism dependency, financial robustness, and global uncertainty are associated with tourism financial resilience across countries.

Design/methodology/approach – The study employs an unbalanced panel of 144 tourism-dependent countries over the 2010–2020 period, comprising 1,533 country-year observations. The Tourism Financial Resilience Index and Financial Robustness Index are constructed using Principal Component Analysis (PCA). Country fixed-effects estimation with Driscoll–Kraay robust standard errors is combined with Panel Quantile Regression to examine conditional-mean and distributional associations. Random Forest, XGBoost, and LightGBM models, interpreted using SHapley Additive exPlanations (SHAP), are further employed to explore nonlinear predictive patterns and model-based predictor importance.

Findings – Global uncertainty exhibits a negative and statistically significant association with Tourism Financial Resilience in the fixed-effects model and across all estimated conditional quantiles. Tourism dependency and financial robustness are not statistically significant in the baseline fixed-effects specification but become positive and significant across the 25th, 50th, and 75th quantiles, indicating that their relationships with resilience are specification-dependent. The interaction between global uncertainty and financial robustness is negative and significant in the fixed-effects model; however, its sign does not support the hypothesized buffering effect of financial robustness, and neither moderation effect is significant in the quantile regressions. The machine-learning models exhibit limited out-of-sample generalization, with negative test R² values, and are therefore interpreted primarily as complementary tools for examining nonlinear predictive structure. Within the LightGBM model, SHAP identifies global uncertainty as the most influential predictor, followed by GDP per capita and tourist-arrival intensity.

Discussion – The findings demonstrate that Tourism Financial Resilience is shaped by the interaction of tourism exposure, macro-financial conditions, and persistent global uncertainty rather than by tourism specialization alone. The study contributes to the tourism-resilience literature by integrating tourism and financial resilience within a multidimensional framework and by combining panel econometric inference with explainable machine learning while maintaining a clear distinction between statistical association, predictive performance, and model-based variable importance.