Data-Driven Modeling of Tourism Demand in an Emerging Mediterranean Destination: A Google Trends Network, Granger Causality, and Forecasting Framework

Abstract

This study proposes an integrated, data-driven systems framework for characterizing and forecasting tourism demand dynamics in Vlora, Albania, an emerging Mediterranean destination. Using 61 monthly observations from February 2021 to February 2026 and 15 tourism-related Google Trends search terms across four thematic attraction dimensions, the framework integrates correlation-network analysis, de-seasonalization, false discovery rate (FDR)-corrected Granger causality, null-model benchmarking, Network Vector Autoregression (Network-VAR), Transfer Entropy, and a newly developed Structural Demand Integration (SDI) Index. The raw network exhibits high connectivity (density = 0.835) and differs significantly from random co-movement (p = 0.001). De-seasonalization reduces 91% of network edges and reveals a robust six-node cross-category structural core, while Kanina Castle becomes fully isolated. Following FDR correction, 18 of 210 Granger relationships remain significant, identifying Gjipe Beach, Sazan Island, and the Bay of Grama as principal demand-leading nodes. Transfer Entropy further supports these directional relationships. The SDI Index identifies St. Mary’s Monastery, restaurants in Vlora, and Sazan Island as the most strongly integrated tourism assets. Network-VAR forecasting with Granger-leading nodes significantly outperforms ARIMA (Diebold–Mariano p = 0.0019). Validation against official INSTAT arrival statistics confirms a strong association with physical tourism demand, including after de-seasonalization (r = 0.82). The framework provides a transferable systems-analytics approach for tourism-demand analysis and forecasting in data-scarce Mediterranean destinations.