Forecasting Post-Pandemic Tourist Arrivals in Himachal Pradesh: A SARIMAX Intervention-Analysis

Abstract

This study forecasts total, domestic, and international tourist arrivals in Himachal Pradesh, a Himalayan state in northern India dependent on tourism. Using 132 months of data (January 2015–December 2025), the analysis addresses two major disruptions in the series: the COVID-19 collapse (February 2020–December 2021) and the July–August 2023 flood event. An intervention-adjusted Seasonal Autoregressive Integrated Moving Average with eXogenous regressors (SARIMAX) model is developed and compared with a Holt-Winters exponential smoothing (ETS) model and a standard seasonal ARIMA model without intervention variables. Stationarity tests confirm that the total-arrivals series requires first-order differencing. A SARIMAX model with COVID and flood dummy variables is selected using the Akaike Information Criterion and validated against 2025 as a held-out test year. The intervention-adjusted model outperforms the alternatives across all evaluation metrics (RMSE, MAE, MAPE, and Theil's U), and its residuals show no significant autocorrelation (p > 0.05). The flood variable is a statistically significant negative predictor (p < 0.01), while the COVID variable, despite its large economic impact, and is not statistically significant, as differencing already accounts for much of the shock. The validated model is re-estimated on the full dataset and forecast forward 36 months, projecting a gradual, seasonally uneven recovery, with annual arrivals reaching approximately 16.1, 17.1, and 18.1 million in 2026, 2027, and 2028—still below the 2017 pre-pandemic peak of 19.6 million. This study offers destination management organizations a practical, evidence-based forecasting tool and highlights the value of intervention time-series methods for mountain tourism economies facing recurrent shocks.