Innovative knowledge-based system for forecasting daily hotel operations amid external events using multi-source data: A time-varying parameter state-space model

Forecasting hotel occupancy during external shocks is particularly challenging due to their disruptive effects. This study develops a forecasting framework that integrates multisource data using a time-varying parameter state-space model (TVP-SSM). In this framework, search engine data (SED) are used to construct exogenous variables, intervention variables are used to reflect the severity of external shocks, and holiday and weekend dummy variables are used to capture the seasonal effect. The empirical study used a dataset from the hospitality industry in Hangzhou, China, covering the period from October 1, 2019, to October 28, 2021, and identified the COVID-19 pandemic as an external shock. The results show that TVP-SSM can effectively simulate the dynamic impact of external events and the periodical effect on hotel occupancy. Additionally, the prediction accuracy of TVP-SSM with intervention variables and periodical variables (TVP-SSM-1) exceeds that of competitive models. Specifically, compared to the naïve model and TVP-SSM without intervention variables and periodic variables (TVP-SSM-2), the prediction accuracy, measured by the root mean square error (RMSE) and mean absolute percentage error (MAPE), increased by 86 % and 87 %, respectively, and by 74 % and 76 %, respectively. These results indicate that the forecasting framework proposed in this study exhibits superior forecasting performance and demonstrates its capability for dynamic impact analysis of hotel occupancy at the industry level under external shocks.