Abstract: The tourism industry has experienced sustained growth in recent years.However,the COVID-19 pandemic led to a sharp decline in global tourism in 2020.As the impact of the virus wanes and epidemic management becomes more standardized,the tourism sector is gradually rebounding.China,the world’s largest outbound tourism market,has significantly contributed to the global recovery of tourism through its policy approach to COVID-19 normalization.Forecasting tourist numbers enables more strategic allocation and adjustment of tourism resources,enhancing service quality.The pandemic has influenced tourism demand,making it essential to analyze these shifting patterns for the sustainable growth of the industry. Research in tourism forecasting has yielded several insights:①Tourism demand forecasts benefit from incorporating relevant external factors,such as online search indices,to boost accuracy and interpretability.②Combining forecasting methods can enhance,or at least maintain,the accuracy of single forecasts. ③Unforeseen events like COVID-19 outbreaks can compromise the accuracy of traditional methods,necessitating specialized forecasting approaches during crises.However,existing methods often overlook the varying impact of exogenous factors on tourism demand over different periods,assuming a stable relationship between explanatory and target variables even in crisis conditions.In reality,the pandemic altered tourists’ risk perceptions,which,in turn,influenced their demand and consumption preferences,impacting their travel behavior.This indicates that an explanatory variable’s effect on tourism demand may differ across pre-pandemic,pandemic,and post-pandemic periods,meaning that variables from distinct periods may need to be treated as independent new variables. This research primarily employed a Vector Autoregression model with exogenous variables (VARX),an adaptation of the Vector Autoregression (VAR) model that allows for the simultaneous analysis of endogenous and exogenous variables.The study began by gathering data on the number of Chinese outbound tourists,with a particular focus on trips to Japan,South Korea,and Singapore.Using web scraping techniques,55 Baidu search terms related to visa applications,trip planning,dining,accommodation,transportation,and shopping were collected; following correlation analysis,52 variables were retained.The classic factor model (FM) and dynamic factor model (DFM) were then used to combine the Baidu search indices into a composite indicator that retained dynamic relationships among the original multiple indicators.To account for the variable effects of exogenous factors over time,the Baidu search index was segmented into three phases—peak,trough,and recovery—based on tourists' search behaviors.Each segment served as an exogenous variable in the VARX model to forecast tourist numbers under different influences.Subsequently,a Stacking approach was applied in machine learning to combine various predictions,evaluated using RMSE,MAPE,and MASE.Finally,out-of-sample forecasts were produced to inform projections for Chinese outbound tourism. The findings indicate that VARX models effectively predict Chinese outbound tourist numbers and that accounting for the time-specific effects of exogenous variables can enhance accuracy.Key insights include:①Aggregating new indicators from numerous web search indices is effective for high-dimensional time series forecasting,with the dynamic factor model proving superior in high-dimensional contexts,reducing prediction error and improving average accuracy by 12.25% over the classic factor model.Practically,managers can compile various search indices to create a comprehensive indicator that reflects market trends and use it to forecast tourism demand.② Segmenting variables based on significant time-based changes is practical,as the segmented combination model improves accuracy over traditional models and is valuable for tourism demand forecasting under crisis impacts.As tourism normalizes,people’s behavior and choices may carry over from previous patterns with some adjustments.③ The study not only evaluates the model’s validity but also projects an increase in Chinese outbound travel in 2024,expecting it to recover to at least 60% of pre-pandemic levels. This study introduces a segmented combination forecasting approach for predicting Chinese outbound tourism.Utilizing a dynamic factor model to manage exogenous variables,it captures the original data’s multiple external factors while preserving their dynamic relationships.This method is not only relevant to tourism demand forecasting in special circumstances like pandemics but is also applicable in other contexts where data may be limited.