Enhancing hotel offer price forecasting with AI: The predictive power of historical and advance booking online rates

This study introduces an AI-driven time series forecasting framework to enhance the prediction of hotel offer prices, defined as the publicly listed prices posted on online travel agencies (OTAs) for future stay dates. Unlike traditional models centered on internal demand or realized average daily rates (ADR), our approach focuses on supply-side pricing decisions and leverages large-scale publicly available pricing data. Specifically, the proposed framework integrates historical posted prices and advance booking prices observed at multiple lead times, complemented by temporal, calendar-based, and event-related variables. Using more than 815 million OTA price observations from 53 hotels over four years (2019–2022), we develop a separate predictive model for each property using a walk-forward one-step-ahead autoregressive approach based on XGBoost, with periodic model recalibration to adapt to evolving market conditions. Additionally, an XAI method (SHAP) is applied to provide a post-hoc interpretation of the model's predictions by quantifying the contribution of each predictor to the forecast. Results show that both historical and advance booking prices—particularly the price from the same weekday of the previous week and the price posted for the same weekday of the following week—are among the most influential predictors of future posted prices. Moreover, based on RMSE, hotel category significantly affects forecasting accuracy, highlighting heterogeneity in pricing dynamics across properties. Overall, the findings demonstrate that publicly available historical and forward-looking offer prices effectively capture market conditions and support accurate AI-driven forecasting of hotel pricing decisions, with direct implications for revenue management systems operating in dynamic and uncertain environments.