ENHANCING SHORT-HORIZON DEMAND FORECASTING USING SYSTEM STATE INDICATORS

Accurate short-horizon forecasts of transactional activity are important for operational planning and decision support in electronic markets.Prior studies have shown that intraday activity exhibits strong temporal regularities and can be forecast using historical demand patterns and time-of-day effects.However, such models typically ignore contemporaneous indicators describing the current state of the system.In this paper, we examine whether incorporating simple system state indicators improves minute-level demand forecasting performance in cryptocurrency markets.Using one year of high-frequency data for three highly liquid cryptocurrencies-Bitcoin (BTC), Ethereum (ETH) and Binance Coin (BNB)-we compare baseline forecasting models based on historical demand and calendar effects with augmented models that additionally include short-term variability and activity measures derived from price movements.Empirical results across linear regression, random forest and gradient boosting models show consistent reductions in forecasting error when system state indicators are included, with improvements ranging from approximately 3% to 5%.The findings suggest that even parsimonious state variables provide incremental predictive value for short-horizon demand forecasting in continuously operating electronic markets and can enhance data-driven operational decision support.