FORECASTING HIERARCHICAL TIME SERIES USING MATHEMATICAL METHODS

Based on the company's beverage sales data in the traditional trade segment, presented in the form of hierarchical time series, models were built using a singlelevel approach (topdown, bottomup, middleout) and a multilevel approach (multilevel), namely MinTrace (Minimum Trace), which had not previously been used in business for forecasting, and compared with the basic ARIMA model with automatic parameter selection. As a result, forecasts were built for 5 months ahead until August 2025 at different levels of the hierarchy, which made it possible to improve forecasting processes, implement a new methodology, and obtain different scenarios for sales development throughout the country and at the regional level for the company.