IMPLEMENTATION OF NEURAL NETWORK METHODS FOR SHORT-TERM FORECASTING OF PRICE MOVEMENTS IN FINANCIAL MARKETS

Abstract

This study addresses the problem of short-term forecasting of order book dynamics for a financial instrument using full L2 data. Instead of traditional time series with simple numerical features, the approach relies on sequences of structured order book snapshots, each capturing market depth across 100 price levels. To facilitate this, a custom data collection tool was developed based on the BookmapAPI, enabling automatic generation of training samples with corresponding target variables, including maximum, minimum, and final bid and ask prices over a defined prediction horizon.