dikibagast/systematic-trading-framework

This is an open-source backtesting framework for systematic trading strategies, built in Python and focused exclusively on Bybit perpetual futures contracts. Backtesting means running a set of trading rules over historical price data to estimate how a strategy would have performed, without risking real money. The framework does not connect to any exchange for live trading and is simulation-only. The pipeline follows a fixed sequence. First, historical candle data is fetched from Bybit's public API and stored locally in Parquet files for fast repeated reads. Subsequent fetches are incremental, pulling only new candles. Then you define your entry and exit signals in Python, run a backtest against the stored data, and finally push the results through a validation pipeline designed to catch strategies that look good on paper but would fail in practice. The validation step has four parts. Walk-forward optimization splits the historical data into rolling train and test windows, then checks whether parameter settings that worked on training data also hold up on the separate out-of-sample windows. Island Volume Selection identifies parameter combinations where performance is consistently good across a range of values, rather than at a single peak that may be coincidental. Monte Carlo simulation resamples the actual trade sequence thousands of times to show the range of outcomes the strategy might produce under different orderings of the same trades. A Multi-Metric Standard then applies a pass or fail verdict based on Sharpe ratio, Sortino ratio, drawdown, profit factor, and several other measures. Speed is a priority. Signal generation and backtesting use vectorized operations on Polars DataFrames, meaning entire columns of price data are processed at once rather than looping bar by bar. A strategy covering 50,000 candles completes in milliseconds. The optimization step runs across all available CPU cores in parallel, which the documentation warns can make a computer unresponsive without a CPU limit configured. You can also describe a trading idea in plain language to an AI coding assistant and ask it to generate the required strategy files, since the framework uses a structured file format that AI tools can follow. Installation requires Python and a set of packages via pip. The project is released under the MIT license.