The project focuses on developing an algorithmic trading model for the Russian market using CVaR (Conditional Value at Risk) to minimize losses during market shocks. The system is designed to adapt to different market regimes and control tail risks while maintaining performance.
Data processing relies on an ETL pipeline that cleans raw exchange data. This includes synthetic back-calculation of the LQDT money market fund price to cover the period before its 2022 launch, using MOEXREPO index data dating back to 2013.
The system addresses historical data anomalies such as stock splits and de-listings. It uses a modified Parkinson estimator based on daily highs, lows, and closing prices to measure asset risk, replacing standard deviation for better sensitivity to market volatility.
The simulation framework includes an orchestrator and a backtester to prevent look-ahead bias. It accounts for transaction costs, market drift of asset weights, and a 15-day lag for dividend reinvestment to reflect real-world settlement processes.
Validation involves an 11-year historical track from 2015 to 2026 across three investment paradigms: lump-sum, dollar-cost averaging, and FIRE. Additionally, a Monte Carlo stress test performs 250 runs on subsets of 70 assets, including bankrupt companies.
