Skip to content

Victor's Project Details

Source: Notion | Last edited: 2025-03-25 | ID: 1c12d2dc-3ef...


SR&ED Claim – Victor’s Work on Financial Time Series Forecasting and Algorithmic Trading Enhancements

Section titled “SR&ED Claim – Victor’s Work on Financial Time Series Forecasting and Algorithmic Trading Enhancements”

Existing Technologies and Their Limitations

Section titled “Existing Technologies and Their Limitations”

The financial forecasting and algorithmic trading space contains many well-established tools, each with significant limitations. Key challenges include:

  • Existing reinforcement learning libraries lack industry-specific customizations for high-frequency financial market data. Mean reversion strategies often fail to perform in volatile markets and during regime changes. Data collection, storage, and real-time adaptation present significant challenges, especially when scaling latency-sensitive trading algorithms.

Victor investigated and integrated the Route28 API for dynamic data acquisition and trade execution. Route28, a Synthetic Institutional Exchange developed by Crossover Markets and Hidden Road Partners, enables efficient trading of perpetual swaps without a Central Limit Orderbook (CLO). This non-reliance on CLO could significantly improve trade efficiency through better slippage control and price optimization.

The project also incorporated TensorTrade, a reinforcement learning library, to enhance the trading models’ predictive accuracy and adaptability. Testing alternative mean reversion strategies, specifically customized for various market conditions, formed a crucial part of this development.

Systematic Investigation & Experimental Development

Section titled “Systematic Investigation & Experimental Development”

Victor enhanced the trading models through:

  • API Integration: Implemented Route28 API for efficient data access and model testing. Testing Reinforcement Learning Models: Optimized TensorTrade algorithms for real-time predictions. Scenario Testing: Thoroughly tested mean reversion strategies to improve market predictions. Backtesting and Validation: Evaluated new strategies against Eon Labs’ trading goals using historical data. Data Synchronization: Resolved real-time data acquisition challenges during live environment transition. Model Stability: Fine-tuned reinforcement learning models for stable real-time execution. Latency Issues: Optimized execution speed while maintaining predictive accuracy. Integration with Existing Systems: Successfully merged new models and APIs with legacy systems.

Several approaches proved inadequate:

  • Insufficient Model Adaptability: Early models lacked rapid adaptation to market changes. Overcomplicated Data Structures: Initial complex structures reduced efficiency, requiring simplification. Inefficient Real-Time Integration: Early Route28 API implementation suffered from latency and update issues. Data Collection and Handling: Standard methods failed to meet high-frequency requirements, leading to custom solutions. These setbacks led to improved strategies focusing on model adaptability, streamlined data structures, and smooth live trading integration.