University of Texas Arlington
Jan 2021 - Feb 2023
Research Assistant
Arlington, TX
Exploring Deep Learning in Finance
Designed a state-of-the-art attention-based LSTM transformer encoder to capture complex dependencies in financial market signals. This architecture improved the model’s ability to forecast market movements by focusing on the most influential features in time series data, enhancing predictive performance for market forecasting applications.
Created a robust data sampling strategy using dollar volume and event-based triggers to ensure that the most relevant data points were selected for training the model. By focusing on high-value trades and significant market events, this approach reduced noise and improved model accuracy, allowing for more reliable predictions of market trends and events.
Integrated the Triple Barrier methodology to label market signals dynamically, providing clear buy, hold, or sell recommendations. This framework optimized trading positions, balancing risk and reward, and helped in maximizing risk-adjusted returns. As a result, profitability improved significantly by adapting to market volatility and managing downside risk effectively.
Spearheaded the development of a custom neural network in PyTorch and utilized Optuna for hyper-parameter optimization, which fine-tuned the model's performance. This process led to a 60% reduction in training time and a 9% increase in prediction accuracy, improving the efficiency of the model and enabling faster iteration and deployment.
