Beating the Benjmark is a Warwick AI Society project undertaken by Jamie Guo and Aaron Moorby which explores combining AI Techniques and traditional portfolio theory to construct an optimal portfolio for a one-month timeline.

The goal of the project was to outperform the Warwick AI Society Pet Fish, Benji:

• Benji makes trades by swimming, swimming to the left of the tank is a sell order, whilst swimming to the right is a buy order.
• Inspired by Michael Reeves Viral Video: https://www.youtube.com/watch?v=USKD3vPD6ZA

We won this competition and made a 6.57% return on the month compared to Benji's 0.45%.

1. Grab enough data
   • The model uses 12 years of daily financial data from yfinance to capture meaningful market patterns.
   • Fun fact: Jamie got banned from yfinance for requesting too much data at one point.
2. Extract and process signals
   • Key features include:
     • dist_sma200 – tells the model if an asset is generally trending up or down (long-term trend).
     • efficiency_ratio – measures how reliable a trend is, helping the model ignore noise.
   • Features were scaled using a RobustScaler to reduce the impact of outliers.
3. Use an LSTM to forecast returns
   • Two-layer LSTM predicts one-month returns based on the processed features.
   • A custom loss function punishes the model for incorrect direction.
4. Black-Litterman Portfolio Optimisation
   • Expected returns and their RMSE values are fed into a Black-Litterman framework.
   • Produces a risk-adjusted portfolio that accounts for uncertainty in the LSTM predictions.