Increasing Analysis: Which Models Perform Best Across Then Entire S&P 500

In Part I, we compared rule-based vs reinforcement learning (RL) trading strategies on AAPL using a 2-year backtest of daily data. Now, we’re scaling that idea across all 500+ tickers in the S&P 500.

What happens when we apply the same logic to the entire market?

We keep the same seven models and run a uniform backtest across all tickers. Our goal: uncover which models are consistently effective, which sectors they perform best in, and where RL models can offer an edge over traditional strategies.

Recap: The Trading Models

Insight:
Model0 (simple SMA rule) remains the most consistently high-ranking model. However, RL models like Model4 and Model1 are not far behind — showing strong adaptability in certain sectors.

Average Return by Sector

Insight:

• Financials and Industrials delivered strong returns across most models.
• Model4 (RL using only SMA) outperformed in Information Technology and Utilities.
• Model5 (Rule Mean Reversion) dominated Real Estate, but performed poorly in Consumer sectors.

Top 2 Model Finishes by Sector

Most Consistent Top Model by Sector

Sector-by-Sector Leaderboard (Best Avg Return)

Key Takeaways

• Model 0 (SMA Rule) was the most consistently successful strategy overall — with the best average return in 6 out of 11 sectors, and the highest top-2 count across all tickers.
• Model 4 (RL using only SMA) was the strongest reinforcement learning model, outperforming in sectors like Information Technology, Utilities, and Energy.
• Model 3 (RL using SMA + Volume + Day of Week) led in the Health Care sector, while Model 1 (RL with SMA + Volume) topped Communication Services.
• While rule-based models dominated overall consistency, select RL models performed best in specific industries — particularly when input features aligned well with sector characteristics.
• Simple features often outperform more complex ones unless RL is well-tuned.

What’s Next?

In Part III, we’ll visualize all of these results using a Streamlit dashboard, where you can explore model performance by sector, ticker, or metric — all with just a few clicks.

GitHub Link to notebook: GitHub