Building New Quant Models With GPT 5

This week, OpenAI released GPT-5 for general public use. I’ve spent the past several days putting it through its paces — running fresh backtests, refining concepts, and improving how I identify broad market pivots. My initial impression: it’s powerful. GPT-5 handles complex rule sets far better than earlier versions, though it’s not flawless. It will still make mistakes, so it’s critical to review its outputs carefully.

One pro tip from experience: ask GPT-5 for any intermediate files it creates before you close out a session. These models generate a lot of working files in the background, and once the session ends, they’re gone. If you don’t save them, you’ll have to start from scratch.

One of the new tools I’ve been testing is a Trough Recovery Score. The idea came from the challenge of dealing with over a decade of backtest data, covering about six key variables in the signal. The combinations are enormous, which means that for some setups, there may only be a handful of identical historical cases to reference. The scoring system solves that problem by creating a more generalized, composite measure of favorable “trough to recovery” conditions.

It works surprisingly well. GPT-5 helped me design the scoring and backtesting methodology in a way that balances robustness with the need to avoid overfitting. The result is a metric that captures the bulk of what my experience tells me to look for — but with historical probability data behind it.

Right now, the Trough Recovery Score for the current market stands at 80 out of 100, a level that historically comes with an increased probability of positive returns over the next two weeks. The chart below shows the system’s historical outputs for this setup.

Author’s note: This article was edited for clarity by AI.

Disclaimer: This content is for educational and informational purposes only and does not constitute financial, investment, or trading advice. I am not a licensed financial advisor. Any strategies, models, or projections discussed may not be suitable for your specific financial situation or risk tolerance. All investments involve risk, including the possible loss of principal. Historical performance and model-based scenarios are not indicative of future results. Always consult a licensed financial professional before making investment decisions.