Building Robust Trading Systems and How to Avoid Overfitting

---

Building robust trading systems isn’t just about finding the best stats for APR, Sharpe, Sortino, Drawdown, on your current dataset. The moment you optimize rules to fit that data, you introduce a risk: overfitting.
​
Overfitting isn’t exclusive to machine learning. It happens in regression analysis, algorithmic optimization, and even human analysis. You can have a system that looks perfect on paper but falls apart in live trading.
I’ve built tools to help predict forward stability, how a system holds up on out-of-sample data after being trained in-sample. Sometimes, the highest Sharpe or Sortino isn’t the most stable choice. Slightly less impressive stats can lead to better real world performance.

Recently, a dissertation by Alexander Pearson Sheppert quantified this effect and proposed a scoring system to predict out-of-sample success. You can read it here:
media.proquest.com/medi…

I liked the idea so much I built a plugin for Wealth-Lab that displays this score for my analysis. Like the paper, I call it the GT Score (Golden Ticket Score). I’ve been testing it alongside my own tools to evaluate trading systems, both new and existing.
I’m considering building a simple spreadsheet tool that calculates this score from user entered system metrics. Would that be useful to other system developers out there?

As someone who works with data for a living, I’ll always welcome another quantifiable data point. More information rarely hurts.

Thanks for reading Trading Time Machine! Subscribe for free to receive new posts and support my work.

Via https://backtest.substack.com/p/building-robust-trading-systems-and