NorQuant

Building systematic futures strategies
Research • Backtesting • Execution
Sharing insights, not signals
Located in Norway

[NorQuant @NorQuant]

One of the first systems I made.

Runs on DAX (15m timeframe)
Backtest: 2006–2026
• 56% win rate
• 1.86 gain/loss ratio

Been trading this live since 2018.
No losing years so far.

[NorQuant @NorQuant]

Because the goal isn’t to fit the past perfectly, it’s to survive new data.

A wise trader once said "Make it fit like a mitten, not a glove" and that quote has stuck with me since forever.

Most overfitted systems look great on one dataset, the trick is to make it look great OOS

[NorQuant @NorQuant]

Then I optimize the strategy independently on each segment.
Instead of picking the single “best” parameter I look for stability across all datasets.
If a parameter is optimal at:
• 11 in dataset 1
• 10 in dataset 2
• 12 in dataset 3
I’ll choose something in that range.
2/3

[NorQuant @NorQuant]

How I avoid overfitting when building trading algos 🧵

When I’m ready to optimize a system, I don’t just run it on the full dataset.

I split the data into multiple segments.
Typically:

• First 40% of the data
• Middle 40%
• Last 40%
(Yes, they overlap)
1/3

[NorQuant @NorQuant]

Been running this algo on S&P 500 futures since 2019.

2019–2026:
• 73% win rate
• 1.86 gain/loss ratio

Backtest 1997–2019:
• 73% win rate
• 1.92 gain/loss ratio

Behavior is consistent across periods, doing what it’s supposed to do.

(2 contracts, MES)