How long does it take to prove a trading strategy works?
For a typical trend-following rule, about 34 months of live trades. Two weeks, or even six months, cannot tell skill from luck.
How this was made: the code, the research runs and the first drafts of this text were produced by an AI assistant (Claude) working under the operator's direction; the operator made every decision and checked the numbers. All trading was on paper: real prices, no real money. Nothing here is investment advice.
The numbers behind it
Our daily and four-hour trend ensemble (AND-3), replayed with costs, made an average of +3.7% per trade with a standard deviation of 18.7%, at about five trades a month. The four-hour breakout rule we later ran on the server looks much the same over nine years (1,037 trades):
- Win rate 36%. Losing trades averaged −3.4%; winning trades averaged +13%.
- The top 10% of trades produced more than all of the net profit; the other 90% lost money together.
The calculation
To tell an average of +3.7% apart from zero, when single trades swing by 18.7%, with 90% confidence and 90% power, you need roughly n = ((1.28 + 1.28) × 18.7 / 3.7)² ≈ 167 trades. At five trades a month that is about 34 months. At 95% confidence and 95% power it is about 276 trades, or 56 months.
What short windows look like
In the nine-year replay of the breakout rule, 60% of all two-week windows ended at or below zero, and 12.9% ended at −3.93% or worse, which is exactly where our live server test landed. A bad fortnight is normal for this kind of rule, and so is a good one.
So what can you check quickly?
- Whether the live system makes exactly the decisions the replay would make (we checked trade by trade).
- Whether fills are close to the prices the replay assumed (ours averaged 0.05% worse).
- Whether the system stays up, on time, with no errors.
Those tell you the system works. Only long experience, or a much bigger edge, tells you the strategy works.