Guide
How long must a track record be to prove skill, not luck?
Short answer
Much longer than most records run. An edge of 1% a month over the market, with a typical monthly swing of 4% to 6%, needs about 8 to 19 years of monthly results before a test can tell it from luck 80% of the time. If the strategy was the best of 10 tries, that becomes 16 to 35 years. A year or two of results mostly measures luck.
- months for a 1%-a-month edge, one try
- 99–223
- months for the same edge, best of 10 tries
- 187–421
- chance luck alone matches 24 months of +2%, best of 10
- 41%
How do you tell skill from luck in a track record?
Compare the average return with how much the results swing from month to month. The standard measure is the t-statistic:
t = average monthly return ÷ (monthly swing ÷ √months)
A record clears the bar when its t is high enough that luck alone would rarely produce it. For one honest try, with a one-sided 5% test, the bar is t 1.64. Use the return beyond the market, not the raw return, where you can: in a rising market raw returns flatter everyone.
How many months does a real edge need?
The table shows how long a real edge of a given size needs before it clears the bar 80% of the time. The swing is how far a normal month moves up or down; for stock strategies it is usually 4% to 8%.
| Edge a month, swing | One try | Best of 10 | Best of 100 |
|---|---|---|---|
| 2%, swing 4% | 25about 2 years | 47about 4 years | 69about 6 years |
| 2%, swing 6% | 56about 5 years | 106about 9 years | 154about 13 years |
| 1%, swing 4% | 99about 8 years | 187about 16 years | 274about 23 years |
| 1%, swing 6% | 223about 19 years | 421about 35 years | 615about 51 years |
| 0.5%, swing 4% | 396about 33 years | 748about 62 years | 1,093about 91 years |
Halving the edge quadruples the months needed. That is why a modest but real edge is so hard to prove, and why a short record with a huge return deserves more suspicion, not less.
Why does the best of many tries need more?
Trying 10 versions of a strategy and keeping the best is like rolling 10 dice and reporting the highest. The bar has to rise with the number of tries: from t 1.64 for one, to t 2.58 after 10 and t 3.29 after 100.
Take 24 months averaging +2% with a 6% swing, a record many sellers would advertise. Its t is 1.63.
- If it was the only strategy ever tried, luck alone produces a record this good 5.1% of the time.
- If it was the best of 10, 41% of the time.
- If it was the best of 100, 99.5% of the time.
A seller with several newsletters, model portfolios or accounts is showing the best of several by construction. So is anyone who tuned a backtest until it looked good.
Is a one-year track record ever enough?
Almost never. In 12 months, a single honest try can reliably show only an edge of about 2.9% a month with a 4% swing, or 4.3% a month with a 6% swing. Almost no real strategy beats the market by that much.
A year of +3% a month with an 8% swing sounds impressive. Its t is 1.30: luck alone does that 9.7% of the time for one try and 64% of the time for the best of 10.
What the calculation leaves out
- Costs. Spreads and fees lower the average, so real records need even longer.
- Streaks and extreme months. The formula treats months as independent. Real returns come in streaks, which makes a simple t look stronger than it is, and one extreme month can carry a short record.
- Missing companies. A backtest that only uses stocks still trading today looks better than reality. See survivorship bias in backtesting.
So treat the months in the table as a floor.
How Autrady judges a forward track record
Every pick on an Autrady track is timestamped on arrival, so the record is out of sample from day one. The verdict follows the same logic as this guide:
- No verdict before 12 months with closed picks.
- "Beats the market" only when the return after costs and the market's own move clears the bar.
- "Does not beat the market" only once the record could have detected an edge of 1% a month with 80% probability. Until then the page says "not enough evidence yet" and roughly how many more months are needed.
- The bar rises with each track the same seller starts, so starting many and keeping the best does not work.
Check a record yourself
The free luck check runs this calculation for any claimed record: average monthly return, monthly swing, months and versions tried. It is also available to AI assistants as the luck_check tool on Autrady's public MCP server at https://autrady.com/mcp.
Before trusting the inputs, check where the record came from: how to check a stock picker's track record.
Selling picks, or buying them?
Sellers get a public record nobody can backdate, free for founding sellers until 31 March 2027. Buyers read every track and audit free, and can check any claimed record for luck.
Autrady records and tests evidence. This guide is research, not investment advice, and past results do not predict future returns. Questions or corrections: [email protected].