What is a trading edge, actually?
An edge is a repeatable statistical advantage that survives on data the rules never saw. That last part is what separates an edge from a story. Most trading advice fails one of three tests: it was measured on the same data it was fitted to, it never beat random entry timing, or it rests on too few trades to mean anything. A real edge passes all three.
Based on the same process behind every Library report. How we test.
Test one: the result has to come from unseen data
Run any set of rules over a decade of history and tune the parameters until the equity curve looks good, and you have learned nothing except which numbers fit the past. Walk-forward testing fixes this by training on one window and testing on the next window the rules have never seen, then rolling forward and doing it again. What you read at the end is out-of-sample performance, which is the only kind that predicts anything. If a strategy has only ever been measured on the data used to build it, you do not have a result yet.
Test two: it has to beat random timing, not just the market
A strategy that made money in the last decade may have done nothing except stay long in a rising market. To separate the signal from the tape, we keep a strategy's exits and randomize its entry timing 100 times, then check where the real entries land in that distribution. Beating 90 or more of the 100 random runs means the timing itself carried weight. Landing mid-pack means the rules were decoration on a rising market. This is the test that quietly kills most popular setups.
Test three: there has to be enough of it to measure
Sample size decides whether a good score is evidence or noise. A setup that fires a handful of times in a decade can post a near-perfect score on luck alone, and no amount of statistical polish rescues it. This is why we report the trade count next to every result, and why a backtest with roughly under 30 trades should be read as a hint rather than proof, however good it looks.
What a failed edge looks like: the Golden Cross
The 50/200 moving average cross is the most preached signal in trading, and it is the cleanest example of all three tests at once. On its best ticker it beat almost every random-entry run, which sounds airtight until you count the trades. Here is the live receipt from our test:
Best showing 99 of 100 vs random on QQQ, but only 4 trades in a decade. A high score on too few signals is a coin flip, not an edge, so it clears the bar on no ticker we ran it on.
Why frequency decides so many of these arguments
Notice that the Golden Cross did not fail because trend following is wrong. It failed because a 200-day average moves so slowly that the cross almost never happens, so there is never enough evidence to judge it. Faster versions of the same idea fire often enough to build a sample you can trust, at the cost of more whipsaws. Once you internalize this, a lot of trading debates resolve themselves: before you argue about whether an indicator works, ask how often it fires.
How to use this definition
- Before trusting any backtest, ask whether the result came from data the rules had already seen. If it did, it is a fit, not a finding.
- Ask what the strategy is being compared against. Beating buy and hold in a bull market is a low bar; beating random entry timing is not.
- Count the trades before you read the return. Few trades plus a great score is the signature of luck.
- Judge a setup by how many tickers it held an edge on, not by its best single result. One hot ticker is a coincidence you can rent, not an edge you can own.
- Apply this to your own trading too. Your instinct that a setup works is a hypothesis, and it deserves the same three tests.
Common questions
What is a trading edge?
A repeatable statistical advantage that shows up on data the rules never saw. In practice that means it survives walk-forward testing, beats randomized entry timing, and rests on enough trades to be measurable rather than lucky.
How do I know if my trading edge is real?
Test it out of sample, compare it against random entry timing instead of just against the market, and check the trade count. Failing any one of those three means you do not have a result you can trust yet.
Why do most backtests overstate an edge?
Because the rules were tuned on the same history used to score them. Walk-forward testing avoids this by always measuring on the next unseen window, which usually produces a far less flattering and far more useful number.
How many trades does a backtest need?
There is no hard threshold, but under roughly 30 trades we treat a result as a hint rather than evidence. Signals that fire rarely can post near-perfect scores on luck, which is exactly what happened to the Golden Cross in our test.
Is beating buy and hold the same as having an edge?
No. In a decade that trended up, plenty of rules beat buy and hold simply by being long most of the time. The harder and more informative comparison is against the same strategy with its entry timing randomized.
The methodology page walks through walk-forward testing and the 100-run random-entry gauntlet, with nothing hidden.