What overfitting means
Overfitting happens when the rules of a strategy have been adjusted, again and again, until they fit one stretch of historical data almost perfectly. The strategy ends up describing the specific quirks of that period — the noise — rather than a pattern that repeats. It looks excellent on the data it was built on, and then performs poorly on data it has not seen.
The tell is a backtest that looks too good, produced after many rounds of tweaking. The rules were shaped to the answer, so of course they match it.
An everyday comparison
It is the difference between memorising the answers to one practice exam and understanding the subject. Someone who memorises the answer key scores perfectly on that paper and poorly on a new one, because they learned the paper rather than the material.
An overfit strategy has memorised the market’s past rather than learned anything that will hold in its future.
Why it is so easy to do
With enough settings to adjust and enough attempts, almost any backtest can be made to look good. Each new combination of rules is another try, and the more combinations you test, the more likely it is that one looks impressive by luck alone. This is why testing many ideas and reporting only the best one is so misleading — the winner may owe its result to chance.
The danger is that the person overfitting usually does not feel like they are cheating. They feel like they are improving the strategy.
How overfitting is caught
The main defence is to judge a strategy on data it was never tuned on — holding back a period of history, building the strategy without it, and only then testing on the part that was kept aside. A strategy that was fitted to noise falls apart on this fresh data; one that found a real pattern tends to hold up.
Checking that results stay stable across different periods, and counting how many variations were tried before the winner was chosen, are part of the same defence. The QS Lab validation system is built around catching overfitting, because it is the failure that undoes more strategies than any other.