Overfitting in backtesting
Overfitting is tuning a strategy to past data so closely that it reproduces the noise of that history rather than a stable pattern; it then fails on new data.
What it means
Signs: the result collapses after a small parameter change, only one period carries the profit, or a filter was chosen because it removed the worst trades.
Protection: fewer free parameters, out-of-sample and walk-forward checks, and a placebo test — the same strategy on shuffled data must not look as good.
How 1nTest computes it
Every 1nTest result shows robustness checks next to it: stability across time windows and other indicators, plus an optional placebo test against entries shifted at random in time.
We publish losing examples deliberately: a library that shows only winners is itself a form of selection bias.