@jsfinancials: In sample is the data your model has already seen. Out of sample is the data it hasn't. Almost every blown up strategy dies on that one sentence. You split your history. You let the model learn on one chunk and you tune the parameters on it. That is your in sample. The numbers there are not performance. They are the model describing data it was optimised to describe. Of course it looks good. You bent it until it did. A beautiful in sample curve does not mean you found an edge. It means you found parameters that perfectly explain the past. That is not prediction. It is memorisation with extra steps. Out of sample is the test that matters. Data the model never touched, every parameter frozen, run once. The moment you peek and then go back to retune, that data is now in sample and you have contaminated the only honest test you had. This is where retail backtests quietly die. The thing that printed in sample falls apart out of sample. Sharpe collapses. The gap between the two is not bad luck. It is the overfit showing itself. The bigger the gap, the more of your backtest was fiction. In sample tells you what your model can fit. Out of sample tells you whether you have anything at all. Everyone has a great in sample. That is exactly why it is worth nothing. #jsfinancials #quant #quanttrading #backtesting #data

jsfinancials

Open In TikTok:

Region: AU

Tuesday 19 May 2026 01:33:23 GMT