I'm pretty experienced with the Tradestation platform and have sufficient easylanguage skill.
I'm running various models in Adaptrade. In general I model 8 years of 5 minute MES.D data over equal-thirds training, test, validation periods. A validation period is therefore 2 years. The test data ends Jan 1 2024.
I look for trade significants, corr coefficient primarily to be passing in the three periods. The "validation" segment of the equity curve looks basically like the training and testing periods.
However....
When I take the easylanguage code into tradestation and try the data on 1 year of unseen data (Jan 2024 - Jan 2025), most of these models are failing spectacularly. I don't mean equity curve that break down a bit, I mean 45-degree straight-down equity curve that don't look ANYTHING like the passing models in Adaptrade.
So what am I missing in translation here? Something isn't right. I'm pretty familiar with how to configure strategies in tradestation and I'm not missing anything obviousl that I can think of.
How can a most models that pass quality robustness tests (not just a good validation equity curve) really look absolutely NOTHING like the source model in AB?
Thanks for suggestions.