How to use strategies, e.g, DWT_Predict.

[MuhammadElsaeed]

Hello All, This sounds a stupid a question and appreciate if anybody can help. I have been playing around with the strategies and to be precise this one -> DWT_Predict, I ran the hyperopt to get the best parameters, ran it in backtesting with different time ranges and I get great results, almost no loss. after investigation seems that the strategy is looking at data in the future during backtesting that is why it performed so well(I later noted it was mentioned in the readme). Of course when I run the bot in dry run I get different behavior for the same time period. My question is, how I should use the strategy? how to backtest it in a good way without looking to future data? what I am missing? Again, I apologise if my question are so stupid 😄 .

[maledicente]

Train/Hyper like 6 months 2022/01/01 -> 2022/06/01 and backtest 2022/07/01 Test 2021/01/01 or another months

[MuhammadElsaeed]

I did exactly that, but that strategy is looking to future data during hyperopt and backtesting. I get the best result in backtesting(98% win rate), but with dry run I am losing since dry run mode does not have the future data, I even backtested the same period when the dry run mode is on and I get different result.

[nateemma]

This is an old strategy that I abandoned precisely because of the lookahead problem - sorry, I should have removed it.

The general logic has been moved to TSPredict.py. This is actually just a base class that implements the general logic and buy/sell/exit callbacks. There are several families of subclasses (Gain, SImple, Coeff and Wavelet) that implement different prediction approaches. Note that I am currently restructuring these in an attempt to eliminate all lookahead bias (it's much harder than you would think).

Cheers,

Phil

On Wed, Jan 3, 2024 at 6:25 AM Muhammad Elsaeed @.***> wrote:

I did exactly that, but that strategy is looking to future data during hyperopt an backtesting. I get the best result in backtesting(98% win rate), but with dry run I am losing since dry run mode does not have the future data, I even backtested the same period which the dry run mode is on and I get different result.

— Reply to this email directly, view it on GitHub https://github.com/nateemma/strategies/issues/43#issuecomment-1875454440, or unsubscribe https://github.com/notifications/unsubscribe-auth/ABD4X546RUK3XVVJNIWXNLTYMVS3ZAVCNFSM6AAAAABBLGTWHGVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMYTQNZVGQ2TINBUGA . You are receiving this because you are subscribed to this thread.Message ID: @.***>

[phuphighter]

@nateemma when backtesting any of the TS_Coeff* strategies, I am getting results. However, in dry-run mode, I'm getting zero trades. For instance, I have dry run since 1/1/24 and have zero trades. But if I backtest the same time period, I get a decent amount of trades.

If I remove the model and re-train, I also get zero trades.

Any help would be great!

[nateemma]

I'm debugging that right now - something is not transferring properly between activations and the training is being reset

Cheers,

Phil

On Tue, Jan 16, 2024 at 1:38 PM Johnny Khai Nguyen @.***> wrote:

@nateemma https://github.com/nateemma when backtesting any of the TS_Coeff* strategies, I am getting results. However, in dry-run mode, I'm getting zero trades. For instances, I have dry run since 1/1/24 and have zero trades. But if I backtest the same time period, I get a decent amount of trades.

If I remove the model and re-train, I also get zero trades.

Any help would be great!

— Reply to this email directly, view it on GitHub https://github.com/nateemma/strategies/issues/43#issuecomment-1894556254, or unsubscribe https://github.com/notifications/unsubscribe-auth/ABD4X55O3HNH2A3WLRMHUZLYO3XNRAVCNFSM6AAAAABBLGTWHGVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMYTQOJUGU2TMMRVGQ . You are receiving this because you were mentioned.Message ID: @.***>

[phuphighter]

Thank you @nateemma, this is working now!

[ThomasGoud]

1 Envoyé depuis Yahoo Mail pour Android

Le mer., janv. 17, 2024 à 21:32, Johnny Khai @.***> a écrit:

Thank you @nateemma, this is working now!

— Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you are subscribed to this thread.Message ID: @.***>

[ElDelak]

@phuphighter : how did you backtesting TS_Coeff* strategies? I'm trying to use it but I have this error : (no module named ''modwt) can you help me?

[phuphighter]

@ElDelak make sure modwt.py is in the utils directory.

[ElDelak]

@phuphighter : Thank you :+1: