[ta12653421]

I bet 100 USD that your tool is not working :-)

Why?

BACKtesting is not comparable to real market scenarios.

What you should do before going live, is a so call Walk-Forward-Test - if this version shows your profits, then launch.

[RayVR]

this is not a trading strategy.

This is a tool that supports simulated execution for strategies. The strategies are left for users of the tool to implement.

I have no idea if this code works or if the features listed are finished but if done well it’s a useful baseline for anyone interested in algorithmic trading.

[nurettin]

Walk-forward is just back testing with out of sample periods where your parameters are constant. You can emulate it with a backtester by reducing parameter space to one in those periods.

[ta12653421]

OK,

i will give you 1000 USD if you deliver an algorithm that delivers similar returns for the same asset class / same timeframe with back-testing and second run with walk-forward-testing :-))

[nurettin]

I don't think you understand the explanation. It isn't the same timeframe.

Months 1-6 backtest and optimize.

Months 7-8 backtest, no optimization. Use best parameters from 1-6.

Months 3-8 backtest and optimize.

Months 9-11 backtest, no optimization. Use best parameters from 3-8.

Walk forward with only backtest.

[ta12653421]

Looks like we are talking about two different things.

What you describe is not a walk-forward-test which adds data continuously on a given frequency, like daily (or hour or whatever)

[nurettin]

What you described sounds like just forward testing. You do this for a few weeks to make sure that live trading will match backtests.

See the docs for walk forward optimization here

https://help.tradestation.com/09_01/tswfo/topics/about_wfo.h...

[ta12653421]

Mh, interesting: i was told this is called "walk forward testing" - nontheless: thanks for clarification!

Exactly what you describe is what we are doing: Since in backtesting, our algo works more than perfect - if we switch to "live-data-scenario", meaning adding data bar by bar (per timeframe), the system behaves differently and the results are a little bit worse.

Therefore my sentence above: During intensive testing we found out that most of those strategies fail (nearly completely) in real-world-market-scenarios.

Tradestation: We have built our own app, we are aware of these "of-the-shelf" solutions, but they are fairly limited regarding what we are doing.