Welcome to NexusFi: the best trading community on the planet, with over 150,000 members Sign Up Now for Free
Genuine reviews from real traders, not fake reviews from stealth vendors
Quality education from leading professional traders
We are a friendly, helpful, and positive community
We do not tolerate rude behavior, trolling, or vendors advertising in posts
We are here to help, just let us know what you need
You'll need to register in order to view the content of the threads and start contributing to our community. It's free for basic access, or support us by becoming an Elite Member -- see if you qualify for a discount below.
-- Big Mike, Site Administrator
(If you already have an account, login at the top of the page)
How do you decide when an automated strategy is good enough to take live?
Out of sample data is any data which is not used to develop and optimize a strategy. I'm not familiar with TradeStation, but in NT you can select a date range when running optimization. OOS data would be anything outside of the date range being used to optimize.
Typically for ES I want a minimum average trade profit of $50 (after 2 ticks slippage and commission if using market orders).
The short side seems pointless, why not just make it a long only strategy?
Nice amount of trades, but usually if you can get more trades without sacrificing average trade profit thats better (more statistically significant).
Personally I wouldn't trade this strategy, and if I did it would be in sim for 6 months. The equity curve's correlation coefficient isn't high enough for me.
A system that trades all the time - i.e not regime dependant, but outperforms in certain regimes, and underperforms in others (while still being good enough to not try and filter), would have a poor correlation coefficient. Likewise, a system that dribbles along in a near perfect linear line may not be worth trading as it would tie up trading margin for not much return.
I agree that it would be lovely to have a high correlation, but I wouldn't put much weight in this metric.
Just for a warm and fuzzy, I would compare a sample set of timeusing high order fill resolution (essentially places the trade on a 1 tick data series. Run the strategy normally for lets say the last 12 months, then rerun using high fill. Running the strategy on High fill from 2011 will take a long time. Compare the two results and if you are not getting similar results, then it might be back to the drawing board.
I completely disagree. To give an example just 2 days ago I was sorting through some short ES strategies with equity curves back to 2004 and correlation coefficients in excess of 0.95, strategies which are regime dependent (i.e do well in 2018, 2008 etc, but have marginal return in big bull years.
Anything with a CorrCoef of 0.99 say in back test is probably not going to survive live trading through. It'just a number like Sharpe.
You are quite right. I was poor at visualising equity curves vs correl. It's not something that I use.
I did just plot some out in excel now to get a feel for it, and it wasn't obvious that it's a useful metric. 50%+ drawdowns would still end up with >0.95 for example. I'll stfu now, as I'll likely not add anything useful given my obvious lack of experience with this metric.
Nah, metrics are metrics. For example strategies with nice sharpe ratios sometimes don't function IRL as they do in backtest. I obviously would use a combination of factors.
Personally I like NetProfit/MaxDD, but everyones got a method.
In TS you can find OOS when you optimize parameters. It is only available there since everything is OOS for a naive strategy. After selecting your optimization range it will give you options for standard or walk-forward as well as exhaustive or genetic. There is also a button for optimization settings. In that dialogue you can select your OOS %, date range, fiddle with the genetic algo inputs, etc