Acceptable Sharpe Ratio for live trading?
#1
Newbie question...

After running GekoWarez's GA to back test and get a good configuration for a nice profit, can someone recommend an acceptable Sharpe Ratio for using the config for live testing?

I know the closer to 0 or higher the ratio, the less risk, but I'm just curios if anyone is actually seeing any success in live trading after seeing a strat back test as well as some of the stuff I'm seeing, like 400% profit in a three month span.    I don't expect those kinds of results in live trading, but if the Sharpe Ratio is say -0.2 is that too much risk when the test results are so great?

I have three live trade bots running against different exchange accounts right  now, but only for a few days which is not long enough to see anything positive.

Wondering if the risk factor is too high...

Thanks
  Reply
#2
A higher sharpe ratio is preferrable however you do also got other metrics (that Gekko doesn't output currently) like win% (amount of trades that was a win) and what the average win and loose was etc. There's also alternate methods of measuring risk such as Sortino Ratio.

But since you do not got that data and only sharpe try using things that at least backtest over 0.

In the regular markets i wouldn't touch anything that had a sharpe below 1. Do note that crypto is insanely volatile though so getting over 1 in sharpe is a lot harder.
  Reply
#3
(04-15-2018, 07:50 AM)tommiehansen Wrote: A higher sharpe ratio is preferrable however you do also got other metrics (that Gekko doesn't output currently) like win% (amount of trades that was a win) and what the average win and loose was etc. There's also alternate methods of measuring risk such as Sortino Ratio.

But since you do not got that data and only sharpe try using things that at least backtest over 0.

In the regular markets i wouldn't touch anything that had a sharpe below 1. Do note that crypto is insanely volatile though so getting over 1 in sharpe is a lot harder.

Thanks!

Most all my profitable tests with different strategies show a ratio below 0.  Sometimes only -0.5 or so.  I usually test across multiple time periods as well to make sure the strat returns a profit during different market trends.  

I'll start retesting to see if I can improve the Sharpe Ratio, which is always different depending on the test's time period.  I know it will be challenging to consistently find a positive ratio.

Any additional tips are welcome!

Thanks again...
  Reply
#4
Gekko's Backtest result page isn't very helpful.  Basing your results on overall Simulated Profit is really asking for trouble.  You end up 'curve fitting' for the data you have.  History is the best we can do, but it is very unlikely to ever be repeated, especially not in the order it first came in.

Try looking more closely at the trades.  Stretches of losses can cause trouble for your account if they come at the wrong time.  Large losses might be controlled with stop-losses or other mechanisms, perhaps reducing overall profit but making the strategy more future-friendly.  Consider what a few more of those large losses might do to your results.  Keep them under control.  Not using some form of stop-loss is unnecessary gambling.  A signal for a trade in the opposite direction can sometimes be effective.

Accepting a lower overall profit for smaller stretches of losses and minimizing large losing trades will serve you well.

Most importantly, the more you squeeze every bit of overall profit from your backtesting the less likely it will work in the future.  That's curve fitting.  It's all about trade offs.  Decent profits versus controlled, limited losses.

There are lots of helpful write-ups and videos on-line dealing with backtesting and curve fitting.  Suggest you check a few of them out.

Here's the first one that came up for me in a search ... it's not bad ...

    https://www.youtube.com/watch?v=kJ_6b_2icBU

Good luck!
  Reply
#5
I appreciate the information, the video was informative.

I do test across multiple time periods to make sure I get a positive result in different market scenarios. I'll be widening those tests now though.

I just learned today that the current version of Gekko has an issue with its Sharpe Ratio results, so the ratio result can't be trusted.

Another thing I do when using GA to back test is to slightly increase the slippage and maker fees to make it more difficult to return as positive result.  Once I get a good result from GA, I then use those settings to manually fine tune a strategy across different time periods.  

It's still too early to tell if my back testing strategy is going to successful.
  Reply


Forum Jump:


Users browsing this thread: