Would You Trade This System?

[feng456]

Quote from abattia:


How does 20th MC percentile of max draw down compare with say 2 x max backtested drawdown?

And what is your broker’s margin per contract for the instrument you are trading?

Margin is really low and irrelevant really. I mean if you trade one contract for 400-500$ ur crazy or a genius.

2 x Max historical drawdown is about 10-20 percentile...or in other words max historical drawdown is about 2500-3000...20th percentile Monte Carlo result was 4000.

 

[gmst]

Quote from feng456:


backtested over 5 years worth of data.

holding period is on average 1 hour. averaging ~0.7 trades per day.

Live results from 2012 was around 110% returns. 2013 is looking very poor..like breakeven.

Alright, from the looks of it, it seems like a semi-decent system.

I won't trade this system however I can give you couple hints as to how to improve this system.

1) Increase the average holding period from 1 hour to 3-4 hours. Your average trade PL will go up and vig(comm+slippage) as a percentage of profits will decrease.
2) Reduce number of trades, you should ideally strive to get <0.5 trades per day. This will lead to higher trade selectivity and you will quickly see that average trade profitability and other parameters like PF etc will go up.

Hint: how do you reduce number of trades you take? Well, do more research and put in 1 or 2 more conditions in your trade selection logic.

I understand I haven't given you any concrete hint, however if you think about and really work on the above 2 points, results will speak for themselves (depending on your market experience, work involved on tweaking your strategy will take between 7-30 days.)

 

[abattia]

Quote from feng456:

Margin is really low and irrelevant really. I mean if you trade one contract for 400-500$ ur crazy or a genius.

2 x Max historical drawdown is about 10-20 percentile...or in other words max historical drawdown is about 2500-3000...20th percentile Monte Carlo result was 4000.

OK, so (assuming you've got the numbers correct) ...

Minimum capital/contract = $500 + 2 * $3000 = $6500

So $5000/contract sounds too leveraged, IMHO.

 

[feng456]

Quote from gmst:

Alright, from the looks of it, it seems like a semi-decent system.

I won't trade this system however I can give you couple hints as to how to improve this system.

1) Increase the average holding period from 1 hour to 3-4 hours. Your average trade PL will go up and vig(comm+slippage) as a percentage of profits will decrease.
2) Reduce number of trades, you should ideally strive to get <0.5 trades per day. This will lead to higher trade selectivity and you will quickly see that average trade profitability and other parameters like PF etc will go up.

Hint: how do you reduce number of trades you take? Well, do more research and put in 1 or 2 more conditions in your trade selection logic.

I understand I haven't given you any concrete hint, however if you think about and really work on the above 2 points, results will speak for themselves (depending on your market experience, work involved on tweaking your strategy will take between 7-30 days.)

Thanks. Definitely gives me something to seriously think about.

 

[feng456]

Quote from abattia:

OK, so (assuming you've got the numbers correct) ...

Minimum capital/contract = $500 + 2 * $3000 = $6500

So $5000/contract sounds too leveraged, IMHO.

Well I would partially agree with you. I do eventually want to lower the leverage but right now being undercapitalised I need to take slightly more risk. I mean 20th percentile + cushion isn't that bad anyway........

 

[dom993]

Quote from feng456:

For a $5000/contract ES setup:

max dd = 4000

(determined after 10,000 runs in Monte Carlo sim where 4000 was at 20th percentile for max DD)

expected profit per year = 50% of that 5000

(determined through 10,000 runs in Monte Carlo sim where 50% was around the 20th percentile for expected profits)

Using the above information and assuming perfect execution fills etc. (bear with me please), would you trade this?

$2500 expected annual profit / year for $4000 system stop (assuming you would stop on a DD of $4000).

No, I wouldn't trade it. In general, I demand a 200% expected annual profit on my system stop.

 

[dom993]

Re. trade frequency, unless you have a P/F >= 2, your best bet is to actually increase trade frequency, as this will lower your MTR (mean time to recover - from DD).

"Trade selection" requires a lot of caution to avoid over-fit.

 

[feng456]

Quote from dom993:


"Trade selection" requires a lot of caution to avoid over-fit.

I know this, which is why I try to keep it as simple as possible...but can you extrapolate on that? Empirically, how would I actually "use caution" to avoid overfit?

Thanks.

 

[dom993]

I pay attention to the information ratio (avg/stddev) ... say your strategy has an IR of 0.20 ... then you consider a filter, and say those trades only have an information ratio of -0.10 (of course, it is negative as on average those trades should be losing ones). This filter IR isn't as good as your basic strategy, discard the filter. On the other hand, if another filter had an IR of -0.3 (better than our initial strategy), then I would keep it.

Also, look at that filter IR on a year by year basis ... over 5 years, at most 1 year could have a positive IR, and that shouldn't be more than 1/2 of the strat IR for that year.

Of course, the % of trades caught by a filter is quite an important consideration, but there are many situations where a narrow filter can still be acceptable (eg. filtering on a gap at the Sunday session open).

I hope this helps.

 

[gmst]

Quote from dom993:

I pay attention to the information ratio (avg/stddev) ... say your strategy has an IR of 0.20 ... then you consider a filter, and say those trades only have an information ratio of -0.10 (of course, it is negative as on average those trades should be losing ones). This filter IR isn't as good as your basic strategy, discard the filter. On the other hand, if another filter had an IR of -0.3 (better than our initial strategy), then I would keep it.

Also, look at that filter IR on a year by year basis ... over 5 years, at most 1 year could have a positive IR, and that shouldn't be more than 1/2 of the strat IR for that year.

Of course, the % of trades caught by a filter is quite an important consideration, but there are many situations where a narrow filter can still be acceptable (eg. filtering on a gap at the Sunday session open).

I hope this helps.

useful way to do filter analysis thanks...

not sure though if i will get so technical in selecting my filters in the near future. When I decide to select a filter, I basically look at impact on a number of parameters like PF, # trades, %win, DD, overall look of graph, average trade etc.

Whereas you look at only one number which is Information Ratio. Its debatable in my mind which is a better approach.