Drawdown Question
Hi Dierk,
I’m backtesting a few some simple models in both Excel and in NT just to double check NT’s performance statistics. I’m basically running a simple crossover “long only” model. I’m concerned that “Draw Down” statistic may not be calculated correctly and or applied correctly.
Excel and NT versions
a)NT and Excel both generate the same trades and cumulative PNL so the models are programmed correctly.
b)If I calculate the drawdown as a typical CTA, I would use the formula
a.Percent Draw Down= Retracement/Peak
i.Retracement=Peak-Subsequent Low.
b.(from Ed Seykota’s site)
c)It is necessary to provide a starting equity to get correct % values, and this would be ideal. (I’m getting the same trade sizes as the Excel version because I’m putting starting equity into the strategy)
The results…
The actual draw down for the simple model is 53%. (Excel)
The optimizer returns a value of 7%
A big difference for anyone looking at drawdown….The NT optimizer returns different dollar amounts and % for the actual Draw Down.
I believe the issue lies in
1) Seeding the Drawdown statistic with starting equity..
2) NT’s current “Draw Down” looks at Peak Equity on a closed basis. (e.g. when a trade is taken off). However, if I’m holding a position for several days or weeks, Peak equity and drawdown are actually measured on a daily basis (e.g. on the close of the bar). Thus Peak Equity always needs to basis open equity.
Simple example: Dear Mr. investor, your account is actually down 50% based on open equity, but since we haven’t taken the position off yet…please discount the 50% figure. We watched the open equity go up for several days and back down....really no drawdown...?
If I adjust the Excel model version to be basis Closed equity…they get the same $ dollar figures. Thus I believe the issue to be constrained to both 1 and 2 above.
My guess is that the DD is some type of intraday/real time DD where users take the position off every day…thus the close of the bar and taking the position off are the same thing.
Regards
Matt
I’m backtesting a few some simple models in both Excel and in NT just to double check NT’s performance statistics. I’m basically running a simple crossover “long only” model. I’m concerned that “Draw Down” statistic may not be calculated correctly and or applied correctly.
Excel and NT versions
a)NT and Excel both generate the same trades and cumulative PNL so the models are programmed correctly.
b)If I calculate the drawdown as a typical CTA, I would use the formula
a.Percent Draw Down= Retracement/Peak
i.Retracement=Peak-Subsequent Low.
b.(from Ed Seykota’s site)
c)It is necessary to provide a starting equity to get correct % values, and this would be ideal. (I’m getting the same trade sizes as the Excel version because I’m putting starting equity into the strategy)
The results…
The actual draw down for the simple model is 53%. (Excel)
The optimizer returns a value of 7%
A big difference for anyone looking at drawdown….The NT optimizer returns different dollar amounts and % for the actual Draw Down.
I believe the issue lies in
1) Seeding the Drawdown statistic with starting equity..
2) NT’s current “Draw Down” looks at Peak Equity on a closed basis. (e.g. when a trade is taken off). However, if I’m holding a position for several days or weeks, Peak equity and drawdown are actually measured on a daily basis (e.g. on the close of the bar). Thus Peak Equity always needs to basis open equity.
Simple example: Dear Mr. investor, your account is actually down 50% based on open equity, but since we haven’t taken the position off yet…please discount the 50% figure. We watched the open equity go up for several days and back down....really no drawdown...?
If I adjust the Excel model version to be basis Closed equity…they get the same $ dollar figures. Thus I believe the issue to be constrained to both 1 and 2 above.
My guess is that the DD is some type of intraday/real time DD where users take the position off every day…thus the close of the bar and taking the position off are the same thing.
Regards
Matt
Comment