APD as a Risk Metric


You must be appalled by “Palsun Anboyd”. He has 14 systems. ALL have closed DD’s > 24%. The poor guy couldn’t create a winning system if his life depended on it.

You and Brian should critic Pally’s systems on your analysts pages.

PS: for e.g., Magnum (Conservative) Stocks/ETFs is a conservative system by this definition…

I would say that is more of a problem in C2 not PERHAPS being able to link the partners in a spread, such that they generate a single max DD and profit in the track record. The APD can only generate what it sees, which is dependent on the trades in the track record.

There is indeed no doubt that hedging will artificially lower the APD, as in your example. But I wonder how strong the effect is. Even the lowered APD of 0.70 would be fairly good in comparison to most other systems. Is it possible to get an APD, say, less than .20 while it would be larger than .70 if the hedging was accounted for?

> PS: for e.g., Magnum (Conservative) Stocks/ETFs is a conservative system by this definition…

The “magnum” is a impotent bb gun. In a roaring bull trend it hasn’t

manged to make a penny. Also isn’t this a "scalping system"

by your (um, Brian’s?, er Pal’s?) definition?

See details for Magnum (Conservative) Stocks/ETFs

Magnum (Conservative) Stocks/ETFs

23 wks 333 6.5 days 55.0% 1.0 $123 -3.5%

Even by Pal’s standards of slimy hypocrisy the idea of someone posing

as someone else to tout his system is the unscrupulous and underhanded act of a low life swindler. But of course Brian, Pal, and

Neil are not the same person are they?

"APD discriminates against "scalping systems" - Keith

"Scalping systems on C2 are notorious for not delivering the published profits, so I do not see the problem. This was actually a nice side effect of APD." - Ross

I very much agree. APD does not discriminate against scalping systems. If anything, so far it has served to portray them accurately.

It seems to me that both you and Ross think scalping systems are "bad" and that APD proves this. Do you really believe scalping systems are inherently bad?

you need to do some backreading on the forums. Especially the autotrading forum. There is plenty about these problems

They’re not inherently bad, but some scalping systems on C2 suffer from the fact that the delay between the time the vendor issues the signal and the execution in a subscriber’s account can easily be 5-10 seconds when using TradeBullet, and over half a minute when trading manually (don’t know what it is when using server-based). In addition, when a scalping system gets popular on C2 and attracts a lot of subscribers it is more vulnerable to the (il)liquidity of the market (i.e. slippage), compared to longer-term systems. Finally, scalping systems using stops and limits might have more problems getting fills than longer-term systems.

Another example (besides hedging) of APD inaccuracy as applied to futures trading:

If you had shorted OJU7 on 5/22, you would have about $4000 in profit with little or no DD. APD for this trade would be, well, very large depending upon the small DD on day of entry.

If you shorted OJN7 instead and rolled the position in late June, you would have an APD similar to the example above for OJN7, but the current OJU7 position would currently be underwater and thus showing a very bad APD.

What is essentially the same trade can have 2 very different APDs.

It possibly could be argued that in the long run these things will tend to balance out. However, I suspect that longer term systems that eventually exit on stops will suffer because the last leg of the trade can often show a loss. If the same methodology were applied to stocks, for example, this problem will never occur (you don’t roll over stocks…)


I believe APD currently does not discriminates against scalping systems. Nothing more, nothing less.

I have not enough knowledge about rolling over: I don’t trade futures. So I can’t really answer to that. But what you say seems to be another case of the general point that APD is a problem when two trades should be considered as one trade.

I think it is reasonable to improve the calculation of the APD by treating simultaneous trades with the same underlying as one trade. Similarly, I can imagine that rollovers are treated as one trade.

But I expect that this requires a lot of programming work from Matthew. So probably it won’t be done very soon. The more pragmatical solution is to warn that the APD is not sound for these systems.

Yes, I have had this conversation with MK and realistically it’s not something that is likely to be implemented. I deal with this problem in my description in Broadsword Macro which will hold futures positions long term and inevitably over rollovers. Obviously it can work both ways but people need to be aware of the issue.

When I update the advanced statistics program I can perhaps add an equity curve based version of the APD. That would be the net return divided by the sum of the DDs. That is similar to an APD where each day counts as a trade. This should solve some of the problems with hedging.

Unfortunately, it is not possible to use intraday data for that. It is conceptually similar to the Sharpe ratio, as I have discussed previously with Science Trader. So it will have similar limitations for daytrading systems.


Regarding your suggestion to use average profit and average DD in the APD calculation - Again there is a problem with futures rollovers. Back in '96 to '98, I was short wheat for nearly 2 years. This entailed somewhere around 10 rollovers. The profit for the trade was somewhere around $7000 as I recall. If the APD calculation used the gross totals as it does now, then it would see all of this $7000 profit. But if it used averages, then the profit in the numerator would have been something like $700. The DDs would have been similar in both cases. But one APD would have been 1/10 the other…


My suggestion won’t solve the problem of rollovers or hedging. It won’t make it worse either. It will deal only with the problem of the n/a’s. My suggestion was to take the average and include the trades with n/a DD in the numerator. In the situation that you describe there are no n/a’s, and then the APD doesn’t change because both the numerator and the denominator are divided by N1 (the number of trades with n/a DD). That is, the net profit is divided by 10 but the sum of the DD’s is also divided by 10.

Ross, I agree that the problem is not with the APD (in theory) when it comes to spreads. But the APD only exists at C2 and since C2 can’t calculate the correct APD for hedges and spreads, then the problem is with the C2 APD. Further, there is a problem if the APD is used to determine risky systems. Someone searching for a good trading system would eliminate any system that hedges with options, uses spreads etc…

I’ve recently discussed some of these issues with MK, due to the fact that various parameters are incorrectly displayed for spread trades (such as those executed by Live Long and Prosper). I also realize it would require that a separate “class” of transaction be created and programmed for. As a result, it is simply something we need to realize and live with for now.

In the case of LL&P, a spread such as “BTO @ $5K, STO @ $12.5K” results in a win ratio of 50% rather than 100%, arbitrary draw-down (as long as market value of an open trade exceeds its initial cost, there is no draw-down) even while profitable, and other calculations that are affected by this shortcoming. I believe APD is affected as well.


Here is another actual spread trade from Cheetah:

Sold the Heating and bought the Crude Oil. This was a net loss of $941. For the two legs, C2 has a total DD of $220 on one leg and $2041 on the other leg for total DD of $2261.

However, the actual DD was much different. The actual DD on the spread was the same as the loss, $941.

Now take this info on this spread trade and add it to the info from the prior spread trade that I noted a few posts earlier:

For C2, the total P/L on the two spreads is $2,693 - $941 = $1752 profit. The total DD for all the trades within the two spreads was $2,261 + $3,794 = $6,055. This would make the APD for these trades, as calculated by C2, at .29.

But here is the actual APD calculation that C2 can not calculate since it does not have spread prices:

The net profit is still $1,752. However, the actual DD is $1,236 + 941 = $2,177. The actual APD for these two spreads is then calculated as being .80. Ouch!! That is a big difference. .29 versus .80 means that an overall hedging or spreading system could be shown to have a very low APD when actually it could be a very good APD. In fact, it is probably impossible for a spreading or hedging system to have a decent APD.

The problem is this—To get an adequate idea of of DD, a snapshot has to be taken of the equity at various times and then that equity is charted. The best way to do this is the way a broker might do this, such as every few minutes. An large DD during the day would show up. The APD is a snapshot, to some degree, on individual trades, but for systems that hedge or spread or roll over futures trades, the APD is a misrepresentative snapshot.


Not sure where your work will eventually take you, but one thing to keep in mind, APD is not useful unless all net profits are totaled separately from the total of all the max DDs.

Trying to do anything like using averages of each trade (NP/maxDD) will yield horrendously skewed results, because some days you will have things like $3000 / $0 and other days $0 / $4750 (infinite or zero-division).

I think in the end, trying to get clever with it will not cause it to be much better, except perhaps for systems that tend to have many no calcs. But on the other hand, the result without the max DDs (no calcs) may only lead to different problems.