There has been on-going discussion about systems with relatively low APD. In general they’ve been called “high risk”. Even C2 uses APD exclusively (I think) in computing “low risk per trade” systems in The Grid. If you sort all the “low risk per trade” futures that come up in The Grid by APD, the lowest number is 0.32.
Though APD can highlight a certain type of problem with a system (adding on to losing positions, or holding on to same through huge open-equity draw-down) not all systems with relatively low APD have these problems.
APD discriminates against “scalping systems”. Scalping systems risk draw-down at every entry to achieve a relatively low profit. Trend-following systems risk draw-down infrequently to achieve a relatively large profit. Consider a market that on five consecutive days opens at yesterday’s close, goes down 10 points to a bottom for the day, then rallies to close 5 points above the open. If a scalping system went long on each open and closed the trade each close, it would make 5 trades, net 25 points, and have seen 10 points of drawdown each day. The APD each day would be 0.5 and the average across all 5 trades would also be 0.5. If a “trend-following” system got in at the open of the first day and held until the close of the last day, it too would net 25 points on the single trade, but it’s APD would be 2.5 (25/10), 5 times higher than the scalping system. Yet these two systems had THE EXACT SAME MARKET EXPOSURE THROUGH THE 5 DAYS; they have an identical risk structure. The only difference is that the trend-folloing system suffered the 10 point drawdown each day out of profit.
I mentioned that I think C2 uses APD exclusively in determining “low risk per trade” system. Looking at the list of futures that comes up, the fifth-ranked system has an APD of 0.44 and an average loss of $16,000. That’s a low-risk per trade system?
If second-by-second equity curves were available, APD would be unnecessary. All the faults of a system would be evident in the equity curve. It’s when the equity curve granularity allows short-term systems to have huge drawdowns but recover without seeing the plunge that other metrics are needed. A famous fund manager in the 80s and 90s had a spectacular record of profits month after month, but he suddenly went debit during the Asian Contagian. Seems he was a counter-trend trader who kept adding on to losing positions until they turned his way. None of the daily details ever got reported, just the monthly.
In my opinion, the vast majority of systems fall into one of two categories: scalping strategies, and trend-following strategies. It’s difficult to compare performance between the two because scalpers generally have a relatively high winning percentage, a relatively low profit-per-trade, and a relatively high frequency of trading, while trend-followers have a relatively low winning percentage, a relatively high profit-per trade, and a relatively low trading frequency. The only easy way to compare is to inspect their respective equity curves and equity curve metrics. Since it is impossible to build second-by-second equity curves across systems that make 1,000s of trades per year, other metrics are needed, and APD is a good one as long as it isn’t taken too far.
Last thought. Scalping systems are being judged by a metric that discriminates against them, yet the biggest wort of trend-following systems (open equity giveback: you make a trading profit high of $3,000, but “give back” $2,000 in profit to exit with a $1,000 profit) isn’t specifically addressed by any C2 metric. Open equity giveback should be added to C2 as a system metric in the same way APD is.
"APD discriminates against "scalping systems"."
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.
"If second-by-second equity curves were available, APD would be unnecessary."
I tend to agree with that, although you would still need statistics to capture the properties of the equity curve in a sorting like the Grid. But indeed this is one source of advantage of APD over Sharpe ratio: it uses more granular data.
The real solution is to scrap the APD in favor of the Profit Factor and scrap the combined-equity curve in favor of the closed-equity curve. The solution already exists but is being evaded time and again.
There are two basic choices: switching the mental machinery on or leaving it passive and stagnant. There is also a third possibility, the aberation of evasion.
Evasion is the act of blanking out, the willful suspension of one’s consciousness, the refusal to think - not blindness, but the refusal to see; not ignorance, but the refusal to know. It is the act of unfocussing your mind and inducing an inner fog to escape the responsibility of judgment - on the stated premise that a thing will not exist if only you refuse to identify it, that A will not be A so long as you do not pronounce the verdict “It is.”
Hmm. The threads summary above said:
7/19/07 (18:11)…Brian Milloy…4…APD as a Risk Metric
But there were only 3 messages and the "Brian" message is nowhere to be seen.
Did the Brian Milloy fake alias (aka Palsun) finally get the heave ho?
I deleted his message. Another one about how people should not care about open trade drawdowns, etc., etc.
I don’t think a scalping vs trend-following debate is going to serve anyone well but I think you raise some interesting points.
Personally I think the APD adds value in assessing trading systems, but my concerns are with it’s application. I think you hit the nail on the head when you said just because “APD can highlight a certain type of problem with a system (adding on to losing positions, or holding on to same through huge open-equity draw-down) not all systems with relatively low APD have these problems”… I think that is right and it is very badly worded in the description where it says “Low APD means a system allows open losses to run in order to achieve relatively small profits”. The point is whilst letting losers run is certainly a bad practice and it may be true that most systems that let losses run have a low APD, it does NOT mean that all systems that have a low APD let losses run. But the wording is definitive. At a minimum it should say that a low APD may mean it does it.
I also think the risk grading doesn’t make sense as the percentages are all inclusive, ie the low risk percentage is included in the high risk percentage. For example in Broadsword Forex it says
16.7% of systems have better ADP (sic) than Broadsword’s 0.33
19.8% of systems have an APD of 0.20 or higher (High Risk)
15.4% of systems have an APD of 0.40 or higher (Med Risk)
9.1% of systems have an APD of 1.00 or higher (Low Risk)
This actually means that of the 19.8% at 0.20 or higher, (supposedly high risk) some of them aren’t High Risk at all but are also included in the Med Risk or Low Risk figure. It would make a lot more sense if it said
4.4% have an APD of 0.20-0.39
6.3% have an APD of 0.40-0.99
9.1% have an APD of 1.00 or higher
In doing this it emphasises just how few systems have a high APD. I also find it slightly curious that it words it negatively saying 16.7% of systems have a better APD than Broadsword Forex when surely the more impressive statistic is to turn it around and say that Broadsword Forex has a better APD than 83.3% of systems. Just a thought.
The other thing I am concerned with is it’s use as the sole criteria for defining a low risk system in both the System Finder and The Grid, I think that is too broad a task for one statistic especially one that is not well known. You could easily use Sharpe for one of them or give people a choice to decide for themselves what statistics they think define a low risk system, drawdown, profit factor etc. Personally I think calculating the % of account risked per trade would be a very simple measure, I find it galling that under the current criteria some accounts risk an absurd amount of capital on one single trade and yet get defined as a low risk system.
Like I said, I think APD can be very useful but I think it’s application has been poor and has possibly contributed to it being a controversial issue, I think that could easily be improved and then APD could serve us better.
Keith,
I agree with your example about two systems with similar risk structure but different APD. That is a problem.
In this case the Sharpe ratios would be equal, because it does not depend on arbitrary closing and opening of trades. I still tend to view the Sharpe ratio as the superior measure, with exception of one point: it does not take momentary intra-days drawdowns into account.
I don’t want to deny the fundamental importance of your example, but for practical interpretation of the APD it is also important to know if such cases like your example happen frequently. I have doubts about that.
"The point is whilst letting losers run is certainly a bad practice and it may be true that most systems that let losses run have a low APD, it does NOT mean that all systems that have a low APD let losses run."
I tend to agree, but I’m not entirely convinced. Can you give an example of a system with low APD that is nevertheless “good”?
One exception I thought of was a system that hedges some positions. But thinking again, although such a system will have a disadvantage in the APD because the trades are regarded separately, even in that case I think the APD won’t be very low.
>I think it’s application has been poor
Of course, its application in practice has been poor; because it is a poor statistic in theory. It has no tie to perceptual reality. Thus it has become a floating abstraction because of a break in the chain connecting advanced concepts to the sensory data they ultimately subsume, building confusion upon confusion, instead of knowledge on knowledge. The chain relating higher-level content to perceptual reality is broken, thus rendering the conceptual structure, or semblance of one, with no grounding; it is detached from facts and from understanding.
Errors of this kind are widespread. The fallacy involved is identified as the fallacy of "stolen concept."
The fallacy consists in using a higher-level concept while denying or ignoring its hierarchical roots, i.e., one or more of the earlier concepts on which it logically depends. This is the intellectual equivalent of standing on the 40th floor of skyscraper while dynamiting the first 39. The higher level concept is termed “stolen,” because the individual involved has no longer right to use it. He is an epistemological parasite; he seizes, without understanding, a term created by other men who did observe the necessary hierarchical structure.
The resaon stolen concepts are so prevalent is that most people (and most philosophers) have no idea of the “roots” of a concept. In practice, they treat every concept as a primary, i.e., as a first-level abstraction; thus they tear the concept from any place in a hierarchy and thereby detach it from reality. Thereafter, its use is governed by caprice or unthinking habit, with no objective guidelines for the mind to follow. The result is confusion, contradiction, and the conversion of language into verbiage.
"I tend to agree, but I’m not entirely convinced. Can you give an example of a system with low APD that is nevertheless “good”?"
Well that will depend on what our respective definitions of a ‘good’ system are. How about we go by C2’s definition. Just look under ‘Best Systems’ and ‘All Systems’, so this is regardless of instrument or timescale and the best 3 systems on C2 out of 3,500 systems are as follows:-
1. LRC Extreme Futures - APD 0.18
2. Big Cat - APD 0.12
3. Fortune Cookie - APD 0.01.
That took me 5mins, I’m sure I could find more, but these are supposedly the best 3 systems on here so that should go some way to demonstrating my point.
So now you are calling yourself “Neil Boyd” instead of Brian Milloy? Who do you think you are fooling? After reading 2 sentecnces of you I already knew that you’re again another alias of Palsun aka Brian.
Your My Analyst page is an exact copy of that of Brian Milloy. Indeed, posts that were previously posted by Brian Milloy do now suddenly appear as being posted by Neil Boyd! You didn’t even take the effort to make a new account, you only changed the name in the old Brian Milloy account. This makes it perfectly clear that both Brian Milloy and Neil Boyd are imposters.
ANOTHER PALSUN RENAME:
"Neil Boyd" – dunno if this will stay - I guess Brian Milloy was found out?
New name, but all his "My Analyst" pages have the same rant.
Palsun goes on, but the psychosis remains.
sorry Jules - I put this here before reading your post. But the Palsun continues to quack like a duck and fall over like a duck…
> Posted by: Neil Boyd
Neil,
You stole all of Brian’s links. You are a very dishonest
person.
No problem. I guess by now all three of us get an allergic response when we see a new alias quacking.
I would like to address the APD being used when a system is hedging or spreading.
This is a real life example:
My Cheetah system did these two spreads on 7/17. In two different spreads, Cheetah bought the Aug Crude Oil and sold the Aug Gasoline. As far as C2 is concerned, there were four separate trades. However, since the trades were intermarket spreads, there were really two spread trades.
For C2, the four trades resulted in DD’s totaling $3,794 and net profit of $2,693. For the four trades the APD would be .70.
But in the nature of spreads, one leg can be losing money but the other leg can be making money, so that there is no drawdown or very little drawdown. But C2 would still compute a drawdown from the one leg even though the spread itself might have very little drawdown.
So what really happened within these two spread trades? The actual DD’s for the two spreads totaled $1,236 and the profit is still the same, $2,693. So C2 has a cumulative APD of .70, but the actual APD for the spreads is 2.17. That is quite a difference!
It might be debatable as to whether APD is unfair to day trading or scalping systems versus longer term holds. But there is not doubt that APD presents unfair stats for a system that is hedging or using spread techniques. This is especially unfair if APD is being used to rank systems within the C2 grid regarding risk.
Ross, are you saying that the real world profits of scalping systems do not tally up with the reported equity curves even with comms and realism factors?
Has this information come through annecdotal reports? Has it come from people attempting to follow systems manually or autotrading or both? If you have any more info about this, I’d like to know.
I’m not Ross but have made this same comment many times. You cannot look at every C2 equity curve, including curves with realism and commissions taken into account, and expect to get those results in a real trading account. I have traded several of these myself (eg. Extreme-OS) and lost money when C2 shows a sizeable profit, and the root cause seems to always be slippage on limit order fills when the order is converted to a market order in order to stay in synch with the system.
A fast limit order system like Shark ES or Big Cat (and many other examples, some of which are no longer active) will have this problem in spades, and the reason I no longer subscribe to such systems is that I’m convinced it is impossible to make any money at all trying to autotrade them. On the other hand, improvements have been made continually to the autotrading mechanism and maybe these systems have a better chance now than they did a year ago.
What we need is someone who will subscribe, trade the system for a month, and report their actual results. That is the only way to find out how a real trading account compares to the C2 results. With a system like Shark ES gunning for often 0.25 ES points profit, I suspect there is no chance at all of reproducing the C2 results in a real account, or anything even close.
>You could easily use Sharpe for one of them or give people a choice to decide for themselves what statistics they think define a low risk system, drawdown, profit factor etc. Personally I think calculating the % of account risked per trade would be a very simple measure, I find it galling that under the current criteria some accounts risk an absurd amount of capital on one single trade and yet get defined as a low risk system.
DrawDown (I have to say this, sigh: realized losses) would be my choice. For e.g., a conservative system/method (<= 24% DD) etc.,
