Maximum Adverse Excursion

Discussions about the testing and simulation of mechanical trading systems using historical data and other methods. Trading Blox Customers should post Trading Blox specific questions in the Customer Support forum.
Post Reply
Howard Brazzil
Roundtable Fellow
Roundtable Fellow
Posts: 54
Joined: Wed Apr 16, 2003 12:45 pm
Location: Houston, TX USA

Maximum Adverse Excursion

Post by Howard Brazzil » Wed Apr 16, 2003 3:54 pm

I’m curious to find out if anyone has had success employing the concept of Maximum Adverse Excursion. My understanding of the concept is that it is a practical application of the adage that, "The best trades never go very far against you."

The literature I saw on the matter some time ago implied that by examining the MAEs of a large number of trades, you would find a region that allowed you to tighten up your initial stop in a manner such that large winning trades remained relatively unaffected, while a significant number of large losing trades were converted into smaller losses.

And now that your disaster stop is much closer to the point of entry, overall risk is reduced. Or is it? It’s a seductive argument, but here are my impressions:

First, each time you tighten your initial stop (for whatever reason), the reliability of the system almost always begins to deteriorate. So when you tighten stops as dictated by MAE analysis, your system will begin to generate more and more losing trades . . . which can, through further MAE analysis (and there will always remain some losing trades which are “largeâ€

Idea Man
Full Member
Full Member
Posts: 12
Joined: Wed Apr 16, 2003 10:34 am

Interesting

Post by Idea Man » Wed Apr 16, 2003 5:59 pm

I've found that it often appears as if by making one change you could make huge improvements in the system.

However and unfortunately, systems and markets behave much more like mature ecosystems where the relationships between the parts are complex and often unknowable absent empirical testing.

Changing something seemingly innocuous often has unintended and perhaps unpredictable outcomes.

I've heard of good results coming by tightening stops over time as the move progresses.

Forum Mgmnt
Roundtable Knight
Roundtable Knight
Posts: 1842
Joined: Tue Apr 15, 2003 11:02 am
Contact:

I concur

Post by Forum Mgmnt » Wed Apr 16, 2003 6:13 pm

Howard,

One of the reasons I believe it is very important to have a very flexible testing environment is that it is rarely sufficient to make one change and then use the results of a single test to make an assessment of the effectiveness of a particular method.

An example:
I've found that one often makes changes that appear to decrease the returns or MAR ratio generated by a system.

However, the change might simply be reducing returns because it lowers the number of trades (or the frequency of the bets).

As you know:

Return = Frequency of Bet * Expectation of Single Bet

Lowering the frequency of the bets leaves other options including increasing the bet size or trading another system simultaneously with the unused risk or equity made available by the change. Testing changes to these at the same time with the initial change might result in a system that has overall better risk-adjusted returns.

Further, there are benefits to be gained when trading multiple systems because of additive non-correlated returns. In other words, a system change might result in less overall return or a worse ratio when looking at the change in isolation but might make it a better candidate for inclusion within a mix of other trading systems.

A flexible testing environment allows you to retest mixes and various other system parameters after making a change to make sure the side-effects of the change are poor and you don't end up throwing out a potential gem unnecessarily.

- Forum Mgmnt

Chuck B
Roundtable Knight
Roundtable Knight
Posts: 469
Joined: Thu Apr 17, 2003 6:34 am

Post by Chuck B » Thu Apr 17, 2003 6:56 am

Howard,

I did a lot of research in MAE a number of years ago; however, I based all my work on initial risk. In other words, instead of using $/contract like the infamous book seller guy, I used what I refer to as R-multiples, initial risk multiples to measure everything by.

At that time I was testing on two long term trend systems with wide initial exits (something like 4ATR for one and 17 day low for other); what I found was that there was little to be gained by trying to come up with some MAE value to cut-off losing trades. The reason was that almost any value of MAE above the maximum of 1R started cutting off trades that turned out to be losers of much smaller scale in the system, AND it started cutting off lots of small winners too. Essentially, my conclusion was that if my initial exit was robust with respect to the objectives of the system, tampering with it like this was pointless to the extent that anything gained was likely offset by adding yet another variable, degree of freedom, to the system.

If I recall correctly, I did my testing on a 16 commodity group over 12 years of data and I also tested against a 12 big-cap stock group over 12 years of data.

I was going to post an excel graph here, but I don't think this board will allow that :( .

Idea Man
Full Member
Full Member
Posts: 12
Joined: Wed Apr 16, 2003 10:34 am

Post by Idea Man » Mon Oct 06, 2003 10:35 am

Chuck, since the board now supports this. Any chance of seeing that Excel Spreadsheet?

Chuck B
Roundtable Knight
Roundtable Knight
Posts: 469
Joined: Thu Apr 17, 2003 6:34 am

Post by Chuck B » Mon Oct 06, 2003 2:14 pm

Here's a quick look at the effect of MAE in terms of R-multiples. The x-scale is the R value where the trade is cut short. A value of 1.0 means there is no MAE threshold while a value of 0.5 means that when a trade reaches an R value of -0.5, it is exited.

Note that this graph is for what I call "Type 1" MAE analysis. This means that once the trade is prematurely exited with the MAE cutoff, the next entry will be the one that the system would normally take had the MAE cutoff not existed. A "Type 2" MAE implementation would allow the entry strategy to become active immediately once the MAE exit is hit -- this is really not an "MAE" method at all, but it is the one people most commonly end up creating when they try to evaluate MAE. In fact, it is essentially a new system.

The graph is for a 70-17 channel breakout system on a group of 20 markets. Realize that although this example covers almost 1000 trades, these results will be different with a different system and/or portfolio and/or timeframe used in the analysis.
Attachments
MAE.gif
MAE.gif (6.95 KiB) Viewed 13033 times

Chuck B
Roundtable Knight
Roundtable Knight
Posts: 469
Joined: Thu Apr 17, 2003 6:34 am

Post by Chuck B » Mon Oct 06, 2003 2:24 pm

Here is a look at the winning trade MAE R-values from the same system analysis as above. This graph shows what maximum adverse excursion each winning trade took in terms of R. This is sorted, from left to right, by increasing size of winning trade R-multiple (not shown). The deep wicks on the right 1/4th of the graph show how far some large winning trades went against the initial entry before going on to be a big winner.
Attachments
winning trade MAE.gif
winning trade MAE.gif (11.93 KiB) Viewed 13033 times
Last edited by Chuck B on Mon Oct 06, 2003 2:30 pm, edited 2 times in total.

Chuck B
Roundtable Knight
Roundtable Knight
Posts: 469
Joined: Thu Apr 17, 2003 6:34 am

Post by Chuck B » Mon Oct 06, 2003 2:28 pm

Now here is the losing trade MAE values in terms of R. These are sorted in terms of losing R value (maximum on left at about -1.8R due to stop slippage) all the way to a very tiny loss at the far right. The wicks show how much a losing trade went in terms of negative R prior to being closed at a smaller negative R.
Attachments
losing trade MAE.gif
losing trade MAE.gif (6.7 KiB) Viewed 13028 times

NJP
Contributing Member
Contributing Member
Posts: 7
Joined: Thu Apr 17, 2003 3:13 pm
Location: Florida

Post by NJP » Thu Oct 09, 2003 11:08 am

I also put a pretty good effort into testing MAE and found it to have a negative effect on my system. The main premise that the so called "evolution of a trade" is shaped like the inverse of the letter "Z" tilted on its side is not correct. At least for a long term trend following system it does not hold true. I also tested taking half positions until reaching a "R" multiple of the MFE then putting on the full position. Once again the premise did not hold up.

Dan G

Post by Dan G » Thu Oct 09, 2003 1:18 pm

Chuck B, I've had results similar to you in completely different systems. Several of my largest trades, in terms of R-multiples, initially went well into negative territory before turning very profitable. Like you, using any MAE based stop seemed to be a waste of valuable money.

For such a seemingly great concept, it doesn't seem to work all that well. At least it didn't in the systems that I tested - and the ones you've tested.

Thanks for bringing this topic to the forum.

Chuck B
Roundtable Knight
Roundtable Knight
Posts: 469
Joined: Thu Apr 17, 2003 6:34 am

Post by Chuck B » Thu Oct 09, 2003 2:15 pm

This is just another one of those typical "book making stories" that someone dreams up, selects a bunch of best case results (i.e. a market/system/timeframe combo that fits the story), and then publishes a book about it. If that goes well, then you can teach seminars, sell software and maybe write a follow-up book or two.

The thing that really bothered me about the original "idea" was that it was based on $/contract MAE which is beyond ridiculous.

Kiwi
Roundtable Knight
Roundtable Knight
Posts: 513
Joined: Wed Apr 16, 2003 1:18 am
Location: Nowhere near

Post by Kiwi » Thu Oct 09, 2003 5:31 pm

Stridsmann in his book also concluded that virtually any stop will reduce your profits.

The value he placed on MAE work was in reducing the standard deviation of trades and decreasing the size of the initial or trailing stops. Both of these measures mean that you can be more agressive in your trading. I havent tested this but it should work for at least some systems.

John

Kitt
Contributor
Contributor
Posts: 1
Joined: Wed Oct 08, 2003 8:45 pm

MAE

Post by Kitt » Thu Oct 09, 2003 7:24 pm

I found in testing futures systems that had performed well using just the system stops that trying to use MAEs as a way of lessening losses or improving an equity curve did not work. It did not improve the percentage of winning/losing trades or increase the profit factor. Taking profits with MFEs also had the same effects from the tests I did.

These concepts may be valuable but I was unable to make them work in the systems I trade and would not use them.

Dan G

Post by Dan G » Fri Oct 10, 2003 11:57 am

It seems like most agree that MAE and MFE are poor solutions to adjusting stops in an existing system.

However, I wonder if using the information that is provided by MAE and MFE could be used as a basis of a trading system. From all of the research that I've done, it seems the exits are by far the more important than entries. Changing entry parameters does have some effect, while changing exit style made huge differences in just about every measure of system perfomance. These differences seem to be large enough that I have begun to classify systems much more by exit style than by how they decide to get into the market. Mark Johnson has said somthing simlar to this in a few of his posts - I actively started thinking this way after one of his posts on using a moving average exit where the MA is shortened by one day every day, rather than keeping the MA constant. This simple difference resulted in very different system performance results. Seeing this post kinda knocked me on the head, "hey, that's what happens in every system I test too", and if it took me a while to put two and two together, well, at least now I 'grokk' the concept.


Even though I can't get it to work(yet), doens't mean there isn't some important information contained in MAE and MFE. The question is how to use this information.

Kiwi :D - Stridsman does point out some good ways to potentially use them. That is a great book on evaluating trading systems. I do think he has a flawed approach in wanting to reduce the Standard Deviation of the trades. As a result, he throws out trades that start to get 'too good', because he wants to keep the STD of the trades low. Is the ability to increase trading size worth throwing out those big trades? Why not just use a Downside deviation rather than STD?

Dina
Contributor
Contributor
Posts: 1
Joined: Mon Oct 11, 2004 8:22 am

MFE & MAE formulas?

Post by Dina » Mon Oct 11, 2004 8:31 am

Please tell me know I can count the MFE & MAE. What are thier formulas?
MFE = Maximum Favorable Excursion
MAE = Maximum Adverse Excursion
Thank you very much!

sluggo
Roundtable Knight
Roundtable Knight
Posts: 2986
Joined: Fri Jun 11, 2004 2:50 pm

Post by sluggo » Mon Oct 11, 2004 8:46 am

They were "invented" and popularized by John Sweeney, who writes a column for Technical Analysis of Stocks and Commodities magazine. Sweeney wrote a book about MAE/MFE, the same book that Chuck B. (post on Oct 9, 2003) thinks is so ridiculous. Amazon has it new for $34 and used for $26:

http://www.amazon.com/exec/obidos/tg/de ... ingblox-20

Or you could use an internet search engine to look for "+Sweeney +MAE" on the web. Maybe you'll find what you are looking for.

choppystride
Full Member
Full Member
Posts: 22
Joined: Thu Jul 17, 2003 4:54 am

Post by choppystride » Wed Oct 27, 2004 8:03 am

For those who work with bar data, how do you calculate the excursion on the bar where your trade is closed with the exit price between the high and the low (and not executed at open or at close)?

In this case you have no way of knowing if your trade has reached the extremes of the bar before its exit.

Should I make a pessimistic assumption? (i.e. assume it has reached the adverse extreme but not the favourable extreme)

Perhaps there are better suggestions (besides the obvious solution of backtesting with higher freq data)?

Post Reply