Blend which other system types w/LTTF for smoother eq curve?

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.
sluggo
Roundtable Knight
Roundtable Knight
Posts: 2987
Joined: Fri Jun 11, 2004 2:50 pm

Blend which other system types w/LTTF for smoother eq curve?

Post by sluggo »

I received a message from a Trader's Roundtable forum member, which asked:
  • QUOTE: As I read through the Blox forum, I find that several of the more professional people
    are trading multiple systems to smooth out the drawdowns that come from long term trend following
    systems. Also if I look at succsessful commodity fund managers on IASG, I see similar kinds of things
    (lower drawdowns and fewer losing months with the normal fluctuations in returns). I have traded
    long enough to be comfortable with the drawdowns and I have my heat rate set to where my
    worst case drawdowns are about 30% in the system I trade. With that long introduction,
    my question is, what types of systems would one trade along with a long term trend following
    system to smooth the equity curve?
    END QUOTE
Instead of merely sending one opinion (mine) to one person (the author of the quote above), I thought
it might be preferable to let anyone and everyone chime in -- and to also let everybody see the answers.

Please feel free to post research results, test data, passages from books, articles, guesses,
things you once heard somebody say, opinions, and half baked thoughts about the question above.

It's a free for all, don't be shy.

EDIT - fixed grievous (!!) spelling error
Last edited by sluggo on Tue Jul 19, 2011 10:08 pm, edited 1 time in total.
MarkS
Roundtable Fellow
Roundtable Fellow
Posts: 70
Joined: Wed Dec 15, 2010 12:10 pm
Location: Chicago, IL

Post by MarkS »

Sluggo -- does it count if a different person (me) sends the person above (author of question) a link to a thread written by you? :)

Seriously, I consider your post in the link below to be a great, succinct summary on the four ways systems can vary (under the assumption one wants to use systems that are not correlated when trading multiple systems):

viewtopic.php?t=6880&highlight=correlation
rhc
Roundtable Knight
Roundtable Knight
Posts: 464
Joined: Tue Nov 18, 2008 8:46 pm
Location: Oz

Post by rhc »

With that long introduction, my question is, what types of systems would one trade along with a long term trend following system to smooth
the equity curve?
I’ve found, (through research not direct experience . .yet!) that blending 2 slightly different versions of the system that I’m currently using with ‘1/2x%’ risk each system is better than trading the one original version at ‘x%’ risk.
Equity curve is indeed smoother.

System_A @ 2% risk is not as ‘good’ as System_A @ 1% risk PLUS System_A_SlightlyModified @ 1% risk.
Note that we still are risking 2% per strategy.

“Slightly modifiedâ€
Chris67
Roundtable Knight
Roundtable Knight
Posts: 1052
Joined: Tue Dec 16, 2003 2:12 pm
Location: London

Post by Chris67 »

i agree with rhc
there is a tendency to seek to add "strongly" non correlated counter trend / mean reversions systems to a portfolio of TF systems. Ive found that smoother curves can also be found by blending lots of TF systems together with subtle differences
If you tared 10 L/T tf systems together that all look to be correlated >0.85/0.90 you may be pleasantly surprised by the results / there and again you may not be annd you may certainly reach the sensible conclusion that the extra work / effort needed to run 10 systems (and believe me its hard) is not worth it to get a slightly smoother equiy curve that effectively produces the same CAGR
If I were only trading my money I would trade 1 system and be done with it - it all depends on your circumstanes/ objectives / AUM
Chris67
Roundtable Knight
Roundtable Knight
Posts: 1052
Joined: Tue Dec 16, 2003 2:12 pm
Location: London

Post by Chris67 »

I want to add a couple of screen shots - I have them saved as snagit files - how do i get them into a post ?
C


- - - - - - -

edited by Moderator -- A post in the Customer Support area covers this fairly well. It includes a .pdf document which is copied below.

"Screen Image Capture for Blox Users" (link)
Attachments
Screen_Capture_For_Blox.pdf
(228.33 KiB) Downloaded 723 times
Moto moto
Roundtable Knight
Roundtable Knight
Posts: 427
Joined: Mon Jun 01, 2009 4:12 am
Location: once again in the UK

Post by Moto moto »

"what types of systems would one trade along with a long term trend following system to smooth the equity curve?"

The holy grail of fund to funds.
A lot will clearly derive from what your ultimate objective is, assuming in this case to smooth the equity curve.....in which case maybe trend following should only form a small part of the suite of systems whilst other systems with higher winning percentages dominate?

these do not need to be systems that are necessarily mean reverting but might be systems that take profits far more quickly. Plus how do you define a trend - I always want to trade with momentum which might be the short term trend without picking tops and bottoms.

(no sure if this helps as I am not completely systematic and so use a lot of discretion, plus I need to draw an income from my trading - so this is what I use as a vast majority of my day to day trading - high leverage, high turnover hit it and quit it trades for income (not too excessive turnover - more swing trading), but my long term trend following is my savings pot. Overall this helps smooth my OVERALL personal equity curve until the savings pot becomes bigger than the short term pot, or I tire of the discretionary element. :) )
Chris67
Roundtable Knight
Roundtable Knight
Posts: 1052
Joined: Tue Dec 16, 2003 2:12 pm
Location: London

Post by Chris67 »

moto moto - after recently examining the correlation between my short term discretionary trading and a basket of instruments including - grey hair, taste of bile in mouth, indegestion problems, sleep problems and general irretability combined with genral IBS - I decided teh discretionary element had to go !!! :D
Anthony Marsland
Roundtable Fellow
Roundtable Fellow
Posts: 83
Joined: Thu Jan 13, 2011 5:33 am
Location: London
Contact:

Multiple sub-systems

Post by Anthony Marsland »

Hi Sluggo,

My experience concurs with rhc. I have split my FX Futures Day Trading System into 6 sub-systems per instrument. Each sub-system trades an entry, risk stop and target at a different price to the other sub-systems.

This has increased Sharp by 27% and added the benefit of keeping, what would have been, double digit orders at a price well under ten lots at a price. Smaller lot sizes = smaller slip, about half a tick (Euro, Yen, Pound) per RT.

My personal business plan has me designing/trading systems within different time frames and different markets to achieve as smooth an equity curve as possible.

E.G. Reversion to mean stocks (Day Trade) a la Connors, LTTF as per Sluggo and AJ Gardener, ETF Portfolio Rebalance (One Month) inspired by Ken Long.

With the positive experience of splitting my FX Futures system into sub-systems I will probably look to do the same with all my systems. As an example: With the ETF Rebalance System I may choose to trade 1st-31st for half the capital and 15th-14th for the other half.
sluggo
Roundtable Knight
Roundtable Knight
Posts: 2987
Joined: Fri Jun 11, 2004 2:50 pm

Post by sluggo »

rhc wrote: ... blending 2 slightly different versions of the system that I’m currently using with ‘1/2x%’ risk each system is better than trading the one original version at ‘x%’ risk.

Equity curve is indeed smoother.
The pros do this too, they call it "Trading a system at several different speeds". For example, trade 20-day breakouts AND trade 40-day breakouts AND trade 80-day breakouts AND trade 160-day breakouts. It is one way among many to skin the cat.

Richard Spurgin of Clark University fooled around with a very simple 1-parameter trading system, and measured the equity curve correlation (to a benchmark equity curve) as he varied the system's speed parameter. He found that changing the speed does indeed change the correlation, and some of the correlations are surprisingly low. Perhaps surprisingly, this gets you some diversification. I've annotated his figure 3, below. Others have made similar tests and gotten similar results, (link)

Spurgin's system:

Code: Select all

If Close(today) > Close(PARAM days ago) then Reverse To Long tomorrow at market on open
If Close(today) < Close(PARAM days ago) then Reverse To Short tomorrow at market on open
This system can be blindingly fast in Trading Blox, because it can be written to use ONLY Indicators (which are precomputed in Blox and therefore, Damn Fast). Specifically, "Bar History Value" and/or "Calculated" Indicators.
Attachments
Spurgin_ex3.png
Spurgin_ex3.png (42.87 KiB) Viewed 13482 times
absret111
Senior Member
Senior Member
Posts: 38
Joined: Mon Mar 29, 2010 8:01 am

Post by absret111 »

Regarding this discussion it may be interresting to have a look at the following IASG website:

http://www.iasg.com/groups/group/conque ... res-select

You can see the performance of the Conquest Managed Futures Select Fund.
The trading methodology is based on a diversified set of Nth-day break-out systems.
rabidric
Roundtable Knight
Roundtable Knight
Posts: 243
Joined: Mon Jan 09, 2006 7:45 am

Post by rabidric »

i have looked at diversified multiple n-day breakouts(i.e. 4+ per product, not 1-2) in the past. My conclusion was that it was horribly inefficient in terms of capital usage, and the diversification was not as spectacular as i would have hoped for..
i.e. it is diversified most of the time and makes rubbish returns then, or everything ends up fully committed in one direction or the other during a period of heightened global market correlations(-ve usd, +ve "risk-appetite", or vice-versa) , and you get the massive runups and drawdowns that come with it.

B-¦
Anthony Marsland
Roundtable Fellow
Roundtable Fellow
Posts: 83
Joined: Thu Jan 13, 2011 5:33 am
Location: London
Contact:

Correlation

Post by Anthony Marsland »

Following on from Sluggo's comments regarding Richard Spurgin here are my FX Futures system correlation results to an e-mini system in the same portfolio.

0.855 Original System Settings for Pound
0.839
0.838
0.850
0.874
0.863
0.733 Original System Settings for Euro
0.693
0.648
0.654
0.672
0.685
0.833Original System Settings for Yen
0.843
0.842
0.831
0.847
0.857

The incremental differences really add up to a powerful reduction in pain to gain ratio.
SimJimons
Senior Member
Senior Member
Posts: 31
Joined: Mon Jun 20, 2011 9:59 am

Post by SimJimons »

We trade 40 different time horizons in our trendfollowing models, with an average correlation between adjacent horizons of about 0.985 (daily observations). Depending on which frequency you use to calculate correlations the correlation between the shortest- and the longest-term system is something like 0.6 to 0.7 (if I remember correctly). Clearly not a Holy Grail, but it definitely helps on the margin. This is especially true if we forget about correlations for a while and study the drift instead, which may be very different (and cyclical) even though correlations are fairly high.

Obviously, adding countertrend systems, or very short-term systems, would give more bang for the buck than simply including additional trendfollowing systems (long-term ones, that is). However, it is usually a little more difficult to find well performing and robust systems in these areas. And while I too belive that correlations are much more robust than measurements of risk-adjusted return, I'm not willing to gain de-correlation at any price. In my view, the most difficult question to answer when blending two very different systems is their relative "curve-fitness". And, according to my belief and experience, trendfollowing systems are the least curve-fit systems there are (in general), which means that stretching beyond these systems may not be a strategy suitable for the average investor (not that I'm saying you are average :) ). And if one subscribes to this notion, then mixing it up with a few extra trendfollowing time-frames may be the most prudent thing to do.

Just one more thing. The small investor obviously has to look at the aggregate signal (what sluggo called "voting" HERE) instead of treating each signal separately. If not, he will likely experience the "horribly inefficient capital usage" rabidric talked about.
sluggo
Roundtable Knight
Roundtable Knight
Posts: 2987
Joined: Fri Jun 11, 2004 2:50 pm

Post by sluggo »

Let us not forget soon-to-be professional trader LeviF's recommendation: Simultaneously trade several copies of your trend-following system, where each copy uses the exact same entries but different exits, arranged thus:
  • Let copy #1 be the original, unmodified system which Lets Profits Run a la trend-following
  • Let copy #2 exit at a profit target of "Z" R-multiples (do NOT let profits run) via a Limit order
  • Let copy #3 exit at a profit target of "Y" R-multiples (do NOT let profits run) via a Limit order
  • Let copy #4 exit at a profit target of "X" R-multiples (do NOT let profits run) via a Limit order
  • etc.
This general idea goes by the name "scaling out" and/or "Taking Profits". As you can easily see, it has zero diversification-of-entries and quite a bit of diversification-of-exits. LeviF has advocated this more than once: (link 1) , (link 2) , (link 3)

LeviF points out that you can allocate different percentages of your capital to these various system copies. For example you can have just two copies (copy1 = the trend-follower and copy2 = the profit-taker), then trade 25% of your money on the trend-follower and trade 75% of your money on the profit-taker. Or invent your own weighting scheme; LeviF reminds you that the weights need not be equal.

The vendor-sold system "Aberration Plus" scales out and uses different weights for the various system copies (link 4)

edit#1: fix typo
Last edited by sluggo on Sat Jul 23, 2011 10:27 am, edited 1 time in total.
LeviF
Roundtable Knight
Roundtable Knight
Posts: 1436
Joined: Mon Dec 22, 2003 12:24 pm
Location: Des Moines, IA
Contact:

Post by LeviF »

sluggo wrote:Let us not forget soon-to-be professional trader LeviF's recommendation: Simultaneously trade several copies of your trend-following system, where each copy uses the exact same entries but different exits, arranged thus:
  • Let copy #1 be the original, unmodified system which Lets Profits Run a la trend-following
  • Let copy #2 exit at a profit target of "Z" R-multiples (do NOT let profits run) via a Limit order
  • Let copy #3 exit at a profit target of "Y" R-multiples (do NOT let profits run) via a Limit order
  • Let copy #4 exit at a profit target of "X" R-multiples (do NOT let profits run) via a Limit order
  • etc.
This general idea goes by the name "scaling out" and/or "Taking Profits". As you can easily see, it has zero diversification-of-entries and quite a bit of diversification-of-exits. LeviF has advocated this more than once: (link 1) , (link 2) , (link 3)

LeviF points out that you can allocate different percentages of your capital to these various system copies. For example you can have just two copies (copy1 = the trend-follower and copy2 = the profit-taker), then trade 25% of your money on the trend-follower and trade 75% of your money on the profit-taker. Or invent your own weighting scheme; LeviF reminds you that the weights need not be equal.

The vendor-sold system "Aberration Plus" scales out and uses different weights for the various system copies (link 4)
Gee, I'm blushing. Haha.
Chuck B
Roundtable Knight
Roundtable Knight
Posts: 481
Joined: Thu Apr 17, 2003 6:34 am

Post by Chuck B »

sluggo wrote:Let us not forget soon-to-be professional trader LeviF's recommendation: Simultaneously trade several copies of your trend-following system, where each copy uses the exact same entries but different exits, arranged thus:
  • Let copy #1 be the original, unmodified system which Lets Profits Run a la trend-following
  • Let copy #2 exit at a profit target of "Z" R-multiples (do NOT let profits run) via a Limit order
  • Let copy #3 exit at a profit target of "Y" R-multiples (do NOT let profits run) via a Limit order
  • Let copy #4 exit at a profit target of "X" R-multiples (do NOT let profits run) via a Limit order
  • etc.
This general idea goes by the name "scaling out" and/or "Taking Profits". As you can easily see, it has zero diversification-of-entries and quite a bit of diversification-of-exits. LeviF has advocated this more than once: (link 1) , (link 2) , (link 3)

LeviF points out that you can allocate different percentages of your capital to these various system copies. For example you can have just two copies (copy1 = the trend-follower and copy2 = the profit-taker), then trade 25% of your money on the trend-follower and trade 75% of your money on the profit-taker. Or invent your own weighting scheme; LeviF reminds you that the weights need not be equal.

The vendor-sold system "Aberration Plus" scales out and uses different weights for the various system copies (link 4)

edit#1: fix typo
Realize this will be a real balancing act in many ways and will alter the basic premise underlying the trend following system (however, I'm not advocating against this method...just blabbing a bit here). What happens to the net outcome with a method like above is that for losing trades you take your maximum loss (i.e. -1R), or perhaps said another way, you realize you normal loss outcome (perhaps on average something like -0.54R as your exit moves up some from entry point on most losing trades), BUT you cut short your chance to realize the whooper R-multiples in huge trend trades that make or break a trendfollowing system.

I did years of research on this type of thing back about 15 years ago, all based on R-multiple exiting, and the net of it was finding what R exit value to profit take a portion (and what portion) of the trade was highly dependent on the amount of data you use for the test (duh), what markets were used in the portfolio (double duh, I know), etc. The big outlier black-swan whatever you want to call them 15+R trades (where R was initially a wide value to begin with) that come along every once in a while would be truncated (partially exited) while the other everyday junk stuff that happens most in a LTTF system would be realized (i.e. that average loss R).

All that said, there is still room for "some other input" to manage open position size imo...I've long used Basso-type risk reduction that limits open risk per asset as a percent of total equity. Measures that look at portfolio open risk, asset level and total, as a function of equity and make decisions based on those values seem to be more robust over the long term than arbitrary R-multiple exit targeting.
LeviF
Roundtable Knight
Roundtable Knight
Posts: 1436
Joined: Mon Dec 22, 2003 12:24 pm
Location: Des Moines, IA
Contact:

Post by LeviF »

Chuck B wrote:I did years of research on this type of thing back about 15 years ago...
This post comes to mind:

viewtopic.php?p=35018&highlight=crosshairs#35018
svquant
Roundtable Knight
Roundtable Knight
Posts: 126
Joined: Mon Nov 07, 2005 3:39 am
Location: Silicon Valley, CA
Contact:

Post by svquant »

If you are really interested in correlations of trading strategies then look up the works by Emmanuel Acar and Pierre Lequeux where they derive what the correlations between trading rules (MA) based on length should be under various time series properties and then look at some of the empirical data to see if that holds with FX rates.

If you are looking to only pick 2-3 MA's in your system it helps guide you in terms of minimizing correlation on a per instrument basis. This is how they derive using a 3 MAs of 31, 62, 117 days in the AFX currency index.

It is a bit different in thought than using a lot of near and highly correlated parameters and a bit more in the spirit of making parameter choices closer to equidistant from each other on a %-age basis. This is how MFS works and many other going back to at least Perry Kaufman in the early 1990's have advocated.
rabidric
Roundtable Knight
Roundtable Knight
Posts: 243
Joined: Mon Jan 09, 2006 7:45 am

Post by rabidric »

svquant wrote:If you are really interested in correlations of trading strategies then look up the works by Emmanuel Acar and Pierre Lequeux where they derive what the correlations between trading rules (MA) based on length should be under various time series properties and then look at some of the empirical data to see if that holds with FX rates.

If you are looking to only pick 2-3 MA's in your system it helps guide you in terms of minimizing correlation on a per instrument basis. This is how they derive using a 3 MAs of 31, 62, 117 days in the AFX currency index.

It is a bit different in thought than using a lot of near and highly correlated parameters and a bit more in the spirit of making parameter choices closer to equidistant from each other on a %-age basis. This is how MFS works and many other going back to at least Perry Kaufman in the early 1990's have advocated.
LOL, the result of their "works" is what? MA's that are basically double/half each other's lengths? clearly genius at work there... :)

Academic works on trading can be so funny. imho they often cover the kind of analysis that most TBB users do in an afternoon on an average day...and still don't come up with anything useful 99% of the time.
svquant
Roundtable Knight
Roundtable Knight
Posts: 126
Joined: Mon Nov 07, 2005 3:39 am
Location: Silicon Valley, CA
Contact:

Post by svquant »

radidric

I obviously simplified and misrepresented their works to elicit such a response. It is up to you to decide to use or ignore the information and work they have generously shared while still working in the industry (not academia) and managing some significant funds and allocations on various bank prop desks and via internal hedge funds.
Post Reply