Many systems + few instruments? Or, Few Sys + Many Instr?

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
bobsyd
Roundtable Knight
Roundtable Knight
Posts: 376
Joined: Wed Nov 05, 2008 2:49 pm

Many systems + few instruments? Or, Few Sys + Many Instr?

Post by bobsyd » Mon Nov 08, 2010 8:03 pm

Assuming Sluggo trades his 156 market “kitchen sinkâ€
Attachments
fcalc.png
fcalc.png (6.54 KiB) Viewed 8214 times
Fills Calculator.xls
(20.5 KiB) Downloaded 271 times

td80
Roundtable Fellow
Roundtable Fellow
Posts: 60
Joined: Thu Mar 08, 2007 8:38 pm
Location: Irvine, CA

Post by td80 » Wed Nov 10, 2010 1:37 pm

I think bobsyd sort of hints at an interesting question.

Assuming you have limited capital initially, is it better to diversify multiple systems across a narrower set of instruments, or is better to diversify a broad set of instruments across a system or few?

This question has implications on all of your other ones.

My personal (with not a real strong empirical backing mind you) opinion is to go with a low system count high instrument count if you have limited capital. Something simple and robust, nothing too exotic or with a lot of moving parts. This of course may indicate my LTTF/Position Trading bias. I like big % wins rather than win % because I hate costs and I am becoming less and less of a fan of dealing with execution (what I naively thought was the exciting and cool part when I was a newbie, I was gonna TRADE!).

The other advantage of starting off with a broad selection of instruments is that you are not constrained by liquidity the same way you will be once you have a large capital base. This gives you one of the only advantages you can get when being a small fry in this game.

I think the ultimate goal for many of us would be to get to where Sluggo (amongst others) is in terms of broad set of systems and instruments, but you need a lot of capital to really make it work.

I would be curious to hear what some of our grizzled old-timers have to say about their early days, and how they approached this issue, assuming they can remember that far back. :o

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

Post by sluggo » Wed Nov 10, 2010 3:23 pm

Twenty seven years ago, Eckhart and Dennis taught the Turtle System to their first class of students. It included an idea (circled in red, see Figure 1 below) to which very few people pay attention. That idea is:
  • You don't have to take every entry signal. It's okay to skip some entry signals and not take those trades. In fact, not taking some of your system's trades can be a very good idea!
The Turtle System included max-position-count limits. Think about this for a minute. If your system has a maximum number of allowed positions, you can trade it on an infinitely large portfolio. Furthermore, you can do so without taking infinite amounts of risk.

You can limit total position count, you can limit sector position count, you can limit total portfolio heat (risk), you can limit sector heat. Each of them results in skipping trades and avoiding risks. They let you trade immense portfolios without immense amounts of capital.

System vendor Dean Hoffman includes this idea in some of his products. He describes it as a "Parking Lot" in which there are a fixed, small number of parking spaces (permitted trades), and a potentially large number of autos (entry signals) swarming around: (LINK)

A few years ago, I uploaded a trading system called "Thermal BBBO" to the Blox Marketplace where other Blox customers can download it. It limits both total portfolio heat, and sector heat, skipping new trade entries if the system is above either heat limit: (LINK)

It's been known for 27 years: You can trade an immense portfolio without having an immense account. All you gotta do is skip some trades.
Attachments
Turtle_pic.png
Figure 1. Max # positions in the Turtle system
Turtle_pic.png (7.32 KiB) Viewed 8042 times

nodoodahs
Roundtable Knight
Roundtable Knight
Posts: 218
Joined: Wed Aug 09, 2006 4:01 pm

Post by nodoodahs » Wed Nov 10, 2010 3:44 pm

A suite of systems on a bunch of instruments could be made to work with limited capital if one of the features of the systems was a “gradingâ€

td80
Roundtable Fellow
Roundtable Fellow
Posts: 60
Joined: Thu Mar 08, 2007 8:38 pm
Location: Irvine, CA

Post by td80 » Thu Nov 11, 2010 4:19 pm

Dynamic portfolios still have issues guys. I mean what is "best" in terms of instrument or signal at any given time? You are treading into the realm of relative ranking (which I am interested in, and use to effect in equities). You are still trading a small set of instruments, you are just swapping out the slots based on some systematic decision making.

This may or may not yield the desired "compromise" of being able to run many systems across a (theoretical) broad set of instruments whilst having limited capital, versus having more desirable results with fewer systems but a big portfolio taking all signals.

I think you may find this hybrid approach suggested above works better for shorter term systems that have more trades and thus are not dependent on catching one or two big moves a year.

nodoodahs
Roundtable Knight
Roundtable Knight
Posts: 218
Joined: Wed Aug 09, 2006 4:01 pm

Post by nodoodahs » Thu Nov 11, 2010 4:36 pm

td80 wrote:I mean what is "best" in terms of instrument or signal at any given time?
That sounds like something determined by testing in the context of your system(s).
td80 wrote:This may or may not yield the desired "compromise" of being able to run many systems across a (theoretical) broad set of instruments whilst having limited capital ...
If you find a differentiator to use in ranking trades for a system, it WILL allow you to run that system against more instruments with some given amount of capital. If you DO NOT find a differentiator, then it won't, and running the system with limited capital means limiting your number of instruments.
td80 wrote:... versus having more desirable results with fewer systems but a big portfolio taking all signals.
Again, this sounds like something to spend some time testing in the context of your systems.
td80 wrote:I think you may find this hybrid approach suggested above works better for shorter term systems that have more trades and thus are not dependent on catching one or two big moves a year.
If momentum is used as a differentiator in a trend-following system, you WILL catch every big move, just not ALL of the move.

alp
Roundtable Knight
Roundtable Knight
Posts: 309
Joined: Mon Aug 27, 2007 8:09 pm

Post by alp » Tue Nov 16, 2010 7:21 pm

I find it interesting that there is a general bias to backtest systems using very large portfolios when actually only a small number of traders or funds will be able to actually trade such portfolios.

I suspect that with unusually large and diversified portfolios almost everything works better and you're likely to get more impressive numbers and also be able to mix different systems or use staggered parameters sets so as to get smoother equity curves or, perhaps, better risk-adjusted returns.

Yet if you can only trade or just want to trade a small number of instruments then you'd better focus your research on what is relevant to you, i.e., your trading portfolio, and use larger portfolios for other purposes such as finding robust parameters sets or testing your system in a different scenario.

Whatever I think the answers you're looking for your research can only be found by doing research, i.e., backtests. And what really matters about the direction you take in your tests is what you want about them.

What do you really want? According to Bob Proctor this is the key to the Secret.

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

Post by sluggo » Tue Nov 16, 2010 7:38 pm

alp wrote: ... to backtest systems using very large portfolios when actually only a small number of traders or funds will be able to actually trade such portfolios.
If your system has a maximum number of allowed positions, you can trade it on an infinitely large portfolio. Furthermore, you can do so without taking infinite amounts of risk.

If you can only trade a small number "N" of instruments simultaneously, you can set the maximum number of allowed positions = N, and then trade a portfolio of (5 x N) or (30 x N) instruments. It's possible, you could do it; all you need do is modify your trading system code and then debug your modifications.

Besides diversification, there's another benefit: reducing the dynamic range of total portfolio heat (REF 1), for example as shown in Figure 1 of the Blox Marketplace system "Thermal BBBO" (REF 2). The unpredictability of total portfolio heat is vastly reduced.
Attachments
thermal_BBBO_figure_1.png
Total portfolio heat is kept within a narrow range
thermal_BBBO_figure_1.png (11.52 KiB) Viewed 7814 times

alp
Roundtable Knight
Roundtable Knight
Posts: 309
Joined: Mon Aug 27, 2007 8:09 pm

Post by alp » Tue Nov 16, 2010 8:10 pm

Thanks for pointing it out, Sluggo. I will check the concept of reducing the dynamic range of portfolio total heat while increasing the average heat more carefully.

lperepol
Senior Member
Senior Member
Posts: 30
Joined: Thu Feb 26, 2004 12:50 pm
Location: Castlegar, BC, Canada
Contact:

Portfolio Construct

Post by lperepol » Tue Nov 16, 2010 9:57 pm

I was wondering if anyone has tried to construct a portfolio using pairs?

For example:
Say System A has optimal parameters X
and
Say System B has optimal parameters Y

To keep things simple let System A, System B be single instrument systems. Combine both systems A and B into one system AB a (two instrument system).

Now run simulation on System AB. And save off the results.

Do this for all N instruments, which should result in N(N-1)/2 outcomes.
(There may be a few more outcomes if one wishes to count (distinguish) uniqueness of such a pair.)

I was thinking that perhaps this approach might work well for constructing a portfolio of instruments of more than 2.

Has anyone tried something similar to the above?
If so what where the results -- worthwhile? not?

svquant
Roundtable Knight
Roundtable Knight
Posts: 126
Joined: Mon Nov 07, 2005 3:39 am
Location: Silicon Valley, CA
Contact:

Post by svquant » Thu Nov 18, 2010 4:01 pm

Well somewhat related to this subject, I was rereading this article by Peter Muller of Morgan Stanley Process Driven Trading and thought it interesting that he prefers a deeper strategy vs a wider strategy, i.e. fewer better systems than many okay systems in order to maximize the IR. He specifically states,

"I would much rather have a single strategy with an expected Sharpe Ratio of 2 than a strategy that has an expected Sharpe Ratio of 2.5 formed by putting together five supposedly uncorrelated strategies each with an expected Sharpe Ratio of 1. In the latter case you’re faced with the risk that the strategies are more correlated than you realize (think Long Term Capital). There is also the increased effort of ascertaining whether each individual strategy really has a Sharpe Ratio of 1."

See the attached PDF - my favorite part of this article is Figure 2 reproduced below.
Attachments
Fig2 Muller.JPG
Fig2 Muller.JPG (27.23 KiB) Viewed 7671 times
Muller_Proprietary_trading.pdf
(98.97 KiB) Downloaded 609 times

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

Post by sluggo » Thu Nov 18, 2010 5:09 pm

I find it interesting that he feels "Max Drawdown" provides useful information (column 2, paragraph 4). Some Roundtable members disagree.

I also find it interesting that says he keeps an eye on the ratio (Net Profit) / (Total Commissions). I suspect this may be a red herring, intended to mislead reverse-engineers. But then again it could be a double double-cross, a fact disguised in such a way that it appears to be a lie, inducing the (wily) enemy to disregard it.

svquant
Roundtable Knight
Roundtable Knight
Posts: 126
Joined: Mon Nov 07, 2005 3:39 am
Location: Silicon Valley, CA
Contact:

Post by svquant » Thu Nov 18, 2010 5:22 pm

Perhaps it is an inverse double double-corss and he really watches (Total Commissions) / (Net Profit) :twisted:

Two things to keep in mind - is that they are heavily equity focused with an emphasis on stat arb type of strategies. So perhaps this is a way of steering away from strategies that live off of rebates or making sure you are picking up a nickel ahead of the steamroller not just a penny.

As for max drawdown most likely driven from the environment with respect to trading desk blow-up. The BoD or risk committee has a hard limit on these things and the newspapers are littered with stories of desks being shut down due to a large loss (maxdd).

Just some thoughts - no actual knowledge (or is that a double cross?)...
Last edited by svquant on Fri Nov 19, 2010 11:50 am, edited 1 time in total.

nodoodahs
Roundtable Knight
Roundtable Knight
Posts: 218
Joined: Wed Aug 09, 2006 4:01 pm

Post by nodoodahs » Thu Nov 18, 2010 5:51 pm

However, there is one advantage: because trading occurs less frequently it’s possible to lead a much better lifestyle than if you’re running shorter horizon strategies!

nodoodahs
Roundtable Knight
Roundtable Knight
Posts: 218
Joined: Wed Aug 09, 2006 4:01 pm

Post by nodoodahs » Thu Nov 18, 2010 6:01 pm

Sorry for the double-post, but some thoughts on DD.

Perhaps, just perhaps, DD is more stable for stab art types of strategies? This could be approximated somewhat through sims of much higher Sharpe return streams over time. I know from MC on both hypothetical "known" return streams and bootstraps from actual backtest return streams that with strategies and blends under Sharpe of 2, the DD is very unstable compared to estimates of standard deviation.

The investor interested in stab art is much more balance sheet fluctuation intolerant than the one interested in a LTTF CTA. As discussed elsewhere, regardless of what makes rational sense to the manager's money from the manager's perspective, the manager has to give the investor what they want (even if he thinks they're stupid to ask for it) in order to get the AUM. So it might be "useful information" solely from an asset-gathering perspective.

Or perhaps he's just never Monte Carlo'd his trades and/or return streams to see just how volatile a state like "max drawdown" is ... Naaahh! Couldn't be that!

Post Reply