System Longevity

[td80]

Let me first get this out of the way: Oh no another parameter/robustness thread!

That said. I've been thinking about longevity in this business of ours (financial speculation).

This message may be a bit of a rant but I hope to inspire some of our resident Maestro's to impart a few bits of wisdom

Below is what I'm currently referring to as my LTTF Fundamentalist Manifesto:

The longer I've been fortunate enough to be around and survive, the more I'm coming to the conclusion that the best way (of course in my personal opinion) to survive and thrive is:

Minimum number of parameters (particularly trading logic).
Maximum number of diversified Instruments (Portfolio of choices),
Fixed Fractional Money Management.
No portfolio level-filtering (yes, that's right, I said it all you MACD/MA filterers).
Lots of data to validate your system, broad parameter(s) results
Tune $ allocation based on projected (90%, 95% confidence?) drawdown instead of using correlation or portfolio risk metrics. If you're really conservative, just trade closed-equity.

I predict your returns over the long haul to be around 30-40% annualized with a MAR of around 1 and a Sharpe of around .9 if you are not over-trading, have enough starting capital to make a full go, and have a very broad range of diversified instruments to trade.

I believe the above can be achieved with just 1 (trade logic) parameter.

Have I become a fundamentalist? Can someone present an argument that would convince me to go down the add-more-whizbang's path? Oh you say but if you were a CTA you can't have 40% drawdown? Well how about 10% returns, 9% drawdown? That RAR sucks you say? I predict if you trade 5+ years with a whiz-bang method that you will watch your Sharpe go under 1 and your MAR probably under 1 as well. The more parameters you have, the more wonky the future may be for you. The better the RAR looks now, the worse it will be in the future.

I wish to close with the bold statement that the Original Turtle method was over-optimized. I don't mean to disparage their students or the founders of the system. Quite the opposite, as they were bold pioneers and many of us are simply copying/tweaking what they found a long time ago in some shape or form. I'm simply benefiting from hindsight and a lot more data/computing power compared to what they had to work with at the inception of the system.

I think for longevity and robustness we must capture the essence of what a lot of us see in the data without over-engineering and trying to eliminate risk but still capture return. We get paid for taking risk!

I hope this starts a rancorous, if not constructive, debate that will be educational for all!

[sluggo]

You appear to dodge the question: Change the suite of systems from time to time? Or leave it alone, and trade the same collection of systems with the same subdivision of risk, forever?

Was Charles D. Robusti right?

[td80]

If your system(s) is simple enough, there are only two things to change.

1. Portfolio of instruments (add/reduce in a mechanical fashion based on ability to trade [regulation, broker availability,etc] and liquidity).

2. Very limited parameter sets (in my example of a single system, 1 parameter, there is only one thing you could change/reoptimize over time, and it is unlikely to change much on the long end of the parameter value). Of course anything is possible (I tend to subscribe to Popper on the topic of absolute certainty about anything).

And for Extra Credit: Add new systems? Maybe. This becomes an interesting notion. On the one hand, I have say, 1 LTTF that I feel could work for a very long time nearly unfettered except for instrument addition/reduction, yet if I add newer (presumably uncorrelated systems) we often seem to get improved risk/return metrics.

The question is, do these newfangled systems meet the criteria of the manifesto? Do I accept that shorter-term systems are in dire need of more knobs (and are thus,,, BAAAD), but concede it is worth the risk of over-fitting on these shorties/medium termers/exotics for possibly better risk-adjusted returns?

The fundamentalist in me says caution is warranted. Limited Data available, many path-dependent variables astern...

Does Mr. Robusti have any further insights?

[td80]

Subsequently I wonder as a student of history (and IASG):

Why do the long term (still in business) operations that we have performance data on seem to have MAR's and Sharpe's below 1? Surely these people have significant resources to out-do the Robusti's?

Could it be they had gangbuster early performance (or marketing...) which attracted epic AUM which then limited their instrument selection which subsequently killed their risk/return metrics?

Or could it be that they serve institutional masters who obsess over risk/return to the point that they over-engineer and chase performance ironically causing a more problematic risk/return number?

Maybe a combination of both in that order? Should we strive to be the cockroach of speculators or the human? I know these become deeply personal questions in some respects but I feel it is worth exploring not just personally but also how others address them.

[sluggo]

Could it be that only the mediocre survive in the long term? The sub-mediocre die. And the better-than-mediocre, well, perhaps this: (LINK)

[td80]

I certainly agree that there are probably people out there with exceptional returns who are not reporting performance, however we also have the countless blowups/shutdowns from lackluster performance that are not in the data set anymore (or ever) as well.

So based on "just the facts", things seem murky at best.

[AFJ Garner]

As a colleague and I were discussing a while ago, when looking at MAR on IASG and comparing it to what you see in back testing, you should make sure you are comparing apples to apples. You must look at both your back tests and IASG figures GROSS of fees or NET of fees. EG, if you are comparing your back tests to a CTA's net results (post 2% and 20% fees) then make sure you also deduct fees of 2 and 20 in your back test. It can make quite a difference to MAR.

The following paragraph was added by some naughty little soul (presumably Tim or another moderator) who did not declare themselves. While the addition is useful, may I politely request moderators editing posts to please declare themselves and to add a note to the effect that the addition/alteration has been made by the moderator rather than the original author.

Here is a Trading Blox drop-in that lets you define your own fee schedule (2-and-20? 0-and-30? 4-and-44?) and then deduct those fees during a backtest: viewtopic.php?t=2998&highlight=fees . You can use Parameter Stepping to run your system with and without fees deducted, and see how performance measures (MAR, Sharpe, Ulcer Index, etc) are affected.

[Moto moto]

When it comes to managing other peoples money - its a different ballgame - marketing and contacts can make all the difference to size - and this is not dependant on performance.

Given the little I know about trend trading compared to others here, please correct me if I am wrong....

If the biggest drawdown is yet to come, then after a period of making money and producing performance numbers - you might view it to be better stopping on a high if you have performed?

[td80]

AFJ you raise a very valid point. Without fee disclosure on IASG it gives a new twist to our problem. 2/20 fee's over a 20 year period can be an enormous burden on CAGR/Total return. Let alone some others have an even higher fee structure! Some others have 2/20 but have a watermark. It certainly can become a sticky wicket as you all over the pond would say.

I still contend that with the a broad enough instrument selection, and a simple system(s) that captures the essence of what we see in futures data in general (non-normal return distributions over longer periods of time), that at the end of the race simple and diversified may very well trounce complicated and diversified. I'm not saying don't innovate, but just differentiate between over-engineering and capturing the essence of the data.

I believe the counterpoint to my argument would be to constantly be tweaking your systems and when you hap across something shiny and (apparently) uncorrelated to add it to the mix in some sort of personal or systematic way. I think there is risk here of mining the same data set 300 different ways. I will say however, that the temptation is so great especially once you start adding systems with a bunch of knobs because then you start to fear they might break. This be-gets adding more systems with knobs to help us sleep better at night...

[AFJ Garner]

I still contend that with the a broad enough instrument selection, and a simple system(s) that captures the essence of what we see in futures data in general (non-normal return distributions over longer periods of time), that at the end of the race simple and diversified may very well trounce complicated and diversified. I'm not saying don't innovate, but just differentiate between over-engineering and capturing the essence of the data.

Rightly or wrongly that is the attitude I have adopted.

[nodoodahs]

As a colleague and I were discussing a while ago, when looking at MAR on IASG and comparing it to what you see in back testing, you should make sure you are comparing apples to apples. You must look at both your back tests and IASG figures GROSS of fees or NET of fees.

DON'T FORGET TO TAKE OUT THE CASH!!!!!

IASG will include the collateral return. While cash is yielding sh-- these days, over the long run it's about +5% to the compounding, annually.

Your personal backtest results may, or may not, include the collateral return, depending on what you do to calculate your returns.

Typically the results of commodity strategies studied in research papers do NOT include collateral returns and only include "excess" returns.

[Tim Arnold]

presumably Tim or another moderator

Not me -- I always make myself known, loud and clear...

[sluggo]

the best way (of course in my personal opinion) to survive and thrive is:

Minimum number of parameters (particularly trading logic).
Maximum number of diversified Instruments (Portfolio of choices),
Fixed Fractional Money Management.
No portfolio level-filtering (yes, that's right, I said it all you MACD/MA filterers).
Lots of data to validate your system, broad parameter(s) results

Someone just uploaded a system to the Blox Marketplace that might fit your personality

• 1 parameter
• >150 global futures markets
• 24,000 trades (1/1/1985 thru today)
• Zero losing years, 1985-2009. 2010 is down 4% to date, but the year is young
• Avg hold time of winning trades = 75 days; avg hold of losing trades = 18 days
• Smooth equity curve long-term, yet just bumpy enough short-term to frighten wussy traders (longest Drawdown = 22 months, deepest Drawdown = -25%)

Test results and system code for Blox are (HERE) in the Blox Marketplace.

[td80]

Sounds like just my cup of tea, thank you Sluggo!

[altomas]

Hi Sluggo, td80,

I don't have access to the Blox market place and I am interested in what you mean by 1-parameter?

By my understanding that means the system would have to have a position in all 150 markets at all times?

I've just started my backtesting journey (after a couple of years learning to program and building a testing platform) and have been exploring very long-term dual MA systems with an ATR stop over large portfolio's - 100+ futures markets.

These systems, however have several parameters - 2 to determine entry, 1 or 2 for the stop, 1 for the % of equity risked per trade.

Then at the portfolio level the max allowed portfolio heat is another parameter.

That leads to a total of 5-6 parameters.

And this does not allow one to re-enter a stopped out trade (i.e. at a new high/low) or to take on a trade that had been rejected by hitting the max portfolio heat - another 2 parameters.

Is my understanding of what constitutes a parameter correct?

[Chris67]

Altomas - This is a long running argument
Choice of portfolio = parameter
Choice to trend follow = is that a parameter ?
The list is endless

If a system is a dula moving average cross over - Id suggest its a 3 parameter system - i.e. 2 MA's and a stop loss
The rest is money management / risk managemet
Howver everyone defines things differently ?

[sluggo]

I don't have access to the Blox market place and I am interested in what you mean by 1-parameter

Blox customers exchange and share code they've written in the Marketplace; everyone who has bought software from Trading Blox has access.

Figure 1 below is an image snipped from page 1 of the Trading Blox User's Guide, which anybody can freely download here (.pdf) It discusses the "parameters" of a trading system simulation; as you can see, the User's Guide treats any and every user-adjustable value or user-chosen setting as a "parameter".

Figure 2 below is also from the User's Guide. It shows the Simulation Parameters Editor for one particular trading system. This system happens to have one user-selected choice in the Portfolio Manager (top right), another user-chosen value in the Money Manager (middle right), and eight more user-specified values in the Entries and Exits section (bottom right).

Figure 3 below is a screen image of a trading system that I created in Blox. I lifted this system straight from page 55 of Chuck LeBeau's book Computer Analysis of the Futures Market , and coded it up in Trading Blox. As you can see, it has fewer parameters.

There are a couple things to notice about Figure 3. First of all, the system has no "stop". (Instead, it exits the day after price crosses the middle Bollinger Band.) Secondly, the system has no max portfolio heat (or max sector heat) constraints. Third, because there are no stops, there are no stopped-out trades, and therefore no rules about how to re-enter a stopped-out trade. Yet despite these absences, Figure 3 is a mechanical trading system which can be traded with real money in real life. Some traders may find it completely unacceptable to them; they may strongly believe they cannot possibly trade a system without a stop, or without max heat constraints. However other traders may feel differently; it takes a difference of opinion to make a market.

How many "parameters" does the LeBeau BBand system have? Different people hold different opinions. Some would say "four" because Figure 3 appears to have four user-adjustable knobs. Some would say "twenty three" because the user made twenty choices when she chose her twenty market portfolio; plus the three other parameters in the Figure. Other people (including me) would say "two", because only the bottom two user-chosen values adjust the mechanical system itself. And you can find other traders who hold other opinions. It's controversial. Search the Roundtable and you can find several earlier discussions, as mentioned in the previous post.

[td80]

Hi Altomas,

I would first like to say, welcome!

Now you are really jumping into the fire here with these sorts of interesting questions. You may find that I am usually in the middle of the fire with a bunch of dry tinder and a decent bit of hot air when it comes to parameter/curve fitting discussions.

I contend that not all parameters are created equal, and this alone (as Sluggo alluded to above) is a highly contentious point among various participants.

If you look back over some of our discussions here, (hopefully) you will see me put a sneaky little phrase in when I'm talking about low parameter count systems. I am referring to trade logic (entry and exit triggers) only.

My mind currently thinks like this (put sunglasses and noise cancelling headphones on now):

System Development (R&D):
Minimal trade logic parameter count (buy/sell/flat?)
Maximum number of liquid (parameter!) instruments
Maximum amount of historical data

Production System Implementation (Execution/Operations):
Include-> System from R&D
RISK (RISK, RISK, RISK!) Management Parameters/Risk Per Trade
Money Management Parameters
Operational Procedures (Did I just buy 55 contracts instead of 5? What do I do!!)
Optional - Dynamic portfolio asset allocation parameters

[altomas]

Chris67, sluggo, td80 - thanks for your replies, and the welcome!

I have already been reading many of your posts and am highly indebted to you all - hope to be able to contribute some value in the near future...

Sluggo - one day I plan to buy software from Trading Blox, but for now I feel that walking the pilgrimage of building my own software and getting down and dirty with the data is necessary for me to maintain confidence when the inevitable drawdown comes.

I've been toying with the idea of systems that don't have 'explicit' stops such as the BBand system you outlined - i.e. you can do something similar with X day breakouts.

td80 - I like where your philosophy is going, I want to travel a similar path of low complexity and many markets.

Methinks that I want the option of 'flat?' but at this stage all will be at the mercy of further testing - is it even possible to go flat using just one entry/exit parameter?

[sluggo]

In your opinion, does the following pseudocode define a zero-parameter system which goes flat?

Code: Select all

If (todays close < todays open) then blackcandle = true ; else blackcandle = false

If (blackcandle = true) then buy tomorrow at (todays high) on a Stop and exit at market on close