backtesting the original turtles

[mind]

Quote from inflector:

mind,

I wrote the rules document on the www.originalturtles.org site, founded that site with a few others, traded the rules as a Turtle for the duration of the Turtle program, and have done fairly extensive recent testing on that system and related ideas.

I wasn't sure when I began testing it how well it would hold up since I had been out of the markets for about 15 years.

I don't find the same results you did at all. While I think you can do better, it generates pretty decent positive returns in my tests over the last 10 years.

Did you test all the rules? In particular, did you test the correlation limits, the pyramiding, and the "Last Trade is Loser" rule?

These have a surprising effect on the performance of the system.

What software did you use to do your testing?

Regards,

Curtis Faith

hi curtis

i thank you very much first for the turtle file and second for posting here. i have a [small] team here and two people deal with that subject at the moment. they did program all of it on our prop C++ surface. inlcuding correlations, though we gave the correlation our own levels [and tested several others].

the two of them will enter this thread today or tomorrow and discuss problems in detail. honestly speaking i am both surprised and grateful for your participation here.

i see the turtle as a starting point for digging deeper. but i was not willing to dig where i do not see value. this is the actual reason for this thread.


good luck to you.


peace

 

[mind]

Quote from Walther:

Any BO system and or method will produce large 30% + drawdowns . If you do not like drawdowns, you have to introduce additional rules to the BO system to eliminate bad entries .That is all. turtle is a good method but it was designed while ago when people were not that jumpy like we are now. Turtle method idea is still valid it just needs little tweaking to bring it up to date.

i do not put all eggs in - so vola is fine for me.

 

[mind]

Quote from damir00:

i wouldn't mind seeing some of the turtle code some of you are testing, because i've tested it myself and it looks fine to me. there is an enormous class of trend-following systems that work over the long run, the turtles are just one of them, the key to all of them is money management and strict discipline.

it is also important to note the turtles were applied simultaneously across a wide variety of markets: that is a vital component of any trend-catching system. and running monte carlo against turtles to characterize the market movements producing poor short-run results is also very illuminating.

interesting post. i am not sure i really got the essence of the monte carlo statement ... . nevertheless we are quite motivated to find the mistake we [obviously] made. my two people would crucify me if they did not have the opportunity to find the mistake on their own. so ... thank you very, very much for the offer. and really might come back to it, if i am allowed to. but they demand [and deserve] the chance to fix it up ...

thanx damir for posting

 

[mind]

Quote from darkhorse:

i'd be curious to see the results of the turtle system backtested for the same time frame against a portfolio of Nasdaq 100 stocks from then til now.

intelligent asset allocation is a big piece of the puzzle imho.

p.s. wouldn't it be possible to add some filters to take into account the 'macro trendiness' (for lack of a better term) of an asset class?

i will move there once have indication that we are not completely dumb in what we are doing on the futures. nevertheless i doubt the stocks will be enlightening since hardly any cta could make it in stocks. theb reason IMHO is, that in the end stocks are not diversified enough.

if we do such test we must as well be very carefull about survivorship- and indexBias. is a huge issue on indices in general on high vola stocks in special.

thanks for posting

 

[inflector]

I did quite a bit of stock testing a while back using variants of the Turtle System.

Since the system is more of a medium-term trend-following system rather than a long-term one, I don't think survivorship bias is a big issue. Nevertheless, we did tests that used volume as the cutoff criteria rather than any particular index component list to partially eliminate this problem.

Our results indicated that a Turtle-Type system did not perform well unless you narrowed down the portfolio. Taking every signal performed about as well as the market in general. Taking short signals was a loser. There is too much upward bias in the stock market.

However, taking signals on only the strongest markets for longs only worked reasonably well and out-performed the market by a significant margin.

We tried various market-strength filters and the one that worked well was sorting the markets by the price difference in the last year divided by the Average True Range of the market one year previous, if I recall. Taking the signals from the top-third works pretty well.

- Curtis

 

[Walther]

Quote from mind:

i do not put all eggs in - so vola is fine for me.

Even I do not agree with you ( that 30% + drawdown will make your boss feeling all fuzzy inside, lol ), I will fight for your right to have your eggs crushed .

 

[mind]

Quote from Walther:

Even I do not agree with you ( that 30% + drawdown will make your boss feeling all fuzzy inside, lol ), I will fight for your right to have your eggs crushed .

do not get the meaning of this post.

 

[mind]

we forgot the rule to ignore the next trade after a winner. changes the picture. we are now at modified sharpe ratios of 0.74 for the period 1994 to 2004. we have some stupid spikes embedded, due to excessive use of leverage. we will fix that by capping the max leverage. then we should come near 0.9 or something, which is the starting point i expected in the first place.

the thing with the next trade encourages a countertrend trade after a winner - will check this out. another issue will be interest rate trends. from my point of view there are special situations, where all interest rates trigger trades at almost the same time. these trades have specific features and i would like to address that in a different way than others.
in general we will try to add some filters to bring sharpes up to 1.5, then start a small and see how things turn out.

thanks to all posters here for participating.

 

[DVB]

So you are saying that the trades in the turtle system have negative serial correlation; that the correlation is statistically significant, and hence the trades are not independent?

Quote from mind:

we forgot the rule to ignore the next trade after a winner. changes the picture.

 

[inflector]

Ah yes....

Serial correlation....

When you think about it, markets are run by humans who have among other attributes two that affect trading: emotion and memory.

Combined these create a serial correlation in participant behavior which has a corresponding effect in the markets.

Absolutely, positively there is serial correlation in the markets and in trades for some types of entries, and perhaps most types.

This is one of the reasons that I've never been a huge fan of Monte-Carlo simulation. It tends to underestimate the risk of a long series of losing trades because the way most people do it assumes that trades are indeed independent.

These series happen more in reality than in simulation because humans are involved in the trading and after a good period of trading you tend to get a longer than normal, and longer than one might expect given a random simulation, series of losing trades as traders chase after the same thing that was making them oh so much money so very recently. Especially those traders who jumped aboard late and are trying to make up for lost time and opportunity.

Most of the statistical math traders use also assumes independence, and conclusions made on that basis are wrong in many cases.

- Curtis