Bradley and Chris,
Congratulations on your outstanding long term historical track record.
Very commendable!
I agree with you that the ignorance level is very high. Due diligence by potential subscribers is absolutely necessary, I can’t emphasize this enough. But alas, the horror stories will probably continue. May God bless us all!
Yes. I agree that due diligence is extremely important, since each "investor" is ultimately responsible for his/her results.
The stories have repeated for the last 3+ years that I have been here, and I expect that they will continue.
Speaking from experience, the school of hard knocks (such as losing money in trading) is one of the best teachers!
@CEO, Thanks for the congrats.
Ms Naivete,
thanks for the link. The paper has some serious flaws and its interesting that Ole Peters himself mentions this in his “Discussion” on page 11: "The assumptions made in this study are likely to lead to an over-estimate of optimal leverage: … certain knowledge of mu and sigma…"
Just one thing: The Kelly Criterion (http://en.wikipedia.org/wiki/Kelly_criterion) works for gambling-type games where your odds are known and constant. Both is not true if we deal with financial price series.
Its funny, this is a post 2008 paper. If Ole Peters had applied his results to a stock portfolio (based on the volatilities known before 2008) he would have gone bancrupt in October 2008 - because volatility suddenly rose, the sigma in his formulas changed dramatically.
And something else (which N.N.Taleb tries to make very clear in his books): In the real world of investments we don’t find gaussian probability distributions. We have “fat tails” instead. So the problems with this paper starts on page 2, Equation (1): … sigma * W_t …
sigma is not constant but varies heavily and W_t is not gaussian white noise but a distribution with (very) fat tails.
What I liked in this paper however was the idea of a “characteristic time scale t_c where the trend of [a series] becomes distinguishable from its fluctuations.”. This time depends on actual growth rate and volatility of an individual trading system and is quite a bit shorter for the systems on C2 than the 4-5 years mentioned by Ole Peters for regular instruments and portfolios.
This provides a nice answer to our discussions here about the “required age of a trading system” before we can judge its performance.
Did you notice that we can produce a combination of Ole’s Figure 2 and Figure 3 here on C2 for every Trading System?
Simply select the “Monte Carlo” mode below a system’s chart.
From this picture its possible to estimate the “age” where the “trend” becomes distinguishable from the “fluctuations”.
I’d name this “Ole’s Criterion”: Search the age of a system where the lowest Monte-Carlo equity curve goes above zero. It does not make much sense to judge a system younger than this age.
For example: For Topaz this age is about two years, for YES3 its about one year.
(Because Monte Carlo Simulations involve a random number generateor results vary a bit from run to run, try several runs and take an average)
And: Because Monte Carlo simulation needs some (reliable) material to work on, don’t try this with a trading system with less than 100 trades, results are simply not meaningful.
Rene,
I can only say "wow" for all that research and analysis you did in that short period of time.
Thanks for sharing your insight, I am sure it will be appreciated by many people.
Karl
Thanks Karl for the kind words.
I would summarize on:
Amount of returns = amount of risk
There are as many kinds of subscribers as developers. Everyone has to know what he wants.
The faster a trading system wins the faster it losses. Consistent wins are made on the long run. Personally I prefer systems that has moderate returns but with consistency in the long term. Not positive returns every month but every year.
I have just start two systems here on C2, one started wining and the other started losing, but maybe next month will be the opposite. The important thing for me is that the two systems will win over the time.
Regards
Rene,
I thought your theory on using the "Monte Carlo" simulation to determine the system trend, was very insightful. I agree that established systems can provide a better track record than a younger system.
However, potential subscribers should realize that nothing is guaranteed. A good example is ETF Timer, where the lowest Monte-Carlo equity curve went above zero close to the real time equity curve peak.
Rene’,
I found your posts very interesting. I was hoping you could elaborate on this comment you made:
"Just one thing: The Kelly Criterion(http://en.wikipedia.org/wiki/Kelly_criterion) works for gambling-type games where your odds are known and constant. Both is not true if we deal with financial price series."
Are you saying the Kelly Criterion should not be used for trading systems? I have found it to be very useful as a guide, but I trade significantly less than the Kelly bet (usually around 1/2) to account for imprecise measurement of the odds and the fat tails associated with the financial markets.
Have you found another method that works better for position sizing?
Thanks much.
Gary
Koch,
You obviously enjoy references to Dr.Taleb. I like them too.
Here is one more for you.
Dr. Taleb said:
âI think physicists should go back to the physics department and leave Wall Street alone.â
Are you familiar with it?
Do you agree with Dr. Taleb?
By the way, Dr. Peters had shown in his paper, as I mentioned earlier, that the characteristic time scale is between 4 and 5 years.
Here it is:
“Under the parameters of Fig. 1 this characteristic time scale is
somewhere between 4 and 5 years. Comparisons between portfolios with similar
stochastic properties are meaningful only on much longer time scales”.
So, as Ole had demonstrated, comparing systems that are as young as Topaz or younger does not make sense.
If you are really a scientist/physicist and understand the math in Ole’s paper, you should not be claiming that your systems are more reliable (less likely to crash) than others on C2, like you did in your promotional "Hint’ to me a few weeks back.
By the way, remind me, does your system hold a leveraged portfolio of stocks?
Hey Koch,
Here are couple of issues for you to consider:
1) You wrote: "And something else (which N.N.Taleb tries to make very clear in his books): In the real world of investments we don’t find gaussian probability distributions. We have “fat tails” instead"
Wow, are not you an energetic supporter of the Sharpe Ratio(SR)?
Are not you forgetting that SR uses the “gaussian pd”? What happened to the “real world” and the “fat tails”?
2) You wrote: "Because Monte Carlo Simulations involve a random number generator results vary a bit from run to run, try several runs and take an average) ".
Well, Koch, it looks like you are one of those who are “fooled by randomness”, as N.N.Taleb says.
You are telling us to use the random number generator, while our friend Taleb is strongly objecting to it in his books and papers.
He calls it a “ludic fallacy”.
N.N.Taleb clearly states that using probability theory( and therefore Monte Carlo simulation) for risk analysis is a fallacy.
Come on, Koch, you can’t have it both ways. You are either agreeing with N.N.Taleb or disagreeing with him.
Yes, ETF Timer is good example.
It demonstrates once again that Koch’s insight is based on the ludic fallacy.
Hi Gary
you wrote:
> Are you saying the Kelly Criterion should not be used for trading systems?
Yes. it is simply not made for trading. Resulting positions sizes are way too large (the “real” risk is underestimated). See the wikipedia article for tons of references, papers, articles…
> Have you found another method that works better for position sizing?
Yes. I use a constant relative position size (x% of available capital) which results in an expected drawdown of, say 10% for normal systems and 20% for “high risk” systems. The key here is to use expected drawdown instead of the usual drawdown calculated by your every-day simulation. A normal simple simulation gives you just one possible outcome for your trading system. But the same statistical properties of your system (Win%, avg winner vs. average loser, etc) can result in many slightly different equity curves with a range of possible drawdowns. (That is what C2’s Monte Carlo Display does).
Because we are not so much interested in past drawdowns but instead of possible future drawdowns it makes not much sense to focus on this one existing, observed, past drawdown.
It is much more important to get an estimate of such an possible future drawdown. This number comes form expected drawdown.
How to calculate this?
Glad you asked. The standard approach is called a “Bootstrap”. You generate a large number of new (possible) equity curves by combining existing parts of the one existing/known equity curve. Caclculate Drawdown for each of this “possible equity curves” and look at something like the 95-Percentile of these drawdowns - thats it.
BTW, this is closely related to C2’s “Probabilities of future account loss” shown in a systems performance statistics. To put the things above differently: When devloping a system (and selecting a position size) I try to get a 5% chance of 10% account loss. (or a 5% chance for 20% loss for agressive settings)
Koch wrote "Glad you asked. The standard approach is called a "Bootstrap". You generate a large number of new (possible) equity curves by combining existing parts of the one existing/known equity curve. Caclculate Drawdown for each of this "possible equity curves" and look at something like the 95-Percentile of these drawdowns - thats it."
Here, once again, he offers us another statistical method - the well-known Bootstrap. He, however, forgets that Bootstrap is still based on randomness( "You generate a large number of new curves…")
Once again, Koch is "fooled by randomness".
Ms. N…
I don’t believe that Koch is “fooled” or once again forgotten.
I am one of the first to jump on a vendor making unsubstantiated claims. However, any vendor that has the track record of some of Koch’s systems deserves due consideration.
I think an exceptional system should include a very smooth equity curve with low drawdown % over a history which includes 2008. And, of course, an annual return greater than the S&P. How would I find these systems on C2? Is this the same as a high Sharp Ratio?
Don’t be "fooled by Ms Naviete"
I wonder why Fred Pearce does not bash Rene directly. Instead he relies on his pseudonym as decribed in these threads by Matthew…
http://www.collective2.com/cgi-perl/board.mpl?want=listmsgs&boardid=9323885&threadhilite=9112&displayblock=
http://www.collective2.com/cgi-perl/board.mpl?want=listmsgs&boardid=9323885&threadhilite=9113&displayblock=
Ceo,
Just read carefully what Koch had written here about risk managers being “fooled by randomness”. The same applies to him - he is obviously fooled by it.
Koch’s record, no matter how long, does not make him immune from the statistical fallacy in risk management as defined by N.N. Taleb.
Besides, Koch is an admirer of Taleb. Koch was the one who brought Taleb’s books into this discussion.
If you fail to see the contradictions in Koch’s posts, it is your choice.
You see, Koch claims that he is a scientist and I am having a public scientific discussion with him here.
So, Koch’s record evaluation is up to you, and has nothing to do with the current discussion.
Yes, don’t be fooled by Ms. Naivete.
Instead, you, if you prefer, be fooled by M.Buster, who can’t tell the difference between a scientific discussion and “bashing”.
I guess, it is because he only does “bashing” himself.
I can’ t post here as Fred, simply because I am not him.
But, then what is in the name?
Call me Fred, if you so prefer.
Koch,
You wrote as follows:
"Just one thing: The Kelly Criterion (http://en.wikipedia.org/wiki/Kelly_criterion) works for gambling-type games where your odds are known and constant. Both is not true if we deal with financial price series.“
Be careful here. Anybody can read Wikipedia.
Have you actually read the original paper by J.L. Kelly
"A New Interpretation of Information Rate” published in 1956?
If you still have not read it, consider reading it. That way you would be able to provide a more accurate information here.
Because, the author is talking about a very special gambler case, not a regular casino gambler. He considers the gambler “with a private wire”.
That is the gambler that has a certain information on a chance situation before it becomes common knowledge.
So, the gambler is almost “an insider trader”, not a casino player.
Therefore, Kelly’s insider trader’s capital grows at a different rate than the one of the regular casino player.
Reading Kelly’s paper will also help you to understand Ole’s paper at a much deeper level.
Have fun.
