We need a statistics forum...

The reason trading is an art not a science is because there is no known statistical distribution that accurately describes price movements. It always changes and is ruled by the tails.

Engineering solutions don't wear out with time like trading systems. A P-51 Mustang is just as awesome now as it was during WWII, but old school trend algos have all but been scrapped for lack of funds. Is anyone putting real money on the Donchian systems in 2012?

This may be why there are no happy quants: the fruits of their long hours datamining for The Grail, all too often die on the vine....

:(
 
I'd love to see a cogent thread...
unfortunately, there are few who want to contribute any actual work...

:(

Dozens of books. I often re-read PDQ statistics (Streiner) as a short coffee table refresher.
If you can grasp the concepts within, it's a great fast practical read.


Till then you can always try ...
stack exchange
http://stats.stackexchange.com/
 
It seems like a good idea. I use stuff like that all the time for macroeconomics but I don't post my more hardcore stuff here because I didn't think anyone would be interested. This might be a good way to put that kind of work forward. I would moderate this forum if it came.
 
I wonder what the statistical probability is Baron will add one?

Maybe he should start a forum where it can be discussed.
 
Quote from Wide Tailz:

The reason trading is an art not a science is because there is no known statistical distribution that accurately describes price movements. It always changes and is ruled by the tails.

Engineering solutions don't wear out with time like trading systems. A P-51 Mustang is just as awesome now as it was during WWII, but old school trend algos have all but been scrapped for lack of funds. Is anyone putting real money on the Donchian systems in 2012?

This may be why there are no happy quants: the fruits of their long hours datamining for The Grail, all too often die on the vine....

:(
More...

Now see, this is how you can tell someone who's at least been down this road a bit.

I "discovered" this the hard way myself. Many traders mistakenly apply vague or intuitive statistical thinking to the development of trading systems. They imagine that there's a "population of trades" out there somewhere, and if only they could get a large enough sample, then they'd have statistical significance. But it doesn't work that way, most of the time. The "trade population" is infinite and always changing. A particular setup can work brilliantly for a couple of years and, then, BAM, you take an unprecedented 6 losers in a row.

Quick stats question: what are the odds that a setup with a 78% win rate over the previous 100 trades will suddenly take those 6 losers in a row? It's kind of a trick question...

Financial market behaviors are not normally distributed, meaning the usual stuff about t-tests, z-scores and all that bull**** don't apply...usually

Having said that, there are likely a very few areas where they will. Things like gap fills and breakouts (ACD method) come to mind.

Statistics is a bit like martial arts: Knowing a little can be dangeous as it can provide a false sense of security and may get you into trouble.
 
Quote from iagainsti:


Quick stats question: what are the odds that a setup with a 78% win rate over the previous 100 trades will suddenly take those 6 losers in a row? It's kind of a trick question...
More...

Stats was my worst class. But what I do remember was that you need a known distribution to even calculate future confidence, based on a known population. So your question is unanswerable unless wild assumptions about the distribution are made. This is how LTCM quants proceeded with their modeling - with 98% correct assumptions. It was like flying an airplane that would only fail 2% of the time.

:eek:

Hence my post...... I don't believe there is any way to calculate this stuff. Textbook stats don't apply.

If the Japanese suddenly made a robotic kickboxer, as soon as it hit the tournament circuit, fighters would eventually figure it out and exploit its vulnerabilities. AI, you say? Wake me when it finally happens. It's been hyped for 20 years and there are still no self-learning robots that do not require engineers to set up, teach, and continually tweak them - and this for manufacturing applications where the static laws of physics are in effect (unlike the changing tendencies of human behavior in groups).

Computerized chess playerz, you say? Chess is a closed system, a known distribution of possibilities. Real life is an open system.... there are no rules, other than love, hate, life and death.

Hence my ultimate conclusion about the trading game:

Traders who truly love the process of continual evolution are the ones who get interviewed after decades of brilliant performance in all market conditions. PTJ was a golden gloves boxer before a floor trader & hedge fund manager......

:)
 
You must log in or register to reply here.
Back
Top