Optimizing or curvefitting

It's not about optimising OR curvefitting, it's about optimising AND not curvefitting. True optimisation takes into account the tolerance (what the system CAN do) and specifications (what the system MUST do and it's not only about TRUE performance - in the future not in the past - there are also constraints) of the system, whereas curvefitting doesn't take into account neither one or the other, not even performance since it is not TRUE performance - in the future, which means inferential probability and not only descriptive statistics - but only performance in the past (only descriptive statistics). To do inferential probability requires in fact to eliminate as far as possible any curvefitting, if not so inference about performance in the future risks just to be wrong and so useless.
 
Typically curvefitting gives performance under a conditional probability see the other thread:

http://www.elitetrader.com/vb/showt...hlight=conditional+and+probability#post462625

Re: RSI 25 Explosions System?
88% perhaps but what is not said is that it is a rather a CONDITIONIAL PROBABILITY noted traditionally P(A | B) (probability of A knowing B) where B is the market context or global trend that is to say one must know before if the market is globally bullish or Bearish; let's remark that it could also be sideways but let's simplify for the demo then I will suppose that P(B)=1/2 since I don't have any other knowledge on the design of the system.
What we really need to know is P(A) and not only P(A | B).

Thanks to Bayes Axiom on conditional probability we know that:
p(A and B) = p(B) * p(A/B) (also = p(A) * p(B/A) but not used)
so that p(A) = 0.5 * 0.88 = 0.44 = 44% is an estimation of the real probability of this system
 
Quote from indahook:

Obviously one of the most important things to consider before
running a trading system is determining the type of mkt you are
trading i.e trending up, down or sideways. Which leads me to this question. When backtesting a
long/short strategy do any of you use a broad mkt directional filter
as one of your first rules? When I run a backtest of a certain
strategy I have written over the past 4 years I get mediocre-poor
returns..barely beating the mkt after commission. And an unacceptable
risk reward. But if I use a filter to determine if we are in Bull
or Bear territory before going long or short the returns are great,
whipping the mkt averages 10-15% and risk reward well over 3-1. Do any system traders have any thoughts on this? Can this be considered curze fitting?

BTW, these are longer term strategies the avg hold for winners is 63 days and losers 70.


sorta new at this system building stuff,
Chris
:D
More...

Optimizing and curvefitting is only way to make more than average (25%) return. Robust system for all markets is only for ordinary return. After some time that optimized sytem will stop working when market change in the future, and you have to find new one, so, test, test, test.

Example, es basically a choppy market compare to currency which is trending market last year, but may change in the future. Now, if you enter breakout in currency there is a good probability of success, but not es where many false breakout happens.

So, you need different systems for currency and es. Every market has her personallity.

:) :p :cool: :D
 
Quote from seldin:

>>I read Pardo's book in 1994 or 1995. It was one of the most useless books on trading that I have ever read. Of course, this is a personal opinion.<<

Your comment reminds me of a Day Trader that I had chatted with. He mentioned that back testing was useless and never worked.

I have to disagree. I have been trading for almost 2 years... A newbie. My first year of back testing, I only found systems, that lost money, or after trying what I thought were good, ended up producing poor results.

The point I would like to make, is after doing lot's of back testing for 2 years, I am able to back test a system, and then trade that system in Real Time. The thing that I like, was I am able to duplicate my back test results in real life trading. This is much tougher than it looks.

The trick is learn as much as you can about developing trading systems, analysis, etc.

The point, is just because something does not work for you, does not mean it is useless. However, I would not use something that did not work for me.

I have re-read many trading books and have learned a lot the 2nd time around.

This is the case for me about Pardo's book.

Just my 2 cents...

larryTAKEOUT@seldin.net
More...


Please let me be clear. I think that backtesting is important. I just happen to think that Pardo's book is worthless. The two are not mutually exclusive.
 
I am going to keep a log of the systems results that I started the thread about.


3/29 - 4/13

"Pressure Cooker Buy" 10 trades -$324.00
"Pressure Cooker Sell" 15 trades +$1431.00
-------------
1107.00
comm -50.00
--------------
1057.00


I wish I could cut and paste %`s, max profit/loss and drawdowns but AIQ doesn't allow you to do that.


I had intended to trade long or short based on the direction of the SPX (intermediate term) but it seems that a simultaneous long/short approach may work OK...time will tell
 
You must log in or register to reply here.
Back
Top