Curve Fitting: The Ultimate Conundrum | Forex Factory

Options

Permalinks
Print Thread

Similar Threads

Optimization, statistics, and avoidance of curve fitting 21 replies

EA Backtesting and Curve Fitting Concern 2 replies

(curve)-fitting and overfitting, good examples wanted 0 replies

why curve fitting doesn't work? 23 replies

"curve fitting" or just improving?? 5 replies

Attachments: Curve Fitting: The Ultimate Conundrum
Exit Attachments

Curve Fitting: The Ultimate Conundrum

  • Poll

  • Would you trade a curve fit system that showed profitable years recently?
  •  
  •  
  •  
  • #1
  • First Post: Jun 2, 2007 12:01pm Jun 2, 2007 12:01pm
Hanover has accurately nailed the backtesting problem we all face when trying to build profitable systems: "If we have no edge, we are simply curve fitting."

I agree with him, and bet most of you do to.

So, is it worth trading a curve fit system that backtested well over the last few years or so? Because there is no true edge, and your system is based on the random luck of the fit.

Please comment freely.
  • #2
  • Jun 2, 2007 12:37pm Jun 2, 2007 12:37pm
read and understand this article:
http://www.forexfactory.com/showthread.php?t=4446

its the only article i ever wrote, and i wrote it because there is such a gap in literature about this subject. hindsight is the enemy of the system developer, and unless you take careful precautions, hindsight will win.
Relax and be happy.
  • #3
  • Edited 4:34pm Jun 2, 2007 3:34pm | Edited 4:34pm
Are news trading straddles the only edge that can exist for us?

Merlin, I've got a sick feeling in my stomach. This is my system performance from '01-'07, all based on curve fitting.... I guess my imagination ran away with me as I thought I would get rich. I realize this system can have a bad year or two, but 6 profitable years in a row leads me to believe there are more profitable years ahead. So much of it is luck, and what I chalk up to a calculated risk.
Attached Image (click to enlarge)
Click to Enlarge Name: TesterGraph.jpg Size: 34 KB
  • #4
  • Jun 2, 2007 4:11pm Jun 2, 2007 4:11pm
Quoting tdion
Are news trading straddles the only edge that can exist for us?
And "edge" is when a certain set of conditions increases the probability of the market moving up or down. Trading news gives you no edge at all, if you had an edge, you wouldn't need a straddle in the first place.
  • #5
  • Jun 2, 2007 8:15pm Jun 2, 2007 8:15pm
Quoting aicccia
And "edge" is when a certain set of conditions increases the probability of the market moving up or down. Trading news gives you no edge at all, if you had an edge, you wouldn't need a straddle in the first place.
At the risk of being pedantic, I think there might be some breakout and/or options traders that might disagree. If we can show that, following a certain (e.g. congestion) pattern, there’s a high probability of a large, clean move, even if we don’t yet know in which direction, a straddle could still yield a positive expectancy that could be profitably exploited.

David
  • #6
  • Jun 2, 2007 8:44pm Jun 2, 2007 8:44pm
Quoting tdion
This is my system performance from '01-'07, all based on curve fitting.... I realize this system can have a bad year or two, but 6 profitable years in a row leads me to believe there are more profitable years ahead.
Tom, to the naked eye, that’s a pretty smooth equity curve. No really deep or prolonged drawdowns, or lengthy recovery times.

I’m no statistician, but there’s a table in the following link that gives statistical error, based on backtest sample size:
http://www.tradejuice.com/system-tra...ing-system.htm

I’m assuming that this could be applied to the likes of win rate or profit factor. In other words, if (for example) after 800 trades my system gives a PF of 1.2, then I can say with 95% confidence that the PF actually lies between 1.2 ± 3%, i.e. 1.164 and 1.236

Hence such a system would be realistically tradable, assuming that one can come to terms with losses, and feels comfortable operating within a 95% confidence level.

I’m not saying whether you should or shouldn’t trade your system. However, I realize that, in my own personal situation, perfectionism (the vain quest for the grail) amounts to procrastination, and time is passing me by. Whatever way I look at it, trading probabilities are always going to hinge around assumption and unmanageable risk, and at some point I must take the plunge, and move from testing demo by putting some real $$$ at stake. Heck, if nothing else, (and assuming I can maintain some detachment) it will likely be a fun exercise!

David
  • #7
  • Jun 3, 2007 8:01am Jun 3, 2007 8:01am
polinomial regress is key.
use forier analyze.
frequincy bin cycle reveel market move next.
ok, excel.
thank you.
  • #8
  • Jun 3, 2007 1:20pm Jun 3, 2007 1:20pm
Quoting jane
polinomial regress is key.
use forier analyze.
frequincy bin cycle reveel market move next.
ok, excel.
thank you.
wtf? lol

whoever this is they talk like a robot.
  • #9
  • Jun 3, 2007 4:03pm Jun 3, 2007 4:03pm
my answer is yes.
in fact, i AM trading a system based on polynomial regression. it's based on the same principle as bollinger bands, but instead of a SMA i'm using a linear and a parabolic regression +/- 2 standard deviations.
works pretty nice.
Nihil Sine Deo.
  • #10
  • Jun 3, 2007 10:57pm Jun 3, 2007 10:57pm
Quoting Scrat
my answer is yes.
in fact, i AM trading a system based on polynomial regression. it's based on the same principle as bollinger bands, but instead of a SMA i'm using a linear and a parabolic regression +/- 2 standard deviations.
works pretty nice.
On the subject of polynomials, there’s a guy called Doru Stoian who uses an advanced polynomial curve fitting technique as a method of forecasting future price direction.

If anybody’s interested, you could search for his (many) posts, under the name ‘stoian’, on the Incredible Charts forum here: http://www.incrediblecharts.com/foru...rd-topics.html

His knowledge of statistics is way beyond mine, which, along with the fact that English is clearly not his preferred language, made his posts a very difficult read for me. Despite his best explanations, I was never able to understand how polynomials could be helpful in forecasting market trends, since they tend to trend away to positive or negative infinity; the higher order the poly, the faster the rate.

How profitable his method is, I do not know. Anyway, I include this link in case anybody finds it useful.

David
  • #11
  • Edited Jun 4, 2007 12:10am Jun 3, 2007 11:54pm | Edited Jun 4, 2007 12:10am
yes, it can also be used to forecast... that's how i'm using it. not a very big fan of high polynomial orders, usually 1 or 2 seems more than enough for me; when i use more, the curve adapts too fast to the price.
long story short, non-linear regression is used a lot in research (that's how i came across it) and i believe it has a lot of potential.
i've been using it for about 2 months now and so far has been the only method that brought me consistent profits.
i'll look into doru's thread, thanks for the tip.

Quoting hanover
On the subject of polynomials, there’s a guy called Doru Stoian who uses an advanced polynomial curve fitting technique as a method of forecasting future price direction.

If anybody’s interested, you could search for his (many) posts, under the name ‘stoian’, on the Incredible Charts forum here: http://www.incrediblecharts.com/foru...rd-topics.html

His knowledge of statistics is way beyond mine, which, along with the fact that English is clearly not his preferred language, made his posts a very difficult read for me. Despite his best explanations, I was never able to understand how polynomials could be helpful in forecasting market trends, since they tend to trend away to positive or negative infinity; the higher order the poly, the faster the rate.

How profitable his method is, I do not know. Anyway, I include this link in case anybody finds it useful.

David
Nihil Sine Deo.
  • #12
  • Jun 4, 2007 9:06am Jun 4, 2007 9:06am
Quoting aicccia
wtf? lol

whoever this is they talk like a robot.
domo agriato, mr. roboto! :
  • #13
  • Jun 4, 2007 9:08am Jun 4, 2007 9:08am
This is a bit off topic, but where is there more information on this method? -Thanks

Quoting Scrat
yes, it can also be used to forecast... that's how i'm using it. not a very big fan of high polynomial orders, usually 1 or 2 seems more than enough for me; when i use more, the curve adapts too fast to the price.
long story short, non-linear regression is used a lot in research (that's how i came across it) and i believe it has a lot of potential.
i've been using it for about 2 months now and so far has been the only method that brought me consistent profits.
i'll look into doru's thread, thanks for the tip.
  • #14
  • Jun 4, 2007 9:31am Jun 4, 2007 9:31am
i don't think you understand what i said. it's not off topic, because polynomial (or non-linear) regression is actually fitting the data (i.e., price) into a smooth curve. the degree of the polynome can be 1 (linear), 2 (parabolic), 3 and so on; from my experience, good results are obtained with linear and parabolic regression.
as for the method, it's the same as bollinger bands but instead of a SMA i am using a linear and parabolic regression as average and watching price behavior at +/- 2 st.dev away from this line.
but since you opened this door, what exactly do you understand by curve fitting? what kind of curve?

Quoting tdion
This is a bit off topic, but where is there more information on this method? -Thanks
Nihil Sine Deo.
  • #15
  • Jun 4, 2007 9:46am Jun 4, 2007 9:46am
My definition of curve fitting is creating a system of rules that work on a random set of data. Your backtests show profits in the past, but the future could (according to Merlin, will likely) be nothing like the backtests.

If you found a pattern (or regression technique) that always won 70% of the time because of some fundamental reason, it would not be curve fitting. But rule based trading (by itself,) with no fundamental information tied to it, is likely worthless.

Quoting Scrat
i don't think you understand what i said. it's not off topic, because polynomial (or non-linear) regression is actually fitting the data (i.e., price) into a smooth curve. the degree of the polynome can be 1 (linear), 2 (parabolic), 3 and so on; from my experience, good results are obtained with linear and parabolic regression.
as for the method, it's the same as bollinger bands but instead of a SMA i am using a linear and parabolic regression as average and watching price behavior at +/- 2 st.dev away from this line.
but since you opened this door, what exactly do you understand by curve fitting? what kind of curve?
  • #16
  • Dec 27, 2007 10:28pm Dec 27, 2007 10:28pm
How quick will you jump on board a system? A quick bandwagon rider, or an overly cautious grandma? This is really a great discussion topic.....
  • #17
  • Jan 8, 2021 10:50pm Jan 8, 2021 10:50pm
Quoting merlin
read and understand this article: http://www.forexfactory.com/showthread.php?t=4446 its the only article i ever wrote, and i wrote it because there is such a gap in literature about this subject. hindsight is the enemy of the system developer, and unless you take careful precautions, hindsight will win.
I like the article very much thank you for recommending
Steps to profitability: Collect data, money management, stay bored
  • #18
  • Last Post: Jan 8, 2021 10:52pm Jan 8, 2021 10:52pm
Quoting tdion
Are news trading straddles the only edge that can exist for us? Merlin, I've got a sick feeling in my stomach. This is my system performance from '01-'07, all based on curve fitting.... I guess my imagination ran away with me as I thought I would get rich. I realize this system can have a bad year or two, but 6 profitable years in a row leads me to believe there are more profitable years ahead. So much of it is luck, and what I chalk up to a calculated risk. {image}
i got a tip:
discretionarily trade the return & dd of your curve-fitted system
so for example if your dd is going 2 standard deviation of the avg. return from your backtesting data then do not trade the system until the return breaks even
Steps to profitability: Collect data, money management, stay bored
0 traders viewing now