Avoiding Curve fitting

Quote from Code7:

I could as well say that changing the lookback period of a moving average to improve backtest results is not curve fitting. "(c(1) + c(2) + c(3) + c(4)) / number of summands" is one particular model you select and cannot be changed afterwards. This absurdity can be expanded to all popular indicators that involve any kind of smoothing and even further.
More...

It is not absurdity, it is rather a gross misunderstanding from your part.

With SMA(n) you can set n to improve backtest results but you essentially are fitting a curve, which is the curve generated by the SMA, to market prices in such a way as for a certain objective function to be maximum (or minimum). Mathematically, the problem can be stated as follows

maxJ, J is the objective function and it can be as simple as the net profit

subject to a set of contraints S

With something like C > C(1), there is no curve generated by the system that can be fitted to market prices. There are only signals points. If you change this to C > C(2), you are changing the signals, not any curve in particular. This is not curve-fitting. This is model selection. You determine that some model is better than another. This is the essense of system trading.

I guess I have to try to be more explicit:

(1) Curve-fitting involves model selection

but

(2) Model selection does not necessarily involve curve-fitting

I am afraid I cannot help further on this particular issue. If someone is still confused about these concepts, better stay away from system trading and trade the news instead.
 
Quote from Random.Capital:

"not necessarily"

Very nice backtrack.

Cheers.
More...

No backtracking. It only looks like backtracking to those that have hard time with the subject. Well,

If it makes you happy to state that everything is curve-fitting I am inclined doing so since Christmas is around the corner. Cheers.
 
Quote from intradaybill:

It is not absurdity, it is rather a gross misunderstanding from your part.
More...

You did not address my main point about your arbitrary assessment.

Signal A
buy if c > c(1)
buy if c > c(2)
buy if c > c(3)

Signal B
buy if c > c(1)
buy if c > (c(1) + c(2)) / number of summands
buy if c > (c(1) + c(2) + c(3)) / number of summands

Why do you declare the 3 lines under Signal A to be different signals with no parameters?
Why do you declare the 3 lines under Signal B to be the same signal with different parameters?
What is the rationale behind this, other than supporting your claims about curve fitting?

If you choose to reply, I prefer to have it without obfuscation and hiding behind semantics.
 
Quote from Code7:

You did not address my main point about your arbitrary assessment.

Signal A
buy if c > c(1)
buy if c > c(2)
buy if c > c(3)

Signal B
buy if c > c(1)
buy if c > (c(1) + c(2)) / number of summands
buy if c > (c(1) + c(2) + c(3)) / number of summands

Why do you declare the 3 lines under Signal A to be different signals with no parameters?
Why do you declare the 3 lines under Signal B to be the same signal with different parameters?
What is the rationale behind this, other than supporting your claims about curve fitting?
More...

@ intradaybill

You did not reply so far. I would still like to clear things up so here is my best guess about the real reason for your claims. It looks like your guru Michael Harris confused you into believing that price patterns were not subject to curve fitting as part of his APS software marketing ploy.

However, denying parameters - by simply proclaiming different and unchangeable signals at will - is no way to evade curve fitting. It only makes a difference in your imagination because the conclusions you draw based on arbitrary assessment are illusive.

Apologies if you find that offensive.
 
Quote from Code7:

@ intradaybill

You did not reply so far. I would still like to clear things up so here is my best guess about the real reason for your claims. It looks like your guru Michael Harris confused you into believing that price patterns were not subject to curve fitting as part of his APS software marketing ploy.

However, denying parameters - by simply proclaiming different and unchangeable signals at will - is no way to evade curve fitting. It only makes a difference in your imagination because the conclusions you draw based on arbitrary assessment are illusive.

Apologies if you find that offensive.
More...

Code7, nobody can confuse me into anything. My apologies, I have not answered because I have been busy trading.

Now, I do use APS and these people do not have any marketing ploy. They ask all future customers to first try their demo before parting with money. Actually, this software Michael Harris has developed, I use for the llast 3 years and it has paid for itself multiple times while you are trying to debate whether patterns are subject to curve-fitting.

I know of another guy in Europe who I have talked with in the past and he also made good money with the same software. The point being that some poeple know the difference between curve-fitting and selection bias.

Now, as far as your examples, signal A and signal B. You are totally, completely off and missing the point. Price patterns do not involve any additions or math operations, only Boolean ones. As soon as you introduce math operations, you introduce curve-fitting because of the mappings that take place. Maybe you do not understand what a pattern is and how it differs from a formula. Maybe you do not understand operators and mappings, functions and curves.

By the way, a pattern that has already happened on a chart several times, let us say 100 times, is a true representation of price behavior. Now, does it make any sense to call that pattern curve-fittied? Maybe to a lunatic, who believes that price curve-fits to itself.

The real issue with patterns is whether they are subject to selection bias, not whether they are subject to curve-fitting because there is no such issue, as they represent true price action and nothing more or less.

If you still do not understand that, I cannot do anything else to help. Good luck to you.
 
I do understand very well that you again try to obfuscate your way out with hiding behind semantics. First it was all about how price action does not require any parameters. Now it is about operators and mappings, functions and curves, patterns versus formulas. What a joke you are.

You claim price patterns do not involve any additions or math operations, only Boolean ones. Why not? You might want to look up the meaning of pattern, any dictionary will do. Again your definition of things is completely arbitrary.

In the end, it all boils down to this. Can APS give me that signal as a result? If yes, then you claim it was not curve fitted. Even if you optimized parameters to improve backtest performance before. All your made up drivel is tailored to show just that.

However, your posts cannot be taken seriously. You have a proven record of shilling for Michael Harris and APS on multiple message boards. A simple search of your user name reveals that. Selecting a shill for assessment clearly does introduce bias. Too bad.
 
Quote from Code7:

I do understand very well...
More...

I suggest instead of concentrating on nonsense and animosity against other successful people to try to trade yourself, if you have ever...

Good luck to you...you will need it...
 
Quote from intradaybill:

I suggest instead of concentrating on nonsense and animosity against other successful people to try to trade yourself, if you have ever...

Good luck to you...you will need it...
More...

Very nice evasion.
Looks like you got called out on your shilling with bogus claims and now want to dive away.
Nevermind, you are just a clueless tool for your guru, so who cares.
 
You must log in or register to reply here.
Back
Top