How many hard-coded parameters in the Holy Grail?

Quote from joethetrader:

I'm learning this stuff so this is a genuine question not meaning to challenge: I read everywhere else that having too many parameters tends to lead to over-fitting models. How would you reconcile your theory with that?
More...

That's a question (how many parameters is too many?), that I don't know that I have a solid answer to, actually. But I think wanting to have as few as possible fits with my hypothesis that the best trading system would have one parameter at its core. That just seems like the most logical outcome, since it would seem that such a system would be based on one unique insight.

One way to think of your question's answer is to realize that, if you take everything that "can" be a parameter in a model, then you are talking about hundreds or thousands of possible parameters. That being the case, any subset of even 10 only scratches the surface of how over-fitted your model could be. Ideally, each one would bring something different to the model.
 
Quote from joethetrader:

I'm learning this stuff so this is a genuine question not meaning to challenge: I read everywhere else that having too many parameters tends to lead to over-fitting models. How would you reconcile your theory with that?
More...

There is a specific and very high utility answer to your question.

All system design operates with a few major principles in mind.

All the troubles mentioned here in this thread so far, do go away if the designer/developer is able to start from the beginning.

To define a system, a set is required.

The word set describes a collection.

The specific and well articulated rules (principles) for doing this defining of a system are crystal clear.

The set must be complete.

And the expresession of each element of the set must be in "like kind".

For trading in markets, the set was defined long ago* and to this day works perfectly.

FYI the set has two elements.

As markets operate, the participants perform as they can by following the rules the owners set forth for their participation. Collectively, each participant contributes as he is allowed by the size of his capital and with regard to the instrument characterisitcs.

The consequence of having the set and using it, is that in markets, it becomes observable that there is no noise, no flaws and no anomalies.

A side topic was introduced by a bystander (intellectually speaking) to this thread. Higgs is still alive. CERN is again working and being brought up to its full capability. It is 8/13th of the way there. When the energy of the system is at full force, It will be possible for a lot of computers to work together to determine how mass (see Einstein) is created by particles in space moving through a Higgs field.

Markets were simpler to figure out and it was done long ago. So I have just posted these known results for you to have. "Hard cored parameters" is a misnomer. Notwithstanding this error, the system of the markets is defined. All markets share all the commonalities required in order that one system results for all markets.

*The people who did it and made it public are long dead. These people were different than the people who did the mathematics to make it possible for the problem solvers to reach their elegant solution.
 
IMO, the problem isn't in the number of parameters - not even in the intrinsic complexity of the algo, which is a better representation of its over-optimization capabilities.

The (mechanical) trader's problem is to:

1. identify (mentally) a recurring pattern with a tradable edge.
2. develop an algo to (mechanically) spot that pattern, and make that algo smart enough to resist all kinds of "noise" and identify all kind of "variants" of the ideal pattern.

Doing 1.) usually requires developing and using 2.) - basically because it isn't possible for the brain to recognize on a chart all possible failed patterns.

But instead of focusing on the 3 aspects of the problem - ideal pattern, manifestations of noise, patterns variants, people throw in a number of indicators that seem related to the pattern, then more indicators to try filter the noise ... in the end, it is as meaningless as it gets, and of course, it doesn't work past the backtesting stage (if it ever does there).

Instead of even looking at the number of indicators or parameters, just modify your strategy timeframe by +/- 5% / 10% / 15% (ie., for a strategy designed to work on 100-vol chart, test-it without any change at 85-vol / 90-vol / 95-vol / 105-vol / 110-vol / 115-vol), and backtest using whatever parameters you are planning on using live ... that will give you an idea of how sensitive your strategy is to "noise" - in other terms, how over-optimized it is.

(of course, the ratio of #trades (in backtest) to #params is another tell about possible over-optimization)
 
Quote from logic_man:

That's a question (how many parameters is too many?), that I don't know that I have a solid answer to, actually. But I think wanting to have as few as possible fits with my hypothesis that the best trading system would have one parameter at its core. That just seems like the most logical outcome, since it would seem that such a system would be based on one unique insight.

One way to think of your question's answer is to realize that, if you take everything that "can" be a parameter in a model, then you are talking about hundreds or thousands of possible parameters. That being the case, any subset of even 10 only scratches the surface of how over-fitted your model could be. Ideally, each one would bring something different to the model.
More...

You may wish to consider the purpose of markets. Then, if you do, you may be able to understand the how and why of markets.

You used three words quite casually: Hypothesis, parameter and logic.

If you wish to consider thinking about and organizing facts, then you will be able to go further into the detail of understanding how the solution was worked out long ago.

Markets turn out to be counterintuitive.

Both the baker and the wheat grower each guarrantee their respective success by using markt tools made availalbe by the owners of markets.

Participants other than bakers and wheat growers participate as parasites to make marginal profits. these parsites are very skilled and make money all of the time. So they create tools to extrct profit segments one after another.

This forum was set up by the owners of ET so the name, strategy (System) design, is also a casual string of words.

The parasitic trader DOES design systemmic strategies to always be taking profit segments out of the markets. This designing person does best by knowing what is going on. He does not guess that there is one something or other. This designer goes through a process of deduction. He never uses induction either.
 
Quote from dom993:

IMO, the problem isn't in the number of parameters - not even in the intrinsic complexity of the algo, which is a better representation of its over-optimization capabilities.

The (mechanical) trader's problem is to:

1. identify (mentally) a recurring pattern with a tradable edge.
2. develop an algo to (mechanically) spot that pattern, and make that algo smart enough to resist all kinds of "noise" and identify all kind of "variants" of the ideal pattern.

Doing 1.) usually requires developing and using 2.) - basically because it isn't possible for the brain to recognize on a chart all possible failed patterns.

But instead of focusing on the 3 aspects of the problem - ideal pattern, manifestations of noise, patterns variants, people throw in a number of indicators that seem related to the pattern, then more indicators to try filter the noise ... in the end, it is as meaningless as it gets, and of course, it doesn't work past the backtesting stage (if it ever does there).

Instead of even looking at the number of indicators or parameters, just modify your strategy timeframe by +/- 5% / 10% / 15% (ie., for a strategy designed to work on 100-vol chart, test-it without any change at 85-vol / 90-vol / 95-vol / 105-vol / 110-vol / 115-vol), and backtest using whatever parameters you are planning on using live ... that will give you an idea of how sensitive your strategy is to "noise" - in other terms, how over-optimized it is.

(of course, the ratio of #trades (in backtest) to #params is another tell about possible over-optimization)
More...


This person "believes" noise exists in markets. He gets the consequences of this unfounded belief.

There may be a definition of noise that he uses; He better check how authoritative it is.
 
Quote from jack hershey:

This person "believes" noise exists in markets. He gets the consequences of this unfounded belief.

There may be a definition of noise that he uses; He better check how authoritative it is.
More...


Jack, thanks for the input, really. When was the last thing of practical value your posted on ET?

I can show you white noise in CL any day / any time. But this is not even what I meant by "noise".

BTW, my fully automated system that separates signal from noise sold CL at 89.61 (1:41pm EST) & covered at 88.11 (2:29pm EST), that is 150-ticks per contract in 50-minutes.
 
Quote from dom993:

Jack, thanks for the input, really. When was the last thing of practical value your posted on ET?

I can show you white noise in CL any day / any time. But this is not even what I meant by "noise".

BTW, my fully automated system that separates signal from noise sold CL at 89.61 (1:41pm EST) & covered at 88.11 (2:29pm EST), that is 150-ticks per contract in 50-minutes.
More...

Congrats on your 50 minute trade. I hope you had a lot of contracts running. Do you ever get partial fills on orders??

I suggested that there were two elements in the required set to cover the market completely. One element is for continuation and the other element is for change; they are mutually exclusive.

Here in this forum in another thread, I suggested that the two elements were orthogonal. That was yesterday.

I did 14 trades today, all reversals. Before and after these reversls I did an entry at open and an exit on bar 78 (16:00 hours EDLST.) I was in the market for about 400 minutes.

Often, I call the next day's open. Tomorrow the open will be short.
 
Quote from logic_man:

Wish I could get the poll option to work, but I say 1. No more, no less.

This will allow maximum flexibility in designing and optimizing other parameters with the ability to mathematically differentiate what is and is not a trading opportunity with a consistent standard.

Any other parameters will be optimizable, but that one will have to be the "cornerstone" of the Grail. It's "line in the sand".
More...
that would be, adding to winners and adding to losers. Be nice if you could do one without the other, but unfortunately that is not the way the world works.
 
You must log in or register to reply here.
Back
Top