Intentional Curve Fitting

Mr_You · Dec 19, 2012

I'm a believer in curve fitting a strategy to the instrument and its current market conditions. BUT you don't want to overly curve fit or under curve fit. As mentioned, ideally a profitable strategy has very few parameters to adjust. See Pardos "Evaluation and Optimization...".

But here's the deal... from my recent research, you can't do it on the number of days... you must do it on the number of trades! Atleast ~100 trades is the way to go.

For some instruments the previous week will provide ~100 trades for other instruments several months will only provide ~100 trades. This is why walk-forward optimization testing based on days alone is not ideal.

And, as mentioned, unfortunately there is still a little bit of risk/luck/art involved. Its not an exact science. YMMV!

Any thoughts?

Here2learn · Dec 19, 2012

Quote from abattia:

IMO looking for the parameter set that is âbest in tune with the market right nowâ is fraught with risk, and not only for the reason outlined above. It is also because it ignores that signal-to-noise ratios for trading systems are usually pitifully small, and there can be long, extended periods when no signal at all is present. In my experience, a valid approach in these no signal situations is simply to keep trading as you were, and accept the drawdown, but remain in a position to benefit immediately once the signal comes back again. With a âwalk forward optimizationâ approach on the other hand, this fallow period with no signal may lead to strategy parameters drifting considerably from what is best over the long term, and may even result in the return of signal being missed altogether. And in my experience to date, the period immediately following return of signal is often the best run, that more than makes up for the drawdown.

yeah, this is a great point that I was thinking on as well. The optimization can add aditional risk that you miss the move your strategy was orignally looking for in the first place. The other thought is that if the parameters don't change enough so that they still capture all signals, just a matter of how effectively/profitably they capture the move.

Mr You, I'd agree that ideally it's on number of trades. This may not be an issue in my case as I'm working an intraday strategy that nearly always takes one trade a day. In that case as suggested by the 20% rule, a rolling 5 days(5 trades) of optimization for trading parameters on the sixth day.

jack hershey · Dec 19, 2012

Quote from Mr_You:

I'm a believer in curve fitting a strategy to the instrument and its current market conditions. BUT you don't want to overly curve fit or under curve fit. As mentioned, ideally a profitable strategy has very few parameters to adjust. See Pardos "Evaluation and Optimization...".

But here's the deal... from my recent research, you can't do it on the number of days... you must do it on the number of trades! Atleast ~100 trades is the way to go.

For some instruments the previous week will provide ~100 trades for other instruments several months will only provide ~100 trades. This is why walk-forward optimization testing based on days alone is not ideal.

And, as mentioned, unfortunately there is still a little bit of risk/luck/art involved. Its not an exact science. YMMV!

Any thoughts?

Thank you for moving the conversation forward.

You suggest having a reliable sample and you close the door on time in one dimension.

So:

1. Take all named amounts of time (dimensions) out of the picture.

2. Switch to only examine a sufficient number of linked trends on one fractal.

3. Some data has to be used twice. Limit this usage to only data that occurs during adjacent trend overlap.

5. Initially divde your sample into two groups so that dta is specifically grouped in like kind trends.

6. Acknowledge that some data is not allowed to be used in taking measurements.

7. At first limit the fractal exaamined to only one fractal.

The conclusions you will reach are that:

a. Money making is inversly related to the trading fractal duration.

b. At some point the premium becomes signifcant.

c. The equity curves are a family of fourth degree poylmomials (the only time you can use continuous functions in your research.

d. Market variables have an interlocking frequency ratio: 1 to 2

e. Fractals have an interlocking event ratio: 1:3.

f. The overlap of trends begins with trend failure and ends with regional independence always.

g. the only measure in markets as a PM of an HS is speed of trends: d trend/d event.

Here2learn · Dec 20, 2012

Quote from jack hershey:

d. Market variables have an interlocking frequency ratio: 1 to 2

e. Fractals have an interlocking event ratio: 1:3.

f. The overlap of trends begins with trend failure and ends with regional independence always.

g. the only measure in markets as a PM of an HS is speed of trends: d trend/d event.

I could use some further explanation, especially regarding conclusions d. -> g.

Thanks

braincell · Dec 20, 2012

Quote from Here2learn:

I could use some further explanation, especially regarding conclusions d. -> g.

Thanks

haha

I had Jacko here on ignore and every once in a while a quote by him pops up. Oh boy, haha that was really funny. What a joker!

Hey Jack, after all these years I'm beginning to figure you must be doing this at least for a bit of laugh right!? Sitting at home in dirty undies laughing you arse off. Sounds so professional don't he! That'd be a great episode of Futurama or something.

Anyway, Here2learn, if you're here to learn put Jack on ignore and you'll save yourself some troubles. Or don't, like in this post he has some comedy gems. still laughing hahah

Mr_You · Dec 20, 2012

I agree on all points.

jack hershey · Dec 21, 2012

Quote from braincell:

haha I had Jacko here on ignore and every once in a while a quote by him pops up. Oh boy, haha that was really funny. What a joker!

Hey Jack, after all these years I'm beginning to figure you must be doing this at least for a bit of laugh right!? Sitting at home in dirty undies laughing you arse off. Sounds so professional don't he! That'd be a great episode of Futurama or something.

Anyway, Here2learn, if you're here to learn put Jack on ignore and you'll save yourself some troubles. Or don't, like in this post he has some comedy gems. still laughing hahah

braincell

Registered: Jul 2011
Posts: 485

08-31-11 06:45 AM

This a question I cannot find a clear answer to.

When I place a buy lmt order at a certain price, I'm not sure I see it in market depth. Maybe that's because of slow updates with IB but this leaves me with a question about how markets operate.

When I place a buy limit, if it's at the current Bid, does it immediately become a bid there directly associated with my account?

Or, do the market makers only make bids, and then once those are filled they find my buy limit order and send me the fill?

Any insight would be appreciated, thanks.

DT-waw · Dec 21, 2012

no matter how you slice it, its all curve fitting, folks.

21 dec 2012 today, no need to fool around yourselves anymore.
the market might get hell lot of different and wipe your non parametric stuff away.

if 95% of volume would be generated by smart algos, whats your choice to survive even with smthing you consider non-fitted?

jack hershey · Dec 22, 2012

Quote from DT-waw:

no matter how you slice it, its all curve fitting, folks.

21 dec 2012 today, no need to fool around yourselves anymore.
the market might get hell lot of different and wipe your non parametric stuff away.

if 95% of volume would be generated by smart algos, whats your choice to survive even with smthing you consider non-fitted?

I feel fortunate that you think the way you do.

Two wrongs do not make a right. They only make the wrong more robust.

Please take a moment to think about the higher ground.

It is way above survival, which itself is an illness syndrome creator.

The higher ground of cognitive intellect is immunity from externalities. You may not ask why, either.

When you have immunity, algos are part of he package. They certainly do not use the types of mathematics that are part of your skills/limitation ratio place on the spectrum.

jack hershey · Dec 22, 2012

Quote from Here2learn:

I could use some further explanation, especially regarding conclusions d. -> g.

Thanks

Okay. You are welcome.

d. For price to complete a trend (completing one gerund), volume expresses both its pssible opposing gerunds.

Price increasing: volume increasing, decreasing.

Add a less significant variable: A/D (Use Welles Wilder's two DMI's for measurement.)

volune increasing: A/D A then D.

So I invented "scoring". With scoring you can trade flawlessly and not deal with either risk or money management.

On the binary raw score is converted to the Hindu/Arabic decimal system which has the invention of the place holding concept as a nice aid for Doing arithmetic and more advanced mathematics.