Algorithmic Trading Group Forum

Algorithmic Trading Group Forum
https://www.algotradinggroup.com/cgi-bin/yabb2/YaBB.pl
Trading Strategies >> Development >> Toward simplicity
https://www.algotradinggroup.com/cgi-bin/yabb2/YaBB.pl?num=1316850377

Message started by Algo Designer on 09/24/11 at 08:46:17

Title: Toward simplicity
Post by Algo Designer on 09/24/11 at 08:46:17

"Everything should be made as simple as possible, but not simpler" -- Einstein

"You know you've achieved perfection in design, not when you have nothing more to add, but when you have nothing more to take away" -- de Saint-Exupery

Does this apply to how you build strategies?

(inspired by http://www.ted.com/talks/george_whitesides_toward_a_science_of_simplicity.html)

Title: Re: Toward simplicity
Post by Algo Designer on 09/26/11 at 02:02:20

A colleague of mine suggested this wonderful page:
http://www.jbox.dk/quotations.htm

Title: Re: Toward simplicity
Post by Lampert on 09/27/11 at 10:51:25

Hi,

I would say it goes back to Occam razor
http://en.wikipedia.org/wiki/Occam%27s_razor

A model should be the simplest that still explain known data.

It could be made into math by using information theory approach that increases entropy according to the complexity of the model.

As an electrical engineer, I constantly live by this pardigm

Title: Re: Toward simplicity
Post by Algo Designer on 09/27/11 at 13:21:47

I am a big fan of this principle as well. Carefully "shaving away" unnecessary complexities and degrees of freedom in a trading strategy (in my case, a statistical model) helps improving its robustness. Knowing what can be safely cut off and how the effect of the simplification can be tested takes years of experience. The same applies to any other problem domain.

Title: Re: Toward simplicity
Post by OVVO on 09/27/11 at 18:57:10

Hi Lampart,

While simplicity is critical in lessening model fragility, explanation is not as critical.

"A model should be the simplest that still explain known data."
This is true for the physical sciences and does not hold for the social sciences due to the lack of repeatable results. Optimization of ex post data (explanation) consistently fails ex ante regardless of calibration frequency.

There is a very old debate in economics between Friedman and Simon (Friedman states assumptions do not matter as long as the predictions are accurate). Herbert Simon (1963) countered Friedman by stating the purpose of scientific theories is not to make predictions, but to explain things - predictions are then tests of whether the explanations are correct.

Both Friedman and Simon's views are better directed to a field other than economics/finance. The data at some point will always expose the frailest of assumptions; while the lack of repeatable results supports futility in the explanation of heterogeneous agents.