Hi, Time Series Analysis (TSA) promises extracting from the data all and each of the following components: Code: 1) trend component 2) seasonal component 3) cyclical component 4) error, or irregular component Ie. from the stock price data it can detect and extract all of the above components. This sounds IMO very interessting. What is your practical experience with this method? Is it useful for market analysis and system development? There seems to be many methods of Time Series Analysis: MA, ARMA, ARIMA, ARFIMA, ARCH, GARCH, TARCH, EGARCH, FIGARCH, CGARCH, ... Which one should one learn/study?
http://ocw.mit.edu/courses/electric...-science/6-003-signals-and-systems-fall-2011/ You have to learn to crawl before you can run.
I can think of no reason to believe that markets are like electronic signals. Electronic signals tend to be highly organized and easily differentiable from noise... so the concept of noise isn't a relative thing, it's well defined. If you can define the noise you can filter it out actually... We have arguments here on ET all the time where one loser is swing trading and another is daytrading. So the daytrader is seeing signals and the swing trader is seeing noise... [and not likely either has a steady income from trading maybe]
Basically its just hard to analyse non-stationary data series. There is a lot of info here: http://quant.stackexchange.com/questions/8875/why-non-stationary-data-cannot-be-analyzed HTH! -gariki
Thx for that interesting thread. There, one participant has posted this interesting link: "Nonstationary Models for Time Series" http://sfb649.wiwi.hu-berlin.de/fedc_homepage/xplore/tutorials/xegbohtmlnode37.html Looks IMO much promising.