Is there any appropriate methodology for detrending daily data over a period of a few years. A simple linear regression is not appropriate since the trend likely changes several times over the timespan.I know with economic data we often apply a Hodrick-Prescott filter with a coefficient of 1600 for quarterly data.Does anyone have an opinions on this?Thanks.

I can't think of any application for which this is a good idea. Prices are what they are, someone paid them. "Detrended" prices are meaningless. Also a Hodrick-Prescott filter will smooth the series rather than detrend it.If you want to adjust market prices, my advice is to do it financially. Detrend by subtracting off an index. Smooth by averaging the investment period or considering an option strategy. If you do things this way, you end up with something real, a price series of a portfolio someone could have owned.In some respects, quantitative finance is the opposite of econometrics. In econometrics, you assume a simple underlying model of which you observe noisy measurements. You get rid of the noise to see the model more clearly. In finance, we know the prices in all their complexity are real; the underlying models are even more complex than the data. From actual observation of a univariate interest rate moving up and down in a small range, we generate a multidimensional implied volatility surface with complex dynamics. Your model is noisier than your data.

I can't think of any application for which this is a good idea. Prices are what they are, someone paid them. "Detrended" prices are meaningless. Also a Hodrick-Prescott filter will smooth the series rather than detrend it.why can't you start by log differencing the series (just working with cont. compounded returns)? you can always transform your model from log returns to prices easily enough.

Aaron," I can't think of any application for which this is a good idea. Prices are what they are, someone paid them. "Detrended" prices are meaningless. Also a Hodrick-Prescott filter will smooth the series rather than detrend it."Yeah, but decomposing the price can sometimes help you in determing what is driving price movements. I'm interested in fluctuations around the trend. A HP filter is a way of fitting a nice smooth flowing trend line. I then detrend by doing : SERIES minus TREND.Anyway, I've thought about it for a while, and I'm now using a fouth order polynomial trend function. That seems to do the job- although I'm on the lookout for any research into this issue.kapital,"why can't you start by log differencing the series (just working with cont. compounded returns)? you can always transform your model from log returns to prices easily enough."Because it throws away the info that I'm after. Cheers guys.

QuoteOriginally posted by: MrQwertyYeah, but decomposing the price can sometimes help you in determing what is driving price movements.I'm not saying you're wrong, but in my experience this is not a fruitful approach with financial data. It's a bit like technical analysis, searching for patterns without considering economic justification. If decomposing into a fourth-order polynomial and idiosyncratic series seems to work, my guess is it won't work well out-of-sample. If you can figure out the financial rationale for the pattern, you get a robust model that may work on future data.But if it works for you it works. Good luck.

I am interested in splitting a financial time series into 2 series using a filter (noise and trend). In particular I want to show that certain aggregate factors only effect the "noise" component of the time series, which filtering technique would be best to achieve this? (I have easy access to Hodrick-Prescott and Kalman filters)

anyone.....?

Try thisWahib Arroum and Owen Jones, Estimation for time-changed self-similar stochastic processes. Proceedings SPIE Vol. 6039, Complex Systems, A. Bender Ed. Article 60390F (Jan. 16, 2006). Regards

Thanks for the reply.. Would it be possible to obtain a copy? I don't have access SPIE digital libraryRegardsG

Send me your email adress I'll send it to you....Regards