On Mon, Apr 22, 2013 at 9:54 AM, Rui Barradas <ruipbarra...@sapo.pt> wrote: > Hello, > > If stock prices are daily data, use frequency = 1. >
Err, maybe... The frequency attribute for a "ts" class time series is somewhat subtle. It's the amount of observations per seasonal period / relevant cycle, where relevant is a bit of a judgement call. E.g., suppose I have hourly readings of some sort of astronomical data. If it's one sort solar data, I might want to have my frequency equal to 24, being the number of hours in a daily. If it's based on moon things, I might want to have my frequency being 24*28 (roughly the hours in a lunar cycle). And if it's based on the movement of the earth through it's orbit, perhaps 24*365 makes sense. Still, this is all "hourly" data.... The assumption underlying this then is that there's a single major structural frequency to your data and that you want modelling functions to respect this frequency. If the OP has daily data, either 21 or 252 might make sense, depending on whether we're trying to model monthly or yearly cycles. (Or perhaps some other frequency, such as quarterly giving 84 for a frequency if we're looking at something driven by earnings announcements or the like). Ultimately, the base "ts" series isn't great for financial data. 252 trading days is not uniform across all markets and assets, nor will "daily" data be truly uniform. (weekends, holidays, etc) There's also some interesting research on what one might call "non-linear time" for modelling financial data, but that's now getting somewhat out of the scope of the original question. In short, I'd advise the OP use "xts" instead which uses truly time stamped (and quite likely irregular) data. Cheers, Michael ______________________________________________ R-help@r-project.org mailing list https://stat.ethz.ch/mailman/listinfo/r-help PLEASE do read the posting guide http://www.R-project.org/posting-guide.html and provide commented, minimal, self-contained, reproducible code.
