Jose-Marcio Martins da Cruz
2011-Feb-12 12:48 UTC
[R] Time unit in ts() and arima() functions
This question is surely trivial, sorry. I'm afraid I'm misunterpreting the information I got with the documentation, and I'm a little bit confused. I'm just an engineer with some little skills in statistics. Well, I have a time series - 600 days long - with some weekly periodicity inside. So far, so good. Well, if I define the time series with, say : a <- ts(b, frequency = 7) and I do "plot(a)", each unit of time seems to be a week, not a day, which is coherent with help(ts) which says "frequency: the number of observations per unit of time." but this isn't what I want. I'd like to retrieve seasonal information but, as I'd like to retrieve also day to day information, the time unit should remain one "day". Also, when I use the "arima(...)" function to fit a model (almost the same kind of algorithm which appeared here some days ago), and I specify in the "seas=" parameter, "frequency = 7", or "frequency = frequency(a)" (a is the time series), I can get arN, maN, sarN, smaN... coefficients. What unit shall be applied to these coefficients ? One day or one week ? Logically (and ideally), for me, one day for arN and maN and one week for sarX and smaX coefficients, but I'm not sure. I have the same kind of doubt about lag units when I apply the acf() function to the time series (a) or the residuals returned by the arima() function. Thanks for your answer.
On Sat, Feb 12, 2011 at 7:48 AM, Jose-Marcio Martins da Cruz <Jose-Marcio.Martins at mines-paristech.fr> wrote:> > This question is surely trivial, sorry. I'm afraid I'm misunterpreting the > information I got with the documentation, and I'm a little bit confused. I'm > just an engineer with some little skills in statistics. > > Well, I have a time series - 600 days long - with some weekly periodicity > inside. So far, so good. > > Well, if I define the time series with, say : > > ? a <- ts(b, frequency = 7) > > and I do "plot(a)", each unit of time seems to be a week, not a day, which > is coherent with help(ts) which says "frequency: the number of observations > per unit of time." but this isn't what I want. I'd like to retrieve seasonal > information but, as I'd like to retrieve also day to day information, the > time unit should remain one "day". > > Also, when I use the "arima(...)" function to fit a model (almost the same > kind of algorithm which appeared here some days ago), and I specify in the > "seas=" parameter, "frequency = 7", or "frequency = frequency(a)" (a is the > time series), I can get arN, maN, sarN, smaN... coefficients. What unit > shall be applied to these coefficients ? One day or one week ? Logically > (and ideally), for me, one day for arN and maN and one week for sarX and > smaX coefficients, but I'm not sure. > > I have the same kind of doubt about lag units when I apply the acf() > function to the time series (a) or the residuals returned by the arima() > function. >Functions which work with ts typically assume that a full cycle is represented by 1 unit so if a full cycle is a week then a week must be one unit and a day must be 1/7th of a unit; however, you can later convert your series to a non-ts representation which supports dates. For example,> b <- 101:121 > a <- ts(b, frequency = 7); aTime Series: Start = c(1, 1) End = c(3, 7) Frequency = 7 [1] 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 [20] 120 121> > time(a)Time Series: Start = c(1, 1) End = c(3, 7) Frequency = 7 [1] 1.000000 1.142857 1.285714 1.428571 1.571429 1.714286 1.857143 2.000000 [9] 2.142857 2.285714 2.428571 2.571429 2.714286 2.857143 3.000000 3.142857 [17] 3.285714 3.428571 3.571429 3.714286 3.857143> > # suppose start date is today > library(zoo) > z <- zooreg(coredata(a), start = Sys.Date()); z2011-02-12 2011-02-13 2011-02-14 2011-02-15 2011-02-16 2011-02-17 2011-02-18 101 102 103 104 105 106 107 2011-02-19 2011-02-20 2011-02-21 2011-02-22 2011-02-23 2011-02-24 2011-02-25 108 109 110 111 112 113 114 2011-02-26 2011-02-27 2011-02-28 2011-03-01 2011-03-02 2011-03-03 2011-03-04 115 116 117 118 119 120 121 -- Statistics & Software Consulting GKX Group, GKX Associates Inc. tel: 1-877-GKX-GROUP email: ggrothendieck at gmail.com
Jose-Marcio Martins da Cruz
2011-Feb-12 21:00 UTC
[R] Time unit in ts() and arima() functions
Thanks for the hint. Helped a lot Gabor Grothendieck wrote:> On Sat, Feb 12, 2011 at 7:48 AM, Jose-Marcio Martins da Cruz> > Functions which work with ts typically assume that a full cycle is > represented by 1 unit so if a full cycle is a week then a week must be > one unit and a day must be 1/7th of a unit; however, you can later > convert your series to a non-ts representation which supports dates.--