[R] stationarity tests
spencerg
spencer.graves at prodsyse.com
Fri May 22 04:50:17 CEST 2009
The following searches for help pages in contributed packages
including terms "stationarity" or "unit root":
library(RSiteSearch)
st <- RSiteSearch.function('stationarity')
ur <- RSiteSearch.function('unit root')
ur. <- st|ur
nrow(st) # 68
nrow(ur) # 122
nrow(ur.)# 180
HTML(st&ur) # Shows the 10 with both terms
summary(ur.) # A summary by package
HTML(ur.) # Shows all 180 sorted by package then score for the help page.
You may also be interested in the Box-Ljung test. For this, try
the following:
bl <- RSiteSearch.function('Ljung')
HTML(bl)
"The partial autocorrelations may be estimated by fitting
successively autregressive processes of orders 1, 2, 3, ... by least
squares ... and picking out the estimated phi.hat[1,1], phi.hat[2,2],
phi.hat[3,3], ... of the last coefficient fitted at each stage." (Box
and Jenkins, 1975, Time Series Analysis, Forecasting and Control,
Holden-Day, sec. 3.2.6; see also
"www.itl.nist.gov/div898/handbook/pmc/section4/pmc4463.htm")
Your rules for reading ACF and PACF sound right to me.
Hope this helps.
Spencer Graves
mauede at alice.it wrote:
> How can I make sure the residual signal, after subtracting the trend extracted through some technique, is actually trend-free ?
> I would greatly appreciate any suggestion about some Stationarity tests.
>
> I'd like to make sure I have got the difference between ACF and PACF right.
> In the following I am citing some definitions. I would appreciate your thoughts.
>
> ACF(k) estimates the correlation between y(t) and y(t-k) like an ordinary correlation coefficient.
> ACF is the simple ( i.e. unconditional ) correlation between a time series and it's lags thus
> y(t)=a+b*y(t-k) gnerates the kth autocoreelation coefficient (b).
>
> If we have form y(t)=a+b*y(t-1)+c*y(t-2) .. then (c) is the PARTIAL AUTOCORRELATION COEFFFICIENT or in other words the
> CONDITIONAL CORRELATION of lag 2 given lag1
> PACF(k) estimates the correlation between y(t) and y(t-k) adjusted for the effects of y(t-1), ..., y(t-k+1).
>
> Model identification is achieved by looking at the pattern of the ACF and PACF.
> - If the ACF dies off exponentially, but the PACF has p spikes, AR(p) is indicated.
> - If the ACF has q spikes and the PACF dies off exponentially, MA(q) is indicated.
>
> The ACF and the PACF for the resulting stationary series is used to determine the best B/J model for the series according to the following rules:
> a. If the ACF trails off and the PACF shows spikes, then an AR model with order p = number of significant PACF spikes is the best
> model.
> b. If the PACF trails off and the ACF shows spikes, then an MA model with order q= number of significant ACF spikes is the best model.
> c. If both the ACF and the PACF trail off then a ARMA model is used with p=1 and q=1.
>
> Thank you very much,
> Maura
>
> Thank you very much.
> Best regards,
> Maura Edelweiss
>
>
>
> tutti i telefonini TIM!
>
>
> [[alternative HTML version deleted]]
>
> ______________________________________________
> R-help at 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.
>
>
More information about the R-help
mailing list