Can anyone help in model estimation ? The following are the ACF,PACF and the plot of the sample respectively.
$\begingroup$
$\endgroup$
4
-
$\begingroup$ What is your question ? What is your problem ? What do you wanna achieve ? You have to be way more precise. $\endgroup$– Julien D.May 29, 2014 at 9:51
-
$\begingroup$ @JulienD.- which arma/arima model would fit best? I cannot make out from the acf graph.. $\endgroup$– user2458552May 29, 2014 at 9:57
-
$\begingroup$ also from the first plot, does it need differencing? or can it be considered stationary? $\endgroup$– user2458552May 29, 2014 at 9:57
-
$\begingroup$ If your series is stationary (it might be nearly so), and you're not worried about the discreteness problem, then it may be approximately AR(1)xSAR(12). But I'd fit the AR(1) and see what the residual looked like. $\endgroup$– Glen_bMay 29, 2014 at 14:43
Add a comment
|
3 Answers
$\begingroup$
$\endgroup$
It seems that the time series is not stationary from the first figure. We should see the time series should have a constant mean and constant variance stable no trend ...
On the other hand, some hypothesis test can be employed, like Ljung-Box test, Augmented Dickey–Fuller (ADF) t-statistic test, Kwiatkowski-Phillips-Schmidt-Shin (KPSS) test
$\begingroup$
$\endgroup$
You can try fitting an AR(10) or AR(11) process. Run Ljung-Box tests for various lags on the residuals after fitting the model to see if the p-values of the Q statistic are less than a certain signifiance level (0.05 perhaps). If not, then your null hypothesis, i.e. that your fitted model describes the sample well, holds.
