0
$\begingroup$

Is it make sense that time series method to be compared with non-time series method ? and if it is possible could somebody tell me which non-times series method can I apply to make comparison between time series method and others for make a decision.

$\endgroup$
  • $\begingroup$ We'd really need to know more about what you're doing in order to answer the second part of your question well. $\endgroup$ – jbowman Sep 20 '13 at 17:08
  • $\begingroup$ @jbowman. I am doing time series prediction. I have used ARIMA, MA, ES, and random walk. I would like to know can i compare these time series based model to non-time series model like multiple linear regression ? $\endgroup$ – rose Sep 20 '13 at 23:03
1
$\begingroup$

Yes. A non-time series method that I have used (as a "hack") to model time series is a generalized additive model (GAM) with a nonparametrically-smoothed time effect. For example, using the mgcv package in R, if y is the response variable and t is time, you could run something like model = gam(y~s(t)).

This is not a standard time series model, probably due to the fact that it makes no explicit accommodation for autocorrelation over time. That is, the model assumes that the observations are i.i.d.

$\endgroup$
  • 3
    $\begingroup$ It is relatively easy to make that model account for autocorrelation. One option is to use gamm() and employ the correlation argument of the lme() function (of package nlme. Another is to follow the example in ?magic (also in mgcv) whereby you make some guesses as to the complexities of the smooths, fit model, estimate the appropriate covariance matrix and use it to update the model fit. $\endgroup$ – Reinstate Monica - G. Simpson Sep 20 '13 at 4:11

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.