Which type of time series to use?

[purifyz]

Hey,
I have a data set which includes weekly/monthly sales and questions about customer satisfaction. I have already done EFA/CFA based on these satisfaction questions (turns out it's just one satisfaction factor)

I would like to regress sales on satisfaction, but I think I should include a time component (based on literature review, i.e. past satisfaction affects future satisfaction and same with financial data)

Which Time Series Model should I use?

Btw. I am a complete newbie in time series, now reading "Time Series Analysis and Its Applications With R Examples" by Shumway and Stoffer, hopefully getting some answers there.

Thanks in advance

 

[noetsi]

Are you trying to use variables to predict some other variable over time or are you predicting some univariate time series like say weekly sales (this would be the only variable, you are just looking at the pattern over time) If the later is true I would suggest some form of exponential smoothing because it is simple and has good accuracy. If the former is true there are no easy choices. Multivariate ARIMA, ARIMA with predictors, is a pain to do, regression with autoregressive errors is probably your best bet again because it is relatively simple. Other types of time series such as GARCH/ARCH error correction models etc are very hard to do IMHO.

 

[david moore]

Are you trying to use variables to predict some other variable over time or are you predicting some univariate time series like say weekly sales (this would be the only variable, you are just looking at the pattern over time) If the later is true I would suggest some form of exponential smoothing because it is simple and has good accuracy. If the former is true there are no easy choices. Multivariate ARIMA, ARIMA with predictors, is a pain to do, regression with autoregressive errors is probably your best bet again because it is relatively simple. Other types of time series such as GARCH/ARCH error correction models etc are very hard to do IMHO.

I agree with you.

 

[GretaGarbo]

May I suggest a look at Stock and Watson Dynamic Factor models?

 

[noetsi]

What statistical software (what package if R) runs that? I never heard of it before.

 

[GretaGarbo]

What statistical software (what package if R) runs that? I never heard of it before.

If that is a question about Stock Watson's models then:

Look at Petris and Petrone's paper and about packages dlm and KFAS and BMR.

I don't know much about this but it would be nice to hear if anybody has any views about these models.

(Maybe is this of interest for Noetsi since survey data is mentioned.)

 

[noetsi]

I suspect it is beyond me, but I intend to read it anyhow Its not really survey data I am interested in. It is time series. I mainly predict things. I have predicted spunky would get no cake as long as Dason is around and I have yet to be wrong