2
$\begingroup$

I want to predict a multivariate daily time series, the target output is the volume of packages that is send and the covariates are day specific information as weather, the distance to holidays but as well lagged values of the target variable. The target output time series is not stationary, when I difference it, it is. So my intention was to just difference every variable. However, some of my covariates are already stationary, so differencing makes them non-stationary.

I am not really sure if I should difference everything or nothing or just some variables, where the latter sounds not really reasonable to me. Could you please help?

Improve this question
$\endgroup$
4
  • $\begingroup$ $\Delta y_{t} = \beta_{0} + \beta_{\text{c}}\left[y_{t-1}-\left(w_{t-1}\right)\right] + \beta_{\Delta w}\Delta w_{t} + \beta_{w}w_{t-1} + \beta_{\Delta h}\Delta h_{t} + \beta_{h}h_{t-1} + \varepsilon_{i}$. Instantaneous short run effect of change in, say, $w$ on $y: \beta_{\Delta w}$. Lagged short run effect of previous level of, say, $w$ on $y: \beta_{w}-\beta_{\Delta w} - \beta_{\text{c}}$. Long run equilibrium effect of previous level of, say, $w$ on trend in $y: \left(\beta_{\text{c}} - \beta_{w}\right)/\beta_{\text{c}}$. $\endgroup$
    – Alexis
    Commented Mar 19, 2019 at 15:29
  • $\begingroup$ Boef, S. D., & Keele, L. (2008). Taking Time Seriously. American Journal of Political Science, 52(1), 184–200. $$\phantom{0}$$Banerjee, A., Dolado, J. J., Galbraith, J. W., & Hendry, D. F. (1993). Co-integration, error correction, and the econometric analysis of non-stationary data. Oxford University Press, USA. $\endgroup$
    – Alexis
    Commented Mar 19, 2019 at 15:31
  • $\begingroup$ The de Boef and Keele article give a nice bestiary of how many different time series models are the model above, but with specific constraints imposed on them a prior. $\endgroup$
    – Alexis
    Commented Mar 19, 2019 at 15:33
  • 1
    $\begingroup$ I have spent a lot of time reading on this topic and I have found various authors disagree. I think the most common view is that you make all the variables stationary unless you have cointegration. $\endgroup$
    – user54285
    Commented Mar 19, 2019 at 22:11

1 Answer 1

Reset to default
0
$\begingroup$

differencing is selected for each series based upon univariate analysis. A Prewhitening filter is the added in order to IDENTIFY . See file:///C:/Users/dave/Downloads/Ruey%20S.%20Tsay-Lec1%20(2).pdf for the motivation to convert potential predictor time series which CAN NOT be initially used to identify transfer structure into pseudo series which can be used because they are each white noise variables.

Differencing and arma structures are used to identify the ARMAX model (TF MODEL) See https://web.archive.org/web/20160216193539/https://onlinecourses.science.psu.edu/stat510/node/75 for another presentation.

Yes differencing may apply to all series.

Improve this answer
$\endgroup$
2
  • $\begingroup$ Irishstat do you difference variables that are cointegrated? $\endgroup$
    – user54285
    Commented Mar 20, 2019 at 22:48
  • $\begingroup$ IrishStat, can you provide a viable link of Tsay's lecture note? The current one is something of a pc's local disk directory. $\endgroup$
    – User1865345
    Commented Jan 12, 2023 at 23:24

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

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