1
$\begingroup$

I have hourly time-series data with n inputs and a target variable. The target should be predicted at 07:00 a.m. for the next 48 hours. Some inputs are only available until 23:00 of the previous day. So if you make the forecast at 07:00 on January 31, you have those inputs only until 23:00 on January 30. The other inputs are available for the whole period and also in the future (i.e. also over the forecast horizon of 48 hours).

What is the best practice which data to include and how?

(Would be even better if you could include some pointers to R code that tackle that problem!)

$\endgroup$
1
$\begingroup$

Section 3.1 in this paper shows the interpolation and extrapolation NNS does for aligning data of disparate frequencies in its NOWCASTING method. https://ssrn.com/abstract=3589816

In short, NNS uses a linear interpolation to the higher frequencies, and then an ensemble of a univariate estimate and a multivariate estimate for the extrapolation points.

$\endgroup$
2
  • $\begingroup$ is it just different frequencies or also that certain data are just only available up to a certain point? $\endgroup$ – vonjd Sep 17 '20 at 15:07
  • $\begingroup$ No, it's not specific to different frequencies. $\endgroup$ – Fred Viole Sep 17 '20 at 15:13

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.