0

I have different time series of hourly measures

  1. ENERGY_PRICE (What we want to predict)

  2. ENERGY PRODUCTION

  3. ENERGY DEMAND

  4. SOLAR RADIATION

  5. WIND SPEED

There's some seasonality in the ENERGY PRICE, more specifically daily, weekly and yearly.

From what I read, it is not easy to model a series with various levels of seasonality using ARIMA (Source).

As the goal is to predict one of the series from the others, @RobHyndman says that what I want is

a transfer function or ARMAX model.

How should one model and fit ARIMA/ARIMAX with multivariable time series?

Ideally, are there Python implementations of it?

I have seen various questions with R, but none with Python.

• Fitting a multivariate ARIMA model with R

• How to fit an ARIMAX-model with R?

Gonçalo Peres
  • 165
  • 1
  • 1
  • 11
  • If your question is about Python per se, it is off topic. If not, could you highlight the statistical aspects of it? ARIMA is a univariate model, so it cannot be fit to a multivariate series regardless of what software package you are using. VARMA/VARIMA would be multivariate analogues, though a simplified version, VAR, is more widespread. – Richard Hardy Dec 15 '20 at 10:57
  • @RichardHardy just reformulated the question shifting the focus from Python, however still found it relevant to mention as it may help others. – Gonçalo Peres Dec 15 '20 at 13:46

1 Answers1

1

You describe a multivariate time series. You can use the VAR model, which basically extends the ARIMA model to a multivariate case. VAR model is implemented in Python via statsmodel library. You can try and look on your case a multivariate time series and also try using each feature or a group of them as exogenous variables to better explain one of the dependent metric.

Richard Hardy
  • 54,375
  • 10
  • 95
  • 219
Arkady Mankovsky
  • 51
  • 3