In what scenarios would you use multivariate time series forecasting over univariate, and why?
Instruction: Discuss the considerations for choosing multivariate forecasting techniques and the advantages they offer over univariate methods.
Updated
Example Answer
I would use multiple series or external predictors when they provide useful information available at the forecast date. For demand, planned promotions may help; actual future competitor sales would not be available unless separately forecast, which adds uncertainty.
I would compare a univariate baseline with the richer model at the required horizon. Related variables do not automatically improve the forecast: redundant predictors, short histories, and unstable relationships can add variance or leakage. A joint model such as VAR can be useful when forecasting several interrelated series, while regression with time-series errors can suit one target with known predictors. I would choose the simpler option unless the added information improves later-period performance and is practical to maintain.
Make it your own
Name the additional variable, when it becomes available, and the baseline it must improve upon.
Why this works
It makes information availability and evidence drive model choice.
Follow-up to practice
What if the added predictor must itself be forecast?