How do you evaluate the accuracy of a time series forecast?
Instruction: Discuss methods for assessing the performance of a time series forecasting model.
Updated
Example Answer
I would evaluate forecasts made from past data against outcomes observed later, using the horizon the business actually needs. A model that fits historical values well may still forecast poorly. I would compare it with a naive or seasonal-naive baseline over several forecast origins.
MAE is easy to explain in the target's units, while RMSE gives more weight to large errors. I would avoid MAPE when actual values are zero or close to zero; scaled errors can help compare series with different magnitudes. I would also inspect bias and performance during important periods, rather than hide those behind an overall average. For planning decisions, I would check prediction-interval coverage and the cost of over- versus under-forecasting.
Make it your own
State your target, operational horizon, baseline, and which errors matter most.
Why this works
It evaluates real forecasts and connects the metric to the decision.
Follow-up to practice
What happens to MAPE when demand is zero?