Abstrakt
Recent research has found that electricity consumption is a very useful variable in economics. In many applications it might be desirable to decompose electricity consumption into unpredictable and deterministic (or highly predictable) component.
We want to find out whether forecasting works better if we model electricity load independently for each hour or if we model in the first step the average daily consumption and in a second step we model for each hour deviation from this average. We therefore compare two simple, yet flexible models for hourly electricity consumption in Germany.
Both models use temperature, industrial production, hours of daylight and dummies for days of the week and month of the year as explanatory variables. We find that the first model, despite being simpler, forecasts hourly electricity demand more precisely.
This indicates that hourly electricity consumption represents various goods, and should be modelled separately for each hour.