Time-series prediction
Problem
Predicting energy prices using: consumption and production of energy, total export/import, and the exogenous factor – average daily temperature (capital city temperature as a proxy), as predictors.
Solution
- Development of statistical models (single/multivariable regression with OLS, robust OLS, ridge and lasso regression) that eliminate (multi) collinearity in the data set.
- Development of (S)ARIMA(X) models that improve performance over models developed in (i).
Results
Many of the proposed methods yielded poor results due to the high level of multi-collinearity in the data set. The metrics derived from shrunk regression models did not show significant variation compared to the metrics from ordinary regression models. Log-transformation and differentiation guaranteed stationary in the time-series data sets.
