Abstract

This paper proposes a new approach to hybrid forecasting methodology, characterized as the statistical recalibration of forecasts from fundamental market price formation models. Such hybrid methods based upon fundamentals are particularly appropriate to medium term forecasting and in this paper the application is to month-ahead, hourly prediction of electricity wholesale prices in Spain. The recalibration methodology is innovative in seeking to perform the recalibration into parametrically defined density functions. The density estimation method selects from a wide diversity of general four-parameter distributions to fit hourly spot prices, in which the first four moments are dynamically estimated as latent functions of the outputs from the fundamental model and several other plausible exogenous drivers. The proposed approach demonstrated its effectiveness against benchmark methods across the full range of percentiles of the price distribution and performed particularly well in the tails.

Highlights

  • In contrast to the extensive research on methods of forecasting electricity spot price expectations, the full predictive specification of price density functions has received much less attention

  • This is because by combining fundamental and statistical models, it is possible to incorporate the impact of both the projected fundamental changes in the market and the empirically revealed behavioral aspects. Such hybrid processes have typically focused only upon adjusting the mean bias and not the full density specifications. It is with this motivation, that we investigate the forecasting ability of a novel, fully parametric model for hourly price densities, which is sufficiently flexible in specification to accurately recalibrate the forecasts from a fundamental market model

  • [12], for example, demonstrated that forward prices can be expressed as a Taylor expansion involving the moments of the spot price distribution. We develop this methodology in the context of medium term electricity price forecasting, which is relatively under-researched compared to the short term, as noted by [13]

Read more

Summary

Introduction

In contrast to the extensive research on methods of forecasting electricity spot price expectations, the full predictive specification of price density functions has received much less attention. As an alternative to quantile regression, we investigate several fully parametric specifications for hourly electricity prices (from the Spanish market) to find a distributional form that fits all observed density shapes acceptably well but is expressible in terms of its first four moments These four moments can, under an appropriately specified distribution, in turn be capable of being estimated as dynamic latent variables from a Linear Additive Model (following [10]) linking them to one or more exogenous variables (e.g., the forecasts from a fundamental market model).

Overview of the Methodology
Implementation of the Proposed Methodology
Density Selection
Selection of the Regressors
Performance Analysis
January
Combinations for Increased Accuracy
Conclusions
Findings
Medium-Term
Full Text
Published version (Free)

Talk to us

Join us for a 30 min session where you can share your feedback and ask us any queries you have

Schedule a call