Long-Term Extrapolation of Electricity Forward Curves - A Novel Approach Utilizing Forecasts and Risk Premia
Master thesis
Permanent lenke
https://hdl.handle.net/11250/3024655Utgivelsesdato
2022Metadata
Vis full innførselSamlinger
Sammendrag
Vi ekstrapolerer forward kurven basert på langsiktige spotprisprognoser og en antakelse om at risikopremien konvergerer i modningsdimensjonen. Hypotesen om en konvergerende risikopremie blir vurdert gjennom en paret t-test for to- og treårskontrakter. Ekstrapoleringene gjennomføres for tre ulike metoder: Level, Log-Return, og Rate premie. For å beregne Level premie, benyttes differansen mellom gjennomsnittlig forward og prognose over perioden. Log-Return premie benytter log-endring mellom gjennomsnittlig forward og prognose. Tredje metode, Rate premie beregnes ved gjennomsnitt av log-endring mellom forward og prognose for hvert modningstidspunkt over perioden diskontert for modningstiden. Vi benytter beregnede premier og langsiktige prognoser til å ekstrapolere forwardkurver fram til 2050. Vi tester nøyaktigheten mellom metodene og forward kurven ved å benytte out-of-sample tester. Resultatet målt i MAPE er 8.364%, 8.256%, and 11.439% for Level, Log-Return, og Rate premie. Basert på resultatene kan vi konkludere at Level og Log-Return metodene er signifikant mer nøyaktighet enn Rate premie. Markedsaktører på både tilbud- og etterspørselssiden av kraftmarkedet kan benytte våre ekstrapolerte forward kurver i forbindelse med risikohåndtering og produksjonsplanlegging. We extrapolate the forward price curve based on long-term spot price forecasts and an assumption of a converging forward risk premium in the maturity dimension. The hypothesis of a converging forward risk premium is examined using paired t-tests on the forward risk premia of two- and three-year contracts. Extrapolations are produced using three distinct forward risk premium methods, measuring the maturities between one and two years ahead. The three forward premium methods are referred to as Level, Log-Return, and Rate premium. To calculate the Level premium, we take the difference between the average forward and forecast over the period. For the Log-Return premium, we calculate the log change between the average forward and forecast. Lastly, to calculate the Rate premium, we take the log-return between the forward and forecast price of every maturity in the period, discounting each by their maturity, and then taking the average of these values. The resulting extrapolated forward curves extend to 2050. We measure the out-of-sample accuracy between the extrapolated forward curves and the elementary forward prices. The accuracy of the extrapolated curves measured in MAPE is 8.364%, 8.256%, and 11.439% for the Level, Log-Return, and Rate premium, respectively. Based on the results, we can conclude that the Level and Log-Return methods provide significantly higher accuracy than the Rate premium approach for all investigated accuracy measures. Market participants on both sides of the market can benefit from the long-term forward prices, for production planning and risk management purposes.