In the first essay I examine whether the occurrences of the extreme price events display any regularities that can be captured using an econometric model. Here I treat these price events as point processes and apply Hawkes and Poisson autoregressive models to model the dynamics in the intensity of this process. I use load and meteorological information to model the time variation in the intensity of the process. The models are applied to data from the Australian wholesale electricity market, and a forecasting exercise illustrates both the usefulness of these models and their limitations when attempting to forecast the occurrence of extreme price events.In the second essays I explain that in the past doubts have been raised as to whether the pre-dispatch process in Australia Electricity Market is able to give market participants and market operator good and timely quantity and price information. It is the purpose of the second essay to introduce a framework to analyse whether the pre-dispatch process is delivering biased predictions of the actual wholesale spot price outcomes. Here I investigate the bias by comparing the actual wholesale market spot price outcome to pre-dispatch sensitivity prices established the day before dispatch and on the day of dispatch. I observe a significant bias (mainly indicating that the pre-dispatch process tends to underestimate spot price outcomes) and I further establish the seasonality features of the bias across seasons and/or trading periods. I also establish changes in bias across the years in our sample period (1999 to 2007). In the formal setting of an ordered probit model I establish that there are some exogenous variables that are able to explain increased probabilities of over- or under-predictions of the spot price. It transpires that meteorological data, expected pre-dispatch prices and information on past over- and under-predictions contribute significantly to explaining variation in the probabilities for over- and under-predictions. The results allow me to conjecture that some of the bids and re-bids provided by electricity generators are not made in good faith.Finally, the third essay investigates whether information from this pre-dispatch process can be useful when predicting next-day price spikes. In a preliminary analysis I establish the effect of pre-dispatch prices on the quantiles of the spot price distribution. A Quantile regression approach reveals that higher pre-dispatch prices signal only to a certain extent an increased probability of higher spot price outcomes. They also signal a higher uncertainty about the resulting spot price outcomes. I further establish whether the inclusion of information from the pre-dispatch process can significantly improve the predictability of price spikes when these are modelled as a point process (as in the first essay). The models used here are Hawkes and Poisson autoregressive models which allow for time variation (correlated to exogenous information) in the intensity process that governs the occurrence of price spikes. It transpires that the pre-dispatch process of the Australian Electricity Market does not provide any information that can be used in a systematic manner to help predicting on what days price spikes are more likely to occur.
|Date of Award||1 Aug 2015|
- The University of Manchester
|Supervisor||Ralf Becker (Supervisor) & Sydney Howell (Supervisor)|
- Electricity, Price Spikes, Australia National Electricity Market