The increase in prices and price volatility observed in NSW and Victoria raise the question whether the source of this change is attributable, to a substantial extent, to South Australia. From a statistical perspective, causality is a difficult issue to address in a complex system. To start, there is need for a supporting physical explanation for the hypotheses to be tested, which cannot be offed here. However, statistical evidence of causality can be a necessary if not sufficient condition in establishing causality.
To consider causality, we need to depart from the visualisation approach that has been adopted so far. Granger causality is a relatively easy to understand statistical test of causality, which is based on temporal correlation (see https://en.wikipedia.org/wiki/Granger_causality). The idea is that what is happening, say in NSW, can be predicted by what has happened recently in that region. If what has happened recently in South Australia adds significantly to the quality of that prediction, then there is evidence of cause and effect.
Briefly, the procedure used for the tests of causality in NSW and Victoria is to: transform our volatility measure, the mean absolute deviation in price from trend, in each region. The purpose is to create a volatility distribution that is approximately normal. To do so we use a cube root (as say opposed to a logarithm square root as deviations are both positive a negative). The next step is to use robust regression (https://en.wikipedia.org/wiki/Robust_regression) to estimate volatility in a given region as a functions of past volatility in all the regions. Lastly, standard errors are corrected for temporal correlation and variation when estimating the precision and the significance of the estimates.
The results are summarised in Table 2. It should be noted that the closer the significance level is to zero, the more significant is the result (the confidence level is equal to one minus the significance level). The results of the NSW regression show that price volatility in South Australia and Victoria have a highly significant and positive impact on price volatility in NSW. The results of the Victorian regression show that price volatility in South Australia has a highly significant and positive impact on the price volatility in Victoria. Lastly the South Australian regression is highly symmetric with respect to that of Victoria, with volatility in Victoria impacting on volatility in South Australia. Volatility in NSW does not significantly impact on South Australia.
On balance, therefore, the results of the causality analysis are consistent with the conclusion that the volatility in spot price in South Australia has been passed on to NSW and Victoria. The results that price volatility in Victoria is impacting on NSE and Victoria suggests that the source is not confined to South Australia and while NSW may not be contributing to price volatility in South Australia and Victoria, it has a stake in sorting out the issue.
It appears that the high share of wind generation in South Australia may play central but not a singular role in creating higher levels of price volatility and consequent higher prices. The next few blogs will look more directly at wind generation and prices in South Australia.
