One of the many challenges for anyone forecasting electricity spot prices, as Energy Link has done since first trading in 1996, is that some key factors affecting spot prices are fundamentally unpredictable in the short to medium term:  inflows to the hydro lakes are a case in point.

By Greg Sise, 25 February 2017

In my post 2016 in Retrospect (5^th Feb) I noted our forecast in December 2015 for Benmore was $75 for the year whereas the actual average turned out to be $50.49 or 33% lower than forecast.  Of course, inflows were consistently high during the year and, in addition, demand was unusually low due to a record warm winter and minimum demand for irrigation (which is now a significant portion of total demand, especially in the South Is).

We know of no way of predicting inflows accurately for the next year, and the same goes for temperatures.  In fact, we have two sayings at Energy Link:  “just when you think you’re in an inflow pattern, that’s when it’s going to change”  and “everything [inflow-wise] can be completely different in three months”.  It is no exaggeration to say that the hydro lakes can go from being so low that there is a national campaign to conserve electricity to spilling excess water within three months:  it has happened before, and vice versa, and it will happen again.

When we publish a headline forecast, we are not effectively saying “this is what we think the electricity price will be”.  Rather, we are publishing the average value from hundreds or thousands of scenarios covering a wide range of inflows, demand, planned outages, gas prices, carbon prices and market behaviour, and hence a wide range of possible pricing outcomes.  The reason we only run this many scenarios is that we can’t afford a supercomputer to run millions!

One could argue that factors other than inflows and demand caused our 2016 forecast to be too high, but we can test this (as we have done many times before) quickly and simply by rerunning the December 2015 forecast with actual inflows and actual demand -  this is called back-casting.  When we do this the average price produced by the forecast model is $56, now only 10% higher than the actual average of $50.49. 

So just by using actual inflows and demand we reduced the error by just under 80%.

We could reduce the error further by rerunning the forecast with other random inputs having their actual values, unplanned plant outages (generation and grid) being a case in point, but the reduction in error would be less than it is for inflows and demand.  When we look at forecast errors over a longer period, typically at least two years and ideally more, the errors should average out close to zero because random inputs like inflows, demand and unplanned outages can be higher or lower than average.  However, there can be long periods when, for example, inflows can remain lower than average, in fact for up to two and a half years according to the historical data.  And dry years can be back-to-back.

The key message here is that factors like inflows and demand are random and volatile, and parties exposed to spot prices need to understand the risks.  It is one thing to take note of a headline forecast, but ideally one should be reading further and taking note of the range (distribution) of possible pricing outcomes included with our forecasts.  AND BE PREPARED!