Energy Markets can be a bit different from financial markets. For example they have more seasonality, mean reversion, and spikes than the stock markets. In general they are a bit more predictable. That means that the modeling techniques will be different although the general concepts are the same. Energy markets and financial volatility markets have a lot in common so you may like to read this market as a special case of the financial markets.

The most interesting energy market to us at DDNUM is the electricity market because New Zealand is one of only a few liberalised electricity markets worldwide so there is the opportunity do do some leading edge research. The New Zealand market seems to have a few problems since there are power shortages about every second year now. That market only became competitive in 1999 so there isn't a lot of data to study. Hence some clever analysis and modeling is required.

There are a number of ways of modeling prices: cost based models which take into account the costs of generating electricity, game theoretic models (which recognise that markets participants are bidding against each other), fundamental models which look at the economic relationships between the fundamental drivers such as weather, stochastic models which take into account the randomness in prices so as to calculate risks, technical analysis models which look at what happened in the past, and statistical learning models which add artificial intelligence and nonlinearity into the mix.

Electricity demand has seasonal components - winter is generally the time of highest demand in temperate climates. Electricity supply also has seasonal components when much of a market's supply comes from natural weather-related resources (or even tidal resources). As well as requiring seasonal modeling these components add to statistical complexity by adding serial correlation to the data. Most statistical techniques cannot handle correlation among the data values so do not work correctly (but may seem on the surface to be giving good results - dangerous). Serial correlation is when today's price change is influenced by last week's price change. Generally in the stock market there is little such correlation but in the electricity market there is.

Mathematically we must use spectral analysis to look at seasonality. At DDNUM we like to use wavelets rather than the older Fourier analysis that engineers tend to use. Wavelet functions are quite localized in time and require less data than Fourier analysis so can respond faster to changing market conditions.

Energy prices are generally regarded as mean-reverting. That is, if they move away from the mean they are subjected to forces pushing them back. This introduces long range dependence in the data - what happens today can affect what happens next year. In contrast, stock prices generally have a lack of memory. Electric prices contradict many of the paradigms used in financial markets. Mathematically, this means that common statistical techniques are not valid. This has given rise to new techniques such as "Detrended Fluctuation Analysis." For us this means plenty of possibilities for new techniques to research. And for electricity customers plenty of opportunities for competitive advantages.

Spikes are sudden and (by definition) unanticipated extreme changes in the spot market price. They (by definition) fall outside the normal statistical range of prices. Statistically they are outliers but, unlike usual statistical techniques, cannot be rejected as untypical. The reason is that a single spike can eliminate the gains of a whole year's trading. Spikes are unique to electricity because other commodities do not jump in this way (you cannot store electricity but you can store grain - this removes an element of smoothing from the electricity prices). Mathematically, spikes are not "continuous" changes in price so the usual statistical techniques cannot be used.

The spot price needs to be forecasted on a half-hourly basis (in New Zealand). Both the magnitude of the load and its geographic distribution need to be predicted. Some of these predictions can be made with very small error. But long range forecasts can have a large error. Different methods are used according to the time frame.

Since humans use power human behaviour has to be predicted. For example, what's on TV (an All Black test match for example) has an influence. And power usage jumps during advertising breaks. So Statistical Learning and other artificial intelligence methods are required. Good results have been achieved by researchers using Support Vector Machines.

There is a risk in buying electricity from one market (say the spot market) for delivery to another market under different terms (say delivering to retail customers at capped prices). This risk is required to be hedged. That is, the risk needs to have a counteracting trade to cancel out the effects of adverse price moves. These trade usually use forward contracts, futures, swaps and options. These have the general name derivatives because their price is derived in some way from the electricity price.

Correctly pricing derivatives is important. Since their price is derived from the electricity price the relationship between the two prices needs to be formulated. This is more difficult than for derivatives in the financial markets because electricity cannot be stored. So the classical spot-forward price relationship is violated. Other concepts from financial markets need to be modified for electricity.