Price Forecasting

We offer our clients proven precise pool price forecasts under three different frameworks:

Short-term Price Forecasting

This framework is aimed at week-ahead forecasting and is based on the most recent market data available such as ST and MTPASA, and the latest bidding profiles from generators in the system.

 

Weather, in addition to temporal and economic effects, has a significant impact on load in Australia. Recent load patterns are examined in conjunciton with weather forecasts and historical weather patterns from the Bureau of Meteorology on a weekly basis and are input to our regression and neural net models to obtain likely load forecasts for the week ahead.

 

The inputs are fed into our 2-4-C market simulation engine and the market is solved for each half-hour of the week to come. An example of our weather-adjusted load trace prediciton (a key input to price models) is shown below for a selected week in May in the Queensland region of the NEM. The entire seven-day period was predicted one week in advance (on May 15th).

 

Medium-term Price Forecasting

Weather and bidding patterns are harder to predict 3-to-24 months ahead. In the medium-term framework, loads are forecast using advanced time-series analysis and multi-variate random processes for load co-incidence between the states. Weather correction via piece-wise regression and neural net outcomes is possible due to ROAM's ongoing subscription to the Bureau's 3-month outlook, which details the likelihood of hot or cool conditions across Australia in the months to come.

 

Bidding and outage behaviours are drawn from long term averages and trends as applied in ourROAM Insight publication, and development is continuing on this component of the model input.

 

The 2-4-C half-hourly dispatch engine is utilised to solve the system for the forecast period. The figure below provides a sample of a 'back-casting' simulation study which we complete regularly to callibrade our data sets. This demonstrates the accuracy of our forecasting methodologies and has been used for 'what if' assessment and to study market anomolies.

 

Long-term Price Forecasting

Beyond about 2 years, generator bidding behaviour and weather conditions are almost impossible to predict based on historical trends and so these inputs are discarded. Half hourly load traces are developed using historical reference load patterns which are manipulated to meet NEMMCO and TNSP forecasts of annual energy and peak summer and winter demand forecasts such as the Low / Medium / High energy growth with 10% / 50% / 90% probability of exceedence demand forecasts.

 

Bidding and outage behaviour are drawn from long term averages and predicted generator behaviour as the market evolves.

 

Using 2-4-C, we are able to extend price forecasts as far out in time as could be practically needed.

Pleasecontact us for more information on any of the above services