Copyright © Gaffney Systems Pty Ltd - ABN 67 134 159 167
Gaffprice uses its own proprietary algorithm to price NEM electricity contracts. Average pool prices for the future are set by market rates,
and the future distribution of half-hourly prices and customer load is based
upon an analysis of historical data.
The inputs required are:
- The forward curve (e.g. ASX electricity futures market)
- Historical half-hourly customer load (provided by the customer, from the
retailer's own database, or Net System Load Profile (NSLP) data)
- Historical pool price and demand data for the customer's state (from
the MMS Infoserver database or AEMO website)
- A list of public holidays
for the capital city of the customer's state (past and future dates).
Although historical data is used in the analysis, and actual results are
presented as part of Gaffprice's output, it is not sufficient to use a
historical figure for either the tariff cost or the weather sensitivity* premium, as:
- These numbers vary greatly from year-to-year (see the table of annual
historical weather premiums below for Energy Australia's NSLP from 2003 to
- They do not represent the market cost, which is what the retailer must pay
(or refrain from selling) to cover the customer's future load. There is little gain
financially in setting a customer energy tariff for Q1 Peak at $30/MWh if forward
contracts for Q1 Peak are $150/MWh.
- This is particularly true for customers with large air-conditioning
loads - their cost outcome for Q1 Peak in a given year could turn out quite
low, but if these results are used to price their future load, the cost of
covering their output during volatile periods may be grossly underestimated.
An example of this with the Energy Australia NSLP data would be if the 2003
weather premium had been used when pricing 2004 load - it would have
resulted in an underpricing of $5.78/MWh, which is more than $5 million per
year for 100 MW of load.
- Gaffprice analyses pool price and customer load, looking at:
- Their average and variance, for different times of day, different days of the
week, and each month of the year.
- How closely their movements are linked to regional demand.
- Gaffprice forecasts the customer's future half-hourly load, based on its
historical data (the more years of data provided, the better).
- For each future period,
Gaffprice assesses how much higher or lower the half-hour of load would be than usual.
- If there is a strong relationship between customer load and regional demand,
and a strong link between regional demand and regional pool prices, a high load period
implies a higher-than-usual pool price for that period.
- For example, if a retail company was consuming a lot of power
for air-conditioning on a summer afternoon, regional demand and pool prices would probably be higher than usual.
- Using this pool price result as a median, Gaffprice calculates the average of a
range of possible pool price outcomes for that period.
- All of these pool prices for each period are shifted up and down until they fit with each contract on the
- For steps 4, 5 and 6, a special Bidstack Effect
is used to determine the distribution of expected pool prices.
Gaffprice does not use detailed bottom-up forecasting to set contract prices, since:
Next Page: Bidstack Effect
*Weather sensitivity (also known as flex) is defined here as the variation in customer load that coincides with changes in regional demand
(and therefore regional pool price),
after accounting for variability due to time-of-day, month, and day type (weekday, public holiday, etc). South Australia tends to have the strongest link between
regional demand and pool prices in the NEM.
- Tariffs are based upon market prices for energy.
- Average pool prices in the NEM
driven by a few expensive days each year. It is very difficult to forecast the
number or severity of these, as they are driven by extreme weather, supply outages,
generator bidding, and complex electricity network constraints. The
future level of the customer's load on these extreme days is also uncertain.
- It takes a large amount of resources to develop
and maintain a forecasting model that
comes even close to representing an accurate view of future outcomes.
for most of the driving factors in forecasting models tend to be based upon historical data,
reducing any possible advantage over top-down modelling.