I’ve previously alluded to the fact that there are some ‘games that self-forecasters can* play’ in several prior articles, and have collated a starting set of articles about these here in this category (to which this one will be added).
* readers should clearly note the deliberate use of the word ‘can’ rather than necessarily ‘do’ … because we’re aware that not all self-forecasters choose to play these games:
(a) indeed we invested in one of those that choose not to play these games back at the start of 2024.
(b) This was also discussed in this review of our 25th year of our service.
In the past 5 months, the AER has provided warning on at least 3 occasions to Semi-Scheduled assets, and other stakeholders, about these games:
- On 4th December 2024, we wrote about how ‘AER sends clear message on self-forecasting behaviours’ during the AEMO Intermittent Generation Forum; and
- On 20th February 2025, we belatedly wrote about how ‘AER published ‘Semi-Scheduled Generator Compliance Bulletin’ on Tuesday 4th February 2025’.
- At the end of February 2025, the ‘AER wrote letters to Semi-Scheduled participants (and others) with respect to ‘Expectations surrounding the use of self-forecasting’’.
As we wind down to the commencement of the new Frequency Performance Payments method** on 8th June 2025 (under 7 weeks away), I thought this article would be timely.
** remember, with reference to these ‘Two diagrams to help illustrate the changes coming with FPP (Frequency Performance Payments)’ that this change is more than ‘just’ a replacement of the (flawed) ‘Causer Pays’ method for allocating Regulation FCAS costs.
I’ve been promising this article for a while, but have not invested time to provide yet – but:
1) Know the clock is ticking down to FPP; and
2) I heard the term ‘commercial optimisation’ used by some to explain self-forecasting in the past week or so …
Hence thought it was timely today to invest a bit of time in this article … which is really just a ‘Part 1’ in what might evolve to be a larger series, if readers are interested (time permitting).
It’s a game with Two Separate Legs
In this article I just want to lay out a general framework, as I see it – starting with the fact that readers need to understand that it’s a game with two discrete legs (both needing to be played in tandem), as follows:
| Dates | Leg #1
Minimising a (Semi-Scheduled) DUID’s share of Regulation FCAS costs |
Leg #2
Avoiding AEMO suppression |
|---|---|---|
|
Until Sept 2001 |
For completeness, note that there were no competitively bid FCAS markets operating in the first few years of the NEM. |
|
|
From Sept 2001 to |
Similarly, during this second period there were FCAS commodities being procured, so Regulation FCAS costs allocated (via the ‘Causer Pays’ method) … but: (a) In the earliest years there were no Semi-Scheduled units (not until 31st March 2009); and (b) Self-forecasting for Semi-Scheduled units was not possible until 2019 (i.e. beyond early ARENA-sponsored trials). So these particular games did not start to evolve until September 2019. |
|
|
Background reading |
With respect to Leg #1, there are a couple articles flagged here that are suggested background reading… On 25th March 2022 in his article ‘“Rise of the Machines” – the increasing role of auto-bidding and self-forecasting in the modern-day NEM’ , guest author Jonathon Dyson included an illustration of what he called ‘purposeful over and under forecasting’. Also, readers might like to re-read the article ‘What inputs and processes determine a semi-scheduled unit’s availability’ that Linton posted on 11th August 2023. The key point to understand here is that (once a DUID has a self-forecast vendor that’s passed the AEMO’s initial 8-week assessment period), then:
In other words, the submission of a self-forecast is optional. |
On 14th April 2023 we wrote ‘Delving deeper into dispatch availability self-forecasting performance’. In that article, Linton explained the 1-week, 4-week and 8-week loops that AEMO use to ascertain whether to suppress a self-forecast, or not.
|
|
From ~Sept 2019 |
Note that it’s the (flawed) ‘Causer Pays’ method for allocating Regulation FCAS costs that is relevant over this time period: …. but only for another ~7 weeks. A common misconception I’ve heard voiced is that the ‘Causer Pays’ method
… however it’s important for readers to understand that:
On 13th March 2025 in the article ‘Three headline observations, about the use of proxies for System Frequency in Frequency Control mechanisms’ we illustrated just how flawed the Frequency Indicator has been, in several respects:
Taken together, readers will understand that this opens the door to Leg #1 of these ‘Games that Self-Forecasters Can Play’. Now readers should note that the AEMO has logic such that some dispatch intervals are excluded from the final calculation … but our sense is that these (after the fact) exclusions do little to reduce the up-front incentive to play Leg #1 of the Game.
|
Following from Linton’s articles above, readers should consider that:
… readers might understand that, in Leg 2 of this Game:
It’s important for readers to understand that the change from ‘Causer Pays’ to FPP represents no change in requirement in this Leg #2 of these ‘Games that Self-Forecasters Can Play’. |
|
From 8th June 2025 |
Like many who will be watching with keen interest to see what unfolds with the transition to FPP (including the AER, as they noted in their letters to stakeholders):
One change that does seem to be an upside is that the AEMO has changed from using a Frequency Indicator to using the new Frequency Measure which is (as discussed here) a much better proxy, because it is both:
… though it’s still not a perfect proxy. — Readers might also like to read and reflect on what Jack Fox flagged as ‘an Emerging Risk?’ in his first article here. |
|
We hope that the above will stand readers in good stead, to understand the multi-dimensional nature of the ‘games that self-forecasters can play’.
What Clues might one look for, to spot these games?
This question is one that might be explored in more detail in a subsequent follow-on article, but readers for now might reflect on the following questions as a starting list…
* CAVEAT about Guessing Motivation
Remember (with respect to the questions below) that, in both cases, it’s impossible to know motivation (but it does not stop One guessing).
Q1) What percentage of time is a self-forecast submitted, and used in dispatch?
Understanding Leg #2 of the Game above, readers might understand that:
(a) If a self-forecast is submitted and used in dispatch a high percentage of time over a long period, then it would seem unlikely that Leg #2 of the Game could possibly be being played?
(b) Whereas if it’s only submitted and used in dispatch a lower percentage of time over a time period:
i. there may be several different reasons (*see caveat above) for this
ii. with Leg #2 of the Game only one possible reason.
On 18th March 2024 in writing ‘How many Semi-Scheduled units were submitting Self-Forecasts through 2023 (and how many of these were actually used in dispatch)?’ we flagged the possible use of the GSD2023 back then to produce this sort of analysis (though note that the GSD2024 was published more recently as an update).
Q2) What particularly happens during the semi-dispatch cap periods?
In other articles (some listed here) we noted that:
(a) The more difficult dispatch intervals to forecast availability is where the Semi-Dispatch Cap (SDC) is in place
i. where the Semi-Scheduled unit is being ‘constrained down’ for economic or constraint-related reasons.
ii. in such case, where a simple ‘persistence forecast’ method is not possible to apply (and where a real forecast of UIGF would be more valuable to the AEMO).
(b) So readers might like to look particularly at what happens to a unit’s self-forecast during SDC periods?
i. In some cases we have seen it continue to be used;
ii. But in other cases we have seen it self-suppressed (*see caveat above), in which case NEMDE just ‘fails back’ to the UIGF already being produced by the relevant ASEFS or AWEFS process.
On 27th February 2023 in writing ‘Some revelations in GenInsights Q4 2022 about Self-Forecasting’ we touched on some of these questions.
Q3) What about outside of semi-dispatch cap periods?
Conversely, outside of SDC periods, do you spot a consistent, contiguous period where the actual output of the Wind Farm or Solar Farm (i.e. the FinalMW) is consistently above (or below) the Availability coming from the self-forecast.
If you find periods like this, then:
(a) Perhaps there is some other reason why the self-forecast is inaccurate; or
(b) Perhaps (*see caveat above) it is another indication of the use of ‘Games that self-forecasters can play’.
Q4) For solar farms, can you spot ‘lunar megawatts’?
Readers will understand that solar farm availability should be non-zero during daylight hours (subject to technical limits and cloud cover and so on), but zero whilst the sun is absent at night.
However on occasions we’ve seen non-zero availability at strange overnight hours
(a) … which might be for a few different reasons (not just due to Leg #1 … (*see caveat above)),
(b) but is certainly an obvious head-scratcher.
Where we’ve written articles that have identified some ‘lunar megawatts’ (and if we have remembered) we might have tagged them with ‘lunar megawatts’.
Q5) Starting with the Causer Pays Factor?
Alternatively, Analysts might like to start at the ‘other end’ of the process, to start with those DUIDs paying low (or no) ‘Causer Pays’ charges and work backwards to ascertain how this was achieved. In this case understand that:
(a) There may well be genuine, valid reasons why low (or even no) ‘Causer Pays’ costs are achieved;
(b) A big one of these can be the (genuine, valid) use of FCAS portfolios
i. … in which a participant might group some badly performing unit with some much more controllable units to net out to zero cost across the portfolio
ii. noting that this won’t be possible in the FPP environment (which is probably a good thing)
(c) But perhaps (remembering the * caveat above) it is another indication of the use of ‘Games that self-forecasters can play’.
Again, through publications like the GSD2024 most recently (and earlier editions) we’ve made it easy for clients to spot DUIDs paying $0 for Regulation FCAS costs.
… and that’s where we’ll leave this article for now.
Be the first to comment on "Games that self-forecasters can* play (Part 1 – An Overview)"