In NSW on 12 October we saw end-of-interval demand and prices take a bumpy ride in the afternoon.
A hypothesis as to what drove the demand and price fluctuations is presented here.
Naturally though, the full picture will be wider than one narrow slice through the data. That why we’ve started using A framework for exploring what factors were contributing to the volatility in NSW on Monday 12th October 2026 .
We are exploring wider factors alongside this one. Find those here.
This article fits into the “on-the-day” timeframe, and hones in on demand-side factors.
Price then demand
A good presentation of price and demand over the spicy intervals on the 12th is collated in A wall of numbers … as the next step in exploring the price/demand oscillations in NSW on Monday 12th October 2026.
For the upcoming analysis here, we work backwards from the price that was set, towards demand and drivers of those changes.
The supply-side bid stack was tight
By way of example, and noting that this is just one interval, when the first price spike happened there was only 335 MW separating the target supply from dispatched offers priced up to $300.
Moreover there was only 159 MW between $300 and $590 (lumped into the $500 to $1000 bucket), and nothing else until the 176 MW priced above $20,000 (lumped into the top bucket).
The point to make with this, is that the bid stack was very tight. A change of 177 MW could have been the difference between a high price and a low(ish) price.
Prices are set based on meeting end-of interval demand
Prices fluctuated from one interval to the next, and between actual and forecast made 5 minutes prior.
As prices are set based on meeting demand, we turn attention to demand.
In the NEM supply targets are set to meet the end-of interval demand. That’s a forecast demand value.
That end of interval demand is based on what’s happening at the start of the interval and incorporates expected movement to the end of the interval via the operational demand forecast.
Here are the NSW demand expectations up to 14:30.
The market (total) demand chart above shows how the end-of-interval demand expectations fluctuated; the zigzag from 14:10 to 14:30 and 15:10 to 15:45 also.
One hypothesis for this is:
- Something drives actual demand to be lower (e.g. 14:10, relative to the expectation).
- We don’t see this published specifically. It is the end-of-interval expectation that is published.
- When actual demand is low, relative to the expectation, the next-interval’s expectation is ‘rebased’, if you will, to that different level.
- If the thing drove lower actual demand stops, we might observe demand to resume. Relative to the new expectation a higher-than-expected actual demand would result. In our example in 14:15, relative to the low expectation.
- A higher actual demand level at the end of 14:15 supports a higher demand expectation for 14:20.
- The cycle may repeat.
- In 14:20 we presume actual demand was lower than the expectation and this contributed to set a lower forecast for the 14:25 interval.
- In 14:25, a resumption in consumption is presumed, leading to higher-than-expected demand, leading to a high 14:30 expectation.
Estimates of actual demand at interval-end
Just to indicate the actual vs expected demand pattern described above I’ve marked up an estimate of where actual end-of-interval demand might have been.
The method for this estimate was to select the level that the forecast would have linearly come back to. The forecasts relating to the first 2 blue dots aren’t shown but the third dot at 14:10 is intended to align with the reverse trajectory of the 14:15 expectation and it’s related forecast starting 14:20.
It appears something in the demand forecast was missing to sufficiently capture these fluctuations.
The ‘errors’ are in the order of 500 MW – enough to drive prices from the low to very high, given the tight supply-side bid stack discussed earlier.
A correlation with price*
We noted the price and demand gyrations in the seesaw volatility article.
Let’s propose three price regimes for understanding how the demand ‘errors’ might be price-sensitive:
- When the price spikes high, from a low-priced level, demand may respond. End-of interval demand could be lower than expected in avoidance of high costs or in response to similar incentives.
- The same in reverse – a price sinking low may initiate a response in demand to resume normal levels. Relative to a ‘rebased’ low demand expectation, however, the actual demand would appear high.
- The * in this section’s heading is to highlight this third category. When the price remains at either a high or low level, all else being equal, demand side responses might continue. A rebased demand expectation may appear accurate, and no demand fluctuation would emerge – steady demand.
Here’s a table of how those categories lined up for selected intervals.
| Interval | Price position | Estimated actual NSW demand relative to expected |
| 14:20 | Price spiked high to $23,107 |
Lower |
| 14:25 | Price sunk low to $673 |
Higher |
| 14:30 | Price spiked high to $23,107 |
Lower |
| 14:35 to 15:00 |
Prices remained high >= $9,999 |
Steady |
| 15:05 | Price sunk low to $611 |
Higher |
| 15:10 | Price spiked high to $9,999 |
Lower |
| 15:15 | Price sunk low to $608 |
Higher |
| 15:20 | Price spiked high to $9,889 |
Lower |
| 15:25 | Price sunk low to $164 |
Higher |
| 15:30 | Price remained low $304 |
Steady |
| 15:35 | Price spiked high to $9,999 |
Lower |
| 15:40 | Price sunk low to $164 |
Higher |
| 15:45 | Price remained low $629 |
Steady |
In other words, the demand forecast ‘errors’ appear neatly aligned with price fluctuations.
In some ways it was lucky that the price fluctuations were so large – to some extent this has surely helped the distinction.
I suppose that somewhere in the mid-range, between about $700 and $9,000, one or more price-response thresholds exist. Going from $600 to $9,999 trips all of them.
Further, the responses fluctuated at 5-minute granularity. That pattern doesn’t pair neatly with typical large loads stopping production for extended periods at a time. The changes indicate the responses were very flexible.
Other drivers of demand fluctuations
Other drivers of demand fluctuations include:
- Weather – temperatures were high but reasonably steady, predictable. Read more about this temperatures and demand in Yes, it was hot in NSW on Monday 12th October 2026 – demand was up, but it was not particularly notable.
- Cloud impacting rooftop PV – clouds were developing over Sydney around 14:30 market time. These would have led to reduced rooftop PV and increased grid demand. However, the development was steady and progressive that afternoon – difficult to attribute to demand fluctuations. Rooftop PV is also reviewed in A 1st pass through, inspecting Supply-Side contributions in NSW on Monday 12th October 2026 where the same view is expressed.
Here is the satellite image of cloud cover that was progressing in a north easterly direction over Sydney and greater NSW. The image time was 14:30 NEM time (local time is displayed on the image).
Changes in demand from indicators
The best indicator of whether supply is matching demand is system frequency.
One caveat for this section is that frequency reflects the system-wide behaviour whereas we are seeking to inspect NSW demand imbalances only. There’ll be inherent uncertainty in what we uncover, due to that.
If frequency moves upward, say, it indicates that there is excess supply. That can be because supply has increased, or because demand has decreased, or a combination of the two.
This example is from the 14:20 interval:
When target demand is accurate and units adhere to their targets supply should be in balance and frequency should be nicely near 50 Hz.
In the 14:20 interval, the mainland fleet responded by reducing supply (in blue), relative to the aggregate dispatch trajectory (red):
This 14:20 interval was selected because the NSW price spiked high.
If the story we are converging upon holds, we can associate the frequency rise with a reduction in demand (relative to expected), rather than an oversupply of generation.
An oversupply of generation to drive high frequency should appear as the blue line travelling above the red dispatch trajectory. That didn’t happen in this interval.
The story lines up
Collating the price-spiked and price-sunk intervals into a larger chart we see the pattern holds.
That pattern:
- When the price spiked high, demand reduced (from expected) as indicated by an increase in frequency. Supply side units responded to output below target.
- When the price sunk low, demand increased (from expected) as indicated by a decline in frequency. Supply side units responded to output above target.
Using the table from earlier:
I’m concluding here that demand-side responses did occur in the face of large NSW price fluctuations.
By interval-end the difference in supply, from target, tended to be between 250 to 500 MW. That’s not far off our estimates of demand forecast ‘error’ from earlier.
Timings suggest prices received by responders 1-2 minutes in
Mainly from the frequency, but supported by the supply deviations from target, we observe in most instances the change in frequency (indicating demand change) happened about 1 – 2 minutes into the interval.
UPRR – Batteries mainly
More formally we might say “Unscheduled Price Responsive Resources” (UPRR). But it is home battery ownership that has changed most significantly in recent years. And these are resources that can can respond incredibly flexibly to prices changing every 5 minutes.
The events of the 12th suggest the impacts of UPRR are already upon us.
In this article we’ve just attributed material impacts in the market to these resources, in response to prices.
It is therefore interesting to revisit the recent report from Operational Forecasting on just those things!
The report identifies barriers to forecasting price responsive load. It raises forecasting improvement opportunities to understand and address how UPRR is accounted for.
The report mentions recommendations that include a data sampling approach to:
“… enable AEMO to better understand the impacts from batteries and subsequent price responsive behaviours to better inform impacts on operational demand forecast performance or even inform operational forecasting models.”



















Be the first to comment on "Yes – Unscheduled Price Responsive Resources did deliver price/demand oscillations in NSW on Monday 12th October 2026"