Every forecast you've ever bought is wrong

Every forecast is wrong, ours included. The difference is that we tell you the range we are wrong within, rather than a single number that pretends otherwise.

TLDR

• VarialView is our own price view for the NEM out to 2040, our assumptions, not the ISP's.
• It is a fan of five scenarios, each on five weather years, not a single number.
• Every scenario runs on one consistent base, so the price gaps you see come from the assumptions we changed, not from a different starting point.
• Buy it and the whole fan becomes a basis you can build on in Prices and value across in Valuations, weighted to your own view.

Why we built this, and who it's for

If you are carrying price risk in the NEM, whether you are financing a project, running a retail book, or trading a position, you need a price curve to value against.

VarialView is our answer to that. It is a standing house view of NEM prices. Central is the POE50, the middle of the fan; the fan around it is a plausibly wide risk range. It is aimed at the people who live with that risk day to day: developers and financiers, large buyers and retailers. Each scenario and weather sensitivity is a full fundamental forecast - the 30-minute data underlying each is available to purchasers for further exploration.

Modelling principles

The Fan

VarialView is an adjusted view on AEMO's ISP Step Change scenario - the pathway much of the market treats as the most believable of the ISP set. We take Step Change as the base, then adjust the key input assumptions on top of it - demand, build pace, delivery, fuel and carbon - so the only thing separating our five scenarios is the set of levers we deliberately changed, not a different starting point. What comes out is a fan: a range of prices, not a single confident-looking line. The only thing we know for certain about a point forecast is that it is wrong, so we publish the width instead: wide enough to hold whatever actually turns up, but still tight enough to price a deal against. It is firm in the near term, where the next few years are largely locked in, and spreads wider every year out as uncertainty compounds.

The percentiles on each scenario describe the fan once it has opened. In the first year or two the scenarios sit close together - the near term is largely locked in - so the tails run nearer the middle than their labels imply. From about 2030 the fan is wide enough that the labels hold.

The fan is useful because it is consistent and resilient, not a point forecast that resets every quarter. Most forecasters throw out last quarter's number and hand you a new one - already a quarter old, its assumptions locked before publication - and leave you guessing what changed and re-basing everything you built on the old one. We do not work that way. The fan is what you build on: unless the market moves materially, the five scenarios stay put. What moves is your conviction, not our curve.

So instead of taking step changes every quarter, you slide your own weighting along a fan that holds still - up toward High and Highest on bullish news, back toward Low and Lowest on bearish, the moment it breaks rather than a quarter behind. Your expected value, the probability-weighted blend, walks up and down inside the fan while the fan itself stays where it is. We re-cut the fan only when the market shifts materially, and at least once a year so it never goes stale - and when it moves, you will know exactly what changed and why. A full methodology write-up is coming separately; the short version is an ISP-like engine with REZ and subregional flow modelling, wrapped around a bespoke, future-proof bidding model with no hardcoded assumptions.

A fair question to ask any fan is whether it would have held. Here is the test we can actually run. Draw a fan at the end of 2020 - widening with horizon, the way ours does, to the dispersion this fan actually carries five years out - and walk New South Wales through it: $73, $183, $96, $131, $103.

Three of those five sit inside the inner band, one sits on the floor, and 2022 goes past the top. That distribution is the point. A walk tests an edge occasionally and spends most of its time near the middle, which is why an envelope survives five years and a point forecast does not.

We have left 2022 outside rather than widening until it fits. It was a fuel crisis, not a market state, and a fan redrawn until it swallows the worst year on record is a fan fitted to the answer. Highest is the shape that comes nearest, because Highest is where every input pushes the same way.

The left half is illustrative and the chart says so: we published no forecast in 2020, and drawing one that quietly contained everything would be worth less than showing you the real thing beside it. So the right half is the fan we publish today, on the same axis, with the ASX base-future strip next to Central - which is what lets you check that the thing being argued for is the thing being sold.

The bidding model

The main part of the Varial Market Model we don't expose for customisation is the bidding model - how every generator in the simulation decides what to offer given the current market conditions. We're looking at ways to provide control of bidding to users, but because getting it right is most of the work and critical to a resilient forecast, we've prioritised getting our house view right first.

Importantly, we aren't committing the model to a bid-stack based on historics. We're in an energy transition, and ultimately that means a transitioning bidstack. A model trained on how the market bid last year has learned a generation stack still predominantly set by coal, and it will keep bidding baseload coal into a grid that no longer has it. And it is not a hardcoded set of bids for a fuel type and time of day, which has its own pitfalls. Ours is a behavioural / game theory model that adapts to the fleet in front of it: as coal leaves, renewables and storage arrive, the bidding logic re-derives what each unit should rationally offer, so it stays coherent in a 2035 grid that looks nothing like today's.

The high-level logic itself is not a secret - you can walk the backbone of it in How the NEM sets price. Run it forward as coal retires and you can see where it goes: a much more binary market. Through the middle of the day, distributed and utility solar crush prices toward the floor, and - contrary to an easy assumption - storage will not lift them (we have a separate piece coming on storage versus solar depth). Then in the evening, as solar rolls off, prices climb to where storage is shadowing gas but protecting the cap. If the storage run out, gas generators have real market power, and that is when you will see the very high prices. Going are the days of coal trips being the big fear - instead it will be storages going empty - and this is a real risk, especially in winter, with low solar output and wind lulls not uncommon.

Continuity is key

A rule we hold ourselves to, and one we would push any forecaster on: if you cannot see the model reproduce the recent past, do not trust what it says about the future. Take VRE dispatched by time of day in NSW: below, the observed history runs straight into the forecast on one axis, and the daily shape carries through the join rather than breaking at it. Before we read a single output price, we run that check across the board - demand, supply, network, commodities assumptions and simulated variables. If a model cannot get recent history roughly right, it has no business telling you what 2035 looks like. That is why observed history sits right next to the forecast on our charts: so you can run the check yourself rather than take our word for it. We believe some forecasters miss this key calibration step - so be careful to check on yours.

Key limitations

A house view is a set of judgements, and we like to explain the limitations:

• Weather, yes; random shocks, no. We run real weather variability across five reference years (2021-2025) in this forecast - demand, wind and solar, and coal availability all move with the weather year - so the funnel carries the year-to-year swing weather drives. But we don't model random shocks. This is intentional - random shocks make asset valuations unfair, since an asset built the quarter before a modelled spike looks better than an identical one built the quarter after, purely on luck.
• No price-responsive build (capacity expansion). In the real world, prices sitting up in the High or Highest range would pull more projects in and temper them; we are not modelling that feedback yet. So read the upper scenarios as what happens if a bullish outcome turns up and build does not chase it - the developer gap and the retailer risk - not a settled long-run equilibrium. Capacity-expansion modelling is on our list once we stabilise our price vs input assumptions debates. We'll cover this in a separate article, but applying capacity expansion to bad inputs and modelled prices is a huge waste of time - something we want to avoid.

Our assumptions

The scenarios

We forecast five plausible scenarios ranging from the lowest expected prices to the highest expected prices. The headline settings, lowest to highest, are below, and each lever is walked through underneath with its input chart. These are inputs, not outputs: the resulting price funnel is in the Prices tab.

Demand

We calibrate on operational demand, the load that actually sets price, and move it with the electrification bundle (EVs and the electrification of transport, industry and buildings) plus an additive data-centre block. The electrification components are tiny today and large by 2040, so the funnel auto-widens over the horizon: the scenarios sit close together in the near term and fan out as the decade runs on. Data centres are a clear upside: the ISP already bakes some in, and we fan around that baseline, from about 1 GW below it in the softest case to roughly +2.5 GW and +6.3 GW above it by 2040 in High and Highest. The native (underlying) demand shape gets a near-term correction across all scenarios, because the ISP steps the midday trough down faster than the actuals support, with a two-sided spread in the out-years. Landing points for operational demand at 2040, relative to the Central case: roughly -15%, -8%, flat, +11% and +31%.

Commissioning

The ISP needs an unprecedented pace of new build held for more than a decade, and we do not assume it all arrives on time. Each scenario carries an annual build-rate ceiling, the most new capacity the industry can commission in a year, and anything over the ceiling carries forward to the next year, subject to that year's cap and the network's hosting limits. The ceiling is front-loaded, because the supply chain we have now is not the one we will have in a decade: it starts near 7.5 GW a year in Lowest and 4.5 GW in Highest, and eases to about 10 GW and 8 GW from the mid-2030s, with Central running 5.5 GW rising to 9 GW. Because commissioning is heavy in 2027 to 2030, the ceiling binds hardest in exactly those years, delaying the VRE and storage roll-out that would otherwise cap prices. On top of the annual ceiling, every committed or anticipated project carries its own delivery delay that grows the further out its commissioning sits, because early-stage projects slip more than near-complete ones, and the chronic over-runners, Snowy 2.0 and Borumba, get a larger delay fan of their own. What we do not do is let price pull extra build on top: that capacity-expansion feedback is deferred (more on that choice in the section above).

Retirements

Coal retirements past 2030 shift earlier in the tighter scenarios and later in the looser ones: a couple of years later in Lowest, our researched announced-reality dates through the middle, and AEMO's earlier Step Change schedule (the earliest exit) in Highest. The near-term closures, Yallourn, Eraring and Gladstone, are locked to their announced dates, so the fan only opens up past 2030. When a unit leaves earlier, its output has to be replaced by the rest of the fleet.

Network

Transmission is the hardest part of the transition to deliver, and the part most likely to slip. We leave the committed links on their dates and push the uncommitted flow-path and REZ (renewable energy zone) projects later in the tighter scenarios, Low six months, Central a year, High eighteen months, Highest two years. When a line slips, the zone behind it is stranded and the generation that depended on it slips too, so congestion and curtailment compound rather than cancel.

Fuel and carbon

The funnel carries higher gas and coal prices in the upper scenarios, with Highest NSW gas reaching roughly $22/GJ by 2040. Coal we keep similar to AEMO Step Change scenario assumptions.

Carbon is a clean add on top of a base that prices it at zero everywhere to start: Lowest, Low and Central stay at zero, while High and Highest fan up to roughly $80 and $130 a tonne by 2040, switching on at different dates. Central carries no carbon price deliberately - we are not willing to put one in the scenario a reader will treat as our base case without a policy to point at. The carbon path references the AER's Valuing Emissions Reduction guidance (May 2024), however, still sits intentionally below this level even in the highest scenario. The market price cap is known through 2028 (the $26,485 cap), then escalates on each scenario's own inflation rate, 1.0% in Lowest up to 3.0% in Highest.

Results

Where prices land

The output is a fan of five price paths for the NEM, region by region, out to 2040. The fan is tight in the near term, where the next few years are largely locked in, and spreads wider every year out as the uncertainty compounds. The headline chart is below; the Prices tab carries it region by region, quarterly and by time of day. As at August 2026, the most relevant ASX prices sat at a premium in the front-end, with 2029-2030 prices at near parity to our Central view, as Yallourn and Eraring closures impact the merchant outlook. That is a point-in-time read against a curve that moves; the structural point - a front-end premium converging by the end of the decade - is what we are pointing at.

The generation mix that clears those prices follows from the assumptions above: a different build and retirement path, slower delivery, and our own fuel and demand settings. You can see the resulting stack in the Supply tab and the load in the Demand tab, scenario by scenario.

Dig into the full results

Everything behind the funnel is in the tabs above, one scenario and reference year at a time: quarterly and time-of-day prices, demand, the generation stack and new build, network flows and commodities. The deeper cuts unlock with the forecast: VWAPs and capture prices by region, MLFs, and curtailment. From there you can trace any scenario from an assumption right through to a captured price.

What's next

The report and the headline charts are open to everyone. The forecast itself, the full basis and the prices behind the funnel, is what you buy, and it is what turns a view into a working tool. Two things come with it.

Unlock it in Prices to build on it. You get the complete input assumptions behind every scenario, wired in as a selectable basis, so you can take our fan and customise it: push a lever we did not, run your own sensitivity, or build a bespoke scenario straight on top of ours in the wizard. Our view becomes your starting point, not your ceiling.

Unlock it in Valuations to value against it. The five price paths become a basis you can value your book or asset across, weighted your way, so you get a probability-weighted expected value and the full distribution around it, the value-at-risk read, in minutes rather than a spreadsheet marathon.

What we're hoping to learn

This is our first house view out in the open, and we really want to learn from how you read it. Three things we would love your honest take on.

• The fan, or points? Does a five-scenario funnel actually help you, or would you rather we published a smaller set of named point scenarios? We lean towards the fan, but we want to know if it lands.
• One basis, or many? We put every scenario on the one Step Change basis, so the price differences you see come from the levers we moved, not from a different starting point. Is that the most useful way to see it, or would you rather each scenario carried its own basis?
• Do the prices ring true? Before we layer in capacity-expansion modelling, letting build respond to price, we want to know whether the price outlooks themselves feel right to the people who trade and build in this market. If a scenario looks off, tell us where and why.

None of these are rhetorical. The next version will move based on what we hear back, so if you have a view, get in touch through our contact page.