Comment: Can long-run equilibrium electricity market models be useful in energy markets that are rarely, if ever, in equilibrium?

By Hugo Batten, Managing Director for APAC, & James Ha, Research Lead for APAC

As electricity markets become more volatile (or are at least perceived as being more volatile), sceptics rightly ask whether models that focus on “equilibrium” outcomes are the right tools to forecast electricity systems. Or to put it another way, our subscribers and clients often ask, “Why are price forecasts a long, smooth line when historical prices have been so up-and-down?”

We will try to answer that perfectly valid question in a few ways:

A) Mapping out the purpose of long-run equilibrium models
B) Disaggregating the type of volatility or shocks over the forecast horizon in electricity systems
C) Designing scenarios that better reflect the reality of an occasionally “messy transition”
D) Quantifying the upside, particularly for flexible assets

As with our last piece on some of the challenges of forecasting battery economics, we have this discussion conversational, simple, and pragmatic. As an example, Prof Sir Dieter Helm (one of Aurora’s founders) has a good new thought-piece in which he questions the whole idea of neoclassical equilibriums in a range of areas (stocks, house prices, and commodities). This line of reasoning is a little too high-brow for this blog, although we accept the underlying challenge that “equilibriums” in complex systems are generally an over-simplification.

Mapping out the purpose of long-run equilibrium electricity market models

Companies like Aurora typically use what are called “capacity expansion & dispatch” electricity models to create long-term scenarios. These models forecast both “capacity expansion” (determining where and when new assets will be built based on which technologies are NPV positive in different regions, yielding a future capacity mix) and “dispatch” (when assets operate given asset-level constraints and power market demand across time, yielding a price forecast).

There is an iterative loop in most of these models between capacity expansion and dispatch leading to a consistent and system cost-minimising forecast for a given electricity market. Put simply, the model finds an ‘equilibrium’, where just enough future investments are made to avoid flooding the market (and making those very investments uneconomic), yet there is enough capacity build to avoid assets making super-profits. A key assumption here is foresight—investors are assumed to know what other plants are being built or retired, allowing them to plan their investments to avoid over/under-supplying the market.

Typically, over the long-run, these models will produce price outcomes that approximately reflect the cost of new entrant technologies—e.g., the time-weighted average wholesale price will typically reflect the new entrant cost of firmed renewables. Our models go beyond that to account for additional drivers of cost (e.g. additional network costs) or reduced revenue (e.g. optimal location saturation, grid losses or curtailment, negative prices), but the general idea still stands. This is broadly what we mean by “equilibrium outcomes”.

What inputs shift long-term outcomes in these types of models? Some fairly obvious examples:

• Long-term commodity prices
• CAPEX and OPEX for new entrant technologies
• Availability of spare network capacity which can materially slow down new entrant renewable capacity
• Long-term WACCs (weighted average cost of capital) for various types of technologies
• Timings of exit of sizeable thermal assets

Typically when we are asked to run electricity market scenarios, clients will want to test the impact of changing these (and many other) variables over the forecast-time horizon. Increasingly, our clients are using Origin to undertake this type of analysis themselves.

Our capacity expansion and dispatch models, underpinned by assumptions of “perfect foresight”, will produce a new price outcome, but it will typically be another smooth line, albeit one that is higher or lower than the original counterfactual. These types of scenarios do not solve the original issue—how do we build real-world volatility into equilibrium models?

Disaggregating the type of volatility or shocks over the forecast horizon in electricity systems

To better design scenarios that reflect the reality of the occasionally “messy” transition we are experiencing, it is worth disaggregating some types of volatility or system shocks. We would argue that, at a very broad level, there are three types:

1) Short-term drivers of volatility

• Many of these drivers have always been part of electricity systems; some are probably becoming more acute as we undergo the energy transition.
• Under our taxonomy, some examples would include the following:
  • Geo-political conflict creating much higher commodity prices
  • Variations in weather conditions creating differences in supply (e.g., a low wind year, coal mines flooding) and demand (e.g., additional cooling demand)
  • Network constraints or issues that impede economic power flows between regions
  • Unexpected plant outages

2) Features of the transition through to approximately 2030–2040 (depending on the market)

• As we turnover large swathes of generation and upgrade transmission, it is unlikely (and probably impossible) that this will all go perfectly to plan, and there will be consequences from either imperfect market responses or central government planning.
• Some examples would include the below:
  • Delays to new transmission that subsequently delays new build renewables
  • Exit of thermal plants without full replacement of adequate dispatchable capacity

3) Persistent elements of renewables-intensive, zero-carbon electricity systems

• Once the transition is completed and the system is either at or very close to zero-carbon, there are some elements of zero-carbon energy systems (at least as we currently envisage them) that may make them more prone to persistent price volatility.
• Some examples would include the following:
  • Demand more influenced by climate change and abnormal weather (e.g., more heat waves, operational demand strongly influenced by rooftop solar output)
  • Structural oversupply of renewables (especially in shoulder seasons) increasing hard-to-forecast zero or negative prices
  • Complexity around opportunity cost bidding for storage as thermal plants exit and are less frequently the marginal price-setter
  • Expensive residual firming given low capacity factors (competing with storage) and potentially very high costs of gas with CCS, hydrogen, biomethane, etc.
• However, at this stage, this is more speculation and there are countervailing factors—for example, a zero-carbon power system is likely to be largely decoupled from volatile commodity markets and that may reduce volatility over time.

Each of these types of volatility or shocks, if unforeseen, can materially add to price volatility, particularly in markets like Australia and ERCOT with relatively high wholesale price caps and where wholesale market volatility is intended to be a key driver of new investment. However is less true than it was of both markets a few years ago as the Australian Government is offering contracts to underwrite revenue for zero-carbon plants through its Capacity Investment Scheme, while ERCOT actually lowered its market price cap following Winter Storm Uri.

Designing scenarios that better reflect reality of an occasionally ‘messy transition’

So how do we build real-world volatility into long-run equilibrium modelling and scenarios? Broadly, there are three ways we try to do this in the Aurora APAC team (some of which has been replicated in other markets Aurora operates in):

Approach 1: Build event-based volatility into price forecasts

• Individual events can generate material price volatility, particularly in markets like Australia with high wholesale price caps and long, thin grids. There have been a range of them in Australia recently—coal plant explosions and interconnector outages being two key examples. In these instances, price volatility is often 10–20x higher than the average intraday volatility. In equilibrium models, we don’t tend to assume that these events will occur in future, as it’s virtually impossible to know when they will occur, but clients quite rightly want to see the impact of minor/major events on flexible asset economics if they were to occur every X or Y years. To address this, we have analysed many historical events to determine what revenue a flexible asset might achieve if similar price outcomes were to be observed again in the future. However, we break out this upside as a separate revenue stream so, for example, clients can discount it more heavily.

Approach 2: Update input assumptions to reflect reality of volatility of inputs

• As an example, we have often run scenarios where we allow the gas price to oscillate around our long-term gas price forecasts based on historical distributions of gas prices around long-term averages. In these long-term scenarios, we are not trying to suggest that the gas price will actually be higher or lower than an equilibrium price in any given year, but that over a 20 or 30 year time horizon we can expect some 1–2 year periods where the gas price will be materially above or below equilibrium levels. When we introduce a more volatile gas price, we then see a wholesale electricity price that oscillates as well, although the exact correlation will depend on the generation mix, etc.

Approach 3: Constrain capacity expansion & dispatch model such that it cannot respond with perfect foresight

• We have the flexibility within our capacity expansion & dispatch models to effectively disable the “perfect foresight” function—for example, we can model scenarios where coal plants close while delaying the build of adequate replacement bulk generation and replacement peak capacity. This reflects real-world challenges for investors in timing new builds and acts to create higher prices after the exit of major coal assets.
• A similar exercise could be conducted with delays to major transmission projects, or deliberately overbuilding technology types, to reflect the pro-cyclical nature of investment cycles in electricity markets.
• As an example of policymakers utilising this approach, the NSW Electricity Infrastructure Roadmap very explicitly examined a counter-factual where coal plants exited without adequate replacement capacity. Our team at Aurora undertook the initial modelling for this policy suite.

We run a regular scenario for subscribers in Australia, Japan, and the Philippines (amongst other markets) that brings many of these elements together—we call it “Messy Transition”. It features higher and/or more volatile commodity prices, unexpected exits of coal assets, and delays to new transmission and generation or storage assets. It has become a popular reference case, particularly for flexible asset developers and investors, as well as policymakers. In other markets, EMEA for example, we have explored similar themes via weather or demand year shifts, or commodity price sensitivities.

Quantifying the upside, particularly for flexible assets

We either use Approach 1 in its standard scenarios, or  we use the “Messy Transition” Scenario—a combination of Approaches 2 and 3—to model upsides for flexible assets. A stylised example is mapped out below for a 2-hr Victorian battery with a 2024 entry year. Both a NEM Central Scenario with major/minor system events included every 4–7 years and a NEM Messy Transition Scenario with the same major/minor events included represent a significant upside for battery economics.

As a final point, all of the above is not to say that equilibrium modelling cannot be useful. Our “Messy Transition” Scenarios aim to show what highly volatile future years might look like, but it is very difficult to predict exactly when such years might occur. Messy Transition scenarios try to illustrate the shape of prices over the next 20 years—an acknowledgement that the future probably won’t be as smooth as our equilibrium scenarios suggest—but over the longer term, price outcomes will likely oscillate around equilibrium prices due to the cyclical nature of investments, and so the long-term average level of prices in our equilibrium scenarios still hold.