By Hugo Batten, Managing Director for APAC, & Raheel Raniga, Research Associate for Australia
We have modelled grid-scale battery economics in Europe, USA, LATAM, and APAC, and across a number of business cases and revenue streams—merchant wholesale and ancillary markets, long-term capacity payments, transmission support, amongst others. Battery investment cases have certainly gotten more complicated over the last 8-9 years. We have learnt lessons from a range of markets: the UK where Enhanced Frequency Response (EFR) markets kicked off the battery boom; CAISO, ERCOT, and Australia where batteries are in some instances leading in terms of additions of MW per annum across technology types (and where saturation issues are now starting to bite); and Japan where battery investments are at an earlier stage but governments are running very specific tenders to rapidly add battery MW capacity.
In addition, it has always been hard to exhaustively assess all the potential upsides and downsides in a battery investment case. In battery modelling, there is a strong risk of “multiplicative complexity” whereby market, asset, financing, dispatch, operational, and revenue stack parameters are individually adjusted to create almost infinite scenarios and very little insight.
After almost a decade of Aurora battery modelling, we are summarising some of our “lessons learnt” and providing a high-level guide meant to ensure battery investment cases are as robust as possible. Although this is not comprehensive, we have mapped out a taxonomy of common pitfalls and simplifications that we have seen and then expanded on each of them.
Before diving into detail, there are broadly two types of markets where we see rapid growth in battery penetration:
1.
Volatile wholesale-only markets with high levels of solar penetration where batteries can arbitrage duck curves and take advantage of event-driven peak volatility (for example, Australia and ERCOT)
2.
Markets with capacity mechanisms with strong decarbonisation levers that then require batteries to replace departing thermal generation (for example, the UK and Japan)
Having said that, as much as market design and other factors vary, there are a lot of common methodological challenges in putting sensible bounds around future battery returns.
Our software solutions, Origin (power market model) and Chronos (battery asset dispatch) address all the issues outlined, as well as provide a higher degree of transparency to clients by enabling them to create market scenarios and configure dispatch algorithms themselves. Fundamentally, we believe integrating bankable, granular price forecasts with battery asset dispatch is essential to ensure consistent revenue forecasts. Our clients stress how helpful it is to have our bankable market scenarios (built with Origin) available for use in Chronos so results can be presented to lenders and investors as an integrated package that reflects the nuances of bidding rules—particularly in ancillary markets.
MARKET MODELLING PITFALLS (ORIGIN)
a.
Ensuring wholesale and balancing/ancillary market outcomes are consistent: As algorithmic dispatch rapidly optimises between available revenue streams (day ahead, real time, balancing, ancillary) and battery penetration increases, intelligent price forecasting models should capture greater correlations between these available markets. This dynamic is already playing out in California and Australia, and is very likely to hit other markets as battery penetration accelerates. Essentially, models should not assume batteries can capture uncorrelated price peaks across markets if those markets are increasingly correlated over time
As an example, raised regulation FCAS prices are typically correlated with wholesale prices, due to the opportunity cost of providing this service, beyond a certain price point
b.
Assuming the future will look like the past or that you can draw a straight-line between them: There are two parts to this, but both mean that past price outcomes (particularly for intraday volatility in wholesale and balancing/ancillary markets) are unlikely to be a good guide to future outcomes.
i.
The electricity generation mix is changing profoundly. Almost all electricity markets are quickly and radically changing their generation mix. This changes both average levels of generation by technology, but in many cases, it will have a greater impact on the marginal technology. Linear extrapolations based on recent history are unlikely to capture these dynamics. This applies even more acutely to ancillary markets which are smaller and more prone to saturation leading to non-linear declines in available revenue.
ii.
The transition is likely to be messy and lumpy. As we turnover almost all the capital stock in electricity markets over the next 20-30 years, there is likely to be a degree of dislocation and subsequently peak price volatility that long-run equilibrium models are not traditionally good at capturing. As a very basic example, as we ask aging coal plants to ramp more around deep duck curves, we are likely to see more unplanned outages, in turn creating more peak price volatility. Failing to factor this type of transition-driven volatility is likely to understate battery returns.
c.
Forecasting negative prices increasingly matter (in some markets): Just as forecasting peak price volatility is crucial for batteries, so is getting bottom prices broadly correct equally important (both the extent of negative prices and the volume). This will vary by market but often negative prices are driven by a series of “out-of-market” factors (e.g., renewable energy certificates, bidding around grid constraints, CFD contract conditions) as well as coal SEL constraints. Modelling this is hard, but vital to battery economics—at the very least, short-term forecast negative prices should align with recent history.
Negative prices have increased over the last 5 years—particularly in SA, QLD and VIC
d.
Getting upside/downside scenarios right to actually test battery returns meaningfully: Designing scenarios for batteries can be more complicated when compared to other technology types.
i.
Downside scenarios. For example, at a very simple level, downside scenarios for solar can be upside scenarios for batteries as they are charging at low (or even negative) middle of the day prices. In addition, stress-testing downside scenarios for balancing and ancillary prices will often involve pulling levers that have less substantial impacts on wholesale prices (e.g., lower long-term ancillary market sizes, short-term battery overbuild). Often we see downsides for baseload prices used by default as battery downsides which in many markets makes almost no sense.
ii.
Upside scenarios. As another example, we are designing scenarios that reflect oscillations between equilibrium and non-equilibrium outcomes in energy markets—for instance, exiting coal plants with delays to replacement firming capacity, deeper fluctuations in commodity prices, or delays to new transmission build. These “out-of-equilibrium” outcomes arguably better reflect the reality of pro-cyclical investment in new generation and storage and the difficulty of correctly managing the transition, and they are typically material upside cases for battery investments.
e.
Capturing specific event-driven volatility: Particularly in wholesale markets with high price caps, long-run general equilibrium models typically do not attempt to capture specific event-driven volatility that might, for example, come from extreme storms and freezing gas infrastructure, coal plant explosions, or interconnector outages. Having said that, particularly in Australia where the price cap is A$16.6k/MWh, these types of events have seen batteries earn 20-30% of their CAPEX in 2-week periods. We do not build these events into base cases, but we do help clients tranche up recent events into tiers and then quantify how often they have happened over the last 5 years to help provide a view on potential upside. This helps to provide a directional sense of what additional revenue may be available from hard-to-predict incidents, particularly in regions with long, thin grids and/or extreme weather conditions.
BATTERY ASSET DISPATCH MODELLING (CHRONOS)
a.
Ensuring dispatch algorithms reflect real-world ability of traders or short-term optimisers to capture value in dynamic markets: We work closely with leading short-term optimisers to ensure Chronos achieves a similar “percentage-of-perfect*” (i.e., actual value capture versus value capture that assumes battery operators can perfectly foresee future prices). In addition, we monitor real-world performance of assets in market and backcast against Chronos performance to test whether our forecasts match reality.
Aurora’s modelled imperfect BESS performance follows closely that of an operational BESS, and is ~15% below an asset with perfect foresight
b.
Trusting black-boxes or limited transparency: Generally, battery investors should request granular wholesale and balancing and ancillary market prices at 30 minutes to run their own stress tests on volatility metrics and correlations between markets. In addition, they should request at least samples of forecast-optimised battery dispatch to make an independent assessment of how dispatch might align to their planned trading strategy and to provide independent verification that the “percentage-of-perfect” metrics make sense.
c.
Accounting for the latest battery warranty and operational constraints: Battery warranties, in particular, have changed dramatically over the last 2-4 years—battery life or cycles under warranty especially. Battery investors should make sure the dispatch engine they are using reflects the detailed warranty conditions (e.g., daily or annual cycle limits, non-linear degradation profiles) as these parameters do matter over a battery’s life.
d.
Simplifying market win-rate and small market saturation: We have reviewed a number of vendor cases where batteries are simplistically dispatched against a forecast price series and it is assumed that these batteries always offer the winning bid in balancing and ancillary markets. The reality is, as ever, more complicated. It is important to factor in evolving levels of competition and win rates, especially as the battery fleet size exceeds the balancing or ancillary markets procurement target. This is particularly salient in markets like the Australian NEM, CAISO, and the UK.
e.
Factoring in operating constraints to reflect contractual terms or offtake: As a diverse range of participants experiment with battery investment cases, offtake requests have become more complex—for example, reserving capacity at particular times of day or year to help support transmission assets, perpetually reserving capacity for provisions of ancillary services or system strength services. These offtake terms also may change over time or be dynamic to reflect a retailers hedging requirements. These need to be built into forecasts in ways that are consistent with merchant dispatch, but they also then require a sufficiently flexible dispatch algorithm.
f.
Failing to factor in grid constraints on dispatch, particularly for co-located assets. The nature and level of grid constraints and losses vary considerably by market, but are an increasing factor for both renewable and storage dispatch. We update Chronos quarterly to ensure it reflects the most recent hybrid asset operational rules provided by any given system operator. In addition, we use grid outcomes from our DC approximation power flow models to co-optimise the hybrid system around expected periods of thermal grid curtailment. In essence, the grid matters and may need to be factored into dispatch decisions.
As grid-scale battery investment cases continue to evolve across markets, it is crucial that modelling approaches keep pace. By avoiding common oversimplifications around market price forecasts and asset dispatch algorithms, investors can build more realistic and robust cases. Though battery economics involve many complex variables, thoughtful scenario design paired with transparent, configurable tools can provide the rigour required.
*Not accounting for non-physical trading
