How to build a merit order using live fuel and carbon price inputs
.webp)
A merit order ranks power plants from cheapest to most expensive based on short-run marginal costs. It's a valuable simplified tool for estimating which technology will set the market price at a given time. However, its accuracy depends on its inputs. Since fuel prices, carbon costs, and plant efficiencies can change quickly within a single trading day, a merit order based on last week's static assumptions may no longer accurately identify the marginal plant. The price levels and plant parameters mentioned below are only illustrative and do not represent market forecasts.
For analysts, traders and portfolio managers, understanding how to build and maintain a live merit order, rather than treating it as a one-off exercise, is central to determining which plants are likely to run.
What a merit order is and why it matters
A merit order ranks generation units by short-run marginal cost, reflecting, in a simplified way, the order in which a system operator would ideally dispatch plants to meet demand at the lowest overall cost. It's used to estimate which technology is likely to be marginal, and therefore likely to set the price, at a given demand level. That's a genuinely useful simplification, but it is still a simplification: the basic version excludes network constraints, must-run requirements and a range of other operational limits that shape real dispatch outcomes.
A wide range of people use the tool for slightly different purposes. Traders and analysts use it to estimate which plants are likely to run and, by extension, where the price is likely to land. Utilities and generation portfolio managers use it to plan dispatch and hedging decisions across their own assets. Modelling and forecasting teams build it into supply-stack tools that must update automatically as conditions change, and investors use it to assess a specific asset's competitive position within the stack. Policy analysts, meanwhile, use it to understand how carbon and fuel price changes are likely to reshape the generation mix over time. The underlying calculation is the same across all of these uses; what differs is how live and how granular the inputs need to be.
Core inputs needed to build the merit order
Building a merit order takes a consistent set of inputs, refreshed for every plant in the stack:
Fuel prices for every relevant technology represented in the stack, typically gas, coal and sometimes oil
The prevailing carbon price from the relevant emissions trading scheme
Plant-specific efficiency or heat rate figures, which convert a fuel price into a cost per unit of output
Plant-specific emissions factors, which convert the carbon price into a cost per unit of output in the same way
Variable operating and maintenance costs where available, a real, if usually smaller, part of a plant's marginal cost
Data quality matters as much as data coverage here. A plant-specific efficiency figure that's several years out of date, for a unit that's since been partially refurbished or is degrading with age, can quietly distort a ranking that otherwise looks well-constructed. The same applies to emissions factors, which can shift with fuel quality or blending. A merit order built from broadly representative, technology-average assumptions is a reasonable starting point, but one built from genuinely plant-specific, regularly reviewed figures is a meaningfully more reliable guide to which individual unit is likely to be marginal.
Calculating short-run marginal cost for each plant
The calculation itself follows a consistent structure across every unit in the model, similar to the spark spread and dark spread logic covered in Clean spark spread versus clean dark spread. Convert the fuel price into a cost per unit of electricity using the plant's efficiency or heat rate; convert the carbon price into a cost per unit of electricity using the plant's emissions factor; add both, plus any variable operating costs, to arrive at the plant's short-run marginal cost. Repeat that calculation across every plant or technology type included in the model, and the result is a comparable cost figure for each one, ready to be ranked.
Assembling and ranking the stack
Once every included unit has a short-run marginal cost, the next step is straightforward in principle: sort them from lowest to highest, and assign each one its available capacity to build a cumulative supply curve. Overlaying expected demand against that curve gives an estimate of which unit is likely to be marginal, and therefore likely to be setting the price, at that demand level.
This is also the point where the model needs to stay live rather than static. As fuel and carbon prices move, the relative ranking between technologies, particularly gas and coal, can shift meaningfully. A gas plant that sits comfortably below a coal plant in the stack under one set of price assumptions can end up above it if gas prices rise or carbon prices climb enough to erode coal's remaining cost advantage. Rankings that were accurate yesterday aren't guaranteed to hold today.
It's worth building the stack at more than one demand level rather than a single point estimate, since the marginal unit at low demand often differs significantly from the marginal unit at peak demand. A stack that only tells you which plant is marginal at today's forecast demand gives comparatively little insight into how price sensitivity would change if that forecast moved, whereas the same stack read across a demand range shows where the steepest, most price-sensitive parts of the curve actually sit.
Keeping the merit order live and accurate
How often "regularly" means depends on the purpose of the merit order. A desk using it to inform same-day trading decisions needs inputs refreshed almost continuously, since even a few percent move in gas prices can be enough to reorder plants that sit close together in the stack. A team using it for longer-term planning or scenario work can reasonably work from inputs refreshed daily or weekly, provided that the cadence is explicit rather than assumed, so anyone reading the output knows how current the underlying prices actually are.
Because fuel and carbon prices can move significantly within a single trading day, regularly refreshing those inputs is what keeps a merit order useful rather than merely illustrative. Renewable generation, which typically has close to zero marginal cost, needs to be accounted for separately, sitting at the bottom of the stack whenever it's available and effectively displacing higher-cost thermal capacity from the margin. Planned and unplanned outages also need to be reflected as capacity removed from the stack, since a plant that's offline isn't part of the available supply curve, regardless of where its cost would otherwise rank it.
It's important to clarify what a simplified merit order still misses, even when inputs are active: network constraints, ramping limits, and must-run obligations influence actual dispatch results in ways that pure cost ranking doesn't capture. A plant with a higher marginal cost than the theoretical margin can still operate because it supports local grid stability, while a cheaper unit may remain idle due to a transmission constraint not represented in the merit order.
There's a useful discipline in treating the basic merit order and its real-world constraints as two separate layers rather than forcing them into a single number. The cost-ranked stack answers "which plant would run if only cost mattered"; the constraints layered on top answer "what actually prevents that from being true in practice." Kept separate, each layer remains interpretable on its own terms, and it becomes much easier to spot when an observed dispatch outcome has diverged from the cost ranking because of a genuine constraint, rather than because the underlying fuel and carbon inputs have simply gone stale.
Conclusion
A live merit order depends on continuously updated fuel, carbon and plant-specific inputs rather than assumptions carried over from an earlier period. The relative rankings it produces, particularly between gas and coal, can shift meaningfully as those inputs change. It remains a simplified planning tool even at its most accurate: real dispatch outcomes depend on network constraints, ramping limits and must-run obligations that sit outside the basic calculation. Used with that caveat in mind and kept genuinely current, it's still one of the more useful shorthand ways to gauge which technology is likely to set the price next.



.jpeg)




.webp)