Renewable Energy Financial Model Mistakes to Avoid

Most renewable project models do not fail because the modeller cannot use Excel. They fail because an assumption was inherited from a previous project, dropped into the wrong row, and never questioned again. By the time a lender's adviser opens the file, the error is three versions old, and the investment committee paper has been written around it.

The same mistakes recur across onshore wind, offshore wind, solar and waste to energy. Below are the ones that change the answer, each with its commercial consequence, because an error that shifts equity IRR slightly is a tidy-up job and one that shifts the debt quantum is a repricing event.

Key takeaways

• Sizing debt on a P50 yield instead of the lender's downside case is the most expensive yield error in project finance.

• Exceedance probabilities are period-specific. A one-year P90 and a ten-year P90 are different numbers.

• One deterministic merchant curve with no capture rate decline is not a forecast. Cannibalisation is what lenders price hardest.

• Mixing nominal and real in a single calculation is the quietest source of wrong IRR, NPV and LCOE.

• Curtailment and constraint are different events with different compensation treatment.

• Integrity failures lose lender confidence faster than an aggressive assumption.

What makes a renewable energy financial model bankable?

A bankable model is one a lender can test without rebuilding it: transparent logic, a single dated timeline, sourced assumptions, downside cases that run without breaking, debt mechanics matching the term sheet clause by clause, and outputs (DSCR, LLCR, PLCR, project and equity IRR) that reconcile to the cash flow waterfall every time.

Bankability is a property of the structure, not the result. A model showing a strong equity IRR but with circular references resolved by iterative calculation and no switch will be treated as unreliable whatever it reports. The practical test: give it with the assumption book to someone who has never seen it and ask for the lender case plus two downsides. Our guide to how to build a bankable onshore wind financial model sets out that architecture.

What are the most common errors in renewable project models?

They cluster in six places: energy yield conventions, degradation and availability, merchant price assumptions, debt sizing and sculpting, inflation and indexation consistency, and tax. A seventh, Excel structure, is not an assumption error at all, yet it causes more failed model reviews than any single input does.

Error	Commercial Consequence	Fix
Debt sized on P50 net yield	DSCR is breached in a normal low-resource year, triggering a lock-up or default.	Size debt using the lender case and report P50 separately for equity analysis.
Gross yield used where net yield is required; wake or electrical losses omitted	Net capacity factor is overstated, making all downstream metrics inaccurate.	Build a gross-to-net yield waterfall and reconcile it with the energy assessment.
Degradation applied to prior-year output instead of the Year 1 base	Late-life revenue and the debt tail are misstated.	Align the degradation base with the warranty wording.
No inverter or major component replacement included	Cash available for debt service is overstated in the replacement year.	Model lifecycle capital expenditure (capex) as dated events.
Single deterministic merchant price curve	The lender may impose its own, more conservative price haircut.	Model three price curves and include a capture-rate sensitivity analysis.
Baseload price used instead of captured price	Revenue is overstated, with the overstatement potentially worsening over time.	Apply a capture rate that declines as market penetration increases.
Debt sculpting based on the wrong cash flow line	Reported DSCR does not match the ratio calculated by the lender.	Sculpt debt repayments using Cash Flow Available for Debt Service (CFADS), as defined in the facility agreement.
IDC circularity without a control switch; DSRA funded without a defined source	The model cannot be reliably audited, and the funding requirement is understated.	Implement controlled circularity resolution and fund the Debt Service Reserve Account (DSRA) in accordance with the term sheet.
Operating expenditure (opex) indexed, but revenue is not, or vice versa	The model shows a margin trend that is inconsistent with the underlying contracts.	Apply the appropriate contractual escalation mechanism to each revenue and cost line.
Capital allowances based on the depreciation schedule	The tax shield is recognised too early, overstating early-period cash flow.	Schedule capital allowances based on the applicable in-use date.

LCOE calculation mistakes

Three errors dominate. Failing to discount the energy denominator at the same rate as the costs, so the result is not levelised at all. Mixing a real discount rate with nominal costs. And leaving decommissioning, grid connection charges or lifecycle capex out of the cost stack. LCOE screens options. It does not price a bid.

How should P50, P75 and P90 energy yields be handled?

As three distinct inputs with three distinct purposes. P50 is the central case and drives equity expectations. P90 and P99, over the period the lender specifies, drive debt sizing and covenant testing. P75 often appears in sponsor base cases and sale processes. Never substitute one for another to improve an output.

The subtler error is period. A P90 derived for one operating year captures inter-annual variability in full and is harsher than a P90 across ten or twenty years, where that variability partly averages out. Use the ten-year figure in an annual DSCR test and you overstate the cover. Perimeter is the other trap: assessments give gross production, then a loss waterfall covering wake, electrical and transformer losses, availability, icing and environmental allowances. Models pick up the wrong row, or deduct an allowance the assessment already took.

What goes wrong with degradation and availability assumptions?

Two things. Degradation gets applied to the wrong base, so a stated annual rate compounds differently from what the warranty describes. And availability is deducted twice, once inside the yield assessment's net figure and again as a model-level haircut, removing revenue that was never there.

Solar warranties generally describe a first-year performance step followed by an annual decline from the initial rating. Applying that rate to the previous year's already degraded output gives a different late-life profile from applying it to the year one base, and across thirty years the divergence moves the debt tail and the terminal value. Inverter replacement is the related omission: flat opex with no dated replacement capex overstates cash available for debt service in the years when the DSCR profile is tightest. The wind equivalents are major component exchanges and the service price step-up when the initial contract ends.

How do people get merchant price curves wrong?

The classic failure is one deterministic curve from one adviser, pasted in as values, with no low case and no capture rate adjustment. It treats a twenty-year forecast as fact, and lenders respond by imposing their own haircut, usually harsher than the sponsor would have chosen.

The technical error is confusing baseload price with the price the asset captures. Wind generates when it is windy, and in a high penetration market those hours are lower priced. Solar does the same at midday. The capture rate therefore sits below one and declines as penetration rises, which is cannibalisation. A flat capture rate across thirty years assumes the generation mix stops changing, hard to defend where EirGrid keeps publishing rising renewable penetration scenarios under I-SEM. Check as well whether the curve is nominal or real and whether it already includes capacity market revenue or route to market fees. Price and volume risk interact, which is why we treat them together in our review of key risks in renewable energy projects.

What are the common debt sizing and DSCR mistakes?

Sculpting against the wrong cash flow line, mismatching debt tenor to contracted revenue tenor, leaving interest during construction in an unresolved circularity, failing to fund the debt service reserve account, and omitting the cash sweep. Each changes either debt quantum or covenant headroom, and all five turn up regularly in first drafts.

Sculpting only works if the cash flow sculpted against matches the facility agreement's definition of cash available for debt service: after opex, after tax, after reserve movements. Get that wrong and the reported DSCR is not the one the lender will calculate. The circularity runs debt quantum to cash flow to interest and back, so resolve it deliberately with a switch rather than by enabling iterative calculation and hoping. The DSRA needs an identified funding source, and the choice between equity at close, a letter of credit or a cash build moves the IRR. Where debt runs past the contracted revenue period, show the tail DSCR on a stress curve. Keeping the ratios lenders actually track across a renewable portfolio beside the model while building it avoids that mismatch.

Why do inflation and indexation assumptions cause problems?

Because nominal and real get mixed inside one calculation. A real cash flow discounted at a nominal WACC, or a nominal merchant curve escalated again by CPI, gives an answer that looks plausible and is wrong by the full inflation differential compounded over the asset life.

The discipline is one convention per block, stated in the row label. If the model runs nominal, everything runs nominal, including the discount rate and the terminal value. Indexation is a separate trap, because contracts do not share an index or a frequency: a service agreement may escalate on a published index with a cap, a land lease at fixed review dates, and support scheme revenue may be only partly indexed. One global CPI cell driving everything invents a margin trend found in none of the documents.

What tax and capital allowance errors appear most often?

Timing leads. Capital allowances are claimed by reference to the tax rules and when assets are brought into use, not when depreciation starts or when the final capex instalment is paid. Running allowances off the depreciation schedule pulls the tax shield forward and overstates early post-tax cash, which is the period that sets debt sizing.

Ireland charges corporation tax at 12.5% on trading income and 25% on non-trading income and income from an excepted trade, per Revenue. Models sometimes apply the trading rate to everything in the SPV, or use a blended rate with no basis. Groups within scope of Pillar Two face a 15% minimum effective rate where consolidated revenue reaches the EUR 750 million threshold. Loss carry-forward is often modelled as an unrestricted offset with no pool tracking, and group relief is sometimes assumed when nothing in the structure provides it. Reconcile the computation to a written tax memo and put that memo in the data room.

How should curtailment and constraint be modelled?

As two separate lines, because they are two different events. Curtailment is a reduction instructed for system-wide reasons. Constraint is a reduction caused by a local or regional network limitation. The distinction matters commercially because entitlement to compensation for dispatch down turns on the category, on firm access status and on the market rules in force.

The usual shortcut is one flat percentage haircut labelled "curtailment", held constant for thirty years. That hides the trajectory of dispatch down as penetration grows, the split between compensated and uncompensated volume, and the interaction with the revenue contract. Build it instead as expected dispatch down volume split by category, the compensated portion, the compensation price and a volume sensitivity, sourced to system operator publications and the project's connection agreement. Rules change, so watch the process run by the Commission for Regulation of Utilities.

What are the classic Excel structure and integrity failures?

Hardcoded numbers inside formulas, inconsistent formulas along a row, no error check block, no balance sheet check, circularity resolved by iterative calculation with no switch, timing flags typed in by hand, no scenario manager, and undocumented assumptions. None of these changes an input, and together they are the fastest way to lose a reviewer.

Row consistency is the first fix: a row where every cell holds the same formula audits in minutes, while one patched in year 7 cannot be trusted anywhere. Error checks should be live and aggregated to a single flag on the cover sheet, covering a balance sheet that balances, a sources and uses reconciliation, and full repayment of debt at maturity. Timing flags are the quiet killer: COD, first and last repayment, PPA start and end, support period and tax year boundaries should all derive from one timeline, so moving COD by a quarter moves everything. The FAST Standard is openly published and answers why the model is built the way it is. We cover review technique in the TRP CPD modules.

Model Review Checklist

Area	Check	Pass Test
Timeline	One dated timeline drives all flags and calculations.	Shifting COD by one quarter automatically shifts all dependent dates and calculations.
Yield	Complete gross-to-net yield waterfall is built into the model.	Net yield reconciles with the energy assessment.
Exceedance	Correct P-value, period, and purpose are used.	Debt-sizing and covenant cases are documented separately.
Degradation	Degradation base, convention, and source are explicitly stated.	Degradation methodology matches the warranty wording.
Opex	Lifecycle capex is scheduled as dated events rather than spread across periods.	Replacement events occur in the correct year.
Revenue	Capture rate is applied, and nominal or real pricing is clearly stated.	Captured price is distinguished from baseload price.
Debt	Debt sculpting uses the cash flow line defined in the facility agreement.	Reported DSCR matches a manual recalculation.
Tax	Capital allowances are timed to the in-use date, and tax losses are tracked.	Tax calculations reconcile with the tax memo.
Integrity	No hardcodes are embedded in formulas, and row structures remain consistent.	Row consistency tests pass.
Checks	A live error-check block is displayed on the cover sheet.	A single consolidated status flag is displayed and is currently green.
Scenarios	Lender cases are managed through a central scenario manager.	Every scenario runs successfully without breaking the model.

How do you model a corporate PPA correctly versus RESS?

They are different instruments and should not share a formula. RESS operates as a two-way difference mechanism settled against a reference market price for a defined support period, with scheme rules on indexation, volumes and settlement. A corporate PPA is a bilateral contract whose price, volume obligation, shape risk and credit terms are whatever was negotiated.

For RESS, model the strike price, the reference price, settlement in both directions, the support duration, the treatment of output beyond the supported volume, and the indexation rules for the specific auction round. Those terms have varied between rounds, so read the conditions for the round the project is in. For a corporate PPA the variables that move the answer are: physical or virtual, pay as produced or baseload, who carries shape and imbalance risk, floors and collars, offtaker credit quality, and what happens at expiry. A baseload commitment from an intermittent asset without a sleeve imports real cost, and a ten-year PPA behind an eighteen-year facility needs a tail case. Our guide to corporate PPAs in Ireland covers the structures.

What does a lender or investor actually check first?

The timeline, the yield case used for sizing, the DSCR profile, the sources and uses, and whether the error checks are live and green. Within the first hour they will also run a downside. A model that breaks on the first sensitivity says more about the sponsor's process than any output could.

Attention then moves to provenance: is every material input sourced, dated and traceable to a third party report, a signed contract or a published figure. Inputs without provenance get replaced by the lender's own, which are rarely kinder. What impresses me is not optimism. It is a model that shows its downside plainly, names the two or three assumptions the answer turns on, and lets a third party reach the same number independently. That is the standard behind our financial modelling and transaction support services.

Frequently asked questions

Should I size debt on P50 or P90?

Neither by default. Size on the case the lender specifies, typically a downside exceedance case tested against minimum and average DSCR covenants, and report P50 separately for equity. Sizing on P50 gives a debt quantum the project cannot service in a normal low resource year.

Is a one-year P90 the same as a ten-year P90?

No. Exceedance figures are defined over a stated period, and inter-annual variability partly averages out across longer periods, so a ten-year P90 is less severe than a one-year P90. Applying the ten-year figure to an annual covenant test overstates cover.

What is cannibalisation and why does it matter?

It is the decline in the price a renewable asset captures as more of the same technology generates in the same hours. Wind output peaks together and depresses prices in those hours. A flat capture rate overstates revenue in later years, which is where the merchant tail and refinancing assumptions sit.

What is the difference between curtailment and constraint?

Curtailment is output reduced for system-wide reasons. Constraint is output reduced because of a local or regional network limitation. System operators record them separately, and compensation treatment differs by category, firm access status and prevailing market rules. Model them as two lines with their own assumptions.

What corporation tax rate applies to an Irish renewable SPV?

Ireland charges 12.5% on trading income and 25% on non-trading income and income from an excepted trade. Groups in scope of Pillar Two face a 15% minimum effective rate depending on consolidated group revenue. The treatment for a specific SPV should come from a tax adviser's memo, not a model assumption.

How do I handle circularity for interest during construction?

Deliberately. Either solve algebraically or use a controlled macro that breaks and restores the loop, with a visible switch. Leaving iterative calculation on with no control means the model can converge to a different answer run to run, or fail to converge, when sensitivities are applied.

What does a model audit cover?

Logic integrity, arithmetic accuracy, consistency with the contracts and term sheet, tax and accounting treatment, and behaviour under sensitivities. It does not opine on whether the assumptions are commercially reasonable unless that is specifically in scope, which is a separate commercial review.

Can I use LCOE to set a PPA price or an auction bid?

No. LCOE levelises discounted lifetime costs over discounted lifetime energy. It says nothing about financing structure, risk allocation, tax position or required equity return, and it moves sharply with the discount rate chosen. Price a bid from the full project finance model against a target equity IRR.

What single change improves most models fastest?

A live error check block aggregated to one flag on the cover sheet, paired with a row consistency test. Neither changes an input. Together they catch most mechanical errors before a reviewer does, and they signal that the model is maintained rather than assembled.

Before a lender sees it

The models we are asked to review are rarely bad. They are usually good models carrying one or two inherited assumptions nobody has owned since the first version, and finding that at the diligence stage costs weeks and sometimes a repricing. If you want independent eyes on yours first, get in touch. Further reading: SEAI for national energy data, and EirGrid on dispatch down.