Base Case, Banking Case, Downside Case: The $53.8m Between Two Spreadsheets
A sponsor's model and a lender's model of the same project rarely disagree about anything dramatic. They disagree about four numbers, each individually arguable, and they arrive at debt capacities $53.8m apart.
Sponsor base case CFADS $30.0m → debt capacity $240,749,224
Banking case CFADS $23.3m → debt capacity $186,988,855
Something else changes between them that neither party usually notices. In the sponsor's case the project's cash yield on cost is 11.05%, which — on the crossover from the first post of Series E — makes it gearing-constrained. In the banking case it is 8.58%, which makes it DSCR-constrained. The two models are not just different in degree; they are in different regimes, with different sensitivities and different negotiations attached.
This final post covers what distinguishes each case, the bridge between them, how downside cases should be built, and why the distinction between a sensitivity and a scenario matters more than either.
ℹ️ Note: The haircuts below are labelled assumptions assembled from the preceding series. Real lender cases vary; the structure of the bridge is what generalises.
What Are the Three Cases?
| Case | Whose | Purpose |
|---|---|---|
| Base case | Sponsor | What we expect to happen |
| Banking case | Lender | What the debt is sized on |
| Downside cases | Lender | What happens if it does not |
They are not three levels of pessimism on one axis. Each answers a different question, is owned by a different party, and changes different inputs.
The Base Case
The sponsor's honest expectation: P50 generation, the consultant's central price forecast, contracted availability, budgeted operating costs.
Its purpose is to value the project and to set the equity return. It is the case an investment committee approves and the case a farm-down is priced against.
It is not conservative and should not be. A base case with conservatism baked in cannot be distinguished from a genuine expectation, and once a model contains hidden prudence nobody can say what the project is actually worth.
The Banking Case, and the Bridge to It
The lender's version, built by applying haircuts to the base case. Every one of them has appeared in this series.
| Revenue | Change | |
|---|---|---|
| Sponsor base case | $33,000,000 | |
| P50 → P90 generation, 15-year exceedance | $30,393,000 | ($2,607,000) |
| Curtailment, node-specific at 4% | $29,177,280 | ($1,215,720) |
| Basis and capture rate | $26,843,098 | ($2,334,182) |
| Availability at 97% against 99% assumed | $26,300,867 | ($542,231) |
| Banking case | $26,300,867 | −20.3% |
Then the operating leverage effect from the merchant tail post, because costs do not fall with revenue:
Base case CFADS $33.0m − $3.0m opex = $30,000,000
Banking case CFADS $26.3m − $3.0m opex = $23,300,867
-----------
Reduction 22.3%
Revenue falls 20.3% and CFADS falls 22.3% — the two-point amplification that follows automatically from a fixed cost base, and which is why every haircut in this series lands harder than it looks.
And the debt capacity that follows, at a 1.30× coverage target over eighteen years:
Base case debt capacity $240,749,224
Banking case debt capacity $186,988,855
-----------
Difference $53,760,369
What Else Is in a Banking Case?
The four revenue haircuts are the visible part. A lender's case adjusts several other things, and the cost side often moves more than people expect.
Operating cost escalation. A sponsor budget frequently escalates opex at a general inflation assumption. A lender will escalate at something higher for components with observed cost pressure — labour, insurance, land rent — and will often add a contingency to the O&M budget on top.
Maintenance capex and major overhauls. Whether these sit above or below the CFADS line, as the waterfall post established, is worth $16m of debt capacity on its own. A lender's case puts committed capital maintenance above the line whatever the accounting treatment.
Tax. The depreciation post's point applies: a project with accelerated depreciation has a cash tax profile that is nothing like a straight line, and a lender sizing on post-tax CFADS needs the actual profile rather than a normalised rate.
Inflation on revenue versus cost. A PPA with a fixed price or a sub-inflation escalator against an opex base escalating at CPI produces a declining real margin. That is a genuine feature of the contract and it only appears in a nominal model, which is why the cost of capital post argued for building nominally.
The merchant tail. Everything the merchant tail post described — the price haircut, the P90 basis, the raised coverage target for uncontracted years — applies here and typically removes more debt capacity than any single revenue haircut.
Refinancing assumptions. Where the structure has a mini-perm, the lender's case will not assume a refinancing at the same terms. It will size the wall and test whether a stressed market clears it.
The practical consequence is that a sponsor expecting the banking case to be its own model with a lower price curve is usually surprised. The revenue haircuts are the ones that get discussed; the cost, tax and escalation adjustments are the ones that arrive quietly and, in aggregate, frequently exceed them.
The Regime Change Nobody Notices
Here is the part worth taking away from this post.
Base case CFADS ÷ project cost = 30.0 / 271.5 = 11.05%
Banking case CFADS ÷ project cost = 23.3 / 271.5 = 8.58%
Crossover (75% gearing, 1.30×, 6.5%, 18 years) = 9.35%
The sponsor's case sits above the crossover, so gearing binds and the debt is limited by the 75% cap — $203.6m rather than the $240.7m the coverage test would allow. The banking case sits below it, so coverage binds and the debt is limited to $187.0m.
That means the two parties are, without realising it, having two different conversations.
The sponsor thinks the negotiation is about the gearing cap, because in its model that is what limits the debt. Arguing the revenue case up does nothing for leverage.
The lender thinks the negotiation is about the revenue assumptions, because in its model those are what limit the debt. The gearing cap is not binding and moving it would change nothing.
Both are correct within their own model, and neither is addressing the other's point. Publishing the binding constraint alongside the debt number — as the first Series E post argued — resolves it in a line, and it is almost never done.
Sensitivities Are Not Scenarios
A distinction that determines whether downside analysis is useful.
A sensitivity moves one variable and holds everything else. It answers "how much does this input matter" and is the right tool for identifying which assumptions drive the answer.
A scenario moves a coherent set of variables in a way that could actually happen together. It answers "what does this state of the world do to us."
Most models produce many sensitivities and no scenarios, which leaves two errors uncorrected.
Independent haircuts understate correlated risk. As the yield post established, resource and curtailment are correlated, and as Series C showed, basis and capture rate and negative pricing all move together because they share a cause. Running each as a separate sensitivity and reporting the worst individually understates what happens when the underlying driver moves.
Independent haircuts also overstate uncorrelated risk. Stacking a P99 resource year on top of a P99 price year on top of a P99 availability year produces a number with a probability so small it carries no information, and it crowds out the scenarios that matter.
The useful downside set is small and coherent:
Low resource + high curtailment (correlated driver: regional build-out)
Merchant prices low + basis wide (correlated driver: local oversupply)
Delay + cost overrun (correlated driver: contractor stress)
Offtaker downgrade + refinancing (correlated driver: credit cycle)
Four scenarios that describe recognisable states of the world beat twenty sensitivities that describe none.
Break-Evens Beat Both
The most useful downside output is not a case at all. It is the point at which the structure stops working.
Every post in Series E and F produced one:
- The basis at which the lock-up covenant breaks — −$3.68/MWh on the worked project
- The accreditation at which capacity revenue stops mattering to the credit
- The CFADS at which leverage starts reducing the equity return
- The step-up in network upgrade costs that triggers the withdrawal exemption — 25%
- The contingency confidence level implied by what is carried
- The crossover CFADS between gearing and coverage constraints — 9.35% yield on cost
A break-even survives a forecast being wrong. A scenario does not: it is a specific set of numbers whose probability nobody can state, and it goes out of date the moment an assumption changes. A break-even says how much room there is, which is the question a credit committee is actually trying to answer and the one a base case cannot address.
If a model produces only one thing beyond the base case, it should produce break-evens.
Who Owns Which Case?
Governance matters here more than it sounds, because a case with no owner drifts.
The sponsor owns the base case and should defend it as its genuine expectation. Allowing lender haircuts to migrate into it destroys the only case that says what the project is worth.
The lender owns the banking case. The sponsor's role is to understand it, model it, and know in advance where it lands — not to negotiate its existence.
The lender owns the downside cases, with the independent engineer typically specifying the technical ones.
Nobody owns the reconciliation, which is the problem. The bridge in this post — four lines, each attributable, adding to $53.8m of debt capacity — is the document that makes the two models comparable, and it usually does not exist. Building it is a morning's work and it converts a disagreement about a number into a discussion about four assumptions, each of which can be examined on its own.
Who Verifies the Model?
A step that sits between the cases and the closing, and which sponsors consistently under-budget for.
The model auditor reviews the financial model itself — the formulas, the logic, the consistency of the calculations with the finance documents. It does not opine on the assumptions. It checks that the debt sizing does what the credit agreement says it does, that the waterfall in the model matches the waterfall in the schedule, that the covenant tests compute what the definitions define, and that the arithmetic is correct.
That last function is more valuable than it sounds, and this series has supplied a list of the things it catches: a sculpt whose final balance is not exactly zero, a circular reference resolving to an order-dependent value, a lock-up flagged but not applied to the distribution line, an IDC accrual on the closing balance, a waterfall position inconsistent with the CFADS definition.
The independent engineer opines on the technical assumptions — the energy yield, the availability, the capital and operating cost estimates, the construction schedule. Its energy assessment is usually the one the debt is sized on, and as the yield post noted, it is frequently below the sponsor's.
The insurance adviser, the tax adviser and the market consultant each opine on their own inputs, and the banking case is assembled from their outputs rather than from the sponsor's.
The consequence worth planning for: the banking case is built from third-party opinions the sponsor does not control, commissioned by a party whose interest is conservatism. A sponsor that engages with each adviser early, understands where their views differ from its own, and rebuilds its case on their numbers before the credit paper is written is in a far better position than one discovering the gap at credit committee.
The model audit in particular should happen before the model is used to negotiate, not after. An error found in diligence is a correction; the same error found after terms are agreed is a renegotiation.
When the Case Moves Mid-Deal
The situation nobody plans for and everybody encounters.
A term sheet is signed on a set of assumptions. Between signing and financial close — commonly six to twelve months — something moves. The independent engineer's yield comes in below the sponsor's. The market consultant revises the price curve. A cluster study lands with higher network upgrade costs. Interest rates move.
The banking case is rebuilt, the debt capacity falls, and the equity requirement rises by the difference. On the worked bridge that difference was $53.8m, and a smaller mid-deal revision of a quarter of that is still $13m of unplanned equity.
Three protections exist and each is worth negotiating at term sheet rather than later.
Specify the case at term sheet. Naming the assumptions the indicative sizing was based on — the yield assessment, the price curve, the cost estimate — converts a later change into a variation that has to be justified rather than a restatement.
Cap the flex. Market flex provisions allow a lender to adjust pricing and sometimes structure to syndicate a facility. A cap on how far sizing can move, or a right to terminate without cost if it moves beyond a threshold, bounds the exposure.
Model the sizing sensitivity in advance. The break-evens this post recommends do exactly this job: knowing that a 5% revenue revision costs $12m of debt capacity means the sponsor can see the risk before it arrives and can decide whether to secure additional equity commitments early.
The general observation, and it applies to the whole of this series: a number produced by someone else's model on someone else's assumptions is a risk, not a fact, and it should be treated with the same planning as any other.
How Do You Build This in Excel?
As one model with a case switch, and an explicit bridge between cases.
The case switch
Case_Selector: 1 = Base 2 = Banking 3 = Downside_A ...
Every haircut input reads from a case table:
Generation_Basis P50 / P90_term / P90_oneyear
Curtailment_Rate base / lender / stress
Basis_$/MWh base / lender / stress
Availability contractual / lender / stress
Never build separate workbooks per case. Separate files diverge, and the divergence is discovered at the worst moment.
The bridge, which is the deliverable
PF_BaseCaseRevenue $33,000,000
less generation basis change ($2,607,000)
less curtailment ($1,215,720)
less basis and capture ($2,334,182)
less availability ($542,231)
PF_BankingCaseRevenue $26,300,867
PF_BaseCaseDebtCapacity $240,749,224
PF_BankingCaseDebtCapacity $186,988,855
PF_CaseBridgeDebtEffect $53,760,369
The regime check
PF_BaseCaseBindingConstraint GEARING
PF_BankingCaseBindingConstraint DSCR
Two words that tell both parties what the negotiation is actually about.
The break-even panel
For each key driver, solve for the value at which:
DSCR = lock-up level
DSCR = default level
Equity IRR = hurdle
Binding constraint changes
ℹ️ Note: Report the number of scenarios and their drivers, not just their results. A downside case with no stated driver is a haircut with a label, and it cannot be argued about productively because nobody can say what would have to happen for it to occur.
To build the case switch, the bridge and the break-even panel, prompt Dezzmond with your base case and the lender's assumptions.
What Do Sponsors and Lenders Actually Check?
- Does a written bridge exist between the base and banking cases?
- Is the binding constraint the same in both? If not, the parties are negotiating different things.
- Is the base case genuinely unconservative, or has prudence leaked into it?
- Are downside cases coherent scenarios or stacked independent haircuts?
- Have correlated drivers been modelled as correlated?
- Does the model publish break-evens as well as cases?
- Is it one workbook with a case switch, or several files that have diverged?
Frequently Asked Questions
What is the difference between a base case and a banking case?
The base case is the sponsor's genuine expectation and sets the equity return. The banking case applies the lender's haircuts — generation basis, curtailment, basis, availability — and is what the debt is sized on.
How large is the gap typically?
On the worked example, four haircuts reduce revenue by 20.3% and CFADS by 22.3%, and cut debt capacity by $53.8m. The amplification from revenue to CFADS follows automatically from a fixed operating cost base.
Why does the binding constraint change between cases?
Because the cash yield on cost falls across the bridge — 11.05% in the base case against a 9.35% crossover, and 8.58% in the banking case. Above the crossover gearing binds; below it, coverage does.
What is the difference between a sensitivity and a scenario?
A sensitivity moves one variable and identifies which assumptions matter. A scenario moves a coherent set that could occur together. Most models produce many of the first and none of the second, which both understates correlated risk and overstates uncorrelated risk.
What does a model audit actually check?
The model's logic and arithmetic against the finance documents — not the assumptions. It catches sculpts that do not amortise to zero, circular references resolving to order-dependent values, lock-ups flagged but not applied to distributions, and waterfall positions inconsistent with the CFADS definition.
What happens if the banking case moves before financial close?
Debt capacity falls and the equity requirement rises by the difference. Specify the assumptions behind the indicative sizing at term sheet, cap the sizing flex, and model the sensitivity in advance so the exposure is known before it arrives.
What is the single most useful downside output?
A break-even — the value at which the structure stops working. Unlike a scenario it survives the forecast being wrong, and it answers the question a credit committee is actually asking.
Closing: Fifty-Four Posts, One Instruction
This is the last post in the series, so it is worth saying what all of it amounts to.
Six series have covered policy and tax, structuring, market design, contracts, financing mechanics and returns. The subjects are unrelated on the surface — a domestic content threshold and a cash sweep have nothing obvious in common. But nearly every post arrived at the same shape of conclusion, and it is this:
The number in the model is usually correct. The thing it describes is usually not the thing the reader assumes.
A P90 without a period. A DSCR computed on a CFADS defined by a waterfall nobody read. A capacity payment on nameplate rather than accredited megawatts. An IRR at an unstated level of the structure. A contingency that is a P84 on one project and a P55 on another. A constant WACC applied to a capital structure designed to change. A balanced sources and uses statement with 72% of its sources committed. A basis assumption averaged across hours the project does not generate in.
None of those is an error of arithmetic. Every one is a correct calculation of something other than what it appears to be — and in every case the fix was the same: state what the number is a measure of, and publish the thing it cannot tell you.
The headroom, not the ratio. The binding constraint, not the amount. The bridge, not the three returns. The confidence level, not the percentage. The break-even, not the central case.
That is not a modelling technique, and no skill or template delivers it. It is a habit of asking what question a number is the answer to — which is available to anyone, costs nothing, and would have prevented most of the expensive surprises described across these fifty-four posts.
Sources: Edward Bodmer — Project Finance Exercises · Wall Street Prep — Distinctive Features of a Project Finance Model · LexisNexis — Project Finance Financial Covenants · Ryan O'Connell, CFA — Building a Project Finance Financial Model