Curtailment Risk, Grid Congestion and Project-Finance
A 30-year project-finance model wrapped in 10,000 Monte Carlo trials, testing whether a 150 MW California solar plant still clears its lender covenant as curtailment rises.
- Date
- Mar 2025
- Role
- Lead analyst
- Methods
- Monte Carlo simulation, Project-finance modeling, Techno-economic analysis
- Tools
- Python, Excel/VBA, NREL SAM
77.6%
DSCR covenant breach
Grid-Constrained trials below 1.20×
69 → 40$/MWh
Delivered LCOE
Grid-Constrained to Grid Relief
10,000
Monte Carlo trials
Per regime, over a 30-year pro forma
2.3 → 17.9%
Equity IRR spread
Same plant, same capital stack
Overview
California curtailed roughly 3.4 TWh of solar and wind in 2024 — a 29% jump on the year before, as build-out outpaced transmission. Curtailment is usually filed under lost generation. This study asks a narrower and more consequential question: at what point does it stop being an efficiency problem and become a credit problem?
The model couples a 30-year cash-flow pro forma for a 150 MW single-axis-tracking plant with a Monte Carlo layer over four uncertain inputs — curtailment rate, capacity factor, PPA price, and CapEx — each given a triangular distribution calibrated to CAISO OASIS data, public PPA bids, and the 2025 NREL ATB. Three grid regimes are tested against an identical capital stack, so the only thing that moves between them is the grid.
The answer is that curtailment is now a first-order credit variable. Under the Grid-Constrained regime the minimum debt-service coverage ratio falls below the 1.20× lender covenant in 77.6% of trials, against 21.4% in the Base Case and 7.5% under Grid Relief. Delivered LCOE moves $69 to $40/MWh across the same span. Financeability turns less on hardware cost curves than on whether transmission keeps pace with solar additions.
The question
Lenders underwrite solar against a minimum debt-service coverage ratio, typically 1.20×. Curtailed energy earns nothing, so every curtailed megawatt-hour lands directly on the cash flow available to service debt. If curtailment keeps climbing, the question a sponsor and a credit committee both need answered is where the covenant actually breaks — and which lever moves that outcome most.
Point estimates cannot answer it. Two projects with near-identical average coverage can diverge sharply once the full distribution is considered: one is bankable, the other default-prone. That is the gap this study is built to close.
Approach
A 150 MW single-axis-tracking plant is modeled over 30 years under fixed financing — 70% debt, 20-year tenor, 6% coupon, 30% ITC, five-year MACRS, 30% blended tax rate. Operating assumptions are held constant across every scenario: $1/MWh variable O&M, $21/kW-year fixed O&M, insurance at 0.15% of CapEx, $5,000/MW-year land lease, a $40/MWh merchant tail, and 2% revenue inflation.
Two layers sit on top. A deterministic layer runs three plant-performance profiles that differ only in resource yield, curtailment, and module degradation. A stochastic layer then runs 10,000 Monte Carlo trials per regime over four triangular-distributed inputs, producing a full distribution of minimum DSCR rather than a single figure.
Because capital cost, tariff structure, and leverage are identical across profiles, any divergence in coverage, LCOE, or equity return is attributable to the grid alone.
Scenarios
The three deterministic plant-performance profiles. Financing, operating costs, and tax treatment are identical across all three — only resource-side variables move.
| Profile | Capacity factor | Curtailment | Annual degradation |
|---|---|---|---|
| Risk-Averse | 22% | 20% | 0.75% |
| Reference | 28% | 9% | 0.50% |
| High-Performance | 30% | 2% | 0.25% |
Source: CAISO (2025); NREL ATB (2024).
Figure 1
Minimum and average DSCR by scenario
Deterministic runs under identical financing (70% debt, 20-year tenor, 6% coupon). The Risk-Averse profile is the only one whose minimum DSCR falls below the 1.20× covenant — at 0.76×, it does not service its debt.
Bar chart of average and minimum DSCR across three scenarios. Risk-Averse: average 1.22×, minimum 0.76×. Reference: average 1.89×, minimum 1.26×. High-Performance: average 2.27×, minimum 1.51×. The lender covenant sits at 1.20×; only the Risk-Averse minimum falls below it.
Source: Jamhar (2025), Table 3. Calibrated to CAISO (2025) and NREL ATB (2024).
Data table
| Scenario | Average DSCR | Minimum DSCR |
|---|---|---|
| Risk-Averse | 1.22× | 0.76× |
| Reference | 1.89× | 1.26× |
| High-Performance | 2.27× | 1.51× |
Results
Deterministic outcomes. The Risk-Averse profile does not merely underperform — at a 0.76× minimum DSCR it fails to service its debt, and its 2.3% equity IRR sits below any plausible investor hurdle.
| Metric | Risk-Averse | Reference | High-Performance |
|---|---|---|---|
| Minimum DSCR | 0.76× | 1.26× | 1.51× |
| Average DSCR | 1.22× | 1.89× | 2.27× |
| LCOE ($/MWh) | 68.98 | 46.79 | 39.67 |
| Pre-tax IRR | 4.00% | 9.32% | 11.81% |
| Post-tax IRR | 3.69% | 8.49% | 10.69% |
| Equity IRR | 2.32% | 12.39% | 17.94% |
Source: Jamhar (2025), Table 3.
Figure 2
Minimum DSCR distribution by curtailment regime
10,000 Monte Carlo trials per regime. The bar spans the simulated minimum to maximum, the solid box is one standard deviation either side of the mean, and the notch marks the mean. What separates the regimes is not spread but level — the Grid-Constrained mean sits below the covenant, so ordinary variation cannot rescue it.
Interval plot of minimum DSCR across three curtailment regimes against a 1.20× covenant. Grid-Constrained: mean 1.10×, range 0.76 to 1.60, 77.6% of trials below covenant. Base Case: mean 1.31×, range 0.92 to 1.79, 21.4% below. Grid Relief: mean 1.383×, range 1.04 to 1.80, 7.5% below.
Source: Jamhar (2025), Table 4. Triangular inputs calibrated to CAISO (2025), CPUC (2024), NREL ATB (2024).
Data table
| Regime | Mean min DSCR | Std. dev. | Range | Trials below 1.20× |
|---|---|---|---|---|
| Grid-Constrained | 1.100× | 0.120 | 0.76–1.60× | 77.6% |
| Base Case | 1.310× | 0.137 | 0.92–1.79× | 21.4% |
| Grid Relief | 1.383× | 0.121 | 1.04–1.80× | 7.5% |
Key findings
- 01
Curtailment is a first-order credit variable
Below roughly 10% coverage margin the entire DSCR distribution drops under the 1.20× covenant. Held near 5%, the same plant carries 70% debt and still returns mid-teens to equity.
- 02
Level beats spread
The Base Case has the widest dispersion yet breaches least, because its mean clears the covenant. Grid-Constrained fails systemically because its mean does not.
- 03
Each point of curtailment is worth $3–4/MWh
Near-linear across the tested range, which puts curtailment mitigation on the same footing as capex reduction.
- 04
Average DSCR hides the risk that matters
Two profiles averaging about 1.3× diverged sharply once full distributions were considered — one bankable, one default-prone.
- 05
Grid headroom deserves equal weight to irradiance
A high-insolation parcel with congested interconnection can underperform a moderate-sun, uncongested one once debt service is counted.
Mitigation
Three layers of defence emerged, each cheapest at a different point in the stack. On the grid side, targeted transmission upgrades and full deployment of CAISO's Extended Day-Ahead Market would relieve midday bottlenecks; moving from Grid-Constrained to Base or Relief conditions all but eliminates covenant-breach events in simulation.
Where transmission is slower to arrive, pairing the plant with a four-hour battery at roughly 20–25% of nameplate absorbs midday oversupply and dispatches it into the evening ramp. At current storage costs this raises LCOE by only a few dollars per MWh while restoring enough coverage cushion to hold the original debt sizing.
Contractually, PPAs carrying take-or-pay or curtailment-compensation clauses shift part of the risk to offtakers; even partial reimbursement for curtailed energy materially lifts coverage in low-revenue years. DSCR-linked insurance and credit-default swaps can then cover the residual gap, trading some IRR for the ability to preserve leverage.
Curtailment data sourced from CAISO's Open Access Same-time Information System (OASIS). Developed under the guidance of Professor Alvar Escriva-Bou, UC Davis. All modeling files and CAISO curtailment data referenced in the report are available in the linked appendix.