Update: This more recent post, Apportioning Discretionary Funding for Expected WIFIA Portfolio Systemic Funding Losses, describes the principles and mechanics involved in the topic in greater detail.
Source: InRecap analysis, EPA website, WH Budget Technical Appendices
The Economic Cost of an Average WIFIA Loan, Expected and Actual
The economic cost [1] of an average EPA WIFIA loan at time of loan drawdown 2018-2025 appears to be straightforward.
The Expected case was about 1% of loan principal amount from taxpayers to cover projected credit losses. The other 99% was assumed to be repaid, with UST interest, by the borrower. Given the characteristics of WIFIA’s mainly Aa3/AA- borrowers, this was conservative.
The Expected case, however, was also based on assumptions about the cost of funding the loan commitment when it was finally drawn, many years later. Since interest rate movements are unpredictable, an estimate of the funding loss or gain on an individual loan was not considered possible. But since over time interest rates are cyclical and revert to mean, eventually individual loan funding losses and gains would balance out. On a portfolio basis, therefore, the WIFIA Program would be costless with respect to funding the loan commitments. Critically, this conclusion relies on the assumption that WIFIA borrowers will draw on loan commitments in a way that was generally correlated to the project construction schedule, and completely uncorrelated to UST rates, relative to the loan commitment’s rate and the borrower’s financing alternatives.
The Actual results over the period, of course, were significantly different. There was no reason to revise credit loss assumptions, and that cost remained at 1%. But the net funding losses covered by taxpayers (i.e., WIFIA loan rate lower than UST funding cost at drawdown) on drawn loans were not zero or minor, but about $2b, or 9% of the $22b loan commitment portfolio at FYE25 [2].
This is not mysterious. Obviously, Aa3/AA- borrowers with significant resources and tax-exempt financing alternatives (both short and long term) will be proactively choosing the least-cost option to finance their project at every point. Decisions will be highly correlated with UST rates. Construction advances will be covered with short-term financing when short-term rates are below the WIFIA loan commitment rate. Drawdown of the WIFIA loan as permanent financing will only occur if its cost is roughly equal to or less than near-UST tax-exempt bond financing — otherwise, the commitment will be cancelled or the rate reset. Funding gains will be rare and minor; funding losses will be frequent and occasionally significant, especially when commitment rates are reset at low points in the rate cycle, as many were in 2020-2021.
The Expected case contains assumptions about the funding cost of WIFIA loans to taxpayers that might have appeared reasonable at the outset of the Program. But now there are Actual results which appear to conclusively invalidate those assumptions. Apart from the $2b funding loss itself, there is the consistent Aa3/AA- water agency profile of the borrowers, the fact that those borrowers did use ST financing for construction advances as opposed to drawing on a WIFIA loan commitment, the opportune timing of interest rate resets and the pressure brought to bear on the Program to offer them, and the overall correlation of WIFIA loan applications with conditions in the tax-exempt bond market. All of this forms a coherent data set on which expectations of future borrower behavior and funding losses may be based.
Incentives to Continue Using Original Expected Case
In light of Actual results, there is no justification for using the original Expected case going forward. Even if the precise funding loss for an individual loan cannot be predicted, the fact that funding losses are more likely than funding gains from WIFIA’s predominant borrower class is sufficient to predict that on a portfolio basis there is a near-certainty of net funding loss.
Regardless of justifications, however, there are incentives. When a WIFIA loan commitment is executed, the credit subsidy estimate for the specific loan is apportioned by OMB from the Program’s discretionary appropriations. Using the Expected case, only the cost of expected credit loss is considered, about 1% of loan amount. This approach minimizes the use of Congressionally approved discretionary funding and supports a ‘narrative’ that the Program is amazingly cost-effective.
Years later, when the loan commitment is drawn, the full cost of the loan to taxpayers is calculated. The credit subsidy amount is ‘re-estimated’ to account for changes from the original estimate, very often including that the loan is funded at a loss. However, this loss will come in the form of a ‘positive interest rate re-estimate’ that automatically receives ‘permanent indefinite authority’ to incur mandatory appropriations [3]. Congress is not involved and the spending itself is buried in technical budget accounts, far away from public notice and the narrative sphere. Hidden, in effect.
Anti-Deficiency Act Violations, Coercive Deficiency
I’ve noted in previous posts that spending triggered by the mechanical operation of FCRA interest rate re-estimates cannot be in violation of the Anti-Deficiency Act, regardless of facts and circumstances, because FCRA law is black-and-white and the ADA simply excludes such actions.
The calculation and apportionment of the original credit subsidy estimate, however, is a different matter. This process occurs in connection with the execution of a WIFIA loan commitment, and that commitment very much requires the allocation of future federal resources, for which funding must be provided. The estimate at time of commitment is meant to include all predictable costs of the loan, and those costs must be covered by an apportionment of the program’s available discretionary funding. EPA WIFIA is not an ‘entitlement’ program — mandatory appropriations are available only in the specific case that a good-faith estimate of a loan’s funding cost turned out to be inadequate upon re-estimate, not for predictable funding losses. Continuing to base estimates on demonstrably wrong assumptions, especially when there’s an obvious incentive to do so, is effectively to create a ‘coercive deficiency’, a straightforward violation of the ADA.
The Principles and Mechanics of Correct Funding Loss Estimates
In all important ways, an estimate of future loan funding loss at loan commitment execution is the same as an estimate for the loan’s future credit loss, a completely routine process.
As with a credit loss estimate, each borrower’s characteristics, and the likely consequences of those characteristics, need to be considered. A large Aa3/AA- water agency with extensive financial resources and sophisticated advisors is more likely than not to draw a WIFIA loan commitment only when it is materially ‘in-the-money’ relative to the then-current UST curve. The likelihood of a smaller, less highly rated agency doing so will be less. A strapped rural co-op or a Baa3/BBB- project financing can be expected to draw the commitment regardless of changing rates because they have no alternatives.
Analogous to credit ratings, borrowers can be roughly categorized with respect to the probability of incurring WIFIA loan funding loss and historical data from the category analyzed. EPA WIFIA now has such data, obviously not as extensive as for credit loss but sufficient for initial estimates. My guess is that most of the $2b of program funding loss 2018-2025 came from the largest, most highly rated borrowers — the correlation of loss with characteristics is probably strong, perhaps even statistically significant in a formal regression analysis.
The mechanics of quantifying estimates of loan funding loss are, I think, equally straightforward. OMB’s Credit Subsidy Calculator (CSC) is not mysterious — it is just a standard model for discounting projected cash flows that incorporates zero-coupon UST curves. When the loan commitment is executed, the loan’s interest rate per program policy must reflect then-current UST rates (either the yield at WAL or the CSC’s ‘single effective rate’ for projected cash flows). But the estimate for apportionment can assume the ‘disbursement scenario’ with a higher UST curve that reflects the borrower’s likely behavior (e.g., will on average draw only when UST curve is 15% higher than at commitment). The projected funding loss will appear as the amount by which the PV of the loan’s debt service is lower due to the higher discount rate, and the apportionment estimate for a funding loss reserve can be based on that.
When drawdown occurs and the cost of the loan is finalized, FCRA interest rate re-estimate mechanics will work as before. If rates are exactly as anticipated under the disbursement scenario, no re-estimate is needed. If rates are lower, some or all of the funding loss reserve will be returned to the program’s general budget authority. If rates are higher, permanent indefinite authority will provide mandatory appropriations to increase the funding loss reserve to the required amount. The only difference from current practice is that the predictable part of funding loss is funded by the program’s discretionary appropriations. To the extent that for a specific loan in a specific year most of the funding loss or gain will be unpredictable, all the re-estimate procedures work as intended. Over time, if the funding loss reserve estimates are correct, the program’s portfolio will tend to balance — mandatory spending will be offset by program gains, and discretionary funding will reflect the true cost of WIFIA loans for taxpayers.
No Excuse for Delay
I don’t see any excuse for EPA WIFIA or OMB to delay the inclusion of near-certain funding losses into apportionment for credit subsidy estimates. The process is analogous to apportioning for expected credit losses, and the mechanics to quantify the funding loss reserve are simple discounting adjustments that can be based on the program’s current data set. The estimates change nothing from the borrower’s perspective (they get the same then-current UST rate at commitment) and if somewhat rough initially, the re-estimate process will in any case provide the final cost numbers. The taxpayer will pay the same regardless — the only difference is an adjustment between mandatory and discretionary spending to improve budget accuracy and demonstrate compliance with the ADA. The estimates and methodology can be continuously improved — but the practice should be started without delay.
From a practical perspective, EPA WIFIA has a huge carryover of discretionary funding, with another $65m added in FY2026 spending bill, for a total of about $300m. That’s far in excess of what’s required for any likely level of loan volume over the next few years. Apportioning some of that pile to a funding loss reserve at loan execution (or perhaps more likely, re-execution when the inevitable resets are done for recently closed loans) will not constrain the program in any way. To the extent that an investigation into ADA violations really would constrain the program, especially with respect to approval for non-statutory resets, I would think that WIFIA stakeholders would support improved estimates.
______________________________________________________________________________________
Notes
[1] I discuss why ‘economic cost’ of WIFIA loans, as calculated using UST curves per OMB methodology, is the correct way to evaluate the utilization of taxpayer resources in the first subsection of this post — The Economic Cost of WIFIA’s Portfolio at FYE 2022
[2] It should be noted that according to the program’s press releases, only about $10b of the $22b portfolio has been disbursed as of mid-2025. It’s not clear if this means final drawdowns (90% per OMB rules) that trigger final interest rate re-estimates and mandatory spending, but assuming it does, the $2b in re-estimates may only have been incurred with $10b of loans — 20% of loan amount. On the other hand, there was a concentration of very low-rate loans executed or re-executed in the Covid Treasury dip 2020-2021, and more normal re-estimate losses will be a lower rate, perhaps even lower than the 9% overall to date. The program and OMB have the data — they should use it.
[3] I discuss some of the mechanics of mandatory spending in the second half of this recent post — Two Technical Clarifications – 55Y/35Y Difference, Mandatory Spending.