Which SOFR does your loan use? A benchmark identity and timing audit

A borrower cannot verify a floating-rate payment by searching for “SOFR today” and copying the first number. The relevant question is which defined series the loan uses, for which period, under which observation rule. A correct number from the wrong series can still produce an incorrect payment.

This reference provides a benchmark identity audit. It deliberately stops before adding a loan spread, floor, fees or hedge settlement. Those separate steps belong in the benchmark-to-coupon reconciliation.

Three similar labels with different information

The New York Fed methodology distinguishes an overnight observation from compounded averages covering historical 30-, 90- and 180-calendar-day windows. Its index permits custom-period compounding. Daily SOFR is published after its value date; the averages and index are published on their value date. That distinction matters when assembling a dated record.

CME’s benchmark statement describes Term SOFR as forward-looking, with 1-, 3-, 6- and 12-month tenors. A one-month term fixing therefore does not become equivalent to a historical 30-day compounded average simply because both describe roughly a month.

The observations in this article are hypothetical. The comparison does not establish which benchmark any particular loan uses or whether a proposed fallback is permitted.

Build a record that another person can reproduce

Create one row for each reset or calculation period. Record the contract’s exact defined term, document version and section; the administrator and series; tenor if applicable; publication date; value date; accrual start and end; observation adjustment; raw value; units; and source location.

Do not compress publication date and value date into a single “as of” field. A reviewer needs to distinguish when market activity occurred from when its result became available. Likewise, keep the date a screenshot was downloaded separate from both. Downloading an old observation today does not refresh it.

If the agreement refers to an observation shift, lookback, lockout or business-day adjustment, transcribe the applicable instruction into a separate implementation field and have the responsible reviewer confirm it. Similar labels can describe different calendars. A spreadsheet should not silently choose a convention because its default function expects one.

Reconstruct a small compounding window

Consider a hypothetical five-calendar-day window with a constant $2 million principal. The first observation applies for three days, followed by two observations that each apply for one day. Assume the administrator-style actual/360 compounding convention described here, no spread and no principal changes.

Multiplying the three factors produces 1.000677908179. The resulting compounded amount is $2 million × (factor − 1), or $1,355.82. Expressing the five-day result as an annualized average gives (factor − 1) × 360/5, or 4.88093889%.

Applying simple interest to the same daily blocks instead produces $2 million × (0.048 × 3 + 0.049 + 0.051)/360 = $1,355.56. The $0.26 rounded difference is small in this short example, but the methods are distinct. The test is whether the implementation follows the required method, not whether an accidental difference happens to be immaterial today.

The three-day block uses one factor with three days of simple accrual. It is not three separate daily compounding factors. The New York Fed explains its business-day compounding and intervening nonbusiness-day treatment in the methodology. Do not extend this illustration to another contract’s calendar without checking its terms.

Trace discrepancies in the right order

First compare series identity and dates. If those differ, investigating cents of rounding is premature. Next reconcile the observation calendar and day counts. Then compare raw precision and the point at which rounding occurs. Finally move to the loan-specific spread, floor, principal changes and cash-payment rules.

An index-based reconstruction can be a useful cross-check, but rounded index observations can introduce small precision differences. Preserve raw inputs and explain the tolerance used; do not repeatedly adjust a rate until the answer matches the lender’s statement. A match achieved through an unexplained plug is not a verified calculation.

When an observation is missing, mark the calculation incomplete. Use the contractual fallback or obtain a written clarification from the responsible party. Carrying forward yesterday’s rate may be allowed under a specific provision, but it is not a universal data-cleaning rule.

What a completed audit establishes

A completed benchmark register establishes that the chosen input can be traced to the specified series and dates, and that the reconstruction follows the stated convention. It does not establish that a benchmark forecasts the loan’s eventual cost, that a rate lock is available, or that a hedge has matching cash flows.

Before changing a model, retain the previous input set, record the correction and recalculate affected periods. That makes the economic effect visible and allows a reviewer to separate a data correction from a change in financing assumptions.