compounded-sofr

SOFR compounding-in-arrears math — the ARRC & ISDA conventions (lookback, observation shift, lockout), the SOFR Index method, the compounded average, and daily-simple accrual. Zero dependencies.

npm install compounded-sofr

Why

SOFR replaced LIBOR, and unlike LIBOR it is an overnight rate — so a three-month SOFR coupon is a compounded average of daily rates set in arrears, with a small zoo of conventions (how many business days to look back, whether to shift the observation window, whether to lock out the final days). The New York Fed publishes a SOFR Index and 30/90/180-day averages built on exactly this math, and npm had no library that implements it. This one does — and reproduces those published numbers.

import { compoundedSofr, sofrAverage, averageFromIndex, indexFromRates } from "compounded-sofr";

// observations: one { date, rate } per business day (rate in %), e.g. from the
// `newyorkfed` package. A Friday's rate carries its weekend weight automatically.
compoundedSofr({ observations, start: "2026-06-09", end: "2026-07-09" });
// 3.634089… — the NY Fed's published 30-day SOFR average for 2026-07-09 is 3.63409

// Exactly the published SOFR Average (window may start on a weekend; rounded to 5 dp):
sofrAverage(observations, "2026-07-09", 30); // 3.63409

// The exact, path-independent method straight from the published SOFR Index
// (index on 2026-06-09 and 2026-07-09, 30 calendar days apart):
averageFromIndex(1.24646364, 1.25023844, 30); // 3.634089… — within ~2e-5 of the published average

// Reconstruct the SOFR Index by compounding the daily series (seeded 1.0 on 2018-04-02):
indexFromRates(observations, 1.0);

API

Rates are percents (3.53 = 3.53%), matching the NY Fed API. Observations are { date: "YYYY-MM-DD", rate }, one per business day.

• compoundedSofr({ observations, start, end, convention? }) — compounded-in-arrears rate for the interest period [start, end). convention is { type: "plain" } (default), { type: "lookback", businessDays }, { type: "observationShift", businessDays }, or { type: "lockout", businessDays }.
  • start/end may fall on weekends or holidays: such a day takes the preceding business day's rate (the NY Fed's rule). end's own rate is never used, so observations need only run to the business day before end (at most 7 calendar days short). Under observationShift both must be business days in observations.
  • lookback: rate from businessDays business days earlier, weights from the interest period. observationShift: rates and weights from the shifted observation period. lockout: the last businessDays business days reuse the rate of the business day just before them (QuantLib's lockoutDays); an ARRC-style Rate Cut-Off Date N business days before the end corresponds to businessDays: N − 1.
  • For lookback / observation shift / lockout the observations double as the business-day calendar, so they must include every business day of the period (and the shift days before it).
• sofrAverage(observations, asOf, calendarDays) — the NY Fed's SOFR Average for asOf: compounded over [asOf − calendarDays, asOf), rounded to 5 decimals as published.
• averageFromIndex(startIndex, endIndex, days) — compounded average between two SOFR Index values: (end/start − 1) × 360/days. Unrounded; because the published index is rounded to 8 decimals it can differ from the published average by ~1–2 units in the 5th decimal.
• indexFromRates(observations, startIndex?) — compound a daily series forward into an index value as of the last observation's date (each rate accrues until the next observation; the last observation's rate is not used).
• compoundInArrears(periods) / dailySimpleAverage(periods) — the underlying primitives over { rate, days }[].
• weightedRates(params) — the per-day { rate, days } schedule a convention produces (inspect or reuse it).
• accruedInterest(notional, params) — notional × rate/100 × interestDays/360, with interestDays the calendar days in [start, end) (also under observation shift).
• withSpreadAdjustment(rate, spreadBps) — add an ISDA fallback spread.

The 360-day basis, the calendar-day weighting (weekends fold into the preceding business day), and in-arrears compounding all follow the ARRC conventions.

Correctness

Validated against independent sources (the test suite carries fixtures from both):

• NY Fed published data — sofrAverage reproduces every published 30/90/180-day SOFR Average exactly (all 4,956 values from 2018-04-02 to 2026-10-09 in the audit cross-check; 750 in the test fixtures), including windows that start on weekends and holidays. Compounding the full daily history from 1.0 on 2018-04-02 with indexFromRates and rounding to 8 decimals reproduces every published SOFR Index value.
• QuantLib (OvernightIndexedCoupon) — plain, lookback, observation-shift and lockout rates and accrued amounts agree to 1e-9 over 1,698 random periods on real SOFR fixings, including non-business-day starts and ends.

Related

Pairs with newyorkfed (fetch the daily SOFR and index). Part of a fixed-income toolkit: treasurydirect · tbill · tips-index-ratio · day-count · accrued-interest · 32nds · sifma-holidays · instrument-identifiers.

Author

Built by Moshe Malka — engineering leader in New York City. Studio work at Quentin.Code.

MIT © Moshe Malka