Methodology

Every calculation on Rate Workbench follows a named convention, and a calculation that uses published data shows the date of that data. Each tool page shows checks that run every time the site is built. Some compare the calculation with numbers an official source published. Others, where no official source publishes the figure, test the arithmetic against a definition or a formula. This page describes the approach. The formula for each tool is on that tool's own page.

One engine behind every page

Apart from the reference values of the checks, every rate, price, yield, duration and amount of interest that the site calculates comes from one calculation library, called the engine. The code of each page and calculator adds only small steps around it, such as converting a period between units of time, calendar arithmetic, formatting a figure for display, drawing a chart, and the sum or difference of two figures the page prints. The page that shows the latest number, the everyday calculator and the practitioner page for the same topic all call the same engine functions. The calculator in your browser runs the same compiled code that built the page. Two pages therefore cannot give different answers to the same question.

The engine only does arithmetic. It fetches no data and does not look up the current date; both are passed in. The same inputs always give the same result.

A result is rounded once, when it is shown. In a check it is compared with a reference value at the number of decimals the row shows. Where a source's own rule rounds, as the Treasury's does for the price at an auction and the price of a bill, the engine rounds as the source does. The engine's functions for the change from one published rate to another, the gap between two, the midpoint of a range and a rate plus a margin also round, at one ten-thousandth of a percentage point, which is a hundredth of a basis point: this clears the stray digits that binary arithmetic leaves in the last places, and it loses no digit of the rates that go in, none of which has that many decimals.

In a few places a later figure is worked from one already rounded, so that the step can be repeated from the figures on the page. The Treasury bill calculator takes the interest from the price paid in cents and the rate after tax from the investment rate as shown. The bond calculator takes the amount paid from the price and the accrued interest as it shows them: the accrued interest and a price computed from a yield at six decimals, and a price entered at the decimals typed, up to eight. The bond fund calculator turns a level for the 10-year yield into a move at the hundredth of a basis point and works every figure from that move. The savings calculator starts from the change in the CPI as shown. The pages on bond funds take the percent between a model's duration and the one the fund's issuer published from the two figures as shown. And a sentence that gives a sum or a difference of figures it prints works it from those figures, as does a table cell that repeats the sentence's figure. One table column adds up as printed in the same way: on the rate change calculator each yearly amount is rounded to the cent, and the yearly totals are the sums of the amounts as shown. Any other table cell is rounded on its own, the monthly column of that table included, so a total need not equal the sum of the cells as printed. The same holds for a figure that follows from others on the page by a product or a ratio, such as a DV01 beside a duration and a price, interest beside a rate and an amount, or a monthly amount beside a yearly one: each is rounded once from the unrounded figures, so the same step on the figures as shown can end in other last digits, and when it applies a rounded rate or percent to an amount, the difference grows with the amount and, if the rate is compounded year after year, can grow with the number of years. The pages that print such figures side by side say what each is worked from: the Treasury bond price and yield calculator, the SOFR compounded calculator, the rate change calculator, the bond ETF interest rate calculator and the six bond fund pages, the savings versus inflation calculator, the Treasury curve lab, Prime rate today and TIPS break-even inflation.

Conventions are named

Bond and interest calculations depend on conventions: how days are counted, how often interest compounds, on which date a trade settles. Each tool page names the conventions it uses and links to the source that defines them.

Where the site has to choose a rule that no official source dictates, the page names that rule too. The main example is the curve. The Treasury publishes par yields at a fixed set of maturities. To get zero rates and forward rates from them, the site interpolates par yields between the published maturities on a semiannual grid with a monotone cubic rule (Fritsch-Carlson), then solves for the zero rates one maturity at a time. Linear interpolation is kept for comparison.

The checks on every tool page

Every tool page has a table of checks. Each row sets a reference value beside the engine's result for the same inputs, and says in its description or its source where the reference comes from.

Published figures. On every tool page except the ones named in the next paragraph, the rows rest on what an official source published: a rate, a yield, an index level, an auction price, a scheduled date, or a worked example the source printed. Where a row takes such a figure as it stands, the engine is given the same inputs and has to arrive at the same figure. Other rows on these pages are worked from published figures or from a published rule by a step the row states, such as the difference of two published yields, the ratio of two published index values or a count of days. Each published figure is recorded with its source and the date it was retrieved.

Definitions and formulas, where nothing is published. The Treasury curve lab, the bond ETF interest rate calculator and the bond fund pages calculate figures that no official source publishes, so their rows have no published figure to match. In the curve lab, a bond that pays the par yield of each published maturity is priced on the curve derived from those yields and has to come back to 100, and one forward rate is set beside the same rate worked from two zero rates. On the calculator and each fund page, the row sets the modified duration of a 10-year par bond by its closed form beside the same duration measured by repricing the bond 1 basis point up and down. That row is about one bond. It does not test a fund's model. Rows of this kind test the arithmetic against a definition, a formula or a second route to the same figure. They do not show that a result is close to an official figure, because there is none.

A row matches when the reference and the engine's result print the same at the number of decimals the row shows. For a published figure that is the precision the source publishes. A row for a date matches when both are the same day. Where a source states that it publishes no figure, the row matches when the engine gives none either.

The table is computed when the site is built, not written by hand. If a row stops matching, the build fails, so a version of the site with a failing check is never produced. What you see in the table is a test that ran.

Fund models compared with the funds' daily moves

Each bond fund page also carries a comparison of its model with the fund's own price moves. This is a measurement made by this site, not a check of the kind described above. It has no reference value to match, published or worked, so the fund pages show its figures as they came out and draw no conclusion from them.

Two series of daily returns are set against each other. The model's is the return of the previous trading day's model portfolio when it is valued again on the day's par curve, with any payments that fell due in between added. The fund's is the return from one exchange closing price to the next. The comparison reports the share of the variation in the fund's daily moves that the model's moves explain, and a slope, which says whether the fund's moves were larger or smaller than the model's.

For this comparison only, each day's par curve is valued for the next business day, the day on which trades at those quotations are taken to settle. That is an assumption of this site's measurement and not something the Treasury states: its description of the curve gives the time of the quotations and not their settlement day. The tool pages keep each curve on the date the Treasury gives it.

The funds were not all measured the same way. The days on which a fund paid a distribution were treated in one of three ways: the amount was added back to the fund's return, those days were left out, or no adjustment was made. Each fund page says which applied to it. The periods are not the same for every fund either. The figures of two funds are therefore not a ranking.

The comparison covers one period and describes the direction and size of daily moves. It does not show a fund's return or its price, and it says nothing about another period. Two of its known limits: the par curve rests on quotations taken earlier in the day than the closing price is struck, so a move in rates in between appears as a move the model does not explain, and the list of securities outstanding on one date is applied to every earlier day.

The closing prices are held privately and are not on this site. The statistics are a record made on the date the block gives, and they are not recomputed when the site's data is updated. Each fund's figures are on its own page, in the block whose heading names the fund's model and ends with the words "compared with the fund's daily moves".

Dates: the as-of date and the retrieval time

Every number from a source carries two times.

  • The as-of date is the calendar date the number describes, in US Eastern time.
  • The retrieval time is the moment the site fetched the number from the source.

The two often differ. The Federal Reserve Bank of New York, for example, publishes the effective federal funds rate for the prior business day at about 9:00 a.m., so a rate retrieved this morning describes the previous business day.

Pages with "today" in their name show the latest published value. The as-of date next to the number says which day that is.

Sources sometimes revise a number after publishing it. The site shows what the source showed when it was last retrieved.

Late data: the stale flag

Each snapshot of data records the moment after which it counts as late: the source's next expected publication plus a grace period, with weekends and holidays taken into account. If that moment passes and no newer data has arrived, the number is shown with a visible flag that says it is late.

The check runs twice. It runs when the site is built, and it runs again in your browser against your own clock. A page that was built on time therefore does not go on presenting an old number as current.

New data is checked before it replaces the old: required fields, plausible ranges, and an as-of date that is neither in the future nor earlier than the one already held. Data that fails a check is not used. The last good snapshot stays in place until it becomes late and is flagged.

Limits

  • Par yields are not tradable prices. The Treasury derives its par yield curve from indicative bid-side quotations, not from actual transactions. A point on the curve is the yield of a theoretical new security and may not match the yield of any one actual security.
  • Models and approximations are labeled. Where a result comes from a model or a shortcut and not from exact pricing, the page says so. The rate scenario tools reprice each security on the shifted curve. They show the duration and convexity approximation beside that result and label it as an approximation.
  • Fund pages use a model portfolio, not the fund's holdings. For each bond fund, the site builds a stand-in portfolio from the Treasury's list of marketable securities outstanding. It follows a rule this site records for that fund, and it values every payment on the day's curve. The TLT, IEF, SHY and GOVT models hold each note and bond in their maturity band in proportion to its whole amount outstanding. GOVT's summary prospectus says its index is market value weighted, based on amounts outstanding reduced by the amounts the Federal Reserve holds, so the model differs from that index there. The summary prospectuses of TLT, IEF and SHY say their indexes are market value weighted and name the Federal Reserve's holdings only in a test of minimum size. All four were read for this site on November 15, 2026. The EDV and ZROZ models hold one principal payment for each bond in their band, all in the same face amount. For ZROZ the band and the weighting are this site's assumption. In every model, a bond with a payment more than 30 years away is left out, because the curve ends there. The site does not show the fund's holdings, price or net asset value, and no tool reads them. The model's duration can differ from the figure the fund's issuer publishes, and you can enter the issuer's figure to use that instead.
  • Scenarios are arithmetic, not forecasts. A tool that shows what happens if rates move by some amount says nothing about whether they will.

Corrections

If you think a number is wrong, the contact page says how to report it. A report is checked against the official source. When a calculation or a displayed number is corrected, the correction is recorded in the change log with its date, the page it affects and what changed.