Key takeaways
- Modified duration is a percentage, not a date. The US Comptroller of the Currency defines it as the approximate price change for a 100 basis point move in yield.
- A 30-year Treasury issued at 1.90% in December 2021 had a modified duration of 22.79. Duration said 2022 cost 47.17% of its price. Exact repricing says 36.11%.
- In the SEC's own worked example, a 100bp fall in rates added $82 to a $1,000 bond and a 100bp rise took $75 away. Duration predicts $78.36 in both directions.
- Three US Treasury ETFs with stated durations of 1.87, 6.95 and 15.31 years returned -3.90%, -15.23% and -31.41% in calendar 2022. The ranking held exactly.
- UK index-linked gilts fell 34.55% in 2022 against 23.83% for conventional gilts, because their duration is 13.08 years against 7.21.
Duration is a percentage wearing the costume of a date
Your bond fund's factsheet says "effective duration: 6.95 yrs" and you want to know what the years have to do with anything. Short answer: almost nothing you'd guess. Duration is not how long you're locked in. It's a price-sensitivity number, and the years are a historical accident of how it's calculated.
The Office of the Comptroller of the Currency, the US regulator that supervises national banks, puts it exactly in its Interest Rate Risk handbook. Modified duration is the "approximate percentage change in a bond's price for a 100 basis point change in yield, assuming that the bond's expected cash flow does not change when the yield changes."
So a duration of 6.95 means that if yields rise by 1 point, the price falls about 6.95%. FINRA says the same thing without the regulator's hedging: "if a bond has a duration of 10 and interest rates increase by 1 percentage point, then that bond's price would be expected to decline by approximately 10 percent." Rates fall instead, and it works the other way.
Two things set the number. A coupon is the fixed interest a bond pays — the UK Debt Management Office's pricing formula sets the number of coupons a year at 2 "for all current gilts outstanding", so a 4% gilt pays £2 twice a year per £100 of face value. FINRA's rule: "the higher the coupon rate, the lower the duration; the longer the maturity, the higher the duration." Big coupons return your money sooner, so less of the price depends on distant cash flows.
The SEC's own example contains the error the rule of thumb makes
The SEC's investor bulletin on interest rate risk works a single bond twice, and the two answers don't mirror each other. Take a Treasury with a 3% coupon paid semi-annually, $1,000 of face value and 9 years left to run, priced at par because its yield is also 3%.
Market rates fall to 2%, and the bulletin puts the price at $1,082. Market rates rise to 4%, and it puts the price at $925. Same bond, same 100bp shock, opposite directions.
Now run the duration rule on it. Discount the 18 remaining half-yearly cash flows, weight each by the time you wait for it, and you get a modified duration of 7.84. The rule of thumb says the price moves 7.84% either way — $78.36 up or $78.36 down.
It gained $82 and it lost $75. The rule was wrong twice, in the same direction each time: it understated the gain and it overstated the loss. That's not a rounding artefact in the SEC's table. It's the shape of the price-yield relationship, and it has a name.
The Comptroller's glossary defines convexity as "the rate at which duration changes with respect to changes in rates." Duration is the slope of the price-yield curve at one point. Prices don't sit on a straight line, they sit on a curve that bends upwards. Extend the tangent far enough in either direction and it drifts below the curve — under-forecasting gains, over-forecasting losses.
Price a 30-year Treasury two ways and the gap is 11.1 percentage points
At a 100bp move the error is small enough to ignore. At the size of move 2022 delivered, it isn't.
Start with figures from the US Treasury's daily par yield curve. On 31 December 2021 the 30-year yield was 1.90%. On 30 December 2022 it was 3.97%. That's a move of 207 basis points, or 2.07 points of yield.
Build the bond the Treasury was selling at the start of that year: $100 of face value, a 1.90% coupon paid twice a year, 30 years to maturity, priced at par. Its modified duration is 22.79 — long, because a 1.90% coupon returns almost nothing before the principal arrives.
Duration's answer is one multiplication. 22.79 × 2.07 = 47.17, so the rule of thumb says the bond lost 47.17% of its value.
The arithmetic answer takes longer and is the one that matters. Discount all 60 remaining half-yearly cash flows at 3.97% instead of 1.90%, the way the Debt Management Office's price/yield formula does it, and the bond is worth $63.89. That's a fall of 36.11%.
Duration overstated the loss by 11.1 percentage points. On a $100,000 position that's an $11,060 difference between the estimate and the answer.
The reason is visible if you recompute the duration at the new yield: 20.45, not 22.79. As yields rose, the bond's own sensitivity shrank. Duration was accurate at the moment you measured it and stale by the time the move finished. Adding the second-order convexity term gets you to a 33.78% loss, which now overshoots the other way. Even the correction is an approximation once the move is this big.
2022 tested the ladder, and the ranking held perfectly
None of the above is a claim about what any fund did. So here's what happened to three US Treasury ETFs that differ mainly in duration, from their June 2026 factsheets.
The iShares 1-3 Year Treasury Bond ETF, ticker SHY, carries an effective duration of 1.87 years. It returned -3.90% in 2022. The 7-10 year fund, IEF, carries 6.95 years and returned -15.23%. The 20+ year fund, TLT, carries 15.31 years and returned -31.41%. The chart above puts those alongside two gilt funds.
Multiply each duration by the move in the matching part of the Treasury curve and you get a rough forecast. The 2-year yield went from 0.73% to 4.41%, a rise of 368 basis points. The 10-year went 1.52% to 3.88%, up 236. The 30-year rose 207. That gives predicted losses of 6.88%, 16.40% and 31.69% against actual losses of 3.90%, 15.23% and 31.41%.
Two honest caveats before you lean on that. Those durations are the June 2026 numbers, not the December 2021 numbers, and every one of these funds had a longer duration at the start of 2022, when yields were lower. The 30-year calculation above shows why: the same bond ran 22.79 at 1.90% and 20.45 at 3.97%. Measure at the old yields and the forecasts get worse, not better.
The second caveat is that 2022 wasn't one move. The 2-year yield rose 368 basis points and the 30-year rose 207. Duration answers a question about a parallel shift in the whole curve, and the curve doesn't shift in parallel. It's the same problem the yield curve's bear flattening and subsequent steepening makes visible at the front end.
Convexity sits where the duration sits
The factsheets report convexity too, and the pattern explains where the approximation is safe. SHY reports 0.05. IEF reports 0.58. TLT reports 3.27.
On a short fund the curvature is close to nothing, so the straight-line estimate is fine to the nearest tenth of a percent. On a long fund it's the difference between a usable estimate and a badly wrong one. Convexity is reported on different scales by different providers, so it's a within-family comparison, not a cross-provider one.
That is the practical shape of it. If your bond exposure is short, the duration number is the whole story. If you hold 20-year paper, duration is a first draft.
UK index-linked gilts fell harder in the year inflation actually arrived
The gilt market ran the same experiment with a twist worth sitting with.
The iShares Core UK Gilts UCITS ETF, IGLT, tracks conventional gilts across all maturities and reports an effective duration of 7.21 years. It returned -23.83% in 2022. The index-linked equivalent, INXG, reports 13.08 years and returned -34.55%.
Index-linked gilts pay a coupon and a principal that rise with the retail prices index. In 2022, UK inflation was doing precisely the thing that linkage exists to protect against. The protected fund still lost 10.72 points more than the unprotected one, because the linker index is stuffed with very long maturities and duration overwhelmed the uplift.
The speed matters as much as the size. A Bank of England staff working paper on the 2022 gilt market crisis records that "in a matter of days, nominal yields across the medium and long maturity spectrum rose by more than 100 bps, which had been unprecedented in recent economic history of the UK." A duration number tells you what a move costs. It says nothing about whether the move can happen in a week.
The strongest objection: duration measures a price you may never realise
There's a serious case that all of this is beside the point, and FINRA makes it in the same article as the rule of thumb. "If you're a 'buy and hold to maturity' bond investor, interest rate changes might have little or no direct impact on your fixed income assets."
That's right, and it's the best argument against treating duration as a risk measure. Hold one gilt to maturity and you receive the coupons and the face value whatever the price did in between. The 36.11% mark-to-market fall on that 30-year Treasury was never a loss for someone who wasn't selling.
The objection is weaker for a fund, which has no maturity date and keeps rolling into new bonds. It isn't empty, though, because the same repricing that cut the capital raised the income. TLT's 30-day SEC yield was 4.89% on 30 June 2026, against 4.29% for IEF and 4.00% for SHY. Over the three years to that date SHY annualised 4.23% and IEF 2.95%, while TLT was still at -1.65%.
Read that as the trade duration actually offers. You get more price damage per point of yield, and you get compensated over a horizon roughly as long as the duration itself. Whether that horizon fits the money is a question about the money, not about the bond. The same tension shows up in how much of a portfolio's risk is drawdown rather than standard deviation.
The Comptroller's handbook makes a second objection, aimed at anyone who thinks one shock number is enough. It warns that "static interest rate shocks consisting of parallel shifts in the yield curve of plus or minus 200 basis points may not be sufficient to adequately assess IRR exposure."
What these numbers cannot tell you
The limitations here are specific, and they run in both directions.
The fund durations are dated 30 June 2026. Applying them to 2022 is a rough test, not a clean one, and the direction of the error is known — real durations were longer then. Nothing in this piece is a reconstruction of what those funds actually held on 31 December 2021.
The 30-year calculation holds maturity constant across the year. A real bond aged 12 months and rolled down a sloping curve, which changes the answer by an amount this arithmetic doesn't capture. It also ignores the coupons received, which is why a fund's total return is less bad than its price return.
Effective duration is not modified duration. BlackRock's factsheets state that its "options-based duration model employs certain assumptions and may differ from other fund complexes." Two funds holding similar bonds can report different durations because two models disagree.
And duration only measures sensitivity to the risk-free rate. It says nothing about credit spreads, which is why a corporate bond fund can lose money in a year when government yields fall. The sample here is one rate cycle in two countries. One cycle is not a distribution, and a year in which yields rise 207 basis points is not a forecast of the next one.
What would change the conclusion
If the moves stay small, none of this matters. At the SEC's 100bp shock the straight-line estimate was inside $4 on $1,000, which is noise against the coupon. The case for arithmetic over approximation is entirely a case about large moves.
If yields fall from here, the same curvature works for you rather than against you. The 30-year bond that lost 36.11% instead of duration's 47.17% would, symmetrically, gain more than duration predicts on the way back. Convexity is not a warning label. It's an asymmetry, and it points the same way in both directions.
If the curve moves the way 2022's did, with the 2-year up 368 basis points and the 30-year up 207, a single portfolio duration stops being the right instrument. Fund factsheets report one number because a portfolio has one number to report, not because one number is sufficient. Where the yield sits along the curve does the rest of the work, which is the argument in real yields and discount rates.
The number to watch is on your own factsheet, next to the words "effective duration", and it changes as yields change. LedgerTouch tracks that alongside the rest of a portfolio. The interesting exercise is to multiply it by 2.07 — the move the 30-year actually made in a single year — and decide whether you'd have stayed put for that.