The Law Held From 2000 to 2006
Covered interest parity is the no-arbitrage condition the BIS once called the closest thing to a physical law in finance. A new series running back to 1963 says it has been broken most of the time, and that the quiet years everyone calibrated against were the exception.
Covered interest parity is the plainest arbitrage condition in international finance, and for most of my reading on the subject it has been described as the one relation that simply held. The trade is easy to picture. You have dollars. You can lend them for three months at the dollar rate. Or you can convert them to yen at today’s spot rate, lend the yen for three months at the yen rate, and in the same breath sell the yen proceeds forward for dollars at a price agreed now. Both routes start in dollars and end in dollars. Neither carries exchange-rate risk, because the forward contract fixes the price of the return journey before you set out. The two payoffs ought to be identical, and if they are not, the gap — the cross-currency basis — is money lying on the floor of one of the deepest markets on earth.
In September 2016 the BIS Quarterly Review described the condition as “the closest thing to a physical law in international finance”; its abstract said the relation “verges on a physical law”. The article was an obituary. Since 2008, the basis had refused to close, and a substantial literature had grown up to explain why: persistent demand for currency hedges on one side, and on the other, balance sheets too expensive to lend to an arbitrage. The shape of the story was always the same. Something broke in 2008.
A paper circulated this summer argues that nothing broke in 2008, because there was nothing in working order to break.
What the long record looks like
Olivier Accominotti, Jason Cen, David Chambers and Victor Degorce have assembled a daily series of spot rates, FX swap rates and money-market rates for 19 advanced-economy currencies running from 1963 to 2025, much of it hand-collected from historical newspapers. Two choices in the construction matter more than anything else. They measure against offshore Eurocurrency rates, which sidesteps the capital controls that made onshore comparisons meaningless for much of the twentieth century, and they use bid and ask quotes rather than midpoints, so that a deviation has to clear the cost of actually trading before it counts.
Their conclusion is that deviations from parity have been the ordinary state of affairs for six decades. Plot the three-month dollar basis for sterling from 1968 to 2025 and you get a line that is away from zero most of the time, on a chart the authors have to truncate at plus or minus 60 basis points to keep it readable. The years before the crisis look like the years after it. The exception — the only stretch in sixty-odd years where the condition holds tightly, where net deviations are both rarer and smaller — is roughly 2000 to 2006.
I could not read the 75 pages themselves; SSRN’s download refused me from where I was sitting, so what follows leans on the authors’ own summary and on the older literature they are arguing with. That gap matters most at exactly the point I would most like to check, which I will come back to.
Set the windows side by side and the shape of the problem is hard to miss.
Du, Tepper and Verdelhan’s paper is the canonical statement of the modern puzzle, and it is careful and quantitative: mean absolute three-month bases of 24 basis points over 2010–16, 27 at five years, a five-year yen basis near −90 basis points at the end of 2015, bases widest on the forwards that sit on banks’ balance sheets over quarter-end. Their three-month series begins in January 2000. Their evidence that parity held beforehand is that the basis was statistically indistinguishable from zero across 2000 to 2006 — which is to say, the pre-crisis baseline of the entire post-crisis literature is the seven-year window that the new dataset identifies as the anomaly.
A law tested only across the years it holds is not a law but a description of those years.
Where I think this is softer than it sounds
The claim that carries all the weight is not “deviations existed” but “deviations exceeded transaction costs”. Gross gaps in old data are unsurprising and nobody ever doubted them. Net gaps are the whole argument, and net gaps are a function of how wide you believe the historical bid-ask spread was.
This is not a new exercise. Jacob Frenkel and Richard Levich ran it in the mid-1970s, with the same offshore-rate trick, and reached the opposite answer: estimate the transaction costs, construct the neutral band they imply, and the apparent arbitrage profits of the tranquil 1960s disappear inside it. Turbulence widened both the costs and the band together. One of these two transaction-cost reconstructions is wrong, and from the outside I cannot tell which. A spread printed in a newspaper table in 1968 is an indicative quote, not a price anyone could deal on, and if the true cost of trading was higher than the printed spread suggests, net deviations are an artefact of the measurement rather than a feature of the market.
There is a second problem that old data cannot escape. Taylor found in 1989 that one-way arbitrage profits largely vanish once the spot and forward quotes are time-synchronised, and Akram, Rime and Sarno, running two million tick observations through eight months of 2004, found deviations that were genuinely there but lived for between thirty seconds and four minutes. A spot rate and a forward rate copied from the same newspaper page were not observed at the same instant, and a basis computed from quotes minutes or hours apart will show a gap whether or not one existed.
And a deviation is not the same thing as an available trade. For most of this sample the balance sheet required to run the arbitrage belonged to a bank, and banks were constrained — which is the paper’s own argument. The authors date substantial regulatory constraints on UK and US banks from the early 1960s to the mid-1990s, a loosening of about a decade, and a tightening after 2008. That makes the result less a refutation of the frictions literature than a considerable extension of it.
What survives my doubts about measurement is the structure in the series rather than its level. The basis moves with the interest-rate differential across currencies and across time, in the same direction in each. Across 265 episodes of speculative pressure on a currency, pressure toward devaluation lines up with the dollar basis widening and pressure toward revaluation with it narrowing. The deviations grow ahead of quarter-end reporting dates and during global credit contractions. Measurement noise does not arrange itself around bank reporting calendars or run the right way through currency crises, and that patterning is harder to explain away than any single number in the series.
Why I keep thinking about it
The part I find genuinely uncomfortable is the coincidence at the centre of it. Clean, high-frequency, electronically captured FX data begins around 2000 because that is when the feeds begin. The window in which covered interest parity held tightly also begins around 2000. The field took the beginning of its data for the beginning of normality, and built a decade of theory to explain a departure from a baseline that may be an artefact of where the tape starts.
I don’t think that makes the post-2008 work wrong. The frictions it identified are real, the quarter-end effects are measured, the hedging demand is documented. What moves is the question. If the condition has been broken for sixty years and briefly intact for seven, then the puzzle worth explaining is not the storm but the calm: what was true of 2000 to 2006 — a particular configuration of bank regulation, a global credit boom, abundant arbitrage capital — that let the law look like a law. That is a better question, and it is one the new data can actually be used on. It also carries a warning I would rather absorb cheaply: when a relation is first tested properly at the moment good instruments arrive, the instruments and the quiet may have arrived together, and nothing in the data will tell you so.
Sources
- Covered interest parity deviations and financial market frictions: Six decades of evidence — Accominotti, Cen, Chambers & Degorce, CEPR VoxEU, 9 Oct 2026
- Covered Interest Parity: The Long Run Evidence — Accominotti, Cen, Chambers & Degorce, CEPR DP 21820 (abstract)
- Deviations from Covered Interest Rate Parity — Du, Tepper & Verdelhan, NBER Working Paper 23170
- Covered interest parity lost: understanding the cross-currency basis — Borio, McCauley, McGuire & Sushko, BIS Quarterly Review, September 2016
- Discussion of covered interest parity research (Frenkel & Levich, Taylor, Akram–Rime–Sarno) — Levich, NBER Summer Institute 2009
- Transaction Costs and Interest Arbitrage: Tranquil Versus Turbulent Periods — Frenkel & Levich, Federal Reserve IFDP No. 87, August 1976