The Forward Premium Puzzle in Ten Currencies, 1979-2026: Exact Variance Bounds, Carry Returns and a Sign Change after 2008

Uncovered interest parity predicts that a currency with higher interest rates should depreciate by the interest differential. I test it for ten currencies against the US dollar at a three-month horizon, with month-end spot rates and three-month interbank rates from FRED, from 1979 (or the currency's first available month) to mid-2026. The Fama slope of the three-month change in the dollar value of the currency on the interest differential is negative for 8 of 10 currencies and below 0.5 for all of them; the mean slope is minus 0.303 and the median minus 0.247. Parity (slope one) is rejected at the 5% level for five currencies. I prove that the variance of the risk premium minus the variance of expected depreciation equals one minus twice the slope, times the variance of the differential, so any slope below one half implies a premium more variable than expected depreciation; the implied ratio ranges from 0.10 for the yen to 3.41 for sterling. A pooled within-currency slope is minus 0.335 (standard error 0.284) over the full sample, minus 0.662 (0.320) for 1979 to 2007, but plus 1.566 (0.920) for 2008 to 2026, consistent with parity. Across currencies, those with higher interest rates than the dollar depreciated more on average (slope of mean appreciation on the mean differential 0.440, standard error 0.185). A quarterly portfolio long the three highest-yield and short the three lowest-yield currencies earns 3.13% a year with a Sharpe ratio of 0.41 and skewness of minus 1.43; its worst quarter, minus 21.2%, began in July 2008.