The Fisher Effect Across 123 Economies, 2000–2022: Averaging, Errors in Variables, and Why It Holds Across Countries but Not Within Them

The Fisher hypothesis says that nominal interest rates move one for one with expected inflation. I test it on World Bank data for bank lending rates (118 economies) and deposit rates (123 economies) against consumer price inflation, using country averages for 2000–2022. In the full samples the least-squares slopes are 0.854 for lending rates and 0.455 for deposit rates, pulled down by a few high-inflation economies; robust and median regressions give 0.933 to 1.046 and 0.549 to 0.597. Restricting to average inflation below 10%, the slopes are 1.883 (lending) and 0.930 (deposit). I prove that averaging over T years shrinks the errors-in-variables bias from unexpected inflation by a factor of about T(1−ρ)/(1+ρ) for AR(1) inflation surprises, and I use it to bound the attenuation: for economies with average inflation below 20%, the factor is at least 0.718 for lending rates and 0.713 for deposit rates, so the true deposit-rate slope lies between 0.769 and 1.080. A Monte Carlo study reproduces the theory (0.979 against 0.978). Within countries the Fisher effect is absent at annual frequency: the median time-series slope is 0.110 for lending rates and 0.124 for deposit rates. India's lending rate has a slope of 0.494 on current inflation over 1978–2022, and the ex-post real lending rate ranged from −3.66% in 2010 to 10.98% in 1978.