Fahmy and Kandil [2003] use a cointegration approach to test for the Fisher effect in US interest rates during the 1980s and early 1990s. Here, I argue that even if nominal interest rates and inflation rates do obey integrated processes, cointegration of these two processes is not a sufficient condition for the Fisher effect to hold as it is consistent with any theory implying a stationary real interest rate. As the Fisher effect ex post implies that nominal interest rates embody an optimal inflation forecast, the sufficient condition for it to hold is the unpredictability of the implied inflation forecast error. This condition may be tested using the signal extraction framework of Durlauf and Hall [1988, 1989].
