Learning Trend Inflation – Can Signal Extraction

It can be shown that inflation expectations and associated forecast errors are character-ized by a high degree of persistence. One reason may be that forecasters cannot directly observe the inflation target pursued by the central bank and, hence, face a complicated forecasting problem. In particular, they have to infer whether the observed movement of the inflation rate is due to a permanent change of policy parameters or whether it is the result of a transient shock. Consequently, it is assumed that agents behave like econo-metricians who filter noisy information by estimating an unobserved components model. This constitutes the trend learning algorithm employed by the forecaster. To examine whether this is a valid assumption, I fit a simple learning model to US survey expectations. The second part contains an out-of-sample forecasting experiment which shows that learning by signal extraction matches survey measures closer than other stan-dard models. Moreover, it turns out that a weighted average of different expectation formation processes with a prominent role for signal extraction behaviour is well suited to explain survey measures of inflation expectations