RDP 2010-02: Learning in an Estimated Small Open Economy Model 6. Conclusion

Rational expectations models assume that economic agents are perfectly informed about the structure of the economy. In this paper we relax this assumption and estimate the effect of learning on the propagation mechanism in a small open economy model. When private agents learn about the economy it is reasonable to assume that they form expectations of macroeconomic variables using statistical forecasting models, which are continuously re-estimated as new data become available. Our results show that learning does enhance the empirical fit of a small open economy model for Australia. Milani (2007) claims that learning is a replacement for the standard ad hoc sources of structural inertia such as price indexation and habit formation in consumption in a stylised closed economy model. However, we find that learning complements rather than replaces these structural features. This is consistent with Slobodyan and Wouters (2009) and Murray (2008) who analyse the effects of learning in relatively large closed economy settings.

Unlike these two papers, however, we show that learning results in impulse response functions that are consistent with those seen in more data-driven models. This is particularly noticeable for the response of the real exchange rate. However, a very persistent risk premium shock must still be added to the uncovered interest rate parity condition in order to fit the real exchange rate.

We also find that since the adoption of inflation targeting, agents appear to be using a longer history of data to form their expectations, consistent with more stable inflation and interest rates.

Although there are still many aspects of learning that require further study, our results suggest that the incorporation of learning into standard structural models warrants further investigation.