RDP 9803: Forward-Looking Behaviour and Credibility: Some Evidence and Implications for Policy Appendix A: A Test for the Rationality of Inflation Expectations
February 1998
- Download the Paper 344KB
In this appendix we formally test for rationality. The standard test for rationality of inflation expectations involves regressing actual inflation on what inflation is expected to be for that period:
If the survey measure is an unbiased predictor of actual inflation, then γ = 0 and δ = 1. The broad conclusion from the literature is that surveys of inflation expectations are usually not consistent with rational expectations (Roberts 1997). Table A1 presents the results of estimating Equation (A1) for the MMS measure of financial market economists' quarterly headline inflation expectations and the Melbourne Institute's households' annual inflation expectations.[16]
| Financial Market Economists (The dependent variable is quarterly headline inflation) |
Households (The dependent variable is annual underlying inflation) |
|
|---|---|---|
| Constant (g) | −0.21 (0.08) |
−2.06 (0.56) |
| Expected inflation (d) | 1.17 (0.06) |
0.99 (0.09) |
| R-bar-squared | 0.89 | 0.71 |
| Test that γ = 0 , δ = 1 | F(2,47) = 3.9 [0.03] | Chi-square(2) = 97.1 [0.00] |
Notes: Standard errors are reported in ( ) and marginal significance is reported in [ ]. Since the Melbourne Institute expected inflation series is an annual rate, estimating a regression on a quarterly frequency induces an MA(3) process in the residual. The standard errors and exclusion tests are Newey-West (1987) adjusted using the Robusterrors option in RATS (and the exclusion statistic in this case is a Chi-square statistic). |
||
In both cases the null hypothesis of rational expectations is rejected, although the rejection is clearly stronger for households' expectations than for financial market economists' expectations.
Footnote
The Melbourne Institute survey measures inflation expected over the coming year, and so in this case annual inflation is regressed on the survey measure one year earlier. [16]