Modelling the general public's inflation expectations using the Michigan survey data

Abstract

In this article we discuss a few models developed to explain the general public's inflation expectations formation and provide some relevant estimation results. Furthermore, we suggest a simple Bayesian learning model which could explain the expectations formation process on the individual level. When the model is aggregated to the population level it could explain not only the mean values, but also the variance of the public's inflation expectations. The estimation results of the mean and variance equations seem to be consistent with the results of the questionnaire studies in which the respondents were asked to report their thoughts and opinions about inflation.

Notes

¹See for example Zarnowitz (1985), Bonham and Cohen (1995), Jeong and Maddala (1996) and Lloyd (1999).

²One may assume that the VAR forecasts are almost same as the forecasts of professionals made available to the public through news articles, but they cannot be directly compared since there is no cost to read those news articles.

³Data are available at http://www.phil.frb.org/econ

⁴Data are available at http://www.phil.frb.org/econ/forecast/readow.html

⁵Data are available at http://econweb.rutgers.edu/nswanson/realtime.htm

⁶Note that we do not assume the SPF nor inflation series to be exogenous. For example, even if the lagged values of the Michigan series do not help forecast the future SPF values, one should not take this as a sign of noncausality, since professional forecasters should use the information offered by consumer expectations when they form their forecasts. For example, they might expect that the high inflation expectations of the general public cause consumer inflation to rise.

⁷Also, note that none of the parameters γ ₂₁ ⁽ⁱ⁾had a posterior distribution deviating significantly from zero (the results were similar when we use mean series). We also estimated the error-correction model with the assumption that the SPF series is exogenous and the results looked similar.

⁸The model is of the form , where y _t is a vector of m variables, ψ is a vector of parameters Γ _i : s are parameter matrices and ε_t is the Normally distributed error vector with zero mean and Σ covariance. We use standard noninformative prior distribution

⁹To analyse the predicted variance of the learning model we need a longer sample period than the period, 1981/3 to 2004/1, for which the professionals' CPI inflation forecast series is available. To obtain a longer forecast period, 1970/1 to 2004/1, we used SPF's GDP deflator forecast series. We regressed the CPI inflation forecast series on the GDP deflator inflation forecast series and a constant and predicted the CPI inflation forecast series for the period 1970/1 to 1981/2 using the estimated regression model. The parameter estimates were 0.68 and 0.89 for the constant and the GDP deflator forecast series, respectively).