Cheryl Schonhardt-Bailey on "Estimating Central Bank Preferences" by Nicole Baerg, Will Lowe, Simone Ponzetto, Heiner Stuckenschmidt, and Cäcilia Zirn:

One important strength of this paper is that it builds from an existing theoretical framework for understanding the preferences of decision making by central bank officials—i.e., the Taylor Rule. The Taylor Rule seeks to estimate the interest rate based upon an equation which includes terms for both expected inflation and the output gap. Conceptually, this has proven attractive, but invariably the problem with Taylor Rule models is that none of the variables are easily measurable.

Similar to its usage elsewhere, the Taylor Rule is set out here as a trade off between the output gap and inflation. The authors then move beyond this to imply that when a member of the FOMC talks more about output than inflation, the member can then be seen as relatively more dovish (and vice versa). The difficulty is that the approach taken here does not provide us with confidence that this implication closely resembles the sentiments of FOMC members. In short, there is a leap of faith in saying that if someone talks about inflation, he/she is hawkish. The words themselves do not provide insight into what the member in question thinks about the underlying trade off.

Moreover, the underlying hawk/dove dimension may be a reasonable approximation when it comes to voting behaviour (where voting itself is a constraint). However, in textual data, when members can express any number of arguments on the relationship between expected inflation and the output gap, the simplification of the Taylor Rule is more suspect.

The approach is, however, promising in that it enables the authors to arrange preferences of members in a single chart (Figure 1). This is clearly attractive in its simplicity and clarity. My key recommendation here is the following. First, get rid of the Fed staff (none of whom have any vote on the committee), since this is just muddying the picture for Figure 1. Second, make very clear that your approach is assuming that inflation words signify hawks and output words signify doves (perhaps an explicit test could be done here?). Third, use the Chappell, et al methodology (Chappell, McGregor et al. 2005) to estimate the reaction functions of FOMC members based upon their voting behaviour, and then compare these results with the results from this textual analysis. That would be a great way to investigate whether what members say in the meeting is reflected in their voting behaviour.

Finally, I agree that this study extends the work on FOMC transcripts in my book (Deliberating American Monetary Policy), but the authors may want to correct where they say that Alceste found 8 topics for the 1979-99 period. Actually, the analysis identified 9 broad themes, and in total, 21 discrete themes (the latter falling into sub-themes…see Table 3.3 in the book). Notably, Alceste did not come up with the nine themes. Rather, these are (as in the case of the present paper) informed by expert knowledge, as a means to group the classes identified by Alceste (but labelled by researchers, using expert knowledge). Lastly, while Alceste is proprietary software (as, too, is a lot of econometric software), the free version (R based) is available in Iramuteq.

Chappell, H. W., R. R. McGregor, et al. (2005). Committee Decisions on Monetary Policy: Evidence from Historical Records of the Federal Open Market Committee. Cambridge, MA, MIT Press.