Alastair Firrell and Kate Reinold
The right stance for monetary policy is highly uncertain, and so it is no surprise that members of monetary policy committees – like the Bank of England’s Monetary Policy Committee (MPC) – regularly disagree about the best course of action. Asking a committee to decide allows different opinions to be aired and challenged, with a majority vote needed to determine policy. But how should we expect those disagreements and votes to change in periods of higher uncertainty? Should we expect more 9–0 unanimous votes? Or more 5–4 close contests? We address these questions in this post and find that the degree of disagreement is little changed in periods of high uncertainty, and nor are dissenting votes. There is, however, some difference in how voting decisions are formed when uncertain, with both individual and committee-wide views having less explanatory power for votes.
One key source of disagreement among policymakers is their views on the outlook for the economy. We would like to be able to analyse those views, but we cannot observe them directly, like we can with votes. MPC minutes include discussions and comments, but these are not attributed – we can’t tell who said what. But MPC members are individually accountable for their decisions and so regularly deliver speeches explaining their thoughts and opinions. In our recent Staff Working Paper, we run a text analysis of these speeches to derive estimates of individual MPC member views about the economy. We then use these – along with collective committee views as measured from minutes – in an analysis of how MPC member votes differ in periods of higher uncertainty, corroborating the lessons from a small voting model. While we focus on exploring the relationship for the Bank of England’s MPC, a similar analysis could be applied to other monetary policy committees internationally.
Extracting views from speeches and minutes using textual analysis
To extract views on the economy from speeches and minutes, we identify moments when MPC members are discussing economic conditions, and their sentiment when doing so: are the speakers talking about factors which would warrant a tightening of policy, or loosening – a.k.a. hawkish or dovish? To do this we build up a dictionary of economic terms combined with directional terms to give us search phrases like ‘higher inflation’, ‘unemployment falling’, and ‘reduced spare capacity’. Each of these is classified as hawkish or dovish. Taking each speech and minute in turn, we count instances of these phrases and calculate an overall net hawkishness index on economic topics. This technique builds on work by Apel and Blix Grimaldi (2014), Malmendier et al (2017) and Picault and Renault (2017) in their analyses of Sveriges Riksbank minutes, FOMC speeches and ECB press conferences respectively.
There are many nuances of language we have to deal with to clean the texts so we can robustly locate these phrases of interest. Stemming words to deal with variation (increases, increasing, increased…), handling interposing words (‘inflation has been increasing’) and clause/sentence boundaries, removing sentences with negatives, and so on. By carefully pre-processing the text in this way, we can more accurately identify instances of hawkish and dovish sentiment on economic topics, even among speeches which are much more varied in topic, style and structure than the more consistent minutes.
Views vs votes, under normality and uncertainty
Having established our measures of individual and committee views about the economy we can explore the relationship with votes, in normal and uncertain times.
Chart 1: Economic sentiment from minutes and speeches and MPC votes comove
Notes: Votes are classified as looser, unchanged or tighter (-1, 0, 1) and an average is taken across members and across quarters. A positive (negative) sentiment index implies more hawkish (dovish) discussion of economic conditions.
Chart 1 shows average sentiment from minutes and speeches against the average policy vote by quarter (where votes have been categorised as -1, 0 and 1, for loosening, no change, and tightening). There is a strong comovement, indicating that when MPC members speak more hawkishly they are also likely to vote for tightening. But there is also a fair dispersion in views.
We use these measures of individual and committee sentiment in ordered probit regressions with individual MPC member votes as the dependent variable. We allow for an interaction between sentiment and an uncertainty indicator coded as ‘high’ when either the options-implied volatility of the FTSE is high, or there is high disagreement between Consensus forecasters about the outlook for GDP. We are by no means the first to look at the determinants of MPC member votes: see for example Besley et al (2008), Berk et al (2010) and Harris et al (2010). However, by constructing a measure of individual views we are able to account for a key missing variable that several papers have ascribed importance to (Gerlach-Kristen (2006), Bhatacharjee and Holly (2005) and Hansen et al (2014)).
Our analysis yields a number of results that we compare to the predictions from a small model of committee voting with noisy signals about the state of the economy (similar to Gerlach-Kristenn (2008)). In this modelling setup, policymakers weight their own signals and those of their colleagues according to how informative they are thought to be about the true state of the economy (using a Kalman filter). This means that as the economy becomes more volatile, or noisier to measure, the relative weight that different information receives adjusts.
Findings
We have three main findings. First, both individual views of the state of the economy and those of the rest of the MPC are important in explaining individual votes. This is consistent with the predictions from our model. Where the true state of the economy is only observed with noise, the views of colleagues contain useful information that can inform votes.
Second, both individual views and committee-wide views are down-weighted in voting decisions in periods of higher uncertainty. This is consistent with MPC members placing less weight on views about which they are less confident. The down-weighting of individual views seems to be greater and more robust to specification than the down-weighting of the views of the broader committee.
Finally, while individual views of the state of the economy become more volatile in periods of higher uncertainty, they do not become more dispersed across committee members. We find no increase in the dissent rates in periods of high uncertainty (Chart 2), and this would be predicted by our model: when the economy becomes more volatile, so do estimates of how the economy is evolving. But with no MPC member better able to deal with this volatility than another, the dispersion in views is unchanged, so dissenting votes also remain unchanged.
Chart 2: No clear relationship between the share of dissenting votes and uncertainty
Notes: The dissent rate is computed as the share of votes against the majority decision at each meeting. The rate is averaged across the quarter.
We assume that views are formed in a way that tries to get as close as possible to the truth. We cannot rule out, however, that there is some strategic behaviour under uncertainty. For example, if it is better to be conventionally wrong than unconventionally right, MPC members could downweight their own views by more than would be optimal (see work by Sibert (2006) and Baddeley et al (2007)). However, studies have also noted the potential for the opposite – anti-herding – as policymakers attempt to stand out from the crowd.
Our findings are important because the outside world often looks to the MPC’s dissent rate as an indicator of how uncertain particular decisions are. Our paper shows that in practice the relationship between uncertainty and voting is more nuanced than that. It also reveals that speeches contain useful information about individual MPC member views. Some argue that that too much individualistic communication adds up to a ‘cacophony of voices’ that may have less power, but we show that this should be weighed against the accountability benefits of using speeches to convey information that votes alone do not.
Alastair Firrell works in the Bank’s Advanced Analytics Division and Kate Reinold works in the Bank’s Monetary Policy Outlook Division.
If you want to get in touch, please email us at bankunderground@bankofengland.co.uk or leave a comment below.
Comments will only appear once approved by a moderator, and are only published where a full name is supplied. Bank Underground is a blog for Bank of England staff to share views that challenge — or support — prevailing policy orthodoxies. The views expressed here are those of the authors, and are not necessarily those of the Bank of England, or its policy committees.
