Working paper

State-dependent Central Bank Communication with Heterogeneous Beliefs

Published on 2 June 2022
Authors : Sylvérie Herbert

Working Paper Series no. 875. This paper studies how state-contingent central bank communication can improve welfare when externalities are at play. In the model, a central banker (CB) wants to influence the private sector beliefs, which are heterogeneous, to generate an upward bias in their action. The CB can engender such welfare-improving bias by providing public information, choosing a signalling strategy that is a function of fundamentals. To study this optimal communication strategy, I introduce heterogeneous priors in an otherwise standard Bayesian persuasion model à la Gentzkow and Kamenica (2011) and characterize the dependence of optimal disclosure on the heterogeneity of beliefs. I show that heterogeneity matters in two ways: (i) it is optimal to send moderating signals, which implies sending signals with positive error probabilities in both states, and constitutes a non-trivial departure from the homogeneous beliefs case; (ii) higher dispersion in beliefs leads the monetary authority to send signals with lower error probabilities. I apply my framework to a central bank communication problem in which the policy maker communicates about aggregate conditions to influence firms' investment decisions in presence of investment externalities. I empirically validate the model's predictions by showing that the FOMC unemployment rate forecasts are systematically biased in opposite directions in recessions and expansions. Also in line with the model's predictions, the forecast biases are decreasing in the degree of private sector disagreement for each state.

Image Distribution of FOMC unemployment rate forecast errors (posterior distribution)

Should central banks always reveal with full transparency the state of the economy to steer private sector expectations? Or should their communication strategy rather depend on the state of the economy? How to persuade a heterogeneous audience?  There may be cases where central banks want to use communication strategically to steer expectations, especially if the economy is hit by an inefficient or temporary cost-push shock, or to correct the coordination failure of the private sector in presence of externalities.  Christine Lagarde, in a recent press conference on September 9th, 2021 talked inflation down by emphasizing its temporality: “inflation increased to 3.0 per cent in August [...] This temporary upswing in inflation [...] The new staff projections foresee annual inflation at 2.2 per cent in 2021, 1.7 per cent in 2022 and 1.5 per cent in 2023...” In a similar manner, Ben Bernanke (Fed), at the onset of the Great Recession, talked the economy up by saying that he “wasn’t willing to use the r-word in public at that point, even though the risk of a downturn was clearly significant. [...] [He] didn't want to add unnecessarily to the prevailing gloom by talking down the economy” (Bernanke in “The courage to Act: a memoire of the crisis and its aftermath”).

This paper studies these questions by modelling how a central bank can improve welfare by designing a communication strategy that is state-contingent (i.e., that varies according to the state of the economy) and a function of disagreement in expectations, when externalities are at play (for instance, investment complementarities, or when there is an inefficient shock hitting the economy such as an inflationary cost-push shock). To do so, I introduce heterogeneous priors in a model of Bayesian persuasion à la Gentzkow and Kamenica (2011) to study how to persuade rational agents by controlling their informational environment as a function of the state, and how the heterogeneity in beliefs influences this communication strategy.

The model yields three policy recommendations: some opacity in both states (e.g., recession, expansion) is optimal such that communication is going to play a moderating role  (this means the CB should be sometimes unduly optimistic in bad states, i.e., sending a good signal in a bad state with positive probability, and pessimistic in good states, i.e., sending a bad signal in a good state with positive probability); complete transparency is never optimal; (iii) more prior disagreement in the economy leads to greater transparency.

Using one example of communication, the FOMC unemployment rate forecasts, this would mean over-predicting unemployment in expansion, but under-predicting it in recessions, but less so the higher the disagreement of the private sector, as figure 1 illustrates. I show that empirically, they are systematically biased in opposite directions in recessions and expansions, in line with the policy recommendations. Also in line with the model’s recommendations, the forecasts’ biases are decreasing in the degree of private sector disagreement in recessions and expansions. A similar exercise on the ECB Staff’s forecasts suggest no state dependency of the forecasts’ biases.

Updated on 23 October 2023