Agenda

Gianni Amisano (Federal Reserve Board) - How to catch a star? Reliability of filtering estimates in linear state space systems (with Thorsten Drautzburg and Andrea Stella)

Abstract:

In macroeconomic policy analysis, estimates of latent variables such as the natural rate of interest and the output gap are important inputs for policymakers. We develop reliability measures to assess the quality of estimates of latent variables in linear state space models that go beyond describing the level of uncertainty surrounding these estimates. Motivated by the (infeasible) population regression of the true state at time t+h on time t information and the time t+h forecast, we propose the (conditional) correlation and the associated R2 as intuitive measures to assess the predictability of the latent states. We relate this measure to the posterior variance of the state estimate. To help understand the source of the posterior uncertainty we propose the half-life and the contribution of the initial state to the conditional forecast error variance as additional diagnostics. Models that estimate the level of the hidden state very imprecisely may still imply predictability at relevant horizons. The Laubach-Williams model features high state uncertainty, but reasonable predictability at medium-run horizons. Since it is not possible to assess whether the models accurately quantify the uncertainty surrounding the latent variables, researchers should evaluate the predictive accuracy of forecasts for observables.

The seminar can be attended also remotely, connecting to ZOOM.

Link Zoom: bit.ly/insem-2425
ID riunione:  880 2639 9452
Passcode: InSem-2425