Academic journal article Federal Reserve Bank of Atlanta, Working Paper Series

The Usefulness of the Median CPI in Bayesian VARs Used for Macroeconomic Forecasting and Policy

Academic journal article Federal Reserve Bank of Atlanta, Working Paper Series

The Usefulness of the Median CPI in Bayesian VARs Used for Macroeconomic Forecasting and Policy

Article excerpt

1 Introduction

A key task for economic forecasters and monetary policymakers is to parse through the incoming data, separate signal from noise, and use these data to make judgments over the likely path of the economy heading forward. In forecasting inflation, for example, many studies in the core literature have shown trimmed-mean inflation statistics to be useful indicators. However, evaluating the usefulness of trimmed-means is almost universally performed in simple univariate and single-equation forecasting applications.

While it may be appropriate in some settings to separately forecast one or two variables of interest, a monetary policy setting requires consistent forecasts of multiple variables. The Federal Open Market Committee (FOMC), for example, requires each member to submit forecasts of real GDP growth, the unemployment rate, headline and core inflation, and the federal funds rate four times a year for publication in their Summary Economic Projections (SEP) materials. (1,2)

In this paper, we evaluate the usefulness of trimmed-mean inflation statistics in a class of multivariate models often used for forecasting and policy analysis--Bayesian Vector Autoregressions (BVARs). We are particularly interested in whether using the median CPI as the underlying inflation measure in the BVAR system leads to any appreciable differences in forecast accuracy of important macro variables (real GDP, inflation, fed funds rate, and the unemployment rate). To the best of our knowledge, the performance of trimmed-mean inflation statistics and their influence on other macro variables has to be investigated in a BVAR setting.

Trimmed-mean inflation statistics were first investigated by Bryan and Pike (1991) and then more formally by Bryan, Cecchetti, et al (1994, 1997). These measures of underlying inflation uncover the inflation signal by stripping away the most volatile monthly relative price swings. These measures are much more systematic in the removal of relative price changes than exclusionary indexes like the ex food and energy "core" CPI. Exclusionary indexes, by design, implicitly assume that price changes in every component other than those they exclude are inflation signal. So, for example, if tobacco prices were to spike in a given month because of an excise tax increase, the core CPI would treat this as signal, whereas trimmed-mean inflation statistics would remove this relative price shock.

There is a fairly sizeable literature on the usefulness of trimmed-mean inflation statistics apart from Bryan and Cecchetti. Smith (2004), using both conditional and unconditional forecasting models, finds that the weighted median CPI outperforms the core CPI. Clark (2001) finds that the 16 percent trimmed-mean CPI and the CPI ex energy are better univariate forecasters than the core CPI or underlying inflation index that excludes the 8 most volatile CPI components. Meyer and Pasaogullari (2010) find the median and the 16% trimmed-mean CPI forecast year-ahead headline inflation about as well as inflation expectations do, and outperform simple forecasting models. Crone, Khettry, Mester, and Novak (2013) found that over longer-horizons (i.e. 24-months and longer), the median CPI yields a forecast significantly superior to that of the headline or ex food and energy CPI index.

Others, such Dolmas (2005) and Detmeister (2011) have investigated the use of trimmed-mean inflation statistics using Personal Consumption Expenditures Price Index (PCE) data. Dolmas (2005) finds that an optimally selected asymmetric trim tracked inflation much more closely than the ex food and energy ("core") PCE price index. And, Detmeister (2011) finds that trimmed-mean measures outperform exclusionary indexes (like the core PCE) in tracking the ex-post inflation trend and forecasting future inflation.

Stock and Watson (2008) engage in a comprehensive inflation forecasting exercise, performing a total of 192 forecasting procedures. …

Search by... Author
Show... All Results Primary Sources Peer-reviewed

Oops!

An unknown error has occurred. Please click the button below to reload the page. If the problem persists, please try again in a little while.