Does Real‐Time Macroeconomic Information Help to Predict Interest Rates?

• Published date: 01 December 2023
• Author: ALBERTO CARUSO,LAURA CORONEO
• Date: 01 December 2023
• DOI: http://doi.org/10.1111/jmcb.13021

DOI: 10.1111/jmcb.13021

ALBERTO CARUSO

LAURA CORONEO

Does Real-Time Macroeconomic Information Help

to Predict Interest Rates?

We analyze the predictive ability of real-time macroeconomic information

for the yield curve of interest rates. We specify a mixed-frequency macro-

yields model in real time that incorporates interest rate surveys and treats

macroeconomic factors as unobservable components. Results indicate that

real-time macroeconomic information is helpful to predict interest rates, and

that data revisions drive a superior predictive ability of revisedmacro data

over real-time macro data. We also nd that interest rate surveys can have

signicant predictive power overand above real-time macroeconomic vari-

ables.

JEL codes:C32, C38, C53, E43, E44, G12

Keywords:government bonds, real-time macroeconomics, forecasting,

survey data, factor models

M    important

information for forecasting the evolution of the yield curve. This is due to both the be-

havior of policymakers, who operate on interest rates to stimulate aggregate demand

and control ination, and market agents, who closely monitor macroeconomic data

We thank the editor,Pok-sang Lam, and two anonymous reviewers for comments that helped improve

the paper. Weare also grateful to Carlo Altavilla, Laurent Ferrara, Gergely Gánics, Domenico Giannone,

Adam Golinski, Michal Horvath, Elmar Mertens, IvanPetrella, Lucrezia Reichlin, Dick van Dijk and sem-

inar and conference participants at the International Institute of Forecasters - Macroeconomic Forecasting

Seminar, the 2021 Oxford-MMF workshop (University of Oxford), the 2nd conference on Forecasting at

Central Banks (Bank of England), the NBP Workshopon Forecasting (Bank of Poland), the Workshop on

Big Data and Economic Forecasting (European Commission), the 8th Italian Congress of Econometrics

and Empirical Economics (University of Salento), the 12th International Conference on Computational

and Financial Econometrics (University of Pisa), the Annual Conference of the International Association

for Applied Econometrics (IAAE 2019, Nicosia), the Conference on Real-Time Data Analysis, Methods

and Applications (National Bank of Belgium), the University of York,and the University of Shefeld for

useful comments. The authors acknowledge the support of Now-CastingEconomics Ltd. in the early stages

of the paper. The viewsexpressed in this paper are those of the authors and do not necessarily reect those

of EY.

A C is at EY and ECARES, Université Libre de Bruxelles (E-mail: al-

berto.caruso01@gmail.com).L C is at Department of Economics and Related Studies,

University of York(E-mail: laura.coroneo@york.ac.uk).

Received April 20, 2020; and accepted in revised form September 19, 2022.

Journal of Money, Credit and Banking, Vol. 55, No. 8 (December 2023)

© 2023 The Ohio State University.

2028 :MONEY,CREDIT AND BANKING

and react to macroeconomic news (Beechey and Wright 2009, Altavilla, Giannone,

and Modugno 2017). Indeed, following the seminal work by Ang and Piazzesi (2003),

there is a consensus in the literature that macroeconomic indicators are successful at

predicting interest rates and excess bond returns (see Mönch 2008, Ludvigson and Ng

2009, Favero, Niu, and Sala 2012, Coroneo, Giannone, and Modugno 2016, among

others). However, Ghysels, Horan, and Moench (2017) nd limited evidence of pre-

dictive ability of real-time macroeconomic variables for excess bond returns: they

argue that the result of the previous literature was an artifact coming from the use of

revised data, instead of real-time macroeconomic data.1

In this paper, we assess the relevance of real-time macroeconomic information to

predict the future path of the yield curve of interest rates. Our contribution is to make

interest rate predictions based on the information set available to agents at each point

in time by taking into account all the characteristics of the real-time macroeconomic

data ow.2First, most macroeconomic data are released in a nonsynchronous way

and with different publication lags; therefore the available information at each point

in time can be described by a data set that has a ragged edge, and it is not balanced.

Second, macroeconomic data are very often subsequently revised: the revisionsmight

be substantial and affect the estimation and the forecast computed using different vin-

tages of the data. Third, in real-time forecasting, soft information provided by surveys

can have an important role as it is timely, not subject to revisions, and can readily in-

corporate any information available to survey participants, such as information about

the current state of the economy or forward-looking information contained in mone-

tary policy announcements. However, one drawback of using survey expectations is

that their projections are only for quarterly averages.

In order to exploit the informational content of real-time macroeconomic data for

interest rate predictions, we specify a mixed-frequency macro-yields model in real

time that incorporates interest rate surveys and treats macroeconomic factors as un-

observable components, which we extract simultaneously with the traditional yield

curve factors. Similarly to Coroneo, Giannone, and Modugno (2016), we identify the

factors driving the yield curve by constraining the loadings to follow the smooth pat-

tern proposed by Nelson and Siegel (1987). More specically, our empirical model

is a mixed-frequency dynamic factor model for Treasury zero-coupon yields, a rep-

resentative set of real-time macroeconomic variables and interest rate surveys, with

restrictions on the factor loadings.

1.Acommon denominator of this literature, in fact, is the use of revised macroeconomic data to predict

interest rates, which involves using an information set that is different from the one available to market

participants when the predictions were made.

2.Adequately specifying the information set availableto agents in real time is particularly important

when evaluating models in macroeconomics and nance, especially when the objectiveis to forecast asset

prices using external information, since according to the efcient market hypothesis asset prices should

already incorporate all the available information about their future evolution (see Orphanides 2001, Or-

phanides and VanNorden 2002, Croushore and Stark 2003).

ALBERTOCARUSO AND LAURA CORONEO:2029

Our model can be estimated by maximum likelihood—see Doz, Giannone, and

Reichlin (2012)—using an Expectation-Maximization (EM) algorithm adapted to the

presence of restrictions on the factor loadings and to missing data. Using U.S. data

from 1972 to 2019, we nd that real-time macroeconomic information is helpful to

predict interest rates, especially short maturities at mid and long horizons, and that

data revisions drive an increase in the predictive powerof revised macro information

with respect to real-time macro information. Moreover, during a period when a for-

ward guidance policy is implemented, we nd that incorporating interest rate surveys

in the model signicantly improves its predictive ability.

Our nding that data revisions drive the increased predictive ability of revised

macro data with respect to real-time macro data is in line with Ghysels, Horan, and

Moench (2017). However,while they nd that real-time macroeconomic information

has only a marginal (and often statistically nonsignicant) role in predicting excess

bond returns, our results show that real-time macroeconomic information is help-

ful to predict interest rates, as its predictive power is similar to that of revised macro

data. The crucial difference between our approach and the one in Ghysels, Horan, and

Moench (2017) lies in how the real-time data set is specied: we use the latest infor-

mation available to market participants at the time in which forecasts are made (that

includes both new releases of data points and revisions of already observed data),

Ghysels, Horan, and Moench (2017) instead use rst releases of data. In general,

when the objective is to forecast macroeconomic variables, rst releases provide ac-

curate predictions Koenig, Dolmas, and Piger (2003). However, to predict nancial

variables, it is important to use all the latest available information, as nancial op-

erators care about the nal revised value of a macroeconomic series (Gilbert 2011).

Indeed, our results indicate that the latest information available on real-time macro

variables has a stronger predictive ability than their rst releases, which is in line with

the intuition that revisions enhance the quality of macroeconomic information.

Lastly, we nd that incorporating interest rate surveys from the Surveysof Profes-

sional Forecasters (SPF) can improve the predictive ability of models that use only

information embedded in the yield curve and in macroeconomic variables. Surveys,in

fact, incorporate soft information about the future path of interest rates—that comes

from policy announcements, for example—that cannot be taken into account by stan-

dard macroeconomic variables. With this in mind, we test the relevance of the infor-

mation contained in the SPF survey forecasts for the real-time macro-yields model.

Results indicate that they enhance the predictive ability of the model in a period in

which the Federal Reserve implemented a forward guidance policy.The resulting im-

provement in predictive ability is statistically signicant. This intuitively appealing

result is in line with Altavilla, Giacomini, and Ragusa (2017), who use the selected

survey forecast value as their forecast for the specic horizon and maturity.However,

our results show that in some periods our model produces more accurate forecasts

than the survey forecasts. Therefore, we incorporate the surveys into the model it-

self. In this way, we combine in a single framework the “soft” information embed-

ded in the surveys with the information carried by interest rates and by the real-time