Log in
E-mail
Password
Show password
Remember
Forgot password ?
Become a member for free
Sign up
Sign up
New member
Sign up for FREE
New customer
Discover our services
Settings
Settings
Dynamic quotes 
OFFON
News: Latest News
Latest NewsCompaniesMarketsEconomy & ForexCommoditiesInterest RatesBusiness LeadersFinance Pro.CalendarSectors

Nowcasting euro area GDP with news sentiment: a tale of two crises

11/25/2021 | 05:10am EST

Working Paper Series

Julian Ashwin, Eleni Kalamara, Lorena Saiz Nowcasting euro area GDP with

news sentiment: a tale of two crises

No 2616 / November 2021

Disclaimer: This paper should not be reported as representing the views of the European Central Bank (ECB). The views expressed are those of the authors and do not necessarily reflect those of the ECB.

Abstract

This paper shows that newspaper articles contain timely economic signals that can materially improve nowcasts of real GDP growth for the euro area. Our text data is drawn from fteen popular European newspapers, that collectively represent the four largest Euro area economies, and are machine translated into English. Daily sentiment metrics are created from these news articles and we assess their value for nowcasting. By comparing to competitive and rigorous benchmarks, we nd that newspaper text is helpful in nowcasting GDP growth especially in the rst half of the quarter when other lower-frequency soft indicators are not available. The choice of the sentiment measure matters when tracking economic shocks such as the Great Recession and the Great Lockdown. Non-linear machine learning models can help capture extreme movements in growth, but require sucient training data in order to be eective so become more useful later in our sample.

Keywords: Text analysis, Forecasting, Machine learning, Business cycles, COVID-19

JEL Classication: C43, C45, C55, C82, E37

ECB Working Paper Series No 2616 / November 2021

1

Non-technical summary

Monitoring the economy in real time is crucial for making informed economic and policy decisions. However, macroeconomic fundamentals like Gross Domestic Product (GDP) are typically measured at a quarterly frequency and released with a substantial delay. Market participants and policymakers have traditionally tracked soft data, particularly business and consumer condence surveys, in order to get a real time assessment of economic conditions. This is widely evidenced by monetary policy communications, which frequently point to survey evidence when describing the current macroeconomic situation. On the academic side, recent advances in data availability and computing power have promoted the use of alternative sets of predictors and novel methods often originated from the machine learning literature to gauge economic activity or detect turning points. These new data and models can act as complements to the current econometric tools used by policymakers and provide substantial aid especially in times of distress.

In this paper we build text-based sentiment indicators for the euro area derived from newspaper articles in the largest four euro area countries in their native languages. These indicators are available at daily frequency and contain timely economic signals which are comparable to those from well-known sentiment indicators such as the Purchasing Managers' index (PMI). In a second step, we test their predictive ability across the quarter and nd that they prove to be highly benecial in the rst month of the quarter when other soft indicators are not available. Their power is diminishing as we proceed within the quarter and new data are released. The sentiment indices are based on counts of words in news articles translated into English and rely on several well-known English language dictionaries.

When it comes to detecting turning points, it appears that the choice of the dictionary matters. A commonly used nance-oriented dictionary captures very well the Great Recession, unsurprisingly given the nancial nature of this crisis, but fails to capture the COVID-19 pandemic crisis. By contrast, the general-purpose dictionary is more consistent and robust across time. Therefore, the nature of economic shocks plays a signicant role in identifying the most appropriate text dictionary to be used.

We test the predictive ability of these daily time series with a number of forecasting models,

ECB Working Paper Series No 2616 / November 2021

2

from straightforward linear regressions to more sophisticated non-parametric machine learning algorithms. We nd that our sentiment metrics provide substantial improvements in nowcasting performance compared to both the ECB's ocial projections and benchmarks based on the Purchasing Managers composite index (PMI). These gains are typically concentrated in the rst half of the quarter, when other indicators are not yet available. These gains are particularly pronounced in the two major crisis periods in our sample: the Great Recession (2008-2009) and the Great Lockdown (2020). More specically, we nd that standard linear methods work well when there are no big shifts on the economic outlook but non-linearities matter when extreme economic shocks occur and the non-linear machine learning models can capture them more fully.

ECB Working Paper Series No 2616 / November 2021

3

  • Introduction

Monitoring the economy in real time is crucial for making informed economic and policy decisions. However, macroeconomic fundamentals like Gross Domestic Product (GDP) are typically measured at a quarterly frequency and released with a substantial delay. Market participants and policymakers typically rely on soft data such as business and consumer condence surveys to get a real time assessment of economic conditions. This is widely evidenced by monetary policy communications, which frequently point to survey evidence when describing the current macroeconomic situation. On the academic side, recent advances in data availability and computing power have promoted the use of alternative sets of predictors and novel methods often originated from the machine learning literature to answer many economic questions Hansen et al. (2017); Baker et al. (2016); Azqueta-Gavaldon et al. (2020)1. This study contributes to this growing body of work which shows that non-traditional datasets and methods can improve macroeconomic forecasts at short forecast horizons in the Euro area.

In this paper, we focus on the use of textual data derived from European newspaper articles to nowcast quarterly real GDP growth in the Euro area. We transform the text into daily aggregate time series of news sentiment for the Euro area. We make use of well-knownlexicon-based methods like the economics-oriented dictionaries of Correa et al. (2017) and Loughran and McDonald (2011) but also more of general purpose like VADER (Gilbert, 2014) and AFINN (Nielsen, 2011). This allows us to create high frequency text-based indicators which are able to capture the current economic conditions in a timely manner.

We test the predictive ability of these daily time series with a number of forecasting models, from straightforward linear regressions to more sophisticated non-parametric machine learning approaches. We nd that our sentiment metrics provide substantial improvements in nowcasting performance compared to both the ECB's ocial GDP projections and benchmarks based on the Purchasing Managers' index (PMI). These gains are typically concentrated in the rst half of the quarter, when other indicators are not yet available, and are particularly pronounced in the two major crisis periods in our sample: the Great Recession (2008-2009) and the Great Lockdown (2020).

  • For a review of alternative datasets we refer to Algaba et al. (2020) while a detailed application of a wide set machine learning methods to forecast US output is presented in Coulombe et al. (2020)

ECB Working Paper Series No 2616 / November 2021

4

This is an excerpt of the original content. To continue reading it, access the original document here.

Disclaimer

ECB - European Central Bank published this content on 25 November 2021 and is solely responsible for the information contained therein. Distributed by Public, unedited and unaltered, on 25 November 2021 10:09:04 UTC.


ę Publicnow 2021
Latest news "Companies"
08:48aSBERBANK : Sber Group to buy Sberbank shares for up to 50 billion rubles in 2022-2025 for the long-term employee remuneration program
EQ
08:47aFerguson out to spark Everton in 2nd spell as interim coach
AQ
08:47aEthiopian delays inflation data for over two weeks
RE
08:46aU.S. preparing for possible future COVID variants -White House
RE
08:46aCogeco and Its Employees Give Near Half A Million Dollars Back to the Community as Part of the 2021 United Way/Centraide Campaign
AQ
08:46aCNS Pharmaceuticals to Present at the Virtual Investor 2022 Top Picks Conference
PR
08:46aZUR ROSE GROUP : Jefferies gives a Buy rating
MD
08:46aKAPLAN/MANHATTAN PREP SURVEY : Business Schools Are Moving Slowly But Surely Toward STEM Designation, in a Boon for International Students
BU
08:46aItron's Intelis 250 Smart Gas Meter Receives Approval from Measurement Canada
BU
08:45aGold stalls after rally to two-month highs, palladium extends gains
RE
Latest news "Companies"