Abstract
Recent literature suggests that news shocks could be an important driver of economic cycles. In this article, we use a direct measure of news sentiment derived from media reports. This allows us to examine whether innovations in the reporting tone correlate with changes in the assessment and expectations of the business situation as reported by firms in the German manufacturing sector. We find that innovations in news reporting affect business expectations, even when conditioning on the current business situation and industrial production. The dynamics of the empirical model confirm theoretical predictions that news innovations affect real variables such as production via changes in expectations. Looking at individual sectors within manufacturing, we find that macroeconomic news is at least as important for business expectations as sector-specific news. This is consistent with the existence of information complementarities across sectors.
This is a preview of subscription content, log in via an institution to check access.
Similar content being viewed by others
Notes
The media research institute Medientenor, which provides us with the data, employs human coders, who read every article in the news and classify the information in these reports according to their topic (we look at business cycle, macroeconomic growth, sectoral information). In addition, they identify the tone of each statement in each article as ‘good,’ ‘bad’ or ‘neutral’. We aggregate this daily information for each month, which gives us a time series of media tonality. This approach is complementary to textual analysis approaches that have been used, for example, in accounting and finance, where written media reports are classified based on dictionaries and machine-learning techniques, see Loughran and McDonald (2016) and references therein. Shapiro et al. (2018) show that textual analysis-based news indicators correlate with the business cycle and have predictive power for future economic activity, which is consistent with the findings in this paper based on our Media Tenor data.
See Nerb (2004) for a description.
More precisely, it is the geometric mean of the indicators derived from the balances to the question: ‘We judge our current business situation for product group XY to be good, satisfactorily, or bad,’ and the question: ‘With respect to the business cycle, our business situation for product group XY is expected to be somewhat better, more or less the same, or somewhat worse in the next 6 months.’ Note that both questions refer to the ‘business climate’ of the firm and do not explicitly ask for developments in profits, or production. How the term ‘business climate’ should be interpreted is left open to the individual firms. Nevertheless, it is generally acknowledged that these qualitative results give a good indication of how actual industrial production evolves over the time.
Hence, rather than focusing on the forecasting ability of Ifo Business Tendency Survey indicators, as is often done in literature (see, e.g., Fritsche and Stephan 2002; Hüfner and Schröder 2002), this paper uses these indicators as direct measures for firms’ sentiment and assessments of their own future development.
We do not consider retail and wholesale trade and the construction sector, because we do not have data from Media Tenor on news reports covering these sectors.
The following news sources are analyzed: Daily press: Frankfurter Allgemeine Zeitung, Welt, Süddeutsche Zeitung, Frankfurter Rundschau, Tageszeitung, Bild, Neue Zürcher Zeitung, Berliner Volksstimme, Sächsische, Westdeutsche Allgemeine Zeitung, Kölner Stadt-Anzeiger, Rheinischer Merkur; daily TV News: ARD Tagesschau, Tagesthemen, ZDF Heute, Heute Journal, RTL Aktuell, SAT.1 18:30, ProSieben Nachrichten; Weekly Press: Spiegel, Focus, Die Woche, Wochenpost, Welt am Sonntag, Bild am Sonntag, Die Zeit.
Although our variables are bounded between \({+}\) 100 and \({-}\) 100 there is little merit in applying the log-odds transformation, as the mass of observations is concentrated and more than two standard deviations away from the bounds.
Note that our measure of macroeconomic news is not the average or aggregate of the sector-specific news items, but instead contains the assessment of news focusing on the German economy as a whole. Moreover, the manufacturing sector accounts for less than 25% of German GDP and our six sectors are only part of that. Consequently, the information contained in our sector-specific variable relative to our news measure on the economy as a whole basically does not overlap.
As a robustness check, we also construct media indicators using data covering the full month. However, this does not alter the results qualitatively.
In Table 4 in the ‘Appendix’ section, we show summary statistics for the stacked series covering the six sectors used in the panel VAR analysis.
We use the Cholesky factorization to identify the primitive innovations. In our case, the ordering is largely predetermined by the construction (timing) of the media variables—as depicted in Fig. 2—and the reference in time of the two Ifo survey questions. By construction, we cannot have a contemporary feedback effect from the two Ifo indicators on media as the media coverage of these indicators has been excluded. Hence, we can clearly separate between these two shocks. With respect to the two Ifo indicators, we assume that the Situation indicator can have a contemporaneous effect on the Ifo Expectations, but not vice versa. Because the Ifo Situation indicator reflects the current actual situation, which then might feed into the calculation of the outlook of the firm, this appears to be a sensible assumption. Only with respect to the macroeconomic and the sector-specific media indicators, it is less obvious which ordering is theoretically more plausible. We hypothesise the contemporaneous effect of macroeconomic news on sector-specific news to be negligible. Hence, the ordering from most exogenous to most endogenous is the two Ifo indicators—Ifo Situation and Ifo Expectations—followed by the two media variables—sector-specific and macroeconomic news. None of the qualitative conclusions depend upon this assumption, or in general upon the ordering chosen. Results are available upon request.
Besides the contemporaneous values, we include up to three lags of macroeconomic and sector-specific industrial production growth.
We opt for the likelihood-ratio test. However, Wald tests lead to qualitatively identical results.
The impulse-response functions are scaled by the standard deviation of the response variable, and each shock equals one standard deviation of the impulse variable.
Other evidence supporting the view that news is highly relevant for economic outcomes is provided by Mora and Schulstad (2007). They find that the information agents have about current GNP, i.e., first releases on these, have a larger impact on their own actions than the true ex-post figures of GNP. The authors study the degree to which expectations affect the evolution of the economy. They find that once GNP first releases are taken into account, the true (revised) value of GNP growth at time t has no predictive power in explaining future growth rates at any time. Thus, all the predictive power lies in the unexpected part of the announcements, and not in the true level of growth.
That is, we concentrate on the \(A_{31}\), \(A_{41}\), \(A_{32}\) and \(A_{42}\) polynomials in Eq. (11). The full set of results is available upon request.
TFP growth in the food sector is basically zero. The average unweighted TFP growth across all sectors is about 1%.
This is underlined by the low degree of volatility in this sector relative to the other sectors in our sample. See Table 1.
The following news sources are analyzed: Daily press: Frankfurter Allgemeine Zeitung, Welt, Süddeutsche Zeitung, Frankfurter Rundschau, Tageszeitung, Bild, Neue Züricher Zeitung, Berliner, Volksstimmer, Sächsische, Westdeutsche Allgemeine Zeitung, Kölner Stadt-Anzeiger, Rheinischer Merkur; daily TV News: ARD Tagesschau, Tagesthemen, ZDF Heute, Heute Journal, RTL Aktuell, SAT.1 18:30, ProSieben Nachrichten; Weekly Press: Spiegel, Focus, Die Woche, Wochenpost, Welt am Sonntag, Bild am Sonntag, Die Zeit.
References
Barsky RB, Sims ER (2011) News shocks and business cycles. J Monetary Econ 58(3):273–289
Barsky RB, Basu S, Lee K (2014) Whither news shocks? University of Chicago Press, Chicago, pp 225–264. http://www.nber.org/books/park14-1
Beaudry P, Lucke B (2010) Letting different views about business cycles compete. University of Chicago Press, Chicago, pp 413–455
Beaudry P, Portier F (2006) Stock prices, news, and economic fluctuations. Am Econ Rev 96(4):1293–1307
Beaudry P, Portier F (2014) News-driven business cycles: insights and challenges. J Econ Lit 52(4):993–1074
Bodnaruk A, Loughran T, McDonald B (2015) Using 10-k text to gauge financial constraints. J Financ Quant Anal 50(5):1–24
Brückner M, Pappa E (2015) News shocks in the data: olympic games and their macroeconomic effects. J Money Credit Bank 47(7):1339–1367
Dräger L (2015) Inflation perceptions and expectations in Sweden are media reports the missing link? Oxf Bull Econ Stat 77(5):681–700
Fan H, Gao X, Xu J, Xu Z (2016) News shock, firm dynamics and business cycles: evidence and theory. J Econ Dyn Control 73(C):159–180
Fritsche U, Stephan S (2002) Leading indicators of German business cycles: an assessment of properties. J Econ Stat 222(3):289–315
Giannini C (1992) Topics in structural VAR econometrics. Springer, Berlin
Görtz C, Tsoukalas JD (2017) News and financial intermediation in aggregate fluctuations. Rev Econ Stat 99(3):514–530
Holsti OR (1969) Content analysis for the social sciences and humanities. Addison-Wesley, Reading
Hüfner F, Schröder M (2002) Forecasting economic activity in Germany—how useful are sentiment indicators? Discussion paper 02-56, ZEW
Kurmann A, Sims E (2017) Revisions in utilization-adjusted TFP and robust identification of news shocks. Working paper 23142, National Bureau of Economic Research
Lamla MJ, Lein SM (2014) The role of media for consumers inflation expectation formation. J Econ Behav Org 106:62–77
Lamla MJ, Lein SM (2015) Information rigidities, inflation perceptions, and the media: lessons from the euro cash changeover. Econ Inquiry 153(1):9–22
Loughran T, McDonald B (2011) When is a liability not a liability? Textual analysis, dictionaries, and 10-ks. J Finance 66(1):35–65
Loughran T, McDonald B (2016) Textual analysis in accounting and finance: a survey. J Account Res 54(4):1187–1230
Mora JVR, Schulstad P (2007) The effect of GNP announcements on fluctuations of GNP growth. Eur Econ Rev 51(8):1922–1940
Nerb G (2004) Survey activity of the ifo institute. In: Sturm JE, Wollmershäuser T (eds) Ifo survey data in business cycle and monetary policy analysis. Physica-Verlag, Berlin
Nierhaus W, Sturm JE (2007) Methods of economic forecasting and business cycle indicators. In: Goldrian G (ed) Handbook of survey-based business cycle analysis. Edward Elgar, Cheltenham, pp 117–142 chap 10
Oppenländer K (ed) (1997) Business cycle indicators. Avebury, Aldershot etc
Petersen B, Strongin S (1996) Why are some industries more cyclical than others? J Bus Econ Stat 14(2):189–198
Pigou A (1927) Industrial fluctuations. MacMillan, London
Shapiro AH, Sudhof M, Wilson DJ (2018) Measuring news sentiment. Working paper series 2017-1, Federal Reserve Bank of San Francisco
Strigel W (1990) Business cycle surveys: a new quality in economic statistics. In: Klein P (ed) Analyzing modern business cycles. Essays honoring. G.H. Moore, M.E. Sharpe, Inc., Arnouk, pp 69–84 (Chap. 5)
Sturm JE, Wollmershäuser T (eds) (2004) Ifo survey data in business cycle and monetary policy analysis. Physica-Verlag, Berlin
Theil H (1955) Recent experiences with the Munich business test: an expository article. Econometrica 23:184–192
Veldkamp L, Wolfers J (2007) Aggregate shocks or aggregate information? Costly information and business cycle comovement. J Monetary Econ 54:37–55
Zarnowitz V (1992) Business cycles: theory, history, indicators and forecasting, NBER studies in business cycles, vol 27. The University of Chicago Press, Chicago
Acknowledgements
We thank an anonymous referee, Laura Veldkamp and the participants of the seminar at the University of Zurich, the European Economic Association Meeting, the Meeting of the Swiss Society of Economics and Statistics and German Economic Association Meeting and the CESifo Area Conference on Macro, Money and International Finance for helpful comments. An earlier version circulated under the title News and Sectoral Comovement.
Author information
Authors and Affiliations
Corresponding author
Additional information
Publisher's Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Appendix
Appendix
1.1 Media data
The media data employed in this paper is collected and created by Media Tenor, a provider of media intelligence. Here, we describe in more detail how the data is collected and created and how it compares to other datasets used in the literature that reflect the content of media reporting.
Media Tenor first collects all articles from major newspapers and media broadcasts in Germany that are at least five lines long in case of printed media or last at least 5 seconds in the case of television reports. We rely on news reports stemming from 26 newspapers, weekly magazines and TV broadcasts.Footnote 22
Then, trained coders read every newspaper article and watch every television report that talks about the economic situation or expected economic development. This may be related to the German economy as a whole or to one or more specific sectors. The coders have to evaluate every statement related to the key topic (economic situation or expected economic developments) on whether it is toned neutral, good, or bad.
The resulting data includes for every article that is related to the German aggregate economy or a specific sector in the German economy the following information: (i) a sectoral identifier, which indicates which sector the article is related to. If the article is related to the aggregate economy, we have an identifier for the aggregate economy. (ii) The date when the article appeared in the media. (iii) Three variables that indicate whether a statement are toned good, bad, and neutral, respectively.
From this information, we construct measures of the frequency and tone of news media per month, which is described in the main body of this paper.
This approach is complementary to other approaches based on textual analysis of media data as described by Loughran and McDonald (2011), Bodnaruk et al. (2015), and in the survey on applications in accounting and finance of Loughran and McDonald (2016). The advantage of using textual analysis is that with a given codebook, there is a unique algorithm that classifies media reports based on dictionaries and sentiment word lists. That means, conditional on the provided media reports, the algorithm would always come up with the same classification. Human coders might come up with two different classifications when reading the same text. To control for this, Media Tenor asks two different coders to code each text independently of each other. Only if the two coders come up with the same classification, the data is final. If not, the coders have to re-read the text or further coders have to classify the same text until they achieve convergence. Since the textual analysis codebook described in Loughran and McDonald (2016) is only available in English, and the media data we use is solely in German, we cannot easily replicate our analysis based on their dictionaries, since simple translations are often not capturing all the different words available in another language. Furthermore, our data includes also TV broadcasts, which are not readily available in electronic text formats, that we could quantify.
1.2 Additional figures and tables
Two-step VAR. Notes: This figure reports impulse-response functions for the Panel VAR specified in Eq. (11). Here, news variables are the residuals of a regression of the news variables used in the main body of the paper on aggregate and sectoral industrial production and three lags thereof. The residual thus captures news that are orthogonal to current industrial production and therefore should capture news related to future economic developments. The first to fourth row indicates responses of each variable to a shock in the business situation indicator, the business expectation indicator, the sectoral news variable and the macroeconomic news variable, respectively. Columns show the responses of each of these variables. The impulse-response functions are plotted together with their 95% bootstrapped confidence intervals based on Giannini (1992)
Rights and permissions
About this article
Cite this article
Lamla, M.J., Lein, S.M. & Sturm, JE. Media reporting and business cycles: empirical evidence based on news data. Empir Econ 59, 1085–1105 (2020). https://doi.org/10.1007/s00181-019-01713-5
Received:
Accepted:
Published:
Issue Date:
DOI: https://doi.org/10.1007/s00181-019-01713-5