Policy Forum

A sentiment analysis of tweets following the ECB governing council meeting

A well-known anecdotical characteristic of European integration is that different member countries hold a different perception of their central bank’s performance. In “Frugal countries” – as Germany, the Netherlands or Finland – a stricter adherence to the official mandate and the pursuit of a hawkish monetary policy are seen as what the central bank should aspire to. On the other hand, “Mediterranean countries” – such as Greece or Italy – often complain about the lack of support from the ECB, which in their view should back countries facing large output gaps by adopting a looser monetary policy. Given such polarization, it might be interesting to make a quantitative assessment to this widespread belief.

A quantitative measure for differences in the perception of ECB performance

We perform a simple sentiment analysis on tweets written on the day of the latest ECB governing council meeting, held on June 4th, conditional on the language the tweets were written in.  Specifically, we consider four languages: German, French, Italian and English. For each of these we scrape a sample of tweets containing the acronym “ECB” as spelled in the specific language [1]. Since the meeting’s press release – followed by a press conference held by its president Mrs. Lagarde – was published at 13:45 CET, we have labeled tweets as written pre or post press release, depending on the time stamp, and categorized them accordingly. This approach allows us to follow two paths: firstly, to assess the generic sentiment towards the ECB across countries; and secondarily, to inquire how the (moderately dovish) decision to extend the Pandemic Emergency Purchase Programme (PEPP) was received in different countries.

Code Features

The simple code, written in Python language, uses Twitter API to fetch tweets based on the standard query, which allows to specify which language should the returned tweets be written into [2]. Tweets are then cleaned and standardized for sentiment analysis by automatically removing links and special characters. Finally, they are classified as “positive”, if the tweet polarity is larger than 0; “neutral”, if polarity is equal to 0; or “negative”, if polarity is less than 0. For German, French and English the sentiment analysis is performed using textblob library and its supports textblob_de and textblob_fr. Unfortunately, no Italian support is available as far as we know. Therefore, we must take a slightly more convoluted path when dealing with Italian tweets and first automatically translate them in English through Google Translate API, then analyze the sentiment in the standard way [3]. Given the simplistic approach used to grasp a quantitative snapshot of ECB approval, some caveats are due:

  • Twitter users population might not represent a stratified sample. In particular, older generations are likely to be underrepresented.
  • Given Twitter API thresholds, we cannot scrape all tweets corresponding to the query of interest, but just a sample of them.
  • The analysis concerns tweets written on June 4th 2020, day of the ECB Governing Council meeting, therefore it lacks long term variation in time and is not intended to estimate a measure of long run support for the central bank measures.

Results

SENTIMENT – FULL SAMPLE

LanguagePositiveNegativeNeutral#Tweets
German23%14.3%62.7%709
French42.8%17.1%40.2%879
Italian36.0%23.7%40.4%1292
English41.1%17.3%41.6%4877

Regarding generic sentiment across countries, we observe how, overall, results are in line with anecdotical evidence: in German speaking countries support for ECB is relatively lukewarm, with negative tweets relatively close to positive ones, and a very high share of neutral tweets. This matches the narrative of “Frugal countries” opposing a supposedly dovish monetary policy stance of the ECB. French speaking countries (i.e. mainly France and part of Belgium) on the other hand show overwhelming support for the ECB. Italy positions itself somewhere in between, yet being also more polarized, as suggested by the high share of both positive and negative tweets. This might signal that, although traditionally being favorable as a country to ECB’s expansive monetary policy, trust in European institutions took a hit following the allegedly insufficient Union’s response to COVID19. Nonetheless, no definitive conclusion can be drawn since we lack a measure of country specific sentiment variation in longer time horizons. Finally, English tweets – which cannot even approximately be matched with specific countries, given the role of English as the global language – can be used to proxy a generic international sentiment [4]. We can observe how positive sentiment clearly outweighs negative one in this category.

The effect of the announcement of PEPP extension

By exploiting the discontinuity given by the announcement of the governing council decisions – mainly an extension of the Pandemic Emergency Purchase Programme – we can also assess the public opinion response on the social media to the new measures.

PRE STATEMENT SENTIMENT

LanguagePositiveNegativeNeutral#Tweets
German20.1%12.3%67.5%153
French46.9%14.6%38.6%253
Italian38.8%27.9%36.4%190
English46.3%19.1%34.6%1276

POST STATEMENT SENTIMENT

LanguagePositiveNegativeNeutral#Tweets
German23.8%14.8%61.4%556
French41.1%18.1%40.8%626
Italian36.0%23.0%41.1%1102
English39.3%16.6%44.1%3601

The first element of interest is the rise in interactions. Tweets quoting the ECB increase between 2.5 and 5.8 times after the press release, a phenomenon which cannot simply be explained by daily seasonality (pre and post press release), thus signaling a clear interest of the public in the actions of the central bank [5]. This was particularly true for Italian, something which can intuitively be justified by the particular importance that the ECB’s low rate policy has for Italy, in light of its higher public debt levels.

LanguageRatio of number of tweets (post vs pre meeting)
German3.6
French2.5
Italian5.8
English2.8

The second element of interest is the change in sentiment following the ECB governing council’s press release. Among German speaking tweeters we can observe how both positive and negative proportion of tweets increase, signaling a polarizing effect of the monetary policy decisions. It can be speculated that both pro-Euro and Eurosceptic users actively reacted to the central bank move. In French tweets we observe a sensible decrease in positive tweets, which are matched by a rise in both negative and neutral ones. Such move is not easy to explain and might be interpreted as a sign of higher expectations of monetary stimulus compared to what was actually delivered. Nonetheless, it should be stressed that French tweets remain overall the most supportive of the central bank. Italian tweets are the most positively affected subgroup: though positive sentiment only slightly increases, negative tweets strongly decline, with a corresponding growth of neutral ones. Such shift may hint to the fact that the ECB move succeeded in placating criticism of part of the Italian Eurosceptic twitter. Lastly, English tweets show a fall in both positive and negative tweets. Partly, this might be explained by the fact that the press release represented noteworthy economic news which, however, did not particularly emotionally involve those tweeting about it but not being citizens of a country part of the Eurozone. Partly, this may be due to the underlying composition of this group which – being English-tweeting – likely comprises a relatively more cosmopolitan and less polarized constituency on the subject compared to the groups tweeting in native languages.

DELTA SENTIMENT

LanguagePositiveNegativeNeutral
German3.7%2.5%-6.1%
French-5.8%3.5%2.2%
Italian0.2%-4.9%4.7%
English-7.0%-2.5%9.5%

A final element to take into consideration is polarization, which we define as the share of non-neutral tweets: following the press release, polarization falls in all subgroups except for the German one, where we observe  a stark fall in neutral tweets while both positive and negative ones rise. Such data indicates how ECB’s actions might exacerbate divisions inside countries which represent the core of European integration. Nonetheless, it should be stressed that polarization remains relatively low once compared with other subgroups, thus indicating that stress levels have not been reached yet among the German-speaking public.

POLARIZATION

LanguagePrePostDelta
German32.7538.6%6.1%
French61.4%59.2%-2.2%
Italian63.6%58.9%-4.7%
English65.4%55.9%-9.5%

Conclusions

A simplistic quantitative measure of ECB perception based on language specific tweets seems to support the narrative of a public opinion in the “Frugal block” being relatively more critic of the central bank action, in opposition to the public opinion in “Mediterranean Block” countries. French support is strikingly high, whereas Italy shows a lower support than France, something which might be due to the allegedly insufficient European institutions common response to COVID19 but also rhymes with longer-term Italian skepticism. By exploiting the discontinuity offered by the ECB governing council, we also observe how a moderately dovish policy shift was received by different public opinions: German-speaking users responded with a bounce in polarization (which nonetheless remains overall relatively low), while Italy saw a significant shift of negative tweets towards neutral ones. Surprisingly, French tweets show some degree of disappointment, possibly due to a failure to address higher expectations. Nonetheless, the subgroup remains the one showing the highest support for the central bank.

As a bottom line, the online crowd appears to be sensible to monetary policy decision, with different responses varying significantly in tone across countries. While more dovish stances seem to be correlated with more positive perception of the central bank in struggling countries, they also seem to polarize opinion in fiscally responsible ones. Although present data signal that we are still far from stress levels, the balance of these opposite forces should not be overlooked by policymakers.


[1] This would be “EZB” for German, “BCE” for French and Italian, and “ECB” for English (note that the query is not case sensitive).
[2] We decide to filter and ignore retweets.
[3] This implies that measures for Italian tweets might be less accurate than those available for other languages.
[4] Based on the assumption that English native speaking users (UK, USA, Australia, etc.), though more numerous than other users tweeting in English, are relatively less interested in EU internal policy with respect to the latter.
[5] Though Twitter API allows to scrape only a sample of tweets, the dimension of such sample is proportional to the number of tweets corresponding to the query, thus the inference remains valid.

ANTONIO FOCELLA – Junior Research Analyst Algebris Policy & Research Forum

The Algebris Policy & Research Forum (the "Forum") is a not-for-profit advisory forum, independent of the commercial activities of Algebris (UK) Limited. The Forum is designed to contribute to the promotion of a strong and a balanced European economy, underpinned by a sound financial system and supported by a transparent regulatory and fiscal environment, for the benefit of societies as a whole. The primary focus of the Forum is on publishing expert reports, recommendations on European economic policy issues and sharing expert knowledge and research with the public, governmental and non-governmental institutions.
This document is issued by the Forum and the information and opinions contained in this document are for background purposes only, do not purport to be full or complete, do not constitute investment advice and are entirely independent of the commercial activities of Algebris (UK) Limited. Under no circumstances should any part of this document be construed as an offering or solicitation of any offer of any fund managed by Algebris (UK) Limited. This information does not constitute Investment Research, nor a Research Recommendation. No reliance may be placed for any purpose on the information and opinions contained in this document or their accuracy or completeness. No representation, warranty or undertaking, express or implied, is given as to the accuracy or completeness of the information or opinions contained in this document by the Forum or Algebris (UK) Limited, its directors, employees or affiliates and no liability is accepted by such persons for the accuracy or completeness of any such information or opinions. The distribution of this document may be restricted in certain jurisdictions. The above information is for general guidance only, and it is the responsibility of any person or persons in possession of this document to inform themselves of, and to observe, all applicable laws and regulations of any relevant jurisdiction.
© 2020 Algebris Policy & Research Forum. All Rights Reserved.