Sentiment and emotion in financial journalism: a corpus-based, cross-linguistic analysis of the effects of COVID

The 2008 Global Systemic Crisis significantly impacted economic discourse in the media, with newspapers and financial reports becoming primary information sources for investors. More recently, climate change and the COVID-19 pandemic have further exacerbated economic challenges, creating supply chain issues, inflation, and other economic disruptions. This study focuses on the linguistic shifts in financial journalism during these periods. By analyzing comparable corpora from *The Economist* (English) and *Expansión* (Spanish), the researchers aimed to quantify changes in sentiment and emotion expressed in financial news articles before and during the COVID-19 pandemic. The hypothesis was that the lexicon and tone would be mildly positive before the pandemic, reflecting the lingering effects of the 2008 crisis, but significantly more negative during the pandemic due to the various economic challenges faced. The study also posits that the polarization of sentiment and expressions of emotion would differ between the publications, potentially reflecting fear or greed as underlying factors influencing market volatility.

The study acknowledges the limited research on the connection between emotions and economic language, despite the established role of emotions in economic behavior and decision-making. While economics has traditionally been viewed as a rational and impartial discipline, recent studies emphasize the importance of emotions in economic performance. The CNN Fear & Greed Index, driven by opposing forces of greed and fear, serves as a key indicator in economic decisions. The researchers integrate the concept of greed into Affect Spectrum Theory (AST), defining it as a combination of joy and anger, representing a willingness to invest despite obstacles. Fear, representing aversion to invest, is simpler to define and is considered a primary emotion. The study further considers other emotions like disgust and trust in relation to risk aversion/attraction.

A corpus-based methodology was employed, compiling a large English corpus (3,736,103 words) from *The Economist* and a Spanish corpus (1,965,154 words) from *Expansión*. Each corpus was divided into pre-COVID (2018-2019) and COVID (2020-2021) sub-corpora. Sentiment analysis was conducted using Lingmotif 2 software to determine the semantic orientation (positive/negative) and intensity of sentences. A smaller sample (around 1 million words per sub-corpus) was initially analyzed with Lingmotif 2 due to software limitations. Subsequently, emotion detection was performed using the NRC Word-Emotion Association Lexicon (EmoLex), based on Plutchik's eight primary emotions. EmoLex was used to assign scores for positive/negative sentiment and for potential context-invariant emotions. The researchers cross-checked word frequency lists from Sketch Engine with the EmoLex list, automatically assigning emotions using Excel conditional search formulas. The study then focused on the Fear and Greed Index framework within the study’s emotional lexicon, with the goal of assessing risk aversion/attraction.

Lingmotif 2 analysis of a smaller sample showed that pre-COVID, *Expansión* exhibited more positive sentiment (64% positive sentences) than *The Economist* (22% positive sentences). During the COVID period, both publications showed a shift towards negative sentiment. *Expansión* showed a relative balance between positive and negative words (around 50% each), while *The Economist* exhibited a significant increase in negative items. Further analysis using EmoLex across the full corpora confirmed the shift to negative sentiment during the COVID period. In the pre-COVID period, Expansión showed higher levels of trust, anticipation, and joy, whereas The Economist had more trust, but also considerable fear. In the COVID period, both newspapers showed an increase in fear and sadness and a decrease in trust and joy. Analysis of the ten most frequent nouns relating to fear and greed (anger + joy) further supported this trend, indicating increased risk aversion during the pandemic. Topic analysis, using Sketch Engine's document co-occurrence frequency (DOCF), revealed a shift in topics covered by both newspapers. *Expansión* shifted from a focus on climate change, economic issues, and politics in the pre-COVID period to a dominance of pandemic-related topics in the COVID period. *The Economist*, while maintaining its broad focus on economics, politics, and social issues, showed an increased emphasis on pandemic-related issues during the COVID period. A paired sample t-test in SPSS confirmed a statistically significant increase in negative sentiment (p < 0.05) in both publications after the pandemic began.

The findings largely support the hypotheses, showing a significant shift towards negative sentiment in both publications after the pandemic's onset. However, the intensity of emotional activity remained high in both periods, suggesting a consistently emotionally charged discourse in financial journalism. The differing approaches of the two newspapers highlight the influence of journalistic scope and origins on topic selection and overall tone. The increased negativity reflects the tangible economic impacts of the pandemic, including health risks, and economic uncertainty and fear. The study's integration of linguistic analysis and market indicators provides a novel approach to understanding the interaction between emotion, language, and financial markets. The preponderance of trust prior to the pandemic, which lessened after the pandemic, suggests moderate risk-taking, but this was significantly reduced with the advent of COVID.

This study demonstrates the significant impact of major economic crises on the sentiment and emotional expression in financial journalism. The shift towards negative sentiment during the COVID-19 pandemic is evident in both English and Spanish publications, although with some variations in tone. The study’s framework, linking linguistic analysis with market indicators, offers a new approach to examining the dynamic relationship between language, emotion, and economic behavior. Future research could explore the effects of other crises and events on journalistic language, broaden the range of publications analyzed, and examine the direct impact of this type of media coverage on investors' behaviors.

The study's focus on two specific publications might limit the generalizability of the findings. Further, the analysis relied on automated tools for sentiment and emotion detection, which might not capture all nuances of human language. The study’s analysis of risk-taking behavior focuses only on the extremes, rather than the full range of investor attitudes, possibly creating a less nuanced image of overall behavior. The use of pre-defined emotion lexicons may not fully account for context-specific interpretations of words. While it attempted to address this issue through manual review, it could be augmented further by future research.