[Paper Review] Words that Matter: The Impact of Negative Words on News Sentiment and Stock Market Index

Wonseong Kim|arXiv (Cornell University)|Apr 2, 2023

Financial Markets and Investment StrategiesEconomics, Econometrics and Finance3 citations

TL;DR

This study examines how negative words in South Korean economic news affect sentiment scores and the KOSPI200 index. Using Word2Vec, cosine similarity, and an expanded lexicon (Sent1000), it shows that incorporating high-similarity negative words significantly improves sentiment prediction of market movements compared to a baseline lexicon (Sent0).

ABSTRACT

This study investigates the impact of negative words on sentiment analysis and its effect on the South Korean stock market index, KOSPI200. The research analyzes a dataset of 45,723 South Korean daily economic news articles using Word2Vec, cosine similarity, and an expanded lexicon. The findings suggest that incorporating negative words significantly increases sentiment scores' negativity in news titles, which can affect the stock market index. The study reveals that an augmented sentiment lexicon (Sent1000), including the top 1,000 negative words with high cosine similarity to 'Crisis,' more effectively captures the impact of news sentiment on the stock market index than the original sentiment lexicon (Sent0). The results underscore the importance of considering negative nuances and context when analyzing news content and its potential impact on market dynamics and public opinion.

Motivation & Objective

• To investigate the role of negative words in shaping sentiment in economic news articles.
• To assess how sentiment derived from news content influences the South Korean KOSPI200 stock market index.
• To develop and evaluate an enhanced sentiment lexicon that captures nuanced negative language.
• To compare the predictive power of a refined lexicon (Sent1000) against a standard one (Sent0) in financial sentiment analysis.
• To understand the contextual impact of specific negative terms on market sentiment and dynamics.

Proposed method

• Constructed a custom sentiment lexicon (Sent1000) by identifying the 1,000 most negative words with high cosine similarity to the word 'Crisis' using Word2Vec embeddings.
• Applied cosine similarity to measure semantic proximity between words and the term 'Crisis' to select contextually relevant negative terms.
• Processed 45,723 South Korean daily economic news articles using NLP techniques to extract sentiment scores from both Sent0 and Sent1000 lexicons.
• Calibrated sentiment scores from news titles and compared their correlation with daily KOSPI200 index movements.
• Used statistical analysis to evaluate the predictive power of sentiment scores derived from the two lexicons on market trends.
• Evaluated model performance by comparing sentiment score negativity and market index correlation across both lexicon versions.

Experimental results

Research questions

• RQ1 How do negative words with high semantic similarity to 'Crisis' influence sentiment scores in economic news headlines?
• RQ2 To what extent does an expanded sentiment lexicon (Sent1000) improve sentiment detection compared to a standard lexicon (Sent0)?
• RQ3 What is the correlation between sentiment scores derived from news content and daily movements in the KOSPI200 index?
• RQ4 How does the inclusion of contextually relevant negative words affect the predictive power of sentiment analysis for financial markets?
• RQ5 Can sentiment derived from news headlines effectively anticipate short-term fluctuations in the South Korean stock market index?

Key findings

• The Sent1000 lexicon, which includes 1,000 negative words highly similar to 'Crisis' in Word2Vec space, produced significantly more negative sentiment scores in news titles than the baseline Sent0 lexicon.
• News sentiment scores derived from Sent1000 showed a stronger correlation with daily KOSPI200 index movements than those from Sent0.
• Incorporating contextually relevant negative words enhanced the sensitivity of sentiment analysis to market-moving news events.
• The study found that semantic similarity to 'Crisis' is a strong indicator for identifying high-impact negative terms in financial news.
• The results demonstrate that nuanced lexical choices in news headlines—especially negative terms—have measurable effects on investor sentiment and market performance.
• The expanded lexicon improved the detection of negative sentiment during periods of market stress, suggesting better responsiveness to crisis-related language.