Financial textual sentiment connectedness: Evidence from alternative data

• This study investigates the connectedness of various firm-level investor sentiment proxies—news, social media, ESG positive (ESGpos), and ESG negative (ESGneg) sentiment—using aggregate connectedness measures for DJIA stocks from 2015 to 2024. • News and social media exhibit high self-connection scores, showing that each is predominantly shaped by its own content, while ESG sentiment proxies are largely independent. • The Financials sector is a net transmitter of sentiment shocks, while the Technology sector acts as a net receiver, with significant shifts in ESG sentiment roles during global events. This study investigates the connectedness of various firm-level investor sentiment proxies—news, social media, ESG positive (ESGpos), and ESG negative (ESGneg) sentiment using aggregate connectedness measures and a sample of DJIA stocks between 2015 and 2024. Our findings reveal that each sentiment proxy maintains strong internal consistency, predominantly shaped by its own sources. Specifically, news and social media exhibit high self-connection scores, indicating that these proxies are primarily influenced by their respective content. ESG sentiment proxies show minimal cross-influence from news and social media, indicating their distinct and independent nature. Network analysis further highlights that news and social media transmit sentiment shocks, while ESG-based proxies are predominantly receivers. The most significant flow of sentiment shocks is from social media to ESG negative sentiment. This reflects the central role of social media in shaping sentiment within the system, in contrast to the more isolated influence of news. During significant global event periods, ESGpos and ESGneg shift roles, with ESGpos becoming a transmitter and ESGneg a receiver of sentiment shocks. Sector-specific analysis shows that the Financials (Technology) sector is a net transmitter (receiver) of sentiment shocks. The practical implications of the findings are discussed. The paper contributes to the literature, which has treated different sentiment proxies as distinct phenomena despite their interconnectedness. Additionally, we find that the aggregate connectedness measures used in this study exhibit stronger connectedness compared to the traditional Diebold-Yilmaz framework.