Building the Korean Sentiment Lexicon for Finance (KOSELF)

Abstract

This study aims to verify and establish a Korean sentiment lexicon suitable for corporate financial analysis. When analyzing existing sentiment lexicons, the KOSAC and KNU (Kunsan University) dictionaries developed based on Korean are weak because they are used for general purposes. The Harvard IV and Loughran and McDonald (2011) have the disadvantage of being translated from English. In this study, the Korean Sentiment Lexicon for Finance (KOSELF) is constructed and presented. To verify its usefulness, text data from about 20,000 analyst reports published in Korea from 2016 to 2018 are collected from the Hankyung Consensus web page. After calculating the sentiment variables of negative and positive word frequency using five sentiment lexicons for each report, the recommendation and target price changes are regressed on these sentiment variables. The sentiment variables from the newly-constructed KOSELF in this study have a significant relationship with the analyst’s recommendation and target price change. Even when the sentiment variables calculated through the other four sentiment lexicons are added, it shows better performance. Our work has practical significance in that it proposes a Korean sentiment dictionary that can be used for finance.