BERT어조에 기반한 애널리스트 보고서 텍스트 정보의 유용성

Yang(2021) reports that the prior tone of the analyst report titles provides additional information on the cumulative abnormal returns in the five days before and after the announcement, even after controlling for investment opinions, target prices, and earnings forecasts. This study adds a summary in addition to the title of the analyst report and checks if there is any additional information when judging the tone by the KR-FinBERT-SC (Pre-trained language model) rather than the prior tone. If we judge the information power of 473,455 analyst reports listed on the KOSPI and KOSDAQ markets during the period 2001～2023 by the cumulative abnormal returns before and after the publication date, the perpetuality of the information in the BERT tone is significantly greater than that of the prior word. Regression analysis shows that the BERT tone absorbs most of the information from the dictionary tone. The permanence of the information in the BERT tone is weaker than the rate of change in investment opinion and target, but stronger than the rate of change in earnings forecasts, and provides additional information on the variables of structured information. The BERT tone can also be measured in reports where structured information variables are not provided, and it is more useful in situations where sell recommendations are scarce because analysts have a burden to make sell recommendations. The title and summary of an analyst’s report are hard to argue with the information flow, and the two pieces of information are complementary to each other, so they should be used together to get the most out of the textual information.