法人說明會語調對未來盈餘反應係數之影響

Abstract

本論文利用未來盈餘反應係數 (future earnings response coefficient, FERC) 探討法人說明會語調是否能夠提供增額資訊，以協助投資人修正對於未來盈餘的預期，進而反應在股價上。在語調衡量方面，本論文使用FinBERT 模型及Loughran and Mcdonald (2011)編製的字典來衡量法說會語調。為比較兩個方法，本論文先進行情感分類的測試，從分類測試的結果可以發現FinBERT 模型不僅整體表現較佳，在各個情感類別上的表現也優於字典法。在實證方面，本論文發現越正向的法人說明會語調具有更高的未來反應係數，顯示正向的語調能夠提供較多關於未來盈餘的資訊，此結果與過往研究一致，顯示法人說明會語調確實具有資訊內涵。本論文並未發現法人說明會不同階段的語調對未來盈餘反應係數有明顯差異，顯示不同階段語調所提供有關未來盈餘資訊量差異很小。另外，本論文發現透過FinBERT模型衡量的語調在實證上較顯著，顯示FinBERT模型不僅能更準確地衡量文本情緒，且更適合用於財經及會計方面的情感分析研究。

This paper uses future earnings response coefficient (FERC) to test whether the tone of the conference calls can assist investors in predicting future earnings and reflect the expectations in stock prices. To accurately measure the tone, this paper utilizes the state-of-the-art natural language processing (NLP) algorithm - FinBERT model, and uses the Loughran and Mcdonald (2011) word lists as a benchmark. To check the performance of them, this paper does a classification task first. The results show that the FinBERT model not only achieves better overall performance but also surpasses the word list method in predicting sentiment for each sentiment class. In terms of empirical results, this paper finds weak evidence that firms with more positive tone will have higher FERC, suggesting that positive tone can provide more information and better assist investors in predicting future performance. The findings are consistent with the results in previous research that the linguistic features in conference calls can provide additional information and support the findings that positive tone can reduce uncertainty of firm’s future value. This paper does not find significant FERC difference between the tone in discussion section and the tone in presentation section. The results may suggest that the tone in different section provide almost same amount of information in predicting future earnings. Notably, the paper demonstrates that the tone measured by the FinBERT model gets greater significance than the tone measured by the word list method, suggesting that the FinBERT model not only gauges the sentiment more accurately but is also more suitable for sentiment analysis in financial and accounting research.