Essays on textual analysis in accounting and finance

Abstract

This thesis comprises of three essays that adopts textual analysis techniques in accounting and finance research. In the first essay, we contribute to the literature on the intersection between the lexical features of firms’ financial disclosures and prices leading earnings. Specifically, we examine the relationship between the tone of financial disclosure narratives and share price anticipation of earnings. In efficient markets, if managers disclose incrementally informative (misleading) narratives on the firm-fundamentals, then the tone is expected to improve (deteriorate) the share price informativeness of future earnings. However, managerial incentives to disclose incrementally informative versus misleading narratives is expected to differ between profit firms and loss firms. Using a sample of US 10- K disclosures, we find that as the tone increases, current period returns of profit (loss) firms become more (less) informative of future earnings. Segregating the tone into separate positive and negative tonal components suggests that with increases in both the positive tone and the negative tone, current period returns in profit (loss) firms become more (less) informative of future earnings. Additional analysis reveals that the association between the tone and the share price informativeness of future earnings is stronger in short disclosures than in long disclosures, suggesting that investors find the tone in short disclosures to be better predictors of future earnings than long disclosures. In the second essay, we contribute to the literature on the relationship between the tone of financial disclosure narratives and capital investments. Specifically, we examine conditions where managers have differential incentives to disclose incrementally informative and misleading investment narratives. First, we argue that managers have fewer incentives to disclose misleading investment narratives if their content can be verified from concurrently disclosed numbers. Consistent with this argument, we find that the tone of a sample of 10-K disclosures is positively associated with current-period investments, suggesting that managers disclose incrementally informative narratives on the investment level. Second, we argue that, when the investment outcomes hamper their interests, managers have heightened incentives to disclose misleading investment efficiency narratives, as investment efficiency is not readily verifiable from concurrently disclosed numbers. Consistent with this argument, we find that the tone is more negatively associated with investment efficiency when firms: (a) undertake large vis a vis small investments (b) undertake vis a vis do not undertake new investments in the year (c) overinvest vis a vis underinvest and (d) decrease vis a vis increase investment efficiency. Overall, our results suggest that managers may disclose misleading narratives when the investment outcomes misalign with their interests. In the third essay, we contribute to the literature on the informativeness of firms’ artificial intelligence (AI) disclosures for reported financial performance. We employ a sample of 10-K disclosures to examine whether the association between AI disclosures and the reported cash flows and earnings numbers is related to the concurrent disclosures of four textual narrative characteristics: (i) forward-looking information (ii) causal attributions (iii) performance indicators and (iv) strategy-related information. We find that AI words exhibit a weak positive association with reported cash flows but no clear association with reported earnings. However, with increased disclosures of: (i) forward-looking words (ii) causal attribution words (iii) performance indicator words and (iv) strategy words, the associations of AI words and both reported cash flows and earnings is more positive. This suggests that although AI words are typically not very informative of the reported performance, their informativeness improves with increased disclosures of these four textual characteristics. Our additional analysis reveals that with the disclosure of positive risk words, but not negative risk words, the association between AI words and reported performance is more positive. We interpret these findings as evidence of the inter-relatedness of the informativeness of AI disclosures with that of other textual narrative characteristics.