Li 2008 - Bloomfield critique: What if losses are simply more complex to discuss - Result on stock option ownership favors obfuscation though - Bloomfield critique: Results on future oriented discussion consistent with: - Conservatism: Bad news recognized as it happens, good news recognized when in the past - FOG >= 18 is unreadable on page 5/225. Median is 19.24, 25th is 18.44 - What does this say about annual reports? What does it say about the measure? - Grab this data file... - Mention: Per Loughran, there is a coding error that downweights FOG. Periods weren't removed from numbers. - Also tried Flesch Reading ease and Flesch Kincaid - Why positive coefficient on FOG in positive earnings earnings persistence table? (main effect of Table 4, panel A, p15/235) - Why does "obfuscation" lead to an increase in earnings? - Table 5: Little evidence that readability predicts future earnings persistence for loss firms; only for profit firms Huang, Teoh and Zhang (Tone Management) - "We view disclosure tone as jointly determined by economic fundamentals and managerial incentives." - We don't know how filings are written - Tone of press releases is more positive on average at 0.0043 (as compared to negative in LM) 9/1091 - Consistent with other channels as well. - No idea what restatement measure they are using... GAO p1098 Key result is p1096 -- abnormal tone negatively predicts future earnings - Why does abtone not predict restatements in year t+1? And why look at longer horizons instead of simultaneous? Dyer, Lang, and Stice-Lawrence (2017) Evolution of 10-K disclosure... - "Available only for the 22,349 documents for which we can identify all sections." 8/228 - Difficulty of identifying 10-K sections - Perplexity is a bad way to test LDA models -- always goes down - Intrusion task: Give a few words from 1 topic and a word from another. Theirs is a bit easier, because they require the intruder to be in the bottom 15% of the topic instead of anywhere. - 150 topics is too much, especially for what they are trying to do.