This paper examines the tone and content of the forward-looking statements (FLS) in the Management Discussion and Analysis section (MD&A) of corporate 10-K and 10-Q filings using a Naïve Bayesian machine learning algorithm. I first manually categorize 30,000 sentences of randomly-selected FLS extracted from the MD&As along two dimensions: (1) tone (i.e., positive versus negative tone); and (2) content (i.e., profitability, operations, liquidity etc.). These manually-coded sentences are then used as training data in a Naïve Bayesian machine learning algorithm to classify the tone and content of about 13 million forward-looking statements from more than 140,000 corporate 10-K and 10-Q MD&As between 1994 and 2007. I find that firms with better current performance, lower accruals, smaller size, lower market-to-book ratio, and less return volatility tend to have more positive forwardlooking

Abstract: In this study, we measure managerial affective states during earnings conference calls by analyzing conference call audio files using vocal emotion analysis software. We hypothesize and find that when managers are scrutinized by analysts during conference calls, positive and negative affect displayed by managers are informative about the firm’s financial future. In particular, we find that managers exhibiting positive (negative) affect are positively (negatively) related to contemporaneous stock returns and future unexpected earnings. However, analysts do not incorporate the information when determining short term earnings forecasts. When making stock recommendation changes, however, analysts incorporate positive affect but not negative affect. We observe market underreaction to negative affect as if market participants follow analyst recommendation changes. Together, this study presents new evidence that managerial vocal cues contain useful information about firms ’ fundamentals, incremental to both quantitative earnings information and qualitative “soft ” information conveyed by the linguistic content. We appreciate the assistance of Amir Liberman and Albert De Vries of Nemesysco for helpful discussions and for

Abstract—Traditional movie gross predictions are based on numerical and categorical movie data. But since the 1990s, text sources such as news have been proven to carry extra and meaningful information beyond traditional quantitative finance data, and thus can be used as predictive indicators in finance. In this paper, we use the quantitative news data generated by Lydia, our system for large-scale news analysis, to help us to predict movie grosses. By analyzing two different models (regression and k-nearest neighbor models), we find models using only news data can achieve similar performance to those use numerical and categorical data from The Internet Movie Database (IMDB). Moreover, we can achieve better performance by using the combination of IMDB data and news data. Further, the improvement is statistically significant. I.

We examine how product similarity and competition influence mergers and acquisitions and the ability of firms to exploit product market synergies through asset complementarities. Using novel text-based analysis of firm 10-K product descriptions, we find three key results. (1) Firms are more likely to enter mergers with firms whose language describing their assets is similar. (2) Transactions in competitive product markets with similar acquirer and target firms experience increased stock returns and real longer-term gains in cash flows and higher growth in their product descriptions. (3) These gains are higher when the target is less similar to the acquirer’s closest rivals, and when firms have the potential for unique products. Our findings are consistent with firms merging and buying assets to exploit asset complementarities and to create new products to increase product differentiation.

Firms often disclose information security risk factors in public filings such as 10-K reports. The internal information associated with disclosures may be positive or negative. In this paper, we are interested in evaluating how the nature of security risk factors disclosed, which is believed to represent the internal information regarding information security, is associated with future breach announcements. For this purpose, we build a decision tree model, which classifies the occurrence of future security breaches based on the textual contents of the disclosed security risk factors. The model is able to accurately associate disclosure characteristics with breach announcements about 77 % of the time. We further explore the contents of the security risk factors using text mining techniques to provide a richer interpretation of the results. The results show that the security risk factors with action-oriented terms and phrases are less likely to be related to future incidents. We also conduct a cross-sectional analysis to study how the market interprets the nature of information security risk factors in annual reports at different time points. We find that the market reaction following the security breach announcement is different

A growing stream of research in accounting and finance tests the extent to which the tone of financial disclosure narrative, also referred to as its qualitative information, affects security prices, over and above the disclosed financial performance. These studies typically measure tone by counting the relative frequency of positive versus negative words in a given disclosure such as earnings press releases. Critical to word-frequency based analysis is the list of words deemed to be positive or negative. Because general wordlists (GI or Diction) likely omit words that would be considered positive or negative in the context of financial disclosure and include words that would not, we expect that these general wordlists be less powerful for hypothesis testing compared to wordlists specifically for the domain of financial disclosure (FD). Using a sample of 29,712 earnings press releases, we find that the context-specific FD wordlist produces a more powerful predictor of market reaction than the general wordlists. Additionally, in smaller samples – demonstrated here with 250 regressions using randomly-selected subsamples ranging in size from 50 to 2,000 – the domain-specific FD wordlist retains predictive ability, with rejection rates exceeding 97 percent for samples of 2,000 while the rejection rates for the general wordlists are less than 30 percent. The FD wordlist also performs better than an alternative, domain-specific wordlist. Overall, our findings indicate that the domain-specific FD wordlist provides an alternative, more powerful measure of tone for capital markets researchers. Finally, we show that equal weighting of word occurrences is more intuitive, easier to implement, and more amenable to replication than alternative sample-dependent weighting methodologies advocated by certain concurrent research.

We use a Naïve Bayesian computational linguistics procedure to code the tone of debt analysts ’ discussions about events that potentially generate debt-equity conflicts of interest. Debt analysts ’ views, as published in their investment reports, are expected to reflect the net effect of such conflict-events on the wealth of debt holders by taking into account the level of protection embedded in debt contracts. We document that debt analysts routinely discuss these conflict events and their interpretation is less negative when debt holders are protected by more restrictive covenants. We also provide evidence that debt analysts ’ views on conflict events are

I investigate variation in how well firm-initiated disclosures are transmitted to investors. Improved dissemination is hypothesized to lower the cost of information acquisition and increase firm visibility. Through instrumentation, I examine the impact of differential dissemination of firm-initiated disclosures by the press on bid-ask spreads, trading volume, and idiosyncratic volatility. I find that improved transmission causally lowers bid-ask spreads, increases trading volume, and lowers idiosyncratic volatility. Ultimately, the results suggest that the economic impact of a firm’s disclosure program is affected by both the level of disclosure and the effectiveness of mechanisms in distributing firm-initiated information to investors. This paper is based on my doctoral dissertation. I would like to thank my committee members, Douglas Skinner (chair), Christian Leuz, Abbie Smith, and Suraj Srinivasan, for their many insightful comments and suggestions. I

Very preliminary Please do not quote without permission We investigate whether vocal and linguistic cues elicited from speech are helpful in detecting misreporting. We conduct an experiment where we provide participants with varying levels of incentives to misreport about their performance in an SAT test and generate a sample of misreporters and truth-tellers. All subjects are subsequently interviewed about their performance and their videotaped responses to a series of automated and pre-scripted questions are analyzed using both vocal emotional analysis software and linguistic analysis software. We predict that individuals who overstated their performance will feel cognitive dissonance during the interview. Nonverbal vocal cues are able to detect this dissonance and successfully classified individuals as misreporters or truth-tellers 71 % of the time. Verbal cues from linguistic analysis show only marginal ability to detect misreporting in our setting. These results are informative to regulators and market participants such as analysts and auditors who are interested in uncovering financial misreporting. We thank Dan Ariely and Laureen Maines for helpful comments and discussions. We also appreciate suggestions

Abstract—Traditional movie gross predictions are based on numerical and categorical movie data from The Internet Movie Database (IMDB). In this paper, we use the quantitative news data generated by Lydia, our system for large-scale news analysis, to help people to predict movie grosses. By analyzing two different models (regression and k-nearest neighbor models), we find models using only news data can achieve similar performance to those using IMDB data. Moreover, we can achieve better performance by using the combination of IMDB data and news data. Further, the improvement is statistically significant. I.