*Radwa Magdy, Cairo University
Discussant: Ke Wang, University of Alberta
In responding to recent research calls (e.g. Beyer et al. 2010; and Baginski et al. 2014), for developing a sound measure of disclosure quality using an innovative natural language processing technique (Berger, 2011), the researcher contributes to disclosure studies in two principal ways. First, a quality- based framework for Management Discussion & Analysis (MD&A) preparation is introduced. A novel feature of this framework is that it captures all qualitative attributes of (MD&A) and develops a measure for each of those attributes. Each MD&A principle is operationalized through measurable definition(s). Each measure is derived from relevant research. These measures are then aggregated in a single corporate score representing the overall quality of MD&A disclosure. Further tests show that the proposed measure is reliable and valid. Second, the extent to which the well settled notion in the literature that disclosure quantity is a proper proxy for disclosure quality is examined. The analysis shows that disclosure quantity is not a good proxy for disclosure quality. In addition, it shows that determinants of disclosure quality and quantity differ. The proposed framework has various research and policy implications. It suggests new research avenues on the determinants and economic consequences of MD&A disclosure quality. Such research may inform both regulators and managers as to the costs and benefits of disclosure quality to both firms and stakeholders.
*Peter Clarkson, University of Queensland, Albert Tsang, York University, Jordan Ponn, Gordon D. Richardson, Frank Rudzicz & Jingjing Wang, University of Toronto
Discussant: Carol Pomare, Mount Allison University
We employ computer-based textual analysis to examine disclosure patterns for a sample of U.S. CSR reports from the period 2002-2016. Using report length, we observe a positive relationship between CSR performance and disclosure level, as predicted by signaling theory. The conclusion is further supported by our Latent Dirichlet Allocation (LDA) model results, namely, good CSR performers cover more topics and exhibit greater homogeneity of topic coverage, compared to poor CSR performers. The two CSR performance types differ not only in “how much they say” in CSR reports, but also in “what they say” and “how they say it”. We find that poor CSR performers devote more of their CSR report to areas of CSR strength but less to areas of CSR concern. This selective disclosure behavior is consistent with the predictions of legitimacy theory (i.e., “greenwashing”). Finally, our machine learning model reveals that various other linguistic features, in addition to the level of disclosure, are important for revealing performance type. In particular, our linguistic analyses suggest that good CSR performers: are more specific and advanced in their writing; are generally more sociable, friendly and cooperative; and exhibit features suggesting greater ambition, achievement, and level of sophistication, consistent with their proactive CSR strategies. Our results potentially expand the information set that can be used to ascertain a firm’s true CSR performance type. Further, our results are potentially useful to analysts and investors when they are provided with CSR disclosures by private firms.
*Karel Hrazdil, Simon Fraser University, Jiri Novak, Charles University in Prague, Rafael Rogo, Indiana University, Christine I. Wiedman, University of Waterloo & Ray Zhang, University of British Columbia
Discussant: Yamin Hao, University of Alberta
In this paper, we present a novel approach for measuring CEO personality traits. Relying on recent developments in machine learning and artificial intelligence, we use the IBM Watson Personality Insights service to measure personality based on transcripts of Q&A sessions of conference calls made by CEOs. We measure the Big Five personality traits – Openness, Conscientiousness, Extraversion, Agreeableness, and Neuroticism – and compute a measure for risk tolerance based on these five traits. We conduct a number of validation tests to show that all of the firm-year level personality trait measures are CEO-specific and not related to firm characteristics and firm performance. We then examine a well-established relationship between risk tolerance and audit fees. We find that our proxy for high CEO risk tolerance is associated with significantly higher audit fees and that this association is consistent across several Big Five index components. In supplementary tests, we find evidence that the influence of the CEO’s risk tolerance on audit fees is greater when CEO power is greater and that firms with more risk tolerant CEOs engage in more aggressive reporting practices. Consistent with upper echelons theory, our results suggest that CEO personalities have an incremental impact on auditors’ assessment of engagement risk.
*Michael H.R. Erkens, Erasmus University Rotterdam, Ying Gan, & B. Burcin Yurtoglu, WHU - Otto Beisheim School of Management
Discussant: Derek Oler, Texas Tech University
Using a comprehensive linguistic analysis, we develop a Clawback Strength Index and show that while some firms adopt unambiguous and strong clawback provisions, others adopt more symbolic and weak ones. We find that strong clawback adopters experience (a) improvements in financial reporting quality, (b) a decrease in the likelihood of CEO turnover, and (c) lower total and incentive-based compensation. We advance two possible explanations for our findings. On the one hand, clawback strength may be primarily responsible for the improvements in reporting quality (the causal explanation). On the other hand, strong clawback provisions may yield benefits because they are part of a broader reform package (the broader reform explanation). While our findings on financial reporting quality and CEO turnover are consistent with either the causal or the broader reform explanation, our results on CEO compensation support only the broader reform explanation.