No full-text available
To read the full-text of this research,
you can request a copy directly from the authors.
Top-Down Influence? Predicting CEO Personality and Risk Impact from Speech Transcripts
Abstract
How much does a CEO’s personality impact the performanceof their company? Management theory posits a great influence, but it is difficult to show empirically—there is a lack of publicly available self-reported personality data of top managers. Instead, we propose a text-based personality regressor based on crowd-sourced Myers–Briggs Type Indicator (MBTI) assessments. The ratings have a high internal and external validity and can be predicted with moderate to strong correlations for three out of four dimensions. Providing evidence for the upper echelons theory, we demonstrate that the predicted CEO personalities have explanatory power of financial risk.
... With ongoing advancements in computation, Large Language Models (LLMs) have experienced a significant surge in popularity. Pre-trained LLMs outperform state-of-the-art models in downstream NLP tasks, particularly personality prediction (Kazameini et al., 2020;Theil et al., 2023). ...
... This development continues to reshape the societal processes across diverse domains including social media, online education, business functions, and the electoral process (Alexander et al., 2020). Various researchers have executed APP in diverse contexts from the study of CEO risk-taking personalities, personality-job fit, and brand-follower personality matches to music recommendation systems tailored to personality (Wynekoop and Walz, 2000;Tang et al., 2018;Tomat et al., 2021;Kleć et al., 2023;Theil et al., 2023). ...
... This big data analysis has led to the fine-grained investigation of critical phenomena such as social network analysis (Letzring and Human, 2014), public opinion (Christian et al., 2021;Berggren et al., 2024), and social influence on political mobilization during electoral events (Cui and Qi, 2017;Tandera et al., 2017;Tadesse et al., 2018). Moreover, time spent on social media can help decipher the emotional states of smartphone users' indicating boredom and loneliness (Kazameini et al., 2020;Theil et al., 2023). In turn, such emotional states and tones can help recognize user demographic traits . ...
Introduction
Recent advancements in Natural Language Processing (NLP) and widely available social media data have made it possible to predict human personalities in various computational applications. In this context, pre-trained Large Language Models (LLMs) have gained recognition for their exceptional performance in NLP benchmarks. However, these models require substantial computational resources, escalating their carbon and water footprint. Consequently, a shift toward more computationally efficient smaller models is observed.
Methods
This study compares a small model ALBERT (11.8M parameters) with a larger model, RoBERTa (125M parameters) in predicting big five personality traits. It utilizes the PANDORA dataset comprising Reddit comments, processing them on a Tesla P100-PCIE-16GB GPU. The study customized both models to support multi-output regression and added two linear layers for fine-grained regression analysis.
Results
Results are evaluated on Mean Squared Error (MSE) and Root Mean Squared Error (RMSE), considering the computational resources consumed during training. While ALBERT consumed lower levels of system memory with lower heat emission, it took higher computation time compared to RoBERTa. The study produced comparable levels of MSE, RMSE, and training loss reduction.
Discussion
This highlights the influence of training data quality on the model's performance, outweighing the significance of model size. Theoretical and practical implications are also discussed.
Automatic detection of the four MBTI personality dimensions from texts has recently attracted noticeable attention from the natural language processing and computational linguistic communities. Despite the large collections of Twitter data for training, the best systems rarely even outperform the majority-class baseline. In this paper, we discuss the theoretical reasons for such low results and present the insights from an annotation study that further shed the light on this issue.
We present a novel approach for measuring executive personality traits. Relying on recent developments in machine learning and artificial intelligence, we utilize the IBM Watson Personality Insights service to measure executive personalities based on CEOs’ and CFOs’ responses to questions raised by analysts during conference calls. We obtain the Big Five personality traits – openness, conscientiousness, extraversion, agreeableness, and neuroticism – based on which we estimate risk tolerance. To validate these traits, we first demonstrate that our risk‐tolerance measure varies with existing inherent and behavioural‐based measures (gender, age, sensitivity of executive compensation to stock return volatility, and executive unexercised‐vested options) in predictable ways. Second, we show that variation in firm‐year level personality trait measures, including risk tolerance, is largely explained by manager characteristics, as opposed to firm characteristics and firm performance. Finally, we find that executive inherent risk tolerance helps explain the positive relationship between client risk and audit fees documented in the prior literature. Specifically, the effect of CEO risk‐tolerance – as an innate personality trait – on audit fees is incremental to the effect of increased risk appetite from equity risk‐taking incentives (Vega). Measuring executive personality using machine‐learning algorithms will thus allow researchers to pursue studies that were previously difficult to conduct. This article is protected by copyright. All rights reserved
Self-report questionnaires are the most commonly used personality assessment despite longstanding concerns that self-report responses may be distorted by self-protecting motives and response biases. In a large-scale meta-analysis ( N = 33,033; k = 152 samples), we compared the means of self- and informant reports of the same target's Big Five personality traits to examine the discrepancies in two rating sources and whether people see themselves more positively than they are seen by others. Inconsistent with a general self-enhancement effect, results showed that self-report means generally did not differ from informant-report means (average δ = -.038). Moderate mean differences were found only when we compared self-reports with stranger reports, suggesting that people are critical of unacquainted targets. We discuss implications of these findings for personality assessment and other fields in which self-enhancement motives are relevant.
Many ecological- and individual-level analyses of voting behaviour use multiple regressions with a considerable number of independent variables but few discussions of their results pay any attention to the potential impact of inter-relationships among those independent variables—do they confound the regression parameters and hence their interpretation? Three empirical examples are deployed to address that question, with results which suggest considerable problems. Inter-relationships between variables, even if not approaching high collinearity, can have a substantial impact on regression model results and how they are interpreted in the light of prior expectations. Confounded relationships could be the norm and interpretations open to doubt, unless considerable care is applied in the analyses and an extended principal components method for doing that is introduced and exemplified.
Facebook is rapidly gaining recognition as a powerful research tool for the social sciences. It constitutes a large and diverse pool of participants, who can be selectively recruited for both online and offline studies. Additionally, it facilitates data collection by storing detailed records of its users' demographic profiles, social interactions, and behaviors. With participants' consent, these data can be recorded retrospectively in a convenient, accurate, and inexpensive way. Based on our experience in designing, implementing, and maintaining multiple Facebook-based psychological studies that attracted over 10 million participants, we demonstrate how to recruit participants using Facebook, incentivize them effectively, and maximize their engagement. We also outline the most important opportunities and challenges associated with using Facebook for research, provide several practical guidelines on how to successfully implement studies on Facebook, and finally, discuss ethical considerations. (PsycINFO Database Record
This paper describes a survey of 280 project managers that reveals both their personality types (via Myers‐Briggs personality inventory) and their success in project management. The results show that a project manager's personality is better suited for functioning with partial data and under ambiguity than the rest of the population. These traits were found for both women and men. The conclusion is that project managers (females and males) have a unique personality‐type distribution that distinguishes them from the general population. The findings can contribute to better understanding the traits that characterize the project management population, and their relationship to project success.
Response bias continues to be the most frequently cited criticism of personality testing for personnel selection. The authors meta-analyzed the social desirability literature, examining whether social desirability functions as a predictor for a variety of criteria, as a suppressor, or as a mediator. Social desirability scales were found not to predict school success, task performance, counterproductive behaviors, and job performance. Correlations with the Big Five personality dimensions, cognitive ability, and years of education are presented along with empirical evidence that (a) social desirability is not as pervasive a problem as has been anticipated by industrial-organizational psychologists, (b) social desirability is in fact related to real individual differences in emotional stability and conscientiousness, and (c) social desirability does not function as a predictor, as a practically useful suppressor, or as a mediator variable for the criterion of job performance. Removing the effects of social desirability from the Big Five dimensions of personality leaves the criterion-related validity of personality constructs for predicting job performance intact. (PsycINFO Database Record (c) 2012 APA, all rights reserved)
The Myers-Briggs Type Indicator (MBTI; Myers & McCaulley, 1985) was evaluated from the perspectives of Jung's theory of psychological types and the five-factor model of personality as measured by self-reports and peer ratings on the NEO Personality Inventory (NEO-PI; Costa & McCrae, 1985b). Data were provided by 267 men and 201 women ages 19 to 93. Consistent with earlier research and evaluations, there was no support for the view that the MBTI measures truly dichotomous preferences or qualitatively distinct types; instead, the instrument measures four relatively independent dimensions. The interpretation of the Judging-Perceiving index was also called into question. The data suggest that Jung's theory is either incorrect or inadequately operationalized by the MBTI and cannot provide a sound basis for interpreting it. However, correlational analyses showed that the four MBTI indices did measure aspects of four of the five major dimensions of normal personality. The five-factor model provides an alternative basis for interpreting MBTI findings within a broader, more commonly shared conceptual framework.
Pi (pi) and kappa (kappa) statistics are widely used in the areas of psychiatry and psychological testing to compute the extent of agreement between raters on nominally scaled data. It is a fact that these coefficients occasionally yield unexpected results in situations known as the paradoxes of kappa. This paper explores the origin of these limitations, and introduces an alternative and more stable agreement coefficient referred to as the AC1 coefficient. Also proposed are new variance estimators for the multiple-rater generalized pi and AC1 statistics, whose validity does not depend upon the hypothesis of independence between raters. This is an improvement over existing alternative variances, which depend on the independence assumption. A Monte-Carlo simulation study demonstrates the validity of these variance estimators for confidence interval construction, and confirms the value of AC1 as an improved alternative to existing inter-rater reliability statistics.
Financial risk, defined as the chance to deviate from return expectations, is most commonly measured with volatility. Due to its value for investment decision making, volatility prediction is probably among the most important tasks in finance and risk management. Although evidence exists that enriching purely financial models with natural language information can improve predictions of volatility, this task is still comparably underexplored. We introduce PRoFET, the first neural model for volatility prediction jointly exploiting both semantic language representations and a comprehensive set of financial features. As language data, we use transcripts from quarterly recurring events, so-called "earnings calls"; in these calls, the performance of publicly traded companies is summarized and prognosticated by their management. We show that our proposed architecture, which models verbal context with an attention mechanism, significantly outperforms the previous state-of-the-art and other strong baselines. Finally, we visualize this attention mechanism on the token-level, thus aiding interpretability and providing a use case of PRoFET as a tool for investment decision support.
Predicting financial risk is an essential task in financial market. Prior research has shown that textual information in a firm's financial statement can be used to predict its stock's risk level. Nowadays, firm CEOs communicate information not only verbally through press releases and financial reports, but also nonverbally through investor meetings and earnings conference calls. There are anecdotal evidences that CEO's vocal features, such as emotions and voice tones, can reveal the firm's performance. However, how vocal features can be used to predict risk levels, and to what extent, is still unknown. To fill the gap, we obtain earnings call audio recordings and textual transcripts for S\&P 500 companies in recent years. We propose a multimodal deep regression model (MDRM) that jointly model CEO's verbal (from text) and vocal (from audio) information in a conference call. Empirical results show that our model that jointly considers verbal and vocal features achieves significant and substantial prediction error reduction. We also discuss several interesting findings and the implications to financial markets. The processed earnings conference calls data (text and audio) are released for readers who are interested in reproducing the results or designing trading strategy.
RESEARCH SUMMARY
We draw upon applied psychology literature to explore inter‐agent differences in perceived risk to their equity when making strategic risk decisions. Our theory suggests behavioral agency’s predicted negative relationship between equity risk bearing and strategic risk taking is contingent upon four personality traits. Our empirical analyses, based on personality profiles of 158 CEOs of S&P 1,500 firms in manufacturing industries, indicates the relationship between executive risk bearing and strategic risk taking crosses from negative to positive for high extraversion, greater openness and low conscientiousness. These findings demonstrate that agency based predictions of CEO risk taking in response to compensation – and board attempts at creating incentive alignment using compensation – are enhanced by integrating insights from personality trait literature.
MANAGERIAL SUMMARY
We study the effect of CEO personality on their behavioral responses to stock option pay. Our findings reveal that CEOs that score high on extraversion or openness and low on conscientiousness are less likely to decrease their firm’s strategic risk taking as the value of their stock options increases. That is, the tendency of CEOs to become more risk averse in their strategic choices as their option wealth increases (due to loss aversion) is weaker for highly extraverted and more open CEOs, but stronger for more conscientiousness CEOs. Overall, our findings suggest that board of directors need to consider personality traits of their CEOs when designing compensation packages with the intention to align incentives of CEOs with shareholder risk preferences.
In the analysis of data it is often assumed that observations y1, y2, …, yn are independently normally distributed with constant variance and with expectations specified by a model linear in a set of parameters θ. In this paper we make the less restrictive assumption that such a normal, homoscedastic, linear model is appropriate after some suitable transformation has been applied to the y's. Inferences about the transformation and about the parameters of the linear model are made by computing the likelihood function and the relevant posterior distribution. The contributions of normality, homoscedasticity and additivity to the transformation are separated. The relation of the present methods to earlier procedures for finding transformations is discussed. The methods are illustrated with examples.
This analysis considers the impact of the top managers in an organization on the organization's outcomes, specifically strategic choices and performance levels. The focus is not on the chief executive alone, but rather on the entire top management team. Using a macro view, these organizational outcomes are perceived to be related to the values and cognitive bases of those high-power individuals in the organization. In developing the model, emphasis is on the background characteristics of the top managers as opposed to the psychological dimensions. A series of propositions that should be tested to support the upper echelons theory are presented. The topics of these propositions include age, functional track, other career experiences, education, socioeconomic roots, financial position, and group characteristics. The creation of this model is just the beginning of the work that is necessary to evaluate and understand the upper echelons theory. Further input is needed from areas such as the executive recruiting industry. Additionally, clinical and statistical studies are both necessary to fully develop this theory. (SRD)
Inter-rater reliability studies are used in a diverse set of fields. Often, these investigations involve three or more raters, and thus, require the use of indices such as Fleiss’s kappa, Conger’s kappa, or Krippendorff’s alpha. Through two motivating examples – one theoretical and one from practice – this paper exposes limitations of these indices when the units to be rated are not well-distributed across the rating categories. Then, using a Monte Carlo simulation and information visualizations, we argue for the use of two alternative indices, the Brennan-Prediger coefficient and Gwet’s AC2, because the agreement levels reported by these indices are more robust to variation in the distribution of units that raters encounter. The paper concludes by exploring the complex, interwoven relationship between the number of levels in a rating instrument, the agreement level present among raters, and the distribution of units that are to be scored.
Earnings call summarizes the financial performance of a company, and it is an important indicator of the future financial risks of the company. We quantitatively study how earnings calls are correlated with the financial risks, with a special focus on the financial crisis of 2009. In particular, we perform a text regression task: given the transcript of an earnings call, we predict the volatility of stock prices from the week after the call is made. We propose the use of copula: a powerful statistical framework that separately models the uniform marginals and their complex multivariate stochastic dependencies, while not requiring any prior assumptions on the distributions of the covariate and the dependent variable. By performing probability integral transform, our approach moves beyond the standard count-based bag-ofwords models in NLP, and improves previous work on text regression by incorporating the correlation among local features in the form of semiparametric Gaussian copula. In experiments, we show that our model significantly outperforms strong linear and non-linear discriminative baselines on three datasets under various settings.
Quarterly earnings conference calls are becoming a more pervasive tool for corporate disclosure. However, the extent to which the market embeds information contained in the tone (i.e. sentiment) of conference call wording is unknown. Using computer aided content analysis, we examine the incremental informativeness of quarterly earnings conference calls and the corresponding market reaction. We find that conference call linguistic tone is a significant predictor of abnormal returns and trading volume. Furthermore, conference call tone dominates earnings surprises over the 60 trading days following the call. The question and answer portion of the call has incremental explanatory power for the post-earnings-announcement drift and this significance is primarily concentrated in firms that do not pay dividends, illustrating differences in investor behavior based on the level of cash flow uncertainty. Additionally, we find that a context specific linguistic dictionary is more powerful than a more widely used general dictionary (Harvard IV-4 Psychosocial).
This study investigated the relationship between two of the most widely used personality measures, the Revised NEO Personality Inventory and the Myers-Briggs Type Indicator. A total of 900 participants completed the NEO PI-R and the MBTI. Correlational analysis of the personality measures showed that NEO PI-R Extraversion was correlated with MBTI Extraversion-Introversion, Openness was correlated with Sensing-Intuition, Agreeableness with Thinking-Feeling and Conscientiousness with Judging-Perceiving, replicating the findings of McCrae and Costa (1989).
This paper considers some appropriate and inappropriate uses of coefficient kappa and alternative kappa-like statistics. Discussion is restricted to the descriptive characteristics of these statistics for measuring agreement with categorical data in studies of reliability and validity. Special consideration is given to assumptions about whether marginals are fixed a priori, or free to vary. In reliability studies, when marginals are fixed, coefficient kappa is found to be appropriate. When either or both of the marginals are free to vary, however, it is suggested that the "chance" term in kappa be replaced by 1/n, where n is the number of categories. In validity studies, we suggest considering whether one wants an index of improvement beyond "chance" or beyond the best a priori strategy employing base rates. In the former case, considerations are similar to those in reliability studies with the marginals for the criterion measure considered as fixed. In the latter case, it is suggested that the largest marginal proportion for the criterion measure be used in place of the "chance" term in kappa. Similarities and differences among these statistics are discussed and illustrated with synthetic data.
The convergent validity between self and observer ratings of the Big Five dimensions of personality was examined by cumulating research findings across studies. The mean correlation corrected for coefficient α in self-ratings and inter-rater reliability in observer ratings was .46 for agreeableness (N=6359, k=53), .56 for conscientiousness (N=6754, k=58), .51 for emotional stability (N=8000, k=55), .62 for extraversion (N=7725, k=50), and .59 for openness to experience (N=5333, k=38). Results indicate that, although there is a high degree of construct overlap, both self and observer ratings have substantial unique variance. Moderator effects were analyzed. The duration of acquaintance (strangers vs close relatives) as well as observer type (peers at work vs relatives) were analyzed. Acquaintanceship had a large moderating effect whereas observer type did not moderate the level of convergence.
Previous research uses negative word counts to measure the tone of a text. We show that word lists developed for other disciplines misclassify common words in financial text. In a large sample of 10-Ks during 1994 to 2008, almost three-fourths of the words identified as negative by the widely used Harvard Dictionary are words typically not considered negative in financial contexts. We develop an alternative negative word list, along with five other word lists, that better reflect tone in financial text. We link the word lists to 10-K filing returns, trading volume, return volatility, fraud, material weakness, and unexpected earnings.
The proportion of U.S. firms paying dividends drops sharply during the 1980s and 1990s. Among NYSE, AMEX, and Nasdaq firms, the proportion of dividend payers falls from 66.5% in 1978 to only 20.8% in 1999. The decline is due in part to an avalanche of new listings that tilts the population of publicly traded firms toward small firms with low profitability and strong growth opportunities—the timeworn characteristics of firms that typically do not pay dividends. But this is not the whole story. The authors' more striking finding is that, no matter what their characteristics, firms in general have become less likely to pay dividends.
The authors use two different methods to disentangle the effects of changing firm characteristics and changing propensity to pay on the percent of dividend payers. They find that, of the total decline in the proportion of dividend payers since 1978, roughly one-third is due to the changing characteristics of publicly traded firms and two-thirds is due to a reduced propensity to pay dividends. This lower propensity to pay is quite general—dividends have become less common among even large, profitable firms.
Share repurchases jump in the 1980s, and the authors investigate whether repurchases contribute to the declining incidence of dividend payments. It turns out that repurchases are mainly the province of dividend payers, thus leaving the decline in the percent of payers largely unexplained. Instead, the primary effect of repurchases is to increase the already high payouts of cash dividend payers.
This study sets out examine the relationship between two personality measures—most popularly used measure in the consultancy and training world (the Myers-Briggs Type Indicator) and one of the most heavily used measures in the academic research area on personality (the five factor NEO-PI). One hundred and sixty adults completed the NEO-PI and the MBTI. The NEO-PI Agreeableness score was correlated only with the thinking-feeling (T-F) dimension; the NEO-PI Conscientiousness score was correlated with both thinking-feeling and judging-perceiving (J-P) dimension; the NEO-PI Extraversion score was strongly correlated with the extraversion-introversion (E-I) dimensions, while the Neuroticism score from the NEO-PI was not related to any MBTI subscale score. The openness dimension was correlated with all four especially sensing-intuitive. These results were related to two other similar comparative studies. Results are discussed in terms of recent criticisms of the MBTI.
Despite widespread agreement that multi-method assessments are optimal in personality research, the literature is dominated by a single method: self-reports. This pattern seems to be based, at least in part, on widely held preconceptions about the costs of non-self-report methods, such as informant methods. Researchers seem to believe that informant methods are: (a) time-consuming, (b) expensive, (c) ineffective (i.e., informants will not cooperate), and (d) particularly vulnerable to faking or invalid responses. This article evaluates the validity of these preconceptions in light of recent advances in Internet technology, and proposes some strategies for making informant methods more effective. Drawing on data from three separate studies, I demonstrate that, using these strategies, informant reports can be collected with minimal effort and few monetary costs. In addition, informants are generally very willing to cooperate (e.g., response rates of 76–95%) and provide valid data (in terms of strong consensus and self-other agreement). Informant reports represent a mostly untapped resource that researchers can use to improve the validity of personality assessments and to address new questions that cannot be examined with self-reports alone.
The five-factor model of personality is a hierarchical organization of personality traits in terms of five basic dimensions: Extraversion, Agreeableness, Conscientiousness, Neuroticism, and Openness to Experience. Research using both natural language adjectives and theoretically based personality questionnaires supports the comprehensiveness of the model and its applicability across observers and cultures. This article summarizes the history of the model and its supporting evidence; discusses conceptions of the nature of the factors; and outlines an agenda for theorizing about the origins and operation of the factors. We argue that the model should prove useful both for individual assessment and for the elucidation of a number of topics of interest to personality psychologists.
BERT: Pre-training of Deep Bidirectional Transformers for Language Understanding
- J Devlin
- M W Chang
- K Lee
- K Toutanova
Devlin, J.; Chang, M. W.; Lee, K.; and Toutanova, K. 2019.
BERT: Pre-training of Deep Bidirectional Transformers for
Language Understanding. Proceedings of NAACL, 4171-4186.
- Y Liu
- M Ott
- N Goyal
- J Du
- M Joshi
- D Chen
- O Levy
- M Lewis
- L Zettlemoyer
- V Stoyanov
Liu, Y.; Ott, M.; Goyal, N.; Du, J.; Joshi, M.; Chen, D.;
Levy, O.; Lewis, M.; Zettlemoyer, L.; and Stoyanov, V.
2019. RoBERTa: A Robustly Optimized BERT Pretraining
Approach. arXiv preprint, arXiv: 1907.11692.
Using Linguistic Cues for the Automatic Recognition of Personality in Conversation and Text
- S M Lundberg
- S.-I Lee
- I Guyon
- U V Luxburg
- S Bengio
- H Wallach
- R Fergus
- S Vishwanathan
- R Garnett
- M R Mehl
- R K Moore
Lundberg, S. M.; and Lee, S.-I. 2017. A Unified Approach
to Interpreting Model Predictions. In Guyon, I.; Luxburg,
U. V.; Bengio, S.; Wallach, H.; Fergus, R.; Vishwanathan,
S.; and Garnett, R., eds., Advances in Neural Information
Processing Systems, volume 30. Curran Associates, Inc.
Mairesse, F.; Walker, M. A.; Mehl, M. R.; and Moore, R. K.
2007. Using Linguistic Cues for the Automatic Recognition
of Personality in Conversation and Text. Journal of Artificial
Intelligence Research, 30: 457-500.
The Development and Psychometric Properties of LIWC2015. White paper, University of Texas at Austin. Plank, B.; and Hovy, D. 2015. Personality Traits on Twitter-or-How to Get 1500 Personality Tests in a Week
- J W Pennebaker
- R L Boyd
- K Jordan
- K Blackburn
Pennebaker, J. W.; Boyd, R. L.; Jordan, K.; and Blackburn,
K. 2015. The Development and Psychometric Properties of
LIWC2015. White paper, University of Texas at Austin.
Plank, B.; and Hovy, D. 2015. Personality Traits on
Twitter-or-How to Get 1500 Personality Tests in a Week.
In Proceedings of the 6th Workshop on Computational Approaches to Subjectivity, Sentiment and Social Media Analysis, 92-98.
- K Theil
- D Hovy
- H Stuckenschmidt
Theil, K.; Hovy, D.; and Stuckenschmidt, H. 2022. Top-Down Influence? Predicting CEO Personality and Risk
Impact from Speech Transcripts. arXiv preprint, arXiv:
2201.07670.
HTML: Hierarchical Transformer-Based Multi-Task Learning for Volatility Prediction
- L Yang
- T L Ng
- B Smyth
- R Dong
Yang, L.; Ng, T. L.; Smyth, B.; and Dong, R. 2020. HTML:
Hierarchical Transformer-Based Multi-Task Learning for
Volatility Prediction. In Proceedings of WWW, 441-451.