The CEO's speech
How does a CEO's personality influence the performance of their firm? In several episodes, the market has punished flippant remarks or polarizing tweets, but can we formally measure personality and detect its financial impact? In a work, we developed a text-based personality prediction model based on CEO speeches. Having predicted the personality of a large sample of American CEOs, we
inspected correlations with the financial risk of their firms. Overall, we found that – all things equal – firms with CEOs of the MBTI introverted, feeling, and sensing type tend to face increased risk. Notably, our results are robust to CEO age and gender and several financial variables such as firm size and industry.
How does this work, exactly? You have probably heard of the Myers–Briggs Type Indicator, a popular tool for self-exploration. It classifies personality along the dimensions "extraversion–introversion" (E–I), "sensing–intuition" (S–N), "thinking–feeling" (T–F), and "judging–perceiving" (J–P). We found a publicly available database of celebrity MBTI scores based on crowd votes. As the data also contained votes for the best-known CEOs such as Elon Musk or Steve Jobs, we were motivated to explore is as an alternative to self-reported personality.
A drawback of the MBTI is that the academic psychology literature disputes its validity. Research shows that it is not meaningful to binarize personality (i.e., extraverted or introverted) or classify it into 16 distinct types. First, we were therefore interested to explore the validity of the crowd-sourced MBTI measure compared to the more established Big 5 model. Confirming prior research, we found that the MBTI and the Big 5 correlate moderately to strongly. This shows that the MBTI seems to be a meaningful proxy for personality assessments despite its criticisms. Furthermore, we found that the overall agreement between the crowd voters was high, most notably for E–I and J–P. This further attests to the validity of our measure.
We then trained a powerful language model predicting the MBTI of all 32 available CEOs in our dataset. As input, we used transcripts of their external investor communication. Then, we automatically inferred the personality of 22K American CEOs with a previously unknown MBTI personality. Taking the predicted personalities together with complementary data such as CEO age, gender, and fundamental financial variables (e.g., industry, firm size and valuation), we found a significant and robust correlation with future financial risk. Looking at the different dimensions of personality, CEOs communicating in an "introverted" and "feeling" manner tend to face increased risk. In contrast, "intuitive" communication tends to be associated with decreased risk.
Our results have far-reaching implications for management theory, as it has long been hypothesized that the characteristics of a firm's top management also reflect in the firm's performance. Until now, however, only a few studies have been able to use labeled personality data of CEOs. As we also collected the speech recordings of external CEO communication, we plan to explore this data in more depth in the future. It is likely that features such as voice pitch and vocal inflection impact perceived trustworthiness and competence, which should also reflect in financial risk measures.
On a cautionary note, our work has several limitations: First, the personality labels are based on crowd voters instead of self-reports. However, we accepted this constraint as it is unfeasible to have CEOs such as Elon Musk answer extensive questionnaires. Second, the training data for the personality prediction model is likely biased, since most CEOs in our sample are male, American, and tech industry leaders with a specific MBTI configuration. Hence, we expect our model to be less generalizable to other demographics, cultures, industries, and personalities. Moreover, we would like to advocate against the use of our or similar models for automated profiling and down-stream applications, e.g., hiring and investment decisions: As text-based personality predictors only work with a certain accuracy on average and overall for a large sample, they cannot be applied to single instances, i.e., persons. The risk of false generalizations, unequal treatment, and increasing societal inequity is too high. Therefore, while our results are encouraging, they also demand further research, validation, and prudent use of the proposed method.