|
|
 |
| |
|
|
|
Statistical Analysis of CEO Sentiment in Indian Annual Reports for Predicting Bank Performance Using Machine Learning |
|
|
|
PP: 669-680 |
|
|
doi:10.18576/jsap/140612
|
|
|
|
Author(s) |
|
|
|
Sunil Sharma,
S. Prasanna,
Murtaza M. Junaid Farooque,
Khaliquzzaman Khan,
|
|
|
|
Abstract |
|
|
| Leadership communication, especially through the use of media such as leaflets and letters, is increasingly being considered as one of the main direct avenues through which CEO relationships with external stakeholders are fostered. In other words, such communication by the management leadership helps to sway public opinion and eventually impacts the organization’s financial viability/performance success, the present work intends to establish this dynamic for the banking sector in India (the top five banks). The research focuses on uncovering the concealed linkage between the bank CEO’s tone in their annual shareholder letters and the bank’s financial performance measured via Return on Equity (ROE). The study covers the seven years annual letters to shareholders of the five leading Indian banks (Axis Bank, HDFC Bank, City Union Bank, ICICI Bank, and Yes Bank) and employs a lexicon-based sentiment analysis method to quantify the sentiment or the mood prevailing in the CEO letters. The study develops the bank sentiment measures from the shareholder letters, in particular the Sentiment Score, the Positive Ratio, and the Negative Ratio, through a lexicon-based sentiment analysis method. These indicators from the letters are then juxtaposed with the corresponding ROE data using both linear regression and Random Forest models. The findings of the research are contrary to what may have been expected as they reveal a mildly negative association between ROE and high positive mood levels. Thus, it reveals the possibility that the tone used in the letters might not always reflect the reality on the ground. The inference to be drawn from this is that positive communication may be utilized to mask or offset the negative performance perception occasioned by bad performing situations. Moreover, the bank-wise regression models show better results as compared to global ones which indicate that factors related to the institution and context play a significant role. The random forest model is better in terms of prediction performance than the standard regression for very low ROE values as it has a lower Root Mean Squared Error (RMSE). The research offers a new innovative model that merges sentiment data with financial indicators to forecast financial performance and measure executive effectiveness. This is very valuable and informative to the said groups of people- the analysts, the investors, and the regulators, that depend heavily on the qualitative as well as the quantitative cues in the rapidly changing data-driven financial industry landscape. |
|
|
|
|
|
|
|
