Feel the Market : An Attempt to Identify Additional Factor in the Capital Asset Pricing Model (CAPM) Using Generative Pre-Trained Transformer (GPT) and Bidirectional Encoder Representations from Transformers (BERT)

In the past year, generative AI, led by Chat GPT by Open AI and Bard, a language experiment by Google, generated tremendous attention amongst the public. Their impact on finance, law and general productivity are difficulty to articulate with words. This paper aims to strengthen the well-roundness of Capital Asset Pricing Model (CAPM) by taking sentiments that the public have on a particular stock into consideration. The impact of stories or narratives have been well discussed in past literature, most notably by Shiller (2018) and Baker and Wurgler (2007). However, the impact of narratives on individual stocks and their pricing have yet to be discussed. This paper aims to fill this gap by using the sentiment analysis, applying GPT and BERT on the Financial Times, a well-respected financial newspaper, to understand the impact of narratives on individual stocks. We found i) an increase in explanatory power in sentiment scaled CAPM compared with an outright CAPM. ii) a Revenue Generating Process that is able to systematically generate trading profit using sentiment analysis as trading signal, under specific assumptions and restrictions. The code used for this paper is available at: https://github.com/lz458/EmotionalMarket