Fundamental Analysis based on Copula Entropy

The efficient market hypothesis asserts that all accessible information is completely integrated into security prices. As a quintessential example, fundamental analysis (or financial statements analysis) provides investors with a comprehensive understanding of the operations of listed companies, thereby facilitating trading strategies in the security markets. However, a crucial issue for financial reports is the inconsistent number of disclosed financial items. Specifically, there might be significant variations in the financial statement items disclosed by different companies in the same period, or even differences in the financial statement items disclosed by the same company in different quarters, which makes it difficult to construct models using all the available data. In another academic domain, the entropy of the copula function (or Copula Entropy) serves as an information proxy to measure the dependency between two variables. Our aim is to utilize as much information from quarterly financial reports as possible to generate trading strategies. We adopt the data in the Wholesale Trade and Retail Trade sectors in North American stock markets as single-industry samples to avoid inter-industry effects during modelling, with the period ranging from January 1st, 1991, to September 30th, 2022, totalling 128 quarters. Firstly, we design an exhaustive framework for data preprocessing. The processes include data collection, feature extraction, data integration, data cleaning, data transformation, and variable selection in sequence. Notably, we also employ Copula Entropy for feature selection in models. Subsequently, we utilize the data from the previous four-quarter sliding windows as training data and predict the 30-day accumulated stock excess returns after the financial announcement day by means of two commonly used basic regression models and two classification models in financial research. Simultaneously, we assess the information content in each sliding window through Copula Entropy in 10 distinct settings. Additionally, we refine the Copula Entropy approach to measure the information content of financial statements. Finally, we construct portfolios based on both predictive outcomes and information content. The core concept is to gauge the reliability of predictive results based on information content. The more available information there is, the more reliable the results will be. As a result, over 51.25% of information-based portfolios are more profitable than benchmark portfolios, depending on the choice of the k-Nearest Neighbour parameter for estimating Copula Entropy, model selection, and trading strategies. Our refined Copula Entropy methods generate higher buy-and-hold abnormal returns on average than the basic ones. Experimental results suggest that portfolios based on information content are valid in simple models and have the potential to emerge as a new essential type of portfolios.