Patterns ID focuses on research-backed approaches to understanding stock market dynamics. Below is a collection of some of the academic papers that inspire, support and critique our approach: using statistics to model asset price changes. Other pages in this section present relevant studies on each indicator used throughout the web app. While the papers here are not exhaustive on the subject, they provide a starting point for Patterns ID users to understand the research behind our approach to modeling markets. Our priority in listing this research is to provide accessible resources for our users, which means at times other papers that aren’t accessible to the public were not included.

Supporting Research:

• Evaluation of Forecasting Methods from Selected Stock Market Returns by M. Mallikarjuna and R. Prabhakara Rao (2019): This paper evaluates the effectiveness of various forecasting methods applied to stock market returns, including linear, nonlinear, artificial intelligence (AI), frequency domain, and hybrid models. The study focuses on the predictive performance of these methods across different market types—developed, emerging, and frontier markets. The findings reveal that while nonlinear models like the Self-Exciting Threshold Autoregressive (SETAR) model often outperform others, traditional linear and nonlinear models generally provide more accurate forecasts than AI and frequency domain models. This emphasizes the enduring relevance of classical statistical approaches such as ARIMA in forecasting, even when compared to more advanced AI models.
• Market Efficiency, Market Anomalies, Causes, Evidences, and Some Behavioral Aspects of Market Anomalies by Madiha Latif, Shanza Arshad, Mariam Fatima, and Samia Farooq (2011): This paper provides an extensive review of the Efficient Market Hypothesis (EMH) and its related anomalies. It discusses various types of market anomalies, including calendar, fundamental, and technical anomalies, and how they deviate from the principles of market efficiency. The authors highlight the contradictions between EMH and observed market behaviors, such as the weekend effect, January effect, and low P/E ratio anomaly. The paper also delves into behavioral finance explanations for these anomalies, suggesting that psychological factors and investor sentiment play significant roles in the persistence of these irregularities. Additionally, the study addresses how different financial theories attempt to explain or account for these anomalies, emphasizing the ongoing debate in the finance literature.
• Machine Learning Techniques and Data for Stock Market Forecasting: A Literature Review by Mahinda Mailagaha Kumbure, Christoph Lohrmann, Pasi Luukka, Jari Porras (2022): This comprehensive literature review investigates the application of machine learning techniques in stock market prediction, analyzing 138 journal articles published between 2000 and 2019. The review categorizes variables used for stock market predictions, such as technical indicators, macroeconomic variables, and fundamental indicators, and offers an in-depth examination of machine learning methods employed in the studies. A key focus of the paper is comparing the efficacy of traditional linear and nonlinear models, such as ARIMA and GARCH, against more modern machine learning approaches. Notably, the review highlights that traditional models often outperform more advanced AI-based methods in providing accurate forecasts, particularly in capturing linear and some nonlinear patterns in stock prices. This finding challenges the growing reliance on AI techniques, suggesting that traditional models still hold significant value in stock market forecasting.

Criticizing Research:

• Stupid Data Miner Tricks: Overfitting the S&P 500 by David J. Leinweber (2000): This paper humorously illustrates the dangers of overfitting in financial data analysis. Using absurd examples, such as correlating the S&P 500 index with butter production in Bangladesh, the author highlights how data mining can lead to spurious and meaningless correlations when enough variables are tested. The paper emphasizes the importance of avoiding overfitting by not relying solely on in-sample performance and encourages the use of out-of-sample testing and holdback samples. The author also discusses how random data or unrelated variables can appear to predict financial outcomes due to sheer chance, underscoring the risks of relying on overly complex models without sound theoretical grounding.