Case Retrieval with Clustering for a Case-Based Reasoning and Inverse Problem Methodology: An Investigation of Financial Bubbles

Abstract

This paper proposes an approach for predicting abnormal asset performance in traded securities, often referred to as ‘financial bubbles’. It uses an ensemble technique based on Case-based Reasoning (CBR) and Inverse Problems (IP), which we term IPCBR. More specifically we propose a Machine Learning formative strategy to determine the causes of stock behaviour, rather than to predict future time series points in fuzzy environments. In so doing, our paper contributes to more robust strategies in investigating financial bubbles. The framework uses a geometric pattern description of historical time series and applies clustering techniques to derive a model that generalizes those patterns onto observations. The model constitutes the forward approach to the IPCBR framework; our results demonstrate that, given the target problem, our CBR model provides a computationally inexpensive description of abnormal asset performance.