Designing of an Early Warning System for the Prediction of Price Bubbles in the Tehran Stock Exchange Using a Deep Learning Approach

Articles in Press

Document Type : Research Paper

Authors

1 Ph.D. Candidate, Department of Financial Management, Kish International Campus, Tehran University, Kish, Iran.

2 Department of Financial Engineering, Faculty of Management, University of Tehran, Tehran, Iran.

3 Department of Statistics, Faculty of Basic Sciences, Bu-Ali Sina University, Hamadan, Iran.

Abstract

The main objective of this study is the design of an EWS based on an LSTM architecture, capable of timely forecasting price bubbles in the TSE. Another objective is to compare the accuracy and quality of the results with an LR model using evaluation metrics such as AUC-ROC and the confusion matrix. The system’s performance was tested on five selected TSE indices. Monthly data for the selected indices from 2002 to 2023 were extracted, and bubble periods were identified using the GSADF test with the creation of a binary variable. Considering price changes of warning indicators, the modeling of the bubble variable was performed. The proposed system, adopting a deep-learning approach, achieved a predictive accuracy in the range of 73% to 81%, with the highest for the Total Index at 81% and the lowest for the Basic Metals Index at 73%. A comparison of predictive accuracy between the LSTM and LR models shows improved accuracy across all selected indices under the LSTM. The evaluation metrics indicate a superior performance of the LSTM model relative to the LR model. To the best of our knowledge, no EWS designed for predicting price bubbles using deep learning has been reported in the literature.

Keywords

Main Subjects

Journal of Asset Management and Financing

Articles in Press, Accepted Manuscript
Available Online from 09 September 2025

Files

Share

How to cite

Statistics