Shahid Beheshti University Financial Management Perspective 2645-4637 11 34 2021 08 23 A study on the characteristics of TSE index return data and introducing a regime switching prediction method based on neural networks A study on the characteristics of TSE index return data and introducing a regime switching prediction method based on neural networks 145 171 102120 10.52547/jfmp.11.34.145 FA Amin Aminimehr Masters student in Financial Management, Ershad Damavand Higher Education Institute, Tehran, Iran. 0000-0002-9539-4617 Saeed Bajalan Assistant prof, Department of Financial Management and Insurance, University of Tehran, Tehran, Iran. Hanieh Hekmat Assistant prof, Department of Accounting, Al Zahra University, Tehran, Iran. 0000-0002-8094-3249 Journal Article 2021 06 05 This research has aimed at studying the characteristics and data generation process of TSE index daily return. Applying various tests showed that return data of TSE index follows a chaotic and clustered behavior. Furthermore, beside the condition of efficiency in this market, a novel prediction method is developed. The method introduced in this paper is formed from two consecutive neural networks; a mixture density neural network and a Long short-term memory neural network. It is worthy of note that the proposed method is associated with the inferred statistical structure from the data.  The entire model is compiled in order to predict TSE index considering various number of regimes using daily data December 2008 up to April 2021. Results from various statistical tests rejected the weak form of efficiency and manifested a chaotic behavior in TSE index return. Furthermore, the developed prediction method gained higher accuracy than the same method without considering regimes. Results from Diebold-Mariano test significantly implied the differences of the accuracy between the models with regimes and without regimes. Finally, a back test by considering transaction cost showed that the strategy based on the predicted direction of the model with regimes gains higher return than market benchmark and the model without regimes. This research has aimed at studying the characteristics and data generation process of TSE index daily return. Applying various tests showed that return data of TSE index follows a chaotic and clustered behavior. Furthermore, beside the condition of efficiency in this market, a novel prediction method is developed. The method introduced in this paper is formed from two consecutive neural networks; a mixture density neural network and a Long short-term memory neural network. It is worthy of note that the proposed method is associated with the inferred statistical structure from the data.  The entire model is compiled in order to predict TSE index considering various number of regimes using daily data December 2008 up to April 2021. Results from various statistical tests rejected the weak form of efficiency and manifested a chaotic behavior in TSE index return. Furthermore, the developed prediction method gained higher accuracy than the same method without considering regimes. Results from Diebold-Mariano test significantly implied the differences of the accuracy between the models with regimes and without regimes. Finally, a back test by considering transaction cost showed that the strategy based on the predicted direction of the model with regimes gains higher return than market benchmark and the model without regimes. https://jfmp.sbu.ac.ir/article_102120_951f6d7750eefdd7df4830317e29cc89.pdf