Chaotic Test and Non-Linearity of Abnormal Stock Returns: Selecting an Optimal Chaos Model in Explaining Abnormal Stock Returns around the Release Date of Annual Financial Statements

For many investors, it is important to predict the future trend of abnormal stock returns. Thus, in this research, the abnormal stock returns of the listed companies in Tehran Stock Exchange were tested since 2008- 2017 using three hypotheses. The first and second hypotheses examined the non-linearity and non-randomness of the abnormal stock returns ′ trend around the release date of annual financial statements, respectively. While, the third hypothesis tested the potential of the chaos model in explaining future abnormal returns based on the past abnormal returns around the release date of the annual financial statements. For this pur-pose, BDS, Teraesvirta Neural Network, and White Neural Network tests were used to investigate its non-linearity. In addition, Lyapunov exponent, correlation dimension, Dickey-Fuller, and Hurst exponent tests were used for testing non-randomness and the fitness of AR, SETAR, and LSTAR models to determine the optimal model in explaining the abnormal returns utilizing R software. Results of these tests represented a non-linear and non-random process and chaos in the abnormal stock returns, implying the predictability of abnormal stock returns. Also, among three used chaos models, the LSTAR model had lower error and more predictability than the other two models.