Evolution Of Learning Behavior Of The Noise Trader Sentiment

Why do observed prices of financial assets deviate from their underlying value? And to what degree can this deviation be eliminated by arbitrage? Both of the questions are very popular in financial research recently. The noise trader theory, especially the "noise trader risk" model of De Long et al (1990), has an impressive influence in this field. However, the assumption of this model that noise traders' sentiment follows a white noise process is severely challenged by real investor behavior and related empirical evidences. In order to preserve and enhance the value of noise trader theory, this paper explores the possibility that the sentiment of noise traders is auto-correlated as well as the influence if it is possible, based on the learning behavior of noise traders.This paper investigates how noise traders learn to form their sentiment and how the evolution of their sentiment influences the financial markets, by introducing two kinds of adaptive learning processes into the two-generation overlapping model of De Long et al. Given the noise traders are irrational agents, the two adaptive learning processes consist of reinforcement learning about sentiment based on price expectation performances and best reply learning about sentiment based on market returns. By the theoretical analysis of these two learning processes and the empirical research on the investor sentiment in China's stock markets, this paper mainly discusses three questions as follows. Firstly, is it feasible to find how the average sentiment of all noise traders form and evolve when they own certain learning ability, and to show that their sentiment is not independent distributed? Secondly, if a part of or all noise traders can form their sentiment by certain learning way, can the group of noise traders survive in the financial markets in a long term and play a significant role in asset pricing? Finally, what is the relationship between market performance and noise traders' auto-correlated sentiment?The theoretical analysis about the two sentiment learning processes shows thatthe noise traders' average sentiment may be auto-correlated, resulting in the auto-correlation of prices and returns, if they have certain learning ability. In addition, whether the noise traders can play a significant role in asset pricing depends on the specific learning way. However, once noise traders can do this, their group can survive in the market in a long run. Finally, the sentiment of noise traders who have certain learning ability may be affected significantly by past market returns regardless of the learning way they adopt. If noise traders form their sentiment based on market returns, this influence exists whether they play a significantly role in asset pricing or not.The empirical evidence about the sentiment of retailed and institutional investors in China's stock markets supports the two theoretical conclusions that noise traders' sentiment is auto-correlated and is influenced by past market returns. However, the role of noise traders in asset pricing can not be tested because the investors' sentiment is pessimistic throughout the full sample period.In summary, this paper is of great significance for financial research and practice by investigating the evolution of noise traders' sentiment in a more realistic perspective. In term of theoretical value, this paper can provide more feasible or better explanations for the found evidence in financial markets, especially for the mean reversion in asset returns, the over-reaction of asset prices to fundamental changes as well as the self-reinforcement and subsequent collapse of technical analysis. In addition, the evidences about the investor sentiment in China's stock markets not only support the theoretical conclusions of this paper, but also show there are obvious differences in sentiment's formation and influence between the retailed and institutional stock investors in China. All of these can help regulators and participants of financial markets understand investor behavior more deeply, which is the practical value of this paper.