Predicting investment behavior: An augmented reinforcement learning model
Predicting investment behavior: An augmented reinforcement learning model
复制标题
DOI：
10.1016/j.neucom.2008.11.031
复制
发表时间：
2009-10
期刊：
影响因子：
6
通讯作者：
T. Shimokawa;Kyoko Suzuki;T. Misawa;Yoshitaka Okano
中科院分区：
文献类型：
--
作者：
T. Shimokawa;Kyoko Suzuki;T. Misawa;Yoshitaka Okano
本文的目标是增强顺序时间差型（TD型）强化学习模型，以检测最合适的学习模型的人类决策过程中的金融投资任务。TD型学习模型的简单性和鲁棒性令人着迷。然而，现有的证据和我们的观察表明，在学习中引入非线性效应的必要性和其他因素可能在投资决策过程中发挥重要作用的可能性。为了扩展有序TD型学习模型，我们采用三层感知器作为基函数和分层贝叶斯方法来校准参数值。预测检验的结果表明，本文所构建的TD型学习模型与其他学习模型相比，能够较好地避免过拟合问题，并能较好地预测人们的投资行为。
The goal of this paper is to augment the ordinal temporal-difference type (TD-type) reinforcement learning model in order to detect the most suitable learning model of the human decision-making process in financial investment tasks. The simplicity and robustness of the TD-type learning model is fascinating. However, the available evidence and our observation suggest the necessity of introducing the nonlinear effect in learning and the possibility that additional factors might play important roles in the investment decision-making process. To extend the ordinal TD-type learning model, we adopt a three-layered perceptron as the basis function and the hierarchical Bayesian method to calibrate the parameter values. The result of the predictive test suggests that the augmented TD-type learning model constructed in this paper can evade the overfitting and can predict people's investment behavior well as compared to other familiar learning models.