چکیده:
Most trades in the capital market are based on investors' expectations of the future value of the financial instrument in question. Therefore, having a correct understanding of how these expectations are formed is of great importance in the field of economics. Almost all economic models, especially macroeconomic models, begin by proposing specific hypotheses regarding how expectations are formed. The aim of this research is to estimate and analyze the pattern of rational expectations in the Tehran Stock Exchange in a non-linear manner. To this end, artificial neural networks, which possess great power in predicting non-linear phenomena, have been utilized. A type of these networks, multi-layer perceptron neural networks with backpropagation training algorithms, which are widely used in the field of mapping behavior modeling, has been used as the main core of the model in this article. The findings show that the actors of the Tehran Stock Exchange have rational expectations and these expectations are at a strong level. Therefore, it can be concluded that market actors follow the pattern of rational expectations in their predictions.
خلاصه ماشینی:
Investigation of Rational Expectations in the Tehran Stock Exchange using Artificial Neural Networks 9 Amir Hortamani 2 Masoud Karimkhani Received date: 2014/03/02 Accepted date: 2014/00/00 Abstract Most transactions in the capital market are carried out based on investors' expectations of the future value of the financial instrument in question.
Rational expectations are a main element of the random walk theory 2 or efficient markets 3 4 and, in relation to security prices, hyperinflation dynamics theories and permanent income consumption theories as well as an essential part in designing economic stabilization policies (Lucas and Thomas J.
In the adaptive expectations hypothesis specified by relationship 92, xt moves asymptotically toward the new level xto meaning that for several consecutive periods, the variable xt is predicted to be below the limit; in the theory of rational expectations, individuals do not make systematic errors.
King 1-2- Linking Perceptron Neural Networks with the Rational Expectations Model Applied economists are generally interested in estimating the parameters of structural equations.
4- Research Methodology In the present research, the rational expectations pattern of the Tehran Stock Exchange is estimated based on the Perceptron neural network.
(Refer to the page image) Equation 28 is converted into equation 21 based on equations 4 to 9 (Refer to the page image) y2: Final output (predicted value) f2: Second layer transfer function (sigmoid function), f1: First layer transfer function (hyperbolic function) w2: Second layer weights, w1: First layer weights x: Input variable (actual price lags), b1: Bias (first layer constant parameter) b2: Bias (second layer constant parameter) Table (2): Optimal function for estimating rational expectations Training Sum of Squares Error 0.
(
, 
, 
)
