چکیده:
Modeling and predicting stock returns has always been one of the challenges for researchers and investors. Hence, different methods and models have been proposed, most of which have been based on assumptions such as the distribution of returns. The kernel distribution and mixture of normal distributions were examined to predict stock return in the present study. To this end, kernel functions and mixtures of normal distributions and related parameters have been estimated using maximization of likelihood function and quartiles 99%, 95% and 90% were computed for each of distributions and for 30 superior enterprises listed in Tehran Security and Exchange (TSE) at first quarter in 2019 as predictor values of stock return. In order to determine precision of prediction methods, MSE and PRED error criteria were employed and the findings showed that mixture of normal distributions and kernel approximation might propose favorable predictions for 5-day stock returns in quartiles 90% of return distribution. Comparison of precision between two methods indicated that kernel approximation, as a non parametric method for prediction of returns, leads to higher precision than mixture of normal distributions.
خلاصه ماشینی:
For this purpose, kernel functions, normal mixtures, and their related parameters were estimated through maximum likelihood estimation, and the 99%, 95%, and 90% quantiles of each distribution were calculated for the top 30 stock market companies in the fourth quarter of the year 1398 as return prediction values.
To determine the accuracy of the prediction methods, MSE and PRED error criteria were used, and the results showed that both the mixture of normal distributions and kernel approximation, through the 90% quantile of the return distribution, can provide desirable predictions of 5-day stock returns.
For this purpose, kernel functions and normal mixtures and their related parameters were estimated through maximizing the likelihood function (equivalently, minimizing the Akaike Information Criterion), and the 99%, 95%, and 90% quantiles of each distribution were calculated for each of the 30 studied companies as return prediction values (the value that the return does not exceed with a specific confidence level).
To determine the accuracy of the prediction methods, MSE and PRED error criteria were used according to the definitions provided in the text, and the results showed that both normal distribution mixture and kernel approximation can provide desirable predictions of 5-day stock returns through the 90% quantile of the return distribution.
The results also showed that the mixture of 17 normal distributions has a more accurate fit on the stock return data of companies compared to the kernel function.
Hypothesis 3: Predicting stock returns based on the mixture of normal distributions has higher accuracy compared to the kernel method.
(
, 
, 
)
