ارزیابی مدل های هیبرید در پیش بینی بازده نقدی سهام شرکت های کشاورزی

Introduction: Providing an alternative route for mobilizing domestic savings, stock markets increase investment growth and, consequently, economic growth. Moreover, allocating better resources (through the concessions of labor division), the stock market plays an important role in economic growth. In addition to the positive effects of the stock market on economic growth, this market has high potential to create employment (direct and indirect). According to the existing conditions of the agriculture sector in meeting the needs of the society have increased the importance of this sector. Thus, it is necessary to use the stock exchange of the agricultural sector as one of the most important tools to solve the problems of this sector, improve market efficiency, increase productivity and provide financial requirements of agricultural companies in this market. Concerning the number of shares, stock volume and variety of products, agricultural exchange is the largest stock exchange. Cash returns can be used as a criterion to identify the economic status of agricultural firms in the stock market; identifying this trend plays an important role in financial planning and investment decisions. So, the main objective of this study was to compare the performance and accuracy of different methods to predict cash returns of agricultural companies in Tehran Stock Exchange.
Materials and Methods: The stock returns prediction models can be divided into two groups of statistical models and artificial intelligence. In this study, prediction accuracy was evaluated for linear (ARIMA and ARIMAX), non-linear (NAR and NARX) and hybrid models (ARIMA-ANN and ARIMAX-ANN) using monthly cash returns data over the period 2009(4)-2015(4) and according to the Diebold and Mariano test. ARIMA linear models have dominated many areas of time series forecasting. The linear function is based on three parametric linear components: auto regression (AR), integration (I), and moving average (MA). The ARIMA models also have the capability to include external independent or predictor variables. In this case, the model is a multivariate model and this model is represented as ARIMAX, where X represents the independent external. Artificial neural networks are capable of learning, generalizing, information parallel processing and error endurance. These attributes make the ANNs powerful in solving complex problems. Moreover, the ANN is able to adapt to the data pattern and relationship between the input and output, resulting in better prediction accuracy than the statistical method. The combined ARIMA(X)-ANN model is able to use the characteristics of both models simultaneously. Adding the explanatory variables, exchange rates, interest rates, inflation, oil prices and government spending, this study tried to answer the question whether performance prediction models improve significantly for linear, nonlinear and hybrid?
Results and Discussion: In this study, augmented Dicky Fuller tests (ADF), dickey fuller generalized least squares (DF-GLS), Phillips-Perron (PP) and Kwiatkowski–Phillips–Schmidt–Shin (KPSS) tests were used to test the unit root. According to results, time series of cash returns was stationary. Next we estimated ARIMA (2,0,0) and ARIMAX (2,0,0) models. Based on the criteria of accuracy, the results showed that the ARIMAX model had a higher predicting accuracy compared to the ARIMA. In this study, two models were designed using artificial neural networks to predict cash returns. the first model only included the first and second lag of cash returns (infact, the first model was Nonlinear Autoregressive model (NAR)); in the second model, in addition to two lags of cash returns, the exchange rates, interest rates, inflation, oil prices and government spending variables were used as the network inputs (this pattern is known as Nonlinear Autoregressive with Exogenous Input (NARX Model)). We find that the performance and accuracy of Auto Regression artificial neural networks increases by adding explanatory variables. In this study also used the ARIMA-ANN and ARIMAX-ANN hybrid models, in order to consider the linear and nonlinear characteristics in the data. After running the hybrid models and predicting, the results indicate that ARIMA-ANN and ARIMAX-ANN hybrid models have better performance in comparison with linear models (ARIMA and ARIMA) and non-linear models (NAR and NARX). Another result shows that ARIMA-ANN model is more accurate compared to ARIMAX-ANN. Finally, we show that hybrid models have greater prediction accuracy according to the Diebold and Mariano test.
Conclusions: The results showed that the addition of external explanatory variables could improve the prediction accuracy in linear and nonlinear models and this improvement in performance was significant statistically. The results of the study revealed that making use of hybrid models increased prediction accuracy significantly. Therefore, researchers, companies and shareholders of stock exchange are suggested to use the hybrid models in their financial planning to forecast economic variables.