Zaman serileri analizi ve yapay sinir ağları ile tahmin: Yabancı portföy yatırımları üzerine uygulama
No Thumbnail Available
Files
Date
2009
Authors
Journal Title
Journal ISSN
Volume Title
Publisher
Sosyal Bilimler Enstitüsü
Abstract
To forecast the amount and direction of movements of money is critical importance for fund managers in global financial market which was affected by information technology every day. Therefore, the model used to predict future value of a variable is playing significant role for managers while making a decision. When fund managers made decision for investing international portfolio, they have to evaluate push and pull factors of the country?s to make the decision about time to enter the country and leave the country and also size of this movements. In this framework, work in foreign portfolio investment are thought to affect a variety of factors, January 1997 - December 2008 covers the period by using monthly data, different evaluation approaches and the models used in forecasting models are compared success. The first method of these is ARIMA models, time series analysis method, based on the lagged values of the variable. It is found in the study that the forecasting performance with ARMA is only 17 percent. This performance, of course, is not enough to make some inferences, so the other method is used, as well as ARMA, which is called VAR. VAR model uses affecting factors on the dependent variable with their lagged values with the lagged values of the variable. The analysis of VAR result gives 81 percent forecasting performance, which is relatively acceptable result. This result is high, but the difference of variables to get the name of ensuring stability of the original variables to bring the values and limitations of interpretation of variables in nonlinear analysis such as in the linear direction of operation is to make predictions with Artificial Neural Networks. Artificial Neural Networks do not have any limitations which were seen in ARMA and VAR models. Variables which are used in VAR model with their lagged values are used in Artificial Neural Networks with their original values. Artificial Neural Networks gave 95 percent of model performance result. These results showed that Artificial Neural Networks have more successful results than Time Series Analysis models.
Description
Keywords
Zaman serileri analizi, Yapay sinir ağları