Görüntüleme
45
İndirme
12
Bu çalışmanın amacı Yapay Sinir Ağları’nın finans alanındaki tahmin problemlerine uygulanabilirliğinin gösterilmesi aynı zamanda da Borsa İstanbul (BIST)-100 Endeksini etkileyen değişkenler arasındaki ilişkilerin modellenmesidir. Çalışmada BIST-100 Endeksi, Gecelik Faiz Oranları ve Dolar Kuru arasındaki ilişkiler Yapay Sinir Ağlarından ileri beslemeli, geriye yayılımlı Çok Katmanlı Algılayıcı (Multi-Layered Perceptron-MLP) kullanılarak modellenmiş ve BIST-100 Endeks değeri tahmin edilmiştir. Çalışmada 4 Ocak 2010 ile 7 Ocak 2020 arasındaki 2511 iş gününe ait veriler Türkiye Cumhuriyet Merkez Bankası Elektronik Veri Dağıtım Sisteminden alınmıştır. Veri setinin % 90’ı eğitim , % 10’u test verileri olarak ayrılmış ve MATLAB 2020a ile veriler analiz edilmiştir. Çeşitli eğitim, aktivasyon ve transfer fonksiyonları arasından en uygun olanları veri seti üzerinde yapılan denemeler ile belirlenmiş, sonrasında çeşitli mimarideki modeller kurulmuş modellerin tahmin başarıları, tahmin değerler ile gerçek veriler arasındaki regresyon ilişkisi ile belirlenmiştir. Kurulan modeller ile eğitilen ağların test verileri üzerindeki simülasyonu sonrasında modellerin performansları tahmin değerlerin gerçek değerlerden sapmasını yüzde olarak veren Ortalama Mutlak Yüzde Hata (Mean Absolute Percentage Error-MAPE) değeri kullanılarak yapılmıştır. Elde edilen sonuçlara göre değişkenler arasındaki ilişkileri modelleyen ve en iyi performansı gösteren ağ, mimarisi 2-10-10-1 olan iki gizli katmanlı, her gizli katmanında 10’ar nöronu olan ağdır.
The aim of this study is to demonstrate the applicability of the Artificial Nervous Networks to predictive problems in the financial field and also to model the relationships between the variables affecting the Bursa Istanbul (BIST)-100 Index. In the study, the relationship between the BIST-100 Index, the Night Rate Rates and the Dolls Dry was modeled using the advanced, backward Multi-Layered Perceptron-MLP (Multi-Layered Perceptron-MLP) and the BIST-100 Index value was estimated. In the study, the data of 2511 working days between 4 January 2010 and 7 January 2020 were collected from the Turkish Republic Central Bank Electronic Data Distribution System. 90% of the data set was tailored to training, 10% to test data, and the data was analyzed with MATLAB 2020a. The most suitable among the various training, activation and transfer functions are determined by experiments on the data set, then models in various architectures are established; the predictive achievements of the models are determined by the regression relationship between the predictive values and the real data. Following the simulation of the test data of the established models and trained networks, the performance of the models was made using the average absolute percentage error-MAPE (Mean Absolute Percentage Error-MAPE), which gives the estimated values a percentage of deviation from the real values. According to the results obtained, the network modeling the relationships between the variables and showing the best performance is a network with two hidden layers, with architecture 2-10-10-1, with 10 neurons in each hidden layer.
The aim of this research is to apply the Artificial Neural Networks in finance problems as well as to model the relationships between variables that effect the Stock Market of Turkey. The relationship between BIST-100 index, Overnight Interest Rates and Exchange Rate of Dollar modelled by using feed-forward back propogation Multilayer Perceptron(MLP) of Artificial Neural Networks and BIST-100 index value is estimated. The data is taken from Turkish Republic Central Bank Electronic Data Distribution system. 2511 data between 4 January 2010 and 7 January 2020 of working days has been used. The % 90 of data set is used for training, % 10 is used for testing. MATLAB 2020a is used for analysis. Many training, activation and transfer functions under many architectures are used to establish the most robust model. The performance is measured by the regression relationship between the exact values and estimated values. The forecasting performance of simulation of networks on test data is measured by Mean Absolute Percentage Error-MAPE which shows the deviation of estimated values from real values as percentage. According to the results the best performed neural network that models the relationship between variables has 2 hidden layers, with 10 neurons in each hidden layer and has the architecture 2-10-10-1.
Alan : Sosyal, Beşeri ve İdari Bilimler
Dergi Türü : Uluslararası
Makale Detay 
Benzer Makaleler
PDF Görüntüle
Dergi Bilgisi
Eseri Dinleyin
Alıntı Yap
Bu Sayfayı Yazdırın
Paylaş
Mendeley
