An Integrated Wavelet Transform-Convolutional Neural Network Based Deep Learning Approach and An Application On Stock Exchange Estimation

Hamit Erdal; Selçuk Korucuk

Makale detayları

An Integrated Wavelet Transform-Convolutional Neural Network Based Deep Learning Approach and An Application On Stock Exchange Estimation

Yazarlar:

Hamit Erdal | Selçuk Korucuk

Yayıncı:

Giresun University

Ana Başlıklar:

Ekonometrik ve İstatistiksel Yöntemler

Dergi:

Karadeniz Sosyal Bilimler Dergisi Vol.15 Issue 29, pp.387-404

Belge Tipi:

Makale

Yayın Tarihi:

12/2023

Belge Dili:

Türkçe

Document ID (Org No.):

2001304731

pISSN:

1309-081X

Anahtar kelimeler:

Özet:

 Stock market estimation is one of the most studied topics in financial estimation studies. Developing a better estimation tool or methodology for the stock market, which is a risk-intensive investment tool, is invaluable for investors. In this study, a stock market estimation study was carried out with the daily data of Borsa Istanbul BIST Industrial index XUSIN. In this context, the opening prices of the 5,000-day index 31.12.2001-31.12.2021 were utilized and three models were developed for estimating. In Model 1, estimations were conducted with values that 1 day behind; in Model 2, with values that 3 days behind and in Model 3, with values that 7 days behind, respectively. Convolutional Neural Network CNN and Wavelet Transform Convolutional Neural Network WTCNN, which are deep learning methods whose effectiveness has been demonstrated in many studies, were utilized as estimation methods. Thus, the effect of wavelet transforms which decomposes dataset into subsets for a stationary situation for estimation performance was investigated. It was concluded that the estimation success could be increased with the DDESA approach and it could be used as an effective estimation tool.

 Stock market estimation is one of the most studied topics in financial estimation studies. Developing a better estimation tool or methodology for the stock market, which is a risk-intensive investment tool, is invaluable for investors. In this study, a stock market estimation study was carried out with the daily data of Borsa Istanbul BIST Industrial index XUSIN. In this context, the opening prices of the 5,000-day index 31.12.2001-31.12.2021 were utilized and three models were developed for estimating. In Model 1, estimations were conducted with values that 1 day behind; in Model 2, with values that 3 days behind and in Model 3, with values that 7 days behind, respectively. Convolutional Neural Network CNN and Wavelet Transform Convolutional Neural Network WTCNN, which are deep learning methods whose effectiveness has been demonstrated in many studies, were utilized as estimation methods. Thus, the effect of wavelet transforms which decomposes dataset into subsets for a stationary situation for estimation performance was investigated. It was concluded that the estimation success could be increased with the DDESA approach and it could be used as an effective estimation tool.

 Stock market estimation is one of the most studied topics in financial estimation studies. Developing a better estimation tool or methodology for the stock market, which is a risk-intensive investment tool, is invaluable for investors. In this study, a stock market estimation study was carried out with the daily data of Borsa Istanbul BIST Industrial index XUSIN. In this context, the opening prices of the 5,000-day index 31.12.2001-31.12.2021 were utilized and three models were developed for estimating. In Model 1, estimations were conducted with values that 1 day behind; in Model 2, with values that 3 days behind and in Model 3, with values that 7 days behind, respectively. Convolutional Neural Network CNN and Wavelet Transform Convolutional Neural Network WTCNN, which are deep learning methods whose effectiveness has been demonstrated in many studies, were utilized as estimation methods. Thus, the effect of wavelet transforms which decomposes dataset into subsets for a stationary situation for estimation performance was investigated. It was concluded that the estimation success could be increased with the DDESA approach and it could be used as an effective estimation tool.

 Finansal tahminleme çalışmalarında üzerinde en fazla çalışılan konulardan biri borsa tahminidir. Risk yoğun bir yatırım aracı olan borsa için iyi bir tahmin aracının veya metodolojisinin geliştirilebilmesi yatırımcılar için paha biçilemez önemdedir. Bu çalışmada, Borsa İstanbul Sınai endeksi günlük verisi ile bir borsa tahmini çalışması gerçekleştirilmiş ve borsanın açık olduğu 5.000 günlük 31.12.2001-31.12.2021 endeks açılış fiyatları kullanılarak tahmin için 3 model kurulmuştur. Model 1.de 1 gün geriden gelen değerlerle, Model 2.de 3 gün geriden gelen değerlerle ve Model 3.de ise 7 gün geriden gelen değerlerle tahmin yapmıştır. Tahmin yöntemi olarak etkinliği pek çok çalışmada ortaya konulmuş bir derin öğrenme yöntemi olan Evrişimsel Sinir Ağları ESA ve Dalgacık Dönüşümü DD ile önişleme tabi tutulmuş ESA DDESA yöntemleri kullanılmıştır. Böylece durağan bir durum için veri kümesini alt kümelere ayrıştıran dalgacık dönüşümünün tahmin performansına etkisi araştırılmıştır. Çalışmanın sonucunda DDESA yaklaşımı ile tahmin başarısının artırılabildiği ve etkin bir tahminleme aracı olarak kullanılabileceği sonucuna ulaşılmıştır.

Makale detayları

An Integrated Wavelet Transform-Convolutional Neural Network Based Deep Learning Approach and An Application On Stock Exchange Estimation

Yazarlar:

Hamit Erdal | Selçuk Korucuk

Yayıncı:

Giresun University

Ana Başlıklar:

Ekonometrik ve İstatistiksel Yöntemler

Dergi:

Karadeniz Sosyal Bilimler Dergisi Vol.15 Issue 29, pp.387-404

Belge Tipi:

Makale

Yayın Tarihi:

12/2023

Belge Dili:

Türkçe

Document ID (Org No.):

2001304731

pISSN:

1309-081X

Stock Exchange

Estimation

Deep Learning

Wavelet Transform

Convolutional Neural NetworkBorsa

Tahminleme

Derin Öğrenme

Dalgacık Dönüşümü

Evrişimsel Sinir Ağları

Sayı

Makale detayları

Yazarlar: Hamit Erdal | Selçuk Korucuk

Yayıncı: Giresun University

Ana Başlıklar: Ekonometrik ve İstatistiksel Yöntemler

Dergi: Karadeniz Sosyal Bilimler Dergisi Vol.15 Issue 29, pp.387-404

Belge Tipi: Makale

Yayın Tarihi: 12/2023

Belge Dili: Türkçe

Document ID (Org No.): 2001304731

pISSN: 1309-081X

Sayı

Yazarlar:

Hamit Erdal | Selçuk Korucuk

Yayıncı:

Giresun University

Ana Başlıklar:

Ekonometrik ve İstatistiksel Yöntemler

Dergi:

Karadeniz Sosyal Bilimler Dergisi Vol.15 Issue 29, pp.387-404

Belge Tipi:

Makale

Yayın Tarihi:

12/2023

Belge Dili:

Türkçe

Document ID (Org No.):

2001304731

pISSN:

1309-081X