Publication: BIST 30 Endeksinin ve Hisse Senetlerinin Değer Tahmini Üzerine Bir Çalışma
Abstract
Borsa, menkul kıymetlerin ve diğer sermaye piyasası araçlarının alınıp satılmasına olanak sağlayan platformlar olarak ifade edilmektedir. Bu alım satım işlemleri ile iki fiyat arasındaki farktan kaynaklanan zarar ve kazanç elde edilmektedir. Bu durum, finansal/para piyasalarının tahmininin hem akademik hem de iş dünyasında dikkat çekici bir konu olarak karşımıza çıkmasına olanak sağlamıştır. Bu tez çalışmasında BIST 30 endeks değeri ve bu endekste yer alan AKBNK, HALKB, GARAN, ISCTR, VAKBN ve YKBNK banka hisse senetlerinin bir gün sonraki kapanış fiyatı tahmin edilmiştir. Tahmin modelinin geliştirilmesindeki iki önemli aşama, veri kümesinin ve yöntemin belirlenmesidir. İlk aşamada tahmin modeli için fiyat verileri, temel ve teknik göstergelerden meydana gelen veri kümesi oluşturulmuştur. Bu değişkenlerden teknik göstergeler, fiyat verilerinden hesaplanmaktadır ve sayıları çok fazladır. Mevcut olan tüm teknik göstergelerin veri kümesinde kullanılmasının aksine teknik göstergelerde bir indirgeme yapılarak sayıları 16'ya azaltılmıştır. İndirgenen teknik göstergelere ek olarak temel gösterge ve fiyat verilerinin de eklenmesiyle veri kümesi son halini almıştır. İkinci aşamada ise, tahmin modeli için Uzun – Kısa Dönemli Bellek Ağları (LSTM) yöntemi kullanılmıştır. Geliştirilen modelin performansı RMSE, MAE ve MAPE metrikleri ile değerlendirilmiştir. İki farklı tahmin için geliştirilen modelin performansı literatürde yer alan çalışmalarla kıyaslanmıştır. Yapılan tahminlerde genel olarak endeks ve hisse senetlerinin yönü (artış/azalış) doğru tahmin edilmiş olup fiyatların dip ve zirve yaptığı noktalarda tahminleme performansının düşük olduğu gözlemlenmiştir. Ayrıca bu çalışmada indirgenmiş teknik göstergeler ile geliştirilen tahmin modelinin başarı performansının literatürde yer alan sadece fiyat verileri kullanılarak oluşturulan tahmin modelinin başarı performansından daha iyi olduğu gözlemlenmiştir. Ayrıca bu model ile 10.03.2020 – 10.04.2020 tarihleri arasında gerçek zamanlı tahminleme yapılmıştır. Bu tahminlemelerde genel olarak gerçek zamanlı endeks ve hisse senedi yönleri doğru tahmin edilirken fiyat ve endeks değer değişim miktarlarının tahmininde bazı sapmalar olmuştur. Ancak oluşan bu az sayıdaki sapmalara karşılık, bazı günlere ait tahminlerde modelin tahmin ettiği değerlerin gerçek değerlere çok yakın olduğu görülmüştür.
The stock exchange expressed as platforms that allow the purchase and sale of securities and other capital market instruments. With these trading transactions, losses, and gains arising from the difference between the two prices obtained. This situation has enabled the forecasting of financial/money markets to come across as a remarkable topic in both the academic and business world. In the thesis, the BIST 30 index and the price of AKBNK, HALKB, GARAN, ISCTR, VAKBN and YKBNK bank stocks in this index have been tried to predict the closing prices of the next day. Two important stages in the development of the prediction model are the determination of the dataset and the method. In the first phase, a data set consisting of price data, basic and technical indicators created for the forecast model. Technical indicators, one of this variables, are calculated from price data and are very numerous. In contrast to the fact that all available technical indicators used in the dataset, their numbers reduced by a reduction in this study. In addition to the 16 technical indicators determined as a result of the reduction, the data set became final with the addition of basic indicators and price data. In the second phase, the method of long-short term memory networks is used for the prediction model. The performance of the developed model evaluated by RMSE, MAE, and MAPE metrics. The performance of the forecast model developed for two different predictions has compared to studies in the literature. In general, the direction of the index and stocks (increase/decrease) is estimated correctly and the performance of forecasting is low at the points where prices dip and peak. In this study, it observed that the successful performance of the forecast model developed with reduced technical indicators was better than the successful performance of the forecast models created using only price data in the literature. Also, real-time index and stock price estimations made between 10 March 2020 to 10 April 2020 date with this model. While the real-time index and stock aspects were estimated correctly in these estimates, there were some deviations in the estimation of price and index value change amounts. However, in response to these few deviations, it was often observed that the values predicted by the model were very close to the actual values.
The stock exchange expressed as platforms that allow the purchase and sale of securities and other capital market instruments. With these trading transactions, losses, and gains arising from the difference between the two prices obtained. This situation has enabled the forecasting of financial/money markets to come across as a remarkable topic in both the academic and business world. In the thesis, the BIST 30 index and the price of AKBNK, HALKB, GARAN, ISCTR, VAKBN and YKBNK bank stocks in this index have been tried to predict the closing prices of the next day. Two important stages in the development of the prediction model are the determination of the dataset and the method. In the first phase, a data set consisting of price data, basic and technical indicators created for the forecast model. Technical indicators, one of this variables, are calculated from price data and are very numerous. In contrast to the fact that all available technical indicators used in the dataset, their numbers reduced by a reduction in this study. In addition to the 16 technical indicators determined as a result of the reduction, the data set became final with the addition of basic indicators and price data. In the second phase, the method of long-short term memory networks is used for the prediction model. The performance of the developed model evaluated by RMSE, MAE, and MAPE metrics. The performance of the forecast model developed for two different predictions has compared to studies in the literature. In general, the direction of the index and stocks (increase/decrease) is estimated correctly and the performance of forecasting is low at the points where prices dip and peak. In this study, it observed that the successful performance of the forecast model developed with reduced technical indicators was better than the successful performance of the forecast models created using only price data in the literature. Also, real-time index and stock price estimations made between 10 March 2020 to 10 April 2020 date with this model. While the real-time index and stock aspects were estimated correctly in these estimates, there were some deviations in the estimation of price and index value change amounts. However, in response to these few deviations, it was often observed that the values predicted by the model were very close to the actual values.
Description
Citation
WoS Q
Scopus Q
Source
Volume
Issue
Start Page
End Page
71
