Türkiye'deki finansal yatırım araçlarına yönelik tahminlemede zaman serileri analizi ve derin öğrenme tekniklerinin karşılaştırılması
No Thumbnail Available
Date
2022
Authors
Journal Title
Journal ISSN
Volume Title
Publisher
Open Access Color
OpenAIRE Downloads
OpenAIRE Views
Abstract
Günümüzde zamanı etkili kullanmanın ve geleceği öngörmenin kişi ve kurumları daha avantajlı konuma getirdiği görülmektedir. Dolayısıyla bir olgunun geçmiş zaman dilimindeki değerlerinin analiz edilerek gelecekte alabileceği değerlerinin tahmin edilmesi önem kazanmıştır. Teknolojinin ve bilgisayar yazılımlarının geliştirilmesi ile bu alanda önemli düzeyde ilerlemeler kat edilmiş olup, yaşamın her alanında etkili tahmin yapmak ve rasyonel kararlar alabilmek adına geçerli ve güvenilir tahminleme yöntemleri geliştirilmiştir. Bu yöntemlerden olan zaman serileri analizi ve derin öğrenme teknikleri sıklıkla kullanılmaktadır. Zaman serisi analizinde ve derin öğrenme tekniklerinde tahmin yapılırken, tahmin yapılan olayın gelecekte alabileceği değerlerin, geçmiş zamanda aldığı değerlere benzer şekilde meydana geleceği öngörülmektedir. Dünya finans piyasasında döviz kurlarında, döviz bazlı finans işlemlerinde ve borsa endeksinde bu yöntemler kullanarak yatırım kararları verilmesi yatırımcılar ve araştırmacılar için önemli bir alternatif oluşturmuştur. Alanyazında yer alan çalışmalarda, veri setinin durağan olması ve mevsimsellik barındırması durumunda zaman serileri analizinde mevsimsel otoregresif bütünleşik hareketli ortalama (SARIMA) modelinin, derin öğrenme tekniklerinde uzun kısa süreli bellek (LSTM) modelinin iyi performans gösterdiği görülmektedir. Bu çalışmada kullanılan Amerikan Doları (USD)/ Türk Lirası (TL) döviz kuru gerçek verileri ve BIST100 borsa gerçek verileri ile elde edilen veri setlerinin de bu özelliklere (durağan olmama, mevsimsellik) sahip olmasından dolayı SARIMA ve LSTM modelleri kullanılarak zaman serisi analizi ve derin öğrenme gerçekleştirilmiştir. Veri analizi işlemleri Python programlama dilinden ve çeşitli Python kütüphanelerinden yararlanılarak yapılmıştır. Analiz kapsamında modellerin geleceğe yönelik tahminlemedeki doğruluğu araştırılmıştır ve bu doğrultuda modeller arasında karşılaştırma yapılmıştır. Anahtar Kelimeler: Zaman serileri analizi, derin öğrenme teknikleri, tahminleme, SARIMA modeli, LSTM modeli, döviz kuru, borsa
At the present time, it is seen that using time effectively and predicting the future brings individuals and institutions to a more advantageous position. Therefore, it has become important to analyze the values of a phenomenon in the past time period and to predict its future values. With the development of technology and computer software, significant progress has been made in this field, and valid and reliable estimation methods have been developed in order to make effective predictions and make rational decisions in all areas of life. Time series analysis and deep learning techniques, which are among these methods, are frequently used. While making predictions in time series analysis and deep learning techniques, it is predicted that the values that the predicted event may take in the future will occur like the values it took in the past. Making investment decisions using these methods in exchange rates, foreign exchange-based financial transactions and stock market index in the world financial market has created an important alternative for investors and researchers. In the studies in the literature, it is seen that the seasonal autoregressive integrated moving average (SARIMA) model in time series analysis and the long short-term memory (LSTM) model in deep learning techniques perform well in case the data set is stationary and has seasonality. Since the data sets obtained with the US Dollar (USD) / Turkish Lira (TL) exchange rate data and Bist100 stock market data used in this study have these characteristics (non-stationary, seasonality), time series analysis and deep learning were performed using SARIMA and LSTM models. Data analysis processes were carried out in Python programming language and using various Python libraries. Within the scope of the analysis, the accuracy of the models in predicting the future was investigated and a comparison was made between the models in this direction. Keywords: Time series analysis, deep learning techniques, forecasting, SARIMA model, LSTM model, exchange rate, stock market
At the present time, it is seen that using time effectively and predicting the future brings individuals and institutions to a more advantageous position. Therefore, it has become important to analyze the values of a phenomenon in the past time period and to predict its future values. With the development of technology and computer software, significant progress has been made in this field, and valid and reliable estimation methods have been developed in order to make effective predictions and make rational decisions in all areas of life. Time series analysis and deep learning techniques, which are among these methods, are frequently used. While making predictions in time series analysis and deep learning techniques, it is predicted that the values that the predicted event may take in the future will occur like the values it took in the past. Making investment decisions using these methods in exchange rates, foreign exchange-based financial transactions and stock market index in the world financial market has created an important alternative for investors and researchers. In the studies in the literature, it is seen that the seasonal autoregressive integrated moving average (SARIMA) model in time series analysis and the long short-term memory (LSTM) model in deep learning techniques perform well in case the data set is stationary and has seasonality. Since the data sets obtained with the US Dollar (USD) / Turkish Lira (TL) exchange rate data and Bist100 stock market data used in this study have these characteristics (non-stationary, seasonality), time series analysis and deep learning were performed using SARIMA and LSTM models. Data analysis processes were carried out in Python programming language and using various Python libraries. Within the scope of the analysis, the accuracy of the models in predicting the future was investigated and a comparison was made between the models in this direction. Keywords: Time series analysis, deep learning techniques, forecasting, SARIMA model, LSTM model, exchange rate, stock market
Description
Keywords
Bilgisayar Mühendisliği Bilimleri-Bilgisayar ve Kontrol, Bilim ve Teknoloji, Computer Engineering and Computer Science and Control, Science and Technology, Döviz kuru tahmini, Exchange rate forecasting