Finansal yatırım piyasalarında fiyat tahminleme: Kripto para piyasasında yapay sinir ağları uygulaması
Price prediction in financial investment markets: Application of artificial neural networks in cryptocurrency market
- Tez No: 915333
- Danışmanlar: PROF. DR. TAYFUN AKGÜL, PROF. DR. AYBEN KOY
- Tez Türü: Yüksek Lisans
- Konular: Elektrik ve Elektronik Mühendisliği, Maliye, Electrical and Electronics Engineering, Finance
- Anahtar Kelimeler: Belirtilmemiş.
- Yıl: 2024
- Dil: Türkçe
- Üniversite: İstanbul Teknik Üniversitesi
- Enstitü: Lisansüstü Eğitim Enstitüsü
- Ana Bilim Dalı: Elektronik ve Haberleşme Mühendisliği Ana Bilim Dalı
- Bilim Dalı: Telekomünikasyon Mühendisliği Bilim Dalı
- Sayfa Sayısı: 145
Özet
Günümüzde bireylerin daha yüksek kazanç ve refah amacıyla finansal yatırım piyasalarına yöneldiği bilinmektedir. Bu piyasalar, sermaye sahibi kişi ve kuruluşların çeşitli ürünleri alıp sattığı yerler olarak tanımlanmaktadır. Teknolojinin gelişmesiyle birlikte, yatırımcılar için bilgiye erişim kolaylaşmış ve piyasalara olan ilgi artmıştır. Ancak yoğun veri akışı, özellikle yeni yatırımcılar için kafa karıştırıcı olabilmektedir. Bu karışıklığın önüne geçebilmek için yatırım karar sürecinin iyi anlaşılması gerekmektedir. Yatırım karar süreci, sermaye temini, veri toplama, değerlendirme ve karar uygulama aşamalarından oluşmaktadır. Bu doğrultuda, yatırımcıların doğru karar verebilmeleri için temel ve teknik analiz yöntemlerini kullanmaları önem taşımaktadır. Temel analiz, bir varlığın içsel değerini belirlemeye çalışırken, teknik analiz geçmiş fiyat hareketlerini inceleyerek birtakım teknik indikatörler yardımıyla geleceği tahmin etmek amacıyla kullanılmaktadır. Son yıllarda, bu analiz yöntemlerinin yetersiz kaldığı anlaşıldıkça yapay zekâ ve özellikle yapay sinir ağları piyasa analizlerinde ve kapanış fiyatlarının tahminlenmesinde kullanılmaya başlanmıştır. İnsan beyninin çalışma biçiminden esinlenerek oluşturulmuş, karmaşık ilişkileri tespit etme yeteneğine sahip olan bu teknolojiler, büyük veri setlerini işleyerek daha tutarlı ve gerçeğe yakın tahminler yapma imkânı sunmaktadır. Bu çalışmada da finansal yatırım araçlarının yapay sinir ağları ile tahminlenmesi amaçlanmış, yatırım aracı olarak ise son dönemde popülerliği daha da artan kripto para piyasaları tercih edilmiştir. Çalışmada kripto para piyasalarında işlem hacmi en yüksek olan para birimleri BitCoin, Ethereum ve LiteCoin kullanılmış, çalışmanın veri seti ise her bir kripto para birimi için 15'er, 30'ar ve 60'ar dakikalık periyotlardaki kapanış fiyatları ile SMA, EMA, RSI, MACD ve BBand indikatör değerleri kullanılarak hazırlanmıştır. Kurulan ağ modeli eğitilip test edildikten sonra, 01 Ocak – 31 Aralık 2022 yılı kapanış verileri girdi olarak modele eklenmiş ve modelin 2023 kapanış verilerini tahminlemesi beklenmiştir. Yapılan tahminleme sonucunda her bir veri setinin hata farkı ortalama yüzdesi (MAPE) değeri hesaplanmıştır. BTC15 için 0.0088, BTC30 için 0.0021, BTC60 için 0.0375, ETH15 için 0.000030205, ETH30 için 0.00022303, ETH60 için 0.00020702, LTC15 için 0.00070347, LTC30 için 0.0053 ve LTC60 için 0.00063393 şeklinde bulunan MAPE değerleri modelin ne kadar tutarlı tahminlemeler yaptığını ortaya koymuştur. Ayrıca tahminlenen ve gerçekleşen verilerin yer aldığı grafik de neredeyse birbiriyle tamamen örtüşmüştür. Bu sonuçlar yapay sinir ağı modellerinin kripto para piyasalarında tahminleme için kullanılabileceğini kanıtlamaktadır. Çalışmanın ayrıca; daha ileriki safhada, bu modellerin yardımıyla düzenli gelen veriyi işledikçe nihai kullanıcıya al-sat sinyali gönderebilecek bir robot uygulamanın geliştirilmesi için de bir kılavuz niteliği taşıyacağı da düşünülmektedir.
Özet (Çeviri)
It is well known that individuals are turning to financial investment markets with the aim of achieving higher earnings and prosperity. These markets are defined as places where individuals and organizations buy and sell various products by using their capital. With the advancement of technology, access to information has become easier for investors and lead them to increase their interest to the financial markets. However, the intense data-flow can be confusing, especially for new investors. To overcome this confusion, it is essential to have a good understanding of the investment decision making process. This process consists of stages such as capital procurement, data collection, evaluation, and decision implementation. In this context, it is important for investors to use fundamental and technical analysis methods to make informed decisions. Fundamental analysis aims to determine the intrinsic value of an asset, while technical analysis is used to predict the future by examining past price movements with the help of various technical indicators. Due to the complexity and dynamic nature of investment markets, influenced by macro-environmental factors, the belief that fundamental and technical analysis methods have become inadequate in evaluating the dynamic structure of these markets has been increasingly prevalent over time. As a result, time series analysis techniques and multiple regression models have also been developed alongside these two analyses. However, it has been observed that these methods, particularly linear models, are more successful in making predictions but fall short in capturing sudden changes in investment markets that do not exhibit linear characteristics due to the constantly changing economic environment. The interaction of factors such as political events, investor expectations, and economic conditions complicates forecasting in investment markets. To overcome this difficulty, it is emphasized that computational artificial intelligence methods, which incorporate regression techniques using technical indicators and candlestick charts, should be employed to analyze this variability and non-linear structure, enabling stronger predictions. In recent years, to enhance the effectiveness of these analysis methods, artificial intelligence, particularly artificial neural networks, has started to be used. These technologies offer the potential for more accurate predictions by rapidly processing large data sets and have become critical tools for investors in their decision-making processes, allowing them to monitor market movements and take strategic actions. Neural networks are modeled based on the parallel processing methods of the human brain. The biological brain consists of billions of interconnected processing elements known as neurons, which transmit information and strengthen as the brain learns. In this context, neural networks utilize interconnected processing elements that enable them to learn from errors, learn from examples, recognize patterns in noisy data, and work with missing information. Artificial neural networks aim to overcome the xix limitations of traditional computers by evaluating the processing capabilities of the human brain. An Artificial Neural Network (ANN), in its simplest form, is defined as a model composed of several highly interconnected computational units called neurons or nodes. Each neuron performs a simple operation on an input to generate an output that will be transmitted to the next neuron. This parallel processing provides significant advantages in data analysis. Artificial neural networks are widely used in various branches of engineering and science and are considered a useful tool in econometric analyses due to their ability to approximate complex and non-linear equations. They have emerged as a powerful tool and statistical modeling technique in modern quantitative finance, serving as an attractive alternative for both researchers and practitioners. They can detect underlying functional relationships within a data set and perform tasks such as pattern recognition, classification, evaluation, modeling, prediction, and control. Additionally, with artificial neural networks, investors have the opportunity to develop customized models that align with their strategies and objectives, making it possible to create more personalized and effective investment strategies. Since artificial neural networks operate independently of human psychology, they help minimize emotional decision making processes, thus facilitating more objective and rational investment decisions. They are also effectively used in portfolio optimization and risk management, enabling investors to create more balanced and profitable portfolios by analyzing the risk and return profiles of different assets. Despite the advantages mentioned above, there are various issues that may arise in the applications of artificial neural networks. Foremost among these is data quality; incomplete, erroneous, or misleading data can negatively impact the model's performance. Investment analysis is typically conducted using time series data, and the nature of time series data can sometimes complicate the accurate application of artificial neural network models. Similarly, sudden fluctuations in markets and unexpected events such as economic crises or political instability can invalidate the predictions made by neural networks. These challenges indicate that careful planning and implementation are required for artificial neural networks to be effectively used in investment analysis. To overcome the challenges faced in investment analysis with artificial neural networks, methods such as improving data quality, optimizing the model, managing computational power, using time series methods, accounting for market anomalies, and enhancing the model's transparency are employed. These methods are effective in overcoming the challenges of using artificial neural networks in investment analysis, ensuring flexibility and adaptation during applications, thereby improving the quality of analysis results significantly. This study aims to predict financial investment instruments using artificial neural networks, with cryptocurrency markets, which have gained increasing popularity recently, being chosen as the investment instrument. The cryptocurrency market is defined as a technology developed around the 2010s that creates virtual trading among investors. Cryptocurrencies are considered digital assets developed using an end-to end encryption method called cryptography, which enhances transaction security and ensures the circulation of excess money within the system during exchange processes. Although they have gained popularity today, with nearly a thousand different cryptocurrencies available, they are not backed by central banks, which means they lack any legal framework or government guarantee. Therefore, they are considered xx difficult to analyze and track financially. Due to these characteristics, this study focuses on predicting cryptocurrencies, specifically examining the cryptocurrencies with the highest trading volumes in the market: Bitcoin, Ethereum, and Litecoin. The study utilized the Bitcoin, Ethereum, and Litecoin cryptocurrencies which have the highest trading volumes in the cyrptocurrency market. The dataset was prepared using the closing prices for each cryptocurrency at 15, 30, and 60-minute intervals, along with the SMA, EMA, RSI, MACD, and BBand indicator values. After training and testing the established network model, the closing data for the year 2022 (from January 1 to December 31) was added as input to the model, which was expected to predict the closing data for 2023. As a result of the predictions made, the mean absolute percentage error (MAPE) values for each dataset were calculated: 0.0088 for BTC15, 0.0021 for BTC30, 0.0375 for BTC60, 0.000030205 for ETH15, 0.00022303 for ETH30, 0.00020702 for ETH60, 0.00070347 for LTC15, 0.0053 for LTC30, and 0.00063393 for LTC60. These MAPE values illustrate how consistent the model's predictions are. Moreover, the graph displaying the predicted and actual data nearly completely overlaps. These results demonstrate that artificial neural network models can be utilized for predictions in cryptocurrency markets. Additionally, it is believed that this study will serve as a guide for the development of a robot application capable of sending buy-sell signals to end users as it processes incoming data with the help of these models in a future works.
Benzer Tezler
- Makine öğrenmesi yöntemleri ile bıtcoın trend dönüşlerinin tahmin edilmesi
Predicting bitcoin trends reversals with machine learning methods
SERGÜL ÜRGENÇ
Yüksek Lisans
Türkçe
2023
İstatistikMimar Sinan Güzel Sanatlar Üniversitesiİstatistik Ana Bilim Dalı
PROF. DR. BARIŞ AŞIKGİL
- Kripto para fiyatlarının arıma ve yapay sinir ağı modelleri ile tahmini
Prediction of cryptocurrency prices with arima and artificial neural network models
EMRE YILMAZ
Yüksek Lisans
Türkçe
2024
İstatistikKırıkkale Üniversitesiİstatistik Ana Bilim Dalı
DR. ÖĞR. ÜYESİ SERAP YÖRÜBULUT
- Derin öğrenme ile ethereum fiyat tahmini
Ethereum price prediction with deep learning
MUSTAFA YALÇIN
Yüksek Lisans
Türkçe
2024
Elektrik ve Elektronik MühendisliğiBatman ÜniversitesiElektrik-Elektronik Mühendisliği Ana Bilim Dalı
PROF. DR. ÖMER FARUK ERTUĞRUL
- Genelleştırilmiş toplamsal modeller ile Bitcoin için yön analizi
Directional analysis of Bitcoin using generalized additive models
İLAYDA ARIKAN
Yüksek Lisans
Türkçe
2024
EkonomiHacettepe Üniversitesiİstatistik Ana Bilim Dalı
PROF. DR. SERPİL AKTAŞ ALTUNAY
- Kripto paralarda seçilmiş sistematik yatırım stratejilerinin karşılaştırmalı analizi
A Comparative Analysis of Selected Systematic Investment Strategies in Cryptocurrency Markets
EGEMEN KAHRAMAN