Algoritmik İşlemler İçin Derin Öğrenme Tabanlı Regresyon Yaklaşımı: BİST30 Örneği

Abstract

Today, one of the common uses of artificial intelligence is financial markets. In these markets, which are known as stock market, making price predictions for the future using machine learning and deep learning, making the rise and fall forecasts of indices, sectors and stocks are the main approaches used in this field. In the near future in the financial markets, artificial intelligence based software robots are expected to operate instead of people. For this purpose, learning models are developed by using trend and stock price movements. Validation studies such as accuracy, error value and portfolio simulation are performed to demonstrate the performance of the developed models. In this study, a regression model using deep learning was developed to make adaptive buy-sell operations on the time series consisting of closing prices using data from Borsa İstanbul (BIST). The 2006-2015 range of the BIST30 index was used for training, the 2015-2018 range was used for testing, and the model portfolio value gained 39% on the test data for 694 trading days and the trend direction was estimated with 82% accuracy.

Deep Transformer-Based Asset Price and Direction Prediction

“…We transform our time-series datasets into an image by using 65 different technical indicators, each with various parameter combinations over a defined period. Some common examples of these technical indicators are MACD [18], RSI [19], Bollinger Bands [20], Fibonacci Retractment, etc. Once turned into a matrix, the resulting 2D representation organizes rows to cluster indicators with similar patterns, ensuring smooth transitions along the y-axis, and capturing consecutive patterns for deep learning techniques in asset price and direction predictions.…”