Prediction of Stock Prices Using Statistical and Machine Learning Models: A Comparative Analysis

Abstract

Abstract With the advent of machine learning, numerous approaches have been proposed to forecast stock prices. Various models have been developed to date such as Recurrent Neural Networks, Long Short-Term Memory, Convolutional Neural Network sliding window, etc., but were not accurate enough. Here, the aim is to predict the price of a stock and compare the results obtained using three major algorithms namely Kalman filters, XGBoost and ARIMA. Kalman filters are recursive and use a feedback mechanism to perform error correction. This correction makes them best suited for making accurate predictions as they can factor in the market volatility, whereas XGBoost is a promising technique for datasets that are nonlinear and can gather knowledge by detecting patterns and relationships in the data. XGBoost is also capable of capturing the time dependency of features efficiently. ARIMA refers to an Auto Regressive Integrated Moving Average model that has become very popular in recent times. It is mostly used on time series data and works by eliminating its stationarity. Finally, a hybrid model combining Kalman filters and XGBoostis discussed and a comparison of the results of each of the four models, are made to provide a better clarity for making investments by forecasting the price of a stock.