Arima Modeling

Financial time-series using ARIMA in R: Choosing the right model and number of samples. Nicolas Badaro American University of Beirut Electrical and Computer Engineering

1.

INTRODUCTION

Financial experts and speculators have always been aware of the power of information in achieving greater returns and beating the market. It should not be so surprising that the usage of Data Mining in financial application

Abstract- The aim of this paper is to approach the problem of stock price prediction using ARIMA (Auto-regressive integrated moving average) modeling. Using machine learning and computational science in the field of stock investing is not new: the literature is quite extensive, especially with the rise of quantitative trading strategies (called quants). We will try to demonstrate in this paper that the ARIMA model can achieve better results when combining the right (p,d,q) model and the right number of samples. We propose an algorithm to improve prediction rates. Keywords- ARIMA, Time-Series Analysis, Stock Price, Financial Modeling.

has been so widespread. While the two most popular approaches to stock investing are fundamental and technical analysis [1], this paper will be oriented towards the technical analysis side. Readers should note that their reproduced experience’s results should only be used as a support to investment decisions and not as an error-free prediction tool. 2. BACKGROUND

1.ARIMA modeling: In statistics, the most popular model that can be applied to time-series is called ARIMA, which stands for Auto Regressive Integrated Moving Average. It was developed in order to account for non-stationary time-series. The original model,which is linear, is called ARMA, and was designed to handle stationary time-series. By stationary, we mean that the values fluctuate around a single mean

and that their variance is in a stable interval [-V;+V]. The ARIMA(p,d,q) model is used to forecast future values in a given time-series given a big...