International Review of Financial Analysis

STATISTICS 1
A Brief Survey on Time Series Data Prediction Using ARIMA Models
Student’s name
Tutors name
Course
Institutional affiliation
City and state
Date

STATISTICS 2
Abstract
The survey explores the effect of ‘Auto-Regressive Integrated Moving
Averages’ (ARIMA), models in prediction of data based on time. ARIMA model
involves modelling of time series data in order to prediction. ARIMA models involves
many tests such as stationarity, normality and linearity. For the time series modelled
data, it is essential for the data to be stationary in order to provide reliable and
consistent results. The survey evaluated various other machine learning methods in
prediction. These techniques include; LSTM and PROPHET. The survey looked at
their differences and the behavior towards time series data. It was observed that
ARIMA is mostly used in prediction especially in business intelligence since it
enhances estimation of the historical, current and future forecasts.

STATISTICS 3
ARIMA models in prediction of time series data
1. Introduction
In the past years, time series data resolved by statisticians minus considering the
effects on their analysis that may be brought about by ‘non-stationeries’. Until
George. P and Jenkins. G came up with a monograph known as “Time series Analysis
forecast and control”. This tool helped in predicting so that the non-stationary data
could be transformed into stationary data. This would be done by ‘differencing’ series.
Time series data plays an essential role in future value forecast. Basing on past results,
time series is applied in forecasting changes in economics, capacity planning, weather
among others (Alsharif et al, 2019). However, features time-series data need specific
statistical methods to be applied. There are concepts that time series data uses and
these are autocorrelation, seasonality and stationary. In this paper, we are to discuss
the prominent time series method in forecasting and this method is known as ARIMA.
The ARIMA component is applied to transform time series data so as to ease the and
forecast future values in time series cheaply. ARIMA models give the best approach to
forecasting time series (Alsharif et al, 2019). The commonly used approaches in
forecasting time series data are ARIMA and exponential smoothing. These two
methods work hand in hand to solve problems. As exponential smoothing model relies
on describing data trends and seasonality, ARIMA models do describe the
autocorrelations within data. ARIMA is described as a group of models that
describes a specific time series relying on the previous values that is to say, its ‘own-
errors that are lagged on prediction errors. In this case, the ARIMA model is applied

STATISTICS 4
in prediction of values in future. The series that are not seasonal denies originality
and not a probability as the ‘white noise’ is transformed using ARIMA models. There
are three models that make an up model of ARIMA and these are presented by p, d
and q as discussed below;
P means the order of the ‘AR’ term
d means the number of ‘differencing’ needed to transform the time series data from
being non stationary to being stationary
q represents the order of the ‘MA’ term
If ‘seasonal pattern’ are observed in a time series, then there is a requirement of
summing up all seasonal terms so that the series become ‘Seasonal ARIMA ‘known as
“SARIMA”. For one to start forming an ARIMA model, the time series is made
stationary. The reason behind this is that the concept ‘Auto Regressive’ in ARIMA
shows it is a ‘linear regression model’ which applies its “lags” as forecasts. It is well
known that linear regression models perform well only if predictors are not correlated
and work separately. A time series is made stationary through finding the difference
between the past values and the current values (past values minus current values). In
cases when series are ‘large’, differencing can be made several times. Therefore, d
shows the highest times for differencing required to turn the series stationary.
However, in cases where time series are stationary, the d is zero (d = 0).
P represents the ‘Auto Regressive’ (AR) order of the term. This is the number of
“lags” of Y that are applied as predictors. On the other hand, q means ‘moving
Average’ order of the term. It is the “lagged” predicated errors that are meant to be