Comparative study for obtaining the best ordered of the mathematical forecast's models in time series analysis with application

Abstract EN

The research aimed to forecasting based on fitting models to past observations in a given time series.

The model that should choose depends on how would the forecast used, the degree of accuracy which had been required from the forecast, the amount of time and capital available to the user, the amount and type of data available to be meaningful application, and how far ahead that must be forecast.

Several forecasting models suggested, such as (Random Walk, Linear trend, Quadratic trend, Exponential trend, S – Curve, Moving Average, Brown’s Linear Exponential Smoothing, Holt’s Linear Exponential Smoothing, ARMA (1,0), ARMA(0,1) and ARMA (1,1).

Application section deals with searching for an optimal of mathematical model that ought to be used for prediction or forecasting the quantities and account of sells the Iraqi's cigarettes depends along the period in the past observations (2000 – 2009) yrs.

Two criteria had been applied, the first statistic measured the magnitude of the errors (MSE) and the second statistic measure bias (ME).

The two series (Quantity & Account) according to the magnitude of the residuals through applying (MSE) indicator showed that the best mathematical models were with Quadratic Trend and Linear trend respectively.

The two series (Quantity & Account) according to the bias through applying (ME) indicator showed that the best mathematical model was with S-Curve.