Arima Word
Arima Word
Arima Word
Stationarity
Identification
Stationarity check
Check the stationarity by graph, correlogram and formal tests like ADF, PP, KPSS
If stationary: ARMA Model(p,q)
If non-stationary: ARIMA Model(p,d,q)
Identification step
Check the correlogram to determine p and q
If it is non stationary:
Appropriate transformation (including difference, variance stationarity, logarithm, square root) should
be used to be converted to a stationary sequence. The number of differences is the order of single
integer I
ARIMA model selection
Autocorrelation order p and the moving average order q of the ARIMA model can be estimated by
ACF and PACF plots respectively
Estimation
Estimation of all possible candidate models
Objective : find a stationary and parsimonious model that fits the data well
Model selection criteria:
Significance of the ARMA components
Compare akalike, Schwartz and Hannan, Ouinn. Smaller value of these parameter is better
Diagnostic and Forecasting
Model can be optimized and diagnosed by performing white noise test on residual. If residual is not a
white noise then, re-select the model. If residual is white noise create multiple models and choose
optimal model from all candidate model of the test.
Forecasting
In equation menu of Eviews software, select the forecast menu where static or dynamic can be
selected. Then select the sequence of forecasting value, and click OK.
Example
Data description:
The number of cars in Delhi from 1952 to 2015 is listed in Table
It can be seen that ADF= 0.098212 is greater than the critical value of the significance level of 0.01,
0.05 and 0.1, that is to say, the original GDP sequence is non-stationary.
Fig. Augmented Dickey-Fuller unit root test on cars at the first order difference(d=1)
It can be seen that ADF=-3.603986 is less than the three critical values of the test level. That is to say,
the cars sequence after the logarithmic change and the first-order difference is a stationary series, and
the significance test of the stationarity is passed. It can be seen that the original sequence is a first-
order single-order sequence, that is, d=1.
Model identification:
With the EViews software, the autocorrelation and partial autocorrelation function graphs of cars
series are plotted in Figure.
Fig. Autocorrelation and partial autocorrelation function graphs of the cars data
The temporary models obtained are ARIMA (1,1,1) , ARIMA (1,1,2) , ARIMA (2,1,2) , ARIMA
(2,1,3), ARIMA (1,1,3) and ARIMA (2,1,1) .
Estimation:
The temporary models are tested for its parameters to get the best model. We have taken 2 models
ARIMA(1,1,1) and ARIMA(2,1,1) for comparison to demonstrate.
We have compared the models based on the significance of the ARMA components and Akalike,
Schwartz and Hannan, Ouinn. Smaller value of these parameter is better.
Estimation results of the ARIMA(1,1,1) model
Based on Table 1 and Table 2, it can be seen that ARIMA (1,1,1) is the best model as its variables C,
AR, MA are around 0.05 and the values of Akalike, Schwartz and Hannan-Ouinn are smaller
compared to the other models.
Diagnostic and Forecasting:
Fig. Autocorrelation and partial autocorrelation function graphs of the residual series.
A white noise test is performed on the residual after fitting the ARIMA (1, 1, 1) model. The
autocorrelation and partial autocorrelation function graphs of the residual series are shown in Figure.
It can be seen that the residual is a white noise, indicating that the model is valid.
The final step is forecasting using a model that has been tested before. Forecasting used for the
number of cars is the ARIMA model (1,1,1) and the results are as follows:
Fig. Forecasted numbers of cars from 2020 to 2030
Similarly, we have followed the same procedure for other types of vehicles as well. The results are
shown below.
2w
Fig. Forecasted numbers of 2W from 2020 to 2030
Summary
ARIMA model forecast is comparatively superior time series prediction method. It can realistically
describe the dynamic alternate rules. It may be used to carry out statistical evaluation and forecast for
time series beneath neath sure conditions. Specially, the model is appropriate for short-time period
predictions. Large deviations arise while the forecasting time scale is long. Based on EViews
software, this work offers time series modelling and forecasting with the ARIMA model. It needs to
be referred to that as for a particular time series this is problem to many factors, model predictions
that depend totally on modern-day values and historical information once in a while have a sure
degree of deviation from the actual situation.
Results
The best ARIMA model for estimating the number of cars is ARIMA (1,1,1).
The best ARIMA model for estimating the number of 2W is ARIMA (1,1,0).
The best ARIMA model for estimating the number of buses is ARIMA (3,0,0).
Forecasting results on the number of cars and 2W shows that the number of the vehicles is
increasing from year to year.