Scribd
Upload
24 views

AMFE Module 4 - Box and Jenkins Methodology

Uploaded by

Aruneema
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PPTX, PDF, TXT or read online on Scribd
24 views

AMFE Module 4 - Box and Jenkins Methodology

Uploaded by

Aruneema
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PPTX, PDF, TXT or read online on Scribd
You are on page 1/ 9

Applied Macro and

Financial Econometrics

Box-Jenkins Method
Course Instructor:
Dr. Devasmita Jena
Box-Jenkins (BJ) Methodology
The estimation and forecasting of univariate TS models are carried out using B-J methodology
The methodology has three broad steps:
 Identification
o Determine if the TS is stationary and to make it stationary if it is non-stationary
o Once the time series is stationary, the order of AR and MA terms are identified
o And competing models are shortlisted
 Estimation
o Parameters of the identified models are estimated
 Diagnostic Checks
o Having chosen the ARIMA model and having estimated the model, justify whether the model fits the data reasonably
 Forecasting
o Forecast future values based on the estimated model
Identification
Raw plot: to identify presence of any TS process => prelim idea about stationarity
Unit root test to validate the stationarity
If TS is non-stationary, make it stationary by differencing
Check whether the transformed, differenced series is stationary
Keep differencing till you arrive at I(0) series
Caution: Avoid differencing more than once
 Loss of data points
 Getting back original series to forecast will entail issues!
 Try to transform the series, before even resorting to difference
ACF/PACF plot of transformed series
Process ACF PACF
White Noise No significant spikes No significant spikes
AR(p) Spikes damp out gradually Spikes cut off at the pth lag
MA(q) Spikes cut off at the qth lag Spikes damp out gradually
ARMA(p,q) Spikes damp out gradually Spikes damp out gradually
Identification
Identification of probable model: An example
 ACF Plot:
• Signinificant spikes at lags
1,5,8,11,15,20,31
 PACF plot:
• Significant Spikes at lags 1,5, 11, 12,20,31
 Competing Models for the data, assuming we
differenced the data only once: ARIMA(1,1,1),
ARIMA(1,1,5), ARIMA(1,1,8),ARIMA(1,1,11),
ARIMA(5,1,1), ARIMA(5,1,5), ARIMA(5,1,8),
ARIMA(5,1,11), ARIMA(11,1,1), ARIMA(11,1,
5), ARIMA(11,1,8), ARIMA(11,1,11)
 We have ignored lags higher than 11 due to
parsimony
 Which among the above could represent the
data closely?
• Use information criterion
Indentification: Information Criteria
Information Criterion tests: Probabilistic statistical measures to assess the model performance
and quantify the complexity of the model
 Based on its log-likelihood method of estimation
Advantage: do not require a hold-out test set
Limitation: do not take the uncertainty of the models into account and may end-up selecting
models that are too simple
IC:
• Akaike information criterion

• Schwartz Criterion/ Bayes information criterion


ln(T)
• Hannan-Quinn information criterion

Here, k is the number of regressors (=p+q+1); T is the number of observation and ,i.e, is the residual sum of
squares
Identification: Information Criteria
Model with smallest SC or AIC is chosen
• Reduces the RSS
Choice between the information criteria:
• SIC and HQIC are stricter in penalizing loss of degree of freedom than AIC
Example
Estimation
OLS methodology can not be used to estimate ARIMA models
Estimation of AR models
• OLS not applicable due to presence of autocorrelation
• By assuming normality of the error terms Maximum Likelihood method can be used for estimating the
parameters
• Under normality, OLS estimates of the coefficients coincide with ML estimates
• MLE are known to be asymptotically normal and asymptotically most efficient
Estimation of MA models
• Since most of the parameters (except intercept) are involved in error component which is not observable, OLS
technique can’t be applied directly
• Method of moments or MLE is used
• Since moments are non-linear functions of the unknown coefficients, closed form estimation is ruled out
• Iterative methods are used to estimate the parameters (softwares makes this easy and a matter of wink of eye!)
Estimation of ARMA models
• Yule walker method
• Method of moments
• MLE method
Diagnostic Checks
Comparison between in-sample based forecasts with out-sample observations
• Issue: loss of information, which will be severe if data points are insufficient
 Residual Analysis
• A simple plot of residuals against time will give an idea whether residuals are white noise
• ACF and PACF plots of residuals: No significant spikes

o H0: all ρ=0


• Ljung Box statistical test : to check all autocorrelations are zero

o H1: at least one ρ≠0


; T is the sample size and m is the lag length
• Root mean square error: is the standard deviation of the residuals (prediction errors)
o Model with lowest RMSE
o RMSE values between 0.2 and 0.5 shows that the model can relatively predict the data
accurately
o Mean absolute percentage error
Forecasting
Forecast future values based on the finalized models
Note: Prior to forecasting undo the “difference” transformation so that we get the level values
and not the changes (not needed in R!)
Interpret the forecast
• May be affected due to presence of volatility in the data
• Structural breaks
• Not enough data
• Etc.

You might also like