In: Economics
Following diagnostic tools are used to determine if ARIMA model is appropriate:
- Correlogram. If the correlogram is not flat, then it means that some information is still needs to be captured and the model needs to re-estimated.
- Checking significance of the lag terms. Model should have least number of parameters. A Parsimonious Model is always preferred
- Conduct Ljung-Box Test for squared residuals. This is also known as Autocorrelation Test
- Checking Adjusted Coefficient of Determination and the variance to see the volatility in the underlying data used. This is used to drop irrelevant variables from the model. HIgher number of variables consume more degrees of freedom.
- Check penalty factors such as AIC (Akaike Information Criterion) and SBIC (Schwartz Bayes Information Criterion). If they are diagnosed as high, then re-estimate the model. Model with least AIC or SBIC is preferred