Hello!
I would like to use ARFIMA for model estimation, as the autocorrelation function is decaying slowly . I am using Eviews, although i read the Eviews explanation of ARFIMA models. But still i cannot understand it well.
In Eviews you just have to add the d keyword to the specification to indicate that you would like to estimate and use a fractional difference parameter.
Which i already did , please find attached the estimation outputs.
How to interpret the d parameter and what does it mean?
Thank you :)
Hi Niematallah,
the value of d is at the limit of validity, as -0.5 < d < 0.5 (lucky you...:) ). It means there is a strong long negative dependence in your data.
There are two interesting feature in result.
1. The error of the coefficients are very small, which may imply that you are using a too conservative model. Try adding coefficients to it. This my explain perhaps also the small value for R2.
2. The huge difference between Log.Likelihood and AIC seams odd.
Thank you :)
based on the value of d it can be decided whether ARFIMA model is suitable or no?
Hi Niematallah,
the value for d has nothing to do in deciding if a model fits the data or not. The only important thing for the coefficients is to be consistent with the theory (|d|>=0.5, no unite root).
Looking only at the coefficient one can say that the model is consistent. How good it describes the data you used, this can't be answered. That's way, you should also look at the residuals and search if there is still information present (you can use ACF again, for example). If yes, it would enable you to improve your model, and thus describe the data more adequate.
what are the implication of using ARIMA models when ARFIMA models should be used
Kelechi Aruah, Niematallah Elamin We use ARFIMA in modeling time series with long memory. When the integration parameter d in an ARIMA process is fractional and greater than zero.
You can check if the process exhibits long memory using different methods such as: rescaled range analysis (R/S), modified rescaled range analysis (MRS), or De-trended fluctuation analysis (DFA)
These papers could be helpful for you
Article An Introduction to Long Memory Time Series Models and Fracti...
http://www.sapub.org/global/showpaperpdf.aspx?doi=10.5923/j.m2economics.20160401.03
when I'm using similar equation, syntax error is coming for "d". Can someone please help me?
- Badges
- Science topic
- Similar topics
- Statistics
- Time Series