NEED HELP FOR ARIMA MODEL AND KALMAN FILTER

[tarox]

Dear all ,,, please need help

i have data from May 2007 to Jul 2014 and there are missing values on my data from SEP 2010- JUL 2012 i know i could use Arima for forecasting my steps were:

1- test for stationery for my data for available months from May 2007 - Aug 2010 ...> the result show stationary at second different

2- i used for series at level ( not on second different ) Arima auto selection and it show AR(2) MA(1) SAR(6) SMA(12)

3- then i estimated Arima model d(d(y)) c ar(2) ma(1) the result show no heteroscedasticity or auto-correlation

4- after that i did forecast the missing value from arima model from SEP 2010- JUL 2012 .

my question are : i heard that i should use kalman filter how could use it? and are my steps correct or not ?

please i need help this is very important for my thesis could any one please show the steps for my situation on eviews i am usong version 7

thank you

[harithstats]

hi tarox
i also use the same method for my thesis, i also stuck at to find the way how to apply the kalman filter on the ARIMA models. I got just a few ideas, first find the best model among ARIMA's. The apply to Kalman filter by using OBJECT --> NEW OBJECT --> SSPACE. Up to this i know about Kalman Filter.
Plz somebody out there, help us. I must submit my thesis this December, but i still dont have any idea how to do it

harith

[tarox]

hi harith ,,, me too i must submit my thesis this December i hope some one could help us

[harithstats]

so far, i just found a book named eviews5 user's guide. i think u can google it in the internet, u can download full version in pdf. im also still learning

[tarox]

thank you harith ,,i did the steps on eviews 5 user guide and this file has my result i still face problem,, mscif show the foretasted data by using arima model ,,and mscifff show the foretasted data by kalmen filter model ,,, please could any one tell me if my steps correct or if i misunderstood something

[harithstats]

good day mate,

in Box-Jenkins u hv specify the estimation part and evaluation part. Usually the estimation part is 75% of data and evaluation part is 15% of data.
Then within the 15% of data construct the error measures to find the smallest one. The smallest one, is the best . So, u hv to compare between the fixed(ARIMA) and the time-varying parameter(Kalman filter)

in mscif and msciff, there is different duration. so i wanna know which is the exactly your evaluation period.
thx, harith

[tarox]

thank you very much harith for helping me

[tarox]

dear all this is the answer for my situation but the problem i use eviews not R program could someone please help me

http://stats.stackexchange.com/question ... ssing-data :(