Ask for help to do estimation
Posted: Mon Jan 17, 2011 11:02 am
Dear all,
I have two easy questions for estimation in Eviews 6. I think these problems may be raised by others before but I cannot locate them in the forum. So, if you have any clues or explanations, please help me out.
I tried to estimate an univariate time series (the data is enclosed). The data shows seasonality and structure change. So, I estimated it by seasonal differencing the data. The modified data was tested to be stationary under ADF test, Phillip-Perron, and ERS. Then, I conducted the Chow's breakpoint test multiple times to locate the (best) breakpoint which shows the largest F-statistics, LR ratio, and Wald statistics. So, I coded my model as d(y, 0, 12) c y(-1) y(-2) y(-3) ma(1) ma(2) ma(3) SMA(12). The programs showed insignificant coefficient for the autoregressive terms. However, if I coded my model as d(y, 0, 12) c ar(1) ar(2) ar(3) ma(1) ma(2) ma(3) sma(12), the coefficients for all autoregressive terms are significant and the r-squared value increases a lot.
My first question is what's the reason that the coefficients for my regressor and the R-squared are different when I code AR(1) vs y(-1)? When I read the Eviews guide, it says the terms are inter-changeable. I know that if I modeled a pure AR(1) series, the terms are inter-changeable and the results are the same. However, in my case, it seems the terms are not interchangeable. Does that mean one way is better than the other?
My second question is about involving the structure change in the model. There are two structure changes for the data. I tried to use dummy variable to specified the model. I know how to do it if there is only one structure change. But how about two structure changes?
Thanks for the help.
I have two easy questions for estimation in Eviews 6. I think these problems may be raised by others before but I cannot locate them in the forum. So, if you have any clues or explanations, please help me out.
I tried to estimate an univariate time series (the data is enclosed). The data shows seasonality and structure change. So, I estimated it by seasonal differencing the data. The modified data was tested to be stationary under ADF test, Phillip-Perron, and ERS. Then, I conducted the Chow's breakpoint test multiple times to locate the (best) breakpoint which shows the largest F-statistics, LR ratio, and Wald statistics. So, I coded my model as d(y, 0, 12) c y(-1) y(-2) y(-3) ma(1) ma(2) ma(3) SMA(12). The programs showed insignificant coefficient for the autoregressive terms. However, if I coded my model as d(y, 0, 12) c ar(1) ar(2) ar(3) ma(1) ma(2) ma(3) sma(12), the coefficients for all autoregressive terms are significant and the r-squared value increases a lot.
My first question is what's the reason that the coefficients for my regressor and the R-squared are different when I code AR(1) vs y(-1)? When I read the Eviews guide, it says the terms are inter-changeable. I know that if I modeled a pure AR(1) series, the terms are inter-changeable and the results are the same. However, in my case, it seems the terms are not interchangeable. Does that mean one way is better than the other?
My second question is about involving the structure change in the model. There are two structure changes for the data. I tried to use dummy variable to specified the model. I know how to do it if there is only one structure change. But how about two structure changes?
Thanks for the help.