Newey-West standard errors for overlapping data
Moderators: EViews Gareth, EViews Moderator
-
noobie_simon
- Posts: 6
- Joined: Tue May 24, 2011 2:32 am
Newey-West standard errors for overlapping data
k
Last edited by noobie_simon on Wed Sep 28, 2011 2:33 pm, edited 1 time in total.
-
noobie_simon
- Posts: 6
- Joined: Tue May 24, 2011 2:32 am
Re: Newey-West standard errors for overlapping data
k
Last edited by noobie_simon on Wed Sep 28, 2011 2:34 pm, edited 1 time in total.
-
EViews Glenn
- EViews Developer
- Posts: 2682
- Joined: Wed Oct 15, 2008 9:17 am
Re: Newey-West standard errors for overlapping data
I'm not certain that I quite understand what you are asking, but let me take a crack...
It sounds as though you want the bandwidth for your kernel estimator to at least be as large as the lag on the regressor variable (which you are terming the horizon). I'm not certain why this is the goal, but in that case you are going to want to set a user-specified bandwidth. The automatic Newey-West lag-selection parameter option sets the number of lags used in the automatic bandwidth selection procedure, which isn't what you appear to want...
(Note that some kernels (for example, the Quadratic Spectral) have support over the entire range of lags so that even if you specify a bandwidth smaller than your horizon, the higher lags appear in the calculation (albeit with smaller weights).)
I should point out that even if you know your horizon, you are going to have to figure out the desired bandwidth corresponding with your horizon. Note that the bandwidth you want for a given horizon is going to depend on the kernel.
Note, for example, that if you specify a bandwidth of 7, then the weights in the Bartlett kernel associated with the first 7 lags are:
(1) 6/7 (2) 5/7 (3) 4/7 (4) 3/7 (5) 2/7 (6) 1/7 (7) 0/7
Thus, if you have a horizon of 7 and you want the first seven lags to have weights using a Bartlett kernel, you should set a user-specified bandwidth of 8.
In contrast, the user-specified bandwidth for a Truncated uniform differs. In this case, setting a user-specified bandwidth of 7 yields
(1) 1/7 (2) 1/7 (3) 1/7 (4) 1/7 (5) 1/7 (6) 1/7 (7) 1/7
So for a horizon of 7 you should set a user-specified bandwidth of 7.
It sounds as though you want the bandwidth for your kernel estimator to at least be as large as the lag on the regressor variable (which you are terming the horizon). I'm not certain why this is the goal, but in that case you are going to want to set a user-specified bandwidth. The automatic Newey-West lag-selection parameter option sets the number of lags used in the automatic bandwidth selection procedure, which isn't what you appear to want...
(Note that some kernels (for example, the Quadratic Spectral) have support over the entire range of lags so that even if you specify a bandwidth smaller than your horizon, the higher lags appear in the calculation (albeit with smaller weights).)
I should point out that even if you know your horizon, you are going to have to figure out the desired bandwidth corresponding with your horizon. Note that the bandwidth you want for a given horizon is going to depend on the kernel.
Note, for example, that if you specify a bandwidth of 7, then the weights in the Bartlett kernel associated with the first 7 lags are:
(1) 6/7 (2) 5/7 (3) 4/7 (4) 3/7 (5) 2/7 (6) 1/7 (7) 0/7
Thus, if you have a horizon of 7 and you want the first seven lags to have weights using a Bartlett kernel, you should set a user-specified bandwidth of 8.
In contrast, the user-specified bandwidth for a Truncated uniform differs. In this case, setting a user-specified bandwidth of 7 yields
(1) 1/7 (2) 1/7 (3) 1/7 (4) 1/7 (5) 1/7 (6) 1/7 (7) 1/7
So for a horizon of 7 you should set a user-specified bandwidth of 7.
Who is online
Users browsing this forum: No registered users and 2 guests