regressions with Newey–West standard errors for a list of countries

[Amir Moayeri]

I have a panel-data and I use ‘asreg’ or ‘rangestat’ to estimate a regression for each country separately. Instead of using the ordinary covariance method, I want to estimate OLS regressions with Newey–West standard errors robust to heteroskedasticity.
For a single time series, the command below is used to estimate it in Stata.
newey y x1 x2, lag(0)
How can I apply this approach in ‘asreg’ or ‘rangestat’? or do you know any other method to estimate this regression for a large number of countries separately?
Thanks.

[Alexander Milton]

Hi, im far from an expert but have you considered making a loop? Something in the lines of
Foreach country in local countrylist
reg...

Hope it helps!

[Nick Cox]

rangestat and asreg are both commands from SSC. I can speak confidently for rangestat and less confidently for asreg, which is by Attaullah Shah.

Newey-West standard errors are not supported directly by rangestat, which is not primarily a regression command but more a framework for window-related calculations that is also programmable. I don't doubt that you could write extra code in rangestat or its sibling rangerun to do what you want.

In contrast asreg is billed as a regression command, rather than a programmable framework, and it does offer more regression bells and whistles. A glance at the help identifies a newey() option, but read the documentation carefully to see if it is what you want.

[Nick Cox]

Note that while a loop may be a good idea, #2 conflates two quite different syntaxes that can't be mixed.

foreach country of local countrylist

is more likely to be what you want, although if country is a string variable that can be fiddly too.

That said, there can be simpler ways to do it. https://www.stata.com/support/faqs/d...-with-foreach/ is not a complete discussion, as (e.g.) statsby is often overlooked.

[Attaullah Shah]

Thank you Nick for tagging me in this post. Here is an example using asreg for cross-sectional regression with newey standard errors. Output from the official newey command is also posted for comparison. For a reasonably large dataset, forecch and statsby commands are slow enough to wear down one's patience. However, asreg is a breeze in usage and lightening in speed.

 h asreg

. webuse grunfeld, clear

.  bys company: asreg invest mvalue kstock,  se newey(1)

. keep if company == 1

. tsset year

.  newey invest mvalue kstock,  lag(1)

Regression with Newey–West standard errors      Number of obs     =         20
Maximum lag = 1                                 F(  2,        17) =      59.50
                                                Prob > F          =     0.0000

------------------------------------------------------------------------------
             |             Newey–West
      invest | Coefficient  std. err.      t    P>|t|     [95% conf. interval]
-------------+----------------------------------------------------------------
      mvalue |   .1192808   .0274023     4.35   0.000     .0614671    .1770946
      kstock |   .3714448   .0499277     7.44   0.000     .2661065    .4767831
       _cons |  -149.7824   108.6338    -1.38   0.186    -378.9797    79.41486
------------------------------------------------------------------------------

. list _*

     +----------------------------------------------------------------------------------------------------+
     | _Nobs         _R2      _adjR2   _b_mvalue   _b_kstock     _b_cons   _se_mv~e   _se_ks~k   _se_cons |
     |----------------------------------------------------------------------------------------------------|
  1. |    20   .92135403   .91210157   .11928082   .37144481   -149.7824   .0274023   .0499277   108.6338 |

[Joro Kolev]

On unrelated note, -newey- with 0 lags, which OP suggested he wants to do, is equivalent to robust. Here:

.  webuse idle2

. 
.         . tsset time

Time variable: time, 1 to 30
        Delta: 1 unit

. 
. 
. newey usr idle, lag(0)

Regression with Newey–West standard errors      Number of obs     =         30
Maximum lag = 0                                 F(  1,        28) =      10.43
                                                Prob > F          =     0.0032

------------------------------------------------------------------------------
             |             Newey–West
         usr | Coefficient  std. err.      t    P>|t|     [95% conf. interval]
-------------+----------------------------------------------------------------
        idle |  -.2281501   .0706372    -3.23   0.003    -.3728437   -.0834564
       _cons |   23.13483    6.43118     3.60   0.001     9.961153     36.3085
------------------------------------------------------------------------------

. 
. 
. regress usr idle, robust

Linear regression                               Number of obs     =         30
                                                F(1, 28)          =      10.43
                                                Prob > F          =     0.0032
                                                R-squared         =     0.5006
                                                Root MSE          =     2.7374

------------------------------------------------------------------------------
             |               Robust
         usr | Coefficient  std. err.      t    P>|t|     [95% conf. interval]
-------------+----------------------------------------------------------------
        idle |  -.2281501   .0706372    -3.23   0.003    -.3728437   -.0834564
       _cons |   23.13483    6.43118     3.60   0.001     9.961153     36.3085
------------------------------------------------------------------------------

[Attaullah Shah]

Joro Kolev
Interesting! So if a user wants to use robust standard errors, here is the asreg code.

webuse idle2
tsset time
newey usr idle, lag(0)

------------------------------------------------------------------------------
             |               Robust
         usr | Coefficient  std. err.      t    P>|t|     [95% conf. interval]
-------------+----------------------------------------------------------------
        idle |  -.2281501   .0706372    -3.23   0.003    -.3728437   -.0834564
       _cons |   23.13483    6.43118     3.60   0.001     9.961153     36.3085
------------------------------------------------------------------------------

. asreg usr idle, robust

. list _*

     +------------------------------------------------------------------------------+
     | _Nobs         _R2      _adjR2      _b_idle     _b_cons   _se_idle   _se_cons |
     |------------------------------------------------------------------------------|
  1. |    30   .50064502   .48281092   -.22815005   23.134828   .0706372    6.43118 |