Graph using e(.) results.

[River Huang]

Dear All, I use (please ssc install) -xtqreg- command

webuse grunfeld, clear
xtqreg invest mvalue kstock, q(0.1(0.1)0.9)
*ereturn list
forvalues q = 1(1)9 { 
  matrix list e(b_`q')
  matrix list e(V_`q')
}

and obtain results as

. xtqreg invest mvalue kstock, q(0.1(0.1)0.9)



                              MM-QR regression results
Number of obs = 200
.1 Quantile regression
------------------------------------------------------------------------------
             |      Coef.   Std. Err.      z    P>|z|     [95% Conf. Interval]
-------------+----------------------------------------------------------------
      mvalue |    .084099    .029583     2.84   0.004     .0261174    .1420805
      kstock |   .2646024   .0552066     4.79   0.000     .1563994    .3728054
------------------------------------------------------------------------------

.2 Quantile regression
------------------------------------------------------------------------------
             |      Coef.   Std. Err.      z    P>|z|     [95% Conf. Interval]
-------------+----------------------------------------------------------------
      mvalue |   .0915338   .0237915     3.85   0.000     .0449033    .1381644
      kstock |   .2775904   .0444176     6.25   0.000     .1905334    .3646474
------------------------------------------------------------------------------

.3 Quantile regression
------------------------------------------------------------------------------
             |      Coef.   Std. Err.      z    P>|z|     [95% Conf. Interval]
-------------+----------------------------------------------------------------
      mvalue |   .0978504   .0224984     4.35   0.000     .0537543    .1419465
      kstock |   .2886248     .04201     6.87   0.000     .2062867    .3709629
------------------------------------------------------------------------------

.4 Quantile regression
------------------------------------------------------------------------------
             |      Coef.   Std. Err.      z    P>|z|     [95% Conf. Interval]
-------------+----------------------------------------------------------------
      mvalue |   .1031741   .0244971     4.21   0.000     .0551607    .1511875
      kstock |   .2979249   .0457377     6.51   0.000     .2082807    .3875691
------------------------------------------------------------------------------

.5 Quantile regression
------------------------------------------------------------------------------
             |      Coef.   Std. Err.      z    P>|z|     [95% Conf. Interval]
-------------+----------------------------------------------------------------
      mvalue |    .112539   .0329018     3.42   0.001     .0480526    .1770254
      kstock |   .3142844   .0614081     5.12   0.000     .1939268     .434642
------------------------------------------------------------------------------

.6 Quantile regression
------------------------------------------------------------------------------
             |      Coef.   Std. Err.      z    P>|z|     [95% Conf. Interval]
-------------+----------------------------------------------------------------
      mvalue |   .1164492   .0373915     3.11   0.002     .0431631    .1897353
      kstock |   .3211153   .0697806     4.60   0.000     .1843478    .4578827
------------------------------------------------------------------------------

.7 Quantile regression
------------------------------------------------------------------------------
             |      Coef.   Std. Err.      z    P>|z|     [95% Conf. Interval]
-------------+----------------------------------------------------------------
      mvalue |   .1249394   .0484885     2.58   0.010     .0299037    .2199751
      kstock |   .3359468   .0904895     3.71   0.000     .1585906    .5133029
------------------------------------------------------------------------------

.8 Quantile regression
------------------------------------------------------------------------------
             |      Coef.   Std. Err.      z    P>|z|     [95% Conf. Interval]
-------------+----------------------------------------------------------------
      mvalue |   .1335321   .0603947     2.21   0.027     .0151606    .2519036
      kstock |   .3509575   .1127063     3.11   0.002     .1300572    .5718578
------------------------------------------------------------------------------

.9 Quantile regression
------------------------------------------------------------------------------
             |      Coef.   Std. Err.      z    P>|z|     [95% Conf. Interval]
-------------+----------------------------------------------------------------
      mvalue |    .149272   .0832815     1.79   0.073    -.0139567    .3125008
      kstock |   .3784536   .1554208     2.44   0.015     .0738345    .6830727
------------------------------------------------------------------------------


. *ereturn list
. forvalues q = 1(1)9 { 
  2.   matrix list e(b_`q')
  3.   matrix list e(V_`q')
  4. }

e(b_1)[1,2]
       mvalue     kstock
y1  .08409896  .26460241

symmetric e(V_1)[2,2]
            mvalue      kstock
mvalue   .00087515
kstock  -6.865e-06   .00304777

e(b_2)[1,2]
       mvalue     kstock
y1  .09153382  .27759041

symmetric e(V_2)[2,2]
           mvalue     kstock
mvalue  .00056604
kstock  5.352e-06  .00197293

e(b_3)[1,2]
       mvalue     kstock
y1  .09785037  .28862482

symmetric e(V_3)[2,2]
           mvalue     kstock
mvalue  .00050618
kstock  .00001066  .00176484

e(b_4)[1,2]
       mvalue     kstock
y1  .10317409  .29792486

symmetric e(V_4)[2,2]
           mvalue     kstock
mvalue  .00060011
kstock  .00001112  .00209193

e(b_5)[1,2]
       mvalue     kstock
y1  .11253896  .31428441

symmetric e(V_5)[2,2]
            mvalue      kstock
mvalue   .00108253
kstock  -3.224e-06   .00377095

e(b_6)[1,2]
       mvalue     kstock
y1  .11644922  .32111525

symmetric e(V_6)[2,2]
            mvalue      kstock
mvalue   .00139813
kstock  -.00002525   .00486933

e(b_7)[1,2]
       mvalue     kstock
y1  .12493937  .33594676

symmetric e(V_7)[2,2]
            mvalue      kstock
mvalue   .00235113
kstock  -.00003722   .00818835

e(b_8)[1,2]
       mvalue     kstock
y1  .13353212  .35095748

symmetric e(V_8)[2,2]
            mvalue      kstock
mvalue   .00364752
kstock  -.00007772   .01270271

e(b_9)[1,2]
       mvalue     kstock
y1  .14927204  .37845363

symmetric e(V_9)[2,2]
            mvalue      kstock
mvalue   .00693581
kstock  -.00013575   .02415561

How can I use e(b) to draw a line across different quantile (0.1 to 0.9), along with 95% confidence intervals (approximately 2 standard errors) from e(V)?

[Andrew Musau]

This will get you starting by exporting the coefficients and variance estimates in the different matrices to a Stata data set. Your task is not difficult, just time-consuming, so your best bet is to invest the time yourself to complete it.

webuse grunfeld, clear
xtqreg invest mvalue kstock, q(0.1(0.1)0.9)
mat R= J(9, 4, .)
local i=1
forval j = 1/9{
mat b= e(b_`j')
mat V=e(V_`j')
matrix R[`i',1]= b[1,1]
matrix R[`i',2]= b[1,2]
matrix R[`i',3]= V[1,1]
matrix R[`i',4]= V[2,2]
local ++i
}
 
forvalues i=1/9{
local rnames  "`rnames' quantile`i'"
}
local cnames  "b_mvalue b_kstock V_mvalue V_kstock"
mat rownames R= `rnames'
mat colnames R= `cnames'
mat list R

svmat double R, names(col)

Matrix of coefficients and variances

. mat list R

R[9,4]
            b_mvalue   b_kstock   V_mvalue   V_kstock
quantile1  .08409896  .26460241  .00087515  .00304777
quantile2  .09153382  .27759041  .00056604  .00197293
quantile3  .09785037  .28862482  .00050618  .00176484
quantile4  .10317409  .29792486  .00060011  .00209193
quantile5  .11253896  .31428441  .00108253  .00377095
quantile6  .11644922  .32111525  .00139813  .00486933
quantile7  .12493937  .33594676  .00235113  .00818835
quantile8  .13353212  .35095748  .00364752  .01270271
quantile9  .14927204  .37845363  .00693581  .02415561

[Weiwen Ng]

My first instinct is to use -coefplot- (by Ben Jann, available on the SSC website), which is capable of plotting from a matrix. The complication here is that the matrix of results that Andrew compiled doesn't have the values of the confidence interval bounds in it. I know you can calculate those from the variance, but it does strike me that each time -xtqreg- runs, it outputs to r(table) - for those who don't know, r(table) is a list of regression results inlcuding betas, their standard errors, the Z score, p-value, and lower and upper 95% CI bounds. It essentially feeds what you see in the command window. However, when you run the -xtqreg- syntax that River typed, r(table) only has the 90th percentile results and none of the preceding ones. Here's one way I can think of to do things:

ssc install coefplot
webuse grunfeld, clear
tempname a b
forvalues quantile = 0.1(0.1)0.9 {
    xtqreg invest mvalue kstock, q(`quantile')
    matrix `a' = r(table)
    matrix `b' = `a'[1,1] \ `a'[5..6,1]
    matrix colnames `b' = "q`quantile'"
    matrix result1 = nullmat(result1), `b'
    }
coefplot matrix(result1[1]), ci((2 3)) vertical recast(connected)

The downsides are that this will require manual operation for each covariate you want to plot. Also, for some unknown reason, the names for the 80th and 90th percentiles are messed up (on my computer, they render as q.8999999999 and q.9999999999999, although you could rename the columns using Andrew's method). And finally, the plot currently lacks labels. Hopefully someone can improve this code!

For those new to matrix operations, here's an annotated version:

forvalues quantile = 0.1(0.1)0.9 {
    xtqreg invest mvalue kstock, q(`quantile')
    *Give the temporary matrix a the contents of r(table)
    matrix `a' = r(table)
    *Give the temporary matrix b the first row and first column of a, then column join (the \ operator) with the 5th and 6th row of a, first column
    matrix `b' = `a'[1,1] \ `a'[5..6,1]
    *Give temp matrix b the column name "q" plus the current value of quantile (which is referred to as a local macro)
    matrix colnames `b' = "q`quantile'"
    *If the matrix result1 doesn't exist, create it and assign the contents of temp matrix b; if it does exist, column join the contents of b
    matrix result1 = nullmat(result1), `b'
    }

[River Huang]

@Andrew Musau and @Weiwen Ng, Many thanks for your help suggestions.

[River Huang]

Dear Andrew, Suppose that I'd like to collection quantile coefficients, say, from 0.05, 0.10, 0.15, ..., 0.95, how can I adjust the code to this end? Thanks.

[Liu Qiang]

webuse grunfeld, clear
xtqreg invest mvalue kstock, q(0.1(0.1)0.9)
ereturn list
forvalues q = 1(1)9 {
  matrix list e(b_`q')
  matrix list e(V_`q')
}

webuse grunfeld, clear
xtqreg invest mvalue kstock, q(0.05(0.05)0.95)
forvalues q = 1(1)9 {
  local j=`q'*10
  matrix  B_`j'=e(b_`q')
  matrix  V_`j'=e(V_`q')
  mat list B_`j'
  matrix list V_`j'
}

forvalues q = 5(10)95 {
  matrix  B_`q'=e(b_`q')
  matrix  V_`q'=e(V_`q')
  mat list B_`q'
  matrix list V_`q'
}

mat R= J(19, 4, .)
local i=1
forval j = 5(5)95 {
mat b=B_`j'
mat V=V_`j'
matrix R[`i',1]= b[1,1]
matrix R[`i',2]= b[1,2]
matrix R[`i',3]= V[1,1]
matrix R[`i',4]= V[2,2]
local ++i
}
 
forvalues i=1/19{
local rnames  "`rnames' quantile`i'"
}
local cnames  "b_mvalue b_kstock V_mvalue V_kstock"
mat rownames R= `rnames'
mat colnames R= `cnames'
mat list R

svmat double R, names(col)

I tried to code if you change the quantile range to 0.05(0.05)0.95. It is quite strange that the macro matrix should report quantile10 as quantile1 which brings difficulty in looping. Hope this can help you dear Huang.

[River Huang]

Dear Liu, Thank you for this helpful suggestion.