How to save the "drift" parameter of a Dickey-Fuller unit-root test (dfuller)?

[Rene Macon]

I would like to save the "drift" parameter of a Dickey-Fuller unit-root test, in order to export it with the help of "putexcel". With "drift parameter" I mean the regression constant of the estimated model.

As far as I see, it is not available under the "stored results" of "dfuller". It can however be displayed using the option "regress" in "dfuller". It is then displayed in a table and has the name "_cons". However "_cons" cannot be saved or exported.

Is there any possibility to save the "drift" parameter or would I have to manipulate the source code of "dfuller"?

Thanks for help!

[William Lisowski]

I think this can start you in a useful direction. The key insight is that along with the results returned by dfuller that are listed by

return list

the estimates produced by the regression are still available as shown by

ereturn list

Example:

webuse air2, clear
dfuller air, regress
ereturn list
matrix b = e(b)
scalar drift = b["y1","_cons"]
display drift

. dfuller air, regress

Dickey-Fuller test for unit root                   Number of obs   =       143

                               ---------- Interpolated Dickey-Fuller ---------
                  Test         1% Critical       5% Critical      10% Critical
               Statistic           Value             Value             Value
------------------------------------------------------------------------------
 Z(t)             -1.748            -3.496            -2.887            -2.577
------------------------------------------------------------------------------
MacKinnon approximate p-value for Z(t) = 0.4065

------------------------------------------------------------------------------
       D.air |      Coef.   Std. Err.      t    P>|t|     [95% Conf. Interval]
-------------+----------------------------------------------------------------
         air |
         L1. |   -.041068    .023493    -1.75   0.083    -.0875122    .0053761
             |
       _cons |    13.7055   7.133673     1.92   0.057    -.3972781    27.80829
------------------------------------------------------------------------------

. ereturn list

scalars:
               e(rank) =  2
               e(ll_0) =  -705.6387933373031
                 e(ll) =  -704.1057594990987
               e(r2_a) =  .0142710546604324
                e(rss) =  158355.9401162112
                e(mss) =  3431.975967704871
               e(rmse) =  33.51256143409272
                 e(r2) =  .0212128077966266
                  e(F) =  3.055828604163919
               e(df_r) =  141
               e(df_m) =  1
                  e(N) =  143

macros:
            e(cmdline) : "regress D.air L1.air"
              e(title) : "Linear regression"
          e(marginsok) : "XB default"
                e(vce) : "ols"
             e(depvar) : "D.air"
                e(cmd) : "regress"
         e(properties) : "b V"
            e(predict) : "regres_p"
              e(model) : "ols"
          e(estat_cmd) : "regress_estat"

matrices:
                  e(b) :  1 x 2
                  e(V) :  2 x 2

. matrix b = e(b)

. scalar drift = b["y1","_cons"]

. display drift
13.705505

[Rene Macon]

Thanks William! I receive the following error message using your code:

. dfuller air, regress
Dickey-Fuller test for unit root                   Number of obs   =       143

                               ---------- Interpolated Dickey-Fuller ---------
                  Test         1% Critical       5% Critical      10% Critical
               Statistic           Value             Value             Value
------------------------------------------------------------------------------
 Z(t)             -1.748            -3.496            -2.887            -2.577
------------------------------------------------------------------------------
MacKinnon approximate p-value for Z(t) = 0.4065

------------------------------------------------------------------------------
       D.air |      Coef.   Std. Err.      t    P>|t|     [95% Conf. Interval]
-------------+----------------------------------------------------------------
         air |
         L1. |   -.041068    .023493    -1.75   0.083    -.0875122    .0053761
             |
       _cons |    13.7055   7.133673     1.92   0.057    -.3972781    27.80829
------------------------------------------------------------------------------

.
. ereturn list

.
. matrix b = e(b)
last estimates not found
r(301);

.
. scalar drift = b["y1","_cons"]
b not found
r(111);

.
. display drift
drift not found
r(111);

[/QUOTE]

What problem do I have?

[William Lisowski]

The problem you have is an oversight rather than insight on my part. It wasn't dfuller that left the regression results in memory - they were there from a previous regress I had run.

However, having done

viewsource dfuller.ado

I find an undocumented "certify" option that leaves the regression results intact as I had misled myself into believing they were. So with one change

webuse air2, clear
dfuller air, regress certify
matrix b = e(b)
scalar drift = b["y1","_cons"]
display drift

the code should now work for you.

[Rene Macon]

The output is now:

. dfuller air, regress certify

Dickey-Fuller test for unit root                   Number of obs   =       143

                               ---------- Interpolated Dickey-Fuller ---------
                  Test         1% Critical       5% Critical      10% Critical
               Statistic           Value             Value             Value
------------------------------------------------------------------------------
 Z(t)             -1.748            -3.496            -2.887            -2.577
------------------------------------------------------------------------------
MacKinnon approximate p-value for Z(t) = 0.4065

------------------------------------------------------------------------------
       D.air |      Coef.   Std. Err.      t    P>|t|     [95% Conf. Interval]
-------------+----------------------------------------------------------------
         air |
         L1. |   -.041068    .023493    -1.75   0.083    -.0875122    .0053761
             |
       _cons |    13.7055   7.133673     1.92   0.057    -.3972781    27.80829
------------------------------------------------------------------------------

. 
. matrix b = e(b)

. 
. scalar drift = b["y1","_cons"]
matrix operators that return matrices not allowed in this context
r(509);

. 
. display drift
drift not found
r(111);

Should I perhaps use a Mata export to get the matrix value of the drift parameter?

[William Lisowski]

The matrices from ereturn and return are standard Stata matrices, not Mata matrices. See help matrix to understand them.

I guess you're using a version earlier than Stata 16. Try the classic subscripting with column names (and dfuller only fits a single equation, a row number of 1 instead of a row name will do)

webuse air2, clear
dfuller air, regress certify
matrix b = e(b)
scalar drift = b[1,colnumb(b,"_cons")]
display drift

For what it's worth to anyone else reading this, Stata 16 not only expanded the applicability of column name and row name subscripts, but also allows subscripts to be applied to e() and r() matrices. So on Stata 16, the following works, to my surprise and joy.

webuse air2, clear
dfuller air, regress certify
scalar drift = e(b)[1,"_cons"]
display drift

[Rene Macon]

Indeed, I'm using Stata 15 and should update. Your code for Stata 15 works perfectly!

Thanks a lot!
Rene