-actest- command (Cumby-Huizinga test for autocorrelation)

[Guest]

Dear all,

I ran the -actest- command to test for autocorrelation and below is my result:

. actest, lags(20)

Cumby-Huizinga test for autocorrelation (Breusch-Godfrey)
  H0: variable is MA process up to order q
  HA: serial correlation present at specified lags >q
-----------------------------------------------------------------------------
  H0: q=0 (serially uncorrelated)        |  H0: q=specified lag-1
  HA: s.c. present at range specified    |  HA: s.c. present at lag specified
-----------------------------------------+-----------------------------------
    lags   |      chi2      df     p-val | lag |      chi2      df     p-val
-----------+-----------------------------+-----+-----------------------------
   1 -  1  |    276.493      1    0.0000 |   1 |    276.493      1    0.0000
   1 -  2  |    305.429      2    0.0000 |   2 |     92.334      1    0.0000
   1 -  3  |    365.152      3    0.0000 |   3 |    109.552      1    0.0000
   1 -  4  |    391.180      4    0.0000 |   4 |     78.919      1    0.0000
   1 -  5  |    422.513      5    0.0000 |   5 |     90.395      1    0.0000
   1 -  6  |    434.675      6    0.0000 |   6 |     80.498      1    0.0000
   1 -  7  |    437.740      7    0.0000 |   7 |     57.399      1    0.0000
   1 -  8  |    437.864      8    0.0000 |   8 |     25.801      1    0.0000
   1 -  9  |    440.781      9    0.0000 |   9 |     26.477      1    0.0000
   1 - 10  |    441.528      10   0.0000 |  10 |     19.186      1    0.0000
   1 - 11  |    446.536      11   0.0000 |  11 |     24.134      1    0.0000
   1 - 12  |    446.605      12   0.0000 |  12 |     17.115      1    0.0000
   1 - 13  |    449.874      13   0.0000 |  13 |     24.292      1    0.0000
   1 - 14  |    454.015      14   0.0000 |  14 |     30.399      1    0.0000
   1 - 15  |    454.054      15   0.0000 |  15 |     16.738      1    0.0000
   1 - 16  |    455.875      16   0.0000 |  16 |     18.457      1    0.0000
   1 - 17  |    458.628      17   0.0000 |  17 |     21.598      1    0.0000
   1 - 18  |    458.731      18   0.0000 |  18 |      6.580      1    0.0103
   1 - 19  |    461.106      19   0.0000 |  19 |     14.337      1    0.0002
   1 - 20  |    461.108      20   0.0000 |  20 |     13.264      1    0.0003
-----------------------------------------------------------------------------
  Test allows predetermined regressors/instruments
  Test requires conditional homoskedasticity

How do I interpret the result above, and, if it tells me that there is autocorrelation, how many lags do I include?

[Mark Schaffer]

Short version: serial correlation as far as the eye can see (=20 lags here).

Longer version:

LHS block: null is no sc. Large test stat means reject the null (there is evidence of sc). Rows are where in time you look for sc. And wherever you look for it (t-1, or t-1 through t-10, or whatever) you find evidence e of it.

RHS block: null is possible sc up to lag q-1 but no sc at lag q. Large test stat means reject the null because there is evidence of sc at lag q as well. And whatever lag you look at, you find evidence of sc at that lag.