Syntax error in a while loop with mata

[Nora Kaimann]

I'm sorry to post this problem, but I'm getting increasingly desperate ...

I'm using a loop to get rid of autocorrelated residuals, but I run in a syntax error with the following code:

regress variable1_variable2
drop resid
predict double resid, res

local xtwelfthpv = 0
local xlags = 0

while `xtwelfthpv' < 0.05 {
quietly regress D.resid L.resid L(0/`xlags')D.resid, nocons
quietly actest, lags(12) robust

mata  {
b = st_matrix("r(results)")
m = b[.,4]
st_numscalar("twelfthpv", m[12])
m[12]
      }
local xtwelfthpv = `twelfpv'
local xlags = `xlags' + 1
                         }

di `xtwelfthpv'
di `xlags' 

The problem seems to be that

st_numscalar("twelfthpv", m[12]) 

does not transfer the value of m[12] to xtwelfthpv, since di `xtwelfthpv' yield "0" while m[12] equals 1.

Can anybody tell me, where my syntax error lies? Thanks a lot!

[Nick Cox]

This just looks like confusion between scalars and locals and a series of typos. You assign a scalar called twelfthpv and then look for it in a local macro called twelfpv. If you have a scalar, there is no need to, and no advantage in, putting it in a scalar that can be guessed from your code.

It is difficult to be sure because other typos appear in your code. The first regress statement looks wrong for a start.

[Nora Kaimann]

Hello Nick! Sorry for the typos, which were the result of a "diagnostic" check of the names. The regression command works. Eliminating the typos and using only scalars does not solve the problem - even though "xtwelfthpv" and "xlags" have now the correct values. The error message reads now "} is not a valid command name" The corrected and modified code looks now like this:

regress ln_CPI_GER ln_CPI_FRA ln_nat_curr_GER_FRA if  month >= 0 & month <= 156
drop resid
predict double resid, res

scalar xtwelfthpv = 0
scalar xlags = 0

while `xtwelfthpv' < 0.05 {
quietly regress D.resid L.resid L(0/`xlags')D.resid, nocons
quietly actest, lags(12) robust

mata  {
b = st_matrix("r(results)")
m = b[.,4]
st_numscalar("twelfthpv", m[12])
m[12]
      }
scalar xtwelfthpv = twelfthpv
scalar xlags = xlags + 1
                         }

di xtwelfthpv
di xlags 

What could be the problem? Thanks for any help!

[Nick Cox]

Still confusion between scalars and locals: for example, you define a scalar xlags and then refer to an undefined local xlags.

I can't test anything (no data example; do please read FAQ Advice #12) but with some further simplifications this may be closer to what you seek. I am a big fan of Mata but I can't see that you need it here.

regress ln_CPI_GER ln_CPI_FRA ln_nat_curr_GER_FRA if inrange(month, 0, 156) 
drop resid
predict double resid, res

scalar twelfthpv = 0
local xlags = 0

quietly while twelfthpv < 0.05 {
   regress D.resid L.resid L(0/`xlags')D.resid, nocons
   actest, lags(12) robust
   mat b = r(results)
   scalar twelfthpv = b[12, 4]
   local ++xlags
}

di twelfthpv

[William Lisowski]

To add to Nick's advice, let me explicitly call to your attention that references to scalars are not to be enclosed in quotation marks (`'). Compare the while command in post #3 with that in post #4. I think Nick considered that a typo but I suspect it's a misunderstanding of the syntax.

[Nora Kaimann]

Thanks a lot Nick! Now it works. The program starts with zero lags and adds one additional lag to an Engle-Granger cointegration test until H0 of no serial correlation (according to the Cumby-Huizinga test) over a rage of 12 month cannot be rejected a 5%. The data for the example are as follows:

* Example generated by -dataex-. To install: ssc install dataex
clear
input float(month ln_CPI_GER ln_CPI_FRA ln_nat_curr_GER_FRA)
 0  3.203153 2.3360198 1.0447346
 1 3.1978564  2.338917 1.0472068
 2 3.1978564  2.338917 1.0475808
 3  3.203153  2.339881 1.0482538
 4  3.205993 2.3379521 1.0488515
 5  3.205993 2.3369865 1.0486274
 6  3.205993  2.340844 1.0488515
 7  3.205993 2.3504224 1.0488515
 8  3.203153 2.3513753 1.0488515
 9  3.205993 2.3532782 1.0486912
10 3.2088256 2.3561258 1.0488406
11 3.2088256 2.3580198 1.0484669
12 3.2172744 2.3589654 1.0499076
13 3.2172744 2.3599102 1.0488515
14 3.2172744 2.3589654  .9995538
15 3.2172744 2.3580198  .9989347
16  3.225653 2.3561258  .9995538
17 3.2311995 2.3542283 1.0013821
18 3.2311995 2.3617969 1.0030102
19 3.2311995 2.3674362 1.0029819
20 3.2311995 2.3730435  1.003411
21 3.2311995  2.384165 1.0031863
22  3.236716  2.394252 1.0064545
23  3.236716  2.397895 1.0062741
24 3.2476575  2.404239 1.0059005
25 3.2503746 2.4051416 1.0068873
26  3.253084 2.4114394 1.0058656
27  3.255786 2.4132316 1.0075009
28  3.255786  2.415914  1.005253
29  3.255786  2.420368 1.0050489
30  3.258481  2.423917 1.0062741
31  3.253084  2.423031 1.0072963
32  3.253084  2.427454 1.0075009
33  3.255786  2.429218 1.0103692
34  3.258481  2.435366 1.0083195
35  3.263849  2.438863 1.0066829
36  3.277145 2.4466856 1.0081148
37  3.285038  2.452728 1.0072963
38  3.285038  2.455306 1.0056614
39  3.287655  2.457878 1.0050489
40  3.285038  2.461297 1.0028065
41  3.279783 2.4680996 1.0021958
42  3.279783  2.472328 1.0034176
43  3.277145  2.475698 1.0023993
44  3.279783 2.4824035 1.0019717
45  3.287655  2.484073 1.0013821
46  3.290266 2.4874036 1.0005691
47 3.2980566  2.489065  1.000602
48  3.303217  2.492379 1.0005691
49  3.305787 2.4932055 1.0001956
50  3.305787  2.494857 1.0009756
51  3.305787  2.495682 1.0007724
52  3.305787  2.496506 1.0005691
53  3.303217  2.498152 1.0009756
54  3.305787  2.500616 1.0011789
55  3.305787  2.503074 1.0011789
56  3.305787  2.507157 1.0007724
57  3.308351  2.510412 1.0009756
58 3.3134584  2.510412 1.0011789
59 3.3210714  2.510412 1.0013821
60  3.323596  2.516082 1.0015855
61  3.323596   2.51689 1.0007514
62 3.3286266  2.519308 1.0015855
63 3.3311324  2.520917 1.0011911
64  3.333632  2.524127 1.0060698
65 3.3410935  2.540026 1.0079101
66  3.343568  2.531313 1.0099589
67  3.343568  2.528126 1.0103502
68  3.343568  2.531313 1.0089126
69 3.3484995  2.532903 1.0071273
70  3.353407   2.53449 1.0070722
71   3.35829 2.5384474 1.0081507
72 3.3631494  2.541602 1.0105923
73   3.36557 2.5431755 1.0105554
74  3.367985  2.545531 1.0113955
75 3.3751955 2.5486636 1.0118064
76 3.3751955  2.552565 1.0095853
77 3.3751955  2.551786 1.0070918
78 3.3751955  2.554899 1.0048448
79  3.372798  2.555676 1.0025882
80  3.372798 2.5587766  .9977664
81 3.3751955  2.560323  .9932367
82  3.384729  2.562639  .9909847
83 3.3870986  2.565718   .990997
84  3.389462  2.569554  .9898063
85  3.389462 2.5703194  .9910139
86   3.39182  2.573375    .99084
87   3.39182  2.574138  .9936731
88  3.394173  2.575661   .999503
89  3.394173  2.575661 1.0036757
90  3.394173 2.5787005 1.0073123
91  3.389462  2.581731   1.00625
92  3.389462  2.585506 1.0072927
93   3.39182  2.590017  1.007946
94  3.394173  2.597491 1.0019639
95  3.394173  2.598979 1.0052426
96  3.403528  2.608598 1.0046728
97  3.405853  2.609334 1.0032475
98  3.405853   2.61007  .9982824
99  3.405853 2.6130066  .9969072
end
format %tm month

The program code is:

scalar twelfthpv = 0
local xlags = 0
quietly while twelfthpv < 0.05 {
   egranger ln_CPI_GER ln_CPI_FRA ln_nat_curr_GER_FRA if inrange(month, 0, 156), lags(`xlags') regress
   actest, lags(12) robust
   mat b = r(results)
   scalar twelfthpv = b[12, 4]
   local ++xlags
} 

Omitting "quietly" shows, how adding lags from 0 to 2 increases the corresponding p-value from 0.0030 over 0.0394 to 0.0848. Maybe this is useful for others too.

But what was the problem with my code? As I see, you have omitted my "mata inlay". I thought mata in a while-loop should work? I admit my problems with local variables. I tried to work myself through the corresponding passages in the reference manual, but found them difficult to understand. Most people learn faster with good examples. I find the examples in the reference manual typically quite complex. As a textbook writer, I always start with simple examples and than add some complexities...

[Nick Cox]

I don't think you can go

mata {

}

[Niels Henrik Bruun]

Hi
I would run Stata inside Mata and I would avoid mixing Stata and Mata code.
Like (code is not tested, code from #4 since it seems to be working):

regress ln_CPI_GER ln_CPI_FRA ln_nat_curr_GER_FRA if inrange(month, 0, 156)
drop resid
predict double resid, res

mata:
    twelfthpv = 0
    xlags = 0
    while ( twelfthpv < 0.05 ) {
        stata(sprintf("regress D.resid L.resid L(0/%f)D.resid, nocons", xlags))
        stata(sprintf("actest, lags(12) robust"))
        twelfthpv = st_matrix("r(results)")[12,4]
        ++xlags
    }
    st_numscalar("twelfthpv", twelfthpv)
end

I have used

mata{ }

inside a Stata while loop