Calculating Standard Deviation of the Residuals

[Mario Marra]

Hi all,
I ran an ECM and after this, I want to inspect the Standard Deviation of the Residuals plot to see if there is volatile in my model.

Run model:

 regress Dy l(1/7).yi l(1/8).yd l(0/7).x1i l(0/8).x1d l.ei l.ed

after I get residuals and apply what I read here in #4

predict res, res
egen daily_sd_res=sd(res), by(date)

but what I get are just missing values.

Here a piece of my dataset. Data are

tsset
time variable:  date, 01jan2015 to 31dec2018
                delta:  1 day

* Example generated by -dataex-. To install: ssc install dataex
clear
input float(date Dy yi yd x1i x1d ei ed)
20089             .            0             0            0             0 0   -.08242687
20090     .02116996    .02116996             0            0             0 0    -.0612569
20091     .01106596    .01106596             0            0             0 0   -.05019094
20092    .010417402   .010417402             0            0             0 0   -.03977354
20093   -.007900596            0   -.007900596            0    -.06892812 0   -.01880446
20094   -.004676282            0   -.004676282            0    -.01835227 0  -.015794123
20095   -.002937913            0   -.002937913            0   -.016029835 0   -.01201814
20096  -.0044448376            0  -.0044448376   .010833502             0 0  -.021000454
20097    -.01337427            0    -.01337427            0    -.03350556 0   -.02034134
20098   -.005440772            0   -.005440772            0             0 0   -.02578211
20099   -.001662135            0   -.001662135            0             0 0  -.027444247
20100     -.0064767            0     -.0064767            0    -.02196181 0   -.02472252
20101    .009394586   .009394586             0            0   -.035164475 0 -.0005997333
20102   -.017572522            0   -.017572522   .019682646             0 0  -.026416086
20103    -.00355804            0    -.00355804    .03872418             0 0   -.04619326
20104    -.01020068            0    -.01020068   .006790996             0 0   -.05923827
20105   -.007450819            0   -.007450819            0             0 0   -.06668909
20106  -.0016266108            0  -.0016266108            0             0 0    -.0683157
20107   -.005518496            0   -.005518496   .006149173             0 0    -.0764097
20108    .000685513   .000685513             0            0    -.02171993 0   -.06662706
20109    .005839169   .005839169             0  .0046247244             0 0    -.0627249
20110    .004988313   .004988313             0            0   -.014763355 0   -.05155314
20111  -.0018172264            0  -.0018172264   .036330104             0 0  -.068586774
20112  -.0009147525            0  -.0009147525            0             0 0   -.06950153
20113 -.00013327599            0 -.00013327599            0             0 0    -.0696348
20114    .002408147   .002408147             0   .006327987             0 0   -.06987706
20115    .001775384   .001775384             0            0 -.00056409836 0   -.06786541
20116    .003261387   .003261387             0 .00052559376             0 0   -.06482416
20117 -.00005555153            0 -.00005555153            0  -.0022336245 0   -.06394418
20118   .0020005107  .0020005107             0    .01455152             0 0    -.0680384
20119  .00044727325 .00044727325             0            0             0 0   -.06759112
20120   .0007092953  .0007092953             0            0             0 0   -.06688183
20121   .0018749833  .0018749833             0    .08742452             0 0    -.1016235
20122    .006665409   .006665409             0    .04742861             0 0   -.11482298
20123    .016419828   .016419828             0  .0019119978             0 0   -.09920397
20124    .009312987   .009312987             0   .028562784             0 0   -.10185414
20125    .009942055   .009942055             0            0   -.015233994 0   -.08553152
20126    .007172227   .007172227             0            0             0 0    -.0783593
20127  -.0004681349            0  -.0004681349            0             0 0   -.07882743
20128    .010677457   .010677457             0    .03367603             0 0   -.08225475
20129     .01244551    .01244551             0            0    -.02256608 0   -.06035772
20130    .011302948   .011302948             0            0    -.04184544 0  -.031528335
20131    .005810559   .005810559             0    .05102551             0 0   -.04708917
20132    .004668951   .004668951             0    .06228065             0 0    -.0685057
20133   .0046052337  .0046052337             0            0             0 0   -.06390046
20134   .0020727515  .0020727515             0            0             0 0   -.06182771
20135    .009067178   .009067178             0    .02068734             0 0   -.06142516
20136    .003512859   .003512859             0            0   -.007105112 0   -.05493642
20137   .0021723509  .0021723509             0            0   -.007207036 0   -.04974549
20138    .002964258   .002964258             0            0    -.03206098 0  -.033352893
end
format %td date

Any suggestions as to why this happens?
Thanks in advance.

[Nick Cox]

Three points:

1. regress gives you RMSE which is as good if not better, and fits with standard regression machinery.

2. With your set-up, residuals are constants for each date: As Stata uses (sample size MINUS 1) not sample size in calculating SD, the result of a division by zero is what you see. There is no point therefore in invoking by(date).

3. Smoothing residuals -- or more precisely some scale like |residual| -- as a function of date seems closer to what you may need.