"no observations r(2000)" error when using Wooldridge test for autocorrelation in panel data set

[Joeri Nelemans]

Dear stata list,

I have a question regarding the Wooldridge test for autocorrelation (command: 'xtserial') (Drukker, 2003)**.

After conducting the test, Stata gives an error of "no observations r(2000)".

To install the Wooldridge test I have followed the following steps:

net sj 3-2 st0039
net install st0039

After that:

. xtset Country_ID Year

panel variable: Country_ID (unbalanced)
time variable: Year, 2014 to 2016, but with gaps
delta: 1 unit

. xtserial DVEgovdevelopment IVLegal IVTechnical IVOrganisational IVCapacitybuidling IVCooperation CVGNI CVCorruption CVInnovation

no observations
r(2000);


I note that other have said that there is a problem with xtserial and unbalanced time periods.

Can this be the case, and if yes, is there another way to test for autocorrelation in an unbalanced panel dataset?
Help would be appreciated!

Regards,

Joeri

[Nick Cox]

Show us the results of

tab Year 

if you have just 2014 and 2016 then no panels can be used without changing the setting. If all values are just 2014 and 2016, then use

xtset Country_ID Year, delta(2)

[Carlo Lazzaro]

Joeri:
the problem here seems to rest on the pretty small time-series (T) dimension: a small number of gaps is enough to make the community-contributed command -xtserial- gasping
(see https://pdfs.semanticscholar.org/234...b5ef75747b.pdf, that shows an example of -xtserial- working successfully in a panel dataset with gaps but with a pretty large T dimension).
Unfortunately, I'm not aware of other options to investigate the serial correlation of the idiosyncratic error in panel datasets.

[Joeri Nelemans]

tab Year

Year | Freq. Percent Cum.
------------+-----------------------------------
2014 | 193 50.00 50.00
2016 | 193 50.00 100.00
------------+-----------------------------------
Total | 386 100.00

So yes indeed, all values are either 2014 or 2016.

I tried

xtset Country_ID Year, delta(2)

But doing the xtserial test afterwards results in the same error

[Nick Cox]

So, no way forward on that route except with more data.

[William Lisowski]

For what it's worth, the following demonstrates that xtserial requires at least 3 waves of data to function.

. use http://www.stata-press.com/data/r8/nlswork.dta, clear
(National Longitudinal Survey.  Young Women 14-26 years of age in 1968)

. gen age2 = age^2
(24 missing values generated)

. gen tenure2 = tenure^2
(433 missing values generated)

. 
. tsset idcode year
       panel variable:  idcode (unbalanced)
        time variable:  year, 68 to 88, but with gaps
                delta:  1 unit

. xtserial ln_wage age* ttl_exp tenure* south

Wooldridge test for autocorrelation in panel data
H0: no first order autocorrelation
    F(  1,    1472) =     88.485
           Prob > F =      0.0000

. drop if year>70
(24,241 observations deleted)

. xtserial ln_wage age* ttl_exp tenure* south

Wooldridge test for autocorrelation in panel data
H0: no first order autocorrelation
    F(  1,     650) =     13.706
           Prob > F =      0.0002

. drop if year>69
(1,686 observations deleted)

. xtserial ln_wage age* ttl_exp tenure* south
no observations
r(2000);

end of do-file

r(2000);

[Ngoc Bui]

Dear stata list,

I have a question regarding the Wooldridge test for autocorrelation (command: 'xtserial') (Drukker, 2003)**.

After conducting the test, Stata gives an error of "no observations r(2000)".

To install the Wooldridge test I have followed the following steps:

net sj 3-2 st0039
net install st0039

After that:

. xtset Country_ID Year

panel variable: Country_ID (unbalanced)
time variable: Year, 2014 to 2016, but with gaps
delta: 1 unit

. xtserial DVEgovdevelopment IVLegal IVTechnical IVOrganisational IVCapacitybuidling IVCooperation CVGNI CVCorruption CVInnovation

no observations
r(2000);


I note that other have said that there is a problem with xtserial and unbalanced time periods.

Can this be the case, and if yes, is there another way to test for autocorrelation in an unbalanced panel dataset?
Help would be appreciated!

Regards,

Joeri

Dear Joeri Nelemans,
I have the same problems with my data.
What should I do now?
Please advise me.
Thanks so much.

[Jeff Wooldridge]

Just to make sure Nick's point gets across: you cannot test for serial correlation when using fixed effects or first differencing when T = 2. More importantly, there is absolutely no reason to test for serial correlation. It has no implications for estimation or inference. When T = 2, there is only one estimator because FE = FD. And what is FD when T = 2? It's an OLS regression on the first differences. There is no time series dimension in this regression. Therefore, serial correlation is irrelevant.

BTW, and this is a bit awkward since my name is attached to the command, but I would've implemented xtserial differently. First, I originally suggested a cluster-robust t test, rather than an F test. Since F = t^2, this is not a big deal, but I'd still like to know the sign of the t statistic. More importantly -- and this is related to not knowing the sign -- the command does not report the estimated serial correlation coefficient! We always teach our students we should care about magnitudes (and signs), not just statistical significance. The null is that the first order autocorrelation in the FD errors is -.5. How far is the estimate from -.5? Is it above it or below it? Plus, it is easy to estimate the autocorrelation in the original errors, u(i,t).

I always implement the test I proposed by hand:

xtset id year
reg D.(y x1 x2 ... xK year2 year3 ... yearT), nocons
predict ehat, resid
reg ehat L.ehat, vce(cluster id)
llincom L1.ehat + .5

The final command gives the estimate of rho in the original AR(1) model.

[Daniel Bembennek]

Hello,

I have the same problem with "xtserial" command, but my dataset contains values over 20 years.
"
tsset
panel variable: pairstates2 (weakly balanced)
time variable: year, 1999 to 2015
delta: 1 year

. xtserial FDI_OUT_dt RESTR_d log_DISTANCE log_TAX_d log_GDP_d, output
no observations
r(2000);
"

Do you have any ideas? I didn't find the solution on another answers on this question on statalist.

Many thanks in advance!

best regards,
Daniel

[Carlo Lazzaro]

Daniel:
could you please provide an example/excerpt of your data via -dataex-? Thanks.

[Saad Aftab]

Hello,
I have the same problem with fourteen years of unbalanced quarterly panel data. The xtserial command along with variables gives the r(2000) no observations error. Here is a snap shot of my data. Kindly help.


tab Quarters

Quarters | Freq. Percent Cum.
------------+-----------------------------------
01jan2006 | 1,180 1.07 1.07
31mar2006 | 1,080 0.98 2.05
30jun2006 | 1,104 1.00 3.05
30sep2006 | 1,123 1.02 4.06
01jan2007 | 1,282 1.16 5.22
31mar2007 | 1,185 1.07 6.29
30jun2007 | 1,203 1.09 7.38
30sep2007 | 1,234 1.12 8.50
01jan2008 | 1,330 1.20 9.70
31mar2008 | 1,282 1.16 10.86
30jun2008 | 1,315 1.19 12.06
30sep2008 | 1,328 1.20 13.26
01jan2009 | 1,457 1.32 14.58
31mar2009 | 1,330 1.20 15.78
30jun2009 | 1,330 1.20 16.98
30sep2009 | 1,352 1.22 18.21
01jan2010 | 1,760 1.59 19.80
31mar2010 | 1,494 1.35 21.15
30jun2010 | 1,571 1.42 22.58
30sep2010 | 1,640 1.48 24.06
01jan2011 | 1,953 1.77 25.83
31mar2011 | 1,786 1.62 27.44
30jun2011 | 1,849 1.67 29.12
30sep2011 | 1,897 1.72 30.84
01jan2012 | 2,060 1.86 32.70
31mar2012 | 1,979 1.79 34.49
30jun2012 | 2,029 1.84 36.33
30sep2012 | 2,059 1.86 38.19
01jan2013 | 2,095 1.90 40.09
31mar2013 | 2,059 1.86 41.95
30jun2013 | 2,057 1.86 43.81
30sep2013 | 2,057 1.86 45.68
01jan2014 | 2,184 1.98 47.65
31mar2014 | 2,096 1.90 49.55
30jun2014 | 2,102 1.90 51.45
30sep2014 | 2,131 1.93 53.38
01jan2015 | 2,330 2.11 55.49
31mar2015 | 2,214 2.00 57.49
30jun2015 | 2,296 2.08 59.57
30sep2015 | 2,296 2.08 61.65
01jan2016 | 2,505 2.27 63.92
31mar2016 | 2,339 2.12 66.04
30jun2016 | 2,368 2.14 68.18
30sep2016 | 2,402 2.17 70.35
01jan2017 | 2,739 2.48 72.83
31mar2017 | 2,549 2.31 75.14
30jun2017 | 2,638 2.39 77.53
30sep2017 | 2,681 2.43 79.96
01jan2018 | 2,768 2.51 82.46
31mar2018 | 2,746 2.49 84.95
30jun2018 | 2,751 2.49 87.44
30sep2018 | 2,760 2.50 89.94
01jan2019 | 2,780 2.52 92.45
31mar2019 | 2,773 2.51 94.96
30jun2019 | 2,781 2.52 97.48
30sep2019 | 2,785 2.52 100.00
------------+-----------------------------------
Total | 110,474 100.00

. xtset StockCode Quarters, delta (2)
panel variable: StockCode (unbalanced)
time variable: Quarters, 1/1/2006 to 9/30/2019, but with gaps
delta: 2 days

. xtserial NumberofbankingRelationships AltmanZscore StateorPrivate FirmAge TotalLoans TotalAssets Undistribut
> edProfit TotalShareholdersEquity TangibleAssetsTotalAssets BondsLTLTA ROE
no observations
r(2000);