Heteroskedasticity problem

[Klodi Gjini]

hello everyone,

I am conducting a VAR model for economic growth and globalization. Data is from 1980 to 2022. I have the AGR (annual growth rate of GDP in %) and KOF index (which was retrieved by Dreher 2006). KOF index (1-100 scale) was not stationary so i transformed the KOF into globalization=KOFt/KOFt-1 and solved the stationary issue. I have performed all the steps including: the selection of lags, the model stability, the autocorrelation issue, but now I am having troubles with heteroskedasticity of the residuals. I cannot perform the test in STATA (estat hettest) which says to me last estimates not found (error r301). My ultimate goal is to create Impulse response functions for my two variables AGR and globalization here below is my code

* Example generated by -dataex-. To install: ssc install dataex
clear
input int year float(AGR KOF globalization residuals)
1980      . 23.378174         .           .
1981      . 23.211115  .9928541           .
1982      . 22.783043  .9815574           .
1983      .  23.00968 1.0099477           .
1984  -1.25 22.789507  .9904312           .
1985   1.78 23.374466 1.0256679           .
1986   5.64  22.59832  .9667952   2.0702236
1987   -.79  23.07755 1.0212063  -4.6659617
1988  -1.42 24.654205 1.0683199   -1.821009
1989   9.84 24.972147 1.0128961    7.706656
1990  -9.58 24.640667   .986726    -16.7239
1991    -28    28.569 1.1594249   -22.26849
1992  -7.19  39.75262   1.39146      .34917
1993   9.56  40.83884 1.0273244 -.035226285
1994    8.3 32.801414  .8031916    1.012783
1995  13.32  31.96126  .9743866   11.984923
1996    9.1  30.01909  .9392337   2.1367476
1997 -10.92  28.93174  .9637781    -13.2087
1998   8.83 32.354862 1.1183171   14.909916
1999  12.89 36.109394 1.1160423   2.0066772
2000   6.95  39.19376 1.0854174    .3804192
2001   8.29  42.17943  1.076177    6.129879
2002   4.54   46.7698 1.1088296  -.01984159
2003   5.53   47.2565 1.0104063    2.845588
2004   5.51  47.91206 1.0138725    2.415801
2005   5.53  48.43103 1.0108316    2.234954
2006    5.9   50.5421 1.0435892     2.63613
2007   5.98  53.14189 1.0514382     2.19221
2008    7.5  56.46306 1.0624963     3.84975
2009   3.35  58.03797 1.0278927  -1.1103575
2010   3.71  62.98748 1.0852805   2.1493583
2011   2.55  61.31259  .9734091  -1.1066909
2012   1.42  60.45172  .9859594  -.15974425
2013      1  60.04907  .9933392   -.9909918
2014   1.77 65.107254 1.0842342    -.348368
2015   2.22  65.10788 1.0000097  -1.2856524
2016   3.31 66.630615 1.0233879   1.0711466
2017    3.8  68.52433  1.028421    .5187215
2018   4.07  68.01423  .9925559    .9109921
2019   2.21  68.49394 1.0070531   -.5588056
2020  -3.13  63.79173  .9313484   -5.208308
2021   8.63         .         .    9.517865
2022      3         .         .           .
end

I actually don't understand what I am doing wrong? Do you see any problem with my data? what can I do?

Looking forward your replies,
thanks in advance

[Klodi Gjini]

any suggestion please...

[Jeff Wooldridge]

You haven’t shown the Stata command you typed and the output given — as requested in the FAQ. It’s difficult to make suggestions without that.

[Klodi Gjini]

hello again,
I am using STATA14. I run a VAR model: var AGR and globalization,lags(3). predict residuals,residual. Now when checking for heteroskedasticity in the residuals, I run the code hettest which provides me with an error (last estimates not found
r(301);

[Rich Goldstein]

not sure where you are getting the -hettest- command (current and recent versions of Stata would use -estat hettest- but, as the original author, hettest was not designed to work after the -var- command

[Jeff Wooldridge]

There's no reason to test for heteroskedasticity unless you want to model the variance. Simply use vce(robust) in the -var- command. Then your inference is robust to any kind of heteroskedasticity.