[Andrew Musau]

If you simply want to graph the results, I do not see the need for suest. The advantage of suest is, e.g., if you intend to do tests for cross-model hypotheses. If you can explain what you precisely want to do, better suggestions may be forthcoming,

[David Buegel]

I constructed a measure for monetary policy shocks (based on the model in Coibion and Gorodnichenko 2017 which itself is based on Romer and Romer 2004) during the time when the ffr was at the ZLB.
To calculate the shocks, I used the Wu-Xia Shadow FFR, or rather the change in the Shadow Rate as the dependent variable. Now I'm analysing the "impact" of these new monetary policy shocks on main aggregates such as GDP and unemployment, I will also look at wealth inequality measures later.

While they look at a horizon of 20 quaters when checking for the impact of a MP Shock on different aggregates, I look at a horizon of 8 quaters as I only have 32 quaters in total (2009-1026). Therefore, I have 9 equations: period 0, period 1, period 2, ... ,period 8. Each equation has the change of the aggregate in 0, 1, 2, ... 8 as the dependent variable and is regressed on the changes of the aggregate in -1 and -2 aswell as the shock in -1, -2, ..., -8. I am focusing on the effect of the shock in -1 on the change of the aggregate across all equations. So, i want to see how the Shock in -1 effects the changes the dependent variable in the following 9 periods after -1.

Using -suest- yields lower standard errors compared to -reg, robust-. Now, I know how to plot the coefficients from -suest-. The joint significance test after suest also yields good result. But, if viable, I would also like to plot the coefficients from suest instead the one from -reg, robust-. That's why I was wondering about the whole t-statistics and z-statistics thing in my last post as I am not entirely sure whether or not I can do that.

[Andrew Musau]

If the sample sizes are less than 30, better to report and graph the confidence intervals based on the t-distribution.

[David Buegel]

Got it, thanks Andrew.
But can I still use -suest- and z-statistics to test for joint significance then? Just want to make sure...

[Andrew Musau]

Yes you can.