Problem with Heteroskedasticity

[Lorenzo Panetta]

Hello!

I have panel data comprising 26 countries and 22 years. I want to test the effect of the political and legal environment on the M&A activity both measured through the number of deals and total volume.
After running the Hausman test, Fixed Effect is suggested but, under "robust" setup all the independent variables lose significance.
I had in mind to take the first difference of my dependent variables, but I have a couple of questions regarding the code and the way to proceed.

 xtunitroot llc num_deals if country, lags (4)

With the Augmented DF, resulting in a 0.58 p-value, it suggests the presence of unit root, while with the second Y variable, I have a p-value of 0.002

xtunitroot llc ln_value_deals if country, lags (4)

1) what's the right set to run the first difference, justifying it according to unit-root test?
2) is there a way in which we can deal with heteroskedasticity to obtain significant results?

Many thanks in advance.

[Carlo Lazzaro]

Lorenzo:
welcome to this forum.
1) see -D.- operator in -help tsvarlist-;
2) nothing, except invoking- robust- because of heteroskedasticity, with no expectation about statistically significant results, as results are what they are.

[Lorenzo Panetta]

rewrote below

[Lorenzo Panetta]

Lorenzo:
welcome to this forum.
1) see -D.- operator in -help tsvarlist-;
2) nothing, except invoking- robust- because of heteroskedasticity, with no expectation about statistically significant results, as results are what they are.

Carlo,

Thank you so much for your answer!
I have already tried d. command to execute the first difference, but when I run ADF variables still shows unit-root presence.
I used a Principal Component Analysis (PCA) to create a new variable out of 4 with high degree of correlation with the following code

pca gov_int gov_eff jud_ind prop_ri // Principal Component Analysis (create a new variable)
screeplot // 1 variable above 1 -> 1 component Eigenvalue
pca gov_int gov_eff jud_ind prop_ri, mineigen(1)
estat loadings
predict pol_var, score // new variable, representing 4 with high correlation
estat kmo // Kaiser-Meyer-Olkin measure of sampling adequacy > 0.5 -> high correlation, good

only under this setting the regression

reghdfe ln_num_deals pol_var tax cap_contr burea_cost gdp_grw infl debt int_rate mkt_ret mkt_vol, absorb (id) vce(cluster id)

shows me significant results.

[Carlo Lazzaro]

Lorenzo:
as you absorbed -id- only and clustered on a single variable, your -reghdfe- code is like:

 
 xtset id timevar xtreg ln_num_deals pol_var tax cap_contr burea_cost gdp_grw infl debt int_rate mkt_ret mkt_vol i.timevar, fe vce(cluster id)

In addition, -timevar- should be included in your code.