IV-panel-xtivreg: zero coefficient; time fixed effects

[Melody Guo]

Hello Statalist! I'm doing the project about UIP(uncovered interest rate parity) and illiquidity risk and using fiscal shock as illiquidity's the instrument variable. Thus, I ran a TSLS regressions using STATA.

The data I process is the unbalanced data with id (countrypair group for exchange rate, 6 groups in total) and ym(year, month) where year from 2000-2015, but some from 2003-2015.
xtset id ym
panel variable: id (unbalanced)
time variable: ym, 480 to 690
delta: 1 unit

My regressions are shown as the following:
1st: xtreg illiquidity TFWDVP_SA l.abslibor bas i.ym, fe vce(robust)
2nd: xtivreg absy l.abslibor_1m bas i.ym (illiquidity=TFWDVP_SA i.ym), fe vce(robust)


whereTFWDVP_SA is the fiscal shock which is time series data, that is, for different countrypair at t they are invariant.
abslibor is a monthly libor rate; absy is the absolute value of return; bas is the bid-ask spread; i.ym is the time-fixed effect.

My question is:
If I put "i.ym" inside the parenthesis like: xtivreg absy l.abslibor_1m bas (illiquidity=TFWDVP_SA i.ym), fe vce(robust) , it will generate a different 1st stage result for the coefficient on illiquidity compared with putting "i.ym" outside of the parenthesis like: xtivreg absy l.abslibor_1m bas i.ym (illiquidity=TFWDVP_SA), fe vce(robust).

In addition, as shown in the attached pictures, xtivreg absy l.abslibor_1m bas (illiquidity=TFWDVP_SA i.ym), fe vce(robust) can get a significant coefficient for "illiquidity"; However, xtivreg absy l.abslibor_1m bas i.ym (illiquidity=TFWDVP_SA), fe vce(robust) get a "0" for "illiquidity" with omitted STD error.

I am confused about the results and why i.ym will lead to different results for inside and outside of the parenthesis. Is there a bug in Stata for causing this issue or other data issue?

Hope you can help me fix this problem. Thank you so much.

[Melody Guo]

Hope anyone can help me solve this problem. Thank a lot.

PS: Stata version-SE15.0

[Phil Bromiley]

You didn't get a quick answer. You'll increase your chances of a useful answer by following the FAQ on asking questions - provide Stata code in code delimiters, readable Stata output, and sample data using dataex. Pictures are not encouraged.

When i.ym is only in the parentheses, it is only used as an instrument for illiquidity. When i.ym is in the main model, it is used as part of the main model. These are very different uses of the variable. Note that putting i.ym as an instrument in xtivreg assumes it is exogenous to the structural equation, but then you put it in the structural equation which violates that assumption.

It looks like i.ym is colinear with the predicted illiquidity. I would look at the connection between i..ym and TFWDVP_SA. I'd also do a regression of illiquidity on TFWDVP_SA and then do a predicted value. Look at the tie between the predicted value and i.ym - try regressing the predicted value on i.ym.