Two-step GMM

[Thuc Nguyen]

I am currently struggling with my thesis on the topic: The impact of liquidity risk and credit risk on the stability of commercial banks in Vietnam. My excel file includes data of 20 banks in Vietnam.

I generated year dummies. The stability is represented by z-score which is denoted as

With: U: average performance of the bank's assets (ROA). K: the capital ratio. σ: The standard deviation of ROA that is defined, as an indicator of the volatility of returns.

I have difficulty running two-step GMM test.
In the 1st part, which is the relationship between liquidity risk and credit risk, the results are satisfactory. However, in the 2nd part, which is the impact of liquidity risk and credit risk on the stability, the results are not satisfactory. Only the liquidity risk is significant, but the credit risk and liquidityrisk*creditrisk are not significant.

Can anyone please help me correct the command to get the result I want? Thank you so much.

[Sebastian Kripfganz]

Can anyone please help me correct the command to get the result I want?

I am afraid this is not how empirical research works. If we tweak the analysis until we get the desired results, then why to do an empirical analysis in the first place? We usually start with a hypothesis and then try to see if the data supports or refutes this hypothesis.

Nevertheless, here are a few comments on your specification:

• Note that all the variables listed in your iv() option are implicitly assumed to be strictly exogenous and uncorrelated with the unobserved bank-specific effect. This is essentially a random-effects assumption, which might be hard to justify. If you want to keep those variables as strictly exogenous but allow for correlation with the bank-specific error components - i.e., make a fixed-effects assumption - then you would need to replace eq(level) with eq(diff) inside that option.
• 20 banks is a very small cross-sectional sample size for such a dynamic panel data analysis. It might be too small to obtain reliable (and significant) results. As a minimum, you should strictly limit the number of instruments further by adding the lag() suboption to the gmm() option. It might also be advisable to only use the one-step instead of the two-step estimator, because the latter is likely based on a poorly estimated weighting matrix with such a small cross-sectional sample size, which could lead to substantial finite-sample distortions. After all, the two-step estimator is used to obtain asymptotic efficiency, but just having 20 banks means you are infinitely far away from asymptotia.

More on GMM estimation of linear dynamic panel data models:

• Kripfganz, S. (2019). Generalized method of moments estimation of linear dynamic panel data models. Proceedings of the 2019 London Stata Conference.