Some querries. A robust panel data analysis was done by my statistician in my study covering two periods:
Panel A : 2008-2018 16 Oil companies and
Panel B: 1988-2018 - Three Major Oil companies only.
I am proving the presence of Market Structure, Conduct and Performance. 2008-2018 also includes the 3 major oil companies (who were the only oil companies from 1988 to 1997 - during the regulated period - no deregulation law yet)
a. How market structure - represented by market share-MS/CR/HHI affects market performance - represented by revenue, ROA and ROE, and
b. If there is reverse causality from market performance to market structure
Independent conduct variables are price, general admn expenses and fixed assets (current and lagged - used separately in the models). Have added independent external variables such as GDP, Crude Price, Foreign exchange, and dummy variable deregulation (D).
I would like to ask how I will explain the ff:
a. There are high, low and zero R squared resulting from the analysis. What are the explanations/conclusions that can be made on the R squared for within/between/overall if they are high, low or zero ?
b. The high Chi2 (greater than 0.05) in one of the models. Does this explain model has a poor fit ?
c. Is the "robust method" of panel data a new and more effective way of analysis (e.g. doing away with the tedious separate fixed effect and random effect model analysis and doing doing Hausmann test) ?
d. Are there other things that need to be done (before or after this robust panel data analysis) to check other diagnostics (autocorrelation/heteroskedascity/etc...) or does Stata already accounted and considered these diagnostics and found no problem already on the entered data ?
d. Is the use of Granger causality acceptable in proving reverse causality from market performance to market structure ?
Thanks for anybody who can help me explain.
Dear Virgilio Castillo ,
The main problem I see here is the low number of groups. Without a large N, you should looking more at time series than at applied microeconometrics.
Regarding your questions:
a. I would not worry very much about the fit of the model. Nowadays, the emphasis is on the marginal effects of the covariates of interes.
b. Idem.
c. In principle, you should go to FE directly, which warrants consistency. We seldom use RE.
d. Granger causality test does not help very much. Like Hausman's test, it is rarely used in reality.
The topic is very far from my interests but I think you are right about your concern on reverse causality. In that case, you would need an instrument variable approach (easy to say, but it is difficult to find a proper one). You can also try to defend that the causality goes in a certain direction in your case.
In panel data, you can easily use robust standard errors clustered at the firm level, which would allows you to deal with serial correlation and heteroskedasticity, but, in this case, N is very small (16 or 3), so this does not work. You should think on a method without this requirement.
I would estimate the model using FE (implemented just pluggin in dummies for the firms) and I would model serial correlation (e.g., using the Newey West estimator available in in Stata)
Here, I did something like that: Article The impact of the legalisation of abortion on birth outcomes...
Best,
José-Ignacio
Thank you very much, José-Ignacio Antón
Will review your article soonest and hopefully I can move on with my paper.
Regards.....
Dear Virgilio,
I assume you are talking about more robust panel data estimations besides the usual fixed and random effects models.
Of course, there are more robust panel data estimations that takes care of panel heterogeneity and endogeneity issues inherent in panel data estimation.
I would suggest you read about IV and GMM estimations in STATA.
And of course, there are post estimation tests to be carried out. These include, test for endogeneity, overidentifying restrictions etc.
Please read more or the instrumental variables (IV) and GMM models.
Should you need any further assistance, kindly let me know.
Kind regards,
- Badges
- Science topic
- Similar topics
- Engineering
- Petroleum Engineering
- Oil