Prof Shi, in most cases, it is difficult to find data that fulfill all the ordinary least squares (OLS) assumptions. This may explain why some researchers choose better methods. It appears that OLS has been relegated to a simple baseline method, especially in finance and economics research.

For estimation purpose, the OLS method often produces the same estimates as more advanced methods do. They are different in that those advanced methods, used when some of OLS assumptions are violated, have smaller standard errors of the estimates (e.g. more efficient). I agree with B.T. Matemilola that OLS could be thought as a baseline.

As I read the answers above, I tended to agree with parts of J. Shi, B.T Matemilola, R. Sharma, and A. Andriansyah. We all recognize the limitations of OLS; however, I doubt any of us would be comfortable ruling it out before seeing the model and the data available, especially in the vast and diverse area of economics and finance research. In addition, I am not sure where I would start in teaching econometrics this semester if OLS was stripped from the course.

I think B.T Matemilola's question begs us to think deeper in a different way. Consider the rate of change in refining theory, technique, and data. It seems to me refinement in the data (i.e., both what we collect and how we collect it), may be the slowest to change in economics. Although I am certain we could improve if we could go back and start over, part of our challenge (especially with time series data) is that we often cannot retroactively refine the data. I suspect few of us would care to sacrifice consistency and/or continuity if we could start over now. We are fortunate that what we research is relatively more easy to quantify than other disciplines and those who preceded us devised more or less effective ways to measure variables. I have been fascinated with the ability to adapt in economics, and especially econometrics, and without consciously doing it, it seems like we have followed the least-costly path of improvement in expanding the frontiers of knowledge.

Even poor R-squared values regressions with significant Student t statistics is a significant indication of a relationship between the predictor and the observed variable.

In corporate Finance endogeneity is very present. So you have to use fixed effects regression to eradicate the prevailing heterogeneity. But you may be also facing reverse causality. For example performance may be high since firm's governance is good, but the reverse may occur. It means that a good performance may have promoted a higher level of governance. Thus GMM (Generalized method of moments) may be necessary to use to disentagle the different effectsl, using lagged variables.

My experience is that OLS may be used in modelling distribution of economic variables. For example if you apply it to Lorenz curves you may obtain interesting models and different complex distributions. There are cases in which OLS method contradict main premises and we need to work it with simpler polinomies. OLS requires to be controlled as many other methods. Thanks, emilio