Skip to content
All library documents

Choosing OLS or SURE for Fama–French Portfolio Regressions

Article Quant Q&A · Author: Biv
Read the original

Summary

The document asks whether estimating factor exposures for 25 portfolios requires seemingly unrelated regression (SURE), especially when the estimated intercepts feed into a GRS test. The response says there is no universally correct regression method: the choice depends on the model’s purpose and the tradeoffs that matter. It contrasts SURE with OLS and notes that finance research and industry practice may favor different estimation setups, including different approaches to standard errors.

The reply does not derive a recommendation for the specific portfolio regressions or explain the assumptions behind the GRS test. It points readers toward related discussions of factor-model construction and estimation. Thus, the main lesson is to choose an estimator based on the research question and inference requirements, rather than assuming one method is mandatory because it appears in a particular academic application. The answer is brief, and its general comments should not be treated as a detailed comparison of estimator properties.

Key ideas

  • Regression choice depends on the model’s purpose and the desired inference.
  • SURE is not presented as a universally required estimator for portfolio factor regressions.
  • The question’s concern about using estimated intercepts in a GRS test does not receive a specific technical resolution.
  • Academic and industry workflows may use different regression and standard-error conventions.

Tags

EquitiesFactor investingStatistics
Full text
# Answer by pyCthon (score 2)


# Is it compulsory to use seemingly unrelated regression method while running the Fama French style 25 portfolios on the independent risk factors?












I have scanned rigorously on the internet but the information regarding the SURE methodology seems surprisingly very little (at least with regard to its application in finance). I would have imagined if Fama French used this technique then, surely there might information on it out there. However, it's the exact opposite case.

My problem is that I have already calculated all the estimates for intercept terms and respective betas for the 25 portfolios using OLS. Only now I am realizing that I should apparently use this SURE model and not the OLS method.

This was the lecture (https://www.youtube.com/watch?v=Csajx7C3m5M&t=3723s) by Prof. Klaus Grobys I was watching when it occurred to me that I should be using the SURE model because the intercepts would go on to be included in the GRS test statistic.

What is the correct method?

## Answer by pyCthon (score 2)

https://quant.stackexchange.com/a/69234

There unfortunately is no "correct" regression method, all methods will have pros and cons with regards to the goals intended of the model. It is important to note what is often cited in academia aka Fama–MacBeth (time-series) regression is not the standard for the industry aka OLS with Newey West standard errors (cross-sectional) regression.

There's a wealth of knowledge from previous posts below on the topic.

How to build a factor model?

Which approach to estimating fundamental factor models is better, cross-sectional (unobservable) factors or time-series (observable) factors?

When should you build your own equity risk model?

Shown in full with attribution under the source's licence. Licence: CC BY-SA 4.0 (Stack Exchange)

This summary was written by Stratmill's research agent from the original; it is not a copy of the source.