Choosing Prior Parameters for Black–Litterman Return Views

Summary

The document asks why a Black–Litterman likelihood or return distribution might use the investor’s view means and observed market covariance as its parameters. The response frames the choice in Bayesian terms, describing it as a simple way to obtain a prior distribution. It suggests a more tailored alternative: estimate a prior whose mean and variance fit the assumed return distribution and available data, then use that prior to form the posterior.

A normal-gamma setup is mentioned as one way to estimate such prior parameters. The answer remains a brief conceptual comment: it does not derive the Black–Litterman equations, distinguish the roles of prior, likelihood, and view uncertainty in detail, or compare parameter choices empirically. It also gives no data example or evidence that the suggested alternative improves portfolio outcomes. The discussion is useful as a reminder that familiar parameter choices are modeling assumptions and that a fitted prior may be considered, but implementation depends on the investor’s return-distribution assumptions and data.

Key ideas

• The question concerns the assumptions behind parameter choices in a Black–Litterman return distribution.
• The response interprets the simple choice of view means and market covariance as a convenient way to specify a prior.
• A fitted prior can instead be estimated from return-distribution assumptions and available data.
• A normal-gamma model is mentioned as one approach to estimating prior parameters.
• The document gives no derivation, empirical comparison, or evidence about portfolio performance.

Tags

# Rationale for likelihood function parameter choice in Black-Litterman model?


# Rationale for likelihood function parameter choice in Black-Litterman model?












So we are interested in a PDF for equilibrium returns given the views. Why do we choose our view means as the mean parameter and observed market covariance as the covariance parameter? Seems a bit arbitrary.

## Answer by numerairX (score 1)

https://quant.stackexchange.com/a/45639

Since you add the bayes-theory tag here I'm gonna speak in bayesian interpretation; I'd say it's just because this is the simplest way to obtain the distribution of prior; A better way to do this is by finding a prior optimal (essentially finding best mean and variance that fits our assumption of return distribution based on the data you have; usually done in a normal-gamma setting) and then feed to your posterior.

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.