Skip to content
All library documents

Why the Inverse-Wishart Prior Simplifies Bayesian Covariance Modeling

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

Summary

The document asks why Bayesian asset-allocation models often assign an inverse-Wishart prior to the covariance matrix. The response gives a computational reason: this choice is conjugate, which simplifies the posterior calculations when updating beliefs about covariance from data. It connects the multivariate case to familiar priors for a single variance, such as the inverse-gamma or chi-square distributions.

The explanation offers a learning route: work through the univariate normal model with a variance prior, then view the Wishart family as its multivariate generalization. The answer does not develop the derivation, specify prior parameters, or compare the inverse-Wishart with alternative covariance priors. It therefore explains convenience rather than establishing that this prior is always an appropriate representation of uncertainty; modelers still need to assess the assumptions and suitability for their application.

Key ideas

  • An inverse-Wishart prior is used for covariance partly because it is conjugate and simplifies Bayesian updating.
  • The answer links multivariate covariance priors to variance priors in the univariate normal model.
  • The Wishart family generalizes familiar chi-square or gamma distributions to the multivariate setting.
  • Computational convenience alone does not establish that a prior is suitable for every asset-allocation model.

Tags

Portfolio constructionStatisticsRisk management
Full text
# Is there an open architecture API or excel solution for calculating and adjusting bond pricing?


# Is there an open architecture API or excel solution for calculating and adjusting bond pricing?












Is there an open architecture API for calculating and adjusting bond prices? I am looking to adjust bond pricing on OAS or adjusted spreads to various indices and need a tool to process large amounts of data.

## Answer by user26408 (score -1)

https://quant.stackexchange.com/a/32241

Yes, quantlib does that. QuantLib.org

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.