The Parameter Tau in Idzorek’s Version of the Black-Litterman Model

While on my sabbatical I spent some time working with Idzorek’s version [1] of the Black-Litterman asset allocation model.  I like Idzorek’s version because it gives an intuitive method for specifying one of the most complicated parts of the Black-Litterman model, the parameter \Omega.

Another interesting aspect of Idzorek’s version is that the parameter \tau disappears from the combined expected return vector, the model’s primary output.  This is nice, as specifying \tau is another of the model’s more complicated aspects.  Because of this, at least one author I know of concludes that \tau is irrelevant in Idzorek’s version of Black-Litterman.

However, it turns out that \tau is not irrelevant to Idzorek’s version of Black-Litterman.  I discovered this while trying to understand how the posterior covariance matrix works in Idzorek’s version.  While \tau does disappear from the combined expected return vector, it remains in the formula for the posterior covariance matrix.

(As an aside, Idzorek never explicitly discusses the posterior covariance matrix, other than to cite its formula derived in Satchell and Scowcroft [3].  Instead, his focus is on the combined expected return vector.)


Here’s the derivation.

If R denotes the excess return on the market, the Black-Litterman model assumes

\displaystyle R \sim N(\mu, \Sigma).

The original version of Black-Litterman also takes \mu to be a random variable having the following distribution:

\displaystyle \mu \sim N(\pi, \tau \Sigma).

The parameter \pi is the implied excess return vector from the capital asset pricing model.  The distribution of \mu given here is taken to be a Bayesian prior.

As shown in Meucci [2], the views, together with the assumed distribution of \pi, produce the following posterior distribution for R:

\displaystyle R|v; \Omega \sim N(\mu_{BL}, \Sigma_{BL}),

where

\displaystyle \mu_{BL} = \pi + (\tau \Sigma) P^T (P (\tau \Sigma) P^T + \Omega)^{-1}(v - P \pi),

\displaystyle \Sigma_{BL} = \Sigma + \left(I - (\tau \Sigma) P^T (P (\tau \Sigma) P^T + \Omega)^{-1} P \right) (\tau \Sigma),

v is the vector of view returns, and P is the “pick” matrix specifying which views are applied to which assets.

Following Idzorek’s approach of estimating \Omega' = \Omega/\tau rather than \Omega and \tau separately, the equation for \mu_{BL} becomes

\displaystyle \mu_{BL} = \pi + (\tau \Sigma) P^T \left(P (\tau \Sigma) P^T + \tau \frac{\Omega}{\tau} \right)^{-1} (v - P \pi) \\  = \pi + (\tau \Sigma) P^T \tau^{-1} \left(P \Sigma P^T + \frac{\Omega}{\tau} \right)^{-1} (v - P \pi) \\  = \pi + \Sigma P^T (P \Sigma P^T + \Omega')^{-1} (v - P \pi).

As we can see, estimating \Omega' rather than \Omega means that \tau is eliminated from the formula for the combined expected return vector.

However, the same calibration does not result in the removal of \tau from the formula for the posterior covariance matrix \Sigma_{BL}.  Instead, the equation for \Sigma_{BL} becomes

\displaystyle \Sigma_{BL} = \Sigma + \left(I - (\tau \Sigma) P^T \left( P (\tau \Sigma) P^T + \tau \frac{\Omega}{\tau} \right)^{-1} P \right) (\tau \Sigma) \\    = \Sigma + \left(I - (\tau \Sigma) P^T \tau^{-1} \left(P \Sigma P^T + \frac{\Omega}{\tau} \right)^{-1} P \right) (\tau \Sigma) \\    = \Sigma + \left(I - \Sigma P^T (P \Sigma P^T + \Omega')^{-1} P \right) (\tau \Sigma).

This shows that only some of the factors of \tau in the expression for \Sigma_{BL} are eliminated by choosing to estimate \Omega' rather than \Omega.  The parameter \tau remains in the formula for the posterior covariance matrix \Sigma_{BL}.


References

  1. Thomas Idzorek, “A Step-by-Step Guide to the Black-Litterman Model: Incorporating User-Specified Confidence Levels,” Working paper, Zephyr Associates, 2004.
  2. Attilio Meucci, “The Black-Litterman Approach: Original Model and Extensions,” in The Encyclopedia of Quantitative Finance, Wiley, 2010.
  3. Stephen Satchell and Alan Scowcroft, “A Demystification of the Black-Litterman Model: Managing Quantitative and Traditional Portfolio Construction,” Journal of Asset Management, September 2000, pp. 138–150.
Advertisements
This entry was posted in finance. Bookmark the permalink.

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out / Change )

Twitter picture

You are commenting using your Twitter account. Log Out / Change )

Facebook photo

You are commenting using your Facebook account. Log Out / Change )

Google+ photo

You are commenting using your Google+ account. Log Out / Change )

Connecting to %s