Introduction
The Black-Litterman Model is an asset allocation model. It has three main features, the use of an
informative prior derived from the ICAPM equilibrium, a mixing model that allows the investor
to specify views on any linear combination of the assets, and a portfolio choice feature to identity
the optimal portfolio for the investor based on their views.
The equilibrium prior estimate is generated using a reverse optimization procedure from the
market portfolio using the Markowitz Mean Variance model. The investor knows the covariance
matrix which all investors are using, the weights of the market portfolio and the risk aversion of
the aggregate market portfolio. The investor is uncertain in this estimate and it is thus specified
as a distribution.
The mixed-estimation model was originally developed by Theil (1971). It is equivalent to the
Bayesian problem of an unknown mean and known variance, DeGroot (1970). The mixing model
allows absolute and relative views, and views may be on any combination of the assets. Just as
the investor is uncertain in their estimate of the prior, they are also uncertain in their estimates of
the returns to the views.
This paper will focus on the role of the factor tau (τ). τ is used to scale the investors uncertainty
in their prior estimate of the returns. There are several different approaches to calibrating it, or
even including it described in the literature. Just to illustrate the difference of opinion, we will
look at comments from three authors. He and Litterman (1999) state they set τ = 0.05. Satchell
and Scowcroft (2000) state many people use a value of τ around 1. Meucci (2010) proposes a
formulation of the Black-Litterman model without τ.
We will use the concept of Reference Models to explain the differences between the various
authors. We will start by presenting the Canonical Reference Model as derived from Theil's
mixed estimation approach in Black and Litterman (1991), and further explained in He and
Litterman (1999). The main difference between the canonical reference model and other
reference models is uncertainty. Next we will present the Alternative Reference Model as
proposed by Meucci (2010) which estimates the returns without τ. A third model, the Hybrid
Reference Model was shown in Satchell and Scowcroft (2000), but we will show that any
expression of the hybrid model can be duplicated in the Alternative Reference Model, so we do
not need to discuss the Hybrid Reference Model further. Finally, we will provide
recommendations on how an investor can to calibrate τ.