term. We do however generate the prior covariance matrix using a sample of returns, and we
have a consistent n to use with our Σ which we could use to estimate a standard error for the
estimate of the mean.
Formula (8) shows the basic relationship between τ and n from our covariance matrix
calculations. Because we are using non-statistical methods to estimate the mean return, this is
not a quantitative answer as to what value we should use for τ, it is just one way to provide some
intuition around the scale of τ.
If we were to perform a statistical bootstrap procedure, for example re-sampling with
replacement, to determine the standard error, the central limit theory assures us that the result
from formula (17) is in fact the value we would compute.
A second approach to establishing a reasonable value for τ is to use confidence intervals. This
has a more direct connection with the model and our estimate of the model. Formula (4)
illustrates the distribution of the estimate of the mean, about the mean return. From this
distribution we can assert a confidence interval for our estimate using basic probability.
A plausible scenario we might encounter would be yearly equity like returns with μ = 8% and σ
= 15%. Table 3 below shows the 95% and 99% confidence intervals for this scenario and various
values of τ.
Table 3