Single Index Model
Overview
The single index model is a model for calculating asset covariances
that relies on each asset being driven by 1 systemic risk factor, usually
taken to be the market return, and a set of independent idiosyncratic
factors. It is closely related to the
Capital Asset Pricing Model.
An extension of the single index model that uses multiple indices are
the
multi index models
Single Index Model
The single index model formulates an assets returns as the sum of exposure to the market risk factor and
an idiosyncratic risk factor.
{% r_i = \alpha_i + \beta_i \times r_m + e_i %}
{% r_i %} is the rate of return of a given asset and {% r_m %} is the return on the market.
Typically, these returns are measured as excess returns over the risk
free rate of return. This is due to considerations that come from
the
capital asset pricing model
{% \mathbb{E} [e_i] = 0 %}
Assumptions
The key assumption that makes the single index model easy to work with
is the assumptions that the idiosyncratic asset returns are uncorrelated
with the market, and with each other.
{% cov(e_i, r_m) = 0 %}
{% cov(e_i, e_j) = 0 %}
Then we have the following:
{% \mathbb{E} [r_i] = \alpha_i + \beta_i \times \mathbb{E} [r_m] %}
{% \sigma_i^2 =\beta_i^2 \sigma_m^2 + \sigma _{ei}^2 %}
Porfolio Beta
Portfolio beta is the weighted sum of the individual asset betas.
{% \beta_{port} =\sum w_i \beta_i %}
Total portfolio risk is then
{% \sigma_{port}^2 =\beta_{port}^2 \sigma_m^2 + \sum w_i^2\sigma_{ei}^2 %}
Estimating Beta
The single index equation
{% r_i = \alpha_i + \beta_i \times r_m + e_i %}
is just a simple linear equation. It can be estimated by
a standard
OLS regression.
Managing to a Benchmark
Using a single index factor model is an effective tool when managing
against a benchmark. When trying to perform against a benchmark,
the typical strategy is to measure your portfolio's beta
(sensitivity) to that benchmark, and then to target a beta
of 1.
Regression of Beta to 1
Using historical data to obtain an estimate of beta, as above,
does have some limitations. In particular, it is known that
there is a certain variability in measured betas that creates
a jumping around effect. In general, betas that are measured
as greater than 1 tend to jump back towards 1, and betas
less than 1 tend to jump back up, in effect regressing toward
the mean. Many authors have suggested corrections to measured
beta to compensate for this effect.
Blume Technique
Blume suggests the following correction to measured beta:
{% \beta = 0.343 + 0.677 \times \beta_{measured} %}