In
mathematical finance
Mathematical finance, also known as quantitative finance and financial mathematics, is a field of applied mathematics, concerned with mathematical modeling in the financial field.
In general, there exist two separate branches of finance that req ...
, multiple factor models are
asset pricing
In financial economics, asset pricing refers to a formal treatment and development of two interrelated Price, pricing principles, outlined below, together with the resultant models. There have been many models developed for different situations, ...
models that can be used to estimate the
discount rate for the valuation of financial assets; they may in turn be used to
manage portfolio risk.
They are generally extensions of the single-factor
capital asset pricing model
In finance, the capital asset pricing model (CAPM) is a model used to determine a theoretically appropriate required rate of return of an asset, to make decisions about adding assets to a Diversification (finance), well-diversified Portfolio (f ...
(CAPM).
Model of Rosenberg and Marathe
The multifactor equity risk model was first developed by Barr Rosenberg and Vinay Marathe. Initially they proposed a linear model of beta
where
is the return to equity asset
in the period
Modifications by Torre
Nicolo G. Torre made a number of improvements to this framework which importantly sharpened the risk control achievable by these means. In Rosenberg's model the risk indices X consisted of industry weights and risk indices. Each asset would be given an exposure to one or more industries, e. g. based on breakdowns of the firms balance sheet or earning statement into industry segments. These industry exposures would sum to 1 for each asset. Thus the model had no explicit market factor but rather the market return was projected on to the industry returns. Torre modified this scheme by introducing an
explicit market factor (with unit exposure for each asset.) To keep the model identified by imposed the condition that the industry factor returns sum to zero in each time period. Thus the model is estimated as
f(i,t)=m(t)+\sum_j X(i,j,t)f(j,t) + e(i,t)
subject to
\sum_k f(k,t)=0 for all
t
where the sum is over industry factors. Here m(t) is the market return. Explicitly identifying the market factor then permitted Torre to estimate the variance of this factor using a leveraged GARCH(1,1) model due to Robert Engle and Tim Bollerslev
s^2(t)=w+a s^2(t-1)+ b1 fp(m(t-1))^2 + b2 fm(m(t-1))^2
Here
fp(x)=x for x>0
0 for x<=0
fm(x)=0 for x>=0
x for x<0
and w, a, b1 and b2 are parameters fit from long time series estimations using maximum likelihood methods. This model provides a rapid update of market variance which is incorporated into the update of F, resulting in a more dynamic model of risk. In particular it accounts for the convergence of asset returns and consequent loss of diversification that occurs in portfolios during periods of market turbulence.
In the risk model industry factors carry about half the explanatory power after the market effect is accounted for. However, Rosenberg had left unsolved how the industry groupings should be defined – choosing to rely simply on a conventional set of industries. Defining industry sets is a problem in taxonomy. The basic difficulty is that the industry is defined by the members assigned to it, but which industry an individual equity should be assigned to is often unclear. Difficulties can be reduced by introducing a large number of narrowly defined industries, but this approach is in tension with the demands of risk estimation. For robust risk estimates we favor a moderate number of industries with each industry representing a few percentage points of market capitalization and not exclusively dominated by the largest company in the industry. Torre resolved this problem by introducing several hundred narrowly defined mini-industries and then applying guided clustering techniques to combine the mini-industries into industry groupings suitable for risk estimation.
In the initial Rosenberg approach factor and specific returns are assumed to be normally distributed. However experience turns up a number of outlying observations that are both too large and too frequent to be fit by a normal distribution Although introduction of a GARCH market factor partly reduces this difficulty, it does not eliminate it. Torre showed that return distributions can be modeled as a mixture of a normal distribution and a jump distribution. In the case of a single factor the mixing model is easily stated. Each time period t there is a binary mixing variable b(t). If b(t)=0 then the factor return in that period is drawn from the normal distribution and if b(t)=1 it drawn from the jump distribution. Torre found that simultaneous jumps occur in factors. Accordingly, in the multivariate case it is necessary to introduce a multivariate shock vector w(i,t) where w(i,t)=0 if the multivariate mixing variable b(i,t)=0 and w(i,t) is drawn from the ith jump distribution if b(i,t)=1. A transmission matrix T then maps w from the shock space into the factor space. Torre found that the market, factor and specific
returns could all be described by a mixture of normal returns and power law distributed shocks occurring at a low frequency. This modeling refinement substantially improves the performance of the model with regard to extreme events. As such it makes possible construction of portfolios which behave in more expected manners during periods of market turbulence.
Extensions to other market types
Although originally developed for the US equity market, the multifactor risk model was rapidly extended to other equity markets and to other types of securities such as bonds and equity options. The problem of how to construct a multi-asset class risk model then arises. A first approach was made by Beckers, Rudd and Stefek for the global equity market. They estimated a model involving currency, country, global industries and global risk indices. This model worked well for portfolios constructed by the top down process of first selecting countries and then selecting assets within countries. It was less successful on portfolios constructed by a bottom up process in which portfolios within countries were first selected by country specialists and then a global overlay was applied. In addition the global model applied to a single country portfolio would often be at odds with the local market model. Torre resolved these difficulties by introducing a two-stage factor analysis. The first stage consists of fitting a series of local factor models of the familiar form resulting in a set of factor returns f(i,j,t) where f(i,j,t) is the return to factor i in the jth local model at t. The factor returns are then fit to a second stage model of the
form
f(i,j,t)=\sum_k Y(i,j,k) g(k,t) + h(i,j,t)
Here Y gives the exposure of local factor (i,j) to the
k^th global factor whose return is g(k,t) and h(i,j,t) is the local specific factor return. The covariance matrix of factor returns is estimated as
\mathbf
where G is the covariance matrix of global factors and H is the block diagonal covariances of local specific factor returns. This modeling approach permits gluing any number of local models together to provide a comprehensive multi-asset class analysis. This is particularly relevant for global equity portfolios and for enterprise wide risk management.
The multifactor risk model with the refinements discussed above is the dominant method for controlling risk in professionally managed portfolios. It is estimated that more than half of world capital is managed using such models.
Academic models
Many academics have attempted to construct factor models with a fairly small number of parameters. These include:
*
Fama–French three-factor model
In asset pricing and portfolio management, the Fama–French three-factor model is a statistical model designed in 1992 by Eugene Fama and Kenneth French to describe stock returns. Fama and French were colleagues at the University of Chicago Boo ...
*
Carhart four-factor model
In Investment management, portfolio management, the Carhart four-factor model is an extra factor addition in the Fama–French three-factor model, proposed by Mark Carhart. The Fama-French model, developed in the 1990, argued most stock market re ...
*
Arbitrage pricing theory
In finance, arbitrage pricing theory (APT) is a multi-factor model for asset pricing which relates various macro-economic (systematic) risk variables to the pricing of financial assets. Proposed by economist Stephen Ross (economist), Stephen Ross i ...
However, there is as yet no general agreement on how many factors there are. There are numerous commercial models available, including those from
MSCI
MSCI Inc. (formerly Morgan Stanley Capital International) is an American finance company headquartered in New York City. MSCI is a global provider of equity, fixed income, real estate indices, multi-asset portfolio analysis tools, ESG and ...
and the
Goldman Sachs asset management factor model.
References
{{reflist
Financial risk modeling
Financial markets
Financial models