2

Estimating the Arbitrage Pricing

Theory Factor Sensitivities Using
Quantile Regression
Zeno Adams, Roland Füss, Philipp Grüber,
Ulrich Hommel, and Holger Wohlenberg

2.1 Introduction

One of the main insights from over 50 years of portfolio theory is the
fact that investors should not hold single securities but should invest
in large portfolios. The idiosyncratic risks that affect asset returns on an
individual level cancel out so that only systematic risks affecting all assets
in the economy have to be considered. The capital asset pricing model
(CAPM) (Sharpe 1964; Lintner 1965; Black 1972) laid the cornerstone for
the theory of asset pricing which has been replaced in the following years
by the Fama–French model (Fama and French 1993) and the arbitrage
pricing theory (APT) starting with Ross (1976).1
Although the investor does not know future return realizations of his
portfolio, the APT is a useful instrument for identifying common factors
influencing portfolio returns. By estimating the sensitivities of a port-
folio with respect to these common factors, the investor can reposition
his portfolio in a way that leads to a more balanced risk exposure. For
instance, a company can reduce its risk exposure to a factor that has a
dominant effect on its overall costs of capital and, at the same time, can
accept a higher exposure to a factor that has a negligible effect on the
required investment return. By bearing a higher risk to the latter factor,
the company can expect overall a higher stock return. Studies that have
tested the APT empirically (Roll and Ross 1980; Reinganum 1981; and
Chen 1983 among others) did not, however, always find positive risk
premiums.
In this study, we use the quantile regression technique to show that
risk premiums may not only be negative but may also have different signs

18
G. N. Gregoriou et al. (eds.), Nonlinear Financial Econometrics:
Forecasting Models, Computational and Bayesian Models
© Palgrave Macmillan, a division of Macmillan Publishers Limited 2011
 Estimating the APT Factor Sensitivities 19

depending on the quantile of the return distribution, that is, they depend
on the asset return relative to the return of the overall market. When
bearing additional risk to a common factor, investors should therefore
not only concentrate on the shifting of their risk exposure but should
also draw their attention to the selection of assets that (1) yield a positive
risk premium on average and (2) still generate a positive risk premium
when underperforming the market to ensure partial compensation for
investors. In addition, we can also show that replacing OLS by the quan-
tile regression approach leads to a much better model fit of the APT
for the lower-end quantiles. The usage of this estimation technique is
therefore particularly advisable for periods of high volatility such as the
2007/2008 financial crisis.
The remainder of this study is organized as follows: Section 2.2 gives
a brief overview of APT and its estimation process. This section will also
review the quantile regression methodology and its application to APT. In
Section 2.3, we present and discuss the results of an exemplary empirical
analysis using this approach. The chapter concludes with final remarks
in Section 2.4.

2.2 The Arbitrage Pricing Theory and its Estimation

The APT is based on the insight that long-run asset returns are system-
atically influenced by unexpected shifts of certain factors such as the
rate of inflation, the level of industrial production, or the slope of the
yield curve. By estimating the sensitivities of a given portfolio to these
factors, the investor can assess his or her current risk exposure. The APT
can then be used to construct a portfolio with a desired exposure to these
common factors by selecting assets with the appropriate risk exposure. A
developer of business software will, for instance, have a relatively strong
exposure to unexpected changes in economic activity in general and thus
in industrial production but less exposure to inflation risk. By investing
in companies that show a higher exposure to inflation risk and a lower
exposure to the productivity factor, the software company can achieve
the well-known diversification effect similar to the case of the CAPM.
Unlike the single-factor CAPM, however, four to five risk factors have
been identified for the APT,

CAPM Ri = rf + βCAPM Rm − rf + ei (2.1)

   K
APT Ri = E R i + βk fk + ei .2 (2.2)
k=1