Fieberg Et Al 17 An Investor's Perspective On Risk-Models and Characteristic-Models

The Journal of Risk Finance
An investor’s perspective on risk-models and characteristic-models

Christian Fieberg, Thorsten Poddig, Armin Varmaz,
Article information:
To cite this document:
Downloaded by STAATS UND UNIVERSITAETSBIBLIOTHEK BREMEN At 12:56 09 May 2017 (PT)
Christian Fieberg, Thorsten Poddig, Armin Varmaz, (2016) "An investor’s perspective on risk-models
and characteristic-models", The Journal of Risk Finance, Vol. 17 Issue: 3, pp.262-276, doi: 10.1108/
JRF-02-2016-0026
Permanent link to this document:
http://dx.doi.org/10.1108/JRF-02-2016-0026
Downloaded on: 09 May 2017, At: 12:56 (PT)
References: this document contains references to 32 other documents.
To copy this document: permissions@emeraldinsight.com
The fulltext of this document has been downloaded 345 times since 2016*
Users who downloaded this article also downloaded:
(2016),"On portfolio optimization: Forecasting asset covariances and variances based on multi-
scale risk models", The Journal of Risk Finance, Vol. 17 Iss 3 pp. 295-309 http://dx.doi.org/10.1108/
JRF-09-2015-0094
(2016),"What transaction costs are acceptable in life insurance products from the policyholders’
viewpoint?", The Journal of Risk Finance, Vol. 17 Iss 3 pp. 277-294 http://dx.doi.org/10.1108/
JRF-02-2016-0017
Access to this document was granted through an Emerald subscription provided by emerald-
srm:514734 []
For Authors
If you would like to write for this, or any other Emerald publication, then please use our Emerald
for Authors service information about how to choose which publication to write for and submission
guidelines are available for all. Please visit www.emeraldinsight.com/authors for more information.
About Emerald www.emeraldinsight.com
Emerald is a global publisher linking research and practice to the benefit of society. The company
manages a portfolio of more than 290 journals and over 2,350 books and book series volumes, as
well as providing an extensive range of online products and additional customer resources and
services.
Emerald is both COUNTER 4 and TRANSFER compliant. The organization is a partner of the
Committee on Publication Ethics (COPE) and also works with Portico and the LOCKSS initiative for
digital archive preservation.
*Related content and download information correct at time of download.

The current issue and full text archive of this journal is available on Emerald Insight at:
www.emeraldinsight.com/1526-5943.htm
JRF
17,3
An investor’s perspective on
risk-models and
characteristic-models
262 Christian Fieberg and Thorsten Poddig
Department of Finance, University of Bremen, Bremen, Germany, and
Received 29 February 2016

Revised 21 March 2016 Armin Varmaz
Accepted 27 March 2016
School of International Business, Bremen, Germany
Abstract
Purpose – In capital markets, research risk factor loadings and characteristics are considered as
opposing explanations for the cross-sectional dispersion in average stock returns. However, there
is little known about the performance an investor would obtain who believes either in the characteristics
explanation (CB-investor) or in the risk factor loadings explanation (RB-investor). The purpose of this
paper is to compare the performance of CB- and RB-investors.
Design/methodology/approach – To compare the competing strategies, the authors propose a
simple new approach to equity portfolio optimization in the style of Brandt et al. (2009) by modeling
the portfolio weight in each asset as a function of the asset’s risk factor loadings or characteristics.
The authors perform an empirical analysis on the German stock market, exploiting the risk factor
loadings from the Carhart (1997) four-factor model and the respective characteristics size,
book-to-market equity ratio and momentum.
Findings – The results show that investment strategies relying on characteristics (particularly on
momentum) outperform risk-based investment strategies in horse races. These findings hold in- and
out-of-sample. Furthermore, the characteristics-based investment strategies outperform a value-
weighted market portfolio strategy in- and out-of-sample.
Originality/value – The authors introduce a portfolio optimization approach that enables investors
to directly link portfolio decisions to the firm’s characteristics or risk factor loadings.
Keywords Asset allocation, Characteristic model, German stock market, Risk model
Paper type Research paper
1. Introduction
Traditional finance theory suggests that assets which have riskier payoffs should earn
higher returns on average to compensate investors for carrying that higher (systematic)
risk. Accordingly, investors should hold a combination of risk factor portfolios.
However, instead of being related to systematic risk as the capital asset pricing model
(CAPM) beta (Fama and French (1992)), stock returns seem to be related to firm-level
characteristics like market capitalization (Banz, 1981), the book-to-market equity ratio
(Stattman, 1980 and Rosenberg et al., 1985) and momentum (Jegadeesh and Titman,
1993). These findings caused an intense debate on whether it is a certain characteristic
The Journal of Risk Finance
Vol. 17 No. 3, 2016
pp. 262-276 JEL classification – G11, G12
© Emerald Group Publishing Limited
1526-5943
This research was conducted while Christian Fieberg was lecturer at the FOM Hochschule. The
DOI 10.1108/JRF-02-2016-0026 authors wish to thank anonymous referees for helpful comments.
(e.g. book-to-market equity ratio) that explains the differences in average stock returns Investor’s
or the loading on a factor-mimicking portfolio (e.g. high minus low (HML)) which is perspective on
constructed on that characteristic[1]. The former explanation is also referred to the
terms mispricing or characteristic model/story, while the latter one is also referred to the
risk-models
terms risk or covariance model/story. Since the early works of Daniel and Titman (1997)
and Davis et al. (2000), the question whether covariances or characteristics determine the
cross-sectional differences in average stock returns is still highly debated in capital 263
markets research.
The covariances vs characteristics debate is not just any debate, it is at the core
of modern asset pricing (Lin and Zhang, 2013) and has severe implications on the
estimation of expected stock returns, performance evaluation and portfolio
formation as discussed in Daniel and Titman (1998) and Daniel et al. (1997). Our
study adds to a number of studies that analyze the practical implications of the
covariances vs characteristics debate for portfolio formation (see Haugen and Baker
(1996), Pástor and Stambaugh (2000), Fletcher (2002) and Chou et al. (2006), among
others). However, the approaches applied in these studies are neither able to
parameterize the weight of a stock as a function of the stocks’ characteristics or
factor loadings nor are the relations between expected returns, variances,
covariances and even higher-order moments on the one hand and characteristics or
factor loadings on the other hand simultaneously captured. We tackle these issues in
applying an extension of the parametric portfolio policy proposed by Brandt et al.
(2009) that directly models the stocks’ portfolio weights as a linear function of firm
characteristics. We extend their approach in parameterizing the portfolio weight of
each stock as a function of the stocks’ factor loadings or characteristics. The
coefficients of the portfolio policy are then estimated by maximizing the expected
utility the investor would have realized by implementing the policy over the
historical sample period. This approach enables investors to directly link portfolio
decisions to the firm’s characteristics or risk factor loadings.
We then optimize portfolios of all German stocks from 1980 to 2014 using the factor
loadings from the Carhart (1997) four-factor model (FFC-model) and the respective
characteristics size, the book-to-market equity ratio and momentum. Our results show
that investment strategies relying on characteristics outperform factor loadings-based
investment strategies. These findings hold in- and out-of-sample. Furthermore, the
characteristics-based investment strategies outperform a value-weighted market
portfolio strategy in- and out-of-sample.
The remainder of the paper proceeds as follows. We review the literature in Section 2
and explain the portfolio optimization approach in Section 3. The asset allocation
strategies, data and results of the empirical application are described in Section 4.
Section 5 concludes.
2. Literature review
The question whether covariances or characteristics determine the cross-sectional
differences in average stock returns is highly debated in capital markets research.
For the US stock market, Daniel and Titman (1997) show that the return premia on
small capitalization and high book-to-market equity ratio stocks is not due to the
co-movements of these stocks with small minus big (SMB) and HML. Instead, it is
the size and the book-to-market equity ratio and not the covariance structure of
JRF stock returns that explain the dispersion in average stock returns. Davis et al. (2000)
17,3 argue that the evidence of Daniel and Titman (1997) in favor of the characteristics
model is special to their rather short sample period. Based on an enlarged sample
period, Davis et al. (2000) find that the risk model provides a better explanation for
the relation between the book-to-market equity ratio and average returns. As
Japanese stock returns are also strongly related to their book-to-market equity
264 ratios, Daniel et al. (2001) replicate the Daniel and Titman (1997) test on a Japanese
stock market sample and find evidence in favor of the characteristics story.
Lajili-Jarjir (2007) aims to provide further empirical evidence by analyzing whether

it is the firm’s factor loadings on SMB and HML or the market capitalization and the
book-to-market equity ratio that determine the cross-sectional variation in the
French stock market returns. They argue that the covariance structure of returns
rather than the characteristics appears to explain the cross-sectional variation in
stock returns. More recently, Fieberg et al. (2016) show that the cross-section of
German stock market returns is better explained by the characteristic model than by
a covariance model. International evidence is provided by Hou et al. (2011). They
examine whether stock returns from 49 countries are due to co-movements of stocks
with portfolios constructed on size, dividends, earnings yields, cash-flow-to-price,
book-to-market equity, leverage and momentum. With the exception of the
cash-flow-to-price ratio, they provide evidence in favor of the characteristic model.
Further evidence in favor of the characteristic model is provided among others by
Mohanram and Rajgopal (2009) (information risk), Hirshleifer et al. (2012) (accruals),
Grundy and Martin (2001) (momentum), Li et al. (2013) (idiosyncratic risk) and Ali
et al. (2003) (residual income). Lin and Zhang (2013) conclude that there is evidence
that the characteristic model dominates the covariance model in horse races.
However, they do also argue that these findings might be preliminary due to
possible measurement errors in the estimation of the firm’s factor loadings.
Although there is still no consensus in the covariances vs characteristics debate,
it has serious implications on how investors should form portfolios (Daniel and
Titman, 1998; Daniel et al., 1997, among others). Haugen and Baker (1996), Pástor
and Stambaugh (2000), Fletcher (2002) and Chou et al. (2006), among others, analyze
the practical implications of this controversy for investors. Haugen and Baker (1996)
provide empirical evidence that variables like risk, liquidity, price-level, growth
potential and stock price history can be successfully exploited by investors in
their portfolio decisions. Pástor and Stambaugh (2000) investigate the portfolio
allocations of mean-variance-investors who use sample evidence to update prior
beliefs based on either Fama and French (1993) risk-based or Daniel and Titman
(1997) characteristic-based pricing models. Specifically, they form portfolios by
sorting stocks on size, the book-to-market equity ratio and factor loadings with
respect to the HML factor of the Fama and French (1993) three-factor model. It is
shown that a modest investor uncertainty about the models’ pricing abilities
eliminates the performance differences between those two approaches. Fletcher
(2002) conducts a mean-variance analysis to compare the performance of risk-based
(Fama and French, 1993) and characteristic-based (Daniel and Titman, 1997)
investment strategies. Like Haugen and Baker (1996), he uses both models to
forecast expected returns which are then used in mean-variance optimization. He
finds that the characteristic model exhibits a better performance than the strategies
that use risk factor models. Other than Haugen and Baker (1996) and Fletcher (2002), Investor’s
Chou et al. (2006) do not assume that expected stock returns are a linear function of perspective on
the characteristics. Instead, they set the expected return of a stock equal to the
average return of a portfolio exhibiting characteristics similar to the ones of the
risk-models
stock. They find that the performance of the characteristics explanation
(CB-investor) dominates the performance of a risk factor loadings explanation
(RB-investor)on Japanese market. 265
The above-discussed approaches are not able to directly parameterize the
portfolio weight of a stock as a function of the stocks’ characteristics or risk factor

loadings. Furthermore, the findings are largely based on expected return
comparisons (Chou et al., 2006). Little is known on how the covariance model and the
characteristic model matter to an investor’s portfolio decision. We tackle this issue
in applying an expected utility approach that is able to implicitly capture the
relations between expected returns, variances, covariances and even higher-order
moments on the one hand and characteristics or risk factor loadings on the other
hand. This approach is described in the subsequent section.
3. Methodology
3.1 Basic approach
Our portfolio optimization approach as described in the following is an extension of the
parametric portfolio policy proposed by Brandt et al. (2009). We extend their approach in
that we parameterize the portfolio weight of each stock not only as a function of the
stocks’ characteristics but also as a function of the stocks’ risk factor loadings or
characteristics. This enables investors to directly link their beliefs in opposing asset
pricing explanations (risk or mispricing) to their portfolio decisions. Assume that at
each date t, an investor has to choose the portfolio weights wi,t for a large number of
stocks Nt to maximize the expected utility of the portfolio’s return rp,t⫹1:
关共兺兲兴
Nt
max Et关 u( rp,t⫹1 ) 兴 ⫽ Et u wi,t ri,t⫹1 . (1)

Nt
兵wi,t其 i⫽1
i⫽1
The optimal portfolio weights are parameterized as a function of the stocks’

characteristics xi,t or factor exposures bi,t:
ˉ i,t ⫹ 1 ( lChar␪ T x̂i,t ⫹ lCov␭ T b̂i,t ),

wi,t ⫽ w (2)
Nt
where w ˉ i,t is the weight of stock i at date t in a benchmark portfolio, x̂i,t represents the
characteristics of stock i at date t, ␪ is a vector of characteristic coefficients to be
estimated, b̂i,t represents the factor loadings of stock i at date t and ␭ is a vector of factor
loading coefficients to be estimated. lChar (lCov) equals one if the characteristic (risk
factor loading) model has to be applied and does therefore allow the investor to make
risk- or characteristic-based portfolio decisions. This approach follows the idea of active
portfolio management relative to a benchmark. The characteristics and factor loadings
are standardized cross-sectionally. This is done to ensure stationarity of characteristics
and factor loadings through time. Moreover, standardizing characteristics and factor
JRF loadings does ensure that the deviations of the optimal portfolio weights from the
17,3 benchmark portfolio weights sum to zero (ensuring that the optimal portfolio weights
sum to one). A normalization (1/Nt) allows the consideration of an arbitrary and
time-varying number of stocks. The coefficient vectors ␪ and ␭ are assumed to be
constant across assets and through time meaning that the coefficients that maximize the
investor’s conditional expected utility at a given date are the same for all dates. This
266 implies that the investor’s unconditional expected utility is maximized, too. Our
approach is insofar an extension of the parametric portfolio policy proposed by Brandt
et al. (2009) in that we parameterize the portfolio weight of each stock not only as a
function of the stocks’ characteristics but also as a function of the stocks’ factor loadings
or characteristics. This approach enables investors to directly link opposing asset
pricing explanations for the cross-sectional differences in average stock returns (risk or
mispricing) to portfolio decisions.
The parametric portfolio policy described above realizes several benefits (Brandt
et al., 2009; Hand and Green, 2011 and Hjalmarsson and Manchev, 2012). As traditional
mean-variance portfolio optimization would require modeling the expected returns,
variances and covariances of all stocks as functions of their factor loadings or
characteristics, the main benefit of this approach is that it concentrates directly on the
portfolio weights and not on the joint distribution of returns and risk factor loadings or
characteristics. The approach is also able to capture the relations between the return
distribution moments and firm-specific characteristics or risk factor loadings because
the expected returns, variances, covariances and even higher-order moments impact the
distribution of the optimized portfolio’s returns.
3.2 Refinements and extensions

The basic parametric portfolio policy approach as described above can be refined and
extended in various ways, such as constraining portfolio weights, applying a specific
objective function, considering nonlinear transformations and interactions of the
characteristics or factor exposures, estimating time-varying coefficients, applying
shrinkage estimation, obtaining standard errors for the coefficients of the weight
function and considering transaction costs. These refinements and extensions to the
basic approach are extensively explained in Brandt et al. (2009) and therefore not
covered here entirely. We focus on the refinements and extensions which are applied in
the empirical application below. First, we assume standard constant relative risk
aversion (CRRA) preferences over wealth in our empirical application:
(1 ⫹ rp,t⫹1 )1⫺␥
u(rp,t⫹1 ) ⫽ , ␥ ⫽ 1. (3)
1⫺␥
Like Brandt et al. (2009), we use a value of five for the relative risk aversion ␥. Second, to
account for transaction costs, we calculate the return to the portfolio net of trading costs
as:
Nt
rp,t⫹1 ⫽ 兺关w r
i⫽1
i i,t⫹1 ⫺ ci,t ⱍ wi,t ⫺ wi,t⫺1 ⱍ , (4)
where ci,t is the proportional transaction cost for stock i at time t. Third, we estimate Investor’s
standard errors for the coefficient vectors ␪ and ␭ by bootstrapping. We generate a large perspective on
number of samples of returns, characteristics, factor loadings and benchmark weights
by randomly drawing monthly observations from the original data sample. The
risk-models
coefficients of the optimal portfolio policy are estimated for each bootstrapped sample to
estimate the bootstrapped standard errors.
4. Empirical application
267
4.1 Asset allocation strategies

In this section, we use our approach to analyze how a German stock market investor
would have performed following a risk factor loading- or characteristics-based asset
allocation strategy (AAS). As discussed in Section 2, there is an intense debate in the
literature on whether characteristics like market capitalization, book-to-market
equity ratio or momentum explain the differences in average stock returns or the
loadings on factor mimicking portfolios like SMB, HML or winner minus loser
(WML). SMB, HML and WML are factor-mimicking portfolios from the Carhart
(1997) FFC-model associated with the size effect, the book-to-market equity ratio
effect, and the momentum effect, respectively. According to the FFC-model, the
expected returns follow a FFC-model:
E(Ri ) ⫺ Rf ⫽ bi[E(RM ) ⫺ Rf ] ⫹ siE(SMB) ⫹ hiE(HML) ⫹ wiE(WML). (5)
We consider two asset allocation strategies to demonstrate how an investor can

make risk factor loadings- or characteristics-based investment decisions. For both
asset allocation strategies, the value-weighted market portfolio acts as a benchmark
portfolio. The empirical results of our analysis do therefore document the relevance
of the firm’s factor loadings or characteristics for explaining divergences of the
optimal portfolio weights from the market capitalization weights. The first asset
allocation strategy (AAS 1, RB-investor) relies on the firm’s factor loadings on SMB,
HML and WML from the FFC-model. It might therefore be regarded as a Davis et al.
(2000)- or Carhart (1997)-like strategy. The second asset allocation strategy (AAS 2,
CB-investor) is its characteristics-based counterpart. The characteristics under
consideration are market capitalization, the book-to-market equity ratio and
momentum. AAS 2 might be regarded as a Daniel and Titman (1997)-like strategy.
Although many more risk factor loadings- and/or characteristics-based asset
allocation strategies might be conceivable we restrict our analysis to the ones above
because the source of the size, book-to-market equity ratio and momentum effect is
the most highly debated in the asset pricing literature.
4.2 Data
We acquired data from Thomson Reuters Datastream for the period from January 1980
to December 2014. The firms from research lists (FGER1, FGER2, FGERDOM,
FGKURS), dead lists (DEADBD1 to DEADBD6) and the Thomson Reuters Worldscope
list (WSCOPEBD) are included in our sample obtaining 2,359 common shares for the
German stock market[2]. Financial firms are excluded (Viale et al., 2009). We apply the
one digit SIC-code “6” to identify financial firms.
The monthly stock returns are calculated from the total return index which accounts
for dividend payments, splits and equity offerings. The formation of factor-mimicking
JRF portfolios is based on firm characteristics size, book-to-market equation ratio (BE/ME)
17,3 and momentum. Size in month t is the market-value of equity at the end of month t.
BE/ME in June of the year t is the book-value of equity divided by the market-value of
equity, both as of the end of December of year y – 1. Firm-years exhibiting negative
book-values are excluded. A lag of six month is imposed for BE/ME to ensure that
the accounting data used to calculate these variables are known to the investors. We
268 take the value of BE/ME as of June in year t for the 12 months immediately following. In
line with Jegadeesh (1990), we calculate momentum in month t as the cumulative past
return from month t – 12 to month t – 2.

The construction of the risk factors SMB, HML and WML for the German stock
market follows closely the procedure as described in Fieberg et al. (2016). The market
return RM for each month is the return on the value-weighted portfolio of all sample
stocks in that month. The risk-free rate Rf is the average of the three-month FIBOR rates
(as provided by Bundesbank). The estimation of the loadings on the risk factors from the
FFC-model for each stock at the end of month t is conducted in rolling five-year
(minimum two years) time-series regressions based on monthly returns.
This regression procedure provides us with monthly factor loadings on SMB,
HML and WML from the FFC-model. We further use the monthly characteristics size
and momentum from the construction of the factor-mimicking portfolios in our asset
allocation study. As the BE/ME characteristic is only available yearly (in June of
year y to avoid look ahead bias), we do also use its values in June of year y from July
of year y to May of year y ⫹ 1. This provides us with monthly size (ME), momentum
(MOM) and BE/ME characteristic values. We exclude all months from our sample
for which information on one of the variables needed in our portfolio optimization
procedure (returns, benchmark portfolio weights, characteristics and factor
loadings) is missing for all firms.
Descriptive statistics for the factor loadings and characteristics applied in our
portfolio optimization study are shown in Table I. Because a minimum of 24 return
Descriptive statistics ME BE/ME MOM s h w
Mean 607.39 0.80 0.02 0.76 0.12 ⫺0.09

SD 366.13 0.40 0.32 0.57 0.51 0.36
Skewness 0.76 0.68 0.64 0.06 0.03 ⫺0.03
Kurtosis 6.15 3.23 3.72 3.67 3.56 3.80
25th Perc. 332.51 0.51 ⫺0.20 0.38 ⫺0.23 ⫺0.32
Median 539.91 0.71 ⫺0.02 0.75 0.12 ⫺0.09
75th Perc. 797.18 1.00 0.20 1.15 0.48 0.14
Table I.
Descriptive statistics Notes: The table reports the mean, standard deviation, skewness, kurtosis, median, 25 per cent-th and
for the German stock the 75 per cent-th percentile for the characteristics and covariances included in our study; the
market covariances characteristics are market capitalization (ME) (in millions) in month m of a year t, book-to-market equity
and characteristics, ratio (BE/ME) calculated as of December of year t – 1, and momentum in month m of a year t, calculated
June 1983 to as cumulative equal-weight past returns from month m –12 to m – 2. The covariances are firms monthly
December 2014 factor loadings on SMB, HML and WML from the Carhart (1997) four-factor model (s, h and w)
observations is required to conduct the time-series regression to estimate the firm’s Investor’s
factor loadings, we do not consider data before June 1983. In the subsequent section, perspective on
it is explained how the factor loadings and characteristics are used in the asset
allocation study.
risk-models
4.3 Results
For each month over the period from June 1983 to November 2014, we use the 269
characteristics or factor loadings of that month as described above, the market
capitalization weights of that month as benchmark weights and the next month returns
as optimization inputs. For the in-sample results, we use monthly returns between July
1983 and December 2014 and estimate one parameter set of ␪ or ␭ that relates the
investor’s optimal portfolio weights to firms’ factor loadings or characteristics. For the
out-of-sample results we use an expanding window approach. Specifically, to estimate
the first parameter set, we use 119 monthly returns between July 1983 and May 1993 and
use that parameter set to determine optimal portfolio weights based on the benchmark
weights and factor loadings or characteristics in May 1993. The optimal portfolio
weights are used to calculate the portfolio return in June 1993. The second parameter set
is estimated based on the 120 monthly returns between July 1983 and June 1993 and used
to determine optimal portfolio weights based on the benchmark weights and factor
loadings or characteristics in June 1993. The optimal portfolio weights are used to
calculate the portfolio return in July 1993. The estimation window is expanded
month-by-month so that the final set of parameters is determined on monthly returns
between July 1983 and November 2014.
Table II presents the results for AAS 1, in which the over- or under-weighting of
each stock, relative to the value-weighted market portfolio, depends on the firm’s
factor loadings on SMB, HML and WML from the FFC-model. The Columns 2-5
show the results for the base case (Scenario 1) in which short sales are allowed (ShSa:
Yes), and no transaction costs are considered (TrCo: No). As this scenario might be
regarded as the most favorable one from an investor’s perspective, we do also apply
a second scenario (Scenario 2) in which short sales are disallowed (ShSa: No), and
transaction costs are taken into consideration (TrCo: Yes). The results for Scenario
2 are presented in the Columns 6-9. To estimate the impact of transaction costs, we
assume a constant 0.1 per cent for all firms and over all months. The first two
columns for each scenario report the in-sample results of the benchmark strategy
(first column) and the AAS (second column), while the last two columns report the
out-of-sample results of these two strategies. The table is divided into five sections
describing separately the parameter estimates and their standard errors,
distribution of the portfolio weights, properties of the portfolio returns, average
factor loadings (or characteristics, respectively) of the portfolio and difference in the
certainty-equivalent returns of the optimal portfolio policy and the benchmark
portfolio with bootstrapped standard errors. This format is the same for the two
asset allocation strategies in this paper.
The signs of the factor loadings coefficients (␭) show that deviations of the
optimal portfolio weights from the benchmark portfolio weights decrease with the
firm’s loading on SMB and increase with the firm’s loadings on HML and WML. The
negative sign for the factor loading coefficient of the firm’s loading on SMB might be
surprising in the first place. But this finding can be explained by a negative average
JRF ShSa: Yes, TrCo: No ShSa: No, TrCo: Yes
17,3 In-sample Out-of-sample In-sample Out-of-sample
Statistics B P B P B P B P
␭s ⫺2.5430 ⫺1.1568 ⫺7.7340 ⫺82.2007

␭h 0.9728 0.3516 3.3152 11.3615
270 ␭w 3.4769 2.9152 5.8158 58.9772
se( ␭s ) 1.1396 1.5604 47.3708 43.4500
se( ␭h )
1.4950 1.9816 9.3661 30.7584

se( ␭w ) 1.5502 1.9796 31.8029 34.6033
Avg.ⱍwiⱍ 0.0034 0.0129 0.0022 0.0072 0.0034 0.0034 0.0022 0.0022
min wi 0.0000 ⫺0.0495 0.0000 ⫺0.0274 0.0000 0.0000 0.0000 0.0000
max wi 0.0997 0.1138 0.1013 0.1060 0.0997 0.0321 0.1013 0.0304
兺wi I(wi ⬍ 0) 0.0000 ⫺1.4011 0.0000 ⫺1.2040 0.0000 0.0000 0.0000 0.0000
兺I(wi ⱕ 0)/Nt 0.0000 0.4189 0.0000 0.4188 0.0000 0.0000 0.0000 0.0000
␮(r) 0.0087 0.0129 0.0093 0.0105 0.0086 0.0072 0.0093 0.0053
␴(r) 0.0503 0.0546 0.0491 0.0457 0.0503 0.0376 0.0491 0.0298
␮(r)/␴(r) 0.1722 0.2363 0.1906 0.2299 0.1715 0.1908 0.1899 0.1767
CER 0.0018 0.0053 0.0030 0.0053 0.0018 0.0033 0.0029 0.0030
s 0.0031 0.0082 ⫺0.7944 ⫺1.4178 0.0031 0.0018 ⫺0.7944 ⫺0.3351
h ⫺0.0047 ⫺0.0141 ⫺0.1011 ⫺0.3793 ⫺0.0047 ⫺0.0023 ⫺0.1011 ⫺0.0906
w 0.0082 0.0223 0.1063 2.6892 0.0082 0.0039 0.1063 0.6228
P-B P-B P-B P-B
CER 0.0035 0.0024 0.0016 0.0001
se(CER) 0.0022 0.0016 0.0016 0.0002
Notes: We use data from Thomson Reuters Datastream database from June 1983 to December 2014. In
the “out-of-sample” results, we use data until May 1993 to estimate the coefficients of the portfolio policy
and then form out-of-sample portfolios using those coefficients in the next month; every subsequent
month, we reestimate the portfolio policy by enlarging the sample; all statistics in columns labeled
“out-of-sample” are therefore reported for the period June 1993 to December 2014; all statistics in
columns labeled “in-sample” are reported for the period July 1983 to December 2014; the columns labeled
“B” and “P” display statistics of the market-capitalization-weighted portfolio and the optimal
parametric portfolio policy, respectively; the first set of rows shows the estimated coefficients of the
portfolio policy with bootstrapped standard errors; the “out-of-sample” results display time-series
averages of coefficients and standard errors; the “in-sample” results display coefficients and standard
errors for one set of parameters estimated based on monthly returns from July 1983 to December 2014;
the second set of rows shows statistics of the portfolio weights averaged across time; these statistics
include the average absolute weight, the average minimum and maximum weights, the average sum of
negative weights in the portfolio and the average fraction of negative weights in the portfolio; the third
set of rows displays average portfolio return statistics: average return, standard deviation, average
return to standard deviation ratio and certainty-equivalent return; the fourth set of rows displays the
Table II. average normalized covariances and/or characteristics of the portfolio; the final set of rows reports the
Linear portfolio difference in the certainty-equivalent returns of the optimal portfolio policy and the benchmark
policy: AAS 1 portfolio with bootstrapped standard errors.
return of the SMB factor. Empirical evidence for a negative average return of SMB
is also provided by Fieberg et al. (2016) for the German stock market. When no short
sale constraints and no transaction costs are considered, the optimal portfolio
exhibits a higher certainty-equivalent return compared to the benchmark portfolio
in- and out-of-sample. This is likely being explained by a higher average return Investor’s
in-sample and a lower standard deviation out-of-sample. perspective on
However, when short sales are disallowed and transaction costs are taken into
account, the optimal portfolio policy is barely able to offer a certainty-equivalent gain
risk-models
relative to the benchmark portfolio. Additionally, neither of the coefficients is
statistically significant out-of-sample due to surprisingly large standard errors. An
in-depth analysis reveals that the estimates of the coefficient vectors may fluctuate 271
considerably when the data sample suffers from a large amount of assets entering and
exiting the data sample (as it is the case for the German stock market) and weight
constraints exist. Therefore, tests of the statistical significance of these variables should
not be taken too seriously in such a case. The estimates might be highly instable (and
thereby, we observe large standard errors) without influencing asset weights or
portfolio performance. Thus, we strongly recommend to test the statistical significance
of the coefficient vectors only in case of no weight constraints. Otherwise, the
performance itself should be tested for statistical significance. For being able to evaluate
the performance of the portfolio optimization results, we estimate bootstrapped
standard errors for the certainty-equivalent return of the return differences between the
optimal portfolio policy and the benchmark portfolio. The standard errors of the
certainty equivalent differences reveal that the optimal portfolio policy AAS 1 does not
provide a significant certainty-equivalent gain compared to the benchmark portfolio in
any of the considered cases.
Table III presents the results for AAS 2. Now, we use the characteristics on which SMB,
HML and WML are constructed (the firm’s market capitalization, book-to-market equity
ratio and momentum, AAS 2) in our optimal portfolio policy to outperform the market. The
results for the characteristic coefficients (␪) show that deviations of the optimal portfolio
weights from the benchmark portfolio weights decrease with the firm’s market
capitalization and increase with the firm’s book-to-market equity ratio and momentum.
However, the under-weighting of large firms is relatively low compared to the
over-weighting of high book-to-market equity ratio firms and high momentum firms
indicating strong book-to-market equity ratio and momentum effects. Further evidence for a
significant effect of the book-to-market equity ratio as well as momentum and an
insignificant effect of the market capitalization on portfolio performance is revealed by the
standard errors of the estimated coefficients when short sales are allowed and transaction
costs are not taken into consideration. No matter whether short sales are constrained or
transaction costs are taken into consideration, in all three cases, the optimal portfolio policy
significantly outperforms the benchmark by at least doubling its certainty-equivalent
return.
The empirical application of our approach to equity portfolio optimization by
modeling the portfolio weight in each asset as a function of the asset’s factor loadings or
characteristics is completed by a final look on a comparison of the out-of-sample
performance (Short-Sales: No, Transaction Costs: Yes) of the two optimized portfolios.
Figure 1 shows the price performance of AAS 1 and 2 starting with a value of one
monetary unit in June 1993. At the end of the out-of-sample period in December 2014, the
characteristic-based strategy AAS 2 (15.74) dominates the solely covariances-based
strategy AAS 1 (3.48).
Finally, Table IV reports the results of a regression of the market model (our
benchmark) on each of the return time-series of the two out-of-sample asset allocation
JRF ShSa: Yes, TrCo: No ShSa: No, TrCo: Yes
17,3 In-sample Out-of-sample In-sample Out-of-sample
Statistics B P B P B P B P
␪ME ⫺2.4678 ⫺3.0943 ⫺6.2286 ⫺4.8804

␪BE/ME 6.8822 7.2727 5.1976 9.9889
272 ␪MOM 6.6382 6.1853 12.2092 13.3453
se( ␪ME ) 1.6400 1.9079 17.4328 37.1266
se( ␪BE/ME )
2.7279 3.1275 12.1228 26.5405

se( ␪MOM ) 1.1327 1.4460 31.5117 49.9133
Avg.ⱍwiⱍ 0.0044 0.0262 0.0023 0.0132 0.0044 0.0045 0.0023 0.0023
min wi 0.0000 ⫺0.0673 0.0000 ⫺0.0369 0.0000 0.0000 0.0000 0.0000
max wi 0.1089 0.3758 0.1036 0.2885 0.1089 0.0607 0.1036 0.0624
兺wiI(wi ⬍ 0) 0.0000 ⫺2.4728 0.0000 ⫺2.3505 0.0000 0.0000 0.0000 0.0000
兺I(wi ⱕ 0)/Nt 0.0000 0.5192 0.0000 0.5167 0.0000 0.0000 0.0000 0.0000
␮(r) 0.0085 0.0474 0.0089 0.0478 0.0085 0.0123 0.0088 0.0113
␴(r) 0.0504 0.0903 0.0493 0.0814 0.0504 0.0394 0.0493 0.0342
␮(r)/␴(r) 0.1685 0.5250 0.1799 0.5879 0.1678 0.3121 0.1792 0.3303
CER 0.0016 0.0269 0.0024 0.0309 0.0016 0.0082 0.0024 0.0083
ME ⫺0.0037 ⫺0.0009 4.0091 1.1032 ⫺0.0037 0.0013 4.0091 0.0600
BE/ME 0.0173 0.0035 ⫺0.1782 7.0360 0.0173 ⫺0.0020 ⫺0.1782 1.3740
MOM ⫺0.0056 ⫺0.0066 0.2720 5.6509 ⫺0.0056 0.0010 0.2720 1.2916
P-B P-B P-B P-B
CER 0.0253 0.0285 0.0067 0.0059
se(CER) 0.0062 0.0028 0.0018 0.0003
Notes: We use data from Thomson Reuters Datastream database from June 1983 to December 2014; in the
“out-of-sample” results, we use data until May 1993 to estimate the coefficients of the portfolio policy and then
form out-of-sample portfolios using those coefficients in the next month; every subsequent month, we
reestimate the portfolio policy by enlarging the sample; all statistics in columns labeled “out-of-sample” are
therefore reported for the period June 1993 to December 2014; all statistics in columns labeled “in-sample” are
reported for the period July 1983 to December 2014; the columns labeled “B” and ”P” display statistics of the
market-capitalization-weighted portfolio and the optimal parametric portfolio policy, respectively; the first
set of rows shows the estimated coefficients of the portfolio policy with bootstrapped standard errors; the
“out-of-sample” results display time-series averages of coefficients and standard errors; the
“in-sample”results display coefficients and standard errors for one set of parameters estimated based on
monthly returns from July 1983 to December 2014; the second set of rows shows statistics of the portfolio
weights averaged across time; these statistics include the average absolute weight, the average minimum
and maximum weights, the average sum of negative weights in the portfolio and the average fraction of
negative weights in the portfolio; the third set of rows displays average portfolio return statistics: average
return, standard deviation, average return to standard deviation ratio and certainty-equivalent return; the
Table III. fourth set of rows displays the average normalized covariances and/or characteristics of the portfolio; the
Linear portfolio final set of rows reports the difference in the certainty-equivalent returns of the optimal portfolio policy and
policy: AAS 2 the benchmark portfolio with bootstrapped standard errors.
strategies. We find that the regression intercept of the covariances-based AAS is

negative and statistically significant different from zero. The regression intercept of the
characteristics-based AAS is positive and statistically different from zero providing
final evidence for the profitability of characteristics-based strategies on the German
stock market. Our results thus indicate that an investor relying on the characteristics
Investor’s
perspective on
risk-models
273
Figure 1.
Out-of-sample
performance
Model ␣ ␤ p(␣) p(␤) Adj. R2
AAS1 ⫺0.0023 0.6748 0.9672 0.0000 0.7972

AAS2 0.0042 0.6387 0.5738 0.0000 0.5429
Note: The table reports alpha, beta, p-value of alpha, p-value of beta (both in %) and R2 of an regression Table IV.
of the market model on each of the return time-series of the five out-of-sample asset allocation strategies Market model
(Short-Sales: No, Transaction Costs: Yes) regressions
market capitalization, book-to-market equity ratio and momentum outperforms an

investor holding the value-weighted market portfolio and that an investor holding the
value-weighted market portfolio outperforms an investor relying on the firm’s factor
loadings on SMB, HML and WML.
We did also rerun our analysis using market capitalization and the
book-to-market equity ratio to conduct the characteristics-based AAS and the factor
loadings on SMB and HML to conduct the covariances-based AAS, respectively.
These two strategies clearly rely on the Fama and French (1993)-model and not on
the FFC-model. We can report that the results are qualitatively the same. The
characteristics-based asset allocation strategies dominate covariances-based asset
allocation strategies. We can further state that AAS 2 gains most of the performance
from the momentum characteristic, while, as expected, market capitalization has the
least impact on portfolio performance.
JRF 5. Conclusion
17,3 In this paper, we analyze the performance of characteristics- and covariances-based
asset allocation strategies applying a portfolio optimization approach that enables
investors to directly link competing asset pricing explanations for the
cross-sectional differences in average stock returns (risk or mispricing) to portfolio
decisions. Our results derived for the German stock market show that an investment
274 strategy relying on characteristics (particularly on momentum) outperforms a
covariances-based investment strategy. However, the results of our portfolio
optimization procedure should not be interpreted from an asset pricing perspective.

Our results do not indicate that the characteristics model dominates the covariance
model on the German stock market, as our approach is not able do distinguish
between risk factor loadings and characteristics. This is because the factor loadings
and characteristics are likely to be correlated. Additionally, factor loadings are
likely being measured with error. Such issues present potential directions for future
research. From a portfolio management perspective, our approach provides
guidance directed towards the question as to how investors can form portfolios
based on opposing asset pricing explanations.
Notes
1. Because the asset’s loading on a risk factor and the covariance of that asset with the respective
risk factor can be converted into each other, the terms “systematic risk”, “factor loadings” and
“covariances” are often used synonymously in the literature and also in this paper.
2. In line with Ince and Porter (2006), we additionally adopted the forms from a search for all
German equities using the following Thomson Reuters Datastream filters: status ⫽ all,
market ⫽ Germany, instrument type ⫽ equity.
References
Ali, A., Hwang, L.S. and Trombley, M.A. (2003), “Residual-income-based valuation predicts future
stock returns: evidence on mispricing vs. risk explanations”, The Accounting Review,
Vol. 78 No. 2, pp. 377-396.
Banz, R.W. (1981), “The relationship between return and market value of common stocks”, Journal
of Financial Economics, Vol. 9 No. 1, pp. 3-18.
Brandt, M.W., Santa-Clara, P. and Valkanov, R. (2009), “Parametric portfolio policies: exploiting
characteristics in the cross-section of equity returns”, Review of Financial Studies, Vol. 22 No. 9,
pp. 3411-3447.
Carhart, M.M. (1997), “On persistence in mutual fund performance”, The Journal of Finance,
Vol. 52 No. 1, pp. 57-82.
Chou, P.H., Li, W.S. and Zhou, G. (2006), “Portfolio optimization under asset pricing anomalies”,
Japan and the WORLD Economy, Vol. 18 No. 2, pp. 121-142.
Daniel, K., Grinblatt, M., Titman, S. and Wermers, R. (1997), “Measuring mutual fund performance
with characteristic-based benchmarks”, The Journal of Finance, Vol. 52 No. 3,
pp. 1035-1058.
Daniel, K. and Titman, S. (1997), “Evidence on the characteristics of cross sectional variation in
stock returns”, The Journal of Finance, Vol. 52 No. 1, pp. 1-33.
Daniel, K. and Titman, S. (1998), “Characteristics or covariances?”, The Journal of Portfolio
Management, Vol. 24 No. 4, pp. 24-33.
Daniel, K., Titman, S. and Wei, K. (2001), “Explaining the cross-section of stock returns in japan: Investor’s
factors or characteristics?”, The Journal of Finance, Vol. 56 No. 2, pp. 743-766.
perspective on
Davis, J.L., Fama, E.F. and French, K.R. (2000), “Characteristics, covariances, and average returns: risk-models
1929 to 1997”, The Journal of Finance, Vol. 55 No. 1, pp. 389-406.
Fama, E.F. and French, K.R. (1992), “The cross-section of expected stock returns”, The Journal of
Finance, Vol. 47 No. 2, pp. 427-465.
Fama, E.F. and French, K.R. (1993), “Common risk factors in the returns on stocks and bonds”, 275
Journal of Financial economics, Vol. 33 No. 1, pp. 3-56.
Fieberg, C., Varmaz, A. and Poddig, T. (2016), “Covariances vs characteristics: what does explain the
cross section of the german stock market returns?”, Business Research, Vol. 9 No. 1, pp. 1-24.
Fletcher, J. (2002), “Characteristics versus covariances: an examination of domestic asset
allocation strategies”, Advances in Investment Analysis and Portfolio Management, Vol. 9
No. 1, pp. 251-271.
Grundy, B.D. and Martin, J.S. (2001), “Understanding the nature of the risks and the source of the
rewards to momentum investing”, Review of Financial Studies, Vol. 14 No. 1, pp. 29-78.
Hand, J.R. and Green, J. (2011), “The importance of accounting information in portfolio
optimization”, Journal of Accounting, Auditing & Finance, Vol. 26 No. 1, pp. 1-34.
Haugen, R.A. and Baker, N.L. (1996), “Commonality in the determinants of expected stock
returns”, Journal of Financial Economics, Vo. 41 No. 3, pp. 401-439.
Hirshleifer, D., Hou, K. and Teoh, S.H. (2012), “The accrual anomaly: risk or mispricing?”,
Management Science, Vol. 58 No. 2, pp. 320-335.
Hjalmarsson, E. and Manchev, P. (2012), “Characteristic-based mean-variance portfolio choice”,
Journal of Banking & Finance, Vol. 36 No. 5, pp. 1392-1401.
Hou, K., Karolyi, G.A. and Kho, B.-C. (2011), “What factors drive global stock returns?”, Review of
Financial Studies, Vol. 24 No. 8, pp. 2527-2574.
Ince, O.S. and Porter, R.B. (2006), “Individual equity return data from thomson datastream: handle
with care!”, Journal of Financial Research, Vol. 29 No. 4, pp. 463-479.
Jegadeesh, N. (1990), “Evidence of predictable behavior of security returns”, The Journal of
Finance, Vol. 45 No. 3, pp. 881-898.
Jegadeesh, N. and Titman, S. (1993), “Returns to buying winners and selling losers: implications
for stock market efficiency”, The Journal of Finance, Vol. 48 No. 1, pp. 65-91.
Lajili-Jarjir, S. (2007), “Explaining the cross-section of stock returns in france: characteristics or
risk factors?”, European Journal of Finance, Vol. 13 No. 2, pp. 145-158.
Li, X., Sullivan, R.N. and García-Feijóo, L. (2013), “The low-volatility anomaly: market evidence on
systematic risk versus mispricing”, Financial Analysts Journal, Vol. 72 No. 6.
Lin, X. and Zhang, L. (2013), “The investment manifesto”, Journal of Monetary Economics, Vol. 60
No. 3, pp. 351-366.
Mohanram, P. and Rajgopal, S. (2009), “Is pin priced risk?”, Journal of Accounting and Economics,
Vol. 47 No. 3, pp. 226-243.
Pástor, L. and Stambaugh, R.F. (2000), “Comparing asset pricing models: an investment
perspective”, Journal of Financial Economics, Vol. 56 No. 3, pp. 335-381.
Rosenberg, B., Reid, K. and Lanstein, R. (1985), “Persuasive evidence of market inefficiency”, The
Journal of Portfolio Management, Vol. 11 No. 3, pp. 9-16.
JRF Stattman, D. (1980), “Book values and stock returns”, The Chicago MBA: A Journal of Selected
Papers, Vol. 4 No. 1, pp. 25-45.
17,3
Viale, A.M., Kolari, J.W. and Fraser, D.R. (2009), “Common risk factors in bank stocks”, Journal of
Banking & Finance, Vol. 33 No. 3, pp. 464-472.
Further reading
276 Fama, E.F. and French, K.R. (2012), “Size, value, and momentum in international stock returns”,
Journal of Financial Economics, Vol. 105 No. 3, pp. 457-472.
Corresponding author
Armin Varmaz can be contacted at: armin.varmaz@hs-bremen.de
For instructions on how to order reprints of this article, please visit our website:
www.emeraldgrouppublishing.com/licensing/reprints.htm
Or contact us for further details: permissions@emeraldinsight.com

Fieberg Et Al 17 An Investor's Perspective On Risk-Models and Characteristic-Models

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Fieberg Et Al 17 An Investor's Perspective On Risk-Models and Characteristic-Models

Uploaded by

Copyright:

Available Formats

The Journal of Risk Finance

An investor’s perspective on risk-models and characteristic-models

*Related content and download information correct at time of download.

Received 29 February 2016

Lajili-Jarjir (2007) aims to provide further empirical evidence by analyzing whether

portfolio weight of a stock as a function of the stocks’ characteristics or risk factor

max Et关 u( rp,t⫹1 ) 兴 ⫽ Et u wi,t ri,t⫹1 . (1)

The optimal portfolio weights are parameterized as a function of the stocks’

ˉ i,t ⫹ 1 ( lChar␪ T x̂i,t ⫹ lCov␭ T b̂i,t ),

3.2 Refinements and extensions

4.1 Asset allocation strategies

E(Ri ) ⫺ Rf ⫽ bi[E(RM ) ⫺ Rf ] ⫹ siE(SMB) ⫹ hiE(HML) ⫹ wiE(WML). (5)

We consider two asset allocation strategies to demonstrate how an investor can

return from month t – 12 to month t – 2.

Descriptive statistics ME BE/ME MOM s h w

Mean 607.39 0.80 0.02 0.76 0.12 ⫺0.09

␭s ⫺2.5430 ⫺1.1568 ⫺7.7340 ⫺82.2007

1.4950 1.9816 9.3661 30.7584

␪ME ⫺2.4678 ⫺3.0943 ⫺6.2286 ⫺4.8804

2.7279 3.1275 12.1228 26.5405

strategies. We find that the regression intercept of the covariances-based AAS is

Model ␣ ␤ p(␣) p(␤) Adj. R2

AAS1 ⫺0.0023 0.6748 0.9672 0.0000 0.7972

market capitalization, book-to-market equity ratio and momentum outperforms an

optimization procedure should not be interpreted from an asset pricing perspective.

You might also like