Professional Documents
Culture Documents
a r t i c l e
i n f o
Article history:
Received 28 April 2014
Received in revised form 5 December 2014
Accepted 10 December 2014
Available online 17 December 2014
JEL classication:
GO
G1
a b s t r a c t
By using industry level data, we examine the relation between equity returns and ination in a frequency dependent framework. Our analysis shows that a positive relation in fact exists between equity returns and high frequency ination shocks for commodity and technology related industries. Since higher frequency shocks are
independent from trend and are transitory in nature, our ndings imply a positive relation between stock returns
and the unexpected component of ination. Furthermore, we show that the results are robust to rm-level data
by using a sample from the oil industry. Hence, our study provides a new look at the impact of ination on equities by showing the sensitivity of conclusions in prior work to frequency dependence in data.
2014 Elsevier Inc. All rights reserved.
Keywords:
Ination
Stock returns
Frequency domain
1. Introduction
The relation between ination and equity prices, motivated by the
Fisher hypothesis, is one of the most frequently investigated topics in
nance and economics. The underlying intuition of this hypothesis is
that stocks represent claims on real assets and therefore, their valuations should increase with ination. Thus, a positive relation is predicted
between ination and stock price movements, in which case it can be
argued that equities provide a good hedge against the ination risk.
Unfortunately, empirical studies, beginning with the early work
of Bodie (1976) and Fama and Schwert (1977), consistently report a
negative correlation between stock returns and ination. Ang et al.
(2011) and Hagmann and Lenz (2004) are examples of more recent
papers that reach similar conclusions. One of the explanations offered
for this counter-intuitive nding is the proxy hypothesis of Fama
(1981) and Kaul (1987). The argument of these authors is that the
documented negative correlation with ination is spurious. It arises
because the stock market anticipates the negative impact of higher
ination on growth, which lowers the market valuations.1
I am grateful to Richard Ashley for providing the code to conduct the frequency
domain decomposition used in this paper. Comments provided by an anonymous referee
signicantly improved the paper. All remaining errors are mine.
Tel.: +1 910 962 7497.
E-mail address: cinerc@uncw.edu.
1
Geske and Roll (1983) and Pearce and Roley (1988) provide discussions of alternative
hypotheses on the negative correlation between ination and stock returns.
http://dx.doi.org/10.1016/j.rfe.2014.12.001
1058-3300/ 2014 Elsevier Inc. All rights reserved.
inationary shocks. For example, Fama and Schwert (1977) use changes
in the short term interest rate to infer the expected ination. Bodie
(1976) relies on ARIMA models to construct proxies for expected ination. Hagmann and Lenz (2004) and Hess and Lee (1999) use vector
autoregression (VAR) models to decompose ination into expected
and unexpected components. In related work, McQueen and Roley
(1993) and Pearce and Roley (1988) rely on forecasts by Money Market
Services International to identify the surprise element in ination announcements. More recently, Wei (2009) uses a time series regression
model that also includes lagged values of monthly unemployment rate
to estimate the unexpected component of ination. These studies in
general support the conclusion of earlier papers that the relation
between stock returns and unexpected ination is also negative.
Furthermore, in papers closely related to the present article, Lee and
Ni (1996) use the Chebyshev lter to obtain estimates of unexpected
ination as the transitory component of inationary shocks. These
authors argue that there are differences in the relation between stock
returns and ination based on the frequency components of ination.
Kim and In (2006) rely on a multi-scale wavelet decomposition
approach to examine the same relation. Interestingly, they detect a
positive relation between ination and stock returns at the shortest
scale, which corresponds to the highest frequency shocks. We provide
a further discussion in the following section on why we have preferred
the AshleyVerbrugge decomposition in the present paper rather than
the abovementioned method of analyses.3
We also examine the stock returnination relation by relying on
industry portfolios. This permits us to determine whether some sectors
of the stock market may be considered better hedges against ination.
For example, Boudoukh et al. (1994) argue that non-cyclical stocks tend
to be more positively correlated with ination than cyclical stocks. Stocks
of natural resource companies are usually considered as better ination
hedges because of the sensitivity of commodity prices to ination. In the
empirical analysis, therefore, we rst calculate the conventional ination
betas using 48 industry portfolios by estimating conventional stock
returnination regressions. Our data cover the period between 1990
and 2013 and hence, a further contribution of our analysis is to update
evidence in the literature for a more recent period. Consistent with
prior work, we nd that, in general, ination betas are negative.
Subsequently, we estimate the same model by replacing ination
with its frequency components obtained by the Ashley and Verbrugge
(2009) decomposition. The results of this analysis lend support to
the central claim of the paper, which is that the stock returnination
relation shows dependency on the persistence of inationary shocks.
Specically, we nd that while long term, trend shocks, replicate the
negative ination betas obtained above as expected, ination betas
for unexpected (high frequency) shocks are in fact positive for 18 industry portfolios. These positive unexpected ination betas exist in
commodity-sensitive (such as coal, mines, oil, gold and agricultural)
and technology-related industries (such as telecoms, software and
chips) among others.
The rst part of the empirical analysis utilizes value-weighted industry portfolios and hence, could simply be an artifact of data, since larger
companies dominate value-weighted indices. To investigate the robustness of the ndings to rm size, we conduct the analysis by using
equally-weighted industry portfolios. We nd largely similar results.
Again, unexpected ination betas for commodity-sensitive industries
(gold, mines and oil) as well as technology-related industries (telecom,
hardware and chips) have positive, and statistically signicant, unexpected ination betas.
In the next section, we outline the statistical method of analysis used
in the paper. In Section 3, we present the data set, and discuss the
empirical ndings in Section 4. We provide the concluding comments
of the paper in the nal section of the study.
13
As;t
8
>
1 1=2
>
>
; for s 1;
>
>
T
>
>
> 1=2
>
>
2
st1
>
>
;
for s 2; 4; 6; ; T2orT1
cos
< T
T
1
>
2 2
s1t1
>
>
;
for s 3; 5; 7; ; T1or T;
sin
>
>
T
T
>
>
>
1
>
>
2
1
t1
>
>
1 ;
for s T when T is even
:
T
AY AX A
A is an orthogonal matrix and the pre-multiplication gives:
2
Y X ; n 0; I :
In Eq. (2), the components of the variables now represent frequencies instead of time periods. Next, the T frequency components are
partitioned into M frequency bands and dummy variables are created
to dene M vectors of length T, which can be written as D1, .. DM.
These dummy variables are used in a manner that for elements, which
fall into the sth frequency band, Ds equals Xj and the elements are
zero otherwise. Consequently, the regression equation can be rewritten
as follows:
Y X f jg f jg m1 j;m D
m
in which X{j} is the X* with its jth column deleted and {j} is the vector
with its jth component deleted. Hence, frequency-dependent coefcients j,1 .. j,M can be estimated and hypotheses tests can be
conducted on the signicance of these parameters.
However, Ashley and Verbrugge (2009) argue that an important
weakness exists in the above approach. In particular, since premultiplying with matrix A mixes past and futures values of the variables,
the M frequency components will be correlated with the error terms, if
there is feedback between the dependent and any of the independent
variables, which is likely in nance and economics data. This would
yield inconsistent estimates if the partitioned frequency components
are used in an OLS regression.
The main contribution of Ashley and Verbrugge (2009) is that they
present a solution to this problem by applying a one-sided, rather
14
3. Data
The data used in our study include the CPI, measured as a monthly
series of annual percentage change in headline ination, and the
monthly returns on 48 stock industry portfolios. The series span the
period between January of 1990 and December of 2012.5 The ination
series are obtained from the Fred database at the St. Louis Federal
Reserve and the industry stock returns are found on Kenneth French's
website. Table 1 provides the summary statistics of monthly stock
returns. It can be observed that the returns are positive and statistically
signicantly indifferent from zero, consistent with the view that stock
prices are a submartingale process. We also detect negative skewness
and excess kurtosis in returns similar to prior work.
As mentioned in the Introduction, the essential part of the empirical
analysis of this study involves decomposing the ination rate into its
frequency components using the AshleyVerbrugge approach. The
36-month window used in the application decomposes the ination
rate into 19 frequency components including the zero frequency. We
could estimate the stock returnination regressions by substituting
the ination variable by its 19 frequency components. However, this
would involve a substantial loss in degrees of freedom and, also, different ndings across individual frequencies could make interpreting the
results difcult. Therefore, we generate four new groups from the
frequency components that have intuitive motivations.
Specically, we create four new variables by regrouping the frequency
components as follows:
Long
Medium1
Medium2
Short
15
Table 1
Summary statistics.
Industry
Mean
Std. dev.
Skewness
Kurtosis
Agric
Food
Soda
Beer
Smoke
Toys
Fun
Books
Hshld
Clths
Hlth
MedEq
Drugs
Chems
Rubbr
Txtls
BldMt
Cnstr
Steel
FabPr
Mach
ElcEq
Aero
Ships
Guns
Gold
Mines
Coal
Oil
Util
Telcm
Persv
Bussv
Hardw
Softw
Chips
LabEq
Paper
Boxes
Trans
Whlsl
Rtail
Meals
Banks
Insur
Real
Fin
1.14 (.00)
.74 (.00)
1.36 (.00)
.84 (.00)
1.22 (.00)
.47 (.24)
1.05 (.04)
.63 (.08)
.86 (.00)
.87 (.03)
.78 (.06)
.85 (.00)
.92 (.00)
.94 (.01)
.92 (.01)
.56 (.30)
.89 (.03)
1.08 (.01)
.78 (.17)
.48 (.36)
1.11 (.01)
1.25 (.00)
1.22 (.00)
1.42 (.00)
1.09 (.00)
.82 (.25)
1.16 (.02)
1.81 (.02)
1.07 (.00)
.77 (.00)
.68 (.05)
.61 (.12)
.72 (.03)
1.32 (.02)
1.18 (.01)
1.09 (.05)
.95 (.04)
.77 (.02)
.79 (.05)
.86 (.01)
.74 (.01)
.88 (.00)
.96 (.00)
.83 (.04)
.84 (.01)
.68 (.17)
1.14 (.01)
6.43
3.97
7.67
4.93
7.23
6.42
8.03
5.59
4.25
6.83
6.61
4.82
4.57
5.82
6.05
8.53
6.53
7.12
8.86
8.32
6.99
6.56
6.33
7.58
6.48
11.24
8.10
12.47
5.53
4.13
5.42
6.20
5.25
8.80
7.75
8.78
7.23
5.31
6.36
5.30
4.78
5.01
4.87
6.33
5.40
7.79
7.39
.52
.13
.24
.41
.14
.23
.23
.26
.39
.17
.33
.92
.15
.08
.06
1.14
.10
.38
.29
.07
.49
.31
.93
.11
.72
1.25
.50
.11
.04
.60
.23
.19
.69
.33
.01
.46
.28
.12
.39
.38
.65
.19
.42
.81
.69
.94
.49
1.84
1.68
3.59
2.03
2.38
1.19
3.57
4.70
2.34
1.86
1.04
2.23
.09
2.14
4.02
11.42
6.16
1.07
1.79
3.08
2.11
1.15
2.56
1.96
2.49
8.47
2.48
.89
.87
.91
1.24
.92
2.27
1.23
.86
1.26
1.19
2.72
1.15
.99
2.33
.53
.51
2.86
3.90
15.01
.89
Fig. 2. Long.
4. Empirical ndings
The primary model to estimate the relation between stock returns
and ination is the following equation:
Rit conv t t
Fig. 3. Medium1.
16
Fig. 4. Medium2.
Fig. 5. Short.
consider six large integrated oil and gas companies: Chevron (CVX),
ConocoPhillips (COP), Exxon Mobil (XOM), Hess (HES), Marathon Oil
(MRO) and Murphy Oil (MUR). We report the ndings for this part of
the analysis in Table 4 and, save for MUR, there is a positive and statistically signicant relation between the unexpected component of ination and these individual stock returns at the 10% level, suggesting that
6
To examine the robustness of our ndings, we also used time series models to decompose ination. Specically, we estimated expected ination using both VAR(1) and
AR(2) models for three selected industries; reit, n and gold. We chose those industries
because they have positive and statistically signicant unexpected ination betas. The unexpected ination betas are .31 (.79) for reit, .96 (.40) for n and 1.93 (.27) for gold when a
VAR(1) model is used. When, an AR(2) model is used, they are 1.01 (.17) for reit, 1.33 (.29)
for n and 2.86 (.13) for gold. These results are consistent with the view that time series
models are unlikely to capture the dynamics uncovered by the AshleyVerbrugge decomposition used in the paper and also, that it is difcult to determine the correct forecasting
model to determine expected ination. We thank an anonymous referee for suggesting
this analysis.
Table 2
Ination betas: value-weighted industry portfolios.
Industry
Conventional
Long
Medium1
Medium2
Short
Agric
Food
Soda
Beer
Smoke
Toys
Fun
Books
Hshld
Clths
Hlth
MedEq
Drugs
Chems
Rubbr
Txtls
BldMt
Cnstr
Steel
FabPr
Mach
ElcEq
Aero
Ships
Guns
Gold
Mines
Coal
Oil
Util
Telcm
Persv
Bussv
Hardw
Softw
Chips
LabEq
Paper
Boxes
Trans
Whlsl
Rtail
Meals
Banks
Insur
Real
Fin
.10 (.81)
.19 (.45)
1.03 (.03)
.29 (.36)
.02 (.95)
.70 (.11)
1.59 (.01)
1.19 (.01)
.34 (.26)
.79 (.07)
.12 (.79)
.52 (.05)
.29 (.26)
.92 (.04)
.88 (.09)
1.20 (.22)
1.01 (.08)
.93 (.04)
1.04 (.08)
.56 (.30)
1.01 (.03)
1.02 (.02)
.75 (.06)
.81 (.14)
.26 (.52)
.99 (.14)
1.61 (.00)
.94 (.27)
.39 (.28)
.10 (.69)
.80 (.01)
.21 (.59)
.64 (.04)
1.16 (.01)
.87 (.01)
.78 (.08)
.64 (.08)
.82 (.04)
.95 (.01)
.26 (.49)
.55 (.07)
.61 (.03)
.51 (.08)
.41 (.84)
.69 (.10)
1.39 (.06)
.94 (.03)
.09 (.86)
.22 (.47)
1.19 (.02)
.45 (.18)
.02 (.96)
.54 (.26)
1.89 (.00)
1.62 (.00)
.58 (.07)
.94 (.06)
.23 (.59)
.66 (.05)
.27 (.38)
1.05 (.04)
1.04 (.05)
1.47 (.13)
.98 (.11)
.69 (.21)
.85 (.26)
.78 (.28)
1.09 (.06)
1.05 (.04)
.70 (.13)
.75 (.22)
.38 (.44)
.31 (.72)
1.33 (.06)
1.75 (.10)
.30 (.49)
.17 (.60)
.81 (.04)
.08 (.86)
.65 (.10)
1.45 (.01)
.95 (.06)
.78 (.20)
.72 (.16)
1.07 (.01)
1.00 (.01)
.43 (.29)
.57 (.14)
.65 (.06)
.40 (.24)
.64 (.21)
.73 (.12)
1.26 (.08)
.87 (.12)
3.24 (.22)
.26 (.68)
1.69 (.25)
.25 (.73)
1.09 (.31)
.75 (.60)
.73 (.69)
1.36 (.34)
.61 (.46)
.31 (.81)
1.91 (.10)
.83 (.32)
.56 (.46)
1.22 (.28)
1.03 (.52)
.93 (.73)
1.37 (.43)
2.03 (.16)
2.20 (.27)
1.60 (.38)
1.79 (.18)
.84 (.54)
.46 (.68)
.74 (.76)
.03 (.97)
2.41 (.15)
2.97 (.07)
2.23 (.17)
1.71 (.07)
.30 (.68)
.99 (.33)
.62 (.61)
1.41 (.13)
.32 (.79)
1.06 (.30)
1.50 (.23)
1.69 (.12)
.85 (.48)
.96 (.40)
.39 (.73)
1.05 (.22)
.84 (.33)
.25 (.78)
.04 (.97)
.45 (.73)
2.46 (.27)
2.34 (.07)
2.54 (.22)
.17 (.85)
.66 (.78)
.82 (.43)
.94 (.59)
1.74 (.41)
.06 (.97)
2.11 (.22)
.67 (.54)
.18 (.92)
2.10 (.22)
.89 (.55)
.13 (.90)
.42 (.80)
.42 (.78)
.62 (.82)
.82 (.63)
1.67 (.37)
1.17 (.65)
1.88 (.41)
.38 (.81)
.76 (.61)
1.25 (.39)
1.21 (.54)
.44 (.79)
4.42 (.16)
2.34 (.30)
6.23 (.05)
.12 (.93)
.59 (.56)
.40 (.75)
.82 (.58)
.21 (.87)
.36 (.83)
.06 (.96)
.02 (.99)
1.00 (.56)
1.10 (.47)
.58 (.74)
.54 (.71)
.10 (.93)
.14 (.90)
1.92 (.20)
.94 (.65)
.51 (.77)
1.20 (.63)
.13 (.94)
24.06 (.00)
4.20 (.46)
.68 (.94)
1.20 (.83)
6.15 (.48)
7.15 (.33)
8.65 (.42)
10.31 (.11)
1.26 (.79)
3.47 (.67)
4.72 (.59)
13.06 (.07)
5.33 (.32)
14.41 (.07)
4.44 (.51)
1.63 (.89)
9.51 (.20)
9.80 (.23)
29.50 (.00)
11.50 (.25)
19.53 (.01)
17.90 (.01)
8.16 (.30)
11.43 (.24)
3.46 (.70)
29.06 (.06)
20.95 (.04)
31.30 (.05)
18.56 (.00)
3.49 (.46)
14.47 (.02)
3.15 (.63)
12.88 (.02)
8.45 (.36)
19.17 (.03)
21.52 (.02)
22.10 (.00)
6.55 (.31)
.51 (.94)
6.97 (.31)
10.59 (.07)
2.43 (.69)
5.82 (.36)
4.00 (.64)
5.09 (.46)
22.98 (.03)
15.94 (.08)
Note This table provides the ination beta for value-weighted industry portfolios
calculated by the conventional and frequencydecomposition approaches. Bold entries
signify statistical signicance.
Conventional
Long
Medium1
Medium2
Short
Agric
Food
Soda
Beer
Smoke
Toys
Fun
Books
Hshld
Clths
Hlth
MedEq
Drugs
Chems
Rubbr
Txtls
BldMt
Cnstr
Steel
FabPr
Mach
ElcEq
Aero
Ships
Guns
Gold
Mines
Coal
Oil
Util
Telcm
Persv
Bussv
Hardw
Softw
Chips
LabEq
Paper
Boxes
Trans
Whlsl
Rtail
Meals
Banks
Insur
Real
Fin
.42 (.43)
.73 (.02)
.94 (.04)
.71 (.05)
.53 (.30)
1.24 (.02)
1.34 (.01)
2.49 (.00)
1.25 (.02)
1.25 (.02)
.53 (.19)
.92 (.02)
1.17 (.04)
1.15 (.02)
1.07 (.04)
1.32 (.09)
.84 (.09)
1.10 (.05)
.76 (.20)
.73 (.25)
.83 (.07)
.89 (.04)
.64 (.15)
.92 (.18)
.94 (.03)
2.35 (.00)
2.04 (.00)
.78 (.41)
.48 (.49)
.17 (.43)
1.66 (.00)
.99 (.02)
1.09 (.00)
1.42 (.00)
1.30 (.00)
1.26 (.01)
.83 (.05)
1.62 (.00)
.93 (.02)
.77 (.09)
1.09 (.02)
1.48 (.00)
1.24 (.01)
.06 (.85)
.68 (.09)
1.33 (.02)
.81 (.03)
.52 (.38)
.72 (.06)
1.20 (.02)
.66 (.09)
.90 (.17)
1.41 (.02)
1.50 (.01)
2.94 (.00)
1.59 (.00)
1.44 (.00)
.51 (.31)
1.09 (.03)
.94 (.18)
1.34 (.02)
1.46 (.01)
2.33 (.00)
.87 (.12)
.89 (.15)
.80 (.26)
.84 (.22)
1.06 (.07)
1.08 (.04)
.69 (.16)
.94 (.20)
.81 (.15)
1.85 (.10)
2.13 (.00)
1.75 (.14)
.63 (.46)
.25 (.37)
1.85 (.00)
.95 (.05)
1.15 (.02)
1.52 (.02)
1.51 (.01)
1.40 (.04)
1.10 (.04)
2.09 (.00)
1.03 (.04)
.84 (.09)
1.16 (.02)
1.56 (.00)
.1.22 (.02)
.15 (.68)
.54 (.22)
1.49 (.01)
.72 (.10)
3.04 (.04)
1.45 (.10)
1.79 (.18)
.96 (.35)
.63 (.74)
1.30 (.31)
2.06 (.19)
2.21 (.21)
1.24 (.45)
.69 (.66)
2.35 (.06)
1.84 (.07)
3.31 (.02)
1.27 (.33)
.08 (.96)
.41 (.85)
1.66 (.27)
3.02 (.08)
2.54 (.17)
1.59 (.40)
1.41 (.30)
1.96 (.13)
1.09 (.39)
1.36 (.52)
2.39 (.03)
6.40 (.00)
4.42 (.01)
1.09 (.54)
2.44 (.12)
.42 (.57)
2.28 (.10)
2.09 (.14)
2.42 (.01)
2.58 (.06)
2.28 (.06)
2.39 (.09)
1.91 (.11)
1.55 (.36)
1.17 (.32)
.91 (.51)
2.24 (.08)
2.16 (.71)
2.28 (.12)
.07 (.95)
1.04 (.41)
2.01 (.27)
1.85 (.09)
2.49 (.19)
.10 (.93)
1.67 (.40)
.77 (.58)
2.11 (.38)
.12 (.94)
.41 (.82)
.64 (.78)
1.31 (.46)
.21 (.91)
.82 (.63)
1.13 (.49)
.44 (.81)
.45 (.81)
.90 (.62)
5.20 (.02)
.17 (.92)
.94 (.64)
1.15 (.62)
.93 (.66)
1.44 (.40)
1.47 (.39)
.19 (.90)
.10 (.96)
.78 (.68)
2.34 (.52)
.73 (.77)
6.60 (.04)
2.46 (.29)
.58 (.52)
.40 (.83)
.38 (.81)
.55 (.71)
.48 (.80)
1.14 (.54)
.89 (.66)
2.11 (.21)
1.78 (.36)
.02 (.98)
.10 (.95)
.39 (.80)
.33 (.87)
.51 (.80)
.58 (.68)
1.33 (.44)
.48 (.83)
.55 (.71)
15.95 (.15)
4.93 (.44)
2.04 (.81)
6.12 (.31)
8.80 (.39)
12.21 (.14)
5.33 (.53)
11.15 (.28)
9.62 (.20)
6.42 (.47)
10.11 (.22)
11.41 (.20)
15.70 (.22)
14.42 (.09)
.03 (.99)
2.32 (.79)
8.98 (.24)
12.75 (.19)
17.02 (.08)
23.88 (.01)
18.59 (.02)
17.86 (.03)
9.31 (.26)
24.48 (.01)
3.51 (.71)
33.27 (.02)
23.46 (.03)
19.87 (.28)
32.05 (.00)
3.56 (.38)
24.37 (.01)
7.68 (.30)
13.13 (.08)
20.19 (.07)
17.27 (.12)
22.33 (.04)
20.61 (.02)
9.41 (.28)
.49 (.95)
5.56 (.47)
8.14 (.30)
4.20 (.65)
1.20 (.89)
1.95 (.75)
6.09 (.33)
10.76 (.29)
10.33 (.14)
Note This table provides ination betas for equally-weighted portfolios using conventional
and frequency decomposition approaches. Bold entries signify statistical signicance.
CVX
COP
XOM
HES
MRO
MUR
Conventional
Long
Medium1
Medium2
Short
.29 (.43)
.35 (.45)
.14 (.61)
.33 (.61)
.37 (.47)
.81 (.31)
.04 (.90)
.39 (.51)
.19 (.56)
.56 (.46)
.03 (.96)
.62 (.35)
1.26 (.27)
.83 (.55)
1.44 (.14)
2.72 (.02)
3.12 (.02)
1.72 (.17)
1.11 (.53)
.61 (.74)
1.40 (.32)
3.47 (.13)
.37 (.86)
1.30 (.55)
13.69 (.05)
23.24 (.00)
9.90 (.09)
26.77 (.01)
29.28 (.00)
16.93 (.13)
Note This table provides the ination beta for oil company stocks using the conventional
and frequency domain approaches. Bold entries signify statistical signicance.
17