does not change the pattern of wavelet coefficients, and it
provides increased resolution at coarser time scales. Despite
this, as noted by Jammazi (2012b), theMODWT also has some
drawbacks. First, it is affected by boundary effects that arise
from applying the wavelet transform near the edge of finite
signals due to the lack of data beyond the boundary. Since
the number of boundary elements increases with scale there
will be many more boundary-affected coefficients at higher
scales. The boundary problems may lead to biased estimates
of the wavelet variance and, hence, to spurious and misleading
results (Percival and Walden, 2000). In order to get
an unbiased estimator of the wavelet variance, the exclusion
of all the wavelet coefficients affected by boundary
conditions has become a very common practice. This loss
of edge information inherent to the MODWT may, therefore,
make it difficult the correct characterization of the
real interactions between variables, especially at the coarsest
scales. Second, the MODWT is typically implemented
with the Daubechies least asymmetric (LA) wavelet filter
of length L = 8, denoted by LA(8), as it allows an accurate
alignment in time between wavelet coefficients at various
scales and the original signal. However, the LA(8) filter is not
appropriate to capture the non-linear and chaotic behavior
typical of many economic and financial time series (i.e. oil
price, inflation, stock returns, etc.).
The present study differs from the above mentioned
papers in the use of the HTW transform, a new discrete and
robust wavelet technique that overcomes the major constraints
of the MODWT and has been rarely applied to date
in economic and finance areas. In fact, Jammazi and Aloui
(2010) and Jammazi (2012a,b) are the only ones who have
used the HTW within a financial framework, particularly with
the purpose of exploring the dynamic relationship between
oil price changes and stock returns.
Most of the literature on the interest rate-stock market
link has focused on a few countries with highly developed
financial markets, especially the U.S. and more recently
Germany, the U.K. or Australia. Concerning the Spanish
case, there are some studies (Ferrer et al., 2010; Jareno,
2008; Soto et al., 2005), based primarily on multifactor linear
regression models, which document a significant linkage
between interest rate movements and firms’ stock returns,
confirming the high interest rate sensitivity of the Spanish
equity market.
Data set
This study deals with the relationship between movements
in interest rates and stock returns in the Spanish case over
the period from January 1993 to December 2012. The starting
date of our analysis is January 1993 to avoid possible
distortions in the interest rate-equity market nexus caused
by the turbulences in financial markets in the context of the
crisis of the European monetary system during the second
half of 1992. Along the lines of, among others, Campbell
(1987), Kim and In (2007), Korkeamaki (2011) and Reilly
et al. (2007), monthly data series are employed (a total of
240 observations). Monthly data (end-of-the month observations)
are preferred to weekly or daily data for several
reasons. Firstly, monthly data are less contaminated by
noise and can therefore better capture interactions between
interest rates and equity prices. Secondly, monthly data
have smaller biases due to non-synchronous trading of some
stocks. Thirdly, the results in terms of smoothness and d ifferentiation
among time horizons generated by the application
of wavelet analysis on monthly data are much harder to
obtain with higher frequency data. In fact, it would be
necessary to use a very large number of decomposition levels
when considering weekly or daily data in order to find
comparable results to those achieved with monthly data in
terms of the time range covered.
In line with previous research on the interest rate-stock
market link (Bartram, 2002; Ferrer et al., 2010; Reilly et al.,
2007; Sweeney and Warga, 1986), this analysis is carried out
at the industry level. Various reasons are usually put forward
to justify an industry-based approach. First, the formation
of industry portfolios provides an efficient way of condensing
a sizable amount of information regarding stock price
behavior. Second, the use of portfolios helps to smooth the
noise in the data produced by transitory shocks in individual
stocks, which leads to more precise estimates. Thus, all
firms listed on the Spanish Stock Exchange for at least one
full year of the sample period (a total of 249 companies) are
assigned to any of the industries considered. Then, valueweighted
industry stock indices are constructed from stock
prices, adjusted for splits and dividends, of individual firms
within each industry portfolio.
The fourteen industries covered are: Consumer Goods,
Consumer Services, Technology and Telecommunications,
Real Estate, Banking, Financial Services, Utilities, Construction,
Chemicals and Paper, Basic Resources, Health Care,
Food and Beverages, Industrials, and Energy. This industry
breakdown is similar to the well-known Industry Classification
Benchmark (ICB), but adapted to the singular features
of the Spanish equity market. For example, the Construction
and Real Estate industries have been added since they
were one of the major growth engines of the Spanish economy
over the period 1996-2007, representing almost 18% of
Spanish GDP in 2007. The use of the official industry classification
of the Spanish Stock Exchange has been discarded in
this study for two main reasons. First, the Spanish industrial
classification has undergone several restructuring processes
over the last two decades, with January 2005 being the most
recent. This implies that there are no sufficiently long time
series of industry stock indices for conducting a credible
empirical analysis with monthly data. Second, the current
industry classification of the Spanish market only distinguishes
six basic industries and, therefore, it seems more
appropriate to use more disaggregated industry level data.
The Indice General de la Bolsa de Madrid (IGBM), the broadest
index of the Spanish equity market, is utilized as an
indicator of the stock exchange as a whole. Equity market
data are collected from the Madrid Stock Exchange
database.
Interest rates used in this study are the yields on 10-year
Spanish government bonds, which have been taken from the
Bank of Spain’s website. This choice has become increasingly
popular in the literature on the linkage between interest
rates and stock market (Ballester et al., 2011; Elyasiani and
Mansur, 1998; Faff et al., 2005; Oertmann et al., 2000) and is
justified for several reasons. First, long-term interest rates
contain market expectations about future prospects for the
economy and determine to a large extent the cost of rowing funds. Thus, long-term rates are likely to have a
critical influence on investment decisions and profitability of
firms and, hence, on their stock market performance. Second,
long-term government bonds are often considered as
closer maturity substitutes to stocks, which may presumably
increase the extent of linkage between the two financial
assets. Industry returns are calculated as the first log d ifference
of industry stock indices. Changes in interest rates
are computed as the first differences in the level of interest
rates between two consecutive observations.
Table 1 presents descriptive statistics for the data. The
average monthly return is positive for most industries as well
as the overall stock market in line with the general increasing
trend in stock prices. The average monthly change
in 10-year government bond yields is, however, negative,
reflecting the clear downward trend in Spanish long-term
rates during the sample period. Based on the standard deviation,
all industry and market returns have, as expected,
higher volatility than the series of changes in 10-year Spanish
bond yields. The measures of skewness indicate that the
majority of industry returns are negatively skewed, meaning
that negative shocks are more common than positive ones.
Furthermore, all industry return series exhibit a kurtosis
significantly larger than three, thereby implying leptokurtic
behavior as compared to the Gaussian distribution. The
Jarque-Bera test statistics corroborate this finding, rejecting
the null hypothesis of normality in all cases at the 1%
level. This result is consistent with that reported for the
Spanish stock market by Miralles et al. (2011). A similar
distributional picture emerges for the 10-year government
bond yield change series. The Augmented Dickey-Fuller
(ADF) and Phillips-Perron (PP) unit root tests indicate that
the series of changes in 10-year bond rates and market and
industry equity returns are all stationary (integrated of order
zero). This finding is in line with that of earlier work on financial
return data (Badillo et al., 2010; Czaja et al., 2009;
Miralles et al., 2012).
Fig. 1 presents the dynamics of the Spanish equity market
index, proxied by the IGBM, and the yield on Spanish
10-year government bonds over the period 1993-2012. The
stock market exhibits a general upward trend during most
of the study period, only interrupted by the Internet bubble
burst in March 2000 and the global financial crisis from
late 2007. On the contrary, the yields on 10-year government
securities display a downward trend up to the start
of the sovereign debt crisis in the euro zone during the
spring of 2010. This figure shows that the interest ratestock
market link over the full sample period is somewhat
unclear. In particular, the Spanish market index and 10-year
government bond yields have moved predominantly in opposite
directions until approximately mid-1998. Since then,
the connection between both va
does not change the pattern of wavelet coefficients, and it
ไม่ได้เปลี่ยนรูปแบบของค่าสัมประสิทธิ์เวฟและมันให้ความละเอียดที่เพิ่มขึ้นในระดับเวลาหยาบ แม้จะมีนี้เท่าที่สังเกตจาก Jammazi (2012b) theMODWT ยังมีข้อบกพร่อง provides increased resolution at coarser time scales. Despite
this, as noted by Jammazi (2012b), theMODWT also has some
drawbacks. First, it is affected by boundary effects that arise
from applying the wavelet transform near the edge of finite
signals due to the lack of data beyond the boundary. Since
the number of boundary elements increases with scale there
will be many more boundary-affected coefficients at higher
scales. The boundary problems may lead to biased estimates
of the wavelet variance and, hence, to spurious and misleading
results (Percival and Walden, 2000). In order to get
an unbiased estimator of the wavelet variance, the exclusion
of all the wavelet coefficients affected by boundary
conditions has become a very common practice. This loss
of edge information inherent to the MODWT may, therefore,
make it difficult the correct characterization of the
real interactions between variables, especially at the coarsest
scales. Second, the MODWT is typically implemented
with the Daubechies least asymmetric (LA) wavelet filter
of length L = 8, denoted by LA(8), as it allows an accurate
alignment in time between wavelet coefficients at various
scales and the original signal. However, the LA(8) filter is not
appropriate to capture the non-linear and chaotic behavior
typical of many economic and financial time series (i.e. oil
price, inflation, stock returns, etc.).
The present study differs from the above mentioned
papers in the use of the HTW transform, a new discrete and
robust wavelet technique that overcomes the major constraints
of the MODWT and has been rarely applied to date
in economic and finance areas. In fact, Jammazi and Aloui
(2010) and Jammazi (2012a,b) are the only ones who have
used the HTW within a financial framework, particularly with
the purpose of exploring the dynamic relationship between
oil price changes and stock returns.
Most of the literature on the interest rate-stock market
link has focused on a few countries with highly developed
financial markets, especially the U.S. and more recently
Germany, the U.K. or Australia. Concerning the Spanish
case, there are some studies (Ferrer et al., 2010; Jareno,
2008; Soto et al., 2005), based primarily on multifactor linear
regression models, which document a significant linkage
between interest rate movements and firms’ stock returns,
confirming the high interest rate sensitivity of the Spanish
equity market.
Data set
This study deals with the relationship between movements
in interest rates and stock returns in the Spanish case over
the period from January 1993 to December 2012. The starting
date of our analysis is January 1993 to avoid possible
distortions in the interest rate-equity market nexus caused
by the turbulences in financial markets in the context of the
crisis of the European monetary system during the second
half of 1992. Along the lines of, among others, Campbell
(1987), Kim and In (2007), Korkeamaki (2011) and Reilly
et al. (2007), monthly data series are employed (a total of
240 observations). Monthly data (end-of-the month observations)
are preferred to weekly or daily data for several
reasons. Firstly, monthly data are less contaminated by
noise and can therefore better capture interactions between
interest rates and equity prices. Secondly, monthly data
have smaller biases due to non-synchronous trading of some
stocks. Thirdly, the results in terms of smoothness and d ifferentiation
among time horizons generated by the application
of wavelet analysis on monthly data are much harder to
obtain with higher frequency data. In fact, it would be
necessary to use a very large number of decomposition levels
when considering weekly or daily data in order to find
comparable results to those achieved with monthly data in
terms of the time range covered.
In line with previous research on the interest rate-stock
market link (Bartram, 2002; Ferrer et al., 2010; Reilly et al.,
2007; Sweeney and Warga, 1986), this analysis is carried out
at the industry level. Various reasons are usually put forward
to justify an industry-based approach. First, the formation
of industry portfolios provides an efficient way of condensing
a sizable amount of information regarding stock price
behavior. Second, the use of portfolios helps to smooth the
noise in the data produced by transitory shocks in individual
stocks, which leads to more precise estimates. Thus, all
firms listed on the Spanish Stock Exchange for at least one
full year of the sample period (a total of 249 companies) are
assigned to any of the industries considered. Then, valueweighted
industry stock indices are constructed from stock
prices, adjusted for splits and dividends, of individual firms
within each industry portfolio.
The fourteen industries covered are: Consumer Goods,
Consumer Services, Technology and Telecommunications,
Real Estate, Banking, Financial Services, Utilities, Construction,
Chemicals and Paper, Basic Resources, Health Care,
Food and Beverages, Industrials, and Energy. This industry
breakdown is similar to the well-known Industry Classification
Benchmark (ICB), but adapted to the singular features
of the Spanish equity market. For example, the Construction
and Real Estate industries have been added since they
were one of the major growth engines of the Spanish economy
over the period 1996-2007, representing almost 18% of
Spanish GDP in 2007. The use of the official industry classification
of the Spanish Stock Exchange has been discarded in
this study for two main reasons. First, the Spanish industrial
classification has undergone several restructuring processes
over the last two decades, with January 2005 being the most
recent. This implies that there are no sufficiently long time
series of industry stock indices for conducting a credible
empirical analysis with monthly data. Second, the current
industry classification of the Spanish market only distinguishes
six basic industries and, therefore, it seems more
appropriate to use more disaggregated industry level data.
The Indice General de la Bolsa de Madrid (IGBM), the broadest
index of the Spanish equity market, is utilized as an
indicator of the stock exchange as a whole. Equity market
data are collected from the Madrid Stock Exchange
database.
Interest rates used in this study are the yields on 10-year
Spanish government bonds, which have been taken from the
Bank of Spain’s website. This choice has become increasingly
popular in the literature on the linkage between interest
rates and stock market (Ballester et al., 2011; Elyasiani and
Mansur, 1998; Faff et al., 2005; Oertmann et al., 2000) and is
justified for several reasons. First, long-term interest rates
contain market expectations about future prospects for the
economy and determine to a large extent the cost of rowing funds. Thus, long-term rates are likely to have a
critical influence on investment decisions and profitability of
firms and, hence, on their stock market performance. Second,
long-term government bonds are often considered as
closer maturity substitutes to stocks, which may presumably
increase the extent of linkage between the two financial
assets. Industry returns are calculated as the first log d ifference
of industry stock indices. Changes in interest rates
are computed as the first differences in the level of interest
rates between two consecutive observations.
Table 1 presents descriptive statistics for the data. The
average monthly return is positive for most industries as well
as the overall stock market in line with the general increasing
trend in stock prices. The average monthly change
in 10-year government bond yields is, however, negative,
reflecting the clear downward trend in Spanish long-term
rates during the sample period. Based on the standard deviation,
all industry and market returns have, as expected,
higher volatility than the series of changes in 10-year Spanish
bond yields. The measures of skewness indicate that the
majority of industry returns are negatively skewed, meaning
that negative shocks are more common than positive ones.
Furthermore, all industry return series exhibit a kurtosis
significantly larger than three, thereby implying leptokurtic
behavior as compared to the Gaussian distribution. The
Jarque-Bera test statistics corroborate this finding, rejecting
the null hypothesis of normality in all cases at the 1%
level. This result is consistent with that reported for the
Spanish stock market by Miralles et al. (2011). A similar
distributional picture emerges for the 10-year government
bond yield change series. The Augmented Dickey-Fuller
(ADF) and Phillips-Perron (PP) unit root tests indicate that
the series of changes in 10-year bond rates and market and
industry equity returns are all stationary (integrated of order
zero). This finding is in line with that of earlier work on financial
return data (Badillo et al., 2010; Czaja et al., 2009;
Miralles et al., 2012).
Fig. 1 presents the dynamics of the Spanish equity market
index, proxied by the IGBM, and the yield on Spanish
10-year government bonds over the period 1993-2012. The
stock market exhibits a general upward trend during most
of the study period, only interrupted by the Internet bubble
burst in March 2000 and the global financial crisis from
late 2007. On the contrary, the yields on 10-year government
securities display a downward trend up to the start
of the sovereign debt crisis in the euro zone during the
spring of 2010. This figure shows that the interest ratestock
market link over the full sample period is somewhat
unclear. In particular, the Spanish market index and 10-year
government bond yields have moved predominantly in opposite
directions until approximately mid-1998. Since then,
the connection between both va
การแปล กรุณารอสักครู่..
