Upload
others
View
1
Download
0
Embed Size (px)
Citation preview
Understanding common factors in domestic and international
bond spreads
Rodolfo Martell*
August 29, 2005
ABSTRACT
I study the determinants of changes in credit spreads for U.S. dollar denominated domestic and foreign sovereign bonds using fundamentals specified by structural models to separate spreads into their credit and non-credit components. I find that the non-default portions of spreads have two common components that are distinct for each type of debt. Using a vector autoregressive model, I find that domestic spreads are related to the lagged first component of sovereign spreads. Consequently, even though there is no contemporaneous common component in bond spreads, there is an asymmetric relation between common components when focusing on the dynamics of these spreads. I also find that several proxies for liquidity are related to the common components, suggesting a liquidity-based explanation for the common component not identified by previous research in bond spreads. These results show that the cost of debt for firms and countries depends to some extent on shocks that affect all types of debt. * Assistant Professor of Finance, Krannert School of Management, Purdue University. I am grateful to Mike Cooper, David Denis, Diane Denis, Francis Longstaff, Jean Helwege, Andrew Karolyi, John McConnell, Bernadette Minton, Raghu Rau and seminar participants at Ohio State, Purdue, Drexel, Fordham, Queen’s University and the 2005 AFA meetings for comments. I am especially grateful to René Stulz, my dissertation advisor. All errors are my own responsibility.
1
1. Introduction.
In this paper I analyze the determinants of credit spread changes of individual U.S
domestic corporate and foreign sovereign bonds. Credit spreads, computed as the difference
between bond yields and the yield of U.S. Treasuries, are a benchmark measure of credit risk.1
Previous research has focused on one type of bonds at a time, making this paper the first one to
bring together the credit spreads on these two types of debt to study their joint dynamics. This is
important because if the market for dollar-denominated credit-risky bonds is integrated, we can
expect credit and non-credit related shocks to affect all bonds.
Extant literature in spread changes of U.S. domestic corporate bonds identified a common
component unrelated to credit risk (Collin-Dufresne, Goldstein and Martin, 2001; Huang and
Huang, 2003); one goal of this paper is to address the question whether the information present in
the time series cross-section of the non-default component of U.S. dollar foreign sovereign debt
spread changes is related to the domestic corporate component documented in prior research. This
should especially be the case if those common components can be explained by liquidity shocks,
since such shocks are pervasive across markets (Chen, Lesmond, and Wei, 2002; Chordia, Sarkar,
and Subrahmanyam, 2003; Kamara, 1994). To investigate this issue, I study whether common
factors that explain credit spread changes for domestic corporate and foreign sovereign debt after
taking into account fundamentals are related and then analyze the determinants of these common
factors. My results allow me to investigate, among other things, whether contagion is a liquidity-
related phenomenon and to address the question whether contagion is better described as a flight
to quality or a flight to liquidity behavior.
Existing research investigates separately the existence of common components in
changes in credit spreads for domestic credit-risky debt and dollar-denominated sovereign debt.
1 The credit spread is often referred to as yield spread, debt spread or simply spread. These terms are used interchangeably in this paper.
2
Scherer and Avellaneda (2000) identify the existence of two common factors for foreign
sovereign debt spread changes. Westphalen (2003) finds evidence of a common factor for
changes in sovereign debt spreads denominated in several currencies after controlling for country
risk proxies. Research on changes in domestic corporate bond credit spreads by Collin-Dufresne,
Goldstein and Martin (2001) finds one common component after controlling for fundamentals.
The relation between these common components has not been examined in the literature.
I extend the research on common components present in bond spreads by examining
whether the information in the dynamics of U.S. dollar denominated sovereign debt spreads is
associated with the common component found in U.S. corporate bond spreads. If that is the case,
this could suggest that shocks could be transmitted through the non-credit portion of debt spreads,
particularly during heightened periods of volatility spillovers, i.e., contagion. Specifically, I
estimate different models of spread changes for each type of bonds – domestic corporate and
foreign sovereign – because these two groups vary in their source of credit risk. Applying
principal component analysis techniques to each debt type, I extract common factors from the
unexplained portion of credit spread changes from these models and then investigate if the
common factors in U.S. dollar denominated sovereign debt are related to the common factors
present in U.S. corporate debt spread changes using regressions explaining contemporaneous
changes in spreads and using a dynamic model of changes in spreads. I hypothesize that the non-
credit component of spreads is related to liquidity in bond markets and test that idea with liquidity
proxies and obtain encouraging results.
To test my ideas, I construct a new dataset that is comprised of all domestic industrial and
U.S. dollar-denominated sovereign debt. This dataset contains data for 233 non-callable, non-
puttable bonds issued by 37 emerging countries and 3,097 domestic corporate bonds issued by
649 different companies that traded between January 1990 and January 2003. This dataset is
different from those used by earlier studies in at least three ways. First, earlier bond studies that
use Datastream bond data do not include ‘dead’ issues, i.e., bonds that have matured or were
3
retired -- I include them in my sample to avoid a survivorship bias. Second, the Fixed Income
Database used in some other studies has limited coverage of high-yield issues because it covers
mainly investment-grade bonds (Huang and Kong, 2003). I do not have this problem because my
dataset contains data for the complete universe of bonds covered by Datastream.2 Finally, this
dataset covers a longer time period than any previous study.
My results help to discriminate between competing explanations for the common
component previously documented for domestic debt, and also suggest new explanations. I find
strong evidence of the existence of two common factors unrelated to credit risk in debt spread
changes of U.S. denominated foreign sovereign debt and in the debt spread changes of domestic
corporate bonds. While principal component analysis shows no evidence of contemporaneous
correlation between the two domestic and the two sovereign factors, a vector autoregressive
(VAR) model shows that domestic spread changes are related to the lagged sovereign spread first
common component. I find that all four common factors are related to the flows of funds into
bond funds, as measured by the Investment Company Institute (ICI), while three of the factors are
related to the net borrowed reserves from the Federal Reserve, a macroeconomic measure of
liquidity, and to a measure of liquidity in international bond markets that I construct using Brady
and Eurobond issues. My results together suggest that liquidity shocks affect foreign sovereign
bonds first and then are transmitted to domestic issues.
The results I obtain improve our understanding of the determinants of the cost of debt for
foreign countries and for domestic firms. For example, these results suggest that the cost of debt
for foreign countries and domestic firms is not only a function of their own creditworthiness but
also depends on shocks that affect all debt. Further, these results help us understand better the
extent to which the sovereign and domestic corporate bond markets are integrated. In a fully
2 Informal conversations with Datastream’s customer service revealed that several large banks, including Lehman Brothers, were among their providers for bond data. Since Lehman Brothers was the provider for the FISD, I feel confident Datastream’s data includes what is covered in the FISD and has broader coverage of high-yield bonds because of the additional data providers. A comparison testing the integrity and consistency between FISD and Datastream sovereign bond data is available from the author.
4
integrated dollar debt market, we would expect the relation between domestic corporate credit
spreads and sovereign credit spreads to be contemporaneous. My results suggest this is not the
case.
Finally, the lack of a contemporaneous relation between the common components of
domestic corporate credit spread changes and foreign sovereign credit spread changes suggests
that the cost of debt for emerging markets depends mostly on country and emerging market
specific considerations. This is surprising in light of a considerable literature that emphasizes the
impact of developed country developments for capital flows into emerging markets (Calvo,
Leiderman, and Reinhart, 1993; Chuhan, Claessens, and Mamingi, 1998).
This paper proceeds as follows. Section II describes the literature, sample, and variables
used to model credit spread changes for sovereign and domestic bonds. Section III investigates,
using a variety of techniques, the existence and nature of the factors affecting debt spread
changes. Section IV analyzes the dynamics of the common factors and investigates whether they
are related to liquidity and/or demand related variables. Section V concludes.
2. Debt spreads of sovereign and corporate bonds
To examine whether the same common factor is associated with the variation in U.S.
domestic corporate and U.S. dollar denominated sovereign spreads, the unexplained variation in
each spread (i.e. residuals) must be calculated. My choice of variables to compute the credit risk
portion of debt spread changes is based on determinants of bond spread changes specified by
structural models.3
The determinants of sovereign debt spreads have been studied by Eaton and Gersovitz
(1981), Bulow and Rogoff (1989), Edwards (1984), and Hernández-Trillo (1995). Cantor and
Packer (1996) and Eichengreen and Moody (1998) study the determinants of bond spreads at the
3 Another approach is the use of reduced-form models, which are based on an exogenously specified intensity process. See Merrick (2000), Pagès (2001) and Duffie, Pedersen and Singleton (2003) for applications of this method.
5
issue level, and more recently, Scherer and Avellaneda (2000), Joutz and Maxwell (2002) and
Cifarelli and Paladino (2002) study selected series from several emerging markets using principal
component analysis and vector-autoregressions.
On the domestic debt side, the first structural model of risky debt is by Merton (1974),
using an option pricing approach to include systematic and idiosyncratic risk in the calculation of
the value of a put option on the firm’s value.4 Different extensions to this basic model have been
introduced by Black and Cox (1976), Longstaff and Schwartz (1995), Anderson and Sundaresan
(1996), Mella-Barral and Perraudin (1997), Leland (1994), and Leland and Toft (1996) among
many others. Recently, Huang and Huang (2003) calibrate several classes of structural models to
be consistent with the recent history of observed defaults and find that different models could
generate the wide range of credit spreads observed in the recent past. Elton, Gruber, Agrawal and
Mann (2001) try to explain domestic corporate spreads using explanatory factors that include the
probability of default, the loss given default, and the difference in tax regimes. Collin-Dufresne et
al. (2001) try to explain changes in the credit risk portion of domestic corporate spreads using
data on spot rates, reference yield curve slope, firms leverage and volatility, estimates for jumps
in the firm’s value and a proxy for the general business climate. Both papers, using different
approaches, find similar results with their models leaving a large portion of the cross-sectional
time variation of debt spreads unexplained. Collin-Dufresne et al (2001) find that a single
common factor could explain up to 75% of the residual variation, yet they fail to identify what
this common factor is proxying for.
2.1 Data description.
I collect monthly data on all U.S. dollar denominated eurobonds with Datastream
coverage. Datastream’s yields are calculated using average market maker prices provided by the
4 Specifically, Merton’s (1974) model states that a risky zero-coupon bond has the same payoff structure as a risk-free bond plus being short a put option on the firm’s value with a strike price equal to the face value of the debt.
6
International Securities Market Association (ISMA). I eliminate all bonds that were callable
and/or puttable at borrower’s option, that had an early redemption feature and/or were extendible
at the bond holder’s option, and that were not issued by a sovereign entity.5 This leaves the
dataset with 181 live and 52 dead bonds. Also, I eliminate all observations with less than one year
to maturity because, as these bonds approach their maturity date, they are less traded, which in
turn dries up their liquidity and distorts prices and yields.6 After all these adjustments, the final
sample contains 9,275 monthly observations from 233 bonds issued by 37 different countries that
traded between January 1990 and January 2003.
The domestic sample contains all U.S. denominated bonds issued by industrial domestic
firms. Applying the same selection criteria used for the sovereign sample, I end up with 2,493
live and 604 dead bonds issued by 649 different firms during the January 1990 – January 2003
period for a total of 71,831 monthly observations. The domestic sample differs from previous
studies in at least three aspects. First, it covers a larger time period than previous research.
Second, I collect data for the entire universe of bonds issued in U.S. dollars by domestic
industrial firms, not only for those traded by any specific group of investors. Third, the Fixed
Income Database used in earlier studies like Collin-Dufresne et al. (2001) mainly covers
investment grade bonds, and so results obtained for high-yield bonds using that database might
not be representative (Huang and Kong, 2003).
For each bond, I collect the monthly redemption yield (datatype 4 in Datastream) and the
monthly U.S. Treasury yield curve. I compute debt spreads as the difference between the
redemption yield of the sovereign bond and the value of a linear interpolation of the U.S. 5 I am not using Brady bonds in this part of the analysis because their characteristics are inherently different from regular sovereign bonds. The existence of collateral as well as the existence of value recovery rights attached to Brady bonds makes them a class on their own. Further, the tendency is for sovereigns to retire par and discount Brady bonds, so that movements in Brady bond’s prices might be reflecting low volume and thin trading problems and not changes associated with the underlying value of the issuer and the overall liquidity of the market For instance, Mexico’s Ministry of Finance and Public Credit announced on April 7, 2003 that it was calling US$3,839 million of its dollar-denominated Series A and B Brady Par Bonds, which were the last outstanding series of Mexican Brady Bonds denominated in dollars. 6 See Sarig and Warga (1989). This effect is even more pervasive when considering that liquidity was not great in the first place.
7
Treasury yield curve to obtain the yield of a U.S. instrument with identical maturity as the bond
being analyzed.7 I collect the time-series of years to maturity for each bond.
Since all the bonds in the dataset are denominated in U.S. dollars, I examine factors that
affect the U.S. yield curve term structure. Litterman and Scheinkman (1991) showed that the U.S.
yield curve level and slope are important explanatory factors of the term structure, and to proxy
for them I collect monthly annualized yields for the on-the-run two and ten year Treasury notes.8
To capture a country’s distance-to-default, i.e., its ability (and/or willingness) to keep
servicing its debt, I collect monthly exports in U.S. dollars from the IMF’s International Financial
Statistics. To construct a debt-to-exports ratio, I collect debt outstanding and foreign reserves
data9 from the joint BIS-IMF-OECD-World Bank statistics on external debt (Eaton and
Gersovitz, 1981; Bulow and Rogoff, 1989; Krugman, 1985, 1989; Krugman and Rotemberg,
1991; Gibson and Sundaresan, 1999; Westphalen, 2002). I use the Economist Intelligence Unit
(EIU) countrywide index as a measure of country risk for emerging markets. It measures political,
economic policy, economic structure, currency, sovereign debt and banking sector risks. This
index can be used as a guide for the general risk of a country because it assesses risks associated
with investing in the financial markets of those economies as well as risks involved in direct
investment.10 To get a measure of monthly local wealth volatility and local risk, I use the
7 I also collect the monthly U.S. Treasury yield curve using CMT (constant maturity treasuries) to calculate spreads and our results are insensitive to the choice of U.S. benchmark curve. 8 Another reason to collect U.S. interest rates data is that in this framework, if a country’s wealth follows a stochastic process analogue to a firm’s value process, the risk neutral drift will be positively related to the risk-free rate. An increase (decrease) in the risk-free rate should increase (decrease) the country’s wealth over time, making default less (more) likely to happen. Since an upward-sloping yield curve slope is, according to the expectations hypothesis theory of the term structure, predicting higher interest rates in the near future, I expect this slope to have some effect on spreads today. Also, a positively sloped interest rate term structure is perceived as signaling increased economic activity in the near future. 9 All figures are expressed in current U.S. dollars. One shortcoming of this database is that not all series are available on a quarterly basis and there are some gaps in the data, especially in the earlier 1990s. 10 The values are derived from measuring the risk associated with four aspects of the country –political risk, economic risk, economic structure risk and liquidity risk. The overall risk rating is measured on a scale from 1 to 100 where 1 denotes the least risk and 100 the most risk possible. For example, in December 2002, the value of the index was 78 for Argentina, 63 for Brazil and 48 for Mexico. Unfortunately, this index is available only since March 1997.
8
volatility of Datastream local equity indices. For more than half of the countries in the sample
(twenty one), I collect daily data for the local Datastream equity index. For eight additional
countries, I collect daily data from their own local equity indices. For the remaining countries,
Datastream’s world total return index is used. To correct for differences in the scales of the
indices the coefficient of variation (sample standard deviation over sample mean) is computed. I
also collect the available history of Standard & Poor’s (S&P) country ratings from Bloomberg,
and follow Eom, Helwege and Huang (2003) for translating S&P ratings into numerical values,
where a rating of AAA has a value of 1, AA+ a value of 2 and so on.
Table 1 presents the predicted correlation signs between the variables previously
mentioned and debt spreads. Table 2 reports summary statistics for the sovereign sample.
Observations are grouped in five different categories according to their S&P rating. It is evident
from panel A that all groups display a high degree of non-normality. Also, as expected, spreads
increase as we move down in ratings. The mean debt spread in the overall sample is 483 basis
points, the maximum spread is 3939 basis points and the minimum is 1.9 basis points.
Interestingly, the standard deviation also increases as the rating deteriorates. Over the sample
period, the standard deviation ranges from as little as 25.3 to a max of 809 basis points. There is
evidence of extreme movements in each group as the 90% and 10% values are away from the
mean by several standard deviations. Panel B reports the mean values, by group and for the
overall sample, of some country specific variables. Debt-to-reserves, debt-to-exports and political
risk all increase in value as move down in rating to signal a worsening of a country’s situation. As
expected, these variables to have higher values as we move from high to low ratings.
For the domestic sample, I construct measures of leverage and volatility of a firm’s
equity. I expect a negative relation between each of these two variables and debt spreads, since an
increase on any of them would make default more likely. To compute the leverage ratio I collect
book value of debt from COMPUSTAT (items 45 and 51) and the market value of equity from
CRSP. Leverage ratios are then computed, following earlier literature, as:
9
)()(
ValueMarketEquityDebtofValueBookDebtofValueBook
+
To measure exposure of firms to the economic cycle I collect monthly returns for the
S&P 500 index. Table 3 presents the predicted relations between the variables used in the
domestic spreads regressions. Table 4 shows descriptive statistics for the domestic sample.
Although it would be desirable to classify this sample also by rating, Datastream’s coverage of
ratings is sketchy. In light of that problem, I classify the data according to leverage. As can be
seen from the leverage columns of panel A, credit spreads increase as firms become more
levered. Further, the standard deviation of credit spreads also increases with leverage. Data on the
‘No leverage data’ column refers to firms that are either private or are not covered by
COMPUSTAT. The presence of heavy tails in each category is evident from the dispersion
observed in the max, min, 10% and 90% values.
2.2 A model for sovereign spreads.
I estimate the following equation for each bond observation in the sovereign sample:
∆Spreadi,t = Constant + β1*∆Debt to foreign reserves ratioi,t + β2*∆Country risk
measurei,t + β3*∆U.S. Treasury yield curve level,t + β4*∆U.S. Treasury yield curve (1)
slope,t + β5*∆Local volatilityi,t-1 + β6*Local returnt-1 + β7*∆Years to maturityi,t + εi,t
and the following equation for each bond in the domestic sample:
∆Spreadi,t = Constant + β1*∆Leverage ratioi,t + β2*∆Stock return volatilityi,t +
β3*∆U.S. Treasury yield curve levelt + β4*∆U.S. Treasury yield curve slopet + (2)
β5*∆S&P index returnt-1 + β6*∆Years to maturityi,t + εi,t
Following extant research, I estimate regressions on debt spread changes.11 Equations
were estimated using an OLS model with Newey-West adjusted errors.12 Sovereign regressions
11 See Colin-Dufresne et al (2001). Another reason why I use spread changes is that for most series I cannot reject the existence of unit roots.
10
results are presented in Table 5. The model appears to have a good fit, as measured by R-squared
measures, which range from 19% to 30%. For brevity, I discuss only the results for the overall
sample. The debt-to-reserves ratio and the political risk measure both have a positive coefficient
(as expected) and are highly significant. Two lags of the political risk variable were included to
account for autocorrelation in this variable. These variables measure the ability to service debt
and the overall political and economic environment of the issuer. An increase in political risk
would signal higher instability and/or the possibility of expropriation and therefore should be
associated with a higher spread. An increase in the debt-to-reserves ratio could be caused by an
increase in the nominal debt amount or a decrease in international reserves, both of which should
be associated with a higher spread. I also find that the coefficient estimates when using debt-to-
exports in place of debt-to-reserves are not significant, so I do not report them.
The coefficient associated with the U.S. Treasury yield curve level is negative and highly
significant. Previous work had obtained insignificant positive coefficients (Cline and Barnes,
1997; Min, 1998; and Kamin and Von Kleist, 1999), and significant negative coefficients
(Eichengreen and Mody, 1998). One interpretation of these negative coefficients is that, as
interest rates go up, low rated countries find it less convenient to issue debt. Also, most structural
models predict a negative relation because higher interest rates increase the drift of the process
followed by the firm’s (in this case, country’s) value.13 A higher firm (country) value should be
associated with a smaller spread and hence the negative sign.
Interestingly, the coefficient associated with the U.S. Treasury slope term is always
positive and significant. Following the expectations theory of interest rates, a positively sloped
yield curve signals higher future rates, which should be associated with smaller spreads. Two
12 I experiment with several other methodologies. I estimate equation 1 using OLS fixed effects, grouping our sample by bond, by country, and by region. I also estimate FGLS (Feasible Generalized Least Squares), OLS with panel corrected standard errors and OLS with Huber/White standard error correction. All methodologies produce quantitatively and qualitatively similar results; results obtained with other methods are not reported in this paper and are available from the author. 13 See Longstaff and Schwartz (1995).
11
reasons for this effect were previously mentioned. On one hand, we could expect the average
quality of sovereign issuers to increase because low rated countries decide not to issue debt and
this increase in overall quality puts downward pressure on spreads. On the other hand, higher
rates will mechanically increase the distance to default in most Merton-based structural models
which would also lead to a decrease in spreads.
Local volatility is positive and highly significant, as expected. The local stock return has
the expected (negative) sign and is significant. The coefficient on changes of years to maturity is
negative and not significant. I interpret this coefficient as evidence of the existence of a
survivorship bias in which only relatively better countries make it to issue longer term debt, as
hypothesized by Helwege and Turner (1999) for the domestic case. It may be the case that
investors think that in the case of a default, short term maturities are more risky than long term
maturities since countries will usually default first on issues with closer maturities, making short
term issues riskier. The lack of consistent cross-default clauses in some countries allows them to
default or re-schedule debt payments selectively. Finally, for a country facing financial
difficulties, a longer time horizon will provide the necessary time and maneuvering room to enact
reforms and measures that will allow the country to return to fiscal stability, effectively making
longer term debt less risky.
The results from the domestic regressions are reported on Table 6. The model performs
better in the highly leveraged group, as evidenced by the higher R-squared value. This result is
consistent with previous studies which find that structural models perform better for longer
maturity, lower rated sub-groups. For brevity, I will only go over the overall sample results. The
negative and insignificant coefficient of changes in years to maturity is consistent with previous
results of Helwege and Turner (1999) and with the basic Merton (1974) model predictions, as
described in Stulz (2003), for conservative levels of debt. The coefficients of lagged leverage and
stock return volatility –contemporaneous and lagged- are positive and strongly significant. The
sign of the U.S. yield curve level is also as expected and similar to the results obtained by earlier
12
studies. In contrast to the sovereign sample, nothing conclusive can be said about the sign and
significance of the coefficient estimated for changes of the U.S. Treasury slope, which is
significantly positive in every specification. Finally, the S&P index return is significant and with
a negative sign, just as predicted by the theory.
3. Analyzing the common factor.
As previously mentioned, one goal of this paper is to investigate whether the common
factor identified in domestic credit spread changes also is present in sovereign debt spread
changes. In this section, I establish the existence of a common factor in both the residuals from
the regressions on sovereign and domestic debt spread changes. In order to investigate whether
common factors are present in the unexplained variation in spreads, I use principal components
analysis. This is a statistical technique for data reduction whose objective is to find unit-length
linear combinations of the original variables that capture the maximum variance. I apply principal
component analysis to the residuals obtained from the regressions discussed in previous sections
to verify whether the unexplained variation is truly noise or whether there is evidence of a
common factor driving this unexplained portion of the variance of credit spread changes.14
The first problem faced when applying principal components analysis is how to organize
unbalanced panels in the most efficient form. Boivin and Ng (2003) show that more data is not
always better when conducting this type of factor analysis. In fact, in their forecast exercise they
show that factors extracted from as few as 40 variables could be more informative than factors
extracted from all 147 series in their setup. Basically, their result obtains because of large cross-
correlation in errors and of small variability of the common components. Unfortunately, there is
no guide as to what data should be included in a principal component analysis or what is the
optimal number of series to include in this exercise. Recent work from Scherer and Avellaneda
14 The serial correlation of the residuals from the regressions for sovereign yield changes is -0.1513 with p value of 0.2290 and 0.0068 with p value of 0.9345 for domestic yield changes.
13
(2000) applies principal component analysis to eight variables only, effectively using one or two
bonds per country in their study. While this might be a solution for the sovereign case where it is
easier to identify benchmark bonds for each issuer, this is not feasible in the domestic sample.
This sample has bonds from 649 different firms, and in many cases there are tens of bonds
outstanding from a given firm. Also, applying principal component analysis to all the bonds in the
domestic sample is not a good idea, since that would most certainly only increase the amount of
statistical noise while adding very little or no new information at all.
I follow the approach implemented by Collin-Dufresne et al. (2001) and create groups or
‘bins’ of data to efficiently summarize the information content of the residuals. I divide each
sample (sovereign and domestic) into three maturity categories and three leverage (debt-to-
reserves, in the sovereign case) categories, creating a total of nine ‘bins’ in each sample. Then,
each observation is assigned to a bin. I estimate again equations (1) (for the sovereign bins) and
(2) (for the domestic bins), compute the residuals, and calculate averages across residuals for each
bin. Table 7 shows the correlation structure for the average domestic residuals (panel A), the
average sovereign residuals (panel B) and for all averages –domestic and sovereign (panel C).
The average correlation for the sovereign sample is 0.75, and 0.87 for the domestic sample. To
investigate whether the relatively high correlations found in panel A and C are caused by a
common component, I conduct principal component analysis.
Table 8 shows the results of applying this eigenvalue decomposition to the bins
constructed earlier. Panel A shows strong evidence of the existence of a common factor in
sovereign spreads. The first common factor explains 76.09% of the variation, as shown by the
proportion of the first eigenvalue. The second common component explains the remaining
variance that is orthogonal to the first common component. It is difficult to interpret the second
component because its eigenvalue is well below the value of the first eigenvalue and is much
closer to the third eigenvalue. However, if this is to be interpreted as evidence of a second
common factor, it would explain an additional 20.37%. According to Scherer and Avellaneda
14
(2000), a number between 65% and 80% for the first common component would be considered to
be indicative of strong co-movement characterized by a high correlation in the spread changes.
My results are consistent with their result – obtained with spreads computed from Brady issues
for selected countries – in which they found evidence of two common factors driving most of the
variation in spread changes.
Panel B of Table 8 also shows strong evidence of the existence of a common factor to all
domestic spreads. The first common component explains 86.23% of the variance. There is weak
evidence on the existence of a second component which explains an additional 8.53% of the
variance. The existence of a first common factor that explains such a large portion of the variance
is consistent with previous research, e.g., Collin-Dufresne et al. (2001). The existence of a second
common factor has not been documented for domestic debt before, but this could be due to the
fact that I am using a larger dataset and that I am looking at a longer time period that earlier
studies.
Finally, panel C has the results of looking at the common components of both groups of
bonds, sovereign and domestic. Interestingly, I find no evidence suggestive of the existence of a
common factor to both groups of bonds. The first common factor explains 42.06% of the residual
variance of spread changes, while the second factor explains an additional 33.12%. As mentioned
before, Scherer and Avellaneda (2000) consider a value of 65% for the first common component
as the lower boundary for a weak coupling, or correlation, between spread changes. The result I
obtain is puzzling because if the market for dollar-denominated credit-risky bonds is integrated,
and if the common components I find can be explained by liquidity shocks, then such shocks
should be pervasive across markets (Chen, Lesmond, and Wei, 2002; Chordia, Sarkar, and
Subrahmanyam, 2003; Kamara, 1994). According to panel C, this is not what is happening.
In order to shed more light on the issue of whether the common factor identified in both
samples is indeed the same in both groups, I extract the first and second common components of
each sample to compare them. These common factors are plotted in figure 1. The interpretation of
15
the units in the y-axis is as follows. To compute the principal components, I analyzed the
correlation matrix. This is equivalent to all the variables to having mean 0 and standard deviation
1. Thus, the common factors are expressed in terms of these standardized variables. Loosely
speaking, the units on the y-axis in both figures can be interpreted as percentage points.
The pattern seems to suggest a lead-lag relation between the first factor from the
domestic sample and the first factor extracted from the sovereign sample, with the latter leading
the former. Figure 2 shows the second common component extracted from both samples. The
figure seems to suggest a weak contemporaneous relation. I investigate these issues further in the
next sections.
3.1 Explanatory power of the extracted components.
In this section, I examine whether these common factors have explanatory power over the
cross-section of debt spreads for the other type of debt. I estimate again equations (1) and (2)
including in each equation the two common components extracted from the other group, i.e., I
include the factors extracted from the sovereign sample in the domestic sample regression and
vice versa. Results are shown on Table 9.
First, I discuss the sovereign regression when the domestic common components are
included. Looking at rating categories, the explanatory power of the equation for the lower rated
group (B- to C) increases, as measured by the increases of R-squared statistic from 22% to 31%.
This sub-sample is the one with the least number of observations. There is, however, no gain in
explanatory power in the overall sample. Further, in every case only the contemporaneous value
of the first component from the domestic sample is significant. In the case of the second
component extracted from the domestic sample, only the lagged value of the second factor is
significant for all sub-groups.
The domestic sample has strikingly different results. In this case, I included in the
domestic spread changes equation both common components extracted from the sovereign
16
sample. The explanatory power of the overall equation is increased almost by 30%, from an R-
square value of 0.09 to 0.12. The only significant common component coefficient is the
contemporaneous effect of the first common component from the sovereign sample. The
explanatory power in most domestic sub-samples increases by a similar percentage as the R-
squared value of the overall sample. Overall, I interpret these results as evidence of the existence
of a relation between the first common component extracted from the sovereign spread changes
and domestic debt spread changes. The dynamics of the relation between the common
components extracted from each type of debt is investigated in the next section.
4. The information content of the common factors.
The principal component analysis conducted so far neither provides information on the
dynamics of the factors identified nor provides an economic interpretation of them. In this section
I investigate the contemporaneous and inter-temporal relation between factors and also
investigate whether these factors might be capturing liquidity and/or supply/demand shocks.
4.1 Lead-lag relations.
The picture shown in Figure 1 suggests the possibility of an intertemporal relation
between the first factor extracted from the sovereign sample and the first factor extracted from the
domestic sample. Using a vector-autoregression approach, I investigate the possibility of one of
these markets acting as an early signal for potential problems that can affect the bond market in
general. Previous work like Joutz and Maxwell (2002) and Cifarelli and Paladino (2002) have
applied VAR procedures in a credit spread framework to study the relation between credit spreads
from different countries. More recently, Longstaff, Mithal and Neis (2003) applied a VAR
framework to study the relation between bond and credit derivatives markets. To explore the
lead-lag relation between the sovereign and the domestic factors, the following simple vector-
autoregression specification is used:
17
11
111
11 εδγβ ∑∑=
−=
− ++++=k
jtjtj
k
jjtjt XFacDomFacSovaFacSov
21
221
22 εδγβ ∑∑=
−=
− ++++=k
jtjtj
k
jjtjt XFacDomFacSovaFacDom
Table 10 shows the results for the simple case when k is equal to two. Both the Akaike
Information and Schwartz criteria suggest that a VAR system of two lags is warranted by the
data. I first run the VAR model without exogenous variables to have an initial idea of the lead-lag
structure. Since the basic lead-lag relation is unchanged when the exogenous variables are
included, and in the interest of brevity, I only report the R-squared value for each equation
without exogenous variables.
The exogenous variables included in the VAR are chosen to capture liquidity and
supply/demand effects. Most previous studies dealing with credit spreads specifically abstain
from liquidity effects because of the lack of consensus on how to measure and model liquidity
premium affecting spreads (Chen, Lesmond and Wei, 2003). Longstaff, Mithal and Neis (2003)
study the consistency of the price of credit risk between the bond and derivatives markets. They
find that the implied cost of credit is higher in the bond market than in the credit derivative
market, and advance a possible explanation for this based on the existence of a liquidity
component in debt spreads. Their measure for this liquidity premium is the difference between
the price of credit risk in the bond and credit derivative markets.15
Since all the bonds in the sample are denominated in U.S. dollars and they all trade in
U.S. financial markets, I am interested in variables that measure the overall liquidity in these
markets. I use three general measures of liquidity. The first proxy is the difference in yield
15 Collin-Dufresne et al. (2001) point that Chakravarty and Sarkar (1999), Hotchkiss and Ronen (1999) and Schultz (1999) found evidence of the existence of relatively high transaction costs and low volume in bond markets, and therefore, Collin-Dufresne et al. (2001) interpret these results as evidence of a liquidity premium.
18
between the on-the-run16 thirty year U.S. Treasury bond and the most recent off-the-run bond is
computed. Off-the-run bonds are bonds that whilst not being the most recently issued in a certain
maturity range, are very similar to the on-the-run issue in all respects. Therefore, any differences
in prices –and therefore in yields- is usually considered to be due to liquidity. As liquidity dries
up, this difference is expected to decrease.
The second proxy for general liquidity in the market is the net borrowed reserves from
the Federal Reserve,17 which is considered a measure of the monetary stance. A loose monetary
policy usually implies an increase in liquidity via the decrease of credit constraints. Harvey and
Huang (2002) showed that the Federal Reserve, through its ability to change the money supply,
impacts the trading of bonds and currencies. Following Chordia, Sarkar and Subrahmanyan
(2003) I define net borrowed reserves as total borrowing minus extended credit minus excess
reserves, divided by total reserves. Since borrowed reserves represent the amount that banks are
short to satisfy the Fed’s requirements, a lower value of this measure indicates looser monetary
conditions.
A final proxy for liquidity that I use is the difference in yield between a Brady bond and a
regular Eurobond issue for a given country. A popular hedge fund strategy in the 1990s, it was
know as the Brady-bond/Euro-bond puzzle. The idea was a simple one and involved taking
offsetting positions in dollar-denominated Brady bonds and dollar-denominated sovereign bonds
traded in Euro market. I compute a natural proxy for liquidity by taking the difference between
the implied sovereign yield of Euros and the sovereign yield of Bradys (stripped yield) and then
averaging across countries. The countries for which this measure was available are Argentina,
Brazil, Bulgaria, Dominican Republic, Ecuador, Mexico, Nigeria, Peru, Philippines, Poland and
Venezuela.
16 An on-the-run bond is the most recently issued (and typically the most liquid) government bond with a given maturity. 17 See Chordia, Sarkar and Subrahmanyam (2003).
19
To capture possible supply/demand shocks I collect data from the Investment Company
Institute (ICI) on the monthly flows into mutual funds. ICI’s statistics are collected from
approximately 8,300 mutual funds, and are divided in flows into equity funds and bond funds.
These measures could potentially capture changes in investor’s attitudes towards risk or any other
supply/demand shocks unrelated to overall market liquidity.18
Table 10 reports the results of the VAR model with the exogenous variables. I start
discussing the first common domestic factor equation of the VAR. The coefficient associated with
the second lag of the first sovereign factor is significant, suggesting that the first sovereign factor
leads the first domestic factor. The coefficients of flows to stocks and flows to funds are negative
and significant. This equation is the one with the smallest gain in R-squared when exogenous
variables are included, going from 36% to 44%.
The first sovereign factor seems to be slightly mean reverting as suggested by the
significant coefficient for its first lag. There is some feedback from the first lag of the second
sovereign factor, and most importantly, the coefficients for the flows bond funds, the net reserves
from the Fed and the liquidity spread from brady and euro bonds are highly significant. This
equation also has the highest increase in explanatory power when the exogenous variables are
inclided, since the R-squared increases from 5% to 60%.
The subpanel with the results for the second domestic factor reports significant
coefficients for the second lag of the first sovereign factor (marginally) as well as for its own
second lag. Flows into bond funds, net reserves and the brady bond spread variables have
significant coefficients, and raise the R2 for this equation from 20% (no exogenous variables) to
56% (with all exogenous variables).
The second sovereign factor equation has the highest absolute R-squared value, 69%. The
increase in R2 gained by the inclusion of the exogenous variables is as big as the one experienced
18 There is a potential endogeneity problem in our variables, since for instance, a change in the Fed’s stance could make certain markets more attractive and influence the flows into those markets. No effort is made at this time to address this issue.
20
by the first sovereign factor. Second lag coefficients for all other factors except the first domestic
have significant coefficients, as well as all exogenous variables except the net borrowed reserves
from the Fed.
The picture that emerges from the VAR system is that my choice of exogenous variables
is capturing a large portion of the time series variation of the common factors, domestic and
sovereign. The only variable that is significantly and negatively related to all four common
factors is the flows to bond funds. This is expected because the factors were extracted from bond
spreads. For all four factors, the sign of the coefficients suggests that as the flows into bond funds
increase, the non-credit related component of debt spreads is reduced, which is consistent with the
idea that the non-credit component of spreads reflects mostly changes in liquidity. The flows into
stock funds have a significant positive coefficient in only two equations, consistent with the idea
that bond liquidity might suffer if investors rebalance from bond funds into stock funds. The
magnitude of these coefficients, however, is at least one order of magnitude smaller than those
coefficients associated with the flows into bond funds. The coefficient associated to net borrowed
reserves is also consistent with what I expected if indeed it proxies for macro liquidity. Recall
that a lower value of this variable is associated with a looser monetary policy. The positive
coefficients imply that as monetary policy becomes tighter, the liquidity –non-credit- related
component of spreads also increases. Finally, the liquidity proxy I constructed from Brady and
euro issues is significant for all factors except the first domestic one. The largest coefficient
(negative) is the one associated with the first sovereign factor. Because of the way the Brady
spread variable is constructed, this coefficient is what I expected. Recall that this variable is
constructed as the difference between Euro and Brady yields, averaged across countries. A
smaller (more negative) value means less liquidity, because either the Brady issue was trading at
a discount (higher yield) or the Euro bond was trading at a premium (lower yield). Since I am
using changes in all regressions, a positive value for this variable means more liquidity which in
turn should be reflected in a smaller non-credit –hopefully liquidity- portion of debt spreads. The
21
effect of the Brady spread variable is larger on the first sovereign factor, which, as shown earlier,
leads the first domestic factor. The Brady spread has positive significant coefficients in both
second factors, which is somewhat puzzling, although the smaller value of the coefficients
alleviates some of the concerns raised by that positive value.
Overall, these results make me confident that the variables I chose to proxy for liquidity
do a good job. With the exception of net borrowed reserves (which Chordia, et al (2003) also find
to be the least informative in their setup) all other variables display significant coefficients with
the expected signs. All this evidence suggests that liquidity is driving most of the time series
variation in the non-default related portion of debt spreads.
The pattern displayed by the first common factors in Figure 1 raises the concern that my
VAR results could be driven by the large spike observed around October 1998. As a robustness
check, I re-estimate the system excluding the observations for September, October and November
1998.19 As expected, the significant loading of the second lag of the first sovereign common
component onto the first domestic common component decreases but does not disappear.
Interestingly, the coefficient associated with the second lag of the second sovereign factor now
becomes significant, whereas it was not significant before. The explanatory power of the
exogenous variables and the overall R-squared values of the system remain unchanged. The
evidence suggest that the asymmetric relation observed between the first common factor from the
domestic sample and the common factors from the sovereign sample remains even after
excluding the large shock of October 1998 from the sample.
5. Conclusions
In this paper I regress debt spreads of corporate bonds and of dollar denominated sovereigns
on credit-related variables, and interpret the residual as the non-default portion of spreads. I then
19 Results are not reported but are available upon request.
22
proceed to examine the extent to which the non-default component of corporates and sovereigns
are related.
My key findings are as follows. First, I find no evidence of a contemporaneous relation
between the non-default portion of sovereign and corporate debt spreads. I do find that the
domestic non-default spread correlates with lagged sovereign non-default spread, and that both
non-default spreads are linked to macro liquidity factors such as flows into bond funds, the net
borrowed reserves from the Federal Reserve and a proxy of liquidity in sovereign debt markets
constructed from Brady and Euro bonds. My results also indicate that the cost of debt for
emerging markets depends mostly on country and emerging-market specific considerations. This
is surprising because previous literature (Calvo, et al, 1993) emphasizes the impact of developed
country developments for capital flows into emerging markets.
I also contribute to the literature by showing that current structural models of debt
spreads can be improved if liquidity variables are incorporated. To the extent that investors
depend on these models to hedge the credit risk of their bond positions, they can benefit from a
better understanding of the determinants of credit spreads changes. My research also shows that,
after taking into account the dynamics of the common components in credit spreads across debt
types, the cost of debt for firms and countries depends to some extent on shocks that affect all
types of debt.
There are issues left for future research. The asymmetric lead–lag relation between the
first domestic common factor and the first sovereign factor needs to be investigated further, for
instance, if differences in liquidity across debt issues and/or infrequent trading help explain that
relation. An alternate approach to measuring the size of the default related component may be to
consider using credit default swaps (CDS) levels as a proxy for the default related component
(Longstaff, Mithal, and Neis (2004), Berndt, Douglas, Duffie, Ferguson, and Schranz (2004),
Blanco, Brennan, and Marsh (2003), Pan and Singleton (2004)). It would be interesting to analyze
23
the lead-lag relation using bond and CDS data to investigate if there is lead-lag in the default
component too, as well as the implications this behavior would have on information flows.
24
References
Anderson, Ron and Suresh Sundaresan, 1996, Design and Valuation of Debt Contracts, Review of Financial Studies, Vol. 9, No. 1, pp. 37-68.
Black, Fisher and John C. Cox, 1976, Valuing Corporate Securities: Some Effects of Bond Indenture Provisions, Journal of Finance, Vol. 31, No. 2, pp. 351-367.
Bodie, Zvi, Alex Kane and Alan J. Marcus, 1999, Investments, Irwin-McGraw-Hill, Fourth Edition.
Boivin, Jean and Serena Ng, 2003, Are More Data Always Better for Factor Analysis?, Working paper, Columbia University.
Bulow, Jeremy and Ken Rogoff, 1989, A Constant Recontracting Model of Sovereign Debt, Journal of Political Economy, Vol. 97, No. 1, pp. 155-178.
Calvo, Guillermo, Leonardo Leiderman, and Carmen Reinhart, 1993, Capital inflows and the real exchange rate appreciation in Latin America: The role of external factors, IMF Staff Papers.
Cantor, Richard and Frank Packer, 1996, Determinants and Impact of Sovereign Credit Ratings, Federal Reserve Bank of New York Review, October.
Chakravarty, Sugato and Asani Sarkar, 1999, Liquidity in U.S. fixed income markets: A comparison of the bid-ask spread in corporate, government and municipal bond markets, Working paper, Federal Reserve Bank of New York.
Chen, Long, David Lesmond, and Jason Z. Wei, 2003, Bond Liquidity Estimation and the Liquidity Effect in Yield Spreads, Working Paper.
Chordia, Tarun, Asani Sarkar, and Avanidhar Subrahmanyam, 2003, An Empirical Analysis of Stock and Bond Market Liquidity, Working paper, Federal Reserve Bank of New York.
Chuhan, Punam, Stijn Claessens, and Nlandu Mamingi, 1998, Equity and bond flows to Latin America and Asia: the role of global and country factors, Journal of Development Economics 55, 439-463.
Cifarelli, Giulio and Giovanna Paladino, 2002, An Empirical Analysis of the Co-Movement among Spreads on Emerging-Market Debt, Working Paper.
Claessens, Stijn and George Pennachi, 1996, Estimating the Likelihood of Mexican Default from the Mexican Prices of Brady Bonds, Journal of Financial and Quantitative Analysis, Vol. 31, No. 3, pp. 109-126.
Cline and Barnes, 1977,
Collin-Dufresne, Pierre, Robert S. Goldstein, and J. Spencer Martin, 2001, The Determinants of Credit Spread Changes, The Journal of Finance, Vol. 56, No. 6, pp. 2177 – 2205.
25
Duffie, Darell, Lasse H. Pedersen, and Kenneth J. Singleton, 2003, Modeling Sovereign Yield Spreads: A Case Study of Russian Debt, Journal of Finance, Vol. 58, No. 1, pp 119 – 159.
Duffie, Darrell and Kenneth J. Singleton, 1997, An econometric model of the term structure of interest rate swap yields, Journal of Finance, Vol. 52, pp. 1287 – 1381.
Duffie, Darrell and Kenneth J. Singleton, 1999, Modeling term structures of Defaultable Bonds, Review of Financial Studies, Vol. 12, pp. 687-720.
Eaton, Jonathan and Mark Gersovitz, 1981, Debt with Potential Repudiation: Theoretical and Empirical Analysis , Review of Economic Studies, Vol. 48, No. 2, pp. 289- 309.
Edwards, Sebastian, 1984, LDC Foreign Borrowing and Default Risk: An Empirical Investigation, 1976-80, American Economic Review, Vol. 74, No. 4, pp. 726-734.
Eichengreen, Barry and Ashoka Moody, 1998, What Explains Changing Spreads on Emerging-Market Debt: Fundamentals or Market Sentiment?, NBER Working Paper No. w6408.
Elton, Edwin, Martin Gruber, Deepak Agrawal, and Christopher Mann, 2001, Explaining the Rate Spread on Corporate Bonds, Journal of Finance, Vol. 56, No. 1, pp. 247 – 277.
Eom, Young Ho, Jean Helwege, and Jay Huang, 2003, Structural Models of Corporate Bond Pricing: An Empirical Analysis", Review of Financial Studies, forthcoming.
Gibson, Rajna and M. Suresh Sundaresan, 2001, A Model of Sovereign Borrowing and Sovereign Yield Spreads, Research Paper EFA 0044, PaineWebber Working Paper Series at Columbia University.
Harvey, Campbell and Roger Huang, 2002, The Impact of Federal Reserve Bank's Open Market Operations, Journal of Financial Markets, Vol. 5, No. 2, pp. 223-257.
Helwege, Jean and Christopher M. Turner, 1999, The Slope of the Credit Yield Curve for Speculative-Grade Issuers, Journal of Finance, Vol. 54, No. 5, pp. 1869 – 1884.
Hernández-Trillo, Fausto, 1995, A Model-Based Estimation of the Probability of Default in Sovereing Loan Markets, Journal of Development Economics, Vol. 46, no. 1, pp 163-179 .
Hotchkiss, Edith and Tavy Ronen, 1999, The informational efficiency of the corporate bond market: An intraday analysis, Working paper, Boston College.
Houweling, Patrick, Albert Mentink, and Ton Vorst, 2002, Is Liquidity Reflected in Bond Yields? Evidence from the Euro Corporate Bond Market, Working paper, Erasmus University.
Huang, Jing-Zhi and Weipeng Kong, 2003, Explaining Credit Spread Changes: Some New Evidence from Option-Adjusted Spreads of Bond Indices, Working Paper, Penn State University.
26
Huang Jing-Zhi and Ming Huang, 2003, How much of the Corporate-Treasury Yield Spread is Due to Credit Risk?, Working Paper.
Joutz, Frederik and William Maxwell, 2002, Modeling the yields on noninvestment grade bond indexes, credit risk and macroeconomic factors, International Review of Financial Analysis, Vol. 11, pp. 345 – 374.
Kamara, Avraham, 1994, Liquidity, Taxes and Short-Term Treasury Yields, Journal of Financial and Quantitative Analysis, Vol. 29, No. 3, pp. 403-417.
Kamin, Steven and Karsten von Kleist, 1999, The Evolution and Determinants of Emerging Market Credit Spreads in the 1990s, Board of Governors of the Federal Reserve System, International Finance Discussion Papers, No. 653.
Krugman, Paul, 1985, International Debt Strategies in an Uncertain World, in J. Cuddington and R. Smith, eds., The World Debt Problem, World Bank.
Krugman, Paul, 1989, Private Capital Flows to Problem Debtors, in J. Sachs, ed., The International Debt Problem, University of Chicago Press.
Krugman, Paul and Julio Rotemberg, 1991, Speculative Attacks on Target Zones, in P. Krugman and M. Miller, eds., Exchange Rate Target and Currency Bands, Oxford University Press.
Lando, David, 1998, On Cox Processes and Credit Risky Securities, Review of Derivatives Research, Vol. 2, pp. 99-120.
Leland, Hayne E., 1994, Risky Debt, Bond Covenants and Optimal Capital Structure, Journal of Finance, Vol. 49, No. 4, pp. 1213-1252.
Leland, Hayne E. and K. B. Toft, 1996, Optimal capital structure, endogenous bankruptcy, and the term structure of credit spreads, Journal of Finance, Vol. 51, No. 3, pp. 987-1019.
Litterman, Robert and Jose Scheinkman, 1991, Common factors affecting bond returns, Journal of Fixed Income, Vol. 1, pp. 54-61.
Longstaff, Francis and Eduardo Schwartz, 1995, A simple approach to valuing risky fixed and floating rate debt, Journal of Finance, Vol. 50, No. 3, pp. 789-819.
Longstaff, Francis, Sanjay Mithal, and Eric Neis, 2003, The credit default swap market: Is credit protection priced correctly?, Working paper, UCLA.
Mella-Barral and Perraudin, 1997,
Merrick, John J. Jr., 2000, Crisis Dynamics of Implied Default Recovery Rations: Evidence from Russia and Argentina, Journal of Banking and Finance, Vol. 25, No. 10, pp. 1921-1939.
Merton, Robert, 1974, On the pricing of corporate debt: The risk structure of interest rates, Journal of Finance, Vol. 29, No. 2, pp. 449-469.
27
Ming, Hong G., 1998, The determinants of Emerging Market bond spread: Do economic fundamentals matter?, Policy Research Working Paper 1899, The World Bank.
Pagès, Henri, 2001, Can liquidity risk be subsumed in credit risk? A case study from Brady bond prices, Bank for International Settlements, BIS working papers, No. 101.
Sarig and Warga, 1989,
Scherer, Paul K. and Marco Avellaneda, 2000, All for One ... One for All?: A Principal Component Analysis of Latin American Brady Bond Debt from 1994 to 2000, Working paper, New York University.
Schultz, Paul, 2001, Corporate Bond Trading: A Peek Behind the Curtain, Journal of Finance, Vol. 56, No. 2, pp. 677-698.
Stulz, René M., 2003, Risk Management and Derivatives, Thomson-Southwestern, First Edition.
Westphalen, Michael, 2002, Valuation of Sovereign Debt with Strategic Defaulting and Rescheduling, Research Paper 43, FAME Research Paper Series.
Westphalen, Michael, 2003, The Determinants of Sovereign Bond Credit Spread Changes, Working Paper, FAME.
28
Table 1. Expected signs on explanatory variables for sovereign sample.
Variables Expected sign Rationale
Life to maturity Uncertain Stulz (2003) explains how the relation between time to maturity and credit spreads depends on the relative size of debt and firm value. Helwege and Turner (1999) establish that due to survivorship bias only relatively better rated countries issue longer-term debt. Also, investors might perceive shorter-term sovereign bonds as having higher probability of default and therefore higher expected losses. This belief is reinforced by the fact that some countries' debt does not incorporate cross-default clauses, making easier for countries facing financial distress default first on issues with closer maturities. Longer-term bonds are, in this setting, perceived as safer since countries could have time to implement reforms that bring them out of financial distress.
Debt to foreign reserves ratio and Debt to exports ratio
Positive A higher ratio of any of these two measures implies a smaller distance-to-default. So, larger values of them should be associated with higher spreads.
Country risk measure Positive This measure has a higher value for countries that are perceived to have higher political risks, for instance, higher expropriation risk. The larger the value of this variable, the higher the debt spread.
Local stock market volatility
Positive This variable is an imperfect proxy of a country’s wealth volatility. Still, we expect a positive relation since more volatility makes default more likely.
U.S. Treasury yield curve level
Negative Assuming that the country’s wealth follows a risk-neutral drift, higher rates should be associated with higher drifts which in turn should reduce debt spreads. Also, Stulz (2003) shows how debt value decreases with maturity. This reduced the probability of default, and ergo, spreads.
U.S. Treasury yield curve slope
Negative We assume a positive slope to signal higher future interest rates. The previous arguments then suggest a negative relation between spreads and the interest rate slope.
World return Negative A world index return is included as a proxy of the world economic climate or business cycle. On average, we would expect smaller spreads when the world as a whole is doing well.
29
Table 2. Summary statistics for sovereign sample. Panel A: Descriptive Statistics for Spreads Credit Spread (%) All sample AAA to A+ A to BBB- BB+ to B B- to C No rating Mean 4.834 0.488 2.098 4.971 12.365 4.478 Std. Dev. 4.555 0.253 1.369 3.261 8.090 3.860 Skewness 3.195 2.263 2.170 2.451 1.599 3.503 Kurtosis 18.001 15.840 12.672 14.145 5.063 18.620 Max 39.390 2.204 15.250 33.181 39.390 37.455 90% 39.173 1.332 10.142 30.967 39.173 23.681 10% 0.040 0.076 0.068 0.094 1.408 1.034 Min 0.019 0.038 0.019 0.046 1.241 1.019 No. of observations 9275 165 2213 5450 829 618 Panel B: Means for Selected Country Variables Mean All sample AAA to A+ A to BBB- BB+ to B B- to C No rating Debt-to-reserves 3.919 0.959 1.742 4.733 4.746 1.997 Debt-to-exports 40.603 3.349 12.330 52.654 39.359 8.207 Political risk 51.260 21.330 40.480 53.546 66.028 62.255 U.S. Treasury yield level (%) 5.285 U.S. Treasury yield slope 1.326 Panel C: Other data Number of bonds 233 Number of countries 37
The sample includes all non-callable, non-puttable sovereign bonds in U.S. dollars that traded between January 1990 and January 2003. All data is from Datastream. The spread over U.S. Treasuries is computed as the difference between the redemption yield of the sovereign bond and the value of a linear interpolation of the U.S. Treasury yield curve to obtain the yield of a U.S. instrument with identical maturity. Debt/Reserves is computed using all outstanding foreign debt (bank loans, Brady bonds and Eurobonds) divided by the total number of international reserves in current U.S. dollars. Political risk is the value of The Economist Intelligence Unit's country index. Local stock market volatility is the standard volatility computed each month from daily stock market returns in U.S. dollars. The U.S. Treasury yield level is the yield of the 10 year U.S. Treasury note. The U.S. Treasury slope is computed as the difference between the yield of the 10 year and the 2 year U.S. Treasury notes. The world stock return is the log return of Datastream's world total return index.
30
Table 3. Expected signs on explanatory variables for domestic sample.
Variables Expected sign Rationale
Life to maturity Uncertain Stulz (2003) explains how the relation between time to maturity and credit spreads depends on the relative size of debt and firm value. Helwege and Turner (1999) establish that due to survivorship bias only relatively better rated countries issue longer-term debt. Also, investors might perceive shorter-term sovereign bonds as having higher probability of default and therefore higher expected losses. This belief is reinforced by the fact that some countries' debt do not incorporate cross-default clauses, making easier for countries facing financial distress default first on issues with closer maturities. Longer-term bonds are, in this setting, perceived as safer since countries could have time to implement reforms that bring them out of financial distress.
Leverage and Equity return volatility
Positive A higher leverage ratio increases the probability of a firm facing financial distress. This should increase spreads. Also, from a contingent claims approach, equity return volatility can proxy for firm's value volatility. A higher volatility increases the chance of the firm's value process to cross the threshold at which a firm defaults on its debt.
U.S. Treasury yield curve level
Negative Assuming that the firm's value follows a risk-neutral drift, higher rates should be associated with higher drifts which in turn should reduce debt spreads. Also, Stulz (2003) shows how debt value decreases with maturity. This reduced the probability of default, and ergo, spreads.
U.S. Treasury yield curve slope
Negative We assume a positive slope to signal higher future interest rates. The previous arguments then suggest a negative relation between spreads and the interest rate slope.
S&P 500 return Negative As the economic environment improves, measured by the S&P return, we expect firms to do better and therefore to reduce the probability of defaulting on their debt.
31
Table 4. Summary statistics for domestic sample. Panel A: Descriptive Statistics for Debt Spreads
Overall Leverage Class
No leverage
Credit Spread (%) sample Low Medium High data Mean 2.393 1.359 1.717 3.263 2.633 Std. Dev. 2.779 1.187 1.319 3.566 3.017 Skewness 4.543 5.621 5.210 3.947 4.007 Kurtosis 31.661 77.578 70.700 22.628 24.890 Max 29.989 26.846 29.636 29.847 29.989 90% 29.849 22.659 26.812 29.812 29.849 10% 0.0012 0.0031 0.0125 0.0279 0.0021 Min 0.0007 0.0012 0.0007 0.0062 0.0010 No. of observations 71831 11061 11247 11332 38191 Panel B: Means for Selected Variables Mean Leverage 0.343 Std. Dev. 0.023 U.S. Treasury yield level (%) 5.285 U.S. Treasury yield slope 1.326 Panel C: Other data Number of bonds 2,930 Number of countries 649
The sample includes all non-callable, non-puttable domestic bonds issued by industrial firms in U.S. dollars that traded between January 1990 and January 2003. All data is from Datastream. The spread over U.S. Treasuries is computed as the difference between the redemption yield of the sovereign bond and the value of a linear interpolation of the U.S. Treasury yield curve to obtain the yield of a U.S. instrument with identical maturity. Leverage is computed as the ratio of book value of debt divided by the sum of book value of debt and market value of equity. Stock market volatility is computed monthly from daily stock market log returns. The U.S. Treasury yield level is the yield of the 10 year U.S. Treasury note. The U.S. Treasury slope is computed as the difference between the yield of the 10 year and the 2 year U.S. Treasury notes.
32
Table 5. Sovereign spreads fixed effect regressions.
∆Spread over U.S. Treasury AAA to BBB- BB+ to B B- to C Overall sample
Constant 0.452 -0.06 -2.769 -0.071 (1.74)* (0.49) (1.55) (0.65) ∆Years to maturity 5.85 -0.619 -34.553 -0.833 (1.87)* (0.42) (1.61) (0.64) ∆Debt to foreign reserves 0.048 -0.009 0.755 0.04 (0.85) (0.62) (7.79)*** (2.78)*** ∆Political risk 0.006 0.052 0.133 0.075 (0.62) (4.46)*** (3.96)*** (7.86)*** ∆Political risk lagged 0.03 0.002 0.008 0.016 (3.06)*** (0.15) (0.00) (1.64) ∆Political risk 2nd lag -0.013 0.002 0.041 0.018 (1.33) (0.14) (1.20) (1.85)* ∆Local volatility lagged 0.003 0.088 0.067 0.075 (0.05) (11.12)*** (3.00)*** (11.69)*** ∆U.S. Treasury level -0.807 -1.088 -0.882 -0.978 (16.24)*** (12.85)*** (2.60)*** (14.32)*** ∆U.S. Treasury slope 0.135 0.664 1.073 0.51 (1.75)* (5.17)*** (2.15)** (4.86)*** Local return lagged -0.79 -5.613 -4.663 -4.664 (5.22)*** (25.88)*** (7.03)*** (26.74)***
Observations 1630 3870 690 6316 R-squared 0.19 0.30 0.22 0.22
This table shows estimates from an OLS regression model with Newey-West adjusted errors. We estimated the following equation to each bond observation: ∆Spreadi,t = Constant + β1*∆Debt to foreign reserves ratioi,t + β2*∆Country risk measurei,t + β3*∆U.S. Treasury yield curve levelt + β4*∆U.S. Treasury yield curve slopet + β5*∆Local volatilityi,t-1 + β6*Local returnt-1 + β7*∆Years to maturityi,t + εi,t. We estimated eight different specifications of this basic equation, substituting duration for years to maturity and debt to exports for debt to reserves. Years to maturity is the remaining life of a bond expressed in years, duration is a Macaulay’s duration expressed in years, debt to reserves is the ratio of total debt outstanding (bank loans, Brady and Eurobond issues) denominated in U.S. dollars divided by the total amount of international reserves also denominated in U.S. dollars. Debt to exports is the ratio of total debt outstanding (bank loans, Brady and Eurobond issues) denominated in U.S. dollars divided by the nominal monthly value of exports in U.S. dollars. Political risk is the value of The Economist Intelligence Unit's country index. Local stock market volatility is the standard volatility computed each month from daily stock market returns in U.S. dollars. The U.S. Treasury yield level is the yield of the 10 year U.S. Treasury note. The U.S. Treasury slope is computed as the difference between the yield of the 10 year and the 2 year U.S. Treasury notes. The world stock return is the log return of Datastream's world total return index. Absolute value of t statistics are in parentheses; *, **, *** denote significance at the 10%; 5%; and 1% level respectively.
33
Table 6. Domestic spreads fixed effect regressions. Leverage Class Overall ∆Spread over U.S. Treasury Low Medium High sample Constant -0.046 -0.031 0.075 -0.019 (2.18)** (1.36) (0.87) (0.85) ∆Years to maturity -0.591 -0.141 0.834 -0.249 (2.35)** (0.52) (0.81) (0.93) ∆Leverage -0.447 -0.005 -0.027 0.018 (1.96)** (0.03) (0.09) (0.12) ∆leverage lagged 3.127 0.941 6.747 4.445 (13.80)*** (5.69)*** (22.32)*** (30.77)*** ∆Stock return volatility 1.073 -0.164 3.133 2.149 (2.43)** (0.31) (4.18)*** (5.84)*** ∆Stock return volatility lagged 4.141 3.323 11.866 8.582 (9.59)*** (6.28)*** (15.69)*** (23.42)*** ∆U.S. Treasury level -0.218 -0.338 -0.421 -0.323 (14.53)*** (22.27)*** (11.46)*** (23.28)*** ∆U.S. Treasury slope -0.002 0.023 0.116 0.031 (0.08) (0.86) (1.83)* -1.28 S&P return lagged -0.002 0.008 -0.007 -0.003 (1.99)** (0.09) (3.32)*** (4.05)*** Observations 9080 9679 8709 27468 R-squared 0.07 0.07 0.13 0.09
This table shows estimates from an OLS regression model with Newey-West adjusted errors. We estimated the following equation to each bond observation: ∆Spreadi,t = Constant + β1*∆Leverage ratioi,t + β2*∆Stock return volatilityi,t + β3*∆U.S. Treasury yield curve levelt + β4*∆U.S. Treasury yield curve slopet + β5*S&P index returni,t-1 + β6*∆Years to maturityi,t + εi,t. We estimated six different specifications of this basic equation, substituting duration and modified duration for years to maturity. Years to maturity is the remaining life of a bond expressed in years, duration is a Macaulay’s duration expressed in years, modified duration is estimated as duration divided by (1+ redemption yield).Stock return volatility is the standard volatility of each firm's stock return computed each month from daily log returns in U.S. dollars. The U.S. Treasury yield level is the yield of the 10 year U.S. Treasury note. The U.S. Treasury slope is computed as the difference between the yield of the 10 year and the 2 year U.S. Treasury notes. The S&P index return is the log return of Datastream's S&P 500 total return index. Absolute value of t statistics are in parentheses; *, **, *** denote significance at the 10%; 5%; and 1% level respectively.
34
Table 7. Correlation structure of residuals.
Panel A: Sovereign bins
s11 s12 s13 s21 s22 s23 s31 s32 s33 s11 1.000 s12 0.826 1.000 s13 -0.007 0.511 1.000 s21 0.997 0.817 -0.004 1.000 s22 0.828 0.953 0.459 0.818 1.000 s23 0.367 0.781 0.895 0.369 0.773 1.000 s31 0.983 0.821 0.011 0.989 0.822 0.370 1.000 s32 0.891 0.896 0.259 0.886 0.966 0.624 0.897 1.000 s33 0.820 0.932 0.394 0.801 0.950 0.708 0.806 0.946 1.000
Panel B: Domestic bins
d11 d12 d13 d21 d22 d23 d31 d32 d33 d11 1.000 d12 0.978 1.000 d13 0.616 0.622 1.000 d21 0.967 0.968 0.744 1.000 d22 0.921 0.955 0.707 0.928 1.000 d23 0.685 0.673 0.856 0.747 0.747 1.000 d31 0.930 0.971 0.679 0.954 0.943 0.633 1.000 d32 0.859 0.920 0.712 0.893 0.973 0.659 0.960 1.000 d33 0.865 0.874 0.876 0.909 0.915 0.797 0.909 0.922 1.000
Panel C: Sovereign and Domestic bins
s11 s12 s13 s21 s22 s23 s31 s32 s33 d11 d12 d13 d21 d22 d23 d31 d32 d33 s11 1.000 s12 0.826 1.000 s13 -0.007 0.511 1.000 s21 0.997 0.817 -0.004 1.000 s22 0.828 0.953 0.459 0.818 1.000 s23 0.367 0.781 0.895 0.369 0.773 1.000 s31 0.983 0.821 0.011 0.989 0.822 0.370 1.000 s32 0.891 0.896 0.259 0.886 0.966 0.624 0.897 1.000 s33 0.820 0.932 0.394 0.801 0.950 0.708 0.806 0.946 1.000 d11 0.227 0.119 -0.168 0.221 0.068 -0.087 0.205 0.101 0.100 1.000 d12 0.248 0.127 -0.173 0.245 0.117 -0.079 0.242 0.157 0.114 0.978 1.000 d13 -0.211 0.063 0.464 -0.182 0.092 0.388 -0.159 -0.046 -0.094 0.616 0.622 1.000 d21 0.064 0.073 0.020 0.065 0.082 0.054 0.065 0.070 0.057 0.967 0.968 0.744 1.000 d22 0.194 0.154 -0.004 0.212 0.150 0.074 0.210 0.142 0.036 0.921 0.955 0.707 0.928 1.000 d23 -0.156 0.193 0.572 -0.141 0.100 0.440 -0.124 -0.079 -0.005 0.685 0.673 0.856 0.747 0.747 1.000 d31 0.159 0.062 -0.143 0.160 0.129 -0.059 0.181 0.173 0.074 0.930 0.971 0.679 0.954 0.943 0.633 1.000 d32 0.161 0.086 -0.058 0.181 0.160 0.032 0.196 0.177 0.005 0.859 0.920 0.712 0.893 0.973 0.659 0.960 1.000 d33 -0.003 0.075 0.173 0.020 0.116 0.172 0.040 0.057 -0.063 0.865 0.874 0.876 0.909 0.915 0.797 0.909 0.922 1.000
This table presents the correlation structure of the residual bins. Each sample (sovereign and domestic) was divided in three maturity categories and three leverage (debt to reserves, in the sovereign case) categories. The cutoff values for each category were determined using the 33rd and 66th centile to ensure an approximately equal number of observations in each bin. Each observation was assigned to a category. To compute the residuals, regressions were conducted in each bin. Then, for each bin, we average across residuals. The bins are named dij and sij for i, j=1, 2, 3, where d stands for domestic and s stands for sovereign, i for maturity category (1 = shot-term, 2 = medium-term, 3 = long-term), and j stand for leverage (debt to reserves) category (1 = low, 2 = medium, 3 = high). For example, d23 refers to a domestic, medium-term, high-leverage bin. Each sovereign bin contains 66 observations, while each domestic bin contains 148 observations.
35
Table 8. Principal component analysis of residuals.
Panel A: Sovereign bins Component Eigenvalue Difference Proportion Cumulative
1 6.8483 5.0152 0.7609 0.7609 2 1.8331 1.6574 0.2037 0.9646 3 0.1757 0.1047 0.0195 0.9841 4 0.0710 0.0406 0.0079 0.9920
Panel B: Domestic bins Component Eigenvalue Difference Proportion Cumulative
1 7.7606 6.9927 0.8623 0.8623 2 0.7679 0.5152 0.0853 0.9476 3 0.2527 0.1142 0.0281 0.9757 4 0.1385 0.0932 0.0154 0.9911
Panel C: Domestic and Sovereign bins Component Eigenvalue Difference Proportion Cumulative
1 7.5705 1.6083 0.4206 0.4206 2 5.9621 2.9935 0.3312 0.7518 3 2.9687 2.3462 0.1649 0.9167 4 0.6225 0.2166 0.0346 0.9513
This table presents the results of applying principal components analysis to the residuals. Each sample (sovereign and domestic) was divided in three maturity categories and three leverage (debt to reserves, in the sovereign case) categories. Each observation was assigned to a category. For each bond in each sovereign bin we estimated the following equation: ∆Spreadi,t = Constant + β1*∆Debt to foreign reserves ratioi,t + β2*∆Country risk measurei,t + β3*∆U.S. Treasury yield curve levelt + β4*∆U.S. Treasury yield curve slopet + β5*∆Local volatilityi,t-1 + β6*Local returnt-1 + β7*∆Years to maturityi,t + εi,t. For bonds in the domestic bins the following equation was estimated: ∆Spreadi,t = Constant + β1*∆Leverage ratioi,t + β2*∆Stock return volatilityi,t + β3*∆U.S. Treasury yield curve levelt + β4*∆U.S. Treasury yield curve slopet + β5*S&P index returni,t-1 + β6*∆Years to maturityi,t + εi,t. Residuals for each bond were computed and averaged across bins. For ease of interpretation, only the first four components are shown for each panel.
Eigenvalue
0.01.02.03.04.05.06.07.08.0
0 1 2 3 4 5Component
Eigenvalue
0.0
2.0
4.0
6.0
8.0
10.0
0 1 2 3 4 5Component
Eigenvalue
0.01.02.03.04.05.06.07.08.0
0 1 2 3 4 5Component
36
Table 9. Sovereign and domestic regressions including the common factors.
Sovereign bonds Domestic bonds
Overall Leverage Class Overall
∆Spread over U.S. Treasury AAA to BBB- BB+ to B B- to C sample ∆Spread over U.S. Treasury Low Medium High sample
Constant 0.203 -0.063 -1.055 -0.067 Constant 0.026 -0.019 0.103 0.019
(0.65) (0.81) (0.35) (0.72) (1.14) (0.83) (1.06) (0.68)
∆Years to maturity 2.875 -0.892 -14.339 -1.213 ∆Years to maturity 0.175 -0.006 1.009 0.109
(0.77) (0.97) (0.40) (1.10) (0.64) (0.02) (0.88) (0.33)
∆Debt to foreign reserves 0.103 0.035 1.047 0.125 ∆Leverage -0.673 0.147 0.198 0.141
(0.86) (2.64)*** (6.80)*** (7.01)*** (2.78)*** (0.86) (0.56) (0.83)
∆Political risk 0.023 0.019 0.086 0.049 ∆leverage lagged 4.122 1.19 7.622 5.393
(2.06)** (2.44)** (2.24)** (5.77)*** (17.10)*** (7.08)*** (21.00)*** (31.64)***
∆Political risk lagged 0.031 0.022 -0.026 0.024 ∆Stock return volatility 0.936 -1.009 2.033 1.495
(2.76)*** (2.89)*** (0.67) (2.83)*** (2.29)** (1.86)* (2.26)** (3.57)***
∆Political risk 2nd lag -0.01 0.004 0.046 0.021 ∆Stock return volatility lagged 4.487 3.281 12.396 9.197
(0.92) (0.56) (1.19) (2.42)** (11.46)*** (6.39)*** (14.37)*** (22.91)***
∆Local volatility lagged -0.005 0.054 0.113 0.052 ∆U.S. Treasury level
-0.085 -0.17 -0.26 -0.18
(0.72) (10.49)*** (3.95)*** (9.20)*** (3.65)*** (6.13)*** (3.25)*** (6.42)*** ∆U.S. Treasury level
-0.59 -0.495 -1.903 -0.59 ∆U.S. Treasury slope
-0.154 -0.021 -0.106 -0.091
(6.80)*** (6.33)*** (2.37)** (6.72)*** (4.26)*** (0.49) (0.87) (2.11)** ∆U.S. Treasury slope
-0.147 -0.207 2.163 -0.035 S&P return lagged -0.004 -0.002 -0.009 -0.005
(1.24) (1.83)* (2.25)** (0.28) (4.55)*** (1.81)* (3.07)*** (4.92)***
Local return lagged -0.641 -3.645 -6.104 -3.601 1st factor sovereign
0.006 0.004 0.006 0.005
(3.73)*** (24.99)*** (7.59)*** (22.78)*** (3.20)*** (1.68)* (2.84)** (2.35)** 1st factor domestic
0.026 -0.039 -0.161 -0.044 2dn factor sovereign
-0.001 -0.008 0.015 0.003
(3.56)*** (5.39)*** (2.72)*** (5.42)*** (0.48) (2.26)** (1.44) (0.89) 2dn factor domestic
-0.024 -0.021 0.141 0.024 1st factor sovereign lagged
0.004 0.001 -0.006 -0.001
(1.75)* (1.60) (1.60) (1.69)* (1.84)* (0.53) (0.77) (0.53) 1st factor domestic lagged
0.013 0.036 -0.192 0.002 2dn factor sovereign lagged
-0.013 -0.015 -0.003 -0.004
(1.85)* (5.07)*** (2.93)*** 0.00 (3.23)*** (2.90)*** (0.17) (0.78) 2dn factor domestic lagged
-0.084 -0.13 0.492 -0.07
(4.04)*** (6.79)*** (2.74)*** (3.23)***
Observations 1405 3460 513 5504 Observations 6630 5735 5864 18229
R-squared 0.20 0.35 0.31 0.20 R-squared 0.11 0.08 0.16 0.12
This table shows estimates from an OLS regression model with Newey-West adjusted errors. We estimated the following equation to each sovereign bond observation: ∆Spreadi,t = Constant + β1*∆Debt to foreign reserves ratioi,t + β2*∆Country risk measurei,t + β3*∆U.S. Treasury yield curve levelt + β4*∆U.S. Treasury yield curve slopet + β5*∆Local volatilityi,t-1 + β6*Local returnt-1 + β7*∆Years to maturityi,t+ β8*1st factor domestict + β9*2nd factor domestict + εi,t. For each domestic bond. we estimated the following equation: ∆Spreadi,t = Constant + β1*∆Leverage ratioi,t + β2*∆Stock return volatilityi,t + β3*∆U.S. Treasury yield curve levelt + β4*∆U.S. Treasury yield curve slopet + β5*S&P index returni,t-1 + β6*∆Years to maturityi,t + β8*1st factor sovereignt + β9*2nd factor sovereignt + εi,t. Years to maturity is the remaining life of a bond expressed in years, debt to reserves is the ratio of total debt outstanding (bank loans, Brady and Eurobond issues) denominated in U.S. dollars divided by the total amount of international reserves also denominated in U.S. dollars. Political risk is the value of The Economist Intelligence Unit's country index. Local stock market volatility is the standard volatility computed each month from daily stock market returns in U.S. dollars. The U.S. Treasury yield level is the yield of the 10 year U.S. Treasury note. The U.S. Treasury slope is computed as the difference between the yield of the 10 year and the 2 year U.S. Treasury notes. The S&P stock return is the log return of Datastream's S&PCOMP total return index. Stock return volatility is the standard volatility of each firm's stock return computed each month from daily log returns in U.S. dollars. The S&P index return is the log return of Datastream's S&P 500 total return index. Absolute value of t statistics are in parentheses; *, **, *** denote significance at the 10%; 5%; and 1% level respectively.
37
Table 10. Vector autoregression model with exogenous variables.
R-squared
Equation Obs No exog.
vars. All vars.
1st factor domestic 59 0.3684 0.4496 1st factor sovereign 59 0.0586 0.6061 2nd factor domestic 59 0.2029 0.5649 2nd factor sovereign 59 0.1490 0.6908 1st factor domestic Coeff. Std. Error z P>z 2nd factor domestic Coeff. Std. Error z P>z 1st factor domestic 1st factor domestic Lag 1 -0.060 0.120 -0.500 0.619 Lag 1 -0.048 0.057 -0.860 0.391 Lag 2 0.013 0.116 0.110 0.909 Lag 2 0.053 0.055 0.960 0.335 1st factor sovereign 1st factor sovereign Lag 1 -0.067 0.132 -0.510 0.609 Lag 1 0.034 0.062 0.550 0.586 Lag 2 0.424 0.108 3.940 0.000*** Lag 2 0.087 0.051 1.710 0.087* 2nd factor domestic 2nd factor domestic Lag 1 -0.229 0.305 -0.750 0.453 Lag 1 0.151 0.144 1.050 0.292 Lag 2 -0.500 0.282 -1.770 0.077* Lag 2 -0.349 0.133 -2.630 0.009** 2nd factor sovereign 2nd factor sovereign Lag 1 0.185 0.230 0.800 0.421 Lag 1 0.086 0.108 0.790 0.428 Lag 2 0.242 0.223 1.090 0.278 Lag 2 0.150 0.105 1.440 0.151 ∆flowstocks -0.032 0.018 -1.740 0.082* ∆flowstocks 0.006 0.009 0.750 0.452 ∆flowbonds -0.135 0.056 -2.420 0.016** ∆flowbonds -0.049 0.026 -1.860 0.063* ∆netreserves 6.764 5.771 1.170 0.241 ∆netreserves 13.067 2.717 4.810 0.000*** ∆onoff 9.857 9.295 1.060 0.289 ∆onoff 3.148 4.375 0.720 0.472 ∆spread brady 0.026 0.558 0.050 0.963 ∆spread brady 0.793 0.263 3.020 0.003*** Constant 0.310 0.223 1.390 0.165 Constant -0.072 0.105 -0.690 0.491 1st factor sovereign Coeff. Std. Error z P>z 2nd factor sovereign Coeff. Std. Error z P>z 1st factor domestic 1st factor domestic Lag 1 0.167 0.109 1.530 0.125 Lag 1 -0.014 0.059 -0.230 0.815 Lag 2 -0.090 0.105 -0.860 0.391 Lag 2 0.066 0.057 1.140 0.254 1st factor sovereign 1st factor sovereign Lag 1 -0.339 0.120 -2.830 0.005*** Lag 1 0.082 0.065 1.250 0.211 Lag 2 -0.148 0.098 -1.520 0.129 Lag 2 0.117 0.053 2.210 0.027** 2nd factor domestic 2nd factor domestic Lag 1 -0.118 0.277 -0.430 0.671 Lag 1 0.267 0.151 1.770 0.077* Lag 2 0.398 0.256 1.550 0.120 Lag 2 -0.369 0.140 -2.640 0.008*** 2nd factor sovereign 2nd factor sovereign Lag 1 0.765 0.209 3.670 0.000*** Lag 1 0.048 0.114 0.420 0.672 Lag 2 -0.216 0.202 -1.070 0.284 Lag 2 0.304 0.110 2.760 0.006*** ∆flowstocks -0.013 0.017 -0.800 0.422 ∆flowstocks 0.048 0.009 5.270 0.000*** ∆flowbonds -0.289 0.051 -5.690 0.000*** ∆flowbonds -0.116 0.028 -4.200 0.000*** ∆netreserves 21.262 5.237 4.060 0.000*** ∆netreserves 13.124 2.854 4.600 0.000*** ∆onoff 6.461 8.434 0.770 0.444 ∆onoff 1.330 4.596 0.290 0.772 ∆spread brady -2.968 0.507 -5.860 0.000*** ∆spread brady 0.506 0.276 1.830 0.067* Constant -0.172 0.202 -0.850 0.394 Constant 0.048 0.110 0.440 0.660
This table shows estimates a vector autoregression model of the following form:
11
111
11 εδγβ ∑∑=
−=
− ++++=k
jtjtj
k
jjtjt XFacDomFacSovaFacSov
21
221
22 εδγβ ∑∑=
−=
− ++++=k
jtjtj
k
jjtjt XFacDomFacSovaFacDom
The first and second domestic factors are the factors extracted from the principal component analysis of the residuals of equation (2) applied to the domestic bins. The first and second sovereign factors are extracted from the principal component analysis of the residuals of equation (1) applied to the sovereign bins. Net borrowed reserves is computed as total borrowing minus extended credit minus excess reserves, divided by total reserves. Onoff is the difference between the on-the-run thirty year U.S. Treasury bond and the most recent off-the-run bond. Flowsstocks is from the IFC’s statistics and is the amount of money flowing into equity mutual funds. Flowbonds is from the same source and represents the flows into bond funds; *, **, *** denote significance at the 10%; 5%; and 1% level respectively.
38
Figure 1. First common component
-6
-4
-2
0
2
4
6
8
10
12
6/1/
1997
9/1/
1997
12/1
/199
7
3/1/
1998
6/1/
1998
9/1/
1998
12/1
/199
8
3/1/
1999
6/1/
1999
9/1/
1999
12/1
/199
9
3/1/
2000
6/1/
2000
9/1/
2000
12/1
/200
0
3/1/
2001
6/1/
2001
9/1/
2001
12/1
/200
1
3/1/
2002
6/1/
2002
Sovereign Domestic
Figure 2. Second common component
-6
-5
-4
-3
-2
-1
0
1
2
3
6/1/
1997
9/1/
1997
12/1
/199
7
3/1/
1998
6/1/
1998
9/1/
1998
12/1
/199
8
3/1/
1999
6/1/
1999
9/1/
1999
12/1
/199
9
3/1/
2000
6/1/
2000
9/1/
2000
12/1
/200
0
3/1/
2001
6/1/
2001
9/1/
2001
12/1
/200
1
3/1/
2002
6/1/
2002
Sovereign Domestic