Universiteit van Amsterdam – Amsterdam Business School
What non-country specific factors drive Sovereign Emerging Market Bond Return?
Master Thesis
Author: Sirianong Peyasantiwong (11375787) Program: MSc Finance: Asset Management
Date: June, 2017
Statement of Originality
This document is written by Sirianong Peyasantiwong, who takes full responsibility for the contents of this document.
I declare that the text and the work presented in this document are original and that no source other than those mentioned in the text and its references have been used in creating it.
The Faculty of Economics and Business is responsible solely for the supervision of completion of the work, not for the contents.
Acknowledgement
I appreciate all insightful suggestions from both supervisors at Neuberger Berman (Europe), Ltd. I would like to express my gratitude to Vera Kartseva for the valuable comments and guidance throughout the completion of this thesis. Special thanks go to Lei Wan for his tremendous support and guidance, especially in the methodology section. I also wish to express my appreciation to Rob Drijkoningen for his support during my thesis internship. My thanks also go to Prof. Liang Zou for providing practical and useful advice. I would like to thank Prof. Esther Eiling for an introduction to Principal Component Analysis during the advance investment course. Finally, I am indebted to Prof. Jeroen Ligterink for his comment on my proposal during the thesis seminar course and his continuous support throughout the master program.
Abstract
We identify the common non-country specific factors, which drive Sovereign Emerging Market Bond Return using Principle Component Analysis in various models. When we perform PCA on both hard currency and local currency bond simultaneously, the first few PCs capture only 62%-68% of overall covariance. This suggested that there is a large residual of idiosyncratic countries that is not captured by the systematic risk. The first component represents Hard Currency Exposure while the second component represents Local Currency Exposures. The differentiating factors are U.S. treasury yield, Fed Fund Rate, and Dollar Spot. Equity and Commodity rising is positive to both components.
Table of Contents
Chapter 1: Introduction ... 7
Chapter 2: Literature Review ... 9
2.0 Introduction ... 9
2.1 Overview of Sovereign Emerging Market Bond Index ... 9
2.2 Factors driving sovereign emerging market bond return ... 11
Chapter 3: Research Methodology ... 14
3.0 Introduction ... 14
3.1 Data Sources, Country Selection, and Time Frame ... 14
3.2 The principal component analysis (PCA) ... 16
3.3 Model Specifications ... 18
3.4 OLS Regression ... 19
Chapter 4: Study Results ... 20
4.0 Introduction ... 20
4.1 The number of PCs from PCA extraction ... 20
4.2 Country loadings ... 20
4.3 RCs time-series movement ... 26
4.4 OLS regressions results ... 27
Chapter 5: Discussion ... 30
5.0 Introduction ... 30
5.1 Conclusion of our study ... 30
5.2 Contribution to existing Literature ... 30
5.3 Implication and limitation of our study ... 31
Appendix 1: The selected countries for PCA analysis ... 34
Appendix 2: Scree Plots of 10 Models... 36
Appendix 3: Results of Model 3-5 ... 38
List of Tables
Table 1 Emerging Markets Bond Index Global Composition and Statistics... ...10
Table 2 Government Bond Index-Emerging Markets Composition and Statistics...11
Table 3 The Factor Correlation Matrix...……….15
Table 4 Model Specification...………...……… 19
Table 5 Model 6-7 Results...………...…………. 24
Table 6 Model 8-10 Results...………...…………. 25
Table 7 Regressions of Factors on RCs of Relative Models...…………..28
Table 8 Regressions of Factors on EMBI and GBI-EM...………29
Table 9 The List of Selected Countries from EMBI...………34
Table 10 The List of Selected Countries from GBI-EM...………35
Table 11 The Correlation between Factors and RCs…...……… .41
List of Figures
Figure 1 Model 1 Results...…...……….……….21Figure 2 Model 2 Results …...………...……….………….…...22
Figure 3 Model 2 Time-series RCs movement of selected models…...…….……..…26
Figure 4 Scree Plots of 10 Models…...….…….……..…….36
Figure 5 Model 3 Results…...……….…….……….…..38
Figure 6 Model 4 Results…...……….…….…………...39
Chapter 1: Introduction
Sovereign Emerging Market Debt is a growing asset class as they offer attractive return and diversification to investors. Sovereign emerging market debts comprise of bonds that the governments in emerging countries issue in either hard currency or local currency. While hard currency bonds are typically traded in New York, London, and Singapore, local currency bonds are typically traded in issuers’ countries. These bonds typically offer a better return than bonds issued by developed country government because of higher risk, potential foreign exchange appreciation, and home biasness of investors based in developed countries. Home biasness is the tendency to invest in a large amount of domestic assets, despite the benefits of diversifying into foreign assets.
This thesis aims to find the non-country specific factors driving returns for sovereign emerging market bonds such as bond yield in the United States, Fed Fund Rates, and S&P index. The sensitivities between emerging countries and global factors are reflected in the return of bonds in these emerging countries. Therefore, the movement in the global factors affects the emerging market sovereign bond return according to the country’s sensitivity to that global factor. The results of this analysis are useful for top-down asset allocation, which does not focus on country-specific factors. We select factors based on literature and expert opinions. Expert opinions are from the portfolio managers in Emerging Market Debt Department at Neuberger Berman (Europe), Ltd. These factors are oil, Commodity, USD spot, US Equities, the yield of US government bond in various maturities, Fed Fund Rate, the 2s10s spread of US government bond, the 10s30s spread of US government bond, real US government bond yield in various maturities.
In the methodology section, we start the analysis by using principal component analysis (PCA). Principal Component Analysis (PCA) is a tool to transforms a number of correlated variables into a small set of uncorrelated variables, which are called principal components (PCs). Principal Component (PCs) allows us to capture the largest proportion of the variance in the dataset, with only a few non-overlapping factors. We aim to capture the dominant global factors associated with the variation of emerging market sovereign bond return. We use the J.P. Morgan Emerging Markets Bond Index (EMBI Global) for the Hard Currency bonds and the J.P. Morgan Government Bond Index Emerging-Market (GBI-EM Global) for the local currency bonds.
The PCA offers two results: clusters of countries based on its co-movement in bond return and time-series of PCs. We draw a characteristic of countries in the same clustered to associate that characteristic with factors. Also, we calculate the correlation between the time-series of PCs and factors that we want to test (e.g. Fed Fund Rate). The results from this analysis give us the perspective on the significant global factors that can explain the variation of bond return. We then use multiple regressions to quantify these factors with the bond index.
We design our study based on the three set of models. The first set of models tries to capture return components of both EMBI index (hard currency) and GBI-EM index (local currency) simultaneously. These models offer an insight into systematic factors of bond return. The second set of models tries to capture return component of EMBI index and GBI-EM individually. These models cluster the countries base on its commonality in co-movement of return. The cluster of countries shows common characteristics. Finally, the third set of the model is based on the relative return between EMBI index and GBI-EM index of the same countries. The Relative Models are designed to derive the factors that differentiate the return between hard-currency bonds and local-currency bonds.
To our knowledge, this study is the first study that aims to capture factors that drive sovereign emerging market hard currency and local currency bond return using principal component analysis (PCA).
Chapter 2: Literature Review
2.0 Introduction
This chapter provides a comprehensive literature review of non-specific factors driving emerging market sovereign bond spread. This chapter is divided into two sections. The first section presents an overview of sovereign bond from emerging countries. In this section, we present JPMorgan Emerging Market Bond Index (EMBI Global) and JPMorgan Government Bond Index – Emerging Market Global (GBI-EM Global). In the subsequent section, a literature review of factors driving emerging market sovereign bond return is presented. In addition, dominating factors from studies using principal component analysis are reviewed.
2.1 Overview of Sovereign Emerging Market Bond Index
The bond markets of emerging countries have grown significantly in the last decade. The investment universe for the investor has been expanded due to an increase in the number of issuers from emerging countries. This can be attributed to an increase in liquidity and quality of bonds from emerging markets.
Our study is based on two indices: J.P. Morgan Emerging Markets Bond Index (EMBI) for hard currency bond and JPMorgan Government Bond Index – Emerging Market Global (GBI-EM Global) for local currency bond.
JPMorgan Emerging Markets Bond Index (EMBI)
JPMorgan Emerging Markets Bond Index (EMBI) consists of USD denominated Brady bonds, Euro bonds, and Traded loans issued by sovereign and quasi-sovereign identity. Currently, JP.Morgan defines “quasi-sovereign” as an entity that is 100% guaranteed or owned by the national government. The weight of each instrument in the index is determined by dividing the issue's market capitalization by the total market capitalization of all instruments in the index.
Indices Inclusion Criteria
Eligible countries: A country’s GNI per capita must be below the Index Income Ceiling (IIC) for
three consecutive years. The Index Income Ceiling as the GNI per capital level is adjusted every year by the growth rate of the World GNI per capita, Atlas method (current US$), provided by the World Bank annually.
Instrument type: fixed and floating rate instruments including amortizing bonds or loans. Current face amount outstanding: US$500 million or more
Time until maturity: at least 2.5 years until maturities at issuance. Once added, an instrument may
remain in the EMBI until 12 months before it matures.
Number of Countries: 65 at the end of March 2017
Table 1
Emerging Markets Bond Index Global Composition and Statistics
This table reports statistics for Emerging Markets Bond Index (EMBI) as of 31 March 2017. This index consists of USD denominated Brady bonds, Euro bonds, and Traded loans issued by sovereign and quasi-sovereign identity. The first three columns show index composition. The latter columns report the current yield, spread, average life, interest rate duration to maturity, and interest rate to worst statistics, respectively.
Source: Linzie, Kumaran, & Patel, 2017
JPMorgan Government Bond Index – Emerging Market Global (GBI-EM Global)
JPMorgan Government Bond Index – Emerging Market Global (GBI-EM Global) is an investable benchmark consisting of local currency government bond in countries that are accessible by most of the international investor base, excluding countries with explicit capital controls. The weight is determined by the component countries’ aggregate normalised market capitalization.
Indices Inclusion Criteria
Eligible countries: A country’s GNI per capita must be below the Index Income Ceiling (IIC) for
three consecutive years. The Index Income Ceiling as the GNI per capital level is adjusted every year by the growth rate of the World GNI per capita, Atlas method (current US$), provided by the World Bank annually.
Instrument type: fixed coupon instruments excluding floating-rate and amortising bonds
Current face amount outstanding: US$ 1 billion equivalent for onshore local currency bonds and
US$500 million equivalent for offshore currency linked bonds.
Time until maturity: only bonds with more than 13 months to maturity at the rebalance date are
included.
Number of Countries: 16 countries at the end of March 2017
Table 2
Government Bond Index-Emerging Markets Composition and Statistics
This table reports statistics for Government Bond Index-Emerging Markets (GBI-EM) as of 31 March 2017. This index consists of local currency government bond in countries that are accessible by most of the international investor base. The first two columns show index composition. The latter columns report the current yield, modified duration, convexity, and the number of issues, respectively.
Source: Linzie, Kumaran, & Patel, 2017
2.2 Factors driving sovereign emerging market bond return
While there is no dearth of studies examining the influence of country-specific factors on sovereign bond return (e.g. Edwards, 1984; Rowland & Torres, 2004), research on the global macro-economic determinants of sovereign spread is relatively new and limited. A few scholars like Arora and Cerisola (2001), Ferrucci (2003) and Bellas, Papaioannou, and Petrova (2010) have studied the important
factors that can drive sovereign bond yields and returns in different emerging economies. Their works are critically reviewed herein in order to identify those global factors and their potential influence on the sovereign bond returns.
Arora and Cerisola (2001) were of the opinion that due to globalization, US economic policies and the condition in global economic market influences the sovereign bond return in the emerging countries, making them vulnerable. They found the significant positive influence of 10 years US treasury yield on sovereign bond yields of Argentina, Brazil, Mexico, Colombia, Poland, Bulgaria, Philippines, Thailand, Korea and Indonesia. While one major strength of Arora and Cerisola (2001) is the inclusion of a large number of emerging market, the study did not include an important factor of short-term US treasury yield.
Ferrucci (2003) modeled a significant negative impact of 10 years US interest rate on emerging market sovereign spreads. They held that higher short-term US Interest rate raise borrowing costs for EMEs and the higher long-term US rates produce strong negative impact. They pointed out that as the slope of the curve of US yield become steeper, the spread of sovereign bond in emerging markets declines. According to Ferrucci (2003), one possible explanation for this effect is the behaviour of leveraged investors who may increase the demand for EME assets by pushing prices up and spreads down when global credit conditions allow cheap borrowing.
Bellas, Papaioannou, and Petrova (2010) who used a sample of 14 emerging markets reported similar results. They found that after controlling the effect of some local liquidity variables, both the 3-month US treasury yield and 10-year treasury yield produce a significant impact. They also reported the higher impact of the slope between 3-month US treasury rates and 10 year treasury yield.
Another factor that was reported in studies to influence the sovereign bond spread of emerging markets is the federal fund rates. Arora and Cerisoal (2001) reported much higher and significant impact of US federal fund target rates and market volatility, as a sign of the uncertainty of US monetary policy that was observed on the sovereign bond return of most developing countries in their sample. Rojas and Jaque (2009) also found a significant impact of US federal fund rates on the sovereign spread in Chili. They held that US federal fund rates have two-fold impacts – one is due to the risk of default and the second is due to the rise in the risk-free rates. Baldacci, Gupta and Mati (2008), however, did not find any significant impact of US federal fund rate on the sovereign spread of emerging market. They used a much larger sample of around 30 countries, so this insignificant impact could be the inclusion of markets that were ignored in the other studies. Min (1998) explained this ambiguity by claiming that the severing bonds do not work like syndicated loans, and are not influenced by short-term dollar rates.
Pratiwi (2015) examined default and non-default factors to predict the bond spread in emerging markets. Using cross-sectional fixed-effect panel model, they found a significant relationship between both types of factors. Default factor is proxy by Hartelius’ Credit Ratings and Outlook Index (CROI), and non-default factors are proxy by the volatility in the Fed Fund Futures Market, and the volatility index of S&P 500.
Csonto and Ivanchenko (2013) analysed local and global factors that dominate sovereign bond spread during a different period. Local factors are economic risk rating, financial risk rating and, political risk rating, while global risk factors are global risk aversion and a liquidity condition. The results show that while both country-specific and global developments are important in the long-run, global factors are the main determinants of spreads in the short-run. Furthermore, global factors tend to drive changes in the spreads during the periods of severe market stress.
Factors dominate sovereign emerging market bond using PCA approach
Rajan & Lawlor (2012) used principal component analysis (PCA) to measure systematic risks of emerging market fixed-income securities. They found that the first two principal components captured 70% to 80% of the covariance. The PC3 and higher can explain only a small incremental additional percentage. The first principal component (PC1) is a systematic directional emerging market macro risk, which can be regarded as “EM Beta”. The PC2 is the divergence of emerging local market returns from those of external debt spread market, which is the opposite movement of local versus hard currency returns.
Jaramillo and Weber (2013) used the principal components analysis to find the dominated factors that explain domestic bond yield in emerging countries. The results suggested that the two most common international factors are global risk appetite and global liquidity. According to the results, “Overall, the analysis indicates a clear correlation between the first common factor and global liquidity as measured by market expectations of 3 month US interest rates. Furthermore, there is a strong correlation between the second common factor and the VIX.”
Chapter 3: Research Methodology
3.0 Introduction
This chapter describes methods used to find the relationship between non-country-specific factors and sovereign bond spread in emerging market. It is divided into four sections. The first section shows the data sources, country selection, and time frame. In section two, the principal component analysis (PCA), which is used as the tool for extracting the most significant factors that drive the bond return, is described. The third section describes three set of models that we use for PCA analyses. The fourth section presents the use of regression to quantify the effect of significant factors.
3.1 Data Sources, Country Selection, and Time Frame
We select factors based on literature and expert opinions. Expert opinions are from the portfolio managers in Emerging Market Debt team at Neuberger Berman (Europe), Ltd.
Data Sources
Data Sources
Variables Source
Emerging Market Sovereign Bond Spread
Index (EMBI Index) JP Morgan
Market Website Government Bond Index - Emerging Markets Series (GBI-EM Index)
S&P 500 total return index
Bloomberg Terminal Fed Fund Rates
DXY index (Dollar Spot) Brent
US Spot Raw Industrials US government 2 year yield US government 10 year yield US government 30 year yield
Spread US government 2s10s year yield Spread US government 10s30s year yield US 10Y real yield
Slope between 2s10s and 10s30s yields.
Country Selection
EMBI Universe for Hard Currency Spread Return
We selected 23 countries. The weight for these selected countries is 75% of the total market capitalization of the EMBI index at the end of January 2017. The list of selected countries is shown in Appendix 1. The EMBI index contains 67 countries. We deselected 44 countries because of a number of missing data and discarded China and Lebanon based on liquidity issues.
GBI-EM universe for USD total return
According to the missing data, we selected 11 countries. The weight for these selected countries is 97.3% of the total market capitalization at of the GEI-EM Global at the end of March 2017. The list of selected countries is shown in Appendix 2. The GBI-EM index contains 17 countries. We deselected 6 countries because of a number of missing data.
Time Frame
We use monthly data starting from February 2005. Since we have decided to keep Russia in our sample, the first availability of Russia monthly return data in EMBI index is February 2015.
Table 3
The Factor Correlation Matrix
This table shows the correlation between factors influencing emerging market bond return. These factors are BRENT (oil index), Commodity, DXY (USD spot), S&P 500 (US Equities), US2YY (the yield of 2-year US government bond index), US10YY (the yield of 10-year US government bond index), US30YY (the yield of 30-year US government bond index), World Cyclical Equity Index, World Defensive Equity Index, FFR (Fed Fund Rate), 2y10y (the spread between 2 year and 10 year US government bond), 10y30y (the spread between 10 year and 30 year US government bond), real US10Y (the nominal yield of 10 year government bond subtract by inflation), 2s10s min 10s30s (the spread between the spread of 2y10y and the spread of 10y30y). The correlation calculation is based on the monthly data between February 2005 and February 2017 for all factors except World Cyclical Equity Index and World Defensive Equity Index. The availability of both indices started from October 2007. For FFR, the quantitative easing policy has kept the Fed Fund Rate at the low level irresponsive to the economy condition. Therefore, we use Federal Funds Rate until December 2006 and US 1Y1Y Swap Rates afterward.
BRENT Commo DXY S&P 500 US2YY US10 YY US30 YY world Cyc EQ* world Def EQ*
FFR** 2y10y 10y30y real US10Y 2s10s min 10s30s BRENT 1.00 Commo 0.56 1.00 DXY -0.51 -0.49 1.00 S&P 500 0.40 0.47 -0.47 1.00 US2YY 0.24 0.18 0.11 0.34 1.00 US10YY 0.35 0.28 0.01 0.30 0.70 1.00 US30YY 0.38 0.35 -0.09 0.32 0.53 0.93 1.00 World Cyc EQ* 0.56 0.61 -0.62 0.92 0.32 0.33 0.38 1.00
World Def EQ* 0.33 0.36 -0.55 0.83 0.14 0.05 0.08 0.73 1.00
FFR** 0.29 0.22 0.13 0.27 0.83 0.68 0.51 0.24 0.05 1.00 2y10y 0.21 0.18 -0.11 0.04 -0.16 0.60 0.69 0.11 -0.08 0.00 1.00 10y30y 0.06 0.15 -0.26 0.01 -0.52 -0.29 0.08 0.11 0.08 -0.52 0.18 1.00 real US10Y -0.07 -0.12 0.30 -0.04 0.50 0.71 0.54 -0.13 -0.25 0.52 0.41 -0.51 1.00 2s10s min 10s30s 0.18 0.10 0.02 0.04 0.11 0.72 0.64 0.06 -0.11 0.26 0.88 -0.31 0.65 1.00
3.2 The principal component analysis (PCA)
Principal Component Analysis (PCA) is a tool to transforms a number of correlated variables into a small set of uncorrelated variables, which is called principal components (PCs). PC1 and PC2 are a linear combination of observed variables. The weight used to form PCs is selected in order to maximise the variance that can be explained by PCs and still keep PCs uncorrelated.
The first principal component (PC1) is the weighted combination of n variables that accounts for the highest variance in the original dataset. The second principal component (PC2) is the weight combination that accounts for the most variance in the original dataset, on the condition that it is uncorrelated to the first principal component. In other words, PC2 must be perpendicular to PC1. The subsequent principal components and the weight combination account for the highest variance in the original dataset, on the condition that it is uncorrelated to the previous principal components.
According to Fabozzi (2002), the factors are defined as
Δ𝜙𝑃𝐶𝐴(𝑡) = 𝑎𝑖1∗ 𝑟1(𝑡) + ⋯ + 𝑎𝑖𝑛∗ 𝑟𝑛(𝑡)
Where 𝑟𝑖𝑛 represents the characteristics of our variable of interest. In this study, 𝑟𝑖𝑛 is the bond spread
for hard currency sovereign emerging bond. 𝑎11, … , 𝑎1𝑛 are the coefficient that we aim to estimate.
The covariance between Δ𝑟𝑘(t) and Δ𝑟𝑙(t) is denoted as
𝜎𝑘𝑙= 𝑐𝑜𝑣(Δ𝑟𝑘(𝑡), Δ𝑟𝑙(t))
For the first principal component, the objective of PCA is to find coefficient 𝑎11, … , 𝑎1𝑛 in the
equation below to maximize the variance of Δ𝜙𝑃𝐶𝐴(𝑡) (Fabozzi 2002).
max 𝑎11,…,𝑎1𝑛 Δ𝜙𝑃𝐶𝐴(𝑡) = ∑ ∑ 𝑎1𝑘𝑎1𝑙𝜎𝑘𝑙 𝑛 𝑙=1 𝑛 𝑘=1
Under the restriction
∑ 𝑎1𝑗2 = 1 𝑛
𝑗=1
The residual 𝜀̂(t) is the amount of variation that the first PC has not explained. Then, the second 𝑖
component is calculated based on the residual from the first PC. Δ𝜙2𝑃𝐶𝐴(𝑡) = 𝑎21𝜀̂(𝑡) + ⋯ + 𝑎𝑖 2𝑛𝜀̂(𝑡) 𝑛
max
𝑎11,…,𝑎1𝑛
Var(Δ𝜙𝑃𝐶𝐴(𝑡))
Under the restriction
∑ 𝑎2𝑗2 = 1
𝑛
𝑗=1
This process continues for the subsequent components.
The goal of this process is to find factors that accounted for a large variation of the variable of interest. Since the calculation of the PC2,…,PCn is based on the residuals of the previous PCs, they are
uncorrelated to the previous principal components.
In this study, we calculate the principal component analysis by the following steps: First, we find the covariance matrix of the bond spread return of each country in JPMorgan EMBI index and the total return of GBI-EM index. Secondly, we calculate the eigen values and corresponding eigen vectors of the covariance matrix. The use of eigen values and the eigen vectors is to facilitate the maximisation of problems and to extract weights for principal components. To solve for the maximisation, the Lagrangean is formed as
𝐿(𝑎1) = 𝑎1′𝑀𝑎1− 𝜆(𝑎1′𝑎1− 1)
The first order condition is
𝜕𝐿(𝑎1)
𝜕𝑎1
= 𝑀𝑎1− 𝜆𝑎1= 0
Where M is the covariance matrix estimated in step one; and is the eigen vectors and the eigen values. The scree plot from this step shows the proportion of variance explained from each PCs. We selected only a few most important principal components based on the eigen values using the parallel analysis. The parallel analysis is a Monte Carlo simulation technique that can be used to determine the number of PCs to retain in Principal Component Analysis. It provides a better alternative to other methods such as scree test and Kaiser’s eigen value-greater-than one rule (Ledesma & Valero-Mora, 2007). In the third step, we extract the most important PCs and rotate the axis using orthogonal rotation. The rotation aids the interpretation of the results. In other words, each rotational component is defined by a limited set of variables with a few larger loadings than its corresponding principal component. In this study, the result from the rotation is called “RCs” instead of “PCs”.
The interpretation of PCA is based on the “loadings” of RCs. The loading is the correlation between the observed variables and the RCs. The common characteristics of the countries with high loadings
can be used as factors. In addition, we examine the time-series of each RCs to find the factors that are highly correlated with the dominated RCs.
3.3 Model Specifications
We designed our study based on the three set of models. The first set is the absolute models to investigate the factors that systematically drive the return of both hard currency bonds (EMBI) and local currency bonds (GBI-EM). We have five models in this first set. The first model is designed to test the PCA on both EMBI spread return and GBI-EM total return in USD. The second model is for PCA on EMBI Spread Return, GBI-EM total return in local currency, and Emerging Market spot return of exchange rate without carry. The calculation of the EM FX is based on the equation.
(1 + 𝑇𝑜𝑡𝑎𝑙_𝑅𝑒𝑡𝐿𝑜𝑐𝑎𝑙_𝐶𝑢𝑟𝑟𝑒𝑛𝑐𝑦)(1 + 𝐸𝑀_𝐹𝑋) = (1 + 𝑇𝑜𝑡𝑎𝑙_𝑅𝑒𝑡𝑈𝑆𝐷)
The third and the fourth models are designed to assess the effect of local interest rates and foreign exchange on the bond return. So, we add GBI-EM total return in local currency, which can be view as a proxy for local rates, in the third model and return from spot foreign exchange of countries in GBI-EM index in the fourth model.
The second set of Models (Model 6-8) comprises of three models. In this set of models, we conducted PCA on EMBI spread return, EMBI total return, and GBI-EM total return in USD, respectively. These models are designed to cluster the countries based on their commonalities within the index. We also attempt to find the factors that drive these commonalities. We expected to see how PCA clustered countries within each index. Then, we will link the factors to each clustered countries.
The third set of Models (Model 9-10) is the set of relative models. In model 9, we first subtract GBI-EM total return in USD from the GBI-EMBI spread return. Then, we conduct PCA on this relative data series. In model 10, we first subtract GBI-EM total return in USD from the EMBI total return. The Relative Models are designed to derive the factors that differentiate the return between hard-currency bonds and local-currency bonds.
Table 4
Model Specification
This table reports our model specifications. The Absolute Models (Model 1-5) are designed to show the factors that systematically drive the overall emerging market bond return in both hard currency and local currency simultaneously. The Absolute Models (Model 6-8) are designed to cluster the countries based on their commonalities within the index. We also attempt to the factors that drive these commonalities. The Relative Models (Model 9-10) are designed to derive the factors that differentiate the return between hard-currency bonds and local-currency bonds. The EM FX (excluding carry) is calculated by the equation.
(1 + 𝑇𝑜𝑡𝑎𝑙_𝑅𝑒𝑡𝐿𝑜𝑐𝑎𝑙_𝐶𝑢𝑟𝑟𝑒𝑛𝑐𝑦)(1 + 𝐸𝑀_𝐹𝑋) = (1 + 𝑇𝑜𝑡𝑎𝑙_𝑅𝑒𝑡𝑈𝑆𝐷) EMBI Spread Return EMBI Total Return GBI-EM total return in USD
GBI-EM total return in local currency
EM_FX Exc. carry
Absolute Model (1): PCA on both GBI-EM and EMBI
Model 1 Selected Selected
Model 2 Selected Selected Selected
Model 3 Selected Selected
Model 4 Selected Selected
Model 5 Selected Selected
Absolute Model (2): PCA on either GBI-EM or EMBI
Model 6 Selected
Model 7 Selected
Model 8 Selected
Relative Model: Difference between EMBI return and GBI-EM total return in USD
Model 9 Selected Selected
Model 10 Selected Selected
Once we extracted the RCs from the PCA analysis based on these ten models, we will perform OLS regression to measure the coefficient of each factor in determining the bond return of the emerging market.
3.4 OLS Regression
A standard OLS model can be written as
𝑦𝑖𝑡 = 𝑋𝑖𝑡𝛽 + 𝜇𝑖𝑡
Where 𝑦𝑖𝑡 represents the index returns for time t. 𝑋𝑖𝑡 is a vector of explanatory variable minus the
constant. 𝛽 is a vector of coefficient parameters. We will estimate 𝛽 to test which factors have explanatory power to the sovereign emerging market bond returns. After running regression, we test for OLS assumptions and a multicollinearity issue.
We run regressions on the factors (Table 3) on RCs of selected models, EMBI spread return, and GBI-EM total return in USD. We select the significant factors based on the results of the correlation matrix (Appendix 4) to perform multiple regressions to quantify the effect of factors on our dependent variables.
Chapter 4: Study Results 4.0 Introduction
In this chapter, we report our findings in four sections. Section 4.1 reports the number of PCs from PCA extraction. Section 4.2 presents PCA results on “Country loadings”, which are the correlation between the variables and the Principal Components based on ten models. Section 4.3 describes RCs time-series movement of selected models. Finally, section 4.4 reported the OLS regressions results.
4.1 The number of PCs from PCA extraction
We selected the number of principal components (PCs) based on the eigen values using the parallel analysis as described in section 3.2. The scree plots of 10 Models are shown in Appendix 2. According to the scree plots of Model 1-5, the first few PCs can capture only 62%-68% of overall covariance. Since Absolute Models (Model 1-5) are designed to show the factors that systematically drive the overall emerging market bonds return in both hard currency and local currency simultaneously, there is a large residual of idiosyncratic countries that is not captured by the systematic risk. In other words, the idiosyncratic risk accounts for more than 30% of Hard Currency and Local Currency Bond returns.
4.2 Country loadings
According to the Model 1, high loadings of RC1 are the bond returns series in the EMBI Index and high loadings of RC2 are bond return series in the GBI-EM index. The RC1 and RC2 are clustered by the indices. Clearly, the RC1 represents Hard Currency Exposure while the RC2 represents Local
Currency exposures. Our interpretation is that the movement among intra-asset (EMBI vs GBI-EM)
class is higher than the inter-asset class. In model 2, the pattern still exists but not as strong as other
models. This pattern is also clear in model 1, 3, 4, and 5.
The correlations of the between RCs and the factors (Appendix 4.3) show that US treasury yields (2 year, 5 year, and 10 year) and FFR are the biggest differentiating factors among the two as RC1 has a positive correlation with US bond yields and RC2 has the negative one. The explanation of the differentiation in sensitivity to US treasury yield and FFR is that we use spread return for EMBI and total return for GBI-EM. Both RC1 (HC) and RC2 (LC) have the same sign of correlation with S&P (Positive) and DXY (negative), but RC1 is more sensitive to S&P, while RC2 is more sensitive to the Dollar Spot (DXY).
Figure 1
Model 1 Results
This figure reports the results of PCA on EMBI Spread Return and USD total return of GBI-EM. The darker shade bars on the left show the correlation between the variables and RC1 while the light shade bars on the right show the correlation between the variables and RC2. Proportion Variances explained by RC1 and RC2 are 40% and 24%, respectively.
Figure 2
Model 2 Results
This figure reports the results of PCA on EMBI Spread Return, Local Currency Total Return of GBI-EM and EM FX excluding carry. The darkest bars show the correlation between the variables and RC1 while the lightest bars show the correlation between the variables and RC2. The last bars show the correlations between the variables and RC3. Variances explained by RC1 and RC2 are 40% and 24%, respectively.
According to Model 2, three components are HC spreads, Local bonds, and FX, respectively. RC1 has high exposure to HC spreads, in line with Model 1, but it also includes exposures to Mexico, South Africa, Indonesia and Turkey FX, which suggest that these currencies represent spread-like behaviour. As in model 1, the factor has positive correlations to US yields, S&P and Commodities. Also, the correlation with Commodities is higher than that one with the oil.
The second factor is mostly concentrated in positive exposures to local bonds, somewhat lower to some FX: Colombia, Indonesia, Malaysia, South Africa, Thailand and Turkey, and also to Turkey spreads. It has negative correlations with US rates, which is natural as the factor is long EM rates. Also, it suggests Turkish HC spreads widen if rates go up, which is explainable by country’s external vulnerabilities. Factor’s correlations with the equities and commodities have an opposite sign compared to RC1. RC2’s correlation with the dollar is also negative but weaker than that of RC1. The third factor is interesting as it has exposures to all groups: HC spreads, Local rates, and FX. The highest exposures are to the currencies, but it also has high exposures to Brazilian, Chile, Colombia, Indonesia, Mexican, Russian and Venezuelan spreads and to Russian bonds. This factor’s largest correlation is the negative one with DXY. This is logical because EM currencies usually are negatively correlated to the dollar. Additionally, its correlation to the oil is higher than that of the first component of the model, suggesting that the currencies are more sensitive to the oil price than spreads.
According to the Model 3 (Appendix 3), high loadings of RC1 are the bond returns series in the EMBI Index and high loadings of RC2 are bond return series in local currency in the GBI-EM index. As with model 1, these two components (RC1 and RC2) represent HC and LC exposures. Egypt and Russia are two idiosyncratic countries in EMBI and GBI-EM, respectively.
For Model 4 (Appendix 3), high loadings of RC1 are the bond returns series in the EMBI Index and high loadings of RC2 are the Emerging market spot exchange rate. As with model 3, Egypt is an idiosyncratic country in EMBI.
We design Model 5 to be similar to Model 1. The only difference is the use of “total return” in place of “spread return” in the EMBI. The parallel analysis of this model suggests the extraction of 3 PCs where only 2 PCs for model 1.
According to model 5 (Appendix 3), the high loadings of RC1 are the bond returns series in the EMBI Index. This is similar to model 1 except there are more idiosyncratic countries, i.e. Argentina, Ecuador, and Pakistan. High loadings of RC2 are bond return series in the GBI-EM index, in line with model 1. However, Russia is no longer an idiosyncratic country in Model 5. The high loadings of RC3 are located in EMBI.
Table 5
Model 6-7 Results
This table reports the results of PCA for Model 6 and Model 7. For Model 6, this table shows the correlation between the variables and RCs of the PCA on EMBI Spread Return. Proportion Variances explained by RC1 and RC2 are 43% and 27%, respectively. For Model 7, this table shows the correlation between the variables and RCs of the PCA on EMBI Total Return. Proportion Variances explained by RC1 and RC2 are 46% and 24%, respectively.
Country
Model 6 (Spread Return) Model 7 (Total Return)
RC1 RC2 RC1 RC2 Brazil 89% 24% 89% 16% Colombia 88% 31% 89% 27% Mexico 86% 32% 90% 19% Peru 83% 41% 87% 31% Panama 82% 45% 87% 35% Indonesia 81% 39% 79% 39% Uruguay 80% 47% 84% 40% Turkey 79% 38% 82% 32% Philippines 79% 37% 82% 27% South Africa 76% 47% 83% 39% Chile 62% 40% 81% 10% Argentina 62% 51% 38% 64% Poland 61% 58% 74% 38% El Salvador 59% 65% 60% 61% Malaysia 57% 58% 77% 33% Russia 54% 62% 45% 67% Dominican Rep 52% 70% 52% 70% Hungary 46% 64% 50% 56% Venezuela 42% 58% 16% 69% Egypt 38% 20% 47% 14% Pakistan 29% 68% 22% 68% Ecuador 26% 75% 14% 77% Ukraine 15% 76% 4% 76%
According to both model 6-7, the countries with high loadings in RC1 seem to be geared towards the “main-stream/higher rated” countries, while the countries with high loadings in RC2 are geared more towards “idiosyncratic/lower rated” countries, but also Poland, Hungary, and Russia.
Table 6
Model 8-10 Results
This table reports the results of PCA for Model 8-10. For Model 8, this table shows the correlation between the variables and RCs of the GBI-EM total return in USD. Proportion Variances explained by RC1 and RC2 are 39% and 28%, respectively. For Model 9, this table shows the correlation between the variables and RCs of the
difference between EMBI spread return and USD total return of GBI-EM. Proportion Variances explained by
RC1 and RC2 are 30% and 27%, respectively. For Model 10, this table shows the correlation between the variables and RCs of the difference between EMBI total return and total return of GBI-EM in USD. Proportion Variances explained by RC1 and RC2 are 29% and 27%, respectively.
Country
Model 8 Model 9 Model 10
RC1 RC2 RC1 RC2 RC1 RC2 Indonesia 84% 12% 74% -3% -7% 67% Turkey 78% 33% 75% 25% 26% 73% Thailand 73% 9% Brazil 68% 41% 73% 16% 21% 75% South Africa 68% 43% 47% 57% 59% 41% Malaysia 68% 44% 54% 41% 47% 46% Colombia 64% 51% 64% 38% 43% 59% Mexico 58% 57% 61% 32% 37% 55% Poland 40% 79% 24% 85% 85% 24% Hungary 37% 74% 14% 85% 85% 9% Russia 4% 84% 12% 66% 68% 11%
When we perform PCA on GBI-EM total return in USD (Model 8), the countries are clustered by the common characteristics. For RC1, some of these countries are high yielders: Brazil, Indonesia, and Turkey. Also, some of these countries are located in South East Asia: Indonesia, Malaysia, and Thailand. For RC2, these countries are in Eastern Europe: Russia, Poland, and Hungary.
The top three countries with the highest loading in both RC1 and RC2 of model 9 are the same as model 8. For RC1, countries with high loadings are Indonesia, Brazil, and Turkey. These are HY countries. For RC2, high loading countries are Poland and Hungary. The economy of Poland and Hungary are linked with EU currency, which is negatively related to USD, which in turn is negatively related to a commodity.
The design of Model 10 is similar to Model 9. The difference is the use of “total return of EMBI” instead of “spread return of EMBI”. The countries with high loadings on RC1 of Model 5 are similar to those of RC2 of Model 9. Since the proportion of variance explained by RC1 and RC2 is roughly the same in Model 5, the level of variance captured by countries with high loadings on both RCs is similar. We only have to be careful when relating loading results with the correlation table. RC1 of
Model 5 should associate to factors in the same fashion with RC2 of Model 9 and the same should go for RC2 of Model 5 and RC1 of Model 9 (Appendix 4). This is the case when we look at the results in the correlation table. Except for US yield, all factors have the same sign with RC1 of Model 5 and RC2 of Model 9. RC1 of Model 9 has a more pronounced negative correlation with oil and S&P, while RC2 of Model 5 correlates with US Yields. This is intuitive as the duration of the spread return is lower than the duration of total return. During the period of this study, the mean of spread duration of EMBI Global Diversified is 6.667 while the mean of IR duration is 6.867. This can be attributable to the difference in “total return” vs “spread return” of EMBI.
4.3 RCs time-series movement
Figure 3
Time-series RCs movement of selected models
These figures show the time-series movement of RCs on selected Absolute and Relative Models. The upper figure shows the time-series plots of RC1 and RC2 of Absolute Models (Model 1-4), respectively. The plots show the how factors that systematically drive the overall emerging market bonds return in both hard currency and local currency simultaneously changes between 2005 and 2017. The lower figures show RC1 and RC2 of Relative Models (Model 9-10), respectively. The plots show the how factors that differentiate the return between hard-currency changes between 2005 and 2017.
80 85 90 95 100 105 1/1 /2005 9/1 /2005 5/1 /2006 1/1 /2007 9/1 /2007 5/1 /2008 1/1 /2009 9/1 /2009 5/1 /2010 1/1 /2011 9/1 /2011 5/1 /2012 1/1 /2013 9/1 /2013 5/1 /2014 1/1 /2015 9/1 /2015 5/1 /2016 1/1 /2017 Model_1_RC1 Model_2_RC1 Model_3_RC1 Model_4_RC1 80 85 90 95 100 105 110 115 120 1/1 /2005 9/1 /2005 5/1 /2006 1/1 /2007 9/1 /2007 5/1 /2008 1/1 /2009 9/1 /2009 5/1 /2010 1/1 /2011 9/1 /2011 5/1 /2012 1/1 /2013 9/1 /2013 5/1 /2014 1/1 /2015 9/1 /2015 5/1 /2016 1/1 /2017 Model_1_RC2 Model_2_RC2 Model_3_RC2 Model_4_RC2 80 85 90 95 100 105 1/1 /2005 9/1 /2005 5/1 /2006 1/1 /2007 9/1 /2007 5/1 /2008 1/1 /2009 9/1 /2009 5/1 /2010 1/1 /2011 9/1 /2011 5/1 /2012 1/1 /2013 9/1 /2013 5/1 /2014 1/1 /2015 9/1 /2015 5/1 /2016 1/1 /2017 Model_9_RC1 Model_10_RC1 80 85 90 95 100 105 1/1 /2005 9/1 /2005 5/1 /2006 1/1 /2007 9/1 /2007 5/1 /2008 1/1 /2009 9/1 /2009 5/1 /2010 1/1 /2011 9/1 /2011 5/1 /2012 1/1 /2013 9/1 /2013 5/1 /2014 1/1 /2015 9/1 /2015 5/1 /2016 1/1 /2017 Model_9_RC2 Model_10_RC2
RC1 of Model 1 and 2 is the vector of simultaneous movement in EMBI spread return and GBI-EM local return in USD. Intuitively, RC1 of Model 1 and 2 should show the same movement because they both represent Hard Currency Exposure. According to the graph, they behave as expected.
To distinguish whether the volatility of FX or local rates dominated the volatility of sovereign EM local bond return, we should pay attention to RC1 of Model 3 and 4. Since RC1 of Model 4 (excluding carry) shows a highly correlated pattern with Model 1 and 2, we may conclude that the volatility of local rates does not have a high impact on the volatility of USD return of local bonds. In contrast, RC1 of Model 3 (excluding FX) deviates from RC 1 of Model 1 and 2. Therefore, we can conclude that the volatility of FX spot dominated the volatility of return more than the volatility of the local rates.
The movement of RC2 seems to be divided into two subgroups. The RC2 of model 2 and model 3 is almost the same across the horizon while the RC2 of model 1 and model 4 exhibits a very similar pattern.
RC1 and RC 2 of both absolute and relative models suggest that the factors that determine the countries with high loadings in different models differ in short-term but long-term trends are similar.
4.4 OLS regressions results
We run multiple OLS regressions to see the coefficient of our variables of interest. The results of the OLS regressions are shown in Table 7 and Table 8. As we have tested, our explanatory variables are not subject to multicollinearity.
In model 9, the two factors clearly distinguish between different environments: one of rising rates (Model 9 RC1) and one of stronger US dollar (Model 9 RC2). Both factors have negative exposure to S&P, but different exposures to commodities: negative for RC1 and positive for RC2. Also, RC1 has significant exposure to Ted Fund Rate, while RC2 has it insignificant.
Thus, RC1 identifies an environment when US treasury rates rise in both long and short end, and commodity prices slightly fall. RC2 identifies an environment of a rise in the dollar and rising commodity prices.
Finally, S&P has a negative exposure on both factors in model 9 which on a standalone basis would suggest preferring Local Currency Bond vs Hard Currency Bond. However, S&P is not coming out as a strong significant factor in these models, which makes its impact difficult to interpret.
The results clearly suggest that the US treasury, the dollar and commodities have an important impact on the HC vs Local relative value. S&P impact is unclear and is being overwhelmed by the yields and the dollar.
Table 7
Regressions of Factors on RCs of Relative Models
This table reports the results of regressions of selected factors on RCs of Model 9 and Model 10. Selected factors are US10YY (the yield of 10-year US government bond index), Commodity Index, S&P 500 (US Equities), DXY (USD spot), FFR (Fed Fund Rate). For FFR, the quantitative easing policy has kept the Fed Fund Rate at the low level irresponsive to the economy condition. Therefore, we use Federal Funds Rate until December 2006 and US 1Y1Y Swap Rates afterward. Model 9 RCs are the results of PCA on the difference between EMBI spread return and USD total return of GBI-EM. Model 10 RCs are the results of PCA the difference between EMBI total return and total return of GBI-EM in USD.
Model 9_RC1 Model 9_RC2 Model 10_RC1 Model 10_RC2
(Intercept) Estimate 0.074 -0.017 -0.025 0.061 p-Value 0.33 0.796 0.705 0.462 US10YY Estimate 1.747 0.071 -0.858 -0.561 p-Value 0.000 0.84 0.014 0.204 Commo Estimate -4.651 2.147 2.679 -5.894 p-Value 0.096 0.372 0.262 0.053 S&P500 Estimate -4.155 -3.951 -4.429 -5.313 p-Value 0.066 0.043 0.023 0.031 DXY Estimate -0.125 25.389 24.25 -0.403 p-Value 0.975 0.000 0.000 0.925 FFR Estimate 0.609 0.254 0.204 0.711 p-Value 0.151 0.487 0.574 0.124 N 144 144 144 144 R2 0.272 0.457 0.466 0.136 adjusted R2 0.246 0.438 0.446 0.105
We run regressions on both absolute models and relative models (Table 8). For absolute models, the dependent variables are either hard currency bond return or local currency bond return in USD. We run both spread return and the total return of hard currency bond return (EMBI index). For relative models, we run regressions on the difference between hard currency bond return and local currency bond return.
The results of regression EMBI total return on selected factors suggest that the significant factors are US treasury yield, commodity, US equities, the dollar index, and Fed Fund Rate. When we regress EMBI spread return index on the same factors, we found that the coefficients of most factors do not change from the EMBI total return model. Intuitively, the US treasury yield was no longer significant in the EMBI spread return. The reason for this is that the interest rate duration is extracted from the spread return but is included in the total return.
The significant factors explaining the USD return of local currency bond, which is represented by the GBI-EM index, are similar to the absolute model of EMBI total return index. These factors are US treasury yield, commodity, US equities, dollar index. Also, the Fed Fund Rate is not significant. Regarding the results of the relative model, when we take the difference between EMBI return minus GBI-EM return in USD, the significant factors are Dollar Index (DXY) and Fed Fund Rates. Intuitively, higher US 10-year bond yield is negative to both EMBI and GBI EM total return but is not negative to EMBI Spread return. So, the relative model has it significant on EMBI spread return minus GBI-EM. Dollar Appreciation (DXY) is negative to EMBI less than GBI-EM because GBI-EM return in USD is the function of local bond yield and foreign exchange with inverse relation to USD.
Table 8
Regressions of Factors on EMBI and GBI-EM
This table reports the results of regressions of selected factors on both Absolute Models and Relative Models, respectively. Selected factors are US10YY (the yield of 10-year US government bond index), Commodity Index, S&P 500 (US Equities), DXY (USD spot), FFR (Fed Fund Rate). For FFR, the quantitative easing policy has kept the Fed Fund Rate at the low level irresponsive to the economy condition. Therefore, we use Federal Funds Rate until December 2006 and US 1Y1Y Swap Rates afterward. We use EMBI Total Return, EMBI Spread Return, and GBI-EM Total Return for Absolute Models. For Relative Models, the dependent variables are the
difference between EMBI total/spread return and total return of GBI-EM in USD.
Absolute Model Relative Model
EMBI Total Return EMBI Spread Return GBIEM Total Return EMBI_Tot_mi_ GBIEM EMBI_Spr_mi_ GBIEM (Intercept) Estimate 0.003 0.001 0.002 0.001 -0.001 p-Value 0.017 0.623 0.284 0.381 0.428 US10YY Estimate -0.052 0.009 -0.037 -0.015 0.047 p-Value 0.000 0.252 0.000 0.071 0.000 Commo Estimate 0.124 0.133 0.144 -0.02 -0.01 p-Value 0.017 0.018 0.038 0.717 0.858 S&P500 Estimate 0.313 0.339 0.39 -0.078 -0.051 p-Value 0.000 0.000 0.000 0.088 0.281 DXY Estimate -0.191 -0.181 -0.637 0.446 0.456 p-Value 0.009 0.022 0.000 0.000 0.000 FFR Estimate 0.014 0.014 -0.003 0.017 0.017 p-Value 0.073 0.103 0.753 0.044 0.055 N 144 144 144 144 144 R2 0.581 0.603 0.661 0.389 0.531 adjusted R2 0.566 0.589 0.648 0.367 0.514
Chapter 5: Discussion
5.0 Introduction
This section describes the conclusion and limitation of this thesis. It is divided into three sections. In Section1, we discuss the conclusion of this study. Subsequently, Section 2 relates our findings with the existing literature. Finally, the implication and limitation of this thesis are presented in Section 3.
5.1 Conclusion of our study
The goal of this thesis is to identify the common factors that explain return variation in both hard currency and local currency emerging market sovereign bond. We used the principal component analysis (PCA) to cluster the countries based on its bond returns. Interestingly, PCA clustered countries across all our models with the universe of GBI-EM into two groups: high yield countries and Eastern European countries. Countries with high loadings are consistent across models but countries with medium loadings change in each model.
In addition, we obtain the time-series of components from the PCA analysis. We calculate the correlation of these time-series of components with the common factors that drives the bond market return such as bond yield. Based on this correlation, we selected factors that are highly correlated with each component to be independent variables in the regression. We regress bonds indices and components on these independent variables. The regression results suggested that significant factors for hard currency bond and local currency bond are US bond yield, commodity, US dollar index, and US equities. In relative models, where we subtract local currency bond return from the hard currency bond return, the differentiating factors are US bond yield, US dollar index, and Fed Fund Rate.
5.2 Contribution to existing Literature
To our knowledge, our study is the first to report findings regarding the loadings by PCA analysis in Model 1 to 10. For Model 1 – 5, where we use PCA on both EMBI and GBI-EM simultaneously, the components are clustered based on hard currency exposure, local bonds, and FX. For Model 6-10, where we use PCA on both absolute return of individual index (either EMBI or GBI-EM) and relative return, the countries are clustered based on its characteristics: high – yield countries and Eastern European countries.
The regression results suggested that significant factors for hard currency bond and local currency bond are US bond yield, commodity, US dollar index, and US equities. Our results are intuitive and in line with other literature studying bond return and bond spread of Emerging Market Sovereign Countries.
Since we have not found the study reports factors differentiating relative return between hard currency and local currency bond, our relative model results can be added to the literature studying bond return.
5.3 Implication and limitation of our study
In model 1-5, where we extract PCs for both HC and LC bond simultaneously. The first few PCs capture only 62%-68% of overall covariance. This suggested that there is a large residual of idiosyncratic countries that is not captured by the systematic risk. In other words, global factors accounted for less than percent of the variation in the return of the emerging sovereign bond. In order to invest based on the factors that we have, the investor should be aware of this.
The goal of this thesis is to identify the common factors that explain return variation in both hard currency and local currency emerging market sovereign bond. This thesis is not a return predictability study. We aim to explore non-specific factors that describe the variation in emerging market sovereign bond returns. The objective of finding factors to generate signals to invest on the basis of these signals is out of the scope of this thesis.
References
Adler, M., & Song, J. (2010). The behavior of emerging market sovereigns' credit default swap premiums and bond yield spreads. International Journal of Finance & Economics,
15(1), 31-58.
Arora, V. B., & Cerisola, M. D. (2000). How Does U.S. Monetary Policy Influence Sovereign Spreads in Emerging Markets? . IMF Working Papers, 48(3), 474-498. Retrieved April 6, 2017, from https://www.imf.org/external/pubs/ft/staffp/2001/03/pdf/arora.pdf.
Bunda, I., Hamann, A. J., & Lall, S. (2010). Correlations in emerging market bonds: the role of local
and global factors. Washington, D.C.: International Monetary Fund.
Csonto, B., & Ivaschenko, I. V. (2013). Determinants of Sovereign Bond Spreads in Emerging Markets: Local Fundamentals and Global Factors vs. Ever-Changing Mis-alignments. IMF
Working Papers, 13(164), 1.
Fabozzi, F. J. (2002). Fixed income securities. New York: John Wiley.
Ferrucci, G. (2003). Empirical Determinants of Emerging Market Economies' Sovereign Bond Spreads. SSRN Electronic Journal, 205.
Gupta, S., Mati, A., & Baldacci, E. (2008). Is it (Still) Mostly Fiscal? Determinants of Sovereign Spreads in Emerging Markets. IMF Working Papers, 08(259), 1.
Houweling, P., & Zundert, J. V. (n.d.). Factor Investing in the Corporate Bond Market. SSRN
Electronic Journal. doi:10.2139/ssrn.2516322
Jaramillo, L., & Weber, A. (2013). Global Spillovers into Domestic Bond Markets in Emerging Market Economies. IMF Working Papers, 13(264), 1.
Kabacoff, R. (2015). R in action: data analysis and graphics with R. Shelter Island: Manning.
Kashiwase, K., & Kodres, L. E. (2008). Emerging Market Spread Compression: Is it Real or is it Liquidity? SSRN Electronic Journal.
Kim, G., Ram, K., & Patel, K. (2015, December 21). EMBI Global and EMBI Global (Publication).
Retrieved February 15, 2017, from J.P. Morgan website:
https://markets.jpmorgan.com/#research.emerging_markets.index
Kim, G., & Pithavadian, S. (2015, September 3). Government Bond Index- Emerging Markets Family of Indices: Rules and Methodology. Retrieved April 10, 2017, from https://markets.jpmorgan.com/#research.page.app_base&page=gbi_em
Ledesma, R. D., & Valero-Mora, P. (2007). Determining the Number of Factors to Retain in EFA: an easy-to-use computer program for carrying out Parallel Analysis. Practical Assessment,
Research & Evaluation., 12(2). Retrieved April 10, 2017, from
http://pareonline.net/pdf/v12n2.pdf
Linzie, J., Kumaran, R., & Patel, K. R. (2017, March). J.P. Morgan Markets: Emerging Markets Bond
https://markets.jpmorgan.com/research/PubServlet?action=open&doc=GPS-2299879-0.pdf Longstaff, F., Pan, J., Pedersen, L., & Singleton, K. (2007). How Sovereign is Sovereign Credit Risk?
doi:10.3386/w13658
Min, H. G. (1999). Determinants of Emerging Market Bond Spread: Do Economic Fundamentals Matter? Policy Research Working Papers. doi:10.1596/1813-9450-1899
Min, H., Lee, D., Nam, C., Park, M., & Nam, S. (2003). Determinants of emerging-market bond spreads: Cross-country evidence. Global Finance Journal, 14(3), 271-286.
Pratiwi, A. S. (2015). Understanding Emerging Market Sovereign Bond Yield Spread: Role of Default and Non-Default Determinants. Indonesian Capital Market Review, 7(1).
Rajan, A., & Lawlor, M. (2012, September). Measuring and Managing Emerging Markets Fixed Income Risk - Prudential Financial. Retrieved January 17, 2017, from
http://www.pgimfixedincome.com/pramerica/fi/pdf/pramerica-fixed-income-whitepaper-em-fi-risk-9-12.pdf
Rojas, A., & Jaque, F. (2002). Determinants of the Chilean Sovereign Spread: Is it Purely Fundamentals? Retrieved April 1, 2017, from
http://www.cemla.org/red/papers2002/RED_VII_CHILE-Jaques-Rojas.PDF
Rowland, P., & Torres, J. L. (2004). Determinants of Spread and Creditworthiness for Emerging Market Sovereign Debt: A Panel Data Study. Retrieved April 1, 2017, from
http://www.banrep.gov.co/docum/ftp/borra295.pdf
Siklos, P. L. (2011). Emerging market yield spreads: Domestic, external determinants, and volatility spillovers. Global Finance Journal, 22(2), 83-100.
Sy, A. N. (2002). Emerging market bond spreads and sovereign credit ratings: reconciling market views with economic fundamentals. Emerging Markets Review, 3(4), 380-408.
Appendix 1: The selected countries for PCA analysis
Table 9
The List of Selected Countries from EMBI
This table shows the list of selected 23 countries and their weight based on the J.P. Morgan Emerging Markets Bond Index Global Diversified Index (EMBI) as of January 2017. The EMBI index contains 67 countries. We deselected 44 countries because of missing data and discarded China and Lebanon based on liquidity issues.
Selected countries Index Weight
J.P. Morgan EMBI Global Diversified Argentina Spread Index 3.78% J.P. Morgan EMBI Global Diversified Brazil Spread Index 5.15% J.P. Morgan EMBI Global Diversified Chile Spread Index 2.38% J.P. Morgan EMBI Global Diversified Colombia Spread Index 2.92% J.P. Morgan EMBI Global Diversified Dominican Rep Spread Index 1.27% J.P. Morgan EMBI Global Diversified Ecuador Spread Index 1.07% J.P. Morgan EMBI Global Diversified Egypt Spread Index 0.37% J.P. Morgan EMBI Global Diversified El Salvador Spread Index 0.63% J.P. Morgan EMBI Global Diversified Hungary Spread Index 2.04% J.P. Morgan EMBI Global Diversified Indonesia Spread Index 7.82% J.P. Morgan EMBI Global Diversified Malaysia Spread Index 2.15% J.P. Morgan EMBI Global Diversified Mexico Spread Index 12.91% J.P. Morgan EMBI Global Diversified Pakistan Spread Index 0.62% J.P. Morgan EMBI Global Diversified Panama Spread Index 1.66% J.P. Morgan EMBI Global Diversified Peru Spread Index 1.72% J.P. Morgan EMBI Global Diversified Philippines Spread Index 3.88% J.P. Morgan EMBI Global Diversified Poland Spread Index 1.83% J.P. Morgan EMBI Global Diversified Russia Spread Index 6.73% J.P. Morgan EMBI Global Diversified S. Africa Spread Index 2.49% J.P. Morgan EMBI Global Diversified Turkey Spread Index 6.32% J.P. Morgan EMBI Global Diversified Ukraine Spread Index 1.96% J.P. Morgan EMBI Global Diversified Uruguay Spread Index 1.42% J.P. Morgan EMBI Global Diversified Venezuela Spread Index 4.03%
Table 10
The List of Selected Countries from GBI-EM
This table shows the list of selected 11 countries and their weight based on the J.P. Morgan Government Bond Index – Emerging Markets Index (GBI-EM) as of March 2017. The EMBI index contains 17 countries. The GBI-EM index contains 17 countries. We deselected 6 countries because of missing data.
Selected countries Index Weight
J.P. Morgan GBI-EM Broad Brazil USD Unhedged 24.1% J.P. Morgan GBI-EM Global Colombia USD Unhedged 5.6%
J.P. Morgan GBI Hungary Unhedged USD 3.4%
J.P. Morgan GBI-EM Indonesia Unhedged USD 9.2% J.P. Morgan GBI-EM Malaysia Unhedged USD 5.3%
J.P. Morgan GBI Mexico Unhedged USD 13.7%
J.P. Morgan GBI-EM Poland USD Unhedged 9.2% J.P. Morgan GBI-EM Russia Unhedged USD 4.4% J.P. Morgan Govt Bond Unhedged USD South Africa 8.1% J.P. Morgan GBI-EM Thailand Unhedged USD 5.3% J.P. Morgan GBI-EM Turkey Broad USD Unhedged 6.0%
Appendix 2: Scree Plots of 10 Models
Figure 4
Scree Plots of 10 Models
These figures show the scree plots of our models. For Model 1, the cumulative variance explained by two PCs is 64%. The proportional variances for the 1st PC and 2nd PC are 54% and 10%, respectively. For Model 2, the
cumulative variance explained by three PCs is 63%. The proportional variances for the 1st PC, 2nd PC, and 3rd PC are 46%, 11%, and 7%, respectively. For Model 3, the cumulative variance explained by two PCs is 62%. The proportional variances for the 1st PC and 2nd PC are 49% and 13%, respectively. For Model 4, the
cumulative variance explained by two PCs is 64%. The proportional variances for the 1st PC and 2nd PC are
55% and 9%, respectively. For Model 5, the cumulative variance explained by three PCs is 68%. The proportional variances for the 1st PC, 2nd PC, and 3rd PC are 53%, 8%, and 7%, respectively. For Model 6, the
cumulative variance explained by two PCs is 71%. The proportional variances for the 1st PC, 2nd PC are 65%,
and 6%, respectively.For Model 7, the cumulative variance explained by two PCs is 70%. The proportional variances for the 1st PC, 2nd PC are 60%, and 10%, respectively. For Model 8, the cumulative variance
explained by two PCs is 68%. The proportional variances for the 1st PC, 2nd PC are 58%, and 10%,
respectively. For Model 9, the cumulative variance explained by two PCs is 58%. The proportional variances for the 1st PC and 2nd PC are 45% and 13%, respectively. For Model 10, the cumulative variance explained
Appendix 3: Results of Model 3-5
Figure 5
Model 3 Results
This figure reports the results of PCA on EMBI Spread Return and Local Currency Total Return of GBI-EM. The darker shade bars on the left show the correlation between the variables and RC1 while the lighter shade bars on the right show the correlation between the variables and RC2. Proportion Variances explained by RC1 and RC2 are 44% and 18%, respectively.
Figure 6
Model 4 Results
This figure reports the results of PCA on EMBI spread return and EM FX (excluding carry) using GBI-EM universe. The darker shade bars on the left show the correlation between the variables and RC1 while the lighter shade bars on the right show the correlation between the variables and RC2. Proportion Variances explained by RC1 and RC2 are 40% and 24%, respectively.
Figure 7
Model 5 Results
This figure reports the results of PCA on EMBI total return and USD total return ofGBI-EM total return in USD.The darkest bars show the correlation between the variables and RC1 while the lightest bars show the correlation between the variables and RC2. The last bars show the correlations between the variables and RC3. Variances explained by RC1, RC2 and RC3 are 34%, 19% and 15%, respectively.
