i

Final Thesis:

EMU Sovereign Bond Yields Valuation

Mathieu MINAUD R. JOLIET

Academic Year 2015 - 2016

ii

Final Thesis:

EMU Sovereign Bond Yields Valuation

Mathieu MINAUD R. JOLIET

Academic Year 2015 - 2016

Iéseg School of Mangement

3, Rue de la Digue

59 000 Lille L’Iéseg n’entend donner aucune approbation ni improbation aux opinions émises dans les mémoires; ces

opinions doivent être considérées comme proper à leurs auteurs.

i

Acknowledgements

I would like to thank the Asset Management Department of the Company Davy in Dublin for

introducing me to Fixed-Income valuation models and especially Chantal Brennan and Oliver

Sinnott for their support and comments on my work.

I am grateful to Deutsche Bank and especially to Gaetan Toulemonde for making me work on

my first research projects.

Special Thanks to the Ieseg Master in Finance department and especially Robert Joliet, my

thesis coordinator for his help and guidance toward the writing of this paper.

ii

Summary

1. LITTERATURE REVIEW ................................................................................................................................. 6

1.1 Fundamental dynamics of valuation of sovereign spreads............................................................................................... 7

1.2 Fundamental economic indicators and sovereign bonds .................................................................................................. 8

1.3 Fundamental liquidity and sovereign bond spreads ....................................................................................................... 12

1.4 Trade-off liquidity – Quality ......................................................................................................................................... 15

1.5 Extra fundamental variables .......................................................................................................................................... 18

2. METHODOLOGY .............................................................................................................................................22

2.1 Choice of Variables ....................................................................................................................................................... 22

2.2 Definition of the model ................................................................................................................................................. 32

2.3 From pillar specific to combined ................................................................................................................................... 35

2.4 Model analysis ............................................................................................................................................................... 36

3. DATA ANALYSIS ............................................................................................................................................38

3.1 Yield Spreads ................................................................................................................................................................ 38

3.2 Fundamental Economic Data ......................................................................................................................................... 39

3.3 Fundamental Liquidity Output ...................................................................................................................................... 44

3.4 Quantifiable Extra-Fundamental Output: ...................................................................................................................... 47

3.5 Fundamental Relative Economic Grade Model ............................................................................................................. 50

3.6 Fundamental Relative Liquidity Grade Model .............................................................................................................. 53

3.7 Extra-Fundamental Relative Grade Model .................................................................................................................... 54

3.8 Time-frame of significance ........................................................................................................................................... 57

3.9 Combined model ........................................................................................................................................................... 57

3.10 The influence of credit ratings ..................................................................................................................................... 59

3.11 Interpretations and implication .................................................................................................................................... 60

3.12 Limits of the study and feedback ................................................................................................................................. 61

4. RESULTS ANALYSIS ......................................................................................................................................62

4.1 Clusters .......................................................................................................................................................................... 62

4.2 Where are we in the cycle? ............................................................................................................................................ 63

4.3 Crises effect on sovereign bond valuation ..................................................................................................................... 64

4.4 Back testing Returns ...................................................................................................................................................... 65

4.5 Return uses and performance implication...................................................................................................................... 69

4.6 Limits of the model ....................................................................................................................................................... 69

4.7 Beyond the model .......................................................................................................................................................... 70

iii

Abstract

The European Monetary Union is a single-currency area regrouping countries with

various financial and economic features. Each country needs to finance its expenditure

through financial markets.

Our study focus on the understanding of the main valuation dynamics of sovereign

bonds inside the Eurozone using spreads relative to Germany.

The valuation of spreads inside the EMU lays on three main pillars; fundamental

economic attributes, fundamental liquidity features and extra-fundamental variables.

In order to assess those valuation dynamics we tested assumed explanatory variables

for each variables provided by Bloomberg Datastream, OECD Database, BIS Database and

Eurostat. Once the preliminary testing done we combined those variables into pillar specific

models. Those models then contributed to the creation of combined model regrouping each

pillar.

The significance of our various models varies over time and captures part of the

changes in investment behaviors through time.

Each model provides an estimate residual value of spreads for 10 countries over 14

years monthly, those estimates could be used in order to create investment cases or to assess

market opportunities.

iv

Introduction

Since its inception in 1999 the European Monetary Union is an economic area with a

unique feature: various states with different economic conditions and wealth generation

capabilities share a single currency. Uniting 19 different countries into a single currency area

reveals itself challenging as by adopting the single-currency each country had to give up

sovereignty of its monetary policy to a single supranational central bank: the ECB.

Our study will focus on the financing of the EMU state members on financial markets,

each state finance an important part of its expenditures by emitting public debt: sovereign

bonds. The aim of our study will be to understand the shifts in dynamics of valuation of EMU

sovereign bond spreads over time and the external and internal events which could trigger

those shifts.

We will analyze some recent empirical studies concerning the valuation of sovereign

bonds in the emerging markets as well as the EMU. Highlighting the fundamental dynamics

of bond valuation, its economic parameters, its liquidity features, as well as its extra-

fundamental variables. The fundamental dynamics of bond valuation divide interest rates

between risk premium and expected loss from default. Fundamental economic parameters are

linked to credit quality, how a specific country is able to found its debt and service it.

Liquidity features are composed by market depth and price elasticity to shifts in demand.

Economic and liquidity variables are significantly explaining the fluctuations of bond prices,

however they get more significant than each other according to the economic environment.

Bond prices cannot be explained simply using fundamental variables, some other parameters

influence fixed-income valuation, such as risk aversion and contagion.

In order to observe the main dynamics influencing bond valuation in the EMU, we

will need to assess the various explanatory parameters that are in play in the fixed-income

markets. Corporate and sovereign bonds have some common features and can be compared in

order to understand fixed-income dynamics; however the sovereign obligation market is

unique by the volume of securities exchanged and the fact that states cannot be compared to

companies in term of financial characteristics and structure.

2

Once our different explanatory variables tested, we will proceed to the creation of

combined models, in order to maximize our explanatory power and significance, those

combined models will provide us with estimated residual values of spreads which would be

useful indicators in order to understand the valuation dynamics of sovereign bonds as well as

predict their future fluctuations. We will then proceed to the analysis of our results and their

potential application towards investment.

In order to study those variations of yields, we are going to use Germany as a risk free

value for yields inside the EMU and observe the fluctuations of the other 9 Eurozone state

members over time and regressing different assumed explanatory datasets.

A large part of the primary data we are going to observe will be provided by

Bloomberg Data stream history function, Bloomberg keeps an accurate history of data stream

information as well as central banks and institution publications, some information such as

average daily bid-ask spread are directly available and computed by Bloomberg and therefore

requires a more detailed description as it is not primary data. Additional data will be provided

by the OECD data base, MSCI Corporation and Bank of International Settlements (BIS)

statistic data base.

The study will start on the 31/10/2001 and end on the 30/03/2015, including 162

monthly data. It will be made on this time range for 10 country spreads: Austria, Belgium,

Finland, France, Germany, Ireland, Italy, Netherlands, Portugal and Spain. It will give us

1620 yield spreads observed on which we will test 24 different independent variables ranging

from monthly to yearly frequency. We will assume a continuous growth rate for the quarterly

and yearly data.

3

1. PREVIOUS EMPIRICAL STUDIES

1.1 Fundamental dynamics of valuation of sovereign spreads

Spreads versus a risk free asset are a sum of two indivisible components: Expected

loss from default and Risk premium. Expected loss from default represents the amount of

value a specific bondholder will have reduced on its asset in case of default of the entity

which issued the obligation. The risk premium is the pricing compensation of a specific

bondholder for taking an extra amount of risk compared to a risk free obligation. Those two

components as the fundamentals variables of spreads valuation are used in many sovereign

and corporate bonds valuation models.

Credit risk are used as a standard in order to proxy the probability of default, it is

based on some fundamental economic parameters, in order to value credit risk, most investors

use credit ratings built by some grading agencies such as Moody’s and Standard and Poor’s.

Grading agencies are using those fundamental economic data, macroeconomic tendencies as

well as policy studies in order to create those scaled ratings.

Remolona, Scatigna and Wu (2007) showed that the most important part of the spreads

of emerging markets sovereign bonds versus risk-free rates can be explained by the risk

premium (80%), while expected loss from default would only represent 20% of the volatility

of spreads.

4

Indeed, using a regression built on fundamental economic parameters such as credit

ratings to define expected loss from default, a significant correlation was found. However it

appears that there is a non-linear correlation between the computed expected loss and spreads,

the lower the rating, the wider the spread.

The result of the study showed us the explanatory power of a variable such as credit

ratings on sovereign spreads, however it also indicates that the classical structure of spread

valuation may insufficient to explain the whole volatility of bonds. Indeed the “risk premium”

mentioned by Remolona, Scatigna and Wu (2007) constitutes an unknown parameter which is

obtained by making the difference between spreads value and expected loss from default.

The study of the fundamentals of bond valuation will help us understand the

unexplained part of bond valuation that needs to be studied through explanatory research.

Risk premium is the main component of interest rates that needs some deeper research

analysis in order to fully understand the dynamics of bond valuation. The lack of explanatory

variables assessed by the author also explains the low R squared found during the analysis.

It seems that there are more than two explanatory variables influencing the volatility

of spreads, we need to define what the risk premium is and how it behaves. In order to do so,

we need to assess some significant variables, the most obvious one are the fundamental

economic indicators that we started to observe with credit ratings which were included in the

analysis made by Remolona, Scatigna and Wu (2007).

5

1.2 Fundamental economic indicators and sovereign bonds

Earlier empirical researches on sovereign bond spreads tried to build up a structural

model made of factors that could influence the valuation of obligations according to

explanatory factors, linked to the macro-economic environment, country specific indicators or

other non-fundamental indicators.

Min (1998) is one of the precursors of empirical testing of group of variables clustered

in different categories to analyze the bond volatility of fixed-income assets, he observed

emerging markets obligations from 1990 to 1998. He studied various variables clustered in 4

different groups: “Liquidity and Solvency variables”, “Macroeconomic fundamentals”,

“External shocks” and “Dummy variables”.

Two groups of parameters were found to influence significantly the yield

spreads:”Liquidity and solvency variables” and “macroeconomic fundamentals”. According

to the empirical results of the study some indicators such as debt to GDP, international reserve

to GDP, inflation rate of growth, current accounts and terms of trade have a significant link to

yield spreads in the emerging markets.

6

The results of the model are comforting us in the idea that fundamental economic

metrics can be used to value yield spreads. Fundamentally states will need to create income in

order to be able to service and reimburse its debt. Hence economic variables are good

indicators of the capabilities of states to generate this income, indeed indicators such as GDP

growth or debt to GDP are directly linked to the debt financing problem, a state whose wealth

doesn’t grow will have some budgeting issues to service its debt, while a state whose wealth

is growing will be able to generate more revenue through taxation. Debt levels are good

indicators of cost of servicing the debt and also the level of indebtedness will give insight

about how close to default a state is. However debt levels are also impacting some other

metrics such as liquidity which can have both positive and negative influence on spreads.

The link between “macroeconomic fundamentals” and yield spreads is the first

hypothesis made up in our analysis, a proven linkage between emerging market yields and

their fundamental economic data reinforce our assumption that the same analysis could be set

up on EMU countries.

Westphalen (2001) follows up on Min (1998) by developing an empirical testing of

various parameters assumed to have positive or negative effects on emerging market spreads.

The study focused first on economic assumptions on effects of various indicators on yield

spreads and then empirically test the hypotheses to check if each variable influence emerging

sovereign bonds spreads in the way it is assumed to be.

According to the study each of the indicators which are “Spot rate”, “Yield curve

slope”, “Distance to default”, ”Volatility” and “World economy” is significant according to

the assumptions made by the model, however the model fails to explain 85% of the variations

of the spreads (89.2% in Emerging Europe.).

7

Westphalen (2001) proved that those 85% of fluctuations of spreads are explained by a

single common unknown factor by analyzing the residuals of his regression, suggesting that

some non-country specific features are influencing the valorization of bonds. Other factors

could explain the R squared found in the analysis, the study of economic variables only as a

meaning of valuation of bond spreads can be different depending on the economic cycle

position the world economy is in the range studied.

The results of the study shows significance of fundamental economic parameters,

however, Westphalen (2001) would perhaps had better significance of his model if he had

tested more variables expected to have explanatory link with spreads. Indeed he regressed

mostly factors known to have a direct impact on yields such as credit quality.

Focusing on expected direct relationships between variables may narrow us too much

on the analysis, while testing multiple indirect parameters such as Min (1998) showed would

result in some parameters having no significant effect, but also in a better explanatory

analysis of yield fluctuations.

The insights given by Min (1998) and Westphalen (2001) is crucial to determine the

variables we will be using to assess credit quality but also to define the model as a whole.

Indeed the explanatory variables used in their models are significant in their range of study,

and there is no recent literature assessing a change of significance of those variables in the

recent years. Even if structure of investment behaviors change over time some fundamental

metrics are crucial to assess risk-exposure and value of assets.

8

However the question is to know if as most of the recent studies focus on spreads of

emerging compared to risk free assets such as developed market, we could ask ourselves the

question if such indicators could also fit a study focused on the EMU. Indeed a study based on

emerging markets spreads could be structurally different from a study on the EMU, because

of the uniqueness of the situation of countries under the single currency area, but also because

of the valorization divergence between emerging economies and developed ones.

The studies mentioned in this section allowed us to assess the explanatory power of

credit quality and broader of fundamental economic variables on bond spreads, however this

is not the only fundamental variable influencing bond spreads valuation. Indeed price

elasticity to demand is vital for investors, liquidity is our second fundamental variable.

1.3 Fundamental liquidity and sovereign bond spreads

Analysts and researchers analyzing the European market or the European Monetary

Union would usually use Germany as a benchmark. Germany is the biggest economy in the

EU and is as well considered to be the standard in macroeconomic and the risk-free

benchmark on fixed income matters.

Gomez-Puig (2006) set up a hypothesis which assumed that two major variables were

influencing the credit spreads valuation over Germany inside the EMU, which were liquidity

and credit risk.

9

Her study highlights the lack of linkage between some fundamental economic

indicators such as relative debt levels to GDP and adjusted spreads, this was one of the

hypotheses of the author which assumed that since EMU the standardization of economic

fundamentals would lead to a liquidity valuation of credit spreads in the Euro-area. Indeed

rigorous criterion in economic standards, such as external balance requirement, debt and

deficit management standards were set up in order to join the Eurozone, and avoid the ECB

and other member states to bail out governments in debt financing distress.

The EMU countries’ interest rate spreads over German increased on average of 12

basis points in 5 years since the creation of EMU. The result of this increase is a lesser than

expected reduction of borrowing costs for the other EMU state members.

Liquidity can be measured using several indicators, however Gomez-Puig (2006)

highlights that liquidity is not only market size but also the price elasticity of an asset. Indeed

an investor will be able to step away from a position easier if he is able to sell his assets

without making the prices fall. Hence measuring liquidity in bonds using only market depth:

for example the total amount of debt outstanding, would not take into account the full

dynamic of liquidity, study of bid-ask spreads is fundamental in this case. A high bid-ask

spread would mean a high cost of stepping away from an asset, and therefore highlight a lack

of liquidity.

The change in liquidity assessment and valuation over the price of bonds could be one

of the reasons of the yield spread increase. As analyzed in the model, the most liquid

countries which represented the biggest debt markets in the Euro-area: Spain, Italy and

France, experienced the lowest increase in their adjusted yield spreads. Hence it seems that an

important reason of that result is the fact that liquidity as a variable of valuation of bonds was

more significant after the EMU creation. According to the result of the study liquidity

mattered, more than credit quality in the 1996-2001 period.

10

The European integration was a one-time event that may have biased the results of the

analysis, investors reacted to this new event by focusing on liquidity and value, thinking that

every country in the EMU was economically nearly equal, however recent changes and debt

crisis would have modified their investment style back to quality. Indeed liquidity mattered,

however investment styles are changing along the cycle and the shape of global economy, it is

important to analyze how those investors react to changes in the economic environment in

order to assess accurately the role of liquidity in bunds valuation. In order to analyze those

shifts we will require a longer range of periods and to add other explanatory variables.

While our results may change with a change in the investor behavior, liquidity is

known to be major in order to assess the price of any security. Moreover the study made by

Gomez-Puig (2006) highlighted the choices of variables in order to proxy liquidity and credit

quality.

If liquidity matters more than quality in the Eurozone, is it the case all along the

economic cycles, or does it have some periods of higher significance?

1.4 Trade-off liquidity – Quality

Shifts from liquidity to quality components of bond valuation are tough to observe,

indeed Ericsson and Renault (2006) highlighted the positive correlation of the liquidity and

credit quality features of corporate obligations in the US market, making it more difficult to

observe each feature separately.

The EMU however, is a unique area to observe the trade-off between liquidity and

credit quality as opposing to the classical link between those two factors, some countries like

Italy and Spain are among the most liquid fixed income assets in the region while being of

poor relative credit quality. This negative link between the fundamental quality and liquidity

makes the Euro-area ideal to study the variation of significance between quality and liquidity.

11

Beber, Brandt and Kavajecz (2008) analyzed the excess components of spreads of 10

EMU sovereign fixed-income markets using this negative correlation between liquidity and

quality, in order to show the different fluctuations of most significant parameters over time,

using country specific CDSs to assess credit quality and an average of 4 liquidity variables:

“bid-ask-spread”, “average quoted depth”, “cumulative limit-order book depth” and “average

quoted depth divided by the percentage bid-ask-spread”.

According to the result of the study on the EMU government bonds on the April 2003-

December 2004 period, three main assessments were made.

Firstly liquidity and credit quality are both significant EMU yield changes

components, as the combination of both factors explains from 22% of the volatility of the

short-maturity obligations to 57% for the long maturity assets. This result also highlights the

duration issue of such studies, indeed short-maturity bonds are sources of close-to-maturity

speculations, while long maturity securities are long-horizon expectancies and could biased to

political and reforms speculations, in order to solve such issues, we would need to define

maturity length boundaries.

Secondly credit quality explains the biggest part of fluctuations of yields, as it makes

for 89% of the explanatory power while liquidity constitute the rest. However the authors

emphasize the fact that positive linkage between liquidity and credit quality may create bias,

as the proportions greatly varies between high credit quality countries and low credit quality

counties. In order to analyze deeper the liquidity-quality puzzle we will require to observe a

full cycle, the 8 month period of the study will not be sufficient to do so.

The last conclusion of the analysis is focused on “flights”, while credit quality seems

to always be significant over the period of study, liquidity’s significance is more volatile and

dependable of the movements of the market. In times of local market insecurity, “flights-to-

liquidity” happen, as investors are seeking the most liquid assets as a safe harbor. The same

type of behavior can occur with quality, “flights-to-quality” happen in times of global markets

distress, quality as opposed to value for yield will become the main investment style.

12

The idea of “flights” is crucial in order to understand bond yield volatility. Indeed

investors change their investment behavior according to the economic environment and tend

to “overreact” to some indirect events. When investors behavior changes, their price and risk

assessments change as well. Investors are making the price, understanding those shifts in

behavior and explaining why they occur will be the central theory of our study. Our study will

focus on highlighting the linear and quantifiable part of those behaviors while the chronic and

shorter term speculation part would not be quantifiable and difficult to isolate and analyze.

Moreover the Eurozone market is an ideal region to observe such changes, investors

all across the world try to hedge themselves on currency risk on the biggest currencies in

order to protect their assets’ value. An Asian or American investor will have to invest in euro

denominated securities in order to do so. While in any other region he would have to make his

position on bond duration accordingly to his assessment of the economy of the country, in the

Eurozone he will have the ability to shift from a liquid country to a credit quality one for

example, or to seek both for the same maturity. This investment structure is unique, as well as

the valuation metrics of fixed-income securities exchanged within the area.

The fluctuations of yields are linked to fundamental economic and liquidity variables,

however some extra-fundamental factors are also influencing those yield levels and are not

only country specific. Those variables cannot be categorized as fundamental as they do not

impact directly the residual value of an asset. They can be either the reflect of the risk feeling

of global investors or linked to spoil over of financial distress.

13

1.5 Extra fundamental variables

While standard analysis of bond spreads would assume that the main component of

valorization are fundamental criterion specific to each country, given the fact that financial

markets are internationalized and auto-correlated some common factors could influence the

volatility of fixed-income markets. Moreover the standardization of some trade regulations

and tax policies made investors think of EMU as a single financial market in the early 2000’s.

Caceres, Guzzo and Segovianno (2010) showed that even if a significant part of bond

volatility could be explained by liquidity and quality fluctuations, a part of the yield spreads

was explained by non-country specific factors. Two systemic factors were quoted to illustrate

those factors: “Global Risk Aversion” and “Contagion”.

While global risk aversion constitutes the pricing of risk premium according to the

feeling of investors on the economic environment. Contagion resides in the fact that a

worsening in the economic fundamentals of a single country in the EMU could spread over

the other countries, increasing interest rates for all the bonds in the Eurozone. The contagion

variable illustrates the behavior of investors which assume EMU is a common market and a

worsening of the economy of a single country in the area would endanger each member

state’s economic situation but also the indirect effect of worsening economies or financial

situations of trading partners. Usually sources of contagion would imply low credit quality

countries such as Greece, Portugal or Spain. Negative information concerning the economy or

debt servicing of such country would lead to an increase in spreads over the whole region.

14

While the research highlights effects of extra-fundamental variables, their effect and

patterns are not linear. Caceres, Guzzo and Segovianno (2010) link periods of study to

economic events to explain that while global risk aversion is rising, its effects are multiple.

The spreads inside the EMU will widen between high quality countries and low quality ones.

Indeed this seems to highlight a shift in investment behavior. Investors will overreact to risk

aversion and change their investment style in order to match a safer positioning. Indeed global

risk aversion could either be positive or negative according to the credit quality of the

country; this leads us to assume that risk aversion could be linked indirectly to credit quality

and liquidity.

Moreover contagion is showed to be only negative for every country in the EMU. Any

sign of distress inside the EMU, or indirectly in the world economy, would lead to an increase

in yields. Contagion, while difficult to assess and quantify would be a major parameter of the

extra-fundamental part of our analysis.

The authors emphasize the fact that liquidity and economic variables are still

explanatory variables, the extra-fundamental proportion of volatility is less quantifiable, and

however the use of dummy variables as well as some indexes and proxy ratios would allow us

to measure some part of those effects of contagion and risk aversion.

While risk aversion is chronic and its duration seems to change according to the

dominating environment in the world economy, contagion is a consistent variable which is a

persistent factor. In order to assess the impact of those parameters we would need to proxy an

index for the risk aversion in order to reflect its volatile and uncertain pattern on the one hand,

and a dummy variable to assess contamination in order to trace its consistent impact.

15

A study led by Attinasi, Checherita and Nickel (2009) showed that in the July 2007 –

March 2009 period, when some local banks were under the threat of bankruptcy due to the US

sub primes, generated the spreads widening inside the Euro area.

The objective of the study was to analyze the change in patterns in the behavior of

investors, which were on the first place considering Eurozone as a single market. The idea of

widening spreads does not necessarily mean a sell-off on the European securities, but more a

shift in bond valuation techniques.

The range of study is crucial on the understanding of the European spreads history,

indeed the first half of the range from 2007 to mid-2008 investors were focused on liquid

markets distress as most of the governments had to step-in to bail out their local banks

infested by sup-prime “trash” holdings in order to safeguard the savings of their countrymen.

As explained by the authors a shift in the risk aversion and its price assessment occurred, risk

was transferred from corporate financials to governments, increasing the global risk aversion

towards sovereign bonds. This shift is assumed by the authors to have occurred when country

publicly announced that they would bail out for their local national banks.

On the second part of the range of dates (mid-2008 to 2009) studies the investment

patterns changes again, credit quality became a prominent factor of sovereign bonds

valuation, creating a widening in spreads. This could be explained by the concern of investors

on some European peripheral countries concerning their debt reimbursement abilities.

The breakdown of explanatory variables of bond spreads was the following: 56%

international risk aversion, 21% expected economic fundamentals, 14% liquidity

fundamentals, 9% political announcement. The main component of spreads was systemic and

international, while 44% of the spreads were explained by country specific factors.

16

The predominance of international risk aversion as an explanatory variable showed us

the difficulty in the valuation of sovereign bonds, indeed common supranational variables

seem to explain most of the variations of the spreads. However the blur assessment of global

risk aversion make us questioning ourselves about two assumptions.

Firstly, because of similarities between the markets in the Eurozone and the common

currency effect, the global risk aversion may be broken down to two sub parts, first one the

Eurozone risk aversion, which could be linked to contagion factors as well as direct market

effects of distress inside the euro-area and specific to itself. The second one would be the

extra-Eurozone risk aversion, and would regroup the effect of indirect global markets

riskiness on EMU sovereign bonds.

Secondly, global risk aversion as a single common factor would not explain spreads

according to us as it is common to every yield, even German one. Then what effects on our

analysis on spread would a removal of the analysis of global risk aversion have?

While liquidity was dominating sovereign bond valuation in the early days of the

Euro-area according to the analysis made by Gomez-Puig (2006), the economic environment

changes produced a shift in investing patterns, according to Attinasi, Checherita and Nickel

(2009), the 2007-2009 period was led by fundamental economic indicators such as its main

one, credit quality.

The joint results of those two studies highlights the significance of the assumptions of

Beber, Brandt and Kavajecz (2008), “flights to liquidity” and “flight to quality” occur, for

various reasons, lead on by shifts in investors feelings on the economy, globally but also

inside the Euro-area. Those changes may have some various explanation and need to be

analyzed deeper in order to get a better understanding of the cyclical effect of world’s

economy on EMU sovereign bond valuation.

17

The analysis of previous empirical studies allowed us to formulate our own

theory in order to build up our testing methodology.

2. METHODOLOGY

2.1 Choice of Variables

2.1.1 Sovereign Yield spreads

In order to set up our model and assumptions we need to firstly define the dependent

variable we are going to study. Indeed the definition of this variable will be the foundation of

our model.

We are studying the dynamics of valuation of sovereign bonds inside the EMU,

therefore studying bond value or interest rates value of those underlying bonds. A rise in bond

yield will have a direct effect, the decrease of its price.

Indeed interest rates are the main components of bond prices and very accurate tools

of bond valuation.

Our study will focus on yields of EMU sovereign bonds relative to Germany. The

difference between a country’s yield and German one will give us the relative spread of that

country versus Germany. The use of yield spreads versus a risk-free asset is widely used as a

valuation technique, it allows the analysis to focus on asset specific features, by removing the

common valuation factor.

18

Germany is the most widely used benchmark on the sovereign markets, because of its

economic structure, its financial stability and its predominance in the intra-European market

as well as international market. Italy is the biggest bond market inside the EMU however it is

not the biggest economy in the EU nor is it a risk-neutral market.

In terms of duration we will focus on the 10 year market because it is the most liquid

duration and is subject neither to short-term speculations nor long-term bets. However we will

bear in mind that maturity is another part of bond valuation.

A generic 10 year yield is provided by Bloomberg DataStream with a 30 years history

for each EMU country. Alternative yield spreads valuation metrics are 5 year credit default

swaps, because of their liquidity and fair valuation, as well as 10 years Asset Swap Spreads

because of their fundamental valuation focus. However those alternative valuation metrics

and other maturity generic yields does not provide us with enough history to be included in

our study.

We will regress numerous variables in order to empirically test their explanatory

power on spreads. However a basic assumption is required: an increase in spreads is the

consequence of a deterioration in either economic, liquidity or extra-fundamental metrics.

Furthermore we will assume that a monthly dataset in the 2001-2015 time range will

give us enough observations in order to build a consistent model.

2.1.2 Fundamental economic data

Since the introduction of our thesis we are mentioning fundamental economic data as

the predominant factor of bond valuations, but what does “fundamental economic” refer to?

19

In order to highlight the fundamental dimension in the valuation of sovereign spreads

of economic data, we have to consider a state as a company. A company would be able to

lower the interest rate at which it borrows by increasing the quality of its financial results,

turnover, leverage, margin, net profit, cash generation etc. A state like any other economic

entity requires financing. Fundamental economic factors are a state’s financial results and

features and therefore the best indicators of the credit quality of a country.

Indeed credit quality is the main valuation factor of an entity’s debt valuation as it

emphasizes its ability to refund its debt and service it.

Hence in order to valuate sovereign credit yields we will need to be able to assess the

credit quality of each country we are observing, and this will be done using published and

reliable economic metrics.

Credit quality of a country is assessed and quantified by observing its economic and

financial health, indeed a state generates revenue like a company and needs cash to

continuously fund its operations and investments through issuance of publicly tradable debt.

However the interest rate at which the state will be able to borrow those funds are highly

dependable on its capacity to generate enough income to repay the debt. Indeed the risk that

the borrower could not pay back its lender is the main credit risk. It is also called default risk.

A good ability to service and pay back its debt will lower the premium expected by the

market in order to lend funds to the borrowing entity.

Therefore in order to assess the credit quality of a specific country we will need to find

a way to proxy its financial shape and compare it to peers in the same way analysts would

compare financial statements of companies in the same industry in order to define its credit

risk. Hence we state the economic analysis of a country as fundamental because it is linked to

the first dynamic of credit valuation: default risk.

20

The assessment of a country’s ability to finance its operations and service its debt

resides on four primary economic metrics:

2.1.2.1 Nominal GDP growth

Nominal GDP is a country’s main economic indicator, it reflects the economic activity

of a country and the amount of wealth created by it. As nominal GDP levels are assumed to

be already valued by the market we will consider studying its percentage monthly growth

variations as a metric of a country’s ability for wealth generation.

Indeed nominal GDP growth is the primary assessment of revenue creation made by a

state, as the main source of income of a government is taxation, citizens and local

corporations need to grow the amount of wealth they are producing in order for state to

increase its income.

Nominal GDP growth would be comparable to sales revenue growth of a company, or

its turnover growth as a first indicator of a top down financial analysis.

We will then assume that positive relative GDP growth would have positive effects on

a country’s financial situation and therefore negative effects on spreads, indeed a county

which is increasing the amount of wealth inside its borders will be able to generate more

revenue in order to finance its operations and repay the debt it is accountable for.

21

Chart 2.1: France GDP Growth spread vs. Yield Spread1

Chart 2.2: Italy GDP Growth spread vs. Yield Spread

Chart 2.1 and 2.2 highlight the negative correlation between spreads and nominal GDP

growth for the two biggest debt markets in the EMU versus Germany.

Nominal GDP growth is preferred to real GDP growth, indeed real GDP implies

currency expectations and inflation rate forecast which are less effective knowing that we

study only countries in the single currency area. Moreover nominal GDP is a primary data and

is not subject to any computations, it is provided by Bloomberg as published by central banks

and local governments.

2.1.2.2 Debt to GDP ratio

A highly indebted country will use more of its wealth creation in order to refund its

current level of debt. It will be harder for it to prioritize its funds towards creation of growth

and therefore revenue increase on the short term; therefore interest rate will tend to increase

because of default risk increase. Moreover this vicious circle tend to increase exponentially

over time as interest rate expenses increase as well and is often a hint of recession, currency

risk and austerity reform perspectives.

1 Source: OECD

Negative Yearly average Long-Term* interest rates

spreads v. Germany

*Long term is defined by OECD as 10 year maturity

interest rates

Correlation factors: France (27%), Italy (47%)

22

Furthermore the EMU is a single currency area where state members cannot lead a

currency devaluation process independently without the support of the European Central

Bank and therefore make debt management even tougher to control.

Debt to GDP ratios are published less frequently than other economic metrics,

Bloomberg provides yearly data emerging from IMF publications, we would however be able

to assume a constant monthly rate of growth which would furthermore be reinforced by the

above mentioned increasing nature of debt-to-GDP without proper debt management reforms.

This indicator could be linked to a company’s leverage ratio or gearing, however debt

level increases are not subject to the same analysis from investor according to whether it is a

company or a sovereign as structures of those entities differ largely. Indeed debt issuance

from a company could be analyzed as investment in future growth operations, the structure of

debt management cannot be compared between the two types of entities and is drastically less

flexible on the government side because of the nature of government public expenses.

An increase in debt to GDP ratio is assumed to have a negative effect on the financial

shape of a country and therefore a positive effect on spreads.

Debt to GDP ratio is not directly linked to liquidity as it is a comparison of debt levels

to wealth creation, it highlights the relative levels of indebtedness according to income

generation abilities despite liquidity features.

2.1.2.3 Budget balance to GDP ratio

Budget balance consists in the difference between sovereign revenue and government

expenses, it is then crucial to analyze the deficit or surplus generated by a country to

understand its future financial needs and abilities.

23

Indeed a country in deficit does not have the found required to afford its expenses and

will therefore require to issue an additional amount of debt in order to finance its operations.

Deficit management was one of the “gold” criterion set up by the EU members in

order to join the European Monetary Union because of the impossibility to lead devaluation

processes as mentioned above in the Debt to GDP section. Hence relative levels of budget

balances are crucial to understand the economic distress of certain countries compared to the

others.

Budget balance data are published quarterly by Bloomberg and assumed to have a

continuous monthly rate of growth.

We could compare country’s surplus (deficit) to a company’s net profit (loss) from

operations, and is therefore positively impacting its financial state, hence negatively

correlated to spreads.

2.1.2.4 Current account to GDP ratio

Current account to GDP ratio highlights the difference between money inflows and

money outflows of a local country including trade balance, salary expenses balance and

money flows; it is not specific to a government but to residents of a state as a whole as

opposed to the budget level mentioned above. Hence it is a complementary indicator of

budget balance.

Indeed the structure of current account levels cannot be considered the same way as

the other indicators because of the complexity of the flows it includes as well as the

combination of direct and indirect effect toward the economy.

Current account levels allow us to draw an evaluation of the relative foreign

attractiveness of each EMU market which is more accurate than the trade balance as it is a

combination of trade balance and other money flows.

24

Moreover an unweighted combination of budget balance and current account levels

would give us the relative “twin-deficit” of each country, widely used by investors in order to

assess the financial state of a country, which is an assessment of the fiscal shape of a country

and directly linked to debt-emission.

Negative current account levels would assume that the country requires to borrow

from foreign entities in order to finance the gap, and would therefore have negative effects on

its financial state and conversely. Indeed we assume that positive current account levels have

negative effects on spreads.

Current account levels information of each EMU member states is provided quarterly

by Bloomberg based on ECB reporting and assumed to be growing at a constant monthly rate.

2.1.2.5 Secondary variables

There are also some secondary variables that we do not consider as fundamental,

however their assumed indirect explanatory power requires us to test them as well. We will

include those factors in our regression in order to have a stronger idea of all the features,

direct and indirect, of economic fundamental valuation dynamics of sovereign bonds.

Trade balance and financial accounts are complementary with the “twin-deficit”

variables mentioned above and could give us some interesting insights on bonds valuations;

however trade balance is already included in the balance of payment and assumed to be

strongly correlated with it.

The unemployment growth is generally closely links to the nominal GDP of a country

and should give us another angle of observation of a country’s economic shape outlook.

Local stock markets are usually negatively related to spreads and could give us an

additional explanation component.

25

Industrial and service production levels are other indicators of the potential of a

country to generate income and should be tested in our empirical analysis.

Consumer confidence and economic sentiment are indirect indicators of the

expectancies of economic growth of a country, they are assumed to be linked to the credit

quality of a country.

The exchange rate versus the dollar could impact sovereign valuation, however it is

not a country specific feature, which would lead us to remove the variable from the study.

GDP per capita, wages level and real GDP are too correlated to nominal GDP to be consistent

in the explanatory power in order to optimize our model, we will however add them to the

analysis in order to check if they bring additional explanatory power to the model.

2.1.3 Fundamental Liquidity data

Liquidity is the second fundamental valuation metrics of sovereign bond spreads as it

is directly linked to transaction cost and distress risk. Indeed more liquid markets tend to have

lower yield for the same credit quality for two main reasons.

First of all the “too-big-to-fail” theory is usually assumed by sovereign bond investors,

biggest markets are essential for too many investors and the global economy to be allowed to

default without negotiation of debt, indeed in times of increase in risk aversion, liquidity

would tend to be considered as a safe harbor for investors and therefore yield of liquid market

will tend to increase less than peers in times of financial distress and high risk aversion.

Secondly because of the elasticity to price, for investors it implies less transaction cost

to leave a position on a liquid asset as the price will decrease less in case of sell-off. Hence

positioning on liquid assets would be considered as a less risky investment.

Therefore because of its predominating position in the sovereign and corporate

valuation, liquidity features can be considered as a fundamental valuation mechanism.

26

Liquidity is more difficult to quantify than economic fundamentals because of varying

definitions according to literature and studies. We will assume that liquidity assessment is the

most accurate according to Gomez-Puig (2006) definition, which is a combination of market

depth and price elasticity.

Therefore our fundamental liquidity assessment will be built on 2 main metrics:

2.1.3.1 Market depth

Primary an asset liquidity is considered by the number of assets traded compared to its

peers. Therefore in order to compare sovereign liquidity, we should compare relative amounts

of public debt outstanding.

Indeed private debt is not issued by government and should be removed from the

analysis, the amount of public securities outstanding will give us a first indicator of liquidity.

Moreover increased relative market depths would mean increased relative liquidity

and therefore lower spreads.

Chart 2.3: France Debt Level vs. Spreads2

2 Sources: OECD, Bank for International Settlements (BIS)

Yearly Average amount of Public Debt Outstanding v. Germany

(million EUR)

Chart 2.4: Italy Debt Level vs. Spread

Negative Yearly average Long-Term* (LT) interest rates spreads

v. Germany

*Long term is defined by OECD as 10 year maturity interest

rates

Correlation factors: France (3%), Italy (93%)

27

While the link between market depth and spreads has yet to be proved for France, Italy

shows a very strong correlation between its relative liquidity to Germany and its long-term

spread.

Total amount of public debt outstanding is published quarterly by the Bank for

International Settlements and assumed to grow monthly at a continuous rate. We will analyze

its growth rate as well as the proportion changes of each country’s total amount outstanding

compared to the total of public debt outstanding in the EMU.

2.1.3.2 Price elasticity

Secondary liquidity is defined by the price elasticity features of such asset. Indeed

market depth is an indicator of liquidity, what investors are looking for through liquidity is to

lesser the transaction cost they are subject to when getting off a position.

The bid-ask-spread is the standard unit of reference for measuring price elasticity of

bond securities, it represents the average daily gap between the highest bid and the lowest ask.

This gap measures the actual transaction cost an investor would endure while selling

off his position.

Bloomberg Datastream provides average daily data considering bid-ask spread since

2006, this metric is directly computed by Bloomberg and would need to be broken down and

detailed. We are going to use a proxy metric for the years past to 2006 in order to complete

down the liquidity from 2001 to 2006. This proxy will consist of shorter term securities’ bid

ask spread for each country.

28

An increase in spread would mean more transaction cost and less liquidity, therefore

an increase bid-ask-spread relative to peers would mean an increase in yield spreads.

2.1.4 Quantifiable Extra-Fundamental variables

Spreads are not only valued from the residual value of Liquidity and Economy, indeed

some less obvious variables are also impacting bond valuation because of indirect links,

expectancies or contagion. Most of that Extra-Fundamental part of bond valuation cannot be

quantified because of the complexity of the factors involved.

Although fundamental variables have strong assumed explanatory power, some extra-

fundamental variables are indirectly linked to the valuation of sovereign spreads and need to

be tested. Most of those factors are unfortunately not consistent with the whole region but

mostly country-specific, or linked to chronic short term events. However some factors could

be analyzed in order to see if we can break down the unexplained part of the model using

some variables with assumed explanatory power.

While some metrics are systematic, such as risk aversion and world economy do not

require a closer analysis as we are focusing our study on spreads. Some other variables have

country specific impacts.

Political risk can be a factor of investors’ decision and therefore is required to be

tested. Bloomberg keeps an annual record of Political risk grades which are computed as an

aggregate of different political, legal and human development metrics.

Another path that needs a closer study on is volatility or country specific contagion,

indeed some sovereign “overreact” to news concerning contagion or risk aversion and are

therefore targets of contagion because of their economic structure or financial situation.

29

Some investors are looking for yield without valuating the risk they are bearing to

take, and the proportion of such investors varies along the economic cycle. Hence it would be

interesting to include the nominal value of yield into the extra-fundamental factors.

The last extra-fundamental that we wish to observe is the Upgrade/Downgrade metric.

An upgrade by one of the leading credit agencies is as much a political decision as an

economic one. Indeed investors tend to overreact to this kind of announcement and it will

certainly impact spreads more than the real deterioration of fundamentals.

2.1.5 Where are we in the cycle?

In order to have a better understanding of valuation dynamics of EMU spreads we will

need to assess the economic cycle moves. The world economy as well as EMU economy

shifts from recessions to recoveries, the shifts in the economic cycle define the behavior of

investors.

Multiple indicators are assumed to have explanatory power of cycle variations. In order to

process to an accurate proxy of the EMU economic cycle we will proceed to the creation of a

cycle index. In order to create this index we will aggregate 4 different indicators from both the

EMU and the US to capture a EMU specific cycle valuation as well as a global outlook:

MSCI monthly returns: Stock markets monthly returns are strong indicators of

economic outlook and investors expectancies on a specific market financial

distress.

M2 –Money supply: Money supply is a strong hint of governments’ economic

shape and quantitative easing strategies, indeed in times of financial distress,

central banks will intervene to sustain its local economy.

Nominal GDP growth rate: Fluctuations of the wealth creation of the whole

EMU area will give us a strong indication of the financial state of the region.

30

OECD LEI (Leading Economic Indicator): The aggregate index create by the

OECD is the main leading economic indicator of economic cycles, it is an

index based on various indicators such as wages, companies new book orders,

building permits, credit indexes and consumer expectations.

Each indicator defines a different feature of the economic cycle, by aggregating those

4 indicators we should be able to provide a general outlook on the different periods of the

cycle. Comparison between the cycle movements and our models will provide us with

understanding of cycle shifts influence on bond valuation dynamics.

Chart 2.5: Cycle Estimates Timeframe3

Chart 2.5 suggests a high correlation between each indicator we chose, we can notice

however a time lag between indicators, for instance the M2 – Money supply indicators carries

a 1 year lag, suggesting that central banks reaction comes after crises, we will need to adjust

this lag in order to get a more accurate cycle proxy index.

3 Source : Bloomberg Datastream

-0,12

-0,10

-0,08

-0,06

-0,04

-0,02

0,00

0,02

0,04

0,06

0,08

0,10

-1,20

-1,00

-0,80

-0,60

-0,40

-0,20

0,00

0,20

0,40

0,60

2001 2003 2005 2007 2009 2011 2013

M2 - Money Supply EMU Nominal GDP Growth Rate

Leading Economic Indicator MSCI Monthly Returns

31

2.2 Definition of the model

2.2.1 The influence of credit ratings

We arbitrary chose not to use the credit ratings as explanatory variables in order to

develop our theory on credit quality and economic fundamentals. However we consider the

three leading rating agencies as a consistent dataset which regroups an analysis on the three

pillars we mentioned with a focus on economic fundamentals.

Therefore the use of credit ratings would be necessary in order to check the

consistency of our different variables. Indeed a strong correlation between our models and an

aggregate of the credit ratings will highlight the strength of our choice of variables. Hence we

will compare our results with credit ratings and test its correlation throughout our analysis and

use an unweighted aggregate of the ratings of the three biggest agencies as a benchmark, and

set up correlation targets in order to check the robustness of our model.

A table of correlation of our ratings compared to the credit ratings will be provided in

the analysis.

This decision not to use credit ratings is motivated by the aim we have, which is to

break down the yield spreads into different dominant variables and therefore optimize our

model in order to identify each shift of predominant factors.

2.2.2 Fundamental Economic Grade

In order to test our fundamental economic variables we will regress each of them using

Ordinary Least Square (OLS) regression using yield spreads as a dependent variable. Using

the following formula:

[Yit – YDEt]= α [TVarit – TvarDEt] + Intercept + ὲ

32

Where Yit is the yield of the country I at time t, YDEt is the yield of Germany at time t,

TVarit is the tested variable of the country I at time t and TvarDEt is the tested variable of

Germany at time t.

Once our variables tested we will be able to assess a first filter and remove variables

with weak explanatory power.

Moreover we have to bear in mind that this study is a relative study and therefore does

not focus on generalized valuation but to identify the valuation dynamics in the EMU using

yield spreads.

The next step of our fundamental economic study will consist in the establishment of a

fundamental economic relative rating which will be included in our final model.

In order to set up this fundamental economic grade we will proceed to the z-scoring of

each variable in order to remove the aggregating issue we could have had.

The next task required is the analysis of the correlation between each variable using a

correlation table. This table will allow us to remove some variables or reduce their coefficient

according to their correlation grade.

We will then build a new model with the weighted average each metric z-score as a

variable and test it again using the same formula mentioned above. As a final step we will use

excel solver in order to optimize the R² of the final coefficient using each weight as a solver

variable. However in order to get a consistent result we will use the correlation matrix we

previously built to set up acceptable conditions such that we get consistent result.

33

We will then analyze the residuals of the last regression in order to assess the most

suitable Fundamental Relative Economic Grade (FREG), to avoid heteroscedasticity or

autocorrelation of the residuals.

The FREG will be a weighted average of each explanatory variable according to the

weight which was assigned by the solver.

SPREADit = α FREGit + Intercept + ὲ

2.2.3 Fundamental Liquidity Grade

In order to assess our Fundamental Liquidity Grade we will follow up the same

process, however the fact that we are testing only two variables will simplify the testing.

We will use another Ordinary Least Square (OLS) regression for both variables and

confirm their explanatory powers on yield spreads using the same formula:

[Yit – YDEt]= α [TVarit – TvarDEt] + Intercept + ὲ

Where Yi is the yield of the country I at time t, YDEt is the yield of Germany at time t,

TVarit is the tested variable of the country I at time t and TvarDEt is the tested variable of

Germany at time t.

We then require to be able to aggregate those two indicators using z-scoring

techniques to avoid any unit issue.

Once more we will create a weighted average, using the solver in order to optimize the

R² of the model using the weights as variables; then proceed to a residual analysis and

correction if necessary.

34

The weighted average of each variable’s z-score will provide us a Fundamental

Relative Liquidity Grade (FRLG).

SPREADit = α FRLGit + Intercept + ὲ

2.2.4 Quantifiable Extra-Fundamental Grade

The last part of the grades assessment will be the valuation of the Extra-Fundamental

Relative Grade (XFRG), which will be done using the same process as the previous grading.

Moreover we will proceed to the use of dummy variables in order to assess

downgrades and upgrades “overreaction”, value for spot rate and volatility.

However we have to bear in mind that the OLS could be biased by the noise of the

combined variables and this is the reason why it is crucial to test the explanatory power of

each variable alone and study correlation of different variables before assessing optimal eights

through the solver.

SPREADit = α XFRGit + Intercept + ὲ

2.3 From pillar specific to combined

2.3.1 Pillar specific model analysis

The primary objective of our model is to specify a valuation model for each of the

previously mentioned pillars: Fundamental Economic data, Fundamental Liquidity data and

Extra-fundamental data.

35

In order to do so we built up Grades for each pillar in order to optimize our

explanatory power for each of the variables.

To sum up we came up with three different explanatory models with pillar specific

variables:

Fundamental Relative Economic Grade (FREG):

SPREADit = α FREGit + Intercept + ὲ

Fundamental Relative Liquidity Grade (FRLG):

SPREADit = α FRLGit + Intercept + ὲ

Extra-Fundamental Relative Grade (XFRG):

SPREADit = α XFRGit + Intercept + ὲ

We will then define a polled regression cross-results table depending on clustered

countries and crisis periods.

The result table will allow us to analyze the efficiency of our analysis as well as

predominant models using the R² over time. It will also allow us to emphasize the country-

specific valuation dynamics. For example we expect peripherals to be valued primary on

extra-fundamental more that economic data and liquidity.

Moreover we will be able to test the various models against credit ratings in order to

back test it and use it as predictive model using forecasts and analysts recommendations.

36

2.3.2 Combined model

Once each pillar specific model created, we will be able to build up a combined model

using a regression on the 3 grades which is aimed on predictive and longer term residual

valuation of spreads.

In order to do so we will build a Multiple-Components regression on the 3 variables

using the following formula:

SPREADit = α1 FREGit + α2 FRLGit + α3 XFRGit + Intercept + ὲ

A pooled regression (using clusters and crisis periods) on the combined model would

give us insights about the shifts in factor dominance, the residual value of spreads and the

breakdown of valuation dynamics over time.

Moreover if the R² is satisfactory with our explanatory power objective, it would be a

useful tool in order to determinate sovereign EMU yield residual value and upside/downside

in order to build up investment cases.

2.3.3 Unquantifiable Extra-Fundamental dynamics

As mentioned on the variable choice part of the thesis, the biggest part of the Extra-

fundamental factors cannot be quantified and therefore observed. However as Westphalen

(2001) suggested, a deep analysis of residuals may highlight a common factor explaining the

gap of R² of an empirical study.

While we could assume that the unexplained part of the combined model (1-Adj R²)

consist in the unquantifiable part of extra-fundamental variables, it would be more accurate to

specify that a part of the unexplained volatility of spreads consists in extra-fundamentals

which we failed to quantify.

37

2.4 Model analysis

2.4.1 Linking clusters

The model we will establish is computed using cross-sectional spreads, however there

may be a disparity of significance and usefulness according to each country. We will break

down the ten countries we observed into three clusters and observe the differences between

each cluster. An analysis of those disparities would allow us to highlight the difference in

valuation dynamics in order to the country. Each cluster would identify countries with similar

features relative to the other EMU country.

2.4.1 Linking crises

Following on the above cluster section, our model is cross-sectional, it includes values

from different periods with different valuation dynamics. Therefore we will proceed to an

analysis of the differences in our model results according to time periods.

We set up two main assumptions, valuation dynamics are different according to

whether the economy is going through an expansionary era or a recessionary one. We will

compare our model results according to crises and recoveries we identified using our cycle

proxy.

Furthermore, following on the literature and specifically according to assumptions

made by Attinasi, Checherita and Nickel (2009), the subprime crisis of 2007 and the event of

governments refunding their national banks changed strongly sovereign bond valuation

dynamics, we will compare our pre-subprime crisis results and post-subprime ones in order to

identify that change in valuation dynamics.

2.4.3 Investment insights

The ultimate end of our combined model is to define residual values of spreads. In

order to test if it can be a useful valuation tool, we will proceed to a back testing of sovereign

bond returns and implement our model in order to create an investment case.

38

In order to test our model, we will use the EFFA / Total Return Indexes provided by

Bloomberg Datastream in order to provide the best estimates of historic returns of each

sovereign bond markets.

We will then use the Euro Government Index provided by Bank of America Merill

Lynch in order to set up portfolio weights and test our valuation model setting up “bets”

according to the residual value estimates we computed. The comparative results will provide

us with the excess profit (loss) return created by our model.

2.4.4 Feedback and analysis on the methodology

Both pillar specific and combined models are useful tools in order to analyze the EMU

sovereign spreads.

Indeed the pillar specific models will give us specific analysis of the dynamics of bond

valuations, moreover a plotting of the different models R²s will give us a timeframe of the

different dominating trends in bond valuations. This analysis of shifts in dominance could

then be linked to our cycle analysis in order to find some patterns and trends in the valuation

of spreads according to the cycle.

The combined model is a broader indicator of residual value which takes into account

the whole dynamics of bond valuation in order to get the most accurate residual value of

spreads and be able to predict corrections in the valuation of bonds.

However another feature to observe is the correlation between one pillar and the other,

we are assuming that each category is independent however, some indirect links may bias the

analysis and the combined model.

39

Furthermore the unexplained part of the model cannot be simply explained by the

unquantifiable extra-fundamental part of our modelling but rather a proportion of the

volatility we failed to capture.

Finally the links to the cycle fluctuations may be blurred by the recent volatility of

stock markets which tend to behave out of cyclical fluctuations, an aggregate index would be

more accurate using stock markets but also other economic metrics and leading indicators.

3. DATA ANALYSIS

3.1 Yield Spreads

We used monthly 10Y Generic Yield Spreads vs. Germany. We dropped the other

maturity ranges, asset swap spreads and CDS because of the length of history provided by

Bloomberg Datastream. Table 3.1 (Appendix) resumes every variable used in our analysis,

their sources and frequencies.

3.2 Fundamental Economic Data

Table 3.2: Analysis summary

We selected 4 main economic indicators and 10 additional ones in order to set up our

economic study of spreads.

Test start date 31/10/2001

Test end date 30/03/2015

Observed date periods 162

Observed countries: 10

Observed Independent Variables: 14

Observed Dependant Variables: 1

40

3.2.1 Correlation output:

The correlation table (Table 3.3, Appendix) gives us reliable insights on the

dependency of each indicator toward the others. It also enables us to highlight the different

independent variables from our 4 primary indicators which are: current account to GDP ratio,

budget balance to GDP, Nominal GDP growth rate and Debt to GDP ratio. This is valuable

information in order to increase the explanatory power of our model by adding secondary

variables.

The highest correlation coefficients found are between consumer confidence and

budget balance as well as unemployment rate and yield spreads

Current accounts are unsurprisingly highly correlated with trade balance and budget

balance, so are nominal GDP growth rate and real GDP growth rate.

Economic sentiment, unemployment level and wage level are highly correlated to each

primary indicator.

3.2.2 Pooled regression output:

Table 3.4: Single variable OLS R² output

R² Output Budget Balance Consumer Confidence Current Account Debt to GDP Economic SentimentIndustrial

Production

3M rolling return Local

MSCI

Austria 12,17% 18,97% 0,14% 20,49% 24,21% 13,56% 4,58%

Belgium 47,92% 7,36% 38,76% 3,93% 6,71% 31,69% 0,00%

Finland 38,77% 46,44% 20,48% 0,94% 35,65% 3,14% 6,09%

France 19,30% 11,06% 29,38% 52,62% 24,59% 47,23% 0,20%

Ireland 3,93% 14,70% 20,15% 41,18% 21,53% 0,03% 2,98%

Ita ly 1,71% 20,12% 3,92% 53,97% 39,28% 50,36% 0,42%

Netherlands 50,20% 33,71% 5,50% 28,86% 33,75% 10,09% 2,91%

Portugal 1,72% 18,99% 17,47% 49,08% 43,70% 47,12% 1,64%

Spain 44,07% 19,38% 32,64% 45,35% 29,66% 62,14% 1,16%

Total Ex. Germany 11,21% 8,51% 2,62% 20,73% 16,03% 3,00% 0,19%

Total Inc. Germany 11,85% 9,09% 4,12% 20,70% 13,67% 2,64% 0,20%

41

Table 3.5: Single variable OLS R² output (2)

We then ran a pooled regression depending on countries for each variable that we

assumed could have an explanatory power on spreads. The R² of each regression is displayed

on the tables 3.4 and 3.5, it is an indicator of the explanatory power of each variable per

country as well as per area.

The level of explanatory power fluctuates from countries and variables. Budget

balance, debt to GDP ratio, economic sentiment indicator and unemployment rate show strong

R² when regressing them against spreads.

In order to capture a cross-country regression we would need to combine strong

individually explanatory variables.

R² OutputNominal GDP 3M

rolling Value

Nominal GDP 3M rolling

growth rate

GDP PPP Per

Capita

Real GDP rolling

3M Growth rate

Nominal Trade

BalanceUnemployment Rate Wages Level

Austria 23,52% 22,29% 20,15% 25,93% 36,31% 0,00% 24,44%

Belgium 34,07% 3,99% 24,03% 16,93% 38,44% 13,34% 13,87%

Finland 18,30% 3,30% 22,54% 32,66% 23,50% 1,36% 22,75%

France 46,58% 23,08% 19,95% 13,96% 42,32% 25,62% 1,37%

Ireland 0,13% 1,19% 0,08% 3,12% 9,41% 44,15% 23,04%

Ita ly 28,51% 26,87% 22,73% 18,54% 5,77% 31,23% 16,88%

Netherlands 34,10% 52,24% 33,35% 45,22% 1,51% 3,69% 0,17%

Portugal 11,39% 47,91% 18,73% 26,18% 38,15% 54,59% 2,54%

Spain 20,08% 51,45% 18,05% 43,32% 52,36% 70,05% 8,13%

Total Ex. Germany 0,21% 3,40% 5,47% 4,31% 0,20% 27,79% 7,72%

Total Inc. Germany 2,11% 2,82% 5,05% 3,87% 0,01% 27,48% 7,25%

42

Table 3.6: Single variable OLS Coefficient output

Table 3.7: Single variable OLS Coefficient output (2)

The study of the pooled regression coefficients, displayed on table 3.6 and 3.7,

highlights the consistency of each indicator and the assumed sign effect it would have on

spreads. However some odd results were found such as for example the link between current

account and spreads in Ireland, Portugal and Spain which was assumed to be negative and

displayed as strongly positive. Those unexpected results show the difficulty of conducting a

cross-country analysis on spreads as each country reacts differently and is more sensitive to

different variables.

Regress ion Coefficient

OutputBudget Balance

Consumer

Confidence

Current

AccountDebt to GDP

Economic

Sentiment

Industrial

Production

3M rolling

return Local

MSCI

Austria -0,10 -0,12 0,01 0,02 -0,02 0,01 -1,36

Belgium -0,21 -0,15 -0,16 0,02 -0,01 0,03 -0,09

Finland -0,03 -0,06 -0,03 0,00 -0,01 0,00 -0,85

France -0,10 -0,14 -0,18 0,02 -0,02 -0,03 -0,39

Ireland -0,05 -0,08 0,32 0,04 -0,12 0,00 8,43

Ita ly 0,21 -0,66 -0,17 0,09 -0,10 -0,08 -2,19

Netherlands -0,07 -0,11 0,01 0,01 -0,01 0,01 -0,81

Portugal -0,19 -1,02 0,30 0,07 -0,23 -0,24 -9,79

Spain -0,20 -0,66 0,23 0,05 -0,09 -0,08 -4,14

Total Ex. Germany -0,13 -0,09 -0,05 0,02 -0,07 -0,03 -1,64

Total Inc. Germany -0,13 -0,09 -0,06 0,02 -0,06 -0,02 -1,64

Expected Sign Negative Negative Negative Positive Negative Negative Negative

Regress ion

Coefficient Output

Nominal GDP

3M rolling

Value (x10 000)

Nominal GDP

3M rolling

growth rate

/10

GDP PPP Per

Capita

Growth Rate

Real GDP 3M

rolling Growth

rate

Trade Balance

Growth Rate

Unemployment

Rate

Wages Level

Growth Rate

Austria 0,17 -0,03 21,83 -0,29 -56,02 0,00 221,75

Belgium 0,28 -0,01 40,70 -0,39 -36,73 -0,37 -406,76

Finland 0,14 0,00 11,01 -0,03 -7,39 -0,02 87,88

France 0,05 -0,02 21,13 -0,24 -48,11 0,18 48,82

Ireland -0,17 -0,01 -7,38 -0,24 38,62 0,33 519,31

Ita ly 0,25 -0,08 85,82 -0,79 60,84 0,35 506,32

Netherlands 0,07 -0,02 14,55 -0,17 5,91 0,02 7,89

Portugal 3,14 -0,23 186,23 -2,15 193,93 0,60 265,46

Spain 0,20 -0,09 78,04 -1,40 114,59 0,18 206,12

Total Ex. Germany 0,00 -0,02 51,44 -0,23 2,41 0,20 310,19

Total Inc. Germany -0,01 -0,01 48,24 -0,21 0,54 0,19 302,52

Expected Sign Negative Negative Negative Negative Negative Positive Negative

43

In order to be able to build up a combined model we will need to have the same sign

effect on spreads, hence adjust all the indicators to have a negative impact on spreads by

changing the sign of the indicators which have a positive sign on spreads such as debt to GDP

ratio and unemployment level.

The next step of our study includes a multiple regression on every variable in order to

analyze how they behave together and filter explanatory secondary variables. In order to mix

up each indicator, we will proceed on a z-scoring techniques on each variable.

3.2.3 Multiple Regression output:

Tables 3.8 and 3.9 (Appendix) display the high explanatory power of 144 economic

fundamental variables on spreads on 10 countries and 14 years monthly: which sums up to

1620 observations. The adjusted R² of the multiple regression explains 51% of the 1620

observed spread values.

An analysis of the statistical testing of each variable on the multiple regression output

will allow us to select secondary variables in order to optimize our model. Economic

sentiment, unemployment rate and consumer confidence level while highly correlated to the

primary variables seem to be source of added value of significance to the model, beyond their

dependency to primary variables.

Moreover the positive coefficient of budget balance and real GDP growth rate on

spreads is not consistent with our assumptions and pooled regressions. It highlights also the

noise made by autocorrelation on those primary selection tests of the study.

Following on the above studies we selected economic sentiment level, unemployment

rate level and consumer confidence level as secondary variables to add to our combined

model.

Charts 3.1 and 3.2 (Appendix) display the explanatory power and coefficients of each

selected fundamental economic variables.

44

3.3 Fundamental Liquidity Output

Table 3.10: Analysis summary

3.3.1 Correlation output:

Table 3.11: Fundamental liquidity correlation matrix

Table 3.11 highlights a strong link between bid-ask spread, bid-ask spread spot ratio

and level of spreads as well as negative link between the proportion of public debt and

spreads. It also displays the very strong correlation between bid-ask spread and bid-ask spread

spot ratio, demonstrating that if we want to avoid bias in the regression we will need to

remove one of the two indicators.

3.3.2 Pooled regression output

Table 3.12: Single variable OLS R² output

Test start date 31/10/2001

Test end date 30/03/2015

Observed date periods : 162

Observed countries : 10

Observed independant variables : 4

Observed independant variables : 1

Correlation Table Ex. Germany Spread Rate

Proportion of

EMU

Outstanding

Outstanding

Growth Rate

Bid-Ask

Spread

Bid-Ask

Spread/ Spot

Rate

Spread Rate -6% -5% 83% 55%

Proportion of EMU Outstanding -3% -11% -10%

Outstanding Growth Rate 24% -15%

Bid-Ask Spread 67%

Bid-Ask Spread/ Spot Rate

R² Output Proportion of EMU Outstanding Outstanding Growth Rate Bid-Ask Spread Bid-Ask Spread/Spot

Austria 42,44% 0,22% 27,76% 0,29%

Belgium 29,01% 0,96% 63,39% 32,12%

Finland 0,60% 0,81% 10,44% 0,60%

France 8,45% 0,01% 4,52% 16,95%

Ireland 21,75% 0,08% 76,84% 60,43%

Italy 64,63% 2,94% 26,61% 17,36%

Netherlands 0,97% 0,08% 23,71% 1,08%

Portugal 11,87% 8,99% 84,08% 86,95%

Spain 51,41% 0,01% 73,13% 55,44%

Total Ex. Germany 0,36% 0,22% 68,49% 30,65%

Total Inc. Germany 1,49% 0,22% 68,26% 31,17%

45

Table 3.13: Single variable OLS coefficient output

Pooled regression output (Table 3.12 and 3.13) displays the strong explanatory power

of both bid-ask indicators and show a positive coefficient link between them and Spreads as

we assumed.

However the output also highlights the lack of explanatory power of market depth.

Indeed market depth is an economic indicator as well as a liquidity one. Previous analysis

highlighted the high positive explanatory power of debt to GDP ratio on spreads, however

proportion of public debt outstanding and its growth rate are assumed to have negative effect

on spreads. The dilemma between liquidity and economic features highlights the

“schizophrenia” of such indicators.

3.3.3 Multiple regression output

Table 3.9 (Appendix) displays the strong explanatory power of 4 liquidity fundamental

variables on spreads on 10 countries and 14 years monthly: which sums up to 1620

observations. The adjusted R² of the multiple regression explains 68% of the 1620 observed

spread values.

Regression output analysis allows us to filter our variables, indeed bid-ask spread

doesn’t exclude the null hypothesis while bid-ask spread/spot does. We retain bid-ask spread

as the most consistent indicators of both.

Coefficient Output Proportion of EMU Outstanding Outstanding Growth Rate Bid-Ask Spread Bid-Ask Spread/Spot

Austria -115,45 -0,72 9,61 0,73

Belgium -48,97 -3,15 43,73 84,02

Finland 1,48 -0,83 10,34 2,40

France -59,36 0,14 15,01 62,83

Ireland 241,86 1,93 18,72 144,56

Italy -44,48 -12,46 34,21 151,32

Netherlands 9,92 -0,23 21,02 3,44

Portugal 509,89 -42,63 13,95 163,77

Spain 67,12 -0,89 62,35 274,87

Total Ex. Germany -0,99 -3,38 16,19 77,10

Total Inc. Germany -1,85 -3,27 16,35 78,46

Expected Sign Negative Negative Positive Positive

46

Charts 3.3 and 3.4 (Appendix) give us a graphical insight of the explanatory power and

coefficient sign of the two retained liquidity indicators on spreads in the EMU.

3.4 Quantifiable Extra-Fundamental Output:

Table 3.14: Analysis summary

3.4.1 Correlation output:

Table 3.15: Extra-fundamental variables correlation matrix

Correlation Table 3.15 highlights the high correlation between each indicator except

the upgrade dummy on spreads.

It also shows us the strong positive link between the downgrade dummy, spot rate and

political risk towards the contagion indicator, which was expected as downgrades and

political risk grades highlight increase in risk.

Moreover the high correlation between spot rate and spreads was assumed and

understood, there is a bias risk using this variable as an explanatory variable.

Test start date 31/10/2001

Test end date 30/03/2015

Observed date periods : 162

Observed countries : 10

Observed independant variables : 5

Observed dependant variables : 1

Correlation Table10Y Spot

Rate

10Y Spread

Rate

Downgrade

DummyPol i tica l Risk

Upgrade

Dummy

Volati l i ty/

Contagion

10Y Spot Rate 68,42% 9,36% 23,22% -3,14% 48,16%

10Y Spread Rate 18,55% 52,35% -0,92% 64,36%

Downgrade Dummy 10,88% -1,64% 22,39%

Pol i tica l Risk 5,49% 26,89%

Upgrade Dummy -3,46%

Volati l i ty/Contagion

47

3.4.2 Pooled regression output

Table 3.16: Single variable OLS R² output

The analysis of the R² output (Table 3.16) enables us to confirm strong relationships

between each indicator and spreads, however the upgrade dummy shows no significant result.

This can be explained by the low number of upgrades compared to the volume of data

observed. It also show the overreaction of investors to negative change of grades while they

react rationally to upgrades.

Table 3.17: Single variable OLS R² output

The coefficient study of table 3.17 allows us to check our sign of influence

assumptions and validates them. However we found an odd result concerning the Downgrade

Dummy in Finland, indeed spreads decreased on the month of the announcement of the

Downgrade.

R² Output 10Y Spot Rate Downgrade Dummy Political Risk Upgrade Dummy Volatility/Contagion

Austria 3,32% 0,98% 24,60% 14,92%

Belgium 1,47% 14,21% 49,27% 0,57% 32,89%

Finland 2,67% 0,09% 0,14% 0,00% 12,04%

France 27,98% 3,94% 1,01% 11,16%

Ireland 68,34% 0,29% 40,78% 0,11% 14,09%

Italy 15,20% 5,59% 32,53% 0,34% 36,43%

Netherlands 11,56% 0,50% 36,71% 12,13%

Portugal 84,76% 2,72% 45,61% 66,93%

Spain 26,58% 8,46% 54,30% 0,09% 26,59%

Total Ex. Germany 46,82% 3,44% 27,40% 0,01% 41,43%

Total Inc. Germany 44,90% 3,69% 28,89% 0,00% 40,52%

Coefficient Output 10Y Spot Rate Downgrade Dummy Political Risk Upgrade Dummy Volatility/Contagion

Austria -0,05 0,28 0,08 5,16

Belgium -0,06 1,81 0,07 -0,36 10,82

Finland -0,02 -0,06 0,00 -0,01 2,62

France -0,16 0,39 0,02 4,70

Ireland 1,02 0,48 0,14 -0,53 15,07

Italy 0,55 1,36 0,13 -0,93 21,43

Netherlands -0,05 0,15 -0,04 2,52

Portugal 1,15 2,13 0,20 28,70

Spain 0,75 1,76 0,15 -0,23 18,38

Total Ex. Germany 0,70 1,64 0,08 -0,16 22,92

Total Inc. Germany 0,66 1,71 0,08 -0,07 22,68

Expected Sign Positive Positive Positive Negative Positive

48

3.4.3 Multiple Regression Output:

The table 3.18 (Appendix) displays the results of the third regression and shows the

strong explanatory power of 5 extra-fundamental variables on spreads on 10 countries and 14

years monthly over the three multiple regressions we built. The adjusted R² of the multiple

regression explains 69% of the 1620 observed spread values. Moreover the critical F

probability result excludes the null hypothesis with a range of confidence of nearly 100.00%.

A glance at the statistic table 3.18 (Appendix) for each variable validates the

consistency of our variables picking, however the upgrade dummy variable’s explanatory

power is not as strong as other variables as expected during our single variable regression

analysis.

Charts 3.5 and 3.6 (Appendix) display our final variables pooled regression results

graphically. We removed the upgrade dummy from our final choice of variables because of its

null explanatory power. We however still assumed that it is an explanatory variable on

spreads, however the very low number of downgrades in the chosen range makes it difficult to

observe it.

We removed contagion from the coefficient chart because of its higher than the other

level. Every variable has however positive coefficient effects on spreads.

3.5 Fundamental Relative Economic Grade Model

Once we selected our final variables and proceeded to the z-scoring of each one in

order to void the unit issue, we can proceed to the elaboration of a combined and optimized

model of each pillar we are studying.

We proceed first to an unweighted combined model, using a non-weighted average of

each variable as an indicator that we test against spreads. In order to adjust the coefficient to

the same expected sign we adjusted both debt to GDP ratio z-score and unemployment rate z-

score by multiplying them by (-1) in order to have each indicator having the same assumed

negative effect on spreads.

49

Table 3.19: Unweighted combined OLS results

The unweighted combined results displayed on table 3.19 give us a first insight on the

explanatory power of our combined model as well as displaying the coefficient of our

variable. In the case of Fundamental Economic variables the average of our variables show an

expected negative coefficient on spreads.

The output also displays the disparities in the explanatory power between each

variable, we will need to optimize the model in order to break down the different dynamics of

spreads valuation according to economic grading.

Table 3.20: Optimized combined OLS results

The second step of our analysis consists in the analysis of the output of a combined

model while optimizing the R² score using a solver without further restrictions, it is showed

on table 3.20.

The result of the analysis show 46% of the variations of spreads explained by the

model, highlighting 12% difference from the multiple regression we made without weights.

This difference of R squares can be explained by the stronger explanatory power of variables

such as debt to GDP and unemployment rate.

Budget Balance Current Account Debt to GDP Nominal GDP growth rate Total Primary

14,29% 14,29% 14,29% 14,29% 57,14%

Economic Sentiment Unemployment rate Consumer Confidence Total secondary

14,29% 14,29% 14,29% 42,86%

100,00%

34%

0,00

-1,02Coefficient

Unweighted Combined

Primary Indicators

Secondary Indicators

Total indicators

Combined Regression results

R square

Intercept

Budget Balance Current Account Debt to GDP Nominal GDP growth rate Total Primary

0,00% 9,24% 28,88% 0,00% 38,12%

Economic Sentiment Unemployment rate Consumer Confidence Total secondary

17,79% 28,74% 15,35% 61,88%

100,00%

46%

0,00

-1,11

R square

Intercept

Coefficient

Optimized Solver Combined (without restrictions)

Combined Regression results

Primary Indicators

Secondary Indicators

Total indicators

50

Table 3.21: Optimized with restrictions combined OLS results

We finally proceed to a final combined model using restrictions on minimum

weighting of each variable and a minimum of 60% of the weighting from primary indicators.

The final result displayed on table 3.21 is 43% R square on 1620 observations broken

down in 10 countries during 14 years. The most weighted indicators are the debt to GDP ratio,

the unemployment rate and the current account to GDP ratio.

Chart 3.7: Fundamental Economic Model Significance (Rolling 1 Year)

Chart 3.7 translates our result into a 1 year rolling R², in order to understand the shifts

in explanatory power over time. While the explanatory power of the model is very low in the

first years that we studied, it increased steadily from 2008 to 2011. Peaking during the

economic crisis of 2008-2009 and European debt crisis of 2011-2012.

Budget Balance Current Account Debt to GDP Nominal GDP growth rate Total Primary

5,00% 14,45% 34,82% 5,73% 60,00%

Economic Sentiment Unemployment rate Consumer Confidence Total secondary

10,98% 20,14% 8,88% 40,00%

100,00%

43%

0,00

-1,10

Intercept

Coefficient

Optimized Solver Combined (with parameters restrictions)

Secondary Indicators

Total indicators

Combined Regression results

R square

Primary Indicators

0,00%

10,00%

20,00%

30,00%

40,00%

50,00%

60,00%

70,00%

80,00%

90,00%

2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014

51

3.6 Fundamental Relative Liquidity Grade Model

We proceeded to the same analysis on Fundamental liquidity in order to translate

multiple variables into a weighted combined model. The retained indicators are bid-ask

spread and proportion of EMU public debt outstanding.

We previously highlighted the fact that we expected the proportion of EMU public

debt outstanding to have both a negative and a positive effect on spreads. However we kept

the negative coefficient on spreads as the focus is on the liquidity part of the explanatory

power and not the debt management.

Table 3.22: Unweighted combined OLS results

The R square of the unweighted combined model displayed on table 3.22 is of 41% of

spreads values explained by the model. 27% lower than the multiple regression we previously

built up. This is explained by the unbalanced explanatory power between the two variables.

Indeed we assume the bid/ask spread to have a stronger explanatory power than the

proportion of public debt outstanding.

Table 3.23: Optimized restricted combined OLS results

Bid-Ask spread Proportion Outstanding Total

50,00% 50,00% 100,00%

41%

0,00

0,83

Unweighted Combined

Indicators

Combined Regression results

R square

Intercept

Coefficient

Bid-Ask spread Proportion Outstanding Total

90,00% 10,00% 100,00%

68%

0,00

0,86Coefficient

Restricted Solver

Indicators

Combined Regression results

R square

Intercept

52

Table 3.23 highlights the previous assumption and validates it, we used a solver in

order to maximize the R² while setting up minimum conditions on the weightings of each

variable to 10%. Bid-Ask Spread represents 90% of the model, validating our thinking and

literature from Gomez-Puig (2006), liquidity is not best defined by market depth but by the

price attractiveness of assets and its elasticity to shifts in demands. Bid-ask spread represents

more accurately this feature of securities.

Chart 3.8: Fundamental Liquidity Model Significance (Rolling 1 Year)

Chart 3.8 resumes the history of the 1 year rolling R² of our combined model. We can

observe a drastically low level of significance from 2006 to 2011. However there is a very

strong rebound of significance from 2010 to 2013. This highlights the impact of crises on

bond valuation dynamics and the increasing attractiveness of liquidity in tough economic

times. We also observe a drop in the liquidity valuation feature from 2013 to 2014,

highlighting a shift back towards other valuation dynamics.

3.7 Extra-Fundamental Relative Grade Model

The final combined model for a pillar specific variable includes the same

methodology. We set up two primary indicators: nominal spot rate and contagion/volatility.

We also set up two secondary indicators: downgrade DUMMY and political risk. We also

bear in mind that some extra-fundamental variables are not quantifiable or not cross-sectional

and therefore a part of the explanatory power of extra-fundamentals is not captured by the

study.

0,00%

10,00%

20,00%

30,00%

40,00%

50,00%

60,00%

70,00%

80,00%

90,00%

2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014

53

We chose to separate downgrade and political risk from spot rate and contagion

because of their high correlation factor.

Table 3.24: Unweighted combined OLS results

Table 3.24 shows a significance of 60%, only 9% below the multiple regression

significance level. This high significance level highlights the effect of 3 correlated factors and

the explanatory power of each variable.

Table 3.25: Optimized combined OLS results

Table 3.25 displays the second OLS output, it shows a very high R² of 70%, it is the

same significance output as our multiple regression. This result validates our choice of

variables as well as the consistency of our weighted-average combined methodology.

Nominal spot rate is the most significant variable of the combined model, while the

downgrade dummy is the least significant one.

Spot Rate Contagion Total Primary

25% 25% 50%

Downgrade DUMMY Political Risk Total secondary

25% 25% 50%

1

60%

0,00

1,18

Combined Regression results

Unweighted Average Combined

Total indicators

R square

Intercept

Coefficient

Primary Indicators

Secondary Indicators

Spot Rate Contagion Total Primary

43% 25% 68%

Downgrade DUMMY Political Risk Total secondary

3% 28% 32%

1

70%

0,00

1,15

R square

Intercept

Coefficient

Primary Indicators

Combined Regression results

Secondary Indicators

Optimized Combined

Total indicators

54

Chart 3.9: Extra-Fundamental Model Significance (Rolling 1 Year)

Chart 3.9 sums up the history of the significance of the combined Extra-Fundamental

model, it is steadily increasing from a low start from 2006 to 2014. Interestingly enough the

biggest increases in significance are noticed in out of crisis times, such as 2006-2007 and

2010-2011.

3.8 Time-frame of significance

Chart 3.10: Pillar specific models (Rolling 1 Year R²)

0,00%

10,00%

20,00%

30,00%

40,00%

50,00%

60,00%

70,00%

80,00%

90,00%

100,00%

2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014

0,00%

10,00%

20,00%

30,00%

40,00%

50,00%

60,00%

70,00%

80,00%

90,00%

100,00%

2002 2004 2006 2008 2010 2012 2014

Fundamental Economic Fundamental Liquidity Extra Fundamental

55

Chart 3.10 displays the significance level of each pillar specific model over time.

Each pillar shows a steadily growth rate over time and high levels of significance from

2009 to 2015. However Liquity’s significance drops drastically from 2012 to 2015.

Chart 3.11: Timeframe of Pillar significance (Z score)

We then proceeded on a z-scoring of each 1 year rolling R square for each pillar in

order to get eras of predominance of each component. Eras of significance highlights periods

of time when each model was higher according than the others adjusted to its own mean.

Chart 3.11 shows the result of our analysis, eras of significance alternate from one

pillar to the other over time, we will need to observe possible causes of such shifts.

3.9 Combined model

Once each pillar specific model built, we are now able to mix them up in order to

create a combined model from the three different pillars mentioned above: Fundamental

Economic Relative Grade (FREG), Fundamental Relative Liquidity Grade (FRLG) and Extra-

Fundamental Relative Grade (XFRG).

3.9.1 Correlation Output

As computed earlier the estimate equation of each pillar specific models are the

following:

Spread Value (E) = -1, 05 * FREG

Spread Value (E) = 0, 97 * FRLG

Spread Value (E) = 1, 15 * XFRG

2002 2004 2006 2008 2010 2012 2014

Fundamental Economic Fundamental Liquidity Extra Fundamental

56

Table 3.26: Pillar correlation matrix

We expected a negative correlation between Fundamental Economic on the one side

and Fundamental Liquidity and Extra-Fundamental on the other according to each pillar OLS

coefficients.

Correlation output from table 3.26 also displays a higher than expected rate of

correlation between Liquidity and Extra-Fundamentals, which could be a source of bias in the

combining model.

3.9.2 Combining models

We will proceed to the same combination methodology as the pillar specific models in

order to mix up the three components of our combined model.

In order to make an unweighted average of the models we will first adjust the

Fundamental economic grades by multiplying them by (-1) in order to have the same expected

coefficient sign on each variable.

Table 3.27: Pillar unweighted OLS output

The unweighted average provided by table 3.27shows a R² of 86%. It is way higher

than each separate pillar model significance. This would be explained by a stronger

significance of each pillar on yield spreads but also highlighting the high complementarity of

each pillar with the others.

Correlation Table Fundamental Economic Fundamental Liquidity Extra-Fundamental

Fundamental Economic -15% -59%

Fundamental Liquidity 48%

Extra-Fundamental

Fundamental Economic Fundamental Liquidity Extra-Fundamental

33,33% 33,33% 33,33%

100,00%

86%

0,00

1,44Coefficient

Primary Indicators

TOTAL

Combined Regression results

R square

Intercept

57

Table 3.28: Pillar optimized OLS output

The final model is optimized with 10% minimum weight per component gives us the

following weighting result (Table 3.28): Fundamental Economic Relative Grade 31,66%,

Fundamental Liquidity Relative Grade 35,12%, and Extra-Fundamental Relative Grade

33,19%.

The level of R² on over 162 dates on 10 countries and using 3 explanatory variables is

86%. This level include a cross-country analysis and a 14 monthly years range.

3.10 The influence of credit ratings

In order to check the consistency of our grading we will compare it to an unweighted

average of the credit ratings of the “Big-Three” rating agencies.

Table 3.29: Big Three correlation matrix

Correlation between our Fundamental Economic grade and the credit ratings of the big

three displayed on table 3.29 is very high: 79%, highlighting the consistency of our variable

choice and validating our methodology for assessing credit quality.

It also displays that the most important dynamics of credit valuation inside the EMU is

the credit quality which can be translated by Fundamental Economic grading. Liquidity and

Extra-Fundamental are also significantly correlated to the ratings. Moreover the whole

combined model is also highly related to the ratings (75%), proving the consistency of the

analysis and our variables selection.

Fundamental Economic Fundamental Liquidity Extra-Fundamental

31,66% 35,15% 33,19%

100,00%

86%

0,00

1,42Coefficient

Primary Indicators

TOTAL

Combined Regression results

R square

Intercept

Fundamental Economic Fundamental Liquidity Extra-Fundamental Combined Model

79% 50% 67% 75%

Correlation Vs. Big Three Credit Ratings

58

3.11 Interpretations and implication

Our modelization of spreads inside the EMU over 3 pillars shows interesting

significance level ranging from 43% for our Fundamental Economic grading to 70% for our

Extra-Fundamental grading. Those coefficients of significance highlight the consistency of

our choice of variables. Moreover the rate of significance seems to be increasing over time

(exception made for Liquidity), increasing the short term consistency of our predictive model.

The goal of this model is to enable us to understand the dynamics of spreads valuation

as well as be able to link them to exterior factors such as cycle variations. It also enable us to

describe the shifts in predominant dynamics over time.

We followed up on Min (1998) using both liquidity and economic indicators to build

up a spreads valuation model. However our study focus on a longer time range and is cross-

country. This implies more data observed which should enable us to have better significance

however the cross-country implication makes it harder to get high levels of significance due

to country specific features in the valuation of spreads.

Caceres, Guzzo and Segovianno (2010) studied the effects of both contagion and

fundamentals, we considered contagion as one part of extra-fundamentals, adding nominal

spot rate, political risk and downgrades.

Our combined model is very highly correlated to credit ratings (over 75%), this

correlation results confirms our data choice and methodology in order to define credit quality

relying on fundamentals.

The combined model and pillar specific ones are both descriptive models and

predictive ones. Indeed by defining a significance level and shaping the model to each

country, it can be used in order to assess fundamental value of each spread, the difference

between the sport value of a spread and its assessed fundamental value would define the

upside/downside to be expected. Moreover forecasts are available on the Bloomberg

Datastream and up to three years range enable us to create forecast lines.

59

3.12 Limits of the study and feedback

We managed to illustrate the predominance of liquidity in the 2000-2002 period

following up on Gomez-Puig (2006) analysis on liquidity, however we failed to reach the

same levels of significance as her analysis on the same period.

Westphalen (2001), used some identical parameters in order to illustrate spreads in

Emerging markets and reached a lower significant level. However the markets have some

features of their own and cannot be compared. Moreover we agree with his theory that the rest

of the variations of spreads must be originated from extra-fundamental factors which could be

country specific or in the case of Westphalen (2001) common factors.

Attinasi, Checherita and Nickel (2009) emphasized the fact that since the 2008-2009

crisis credit quality defines bond valuation and explain the widening of spreads since that

time. Our model results seems to be correlated to that analysis, as our significance level is

very low before the 2007-2008 period and drastically increased since the crisis.

Our analysis is cross-country and made on a 14 years range, both time range and

cross-country methodologies are possible reasons of the 86% level of significance of our

model. Yearly and country specific models could be solutions to those issues. However it will

then be impossible to compare countries or years to each other.

Other potential reasons of the lack of significance in the explanatory power are the fact

that we included Germany and mixed up different data frequencies. Indeed we included

Germany in the cross-country models because we have not reasons exclude it even if we are

using spreads. Indeed its value of spreads remains 0 while its fundamental variables

fluctuates. Excluding Germany from the analysis did not increase significantly the R² of our

model, this is the reason why we kept it included. We used different frequencies database as

indicators, assuming constant growth rate for quarterly and yearly data while our study is

based on monthly spreads. This difference of frequency is according to us the most important

reason of the lack of significance, indeed we lost 2/3rd (and in some cases 11/12th) of the

volatility of most of our indicators, and then lost significance. However this explanation

remains uncertain, as price valuation of assets is not reassessed monthly and therefore

continuously shifting, confirming our assumption that we could mix monthly and quarterly

data.

Concerning our variables choices, we used both theoretical thinking and statistical

testing in order to assess those variables, databases could be contested and changed, however

we felt like it was the best trade-off between theory and statistics.

60

4. RESULTS ANALYSIS

4.1 Clusters

In order to separate each of the three clusters of countries, we will use two variables:

Big three credit agencies average ratings and average nominal spot rate. We will then proceed

to a Rating – Nominal spot ratio. The ratio includes credit quality features and risk – value

features; a high ratio means low risk and high quality.

Chart 4.1: Big Three Rating – Nominal spot rate ratio

The visual output of each ratio is displayed on chart 4.1, those ratios allow us to

separate the countries and create 3 clusters as following according to each countries rank:

Core Higher Quality: Germany, Finland, Netherlands

Core Lower Quality: France, Austria, Belgium

Peripherals: Ireland, Spain, Portugal, Italy

Table 4.1: Comparative R² Output according to clusters

0

0,2

0,4

0,6

0,8

1

1,2

1,4

1,6

1,8

2

BIG

TH

REE

/ N

OM

INA

L SP

OT

Rat

io

RSQ Fundamental Economic Fundamental Liquidity Extra-Fundamental Combined Model

Core - Higher Quality 1% 18% 0% 3%

Core - Lower Quality 29% 20% 8% 41%

Peripherals 43% 70% 80% 88%

Total 43% 68% 70% 86%

Comparative significance

61

Table 4.1 displays the comparative significance of each cluster’s models. We can

observe a very strong significance for each model on peripherals, while Core - lower quality

displays average to low significance. Core – higher quality’s significance is strongly lower

than the two other clusters.

Moreover each model shows different sensibility according to each pillar:

Core Higher Quality shows higher relative significance on fundamental liquidity.

Core Lower Quality displays higher relative significance on economic features

Extra-fundamental relative significance is higher on extra-fundamental

Clusters are not the only parameters that affect bond valuation dynamics, cycle shifts from

expansionary periods to recessions also contributes to a change in investment behavior.

4.2 Where are we in the cycle?

In order to compute our cycle index, we z-scored 4 economic cycle indicators in order

to aggregate them:

MSCI monthly returns

M2 –Money supply

Nominal GDP growth rate

OECD LEI (Leading Economic Indicator)

Table 4.2: Cycle Index Weighting output

We then proceeded to the creation of an index combining those indicators for both the

US and the EMU. The weights of each indicator is detailed on the table 4.2.

US EMU

MSCI Monthly Returns 10,00% 2,50% 7,50%

M2 Money Supply 20,00% 5,00% 15,00%

OECD Leading Economic Indicators 30,00% 7,50% 22,50%

Nominal GDP Growth Rate 40,00% 0,00% 40,00%

Total 100,00% 15,00% 85,00%

Geographic weight Breakdown

Aggregate Cycle Index Weights

Indicator weightIndicators

62

Chart 4.2: Aggregate Cycle Index

Chart 4.2 resumes the variations of our combined cycle index, we can notice the strong

effect of the sub-primes crisis on the indicator as well as the Eurozone crisis. The weightings

of our indicators is consistent with both a global outlook as well as an EMU specific analysis.

4.3 Crises effect on sovereign bond valuation

Thanks to the aggregate cycle index we built, we were able to assess periods of crisis

and periods of recovery. According to the slope of our cycle index, we define crisis and

recovery periods.

Chart 4.3: Crisis / Recovery Timeframe

Chart 4.3 highlights the crisis/recovery breakdown over time and allows us to run our

model on both crisis and recovery periods.

-0,80

-0,60

-0,40

-0,20

0,00

0,20

0,40

2001 2003 2005 2007 2009 2011 2013

Crisis Period Recovery Period

63

Table 4.3: Crisis / Recovery periods: Comparative significance output

Table 4.3 displays our regression output according to the periods we are in. The

significance of our combined model as well as each pillar specific model is drastically

stronger in crisis periods, highlighting the change in investors’ behaviors. While fundamental

economic features are strong over both periods, all other model shows that investors tend to

invest more on liquidity and extra-fundamentals in times of financial distress, while they tend

to invest more on feelings and less rational indicators in time of recovery.

Table 4.4: Pre-subprime / Post-subprime periods: Comparative significance output

Table 4.4 displays a second analysis made before the subprime crisis and after. This

results shows us that the subprime crisis had a significant impact on investment style

concerning sovereign bonds, and moreover this investment style seems to have kept going

even in recovery periods.

4.4 Back testing Returns

Our combined model as well as each pillar specific model provides a tool in order to

estimate sovereign bond spreads in the EMU. According to our study results, our combined

model allows us to assess 89% of the fluctuations of spreads.

In order to use the model to define residual value of spreads, we should bear in mind

the following equations:

Spread Value (E) = Combined Regression Coefficient * Combined Regression Score

Spread Value = Combined Regression Coefficient * Combined Regression Score + ὲ

RSQ Fundamental Economic Fundamental Liquidity Extra-Fundamental Combined Model

Crisis Period 50% 76% 78% 89%

Recovery Period 48% 41% 52% 72%

Spread 2% 35% 26% 17%

Comparative significance

RSQ Fundamental Economic Fundamental Liquidity Extra-Fundamental Combined Model

Pre-Subrpimes

crisis Period 1% 8% 10% 2%

Post-Subrpimes

crisis Period 49% 69% 79% 90%

Spread -48% -61% -69% -87%

Comparative significance

64

Spread Value (E) t = Spread Value t - ὲt

Our model assesses a residual value estimate of spreads, the actual value of spreads is

not necessarily equal to its residual as some other chronic parameters can cause a

misevaluation.

Indeed starting from that assumptions we can assess area of opportunities, which are

the gaps between the actual value and our estimate. As we consider this gap a misevaluation,

we think that spreads are going to tend towards the residual value estimate we computed.

Chart 4.4: Combined Spread Estimate - Actual - Austria

Chart 4.5: Combined Spread Estimate – Actual - Belgium

Charts 4.4 and 4.5 highlight the differences between our residual estimates and the

actual value of spreads for Austria and Belgium. Those gaps are considered in our analysis as

opportunities for investment.

-0,25

-0,2

-0,15

-0,1

-0,05

0

0,05

0,1

0,15

0,2

2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014

Spread ValueCombined Spread Estimate

-0,2

-0,1

0

0,1

0,2

0,3

0,4

0,5

2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014

Spread Value

Combined Spread Estimate

65

Once we computed each country’s estimate – actual spreads, we obtain “value grades”

which are according to our theory areas of opportunities for investments. We then used the

Euro Government Index provided by Bank of America Merill Lynch in order to set up a

benchmark and run our model against it, the weights used in the benchmark are displayed on

chart 4.6 (Appendix). Our model will adjust the weights of the benchmark according to the

“bets” it is taking, those “bets” consist in the spreads between our estimate and the actual

value of spreads at a time t.

Table 4.5: Investment case output

Table 4.5 displays the result of our back test against the Euro Government Index using

our combined model as an investment tool from 2001 to 2015 annualized and resumed our

average relative positioning and weights over the same period to finally assess an annualized

excess return over the period.

While our output table shows a positive excess return of 0.6 basis point per year, chart

4.7 (Appendix) displays the inconsistency of our excess return index and chart 4.8 (Appendix)

highlights the volatility of monthly excess returns.

Benchmark

Weight

Weighted

Return

Real-Combined

Model Spread Position

Adjusted

Weight

Excess

Return

Austria 3,82% 0,07% 0,03 UW 3,73% -0,0008%

Belgium 6,24% 0,12% 0,09 UW 5,77% -0,0078%

Finland 1,49% 0,03% -0,08 OW 1,63% 0,0028%

France 23,72% 0,42% 0,02 UW 23,43% 0,0037%

Germany 18,43% 0,34% -0,01 OW 18,80% 0,0123%

Ireland 1,96% 0,04% -0,05 OW 2,08% 0,0057%

Italy 24,60% 0,49% 0,02 UW 24,28% -0,0101%

Netherlands 6,19% 0,11% -0,03 OW 6,47% 0,0069%

Portugal 0,00% 0,00% 0,04 UW 0,00% 0,0000%

Spain 13,56% 0,27% -0,01 OW 13,83% -0,0073%

Total 100% 1,89% 100% 0,0060%

Investment Characteristics (Average Annualized)

66

Chart 4.9: Combined Cycle Index vs. Excess Return Index

Chart 4.9 displays a comparison between our excess return index and our cycle proxy,

we can observe two strong period of loss, firstly during the 2001 to 2003 period, consistent

with the low significance of our combined model during that period and secondly during the

2008-2010 recovery period. The excess return index however strongly performed since the

euro-debt crisis as fundamentals became the main factor of valuation of bonds.

4.5 Return uses and performance implication

Our model is a tool which enables investors to adjust their investment behavior over

time and to assess residual value of spreads based on fundamentals. It is however not an

investment model on its own. Indeed the lack of consistency during recovery periods, and

over some countries shows us that the cross-sectional analysis we made carries some gaps.

It is however an interesting tool in order to provide investment guidance

complementary with other analyses made by investors.

94

95

96

97

98

99

100

101

-0,80

-0,60

-0,40

-0,20

0,00

0,20

0,40

2001 2003 2005 2007 2009 2011 2013

Exce

ss R

etu

rn In

dex

(b

ase

10

0)

Co

mb

ined

cyc

le In

dex

Lev

el

COMBINED EMU CYCLE INDEX Excess Return Index

67

4.6 Limits of the model

While the model showed strong significance over 10 countries on 14 years, it is based

on historical values, in order to be able to use the model to assess investment behaviors, we

would need to use forecasts instead of historical data. Precise consensus forecasts can be

found easily for the fundamental economic variables we used, however liquidity and extra-

fundamental variables accurate forecasts will not be found, making the model difficult to use

on the predictive area.

Moreover while the overall significance of our models is high over the studied period,

we do not know if investment behaviors are going to be the same in the futures, hence our

model could become obsolete in the future.

Finally our model does not take into account a main characteristic of bond value,

maturity, indeed we ran our model only for the 10 year bonds and have no insight on how it

would perform on other maturities. Furthermore, our “bets” and excess returns do not include

transaction costs and change monthly, implying high losses from transaction costs.

4.7 Beyond the model

We argued in the section 4.4 of the report that gaps between our estimates and the real

value were areas of opportunities, another theory would be to argue that the opportunity is the

highest when spreads are fair valued, indeed as investment behaviors change over time, a

correct anticipation of investment behaviors while bond spreads are fair valued would be a

significant opportunity for an investor aware of it. Our model could be used on the opposite

way of what we did in section 4.4, by predicting a shift from fair value and making the

opposite “bets” maximizing the return of riskier assets before their deviation from fair value.

Other investigation areas include the same analysis yearly, once done, we should be

able to assess in a more accurate way the shifts in valuation dynamics over time.

Moreover we could proceed to the building of similar models for different maturities

and areas; including other currency areas, and compare the relative outputs in order to get

wider clusters and anticipate shifts in regional valuations.

68

Conclusion

In order to understand the main valuation dynamics of spreads, we analyzed the

previous research works on wider different geographic areas as well as researches focused on

the sovereign bonds inside the EMU. We assumed from the past literature 3 main valuation

features: Fundamental economic variables, fundamental liquidity variables and extra-

fundamental variables.

We then tested our theory using empirical analyzes on EMU sovereign spreads on the

10 year maturity of 10 countries over 14 years monthly. The results of our testing validated

our theory: the three pillars we selected show significant explanatory power. Our results lead

us to the building of four combined models: 3 pillar specific combined model and a final

model regrouping every pillar specific model into one.

This study has both a descriptive and a predictive goal, as we tend to identify the

changes in valuation dynamics of spreads in order to be able to predict future residual value of

spreads.

Each model we built up is an estimate of yield spreads in the EMU according to

specific sets of variables. By analyzing the significance of our results we observed that the

significance of our model was not steady over time and depending on external as well as

internal events.

Those models are tools in order to get a better understanding of valuation dynamics

and could be used as an investment model. However the best use for it is to provide reliable

information on the residual estimate value of spreads. This estimate combined with market

knowledge and complementary financial analysis would allow investors to adjust their

investment behavior and knowledge of the European economy.

We assume that changes in significance of the model originate from changes in the

economic cycle as well as country dependent, limiting therefore a long term and cross country

use of the model. Any investor using this model would have to bear in mind the cluster

influence on the model as well as its higher significance in crisis times.

Alternatively, it would be interesting to compare the model beyond the European

boundaries, adding up other region countries in order to analyze the main differences in the

models would allow us to understand better which valuation parameters are common only in

the EMU area.

69

Bibliography

- Attinasi M-G., Checherita C., Nickel C., (2009) What explains the surge in Euro Area

sovereign spreads during the financial crisis of 2007-09? , ECB working paper.

- Beber A., Brandt M.W., Kavajecz K.A., (2008) Flight-to-Quality or Flight-to-

Liquidity? Evidence from the Euro-Area Bond Market, Oxford University Press.

- Caceres C., Guzzo V., Segoviano M., (2010) Sovereign Spreads: Global risk

Aversion, Contagion or Fundamentals? , IMF working papers.

- Ericsson J., Renault O., Liquidity and Credit Risk, (2006), the Journal of Finance

- Gomez-Puig M., (2006) Size Matters for Liquidity: Evidence from EMU Sovereign

Yields Spreads, Economic Letters.

- Min H.G., Determinants of Emerging Market Bond Spread, (1998), World Bank

vol.1899.

- Remonola E.M., Scatigna M., Wu E., (2007) Interpreting sovereign spreads, BIS

quarterly review.

- Westphalen M., (2001) The Determinants of Sovereign Bond Credit Spreads Changes,

Ecole des HEC, Université de Lausanne, and Fame.

70

Index

Acknowledgements ..................................................................................................................... i

Summary .................................................................................................................................... ii

Abstract ..................................................................................................................................... iii

Introduction ................................................................................................................................ 4

1. LITTERATURE REVIEW .................................................................................................... 3

1.1 Fundamental dynamics of valuation of sovereign spreads ............................................................ 3

1.2 Fundamental economic indicators and sovereign bonds ............................................................... 5

1.3 Fundamental liquidity and sovereign bond spreads ...................................................................... 8

1.4 Trade-off liquidity – Quality ....................................................................................................... 10

1.5 Extra fundamental variables ........................................................................................................ 13

2. METHODOLOGY ............................................................................................................... 17

2.1 Choice of Variables ..................................................................................................................... 17

2.1.1 Sovereign Yield spreads .......................................................................................... 17

2.1.2 Fundamental economic data .................................................................................... 18

2.1.3 Fundamental Liquidity data ..................................................................................... 25

2.1.4 Quantifiable Extra-Fundamental variables .............................................................. 28

2.1.5 Where are we in the cycle? ...................................................................................... 29

2.2 Definition of the model ............................................................................................................... 31

2.2.1 The influence of credit ratings ................................................................................. 31

2.2.2 Fundamental Economic Grade ................................................................................. 31

2.2.3 Fundamental Liquidity Grade .................................................................................. 33

2.2.4 Quantifiable Extra-Fundamental Grade ................................................................... 34

2.3 From pillar specific to combined ................................................................................................. 34

2.3.1 Pillar specific model analysis ................................................................................... 34

2.3.2 Combined model ...................................................................................................... 36

2.3.3 Unquantifiable Extra-Fundamental dynamics ......................................................... 36

2.4 Model analysis ............................................................................................................................. 37

2.4.1 Linking clusters ........................................................................................................ 37

2.4.1 Linking crises ........................................................................................................... 37

2.4.3 Investment insights .................................................................................................. 37

71

2.4.4 Feedback and analysis on the methodology ............................................................. 38

3. DATA ANALYSIS .............................................................................................................. 39

3.1 Yield Spreads .............................................................................................................................. 39

3.2 Fundamental Economic Data ...................................................................................................... 39

3.2.1 Correlation output: ................................................................................................... 40

3.2.2 Pooled regression output: ......................................................................................... 40

3.2.3 Multiple Regression output: ..................................................................................... 43

3.3 Fundamental Liquidity Output .................................................................................................... 44

3.3.1 Correlation output: ................................................................................................... 44

3.3.2 Pooled regression output .......................................................................................... 44

3.3.3 Multiple regression output ....................................................................................... 45

3.4 Quantifiable Extra-Fundamental Output: .................................................................................... 46

3.4.1 Correlation output: ................................................................................................... 46

3.4.2 Pooled regression output .......................................................................................... 47

3.4.3 Multiple Regression Output: .................................................................................... 48

3.5 Fundamental Relative Economic Grade Model .......................................................................... 48

3.6 Fundamental Relative Liquidity Grade Model ............................................................................ 51

3.7 Extra-Fundamental Relative Grade Model .................................................................................. 52

3.8 Time-frame of significance ......................................................................................................... 54

3.9 Combined model ......................................................................................................................... 55

3.9.1 Correlation Output ................................................................................................... 55

3.9.2 Combining models ................................................................................................... 56

3.10 The influence of credit ratings ................................................................................................... 57

3.11 Interpretations and implication .................................................................................................. 58

3.12 Limits of the study and feedback .............................................................................................. 59

4. RESULTS ANALYSIS ........................................................................................................ 60

4.1 Clusters ........................................................................................................................................ 60

4.2 Where are we in the cycle? .......................................................................................................... 61

4.3 Crises effect on sovereign bond valuation ................................................................................... 62

4.4 Back testing Returns .................................................................................................................... 63

4.5 Return uses and performance implication ................................................................................... 66

4.6 Limits of the model ..................................................................................................................... 67

4.7 Beyond the model ........................................................................................................................ 67

72

Conclusion ................................................................................................................................ 68

Index ......................................................................................................................................... 70

Bibliography ............................................................................................................................. 69

73

Appendix

Table 3.1: Data sources

Name Source Frequency

Generic 1Y Yield Bloomberg Monthly

Generic 5Y Yield Bloomberg Monthly

Generic 10Y Yield Bloomberg Monthly

Generic 20Y Yield Bloomberg Monthly

10Y ASW Bloomberg Monthly

5YCDS Bloomberg Monthly

Budget Balance Bloomberg Yearly

Consumer Confidence OECD Quarterly

Current Accounts Bloomberg Quarterly

Debt to GDP Bloomberg Quarterly

Economic Sentiment Eurostat Monthly

Industrial Production OECD Quarterly

3M Rolling local MSCI Monthly returns Bloomberg Quarterly

Nominal GDP 3M Rolling Growth Rate Bloomberg Quarterly

GDP PPP Per Capita OECD Yearly

Real GDP 3M Rolling Growth Rate Bloomberg Quarterly

Nominal Trade Balance Bloomberg Monthly

Unemployment Rate OECD Monthly

Wage Level OECD Yearly

Outstanding Amount of Public Debt BIS Quaterly

Outstanding Amount of Public Debt BIS Quaterly

Bid-Ask Spread Bloomberg Monthly

Bid-Ask Spread/ Spot Rate ratio Bloomberg Monthly

Generic Nominal Spot Rate Bloomberg Monthly

Upgrade DUMMY Bloomberg Monthly

Downgrade DUMMY Bloomberg Monthly

Political Risk Bloomberg Semi-Yearly

Volatility / Contagion Bloomberg Monthly

EFFA / Bond Total Return Index Bloomberg Monthly

OECD / Leading Economic Indicator Bloomberg Monthly

M2 / Money Supply Bloomberg Monthly

DATASET SOURCES

74

Table 3.3: Fundamental economic correlation matrix

Correlation Table10Y Spread

Budget

Balance

Consumer

Confidence

Current

AccountDebt to GDP

Economic

Sentiment

Industrial

Production

3M rolling

return MSCI

World

Nominal

GDP 3M

rolling

growth rate

GDP PPP

Per Capita

Real GDP

Growth rate

Nominal

Trade

Balance

Unemploym

ent RateW

ages Level

10Y Spread-33%

-29%-16%

46%-40%

-17%-4%

-18%23%

-21%4%

53%28%

Budget Balance59%

31%-24%

40%5%

10%28%

-42%40%

-6%-43%

-25%

Consumer Confidence

10%0%

27%12%

9%18%

-24%19%

-59%-18%

10%

Current Account4%

9%-37%

7%2%

-10%13%

44%-19%

-4%

Debt to GDP-17%

-12%7%

-10%17%

-18%-17%

35%40%

Economic Sentim

ent31%

2%33%

-31%53%

1%-21%

-11%

Industrial Production-8%

11%2%

15%-34%

-6%-4%

3M rolling return M

SCI

World

9%-17%

9%0%

6%18%

Nominal GDP 3M

rolling

growth rate-25%

42%3%

-15%-5%

GDP PPP Per Capita-28%

-7%20%

0%

Real GDP Growth rate13%

-12%-9%

Nominal Trade Balance

-15%-26%

Unemploym

ent Rate33%

Wages Level

75

Table 3.8: Statistical assumptions legend

Table 3.9: Fundamental economic variables, multiple regression output

x Assumed Null explanatory power

x No rejection of Null explanatory power

x Rejection of Null explanatory power

Degré de

liberté

Somme des

carrés

Moyenne des

carrés F Valeur critique de F

Régression 13,00 68,83 5,29 128,65 0,000000%

Résidus 1606,00 66,09 0,04

Total 1619,00 134,92

Coefficients Erreur-type Statistique t (Abs t stat) Probabilité

Constante 0,00 0,01 0,00 0,00 100%

BudgBal 0,20 0,03 7,00 7,00 0,00%

ConsuConf -0,20 0,03 -5,92 5,92 0,00%

CurrAcc -0,20 0,03 -7,71 7,71 0,00%

DebttoGDP 0,30 0,02 13,51 13,51 0,00%

EcoSent -0,23 0,02 -9,82 9,82 0,00%

InduProd -0,08 0,02 -3,36 3,36 0,08%

MSCI3M -0,10 0,02 -5,40 5,40 0,00%

NominalGDPgr -0,03 0,02 -1,43 1,43 15,37%

PerCapita -0,18 0,03 -5,67 5,67 0,00%

RealGDP 0,03 0,02 1,23 1,23 21,83%

TradeBal 0,21 0,03 6,58 6,58 0,00%

Unemploy 0,30 0,02 12,87 12,87 0,00%

Wages 0,09 0,02 4,21 4,21 0,00%

Statistiques de la régression

Coefficient de détermination multiple

Coefficient de détermination R^2

Coefficient de détermination R^2

Erreur-type

Observations

ANALYSE DE VARIANCE

71,42%

51,01%

50,62%

0,20

1620

76

Chart 3.1: R² Spreads determinants per country

Chart 3.2: R² Spreads determinants per country (2)

0,00%

10,00%

20,00%

30,00%

40,00%

50,00%

60,00%

Budget Balance Current Account Nominal GDP 3M rolling growth rate Debt to GDP

0,00%

10,00%

20,00%

30,00%

40,00%

50,00%

60,00%

70,00%

80,00%

Economic Sentiment Unemployment Rate Consumer Confidence

77

Table 3.9: Fundamental liquidity variables, multiple regression output

Chart 3.3: R² Spreads determinants per country

Degré de liberté Somme des carrés Moyenne des carrés F Valeur critique de F

Régression 4,00 92,12 23,03 869,13 0,000000%

Résidus 1615,00 42,79 0,03

Total 1619,00 134,92

Coefficients Erreur-type Statistique t (Abs stat T) Probabilité

Constante 0,00 0,00 -0,90 0,90 36,58%

Bid-Ask Spread / Spot Rate 0,00 0,02 0,17 0,17 86,34%

Bid-Ask Spread 0,78 0,02 43,24 43,24 0,00%

Proportion of EMU Public Debt

Outstanding -0,01 0,01 -0,89 0,89 37,36%

Public Debt Outstanding Growth Rate 0,00 0,00 0,59 0,59 55,31%

ANALYSE DE VARIANCE

Statistiques de la régression

Coefficient de détermination multiple

Coefficient de détermination R^2

Coefficient de détermination R^2

Erreur-type

Observations

83%

68%

68%

0,16

1620

0,00%

10,00%

20,00%

30,00%

40,00%

50,00%

60,00%

70,00%

80,00%

90,00%

Proportion of EMU Outstanding Bid-Ask Spread

78

Chart 3.4: Coefficient Spreads determinants per country

Table 3.18: Extra-fundamental variables, multiple regression output

-200,00

-100,00

0,00

100,00

200,00

300,00

400,00

500,00

600,00

Proportion of EMU Outstanding Bid-Ask Spread

1620

Degré de libertéSomme des carrésMoyenne des carrés F Valeur critique de F

Régression 4 93,83 23,46 922,09 0,000000%

Résidus 1615 41,09 0,03

Total 1619 134,92

Coefficients Erreur-type Statistique t (Abs t stat) Probabilité

Constante 0,00 0,00 -0,15 0,15 88,33%

Downgrade Dummy 0,50 0,02 30,02 30,02 0,00%

Political Risk 0,29 0,02 18,87 18,87 0,00%

Upgrade Dummy 0,04 0,01 2,77 2,77 0,57%

Volatility / Contagion 0,33 0,01 21,82 21,82 0,00%

Statistiques de la régression

Observations

ANALYSE DE VARIANCE

Coefficient de détermination multiple

Erreur-type

Coefficient de détermination R^2

Coefficient de détermination R^2 70%

69%

0,16

83%

79

Chart 3.5: R² Spreads determinants per country

Chart 3.6: Coefficient Spreads determinants per country (Ex Contagion)

0,00%

10,00%

20,00%

30,00%

40,00%

50,00%

60,00%

70,00%

80,00%

90,00%

10Y Spot Rate Downgrade Dummy Political Risk Volatility/Contagion

-0,50

0,00

0,50

1,00

1,50

2,00

2,50

10Y Spot Rate Downgrade Dummy Political Risk

80

Chart 4.6: Bank of America - Merrill Lynch Euro Government Index Weights

Chart 4.7: Excess Return Index - Using BofA Merill Lynch Index Weights

AT4%

BE6%

DE18%

ES14%

FI1%

FR24%

IE2%

IT25%

NL6%

94

95

96

97

98

99

100

101

2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014

81

Chart 4.8: Monthly Excess Return Timeframe

-1,00%

-0,80%

-0,60%

-0,40%

-0,20%

0,00%

0,20%

0,40%

0,60%

0,80%

1,00%

2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014

Monthly Excess Return Timeframe

82