Práce porovnává základní přístupy a možnosti v řízení portfolií aktiv institucionálními investory, přičemž se zaměřuje zejména na penzijní fondy a pojišťovny. V jednotlivých kapitolách se nejprve věnuje omezením, na něž portfolio manažeři naráží při umisťování prostředků uvedených finančních institucí. Dále přibližuje třídy aktiv, tradiční i alternativní, do nichž je možné investovat volné prostředky, s nimi spojená rizika a nástroje pro jejich měření. Podrobně rozebrány jsou základní přístupy pro správu portfolia: Benchmarking, ALM a Risk budgeting. V závěru jsou uvedeny nejpoužívanější investiční strategie pro řízení části portfolia s dodatečnými požadavky na riziko nebo výnos.

L’objectif de cette thèse est de proposer des schémas de pondérations alternatives visant à prendre en compte le risque dans la construction d’indices obligataires. Nous partons du constat suivant : les indices obligataires qui existent sur le marché sont pondérés en fonction de la capitalisation des émetteurs. L’implication n’est pas négligeable, dans la mesure où utiliser cette approche implique de sur-pondérer les entités les plus endettées. Sur cette base, nous proposons dans le premier chapitre de pondérer les entreprises au sein de l’indice en fonction de leur solvabilité. Dans le deuxième chapitre, toujours sur l’univers des obligations d’entreprises, nous appliquons le principe du risque en parité. Plus précisément, les secteurs sont pondérés de façon inversement proportionnelle à une mesure du risque de crédit innovante : la Duration Times Spread. Enfin, le dernier chapitre s’intéresse à l’application de cette même technique du risque en parité, mais cette fois-ci à l’univers des obligations souveraines. Nous nous engageons dans la modélisation d’une structure de taux à terme, permettant de mesurer le risque de taux d’intérêt dans un contexte global. Plus généralement, nous démontrons que ces pondérations alternatives, qui intègrent une notion de risque (crédit ou taux) et s’éloignent ainsi du pur aspect “niveau d’endettement”, fournissent une nouvelle grille de lecture pour la compréhension de la dynamique des marchés obligataires ainsi que des améliorations significatives dans le profil rendement-risque. / In this thesis, we are keen to explore alternative weighting schemes that account for risk in the fixed-income indexing market. We start with the following observation: bond indexes that exist on the market are generally cap-weighted. The implication is not trivial: when holding such index, an investor is exposed to the most indebted issuers. From that standpoint, in the first chapter we make the proposal to consider an issuer’s creditworthiness as a weighting metric. Then in the second chapter, still working on the corporate bond market, we decide to turn to risk-parity indexing. More precisely, sectors are weighted inversely proportional to an innovative credit risk measure. Finally, the third chapter is devoted to the transposition of such risk-based philosophy to the sovereign bond universe. Particularly, we examine term structure modeling to appraise interest rate risk in a global framework. On a more general note, we show that these alternative indexing schemes - that do not emanate from pure indebtedness, but that are rather based on more sensible definitions of risk (credit or interest rate) provide a new reading grid for understanding bond market’s dynamics as well as appealing improvements in the indexes’ risk-return profile.

Cette thèse porte sur la modélisation, la prévision et la couverture du risque de contagion financière. Après une présentation générale des fondements théoriques et des mécanismes de propagation relatifs à la contagion financière, nous introduisons une modélisation fondée sur les modèles de cointégration non linéaire et de causalité non linéaire dans lesquels, les variables et le terme d’erreur du modèle à correction d’erreur obéissent à la dynamique de processus auto-régressifs à changement de régime de type TAR et M-TAR pour capter l’effet de contagion. Une extension de cette modélisation au cadre de prévision probabiliste conditionnelle a été faite par la suite à travers les réseaux de croyance Bayésienne pour renforcer le pouvoir prédictif. Ensuite, nous montrons comment une institution financière peut couvrir son portefeuille contre ce type de risque par de nouvelles approches. Nous proposons pour cela, une stratégie de couverture purement statique dans une perspective règlementaire à l’aide de modèles génératifs de type Vines-copula, une stratégie de couverture semi-statique fondée sur la budgétisation des risques et une stratégie de couverture dynamique à partir des processus de diffusion à sauts mutualisés. Ces nouvelles modélisations sont testées empiriquement sur un ensemble d’indices boursiers. / This Ph.D thesis focuses on modeling, forecasting and hedging financial contagion. After an overview of the theoretical foundations and spread mechanism relating to financial contagion, we introduce modeling based on nonlinear cointegration and non-linear causality models in which the variables and the error term in the correction model error obey at the dynamics of autoregressive regime change process of type TAR and M-TAR to catch the contagion effect. An extension of this model to conditional probabilistic forecasting framework was done through Bayesian belief networks, to enhance the predictive power. Then we show how a financial institution can hedge its portfolio against this risk by new specifications. Therefore, we offer a purely static hedging strategy in a regulatory perspective using generative models Vines-copula, a semi-static hedging strategy based on risk budgeting and dynamic hedging strategy based on mutually exciting jumps diffusion process. These new models are tested empirically on set of market indices.

Risk and tracking error budgeting was originally adopted by large institutional investors, including pension funds, plan sponsors, foundations, and endowments. More recently, risk and tracking error budgeting have gained popularity among financial advisors, multi-managers, fund of funds managers, high net worth individuals as well as retail investors. These techniques contribute to the portfolio optimisation process by limiting the extent to which a portfolio can deviate from its benchmark with regards to risk and tracking error. This is an ambitious paper that attempts to determine the optimal strategy to practically implement risk and tracking error budgeting as a portfolio optimisation technique in South Africa. This study attempts to bridge the gap between active, passive, and smart-beta investment management styles by introducing a low-cost portfolio construction technique, for core-satellite portfolio management, which contributes to the risk and tracking error budgeting process. Core-satellite portfolios are designed to expose the portfolio to a low-cost primary “core” consisting of passive and enhanced index funds, thus systematic risk “beta”, limiting the tracking error of the portfolio. The secondary “satellite” component is allocated to active and smart-beta managers to exploit expected excess return “alpha”. The primary aim of this research is to construct a rule-based product range of core-satellite portfolios called “replica portfolios”. The product range builds on the foundation of the Association for Savings & Investments South Africa (ASISA) framework. The study identifies three “target portfolios” from ASISA's framework, namely (1) High Risk: SA General Equity, (2) Medium Risk: SA Multi-Asset High Equity and (3) Low Risk: SA Multi-Asset Low Equity. Through this framework, active managers from each category are shortlisted using a Sharpe and Information Ratio filter. A secondary filtering technique, namely Returns Based Style Analysis (RBSA) is used to determine the style, R-squared and alpha-generating ability of active managers versus the passive asset classes and style indices they seek to replicate. Applying Euler's theorem for homogenous functions, we decompose the risk of the coresatellite portfolio into the risk contributed by each of its components. The primary mandate of the core-satellite portfolios in the product range is to allocate risk and tracking error efficiently across several investment management styles and asset classes in order to maximise returns while remaining within the specified risk parameters. iii The results highlighted that active managers, after fees, predominantly failed to outperform their benchmarks and passive building blocks, as identified through RBSA over the sample period (October 2009 – September 2019). However, only a small number of active managers generated superior risk-adjusted returns and were included in the core-satellite range of products. This study recommends to investors that they exploit the “hot-hands effect” by investing in specialised, benchmark agnostic active managers who consistently produce superior risk-adjusted returns. By blending active, passive and smart-beta strategies, investors are exposed to less total risk, less risk per holding and a lower tracking error. The three coresatellite portfolios developed in this study generated absolute and risk-adjusted returns that are more significant than their active and passive counterparts. Fee arbitrage was derived through the range of core-satellite products, resulting in tangible alpha over the sample period. The study encourages investors to use smart-beta strategies alongside active and passive funds since it improves Sharpe and Information ratios while enhancing the original portfolio's characteristics.

Submitted by Yugo Watari (ywatari@gmail.com) on 2017-09-19T16:23:48Z No. of bitstreams: 1 main.pdf: 2611498 bytes, checksum: 1f50a4c20e7433334a4e2b45acd23424 (MD5) / Approved for entry into archive by Thais Oliveira (thais.oliveira@fgv.br) on 2017-09-19T16:31:51Z (GMT) No. of bitstreams: 1 main.pdf: 2611498 bytes, checksum: 1f50a4c20e7433334a4e2b45acd23424 (MD5) / Made available in DSpace on 2017-09-19T17:31:18Z (GMT). No. of bitstreams: 1 main.pdf: 2611498 bytes, checksum: 1f50a4c20e7433334a4e2b45acd23424 (MD5) Previous issue date: 2017-08-21 / We approach portfolio construction with risk based allocation, using volatility as the measure of risk, and applying to the stock markets. We start by obtaining generic risk factors based on the approach of Fama&French; and them we decompose the volatility in risk contributions of those generic risk factors. Differing from previous works, instead of allocating in indexes that represent the generic risk factors, we allocate at the asset level, in hopes that this will lead to reproducing the effects of inveting on those indexes, which brings additional complexity to the problem. This was motivated by investors not always having access to invest in theses indexes. Finally, for the purpose of illustration, we apply the metodology to the brazilian stock markets, selecting as risk factors, the five Fama&French risk factors. We obtain portfolios with the desired risk contributions, but as we look in to the weights of each risk factor, there is alocations of weights in the risk factors not related to those of Fama&French, even though the risk contributions are neutralized. We argue that these allocations are preventing from obtaining exposures to the distinct characteristics of each Fama&French risk factor. / A construção de portfólios, ou seja, a definição da composição de uma carteira de ativos, é abordada, nesse trabalho, pela ótica da alocação baseada em contribuições do risco, medida via volatilidade, aplicada a uma carteira de ações. O objetivo é a construção de portfolios, via as contribuições de riscos; para isto construímos fatores de riscos genéricos baseados na abordagem de Fama&French; na sequência aplicamos uma metodologia para distribuir a volatilidade como contribuições de risco destes fatores genéricos. Diferentemente de outros trabalhos, ao invés de alocar em índices que representem estes fatores de riscos genéricos, alocamos diretamente nos ativos na expectativa de conseguirmos reproduzir o efeito de investir nestes índices, o que traz uma complexidade adicional. Esta abordagem foi motivada por nem sempre termos acesso à investir nesses índices. Finalmente, a título de ilustração, a metodologia foi aplicada ao mercado brasileiro de ações, em particular utilizando os fatores do modelo Fama&French de 5 fatores. Obtivemos portfolios com as contribuições de riscos desejadas em relação aos fatores de Fama&French, mas ao se analisar a alocação dos pesos dos fatores de riscos sobre os portfolios obtidos, verificamos que são alocados pesos a fatores que não estão relacionados aos de Fama&French, apesar das contribuições de risco destas estarem neutralizadas. E por fim argumentamos que estas alocações evitam a captura das características distintas de cada fator que gostaríamos de reproduzir.