The research work presented in my thesis has been articulated around the goal of building financial crisis indicators with the ability to forecast as accurately as possible future market events and then use that information to devise systematic trading strategies. Those financial crisis indicators work from multiple points of view that complement one another. They rely on the correlation and the volatility inside a basked of asset or the components of an equity index, like the SP500 and, in one part of the thesis, we also develop indicators based on the distribution of the spreads of the components of a CDS index, like the Itraxx Europe 125. Those financial crisis indicators are then applied to many different datasets, large and small, and the signal that they provide is used as the basis for the construction of an active trading signal.

This thesis is made of three research papers constituting its three chapters. Each of those three papers is, at the date of my defence, about to be independently submitted or already undergoing review at a peer reviewed publication, albeit sometimes in a shortened form.

The first chapter deals with the construction of two kinds of financial crisis indicators. The first kind of financial crisis indicators is based on the comparison of the empirical spectrum of a rolling covariance matrix to a distribution of reference that may represent either a calm or an agitated market reference. The second kind of financial crisis indicators is based on the computation of the trace or of the spectral radius of the covariance matrix, the correlation matrix or a weighted version of the correlation matrix. The weights that we use in this first chapter are the market capitalization and the volume traded. After defining a total of nine financial crisis indicators, of both kinds, we then proceed to demonstrate out-of-sample predictive power for one of them, which we choose to be the spectral radius of the correlation matrix weighted by volume traded applied on our best and most detailed dataset that contains the SP500 index and its stock components. The most interesting aspect of the demonstration of the prediction power of our financial crisis indicator is the implementation of a successful protective put systematic trading strategy based on its signal. While the worth of our approach is demonstrated and the prediction power of our financial crisis indicators clearly established, we also underline the limitations of our approach, which in particular may take the form of a significant number of false positive errors in the signal provided by our financial crisis indicators.

The second chapter is constituted of a paper that builds upon the framework and financial crisis indicators constructed in the first chapter. In this second paper, we expands the use of our financial crisis indicators by combining the signals provided by 29 of them and create a decision process designed to govern a portfolio constituted of a mix of cash and ETF shares. Since the main limitation of our financial crisis indicators, while considered individually, is the presence of false positives in their predictions, we aim in this paper at combining the signals provided by many of them and make a systematic trading strategy act on the composition of the portfolio, selling the shares when the risk of a crisis is high and converting the cash into shares when the risk of a crisis is low, only when the indicators reach some kind of consensus in their forecasts. We then apply this approach to five datasets containing the stock components of five major equity indices. The success of the systematic trading strategies based on our financial crisis indicators is demonstrated by comparing their performances to a buy and hold strategy as well as to a large number of paths of a strategy where the choices to convert the cash into shares or the shares into cash is random. The main result in this chapter is the validation of our framework and the demonstration of the usefulness and prediction power of our financial crisis indicators through the production of winning investment strategies based on those financial crisis indicators. The added value of our systematic active strategies, both in comparison to the static buy and hold references and to the random paths is clear in terms of Sharpe ratio, reduced volatility, increased overall performance and Calmar ratio.

The third and final chapter talks about another and novel approach to financial crisis indicators, this time by using the dynamic evolution of the distribution of the spreads of the components of a CDS index, like the Itraxx Europe 125. After establishing some results that allow us to work with dynamic distributions on solid theoretical ground, we fit the empirical distribution of the spreads of the components of the index with a mixture of two lognormal distributions. From the study of the dynamics of the coefficients of the decomposition of the empirical distribution of the spreads on the basis constituted of the two chosen lognormal distributions, we then build a lower and an upper boundary around the fitted empirical cumulative distribution function of the spreads of the components of the CDS index. This approach defines a Bollinger band around the fitted empirical cumulative distribution function and the crossing of either boundary defining by that band is interpreted in terms of risk and therefore translated into a trading signal. While the establishment of a complete and fully functional active trading strategy using that Bollinger band upper and lower boundary crossing signal is going to be presented in a mature form only in future revisions of this work, the results obtained are still attractive enough to be considered by the asset management industry, to which we believe this work can be extremely useful in order to navigate through a globally uncertain environment.