Past PhD theses

Bears and bulls are deeply ingrained symbols in financial culture, representing market sentiment and trends. Meanwhile, monkeys have become emblematic of chance and randomness within both financial and statistical contexts. In finance, the metaphorical image of monkeys throwing darts reflects the stochastic process of sampling -- a central theme of this dissertation.

The thesis comprises four self-contained chapters focused on computational tools, which are examined both as instruments for addressing inquiries in finance, economics, and statistics, and as subjects of study in their own right. The overarching theme across all chapters is a geometric perspective, which is relevant to both the modeling of economic questions and the analysis of algorithmic sampling procedures.

Specifically, the work examines geometric random walk algorithms, a powerful class of Markov chain Monte Carlo procedures designed to sample from high-dimensional and truncated distributions. These methods facilitate the creation of portfolios with particular characteristics, such as targeted variance levels or strict compliance with investment guidelines, thereby enabling the investigation of empirical asset pricing anomalies under constraints.

The geometric perspective offers analytical insights into one of the most renowned sampling-based methods for statistical inference: the bootstrap. The corresponding chapter provides exact solutions that eliminate randomization error by establishing a link between resampling plans and the distribution of random weights and by drawing upon concentration phenomena from high-dimensional geometry.

Finally, a distance-based measure of financial turbulence is proposed, capable of distinguishing between beneficial and detrimental turbulence episodes and proving useful for dynamic hedging.

This thesis consists of three applications of contemporary high-performance computing to accelerate large-scale dynamic models in economics and finance. The first chapter is entitled “Scalable high-dimensional dynamic stochastic economic modeling” and presents a highly parallelizable and flexible computational method to solve high-dimensional stochastic dynamic economic models. By exploiting the generic iterative structure of this broad class of economic problems, we propose a parallelization scheme that favors hybrid massively parallel computer architectures. Numerical experiments at the Swiss National Supercomputing Centre show that high-dimensional international real business cycle models can be efficiently solved in parallel up to 2,048 compute nodes. The second chapter is called “Rethinking large-scale economic modeling for efficiency: optimizations for GPU and Xeon Phi clusters” and proposes a massively parallelized and optimized framework to solve high-dimensional dynamic stochastic economic models on modern GPU- and KNL-based clusters. Numerical experiments show that our framework scales to at least 4,096 compute nodes. The third chapter, titled “GPU-Accelerated Dynamic Human Capital Models” develops a generic computational method for dynamic discrete-choice models. We align the generic numerical properties of the models under consideration with the recent advancements in GPU computing hardware in order to solve, simulate, and calibrate models of great complexity in relatively short times. Our tests show a speedup of at least three orders of magnitude over the previous state of the art.

Today there is significant pressure to transition electricity systems from fossil fuel and/or nuclear generation towards renewables. This raises socio-economic, political, and technical questions. Few countries have developed 100% renewable generation systems; exceptions being Ireland, Norway, and Paraguay. The main challenge is the need to match demand and supply at all times, in the presence of intermittent sources (e.g., wind, solar), which implies a need to store excess generation.
In this research we develop a stylized simulation model to analyze the feasibility and the long-term consequences of the transition of electricity systems towards renewables, using Switzerland as a study case. Given this country’s geographical characteristics, our model considers hydropower and PV generation, and pumped hydro-storage. This model allows us to include the possible effects of climate change on demand and generation capacity.
Our overall conclusion is that a 100% renewable generation system is technically feasible; shortages do occur at some point under all climate scenarios. These can be eliminated by subsidizing PV, which indirectly incentivizes storage. We also test a demand side management program; while useful to reduce unmet demand and mitigate the increase of the electricity price, the impact is limited. Our analysis provides useful insights for policymakers regarding managing the transition towards renewables. Though calibrated for Switzerland, our model can be adapted to other regions, with different energy policies and technologies.

This thesis develops new methodology in the statistical analysis of event data, and extreme events in particular. The first contribution is a testing method for rigorous comparison between risks of different natures, such as classes of insurance policies. As a second contribution, advanced methods for event data analysis are proposed, drawing on classical results in stationary time series. The issue of event prediction based on past history is treated in detail. All the proposed theoretical tools are evaluated by two applications: the first evaluates meteorological and climatological risk in Lausanne via our proposed test; the second studies factors responsible for overcrowding in a hospital emergency department, by means of a flexible patient flow model. Finally, avenues for further research are briefly discussed.

Logistics (or supply chain management) refer to the process of moving goods, materials, and people within a company. The optimization of such processes enables firms to gain competitiveness, and constant innovative methodological approaches must be unveiled to sustain this potential for improvement. Currently, to be optimized, the supply chain is divided into several echelons (or levels) (e.g., production, distribution, storage, etc.), each of them being optimized individually and independently. As a result, the obtained solutions for one echelon become then the input data for the next echelon, which leads globally to sub-optimality. In that regard, synchronization (i.e., aiming at a harmonious collaboration of various actors at different levels of the supply chain) appears to be a promising avenue for improvement. In this thesis, three practical cases, inspired by different industrial partners, are analyzed. The considered situations range from the transportation of raw materials from suppliers to production plants to home parcel delivery. In each of these cases, we quantify the potential offered by adopting a methodological approach based on synchronization and we compare the obtained solutions to current industrial practices. Results show that aspects related to costs, labor conditions, or even environmental impacts of the supply chain can be significantly improved by obtaining and introducing synchronized solutions in the field of logistics. In this respect, this thesis focuses also on building the cutting-edge algorithms that are necessary to solve such resulting problems and to obtain such synchronized solutions.

While electricity markets have evolved constantly since their restructuring in the late 1980s, their regulation does not always progress at the same pace, creating a mismatch which results in market failures. Current climate change mitigation efforts and the pace of technological change exacerbate this problem. This thesis tackles the fit of regulation to its market in the context of electricity. The first paper identifies four stages in the evolution of electricity markets: monopoly, wholesale competition, retail competition and reregulation. We then discuss experiences from different jurisdictions and problems arising from inappropriate regulation at each stage. Based on this discussion, we propose a behavioural regulatory framework which results in a more agile, forward looking regulation, requiring less frequent adaptations. We then focus on the last stage of market evolution, reregulation, to analyse the transition towards an increasing green share of generation. Using a system dynamics model, we explore the effect of roadmaps (a set of intermediary objectives) on the achievement of long-term environmental targets and electricity market performance. We first optimize regulatory decisions, and then compare these findings with the results from laboratory experiments. We find that the presence and time granularity of roadmaps translate into different subsidy decisions and costs for customers.  From a policy point of view, this thesis illustrates the importance of providing roadmaps together with a long-term target, as these roadmaps shape the path of the markets towards the final point.