Monday 14 October 2024, 12:00-13:00
UNIH-205 University House
We find that good news extracted by ChatGPT from the front pages of Wall Street Journal can predict the stock market and is related to macroeconomic conditions. Consistent with existing theories, investors tend to underreact to positive news, especially during periods of economic downturns, high information uncertainty and high novelty of news. In a related problem, we propose a supervised learning method to find the optimal predictive factor, which is the best linear combination of a large set of predictors.
Our approach outperforms not only both the PCA and PLS, but also all those commonly used state of are machine learning methods. Furthermore, given a large set of factors or strategy returns, we provide a general approach to select factors that achieve the mean-variance efficiency, yielding the pricing kernel.
Monday 21 October 2024, 16:00-17:00
Watson Building, B16
Overdamped Langevin dynamics are stochastic differential equations, where gradient dynamics are perturbed by noise in order to sample high-dimensional probability measures such as the ones appearing in computational statistical physics and Bayesian inference. By varying the diffusion coefficient, there are in fact infinitely many overdamped Langevin dynamics which preserve the target probability measure at hand. This suggests to optimise the diffusion coefficient in order to increase the convergence rate of the dynamics, as measured by the spectral gap of the generator associated with the stochastic differential equation. We analytically study this problem here, obtaining in particular necessary conditions on the optimal diffusion coefficient. We also derive an explicit expression of the optimal diffusion in some homogenised limit. Numerical results, both on discretisations of the spectral gap problem and Monte Carlo simulations of the stochastic dynamics, demonstrate the increased quality of the sampling arising from an appropriate choice of the diffusion coefficient.
Wednesday 13 November 2024, 12:00-13:00
University House G07
Traffic emissions are one of the most important sources of urban air pollution. One widely adopted strategy for mitigating urban air pollution is traffic management through Low Emission Zones (LEZs). As many other UK cities are either considering or already implementing similar policies, a rigorous evaluation of its effectiveness is crucial. Using Birmingham's Clean Air zone and London's Ultra Low Emission Zone as case studies, we propose a two-step data driven approach to assess the causal impact of these interventions on local air quality. Our goal is to develop a quantitative, user-friendly tool for evaluating the effects of clean air policies on air pollution levels based on observational data.
Thursday 28 November 2024, 15:00-16:00
Watson R17/18
We propose a deep learning method for solving the American options model with a free boundary feature, leveraging the Landau transformation to extract the early exercise boundary. Our approach employs an implicit dual solution framework, incorporating a novel auxiliary function and free boundary equations. The auxiliary function, utilising a feed-forward deep neural network (DNN) output, mimics far boundary behaviour and satisfies key conditions. As the early exercise boundary is not known a priori, we approximate it and its derivative from the DNN output. Option Greeks are derived from the auxiliary function's derivatives. Comparative testing with existing methods demonstrates the efficiency of our approach in pricing options with early exercise features.
Thursday 5 December 2024, 14:00-15:00
Watson Building, LTA
We estimate the country-level risk of extreme wildfires defined by burned area (BA) for Mediterranean Europe and carry out a cross-country comparison. To this end, we avail of the European Forest Fire Information System (EFFIS) geospatial data from 2006 to 2019 to perform an extreme value analysis. More specifically, we apply a point process characterisation of wildfire extremes using maximum likelihood estimation. By modelling covariates, we also evaluate potential trends and correlations with commonly known factors that drive or affect wildfire occurrence, such as the Fire Weather Index as a proxy for meteorological conditions, population density, land cover type, and seasonality. We find that the highest risk of extreme wildfires is in Portugal (PT), followed by Greece (GR), Spain (ES), and Italy (IT) with a 10-year BA return level of 50,338 ha, 33,242 ha, 25,165 ha, and 8,966 ha, respectively. Coupling our results with existing estimates of the monetary impact of large wildfires suggests expected losses of €162–439 million (PT), €81–219 million (ES), €41–290 million (GR), and €18–78 million (IT) for such 10-year return period events.