Dr Patrick E McSharry
Head of Catastrophe Risk Financing
Centre of Catastrophe Risk Financing
mcsharry AT maths.ox.ac.uk
Tel: +44 1865 614943
My research takes a multidisciplinary approach to modelling complex dynamical
systems. Typical characteristics include non-normal distributions, nonlinear interactions
and spatio-temporal dynamics. This research has practical predictive applications in
quantifying risk and uncertainty for effective decision-making.
The general aim is to construct parsimonious data-driven models from
Information about the underlying dynamics, such as existing first-principle models or
conservation laws, may be imposed to constrain the model class.
This approach uses a diverse range of techniques from areas including time series analysis,
signal processing, forecasting, operations research, data-mining and machine learning.
Observations are invariably contaminated by measurement errors,
leading to uncertainty in the data and structural errors in all models derived
from this data.
I am developing an approach that propagates this uncertainty through all the
different levels of the modelling process in order to provide the end-user
with the maximum amount of information available for decision-making.
In particular, I have developed
techniques for the construction and evaluation of probabilistic forecasts.
Probabilistic forecasts are required for understanding the inherent
uncertainty in the modelling process and have the additional
value of providing estimates of the implied uncertainty.
Furthermore, such forecasts are essential for risk management and the
maximisation of utility functions.
Models and Techniques
The model structures employed are linear stochastic models such as ARIMA
models and nonlinear models such as local polynomials, radial basis
functions, neural networks, threshold auto-regressive (TAR) models
and Markov switching models.
Techniques used for classification, clustering and visualisation include k-means,
self-organising maps, support vector machines and genetic algorithms.
Applications of this research include forecasting, classification,
signal processing, fault detection and the analysis of biomedical,
economic, energy and financial time series. Examples include:
(i) risk management in healthcare and energy sectors;
(ii) diagnosing, classifying and monitoring states of health/illness using non-invasive
(iii) biomedical engineering (prediction and classification of epilepsy and
(iv) telemedicine (influence of weather on asthma and self-management of diabetes);
(v) decision science (financial trading strategies, weather
forecasting, risk management, forecasting hospital bed availability);
(vi) energy sector (forecasting demand, price dynamics);
(vii) systems biology (gene networks, biochemical reaction networks);
(viii) speech modelling (vocal fold dynamics, speech processing, patent pending);
(ix) fault detection and condition-monitoring (aircraft failure, reactor disruptions,
nucleate boiling, medical disorders) and
(x) social science (policy optimisation for the
World Bank and the analysis of human development indicators).
More details about techniques and applications are available from:
Systems Analysis, Modelling and
| OCIAM |