Events Calendar

Robust Simulation and Catastrophe Diagnostic for Accounting for Uncertainty in Catastrophe Risk Analysis

Speaker: Dr. Craig Taylor, Director, Research for Baseline Management Company, Inc. and Research Professor, Astani Department of Civil and Environmental Engineering, Viterbi School of Engineering

Abstract:

The speaker has given two previous seminars in the USC CEE Department:

• 2000: On Acceptable Risk Procedures for Ports and
Airports
• 2008: On A Non-Parametric Approach to Evaluating
Catastrophe Risk and Decisions: Financial and
Infrastructure Systems

In these seminars, the speaker outlined catastrophe risk procedures for infrastructure systems and showed how the uncertainties in the models involved could be accounted for in the overall loss distribution. Methods were outlined to manage the uncertainties in the parameters that are explicitly considered within each part of the model (e.g., hazards, response of components, systems response). These are the “nominal” or “endogenous” uncertainties. With the introduction of alternative models, based on different assumptions, parameters, or data, one may begin to account for the remaining “exogenous” uncertainties that lie within the bounds of current knowledge. Of course, no domain of science is or should be ever “settled,” so exogenous elements will persist, contributing to some residual uncertainty.

In past CEE seminars, the speaker described the weakness of one conventional approach that parses uncertainty into “aleatory” and “epistemic” elements. Robust simulation provides an alternative approach to the management of uncertainty in catastrophe risk analysis, as well as overcome severe weaknesses that may occur in the use of logic trees and weighting systems. The speaker will further reiterate briefly weaknesses that can arise through the numerous smoothing techniques that can arise. These have arguably contributed significantly to the collapse of Long-Term Capital Management and to the recent severe recession resulting in part from the egregiously high ratings of mortgage-backed securities containing sub-prime loans.

The previous seminars assumed that the endogenous uncertainties “vanish” as numerous simulations are performed. Thus, topics of “infinite variance” or “infinite mean” were ignored. To a very large extent, a priori modeling of exposures subject to catastrophes may postulate “unstable” distributions (ignoring here for instance alpha-stable distributions). The speaker instead has devised a very simple method for evaluating the “stability” or “danger” of a distribution with a simple one-parameter Pareto. Comparison of the “tail” (99th centile in the severity distribution) of the Pareto distribution with the simulated catastrophe loss distribution can provide a diagnostic (similar to a modified QQ diagnostic) helpful in testing how the degree of “danger” of a catastrophe loss distribution. Typically there will be limits (e.g., limits on the amount of capital at stake, limits on the magnitude of an earthquake, limits on the total loss for a specific property) that will render a catastrophe loss distribution more stable than might be modeled if extreme value distributions are postulated in advance of such considerations.

Robust simulation then begins with a preferred set of models and a test of the “danger” of the loss distribution given an extremely large number of simulations. The “exogenous” uncertainties in the catastrophe loss distribution are illustrated in ongoing research as in missile risk analysis, global climate change, climate conditioning for hurricanes and other severe weather events, and alterative seismicity. Each model is rendered as coherent as possible; mixing models as through “weights” may produce less than coherent results. If weighting is required, as for “official” results, this should be performed at the end of the process. The result of this process yields “bounds of uncertainty.” Unless one imposes a distribution on these outcomes, these uncertainties do not represent confidence intervals.

This procedure is not altogether felicitous, but represents a mature viewpoint. Many models that were once disregarded because they were not good enough for some reason or other now come into play to assist in defining bounds of uncertainty. In the selection of alternative credible models, merits begin to count as well as demerits. Encouraging competition among models is salutary in science and engineering. In this probabilistic realm, selection of one model over another often involves tradeoffs—with pros and cons of various fitting criteria, parameters, assumptions, and the like.





Biography:
Dr. Craig E. Taylor has had over thirty years of experience in catastrophe risk analysis with an emphasis on infrastructure systems, finance, policy, and earthquakes. Currently Director of Research for Baseline Management Company, Inc. and Research Professor at the University of Southern California, he has taught an advanced Civil Engineering course on risk and decision analysis for infrastructure systems. His 200 or so publications and over sixty papers include as contributor and editor four monographs for the American Society of Civil Engineers (ASCE) and major reports on earthquake mitigation for a federal earthquake insurance program, should one be established. Belonging to four professional organizations, and previous chair of the ASCE Council on Disaster Risk Management (CDRM) as well as past chair of several committees, he has received several awards including a lifetime achievement award from the Technical Council on Lifeline Earthquake Engineering (TCLEE). In October 2008 he served as an ASCE representative to look at impacts of the Wenchuan earthquake and to participate at a Tongji University workshop on reconstruction alternatives. In September 2010 he returned to Tongji University to give a plenary presentation at the International Symposium on Reliability Engineering and Risk Management (ISRERM 2010). He earned his doctorate at the University of Illinois.

Host: Prof. Jean-Pierre Bardet

Audiences: Everyone Is Invited

Contact: Evangeline Reyes