When you can’t cut, estimate your losses

Photo by born1945 via Flickr.

Hur­ri­cane Sandy has laid bare the frailty of an urban infra­struc­ture not accus­tomed to large-​​scale nat­ural dis­as­ters. As others have recently explained, cli­mate change mod­eling sug­gests that the fre­quency of this kind of cat­a­strophe will only rise in the coming decades.

These frail­ties have lead to enor­mous and unex­pected finan­cial losses. But what if we could esti­mate the con­se­quences of these events before they strike? The chal­lenge obvi­ously lies in the lack of pre-​​existing data. Hur­ri­canes in the South­east are nothing new, so mod­eling the finan­cial impact in that region isn’t ter­ribly dif­fi­cult. Not so in the Northeast.

Last week, AIR World­wide exec­u­tive Jay Guin gave a guest lec­ture on this topic in civil and envi­ron­mental engi­neering pro­fessor Auroop Ganguly’s grad­uate class, Applied Time Series and Spa­tial Sta­tis­tics. Guin leads the research and devel­op­ment team at AIR, which develops cat­a­strophe mod­eling and ana­lytics tools for the “rein­sur­ance” industry.

Rein­sur­ance, Guin explained, pro­vides insur­ance for insur­ance com­pa­nies. The industry picked up steam after 1992’s Hur­ri­cane Andrew, which led to an unprece­dented $16 bil­lion in losses that insur­ance com­pa­nies could not handle alone. In their bid to develop nat­ural cat­a­strophe or “nat cat” models, rein­sur­ance pro­fes­sionals face sig­nif­i­cant chal­lenges, Guin said. Beyond a scarcity of data, these include a high level of uncer­tainty and sen­si­tivity in loss esti­ma­tion along with the dif­fi­cult task of inte­grating sta­tis­tical and phys­ical mod­eling approaches.

North­eastern PhD can­di­date Evan Kodra (left) and research asso­ciate Rachindra Mawalagedara (right) attended the lec­ture and con­tributed sig­nif­i­cantly to this post. Here they are with Jay Guin, AIR Worldwide’s R&D lead.

AIR runs tens of thou­sands of plau­sible model sim­u­la­tions to under­stand the pos­sible risks to unex­pected areas. For example, they’ll use hur­ri­cane data from 1900 onwards to under­stand and pre­dict pos­sible hur­ri­cane tracks and then deter­mine the poten­tial dam­ages that can happen if a hur­ri­cane fol­lows a par­tic­ular track. Inte­gra­tion of sta­tis­tical and physics based models give a more accu­rate pic­ture for impact assessment.

Research taking place in Ganguly’s Sus­tain­ability and Data Sci­ences Lab com­ple­ments that of the rein­sur­ance industry.  They solve  prob­lems in cli­mate change, nat­ural haz­ards and com­pu­ta­tional data sci­ences. The insights they gain can help com­pa­nies like AIR to solve prob­lems for insur­ance and rein­sur­ance clients dealing with nat­ural haz­ards. At the same time, data about robust­ness, vul­ner­a­bility and expo­sure coming from the rein­sur­ance industry helps Ganguly’s team with its research about extreme weather.

Together, the research field and the rein­sur­ance industry are con­stantly making improve­ments to pre­dic­tion and esti­ma­tion tech­nolo­gies. Hope­fully these tools will one day allow us to steer clear of the finan­cial destruc­tion we’re seeing with Sandy.