Академический Документы
Профессиональный Документы
Культура Документы
net/publication/257875992
CITATIONS READS
12 402
2 authors:
Some of the authors of this publication are also working on these related projects:
CRUST: Cascading Risk and Uncertainty assessment of earthquake Shaking and Tsunami View project
Force- and Performance-Based Seismic Design Guidelines for Timber-Steel Hybrid Structures View project
All content following this page was uploaded by Katsuichiro Goda on 06 December 2018.
Abstract: The role of financial seismic risk analysis and risk quantification has grown rapidly in dealing with important risk-management
decisions for building portfolios. Recent developments in regional seismicity modeling, ground-motion modeling of spatially correlated seis-
mic effects, and seismic fragility analysis have enabled more accurate estimation of earthquake risk exposure of building portfolios. The result
from quantitative seismic loss estimation can be represented by a seismic loss curve and various scalar risk metrics, such as annual expected loss,
Downloaded from ascelibrary.org by University of Bristol on 04/17/14. Copyright ASCE. For personal use only; all rights reserved.
value at risk, and tail value at risk, which are all expressed in monetary terms. Their use facilitates the risk-based decision making and risk
communication. This study illustrates the current risk-quantification approaches through a case study for a group of wood-frame houses in Can-
ada. Potential pitfalls in using simple risk metrics for decision analysis on seismic risk-transfer and risk-mitigation strategies are discussed. The
results indicate that the use of tail value at risk may be appropriate as it provides more consistent risk-comparison results by reflecting the
expected risk of rare events. DOI: 10.1061/(ASCE)NH.1527-6996.0000129. © 2014 American Society of Civil Engineers.
Author keywords: Earthquake engineering; Risk management; Seismic analysis; Wood structures.
relation of seismic excitations) to identify potential pitfalls in using seismological evidence (McGuire 2004). A seismic hazard model,
simple risk metrics for decision making, and (2) discuss usefulness consisting of seismic source zones and associated magnitude-
and limitation of seismic risk curves and risk metrics for asset man- recurrence relationships (Step 1 in Fig. 1), provides a convenient
agement under catastrophic earthquake risk. Finally, conclusions and mathematical representation of regional seismicity in space and time.
future challenges of financial seismic risk analysis are highlighted. A direct and comprehensive way to identify significant EQS is Monte
An idealistic view that various components of risk quantification Carlo sampling from an adopted seismic hazard model (Atkinson and
for rare earthquake disasters are possible is adopted in this study for Goda 2011), producing a synthetic earthquake catalog (list of oc-
practical purposes. Numerous assumptions and simplification are currence time, location, magnitude, source dimension, and so forth).
made in the assessment, leading to considerable uncertainty asso- This step is a critical component in quantifying seismic risk; usually,
ciated with risk estimates. Therefore, one should be cautious when estimation of occurrence of future major earthquakes is highly un-
results from risk assessment are used for making decisions related to certain, and careful consideration of adopting simplified models is
seismic risk mitigation. Nevertheless, one should strive to enhance required (e.g., time-invariant versus time-dependent models and upper
seismic risk modeling capability further, because such a proactive magnitude truncation).
approach will advance our understanding of highly complex and One of the key requirements for earthquake risk assessment of
uncertain risk processes and, eventually, improve risk-management building portfolios is the consideration of spatially correlated seis-
skills. mic IM (e.g., spectral acceleration) at multiple building sites. This is
beyond the capability of conventional ground-motion prediction
equations, because they have been developed for a single site. Using
PSRA appropriate spatial correlation models of ground-motion parameters
(Goda and Hong 2008; Goda and Atkinson 2009), simultaneous
A PSRA methodology for evaluating a seismic risk curve for building seismic intensities of buildings with different fundamental vibration
portfolios, which quantifies an adopted measure of seismic conse- periods at different locations can be evaluated for a given EQS.
quences as a function of annual occurrence probability, is described Steps 1 and 2 in Fig. 1 generate probabilistic estimates of IM at
herein. An extension of the conventional PSRA for a single structure multiple sites for many seismic events over a certain period (i.e.,
to that for multiple structures is emphasized. Following the analysis multiple seismic intensity maps).
framework, several scalar risk metrics, which can be derived from Nonlinear dynamic analysis of structural models subjected to a
a seismic risk curve, are defined, and decision-analysis techniques set of ground motion records facilitates the probabilistic character-
based on stochastic dominance (Levy 2006) are introduced. These ization of inelastic structural response (EDP, e.g., maximum inter-
approaches are implemented subsequently in a case study for wood- story drift) for a given seismic hazard level (IM). Various numerical
frame houses in Canada. methods have been developed. One of the most computationally
intensive approaches is incremental dynamic analysis (IDA;
Vamvatsikos and Cornell 2004). In IDA, samples of EDP and IM are
Seismic Risk Analysis of Multiple Buildings
generated by conducting numerous nonlinear dynamic analyses with
PSRA is an essential part of quantitative assessment of earthquake records scaled at various IM levels, and are used to develop a
risk impact on built environments. It can be formulated in various probabilistic relationship between EDP and IM (Step 3 in Fig. 1). A
forms, depending on the scope of assessment, structures, available careful record selection is important to avoid bias induced by ex-
data, and applicable models (Moehle and Deierlein 2004; Porter cessive record scaling (Luco and Bazzurro 2007). For this purpose,
et al. 2006; Goulet et al. 2007). An initial step, prior to PSRA cal- conditional mean spectrum can be used as target response spectrum
culations, involves the collection of building portfolio information by taking into account detailed characteristics of significant EQS
(e.g., location, structural/material type, year built, occupancy, and site and their response spectral shapes (Baker 2011; Goda and Atkinson
condition). A standard computational method involves several steps 2011).
as follows: (1) generation of significant earthquake scenarios (EQS) Structural vulnerability models quantify the extent of structural
based on stochastic regional seismicity models; (2) estimation of damage (DF) for a given inelastic structural response (EDP), and
seismic intensity measure (IM) at multiple sites; (3) estimation of have significant impact on seismic loss estimation. Damage analysis
engineering demand parameter (EDP) given IM via applicable is carried out by comparing sustained inelastic seismic demand to
seismic demand prediction models; (4) evaluation of damage factor uncertain structural capacity (Step 4 in Fig. 1). In displacement-
(DF) by comparing seismic demand and seismic capacity; and based damage assessment, structural capacity typically is specified
(5) estimation of economic loss (EL) via damage-loss and cost models. in terms of yield and ultimate displacements, and these variables are
This framework is shown in Fig. 1. It is noted that the framework is associated with specific damage states (e.g., slight, extensive, and
generic and alternate procedures can be developed. For example, in collapse). A simple form of displacement-based DF can be defined
struction cost but much reduced seismic damage cost. As three risk
VS30 5 250 m=s, i.e., National Earthquake Hazards Reduction Pro-
curves shown in Fig. 3(a) intersect one another, none of these options
gram site class D).
is preferred according to FSD (i.e., there are some pros and cons for
Four typical classes of wooden houses in Canada have been
decision makers with u9ðxÞ $ 0). Fig. 3(b) shows the integral of
identified by White and Ventura (2006), i.e., Houses 1 to 4. The 500
cumulative distribution functions for the three options. In this figure,
buildings belong to the same house model class. It is considered that
the option with high resistance is identified as optimum because it
these four house types are prevalent across Canada (Paultre et al.
lies below the other two options for all values of x. In other words, all
1993; Ventura et al. 2005). House 1 has blocked plywood/oriented
risk-averse decision makers prefer the high-resistance solution.
strand board (OSB) shear walls with exterior stucco cladding and
gypsum wallboard (GWB) interior finish; House 2 has blocked
Case Study: Financial Seismic Risk Analysis of plywood/OSB shear walls with GWB interior finish; House 3 has
Wood-Frame Houses unblocked plywood/OSB shear walls with GWB interior finish; and
House 4 has horizontal boards with GWB interior finish. In terms of
This case study presents financial seismic risk analysis of wood- seismic capacity, House 1 has superior seismic resistance with
frame houses in Canada. Four locations are considered for the higher stiffness/strength and ductility capacity; Houses 2 and 3 have
building portfolio: Vancouver, Victoria, Montreal, and Ottawa similar seismic resistance (but House 2 has higher capacity than
(i.e., major population centers). The main objectives are to in- House 3); and House 4 has low seismic resistance in terms of both
vestigate the effects of different seismic hazard information and stiffness/strength and ductility capacity. The seismic capacities of
spatial correlation models on seismic risk curves and risk metrics, the four house models are represented by the yield and ultimate
and to assess their usefulness and limitations for earthquake risk interstory drift ratios (both parameters are modeled by the lognormal
management. Moreover, a practical situation where a building distribution); to account for uncertainty of these parameters for
portfolio manager has several risk-mitigation options, such as different house models, their mean and coefficient of variation are
seismic retrofitting and purchasing earthquake insurance, are varied based on the assumed statistical information of these con-
discussed. struction types (Goda et al. 2011). The damage-loss functions and
where CR 5 total retrofitting cost; and the interest rate g and duration Fig. 4. Comparison of seismic loss curves for House 4 in Vancouver
T are set to 0.05 and 30 years, respectively. by considering three spatial correlation models
Earthquake insurance can be useful in transferring seismic risk
to an insurer, who has a higher financial risk-bearing capacity. When
purchase of earthquake insurance is considered, deductible and limit Table 1. Summary of Risk Metrics for House 4 in Vancouver by
(or cap) are set to 0.1 and 0.5, respectively (i.e., seismic loss up to Considering Three Spatial Correlation Cases
10% of the total insured value is incurred entirely by an owner,
seismic loss between 10 and 50% of the total insured value is cov- Correlation Annual rate AEL VAR0:99 VAR0:999 TVAR0:99 TVAR0:999
ered by an insurer, and, if the seismic loss exceeds 50% of the total Partial 0.078 249 6,072 41,793 20,079 66,982
insured value, the excess loss is borne by the owner). This setup is No 0.138 247 6,614 29,666 16,116 42,351
typical for earthquake insurance. In exchange for earthquake risk Full 0.050 250 6,149 43,680 21,628 75,146
coverage, an insurer collects insurance premium, consisting of Note: Unit 5 CAD$1,000.
pure premium and risk premium (including transaction costs). The
pure premium is equivalent to the expected payout for earthquake
loss coverage [EðLI Þ], whereas the risk premium is the additional spatial correlation modeling for seismic loss estimation of building
cost, demanded by an insurer to undertake low-probability high- portfolios. An important issue to be pointed out, related to decision
consequence risks. In this case study, the total insurance premium making, is that when seismic loss curves intersect one another, AEL
PI is calculated as and VAR may fail to capture key features of underlying risk curves.
For instance, insensitivity of AEL to different correlation cases is not
PI ¼ ð1 þ aÞEðLI Þ (4) effective as a risk metric, whereas dependency of VAR on selected
probability level can lead to different risk comparisons (e.g., 0.99
where a 5 safety-loading factor set to 1.0. level versus 0.999 level). In contrast, the risk ordering based on
TVAR is intuitive and consistent at 0.99 and 0.999 levels (i.e.,
full . partial . no; Table 1). Moreover, this trend is preserved for
Effects of Spatial Correlation Model and
all probabilities of interest for risk-management decisions, which
Regional Seismicity
is not the case for AEL and VAR. This is important to avoid mis-
Using an appropriate spatial correlation model is a modeling issue informed decision making related to seismic risk assessment.
and, thus, should be overcome; a partial correlation case, reflecting Next, seismic loss curves for House 4, which is located in Van-
spatial distribution of a building portfolio in the assessment, is more couver, Victoria, Montreal, and Ottawa, are compared in Fig. 5. It is
accurate. To inspect their effects on estimated aggregate seismic noted that seismic hazard levels as well as dominant EQS in terms of
loss, seismic loss curves for House 4 in Vancouver (considering expected magnitude and event type are different in the west and the
three correlation cases of partial, no, and full) are compared in Fig. east. In western Canada, seismic hazard levels are relatively high
4. A summary of risk metrics for the three cases is given in Table 1. because of three active earthquake sources (i.e., shallow crustal events,
Two important observations are as follows: (1) three curves intersect deep inslab events, and interface events from the Cascadia sub-
one another, indicating that the order of the loss curves in terms of duction zone). For example, uniform hazard spectra for Vancouver
severity is full , partial , no at probability levels above the in- and Victoria at the return period of 2,500 years (i.e., probability
tersection point, whereas the order is reversed at probability levels level adopted for the national seismic hazard maps in Canada) are
below the intersection point; and (2) differences of the loss curves in between 1.0 and 2.0g in the short-vibration-period range (0.1 to 0.5 s)
the upper right tail (i.e., small annual probability) are significant. and between 0.25 and 0.8g in the long-period range (1.0 to 3.0 s) for
Regarding the calculated values of risk metrics, AEL for the three soft soil (Atkinson and Goda 2011). On the other hand, seismic
curves is almost identical (Table 1), whereas VAR and TVAR vary hazards in the east are mainly contributed by local moderate earth-
significantly depending on the probability level of interest. It is quakes (Mw 5 5:0e6:5) and potential large earthquakes originated
noted that the annual loss occurrence probability differs for the from the St. Lawrence rift zone (Mw 5 7:0e7:5). Because of less-
three cases (Table 1). The results clearly show the importance of active seismicity in eastern Canada, uniform hazard spectra for
alistically, because both measures integrate an area under the seismic upgrading and earthquake insurance may be attractive for those who
loss curve (although AEL fails to quantify the upper-tail risk).
to provide more consistent risk comparison results for different (smaller values are preferred). Inspection of Fig. 8 and Table 2
seismic demand modeling cases, seismic capacity cases, and in- indicates that the seismic loss curves intersect one another; the cases
surance cases. with insurance/upgrading have greater reference cost due to in-
surance premium or upgrading costs, but are associated with reduced
seismic losses in cases of significant earthquake. This tendency is
Decision Analysis for Seismic Risk Mitigation manifested in the calculated values of AEL (Table 2); the base case is
A decision analysis is given for a hypothetical portfolio of 1,000 associated with the least expected cost. The impact of implementing
houses in Vancouver and Montreal. A risk manager of the building risk-transfer/mitigation measures for the houses in Montreal is
portfolio is concerned about earthquake risk exposure to a combined relatively minor because of the dominant influence of the houses in
portfolio of 500 Houses 3 in Vancouver and 500 Houses 4 in Vancouver (Fig. 5).
The risk ordering in Table 2 suggests that the ranking based on
AEL is different from those based on VAR and TVAR (note: de-
cision making based on AEL is a risk-neutral approach), and that, in
cases where potential large earthquake loss is concerned, the risk-
transfer/mitigation for the houses in Vancouver is a viable solution
based on the VAR and TVAR values (about 10–70% reduction for
VAR0:99 , VAR0:999 , TVAR0:99 , and TVAR0:999 values in terms of
no risk-mitigation option). It is noteworthy that the VAR-based risk
ordering at relatively small probability levels (around 0.99–0.999) is
more variable compared to the TVAR-based risk ordering. Further,
an application of the stochastic dominance rules to this problem
indicates that the Van-H3&Mont-H2Upg option is dominated by the
Van-H3Ins&Mont-H4 and Van-H1Upg&Mont-H4 options based
on the SSD rule (note: the FSD rule cannot eliminate any of the
options as the seismic loss curves for the five cases intersect). In light
of the result based on the SSD rule, the VAR-based risk ordering
can be misleading, if only a few risk metrics (in terms of probability
level) are used for decision making.
Finally, it is emphasized that the decision analysis described pre-
viously can be extended. Different risk-transfer and risk-mitigation
options can be assessed easily [for instance, different insurance
parameters, in terms of deductible, limit, and safety-loading factor,
Fig. 8. Comparison of seismic loss curves for a hypothetical portfolio
and different retrofitting options, in terms of proportion of upgra-
of 1,000 houses in Vancouver and Montreal by considering five risk-
transfer/mitigation cases
ded buildings and other available solutions (e.g., friction and fluid
dampers), which are not considered in this case study].
Table 2. Summary of Risk Metrics and Risk Comparisons for a Hypothetical Portfolio of 1,000 houses in Vancouver and Montreal
Option AEL and SDa VAR0:99 VAR0:999 TVAR0:99 TVAR0:999
Van-H3&Mont-H4 169 and 1,821 (1 and 5) 1,304 (5) 25,211 (5) 10,317 (5) 47,263 (5)
Van-H3Ins&Mont-H4 200 and 1,213 (3 and 1) 1,236 (3) 15,478 (2) 7,232 (2) 30,007 (1)
Van-H1Upg&Mont-H4 234 and 1,213 (4 and 2) 406 (1) 13,924 (1) 5,508 (1) 31,067 (2)
Van-H3&Mont-H4Ins 178 and 1,665 (2 and 3) 1,283 (4) 21,598 (4) 9,355 (4) 42,933 (3)
Van-H3&Mont-H2Upgb 238 and 1,666 (5 and 4) 846 (2) 21,502 (3) 8,689 (3) 43,671 (4)
Note: Unit 5 CAD$1,000; number in parentheses represents the risk order based on the risk metrics.
a
SD of annual seismic loss of the portfolio.
b
This option is dominated by the options Van-H3Ins&Mont-H4 and Van-H1Upg&Mont-H4 based on the SSD rule.
attitude, societal tolerable risk level, and life quality criterion.” J. Struct.
quantification approaches were illustrated through a case study of Eng., 10.1061/(ASCE)0733-9445(2006)132:12(2027), 2027–2035.
a group of wood-frame houses in Canada. Important conclusions Goda, K., and Hong, H. P. (2008). “Spatial correlation of peak ground
from the case study are that exclusive reliance on AEL in decision motions and response spectra.” Bull. Seismol. Soc. Am., 98(1), 354–365.
making (i.e., risk-neutral approach) is not appropriate in dealing with Goda, K., and Yoshikawa, H. (2012). “Earthquake insurance portfolio
low-probability high-consequence risks; risk ordering based on VAR analysis of wood-frame houses in south-western British Columbia,
is dependent on probability level and can be misleading; and TVAR Canada.” Bull. Earthquake Eng., 10(2), 615–643.
Goulet, C. A., et al. (2007). “Evaluation of the seismic performance of a
tends to provide more consistent risk-comparison results by reflecting
code-conforming reinforced-concrete frame building—From seismic
the expected risk of rare events (e.g., Fig. 4 and Table 1 for the effects hazard to collapse safety and economic losses.” Earthquake Eng. Struct.
of spatial correlation; Fig. 8 and Table 2 for the illustrative problem Dynam., 36(13), 1973–1997.
for seismic risk-mitigation decision making). Grace, M. F., Klein, R. W., Kleindorfer, P. R., and Murray, M. R. (2003).
The authors envisage the role of financial seismic risk analysis as Catastrophe insurance: Consumer demand, markets and regulation,
greater in the future owing to a major paradigm shift from life safety Kluwer Academic Publishers, Boston.
to damage control and loss reduction. The future extensions of the Hazus-MH 2.1 [Computer software]. Washington, DC, FEMA.
framework should consider incorporation of the effects of after- Kanda, J., and Okamura, S. (2006). “Estimation of PML for buildings with
various strengths in Japan.” Proc., 2nd Int. Forum on Engineering
shocks and multihazards (ground failure, fire, tsunami, and so forth)
Decision Making, International Forum on Engineering Decision
from hazard- and risk-modeling perspectives. From consequence- Making, Delft, Denmark.
modeling viewpoints, interdependency of functionality of inter- Lalonde, D. (2005). “Risk financing.” Catastrophe modeling: A new ap-
connected infrastructure systems, dynamic modeling of recovery proach to managing risk, Springer, New York.
process, and modeling indirect economic effects are of paramount Levy, H. (2006). Stochastic dominance: Investment decision making under
importance. These issues have been addressed partially by various uncertainty, 2nd Ed., Springer, New York.
researchers; yet, full consideration of these challenging problems Luco, N., and Bazzurro, P. (2007). “Does amplitude scaling of ground
needs to be achieved. motion records result in biased nonlinear structural drift responses?”
Earthquake Eng. Struct. Dynam., 36(13), 1813–1835.
McGuire, R. K. (2004). Seismic hazard and risk analysis, Earthquake
Acknowledgments Engineering Research Institute, Oakland, CA.
Moehle, J., and Deierlein, G. (2004). “A framework methodology for
performance-based earthquake engineering.” Proc., 13th World Conf.
The authors thank two anonymous reviewers for their insightful sug-
on Earthquake Engineering, International Association for Earthquake
gestions on the original manuscript. This research project was sup- Engineering, Tokyo.
ported by the Kajima Corporation Research Grant. The second Paultre, P., Lefebvre, G., Devic, J. P., and Côté, G. (1993). “Statistical
author is grateful for financial support provided by the Leverhulme analyses of damages to buildings in the 1988 Saguenay earthquake.”
Trust through the 2011 Philip Leverhulme Prize. Can. J. Civ. Eng., 20(6), 988–998.
Porter, K. A., Scawthorn, C. R., and Beck, J. L. (2006). “Cost-effectiveness
of stronger woodframe buildings.” Earthquake Spectra, 22(1), 239–
References
266.
Vamvatsikos, D., and Cornell, C. A. (2004). “Applied incremental dynamic
Artzner, P., Delbaen, F., Eber, J. M., and Heath, D. (1999). “Coherent analysis.” Earthquake Spectra, 20(2), 523–553.
measures of risk.” Math. Finance, 9(3), 203–228. Ventura, C. E., Finn, W. D. L., Onur, T., Blanquera, A., and Rezai, M.
Atkinson, G. M., and Goda, K. (2011). “Effects of seismicity models and (2005). “Regional seismic risk in British Columbia—Classification of
new ground-motion prediction equations on seismic hazard assessment buildings and development of damage probability functions.” Can.
for four Canadian cities.” Bull. Seismol. Soc. Am., 101(1), 176–189. J. Civ. Eng., 32(2), 372–387.
Baker, J. W. (2011). “Conditional mean spectrum: Tool for ground-motion White, T. W., and Ventura, C. E. (2006). “Seismic performance of wood-
selection.” J. Struct. Eng., 10.1061/(ASCE)ST.1943-541X.0000215, frame residential construction in British Columbia.” Earthquake En-
322–331. gineering Research Facility Rep. No. 06-03, Univ. of British Columbia,
Bargès, M., Cossette, H., and Marceau, É. (2009). “TVaR-based capital Vancouver, BC, Canada.
allocation with copulas.” Insur. Math. Econ., 45(3), 348–361. Wilkinson, S., et al. (2013). “Observations and implications of damage
Daniell, J., and Vervaeck, A. (2012). “Damaging earthquakes database from the magnitude Mw 6.3 Christchurch, New Zealand earthquake
2011—The year in review.” Æhttp://www.cedim.de/download/ of 22 February 2011.” Bull. Earthquake Eng., 13(1), 1–35.
CATDATDamagingEarthquakesDatabase2011AnnualReview.pdfæ (Jan. Yoshikawa, H. (2013). “Seismic risk and possible maximum loss (PML)
3, 2013). analysis of reinforced concrete structures.” Handbook of seismic risk
Ellingwood, B. R., Rosowsky, R. V., and Pang, W. (2008). “Performance analysis and management of civil infrastructure systems, Woodhead
of light-frame wood residential construction subjected to earthquakes Publishing, Cambridge, U.K.