Risk caused by the propagation of earthquake losses through the economy

Economic production (indirect) losses from disasters can equal or surpass direct physical damage, yet are hard to quantify from scarce historical data. While catastrophe models for physical earthquake damage are mature, translating property damage into indirect economic losses remains challenging and often relies on crude ratios. The growing complexity of global supply chains heightens the importance of accounting for indirect losses. Existing studies on economic impacts of disasters and new modeling frameworks have advanced the field, but often analyze individual events without incorporating event frequency, and give limited attention to linking physical damage to higher-order losses. Given these gaps, the paper introduces a systematic, probabilistic integration of comprehensive seismic risk assessment with spatial Computable General Equilibrium (CGE) models to estimate higher-order, economy-wide losses. The approach treats earthquakes as a stochastic time process, explicitly accounts for event frequencies, and links physical damage to reductions in productive capital stock. Using a spatial CGE (SCGE) framework that captures the geographic distribution of agents and endowments, the study computes AAL and LECs for multiple economic variables across country, regional, and sectoral levels. Chile is used as a detailed case study with 44,350 earthquake events describing national seismic hazard.

A substantial literature examines economic impacts of natural and man-made disasters, with recent advances compiled by Okuyama and Rose. Efforts include improving quantitative models for cyber-attacks, extreme weather, earthquakes, floods, climate change, and terrorism; integrating transportation, critical supply chains, and community demand models; and developing methods to estimate economic losses from physical hazard data or proxy measures (e.g., nighttime lights). European initiatives (ESPON-TITAN), the European Commission, and the World Bank employ multi-regional input-output models to assess direct and indirect disaster impacts. Despite this rich framework, most analyses focus on single events and neglect frequency of occurrence, which is critical for risk-based decision-making. Moreover, limited attention has been given to explicitly linking infrastructure/economic component damage to higher-order economic outcomes. The paper positions spatial CGE models as going beyond input-output models by incorporating agent behavior, price effects, and competition for scarce resources, which prior work shows are better suited for detailed disaster impact assessments and may avoid overestimation of losses seen in conventional MRIO analyses.

General approach: A stochastic earthquake catalog consistent with a regional seismic hazard model is generated. For each event, probabilistic physical damage (direct losses) to assets relevant to production (non-residential buildings, factories, infrastructure) is estimated using conventional probabilistic seismic risk assessment. These direct losses are then mapped to exogenous reductions in capital stock in a spatial CGE model. The CGE model is rerun to a new equilibrium, yielding indirect (higher-order) economic impacts. Repeating across all events produces samples to compute probabilistic risk metrics (AAL, LEC) for multiple variables. Seismic risk model: The exposure database identifies asset location, vulnerability, and economic sector. The hazard component provides a collectively exhaustive event set with annual occurrence frequencies and intensity fields (e.g., PGA). The loss component uses sector-specific vulnerability functions to derive probabilistic losses per asset. Given CGE’s coarser spatial resolution, losses are aggregated by sector-region accounting for correlation among asset losses for the same event. Economic model: The Chilean BMCH spatial CGE model (a B-MARIA variant) represents 15 regions, 12 production/investment sectors, regional households and governments, a central government, and a single foreign region. Production uses regional capital and labor; the model is Johansen-type solved via linearized equations (GEMPACK/CRunGEM). Interregional trade flows and elasticities calibrate interregional linkages. Direct physical losses by sector-region are introduced as exogenous capital stock shocks (ratio of loss to capital stock). The model computes percentage changes in endogenous variables (e.g., output, prices, employment) relative to the pre-shock baseline. Indirect loss for sector i in region j is the drop in production after the shock relative to baseline. Risk measures: For event k with expected direct loss Ld_k and indirect (production) loss Lp_k and annual frequency Fk: AAL_d = Σ E(Ld_k)Fk; AAL_p = Σ E(Lp_k)Fk. Loss exceedance curves ν_d(l) and ν_p(l) sum exceedance probabilities times Fk across events. The procedure extends to other CGE variables (employment, GDP/GRP, CPI, exports, tariff revenue) and to positive effects (gains), counting only positive outcomes. Implementation: Calculations used DIRAS-2020 to integrate the seismic risk model and the SCGE (BMCH) via CRunGEM/GEMPACK. The Chile case considers 44,350 events. Direct losses are restricted to non-residential buildings to reflect productive capital stock damage relevant for propagation through the economy.

- National aggregates (AAL): Direct AAL for non-residential buildings: 302 million USD (0.290% of exposed non-residential building value of 103,720 million USD). Production AAL: 583 million USD (0.132% of yearly production of 442,805 million USD). GDP AAL: 305 million USD (0.122% contraction of 251,020 million USD). Employment AAL: 7,786 workers (0.115% of 6,751,073). Export volume AAL: 62 million USD (0.075% of 83,102 million USD). - Loss exceedance: At 250-year return period, production loss ≈ 15,870 million USD (3.58% of yearly production); direct loss ≈ 5,025 million USD (4.85% of non-residential building value). At 1,000-year, production ≈ 28,760 million USD; direct ≈ 9,835 million USD. - Relationship of indirect to direct losses: For return periods up to ~50 years, production losses scale positively with direct losses up to ~0.74C (C = total production / total non-residential building value). Between ~50–400 years, production losses exceed direct losses. For rarer, more severe events (>400-year), the ratio declines slowly (≈0.6C at 10,000-year; ≈0.5C at 100,000-year). - Sectoral risk: Direct losses AAL share is dominated by S6 “Commerce, hotels and restaurants” (~35%). For production losses, the riskiest sector is S7 “Transport, communications and information services,” while S3 “Manufacturing industry” contributes the largest share to total production AAL (~23%). - Regional risk: The Metropolitan Region (R7) concentrates the largest AAL: ~40% of direct and ~41% of production AAL. However, relative risk rankings differ: Atacama (R4) is riskiest by direct losses (relative AAL), whereas Valparaiso (R6) is riskiest by production losses. Antofagasta (R3) ranks 5th by direct but 12th by production losses, highlighting propagation effects. - Regional/sectoral LECs: At 250-year, R7 production loss ≈ 10,674 million USD (5.45% of its annual production). In relative terms, up to ~450-year RP the most affected region is R6; beyond ~450-year, R1 becomes most affected. Sectorally, S3 (manufacturing) has highest absolute production losses; S7 (transport/communications/information) has highest relative losses for all RPs. - Complementary indicators: LECs computed for employment, GDP/GRP, exports, CPI. Example: at 250-year RP, export volume loss is ~2.25% of annual exports; model indicates average GDP contraction AAL of 0.122%. The CPI indicator shows central-north regions more price-sensitive to earthquakes. - Positive effects: Some regions/sectors exhibit gains due to substitution and relative price effects. Average Annual Gain (AAG) of national production ≈ 18.32 million USD (0.0041% of yearly production), smaller than losses since recovery/reconstruction is not modeled. - Scenario analysis and coherence with observations: For an Mw8.8 Maule-2010-like event, maximum price increase occurs in Biobío (R10) at ~1.9%; average national production loss ~1.7% of yearly production. Modeled direct non-residential losses (~2.5% of stock) align with Central Bank estimates (~2.6%). Modeled yearly GDP contraction ~1.65% is consistent with government’s 7.6 billion USD GDP loss over four years (implying ~1.74% first-year contraction given 2010 GDP of 218.5 billion USD).

The integrated probabilistic seismic risk–spatial CGE framework provides a robust, systematic view of how earthquake-induced physical damage propagates through intersectoral and interregional linkages to generate higher-order economic impacts. By computing AAL and LECs for production and multiple macro and regional variables, the study offers risk metrics that complement traditional physical damage indicators, improving comprehensive disaster risk management and informing financial hedging for governments and insurers. Results emphasize supply-chain disruption as a key propagation channel and show that regions not most physically damaged can be most affected economically. The framework also identifies potential positive spillovers (gains) via substitution and relative price effects, aiding planning for uneven regional impacts. The approach is fully probabilistic on the hazard/damage side and currently deterministic on the CGE parameter side; nonetheless, indirect losses remain probabilistic because inputs are stochastic. The methodology is flexible, extendable to other hazards (floods, hurricanes, droughts), and supports both short-run and long-run closures (e.g., with labor migration), illustrating magnification of losses absent mitigation and capital recovery. Partial validation against observed outcomes (e.g., Maule 2010) indicates coherence of modeled losses and price effects.

This study advances catastrophe risk assessment by integrating probabilistic seismic risk modeling with a spatial CGE framework to quantify both direct and higher-order economic impacts of earthquakes. It introduces production AAL and production LECs, alongside complementary indicators (employment, GDP/GRP, exports, CPI), at national, regional, and sectoral scales. Applied to Chile’s 44,350-event catalog, the approach reveals how physical capital damage translates into systemic losses through supply chains, identifies differential regional-sectoral vulnerabilities, and captures occasional positive effects from substitution. Future research directions include: incorporating uncertainty into CGE behavioral parameters and structural coefficients; enhancing empirical validation strategies; modeling human-physical feedbacks (e.g., risk perception and adaptation) within utility/production functions; and systematically representing recovery and reconstruction dynamics in long-run scenarios to assess mitigation and resilience policies.

- CGE side treated deterministically (behavioral parameters and structural coefficients fixed); uncertainty characterization in CGE not yet incorporated. - Positive economic effects are likely underestimated because recovery and reconstruction processes of lost capital are not modeled; analysis is primarily short-run (though long-run closure is feasible and explored illustratively). - Empirical validation is inherently limited due to rarity of catastrophic events and evolving urban/building conditions; direct exceedance-frequency validation is infeasible. - Direct loss modeling restricted to non-residential buildings (productive capital), excluding residential asset damage from direct loss metrics. - Aggregation from detailed asset losses to sector-region requires handling correlations; CGE’s coarser spatial resolution may smooth local heterogeneity. - Transferability of macro assessments to regional policy levels can be constrained; nevertheless, the spatial CGE approach mitigates this to some extent.