Time windows of opportunities to fight earthquake under-insurance: evidence from Google Trends

The study addresses the persistent challenge of earthquake under-insurance despite the recognized role of insurance in disaster risk management and post-disaster recovery. Prior work has emphasized high premiums, socio-demographic and geographic factors, and low risk perception as barriers to purchase. Traditional research has relied on top-down conceptual models with limited data, often simplifying key components like perceived risk. This paper proposes a bottom-up, data-enabled approach using Google Trends (GT) to monitor public interest in earthquake insurance at daily resolution. Focusing on Italy—a high-hazard country with among the lowest earthquake insurance penetration in the OECD—the research investigates how direct, indirect, vicarious, and life experiences with earthquakes, as well as policy actions, shape online information seeking. The aims are to identify triggers and temporal patterns in insurance-related searches, explore links with actual purchasing behavior, and inform insurers and policymakers about time windows of opportunity to increase uptake.

The paper reviews research on disaster insurance demand, noting variations between developing and developed contexts and frequent unwillingness to purchase due to cost, demographics, geography, and low perceived risk. OECD highlights the difficulty of achieving comprehensive, affordable coverage. Conceptual top-down models have simulated interactions among premiums, socio-economics, and risk perception, but validation is constrained by data limitations. Psychological and behavioral literature identifies four experience types (direct, indirect, vicarious, life experience) that influence preparedness, including insurance uptake, with direct experience often most impactful and vicarious experiences acting as important predisposing factors. Prior empirical works show that experiencing shaking without losses can raise insurance demand. Google Trends has been used as a proxy for public awareness/interest in hazards (earthquakes, drought) across scales, supporting its utility for tracking attention dynamics relevant to insurance.

Study area and data: Italy was selected due to high seismic hazard and low insurance penetration. Data sources include: Google Trends (GT) monthly and daily search activity volumes from 2004–2019; GT GeoMaps (GM) for regional diffusion; World Bank annual percentages of internet users in Italy; and earthquake catalogs (INGV/CPTI). Search configuration: Two Italian search terms were used—"terremoto" (earthquake) and "assicurazione terremoto" (earthquake insurance)—with GT options "search term" and "search topic". Monthly GT data (2004–2019) were retrieved, and daily GT data were collected in contiguous 6-month windows (32 intervals from Jan 2004 to Dec 2019), saved on January 28, 2020. GT GeoMaps were retrieved for key dates (Feb 9–12, 2020) to visualize spatial diffusion of interest by region. Accounting for internet adoption and GT scaling: To correct for (a) GT’s relative scaling (0–100 within selected periods) and (b) the growth in the share of internet users (34% in 2004 to 73% in 2018), the authors constructed a weighted daily GT series (wDGT): (1) Monthly GT series MGT(y,m) were expanded to daily MGT(y,m,d) by assigning each day in a month the monthly value. (2) Annual internet user percentages API(y) were expanded to daily API(y,m,d)=API(y). (3) Daily weights wgt(y,m,d) were computed as [MGT(y,m,d)/100]×[API(y,m,d)/100]. (4) Raw daily GT values DGT(y,m,d) were multiplied by wgt(y,m,d), and the product was normalized by its maximum over the full period, then scaled to 0–100 to obtain wDGT(y,m,d). This produced comparable daily time series for both search terms across 2004–2019. GeoMap method: For each significant event, cumulative GMs were built by iteratively expanding the date window and recording which Italian regions contributed to the national search volume, illustrating temporal-spatial diffusion of interest. Analysis: The wDGT series for "terremoto" (TR) and "assicurazione terremoto" (ATR) were compared with daily earthquake counts and magnitudes. Temporal correlations (Spearman rank and Pearson linear) were computed over four periods segmented by major earthquakes. Peaks were classified as: (1) immediate earthquake-related; (2) not immediately earthquake-related (e.g., overseas events or policy decisions); and (3) no peaks at earthquake occurrence. Descriptive analyses of peak magnitudes, durations, and spatial diffusion were conducted.

• Four temporal periods of public interest were identified based on major events, with increasing persistence and magnitude over time (Table 1 summary): 1) 1st (A): 01/01/2004–04/05/2009, 1922 days; ATR activity essentially absent (days with volume>0 = 10; 0.5%); ATR range 0.0–0.1; total ATR volume 1.0. 2) 2nd (B): 04/06/2009–05/19/2012, 1140 days; very low interest (112 days; 9.8%); ATR range 0.0–3.8; total 75.0. 3) 3rd (C): 05/20/2012–08/23/2016, 1557 days; low–moderate interest (291 days; 18.7%); ATR range 0.0–47.8; total 1568.1. 4) 4th (D): 08/24/2016–12/31/2019, 1225 days; moderate–high interest (359 days; 29.3%); ATR range 0.0–100.0; total 4223.8. - Correlations between wDGT and daily earthquake frequency strengthened in later periods compared to early 2000s, indicating increasing coupling between seismicity and public search behavior. - Event-triggered peaks and durations: • 2009 Abruzzo earthquake (Mw 6.3): TR peaked same day; ATR rose modestly (max ~2.4) with low volumes for ~20 days; spatial diffusion reached 50% of regions cumulatively, with affected region (Abruzzo) appearing in searches after ~3 weeks, likely due to immediate emergency focus. • 2011 Tohoku, Japan (M9.0 and >20 M6+ events): despite no domestic seismicity, a small ATR deviation from baseline occurred, evidencing vicarious experience via media. • Policy triggers: May 15, 2012 Decree discussion linked with ATR peak 3.8 (no significant seismicity); Aug 1, 2016 Resolution on disaster compensation correlated with ATR peak 17.1. • 2012 Emilia sequence (Mw 6.1 on May 20; Mw 5.9 and 5.5 on May 29): ATR jumped to 20.6–26.3, then to 47.8 on May 29; notable persistence (ATR>10 until June 5). Initial searches concentrated in affected and bordering regions (Emilia-Romagna, Lombardy, Veneto, Tuscany, Marche), consistent with higher pre-existing insurance awareness and policy penetration. • Subsequent moderate events (e.g., Tuscany 2013 Mw 5.4; Piedmont 2014 Mw 4.7; Tuscany 2014 Mw 4.1; Emilia area 2015 Mw 4.3) produced varying ATR responses, generally higher where prior awareness/coverage was stronger. • 2016–2017 Central Italy sequence (Mw 6.2 on Aug 24; Mw 6.1 on Oct 26; Mw 6.6 on Oct 30): Strongest ATR response observed—88.9 on Aug 24; sustained ATR>10 until Sep 3; maximum ATR=100 following Oct 30 event; persistent attention over ensuing weeks amid M4+ aftershocks. • 2017 Ischia (Casamicciola) Mw 3.9: TR peaked day after (22.8); ATR peaked on event day (36.6) and remained high next day (35.8), illustrating time-of-day and media effects. • 2019 Lazio Mw 3.6 near Rome: despite low magnitude, ATR rose and persisted for about a week, likely reactivating memories of prior local and Central Italy events and contemporaneous media discourse on reconstruction delays. - Three trend types for ATR were identified: (1) immediate earthquake-related peaks; (2) peaks decoupled from immediate domestic seismicity (overseas quakes, policy decisions/media); and (3) no peaks at time of some earthquakes. - Spatial diffusion: GM analyses show searches originate in affected/bordering regions and then spread nationally; vicarious experiences enhance diffusion. Some southern regions with medium-high seismic hazard (Molise, Basilicata, Calabria) showed minimal ATR activity, aligning with low internet access and very low insurance penetration (<2%). Low-hazard regions (e.g., Piedmont) still exhibited interest, suggesting effects of "no-loss" experience and opportunities to manage adverse selection. - Duration of interest: high-moderate ATR levels typically lasted 8–10 days; low levels persisted from 3–4 months up to 3 years. - Indicators of intent to purchase: GT associated queries included cost-related searches and brand-specific queries (e.g., "Generali assicurazione terremoto"). - Insurance uptake data: Insured houses (earthquake and floods) were ~35,000 (~0.1%) in 2009; earthquake-insured houses reached ~608,000 in 2018 and 781,000 in 2019, mirroring increased online interest, especially during the fourth period.

The findings demonstrate that a bottom-up, data-enabled approach using GT can capture temporal and spatial patterns of public interest in earthquake insurance and identify windows of opportunity following direct experiences (domestic earthquakes), vicarious experiences (overseas events and media coverage), and policy decisions. Increasing correlations over time suggest growing responsiveness of public information seeking to seismic activity. The persistence and spread of interest indicate that targeted communication and offers could be timed to capitalize on heightened awareness for up to 1–2 weeks post-event, with lower-level attention persisting much longer. Observed interest in low-hazard regions indicates that campaigns there may diversify risk pools and mitigate adverse selection, while absent interest in some high-hazard southern regions highlights areas where education and access gaps may need addressing. The alignment between search behavior and rising insurance uptake suggests that online interest can translate into action, though the causal pathway needs further validation.

This study introduces and validates a bottom-up methodology that integrates Google Trends with contextual data to analyze earthquake insurance interest at daily and regional scales. Applied to Italy, the approach identifies clear triggers (domestic and overseas earthquakes, policy actions) and temporal windows during which public interest peaks and diffuses geographically, offering actionable insights for insurers and policymakers to counter under-insurance. Evidence suggests that interest can persist beyond immediate events and that low-hazard regions can be effective targets to broaden coverage and address adverse selection. While correlations with increased policy uptake are encouraging, future research should combine GT with survey and social media data to test causal links between information seeking and purchase behavior, and to refine strategies for timing, targeting, and messaging in different regional and hazard contexts. The methodology is transferable to other countries facing low earthquake insurance penetration.

• Google Trends opacity: GT provides post-processed, relative volumes without detailed documentation of data generation, and filters repeated queries, limiting interpretability and some causal/network analyses. - Aggregate data: Inability to track individual behavior (user flows, engagement, conversion) constrains direct inference about the information-seeking to purchasing pathway. - Scaling and comparability: GT’s relative scaling across time windows necessitated weighting/normalization; residual biases may remain. - GeoMap sensitivity: Regional GM outputs can be influenced by population size and internet access disparities, potentially underrepresenting low-population or low-access regions. - Need for complementary data: Validation with transaction-level insurance data (regional/monthly) and carefully designed surveys, as well as integration with other social media platforms, is needed to robustly link search interest to actual policy uptake.