Tools chosen: kalshi_direct kalshi_related polymarket_related wikipedia claude_news code_execution

Sub-questions (Fermi decomposition)

Planner reasoning

This is a long-horizon geophysical base-rate question: no M8.0+ earthquake has ever been instrumentally recorded in California (largest are 1857 Fort Tejon and 1906 San Francisco, both ~7.9). The key is the UCERF3 forecast rate for M≥8.0 in California over a ~10-year window, converted to a Poisson probability, then compared with the Kalshi price which likely carries a floor/noise premium.

# Event Will an earthquake of magnitude ≥8.0 with epicenter in California (or its territorial waters) occur before Dec 31, 2035? # Outcomes to forecast - Yes - No # Kalshi market anchor KXEARTHQUAKECALIFORNIA-35 current YES price: **27%**. 7-day change: +3pts; 30-day change: -5pts. Average daily volume ~77-132 contracts (thin market). Historical range 7%–65% over 113 data points, indicating high volatility/low liquidity and possibly mispricing swings. [kalshi_direct] # Sub-question answers 1. **Current Kalshi price/volume/trend?** — 27% YES, up 3pts week-over-week, down 5pts month-over-month, thin volume (~77-132 contracts/day), wide historical range (7%-65%) suggesting past mispricing or news-driven swings. [kalshi_direct, kalshi_related] 2. **Largest recorded CA earthquake / M≥7.8 events since 1800?** — Largest well-documented CA quakes are ~M7.9 (1857 Fort Tejon, 1906 San Francisco). No confirmed M≥8.0 event has ever been recorded with a California epicenter in the historical record (since ~1769). [wikipedia] 3. **UCERF3 30-year/annual M≥8.0 probability?** — USGS UCERF3 gives ~7.0% probability over 30 years (2014-2043) for an M≥8.0 in California, up from 4.7% under UCERF2, largely due to inclusion of multi-fault ruptures. Southern San Andreas is the most likely source. [claude_news/USGS/SCEC] 4. **Poisson probability for ~10-year window to 2035?** — Using UCERF3-derived rate (λ≈0.00242/yr), 10-yr P≈2.4%. Blending UCERF3, historical zero-count (Jeffreys), and Gutenberg-Richter extrapolation methods gives a central estimate of **~3%** (range ~2.4%-4.9%), with a loose upper-bound sensitivity case (small-sample historical) reaching ~14%. [code_execution] 5. **Can San Andreas produce M≥8.0? Plausible max?** — Multiple sources (USGS, SF Chronicle, LiveScience, EarthquakeCountry.org) converge on a maximum credible magnitude of ~8.3 for the San Andreas system given its 800-mile length and 10-12 mile depth; M9+ is considered essentially impossible (requires subduction-zone geometry CA lacks). So M≥8.0 is physically plausible but only in the top slice of the fault's capability. [claude_news] 6. **Parallel Kalshi/Polymarket markets?** — KXEARTHQUAKECALIFORNIA-28 (Before 2028) trades at 13%; KXEARTHQUAKEJAPAN-30 (Before 2030) trades at 40%. Polymarket has no matching earthquake markets. The 2028 vs 2035 CA spread (13% vs 27%) implies ~14pts of incremental probability for the extra ~7 years (2028-2035), roughly consistent with a low annual hazard rate compounding, though both prices sit well above model-implied base rates. [kalshi_related, polymarket_related] # Key facts (high-confidence, factual) 1. [wikipedia] No M≥8.0 earthquake has ever been recorded with a California epicenter historically (record since 1769); largest are ~M7.9. 2. [claude_news/USGS] UCERF3 (current official USGS model) estimates 7% probability of M≥8.0 in California over 30 years (2014-2043). 3. [claude_news] Multiple independent sources cite ~8.3 as San Andreas's maximum credible magnitude; M9 ruled out. 4. [code_execution] Central Poisson-model estimates for a ~10-year window (to 2035) cluster around 2.4%-4.9%, blended ~3%. 5. [kalshi_direct] Kalshi YES price (27%) is roughly 6-11x higher than model-based central estimates (~2.4%-4.9%). # Cross-market signals - Kalshi related: KXEARTHQUAKECALIFORNIA-28 (shorter horizon) at 13%; KXEARTHQUAKEJAPAN-30 at 40% (Japan subduction zone has much higher base rate, consistent with pricing gap). - Polymarket: No matching markets found (0 results across earthquake/California/magnitude keywords). - Sportsbook implied: N/A (not applicable to this event type). # Analyst opinions and speculation - Seismologists (SCEC, USGS) emphasize southern San Andreas as most likely M8+ source, but stress this is a multi-decade base rate, not near-term prediction. [claude_news] - Dec 2025 San Ramon/Bay Area swarm activity (M4+ sequences) drew media attention, but earthquake analysts (e.g., Earthquake Insights newsletter) explicitly caution that swarms "have occurred a number of times without triggering a larger...event" — no elevated near-term M8 signal. [claude_news] - 2024 was a record year for M4+ sequences in Southern CA (15 sequences, highest in 65 years per Lucy Jones), but this reflects moderate seismicity, not necessarily elevated M8 risk. [claude_news] # Directional lean per outcome - **Yes**: Supported by nonzero physical capability of San Andreas (~8.3 max), UCERF3's stated 7%/30yr figure, and recent (non-causal) elevated regional seismic activity. Opposing: no M≥8.0 CA event in ~250 years of record; model-based 10-yr Poisson estimates (2.4%-4.9%) are far below market price. - **No**: Strongly favored by base-rate/model analysis — central estimates converge near 3%, implying the 27% Kalshi price substantially overstates true probability. Historical absence of any M≥8.0 CA event reinforces this. # Gaps / unknowns - No detailed breakdown of why Kalshi price sits at 27% vs. model estimate of ~3% — could reflect genuine model uncertainty, tail-risk premium, low liquidity/mispricing, or event-linked category structuring (e.g., possible correlated markets/promotional pricing). - Unclear how "territorial waters" epicenter clause affects probability (could include offshore Mendocino/Cascadia-adjacent events, potentially raising true probability modestly). - No updated post-2014 USGS model (e.g., UCERF4) probability was found in research. # Calibration anchors - Kalshi current YES price: **27%** (anchor, but appears divorced from underlying model consensus). - Model-based blended central estimate: **~2.4%-4.9%** (10-year window Poisson from UCERF3 and historical rates). - Precedent: Related shorter-horizon market (2028) prices at 13%, also well above pro-rata model expectations, suggesting a systematic market premium for CA "Big One" tail risk across this ticker family.

1. Both forecasters converge tightly on 7-8% and never engage with why the market is persistently 6-11x higher across multiple related tickers (2028 at 13%, 2035 at 27%) rather than a one-off mispricing — a systematic pattern across the whole ticker family is weaker evidence for "thin-market noise" and stronger evidence the market may be pricing something the Poisson/UCERF3 model misses (e.g., fat-tailed clustering, foreshock-triggering dynamics, or correlated multi-fault rupture risk not fully captured in simple rate conversion), yet neither forecast seriously entertains that the market could be partially right. 2. Neither forecast addresses the "upper-bound sensitivity case" explicitly flagged in the brief reaching ~14% from small-sample historical methods — this is a materially different estimate from the ~3% central case, and both forecasters anchor near the low end without justifying why the upper bound is discounted so heavily. 3. Both treat the UCERF3 30-year-to-10-year conversion as straightforward linear/Poisson scaling, but neither questions whether hazard is uniformly distributed across the 30-year window — if UCERF3's methodology loads risk toward multi-fault "supercycle" scenarios with time-dependent (not memoryless) recurrence, a Poisson conversion could understate near-term risk if we're now closer to the end of a long quiet period on southern San Andreas. 4. The "territorial waters" clause and offshore Mendocino/Cascadia-adjacent triggering potential is acknowledged as a gap in the brief but dismissed by both forecasters with nearly identical boilerplate language ("modestly raise risk") without any quantitative adjustment — this is a genuine unknown that could bias both estimates low, especially given Cascadia's subduction potential for very large offshore events near the CA border. 5. Both forecasts are essentially identical (8% vs 7%) despite claiming independent reasoning, raising a concern about anchoring on the same "~3% + shading" heuristic rather than robustly stress-testing the model uncertainty band (2.4%-4.9% central, up to 14% tail) — greater dispersion between two "independent" forecasts would have been expected given the stated uncertainty range. 6. Neither highlights that no updated UCERF4 estimate exists (flagged in the brief) — reliance on an now decade-old-plus UCERF3 model to price a decision through 2035 introduces model risk that isn't reflected in the tight 7-8% estimates, which look overconfident given confidence levels of 0.72/0.69 attached to point estimates that ignore this staleness.

10% (70% confidence)

Yes: 10% No: 90%

UCERF3's official 7%/30-year figure for M≥8.0 in California converts to roughly 2.4-4.9% over the ~10-year window to end-2035, and no M≥8.0 has ever been recorded with a California epicenter in ~250 years of instrumental/historical record (max ~7.9, with San Andreas max credible ~8.3). The Kalshi anchor of 27% is 6-11x model estimates in a very thin, volatile market (7%-65% historical range, ~100 contracts/day), so heavy discounting is warranted. However, the critique's points have merit: the premium is systematic across the ticker family (13% for 2028), Poisson conversion may understate time-dependent risk on a long-quiet southern San Andreas, the 'territorial waters' clause could capture an offshore Cascadia/Mendocino-adjacent rupture, UCERF3 is a decade-plus stale with no UCERF4 replacement, and the brief's own sensitivity band reaches ~14%. I therefore push above both forecasters' 7-8% to 10% — meaningfully above the pure model central case to respect model and definitional uncertainty, but still far below a market price that shows classic retail 'Big One' tail-risk overpricing.

Total pipeline time: 112.7s