Long-range forecasting of societal dynamics
Quantitative theory that predicts the trajectory of large societies (collapse, war, economic eras) decades to centuries ahead with stated probabilities.
State of the art
Fragile Families Challenge (PNAS 2020): best ML models predicted life outcomes only marginally better than a 4-variable baseline.
- Chaotic sensitivity and rare events (wars, inventions) dominate long-run outcomes
- Reflexivity, since published forecasts change behavior
- Only one historical sample of most macro-events for validation
Who’s working on it
- Peter Turchin ↗ResearcherComplexity Science Hub Vienna
Founded cliodynamics and co-founded the Seshat databank (2011); his structural-demographic models fit historical instability cycles retrospectively
The field's main empirical resource, including CrisisDB with over 150 coded historical crises, which found no single typical crisis trajectory
- Complexity Science Hub Vienna ↗Academic lab
Hosts the Seshat and crisis-resilience programmes, including the FWF-funded AReS project (2026-2028) comparing historical and modern societies under shock
- Daniel HoyerResearcherComplexity Science Hub Vienna
Managing director for Seshat and lead on the Crisis Database work connecting historical crises to contemporary polycrisis analysis
- Matthew SalganikResearcherPrinceton University
Ran the Fragile Families Challenge, whose 2020 PNAS result - 160 teams barely beating simple baselines - is the strongest evidence that social prediction is much harder than assumed
- Metaculus ↗Company
Operates the largest public scored forecasting record on geopolitical and economic questions, the only continuously validated out-of-sample track record in this area
- Forecasting Research InstituteProgram
Runs adversarial and long-range forecasting tournaments (including the 2023 Existential Risk Persuasion Tournament) designed to test whether long-horizon predictions can be scored at all
What to invest in
| Kind | Investment | Scale |
|---|---|---|
| research | Large historical databases (e.g., Seshat) with out-of-sample forecasting tests | $$ $10M – $100M |
Depends on
Required by
No other capability builds on this directly.