Forecastability of Technological Progress
Forecastability of Technological Progress
One-line summary: Technological cost progress is genuinely forecastable — Wright's Law beat five rival laws across 62 technologies, with forecast error growing predictably — but adoption timing and the ceiling are far less forecastable, the key asymmetry for any investing use.
The insight
The Santa Fe / MIT work (Nagy, Farmer, Trancik) tested six candidate "laws" of technological improvement by hindcasting: Wright's Law produced the best forecasts, with Moore's Law close behind, and progress is forecastable with the root-mean-square logarithmic error growing roughly linearly at ~2.5%/year of horizon. Farmer & Lafond formalized this as a correlated geometric random walk whose error distribution collapses many technologies onto one universal curve — enabling a probability that one technology will out-cost another by a future date. The asymmetry that matters: the cost trajectory is the reliable part; the adoption date and the ceiling are the unreliable parts (Bass M/p/q are badly identified; learning rates change). So an evaluation lens should weight cost-curve evidence heavily and treat inflection timing as a wide distribution, not a point.
Evidence
- From 2026-06-16-academic-research-technology-adoption-s-curves: "Wright's law produces the best forecasts, but Moore's law is not far behind"; error "growing linearly with the forecasting horizon at a typical rate of 2.5% per year" (Nagy et al. 2012); universal error distribution and tech-vs-tech probability (Farmer & Lafond 2015).
- From 2026-06-16-academic-research-technology-adoption-s-curves: the offsetting unforecastable side —
M/p/qinstability (Massiani 2015), learning-rate change (Carlino 2025), asymmetric curves (Easingwood 1981). - From 2026-06-16-autoresearch-tony-seba-technology-disruption-s-curves: the live confirmation that cost is forecastable (LLMflation 10×/yr) while timing is not (tony-seba's cost calls right, adoption-timing calls wrong).
Design implications
- Anchor conviction on the cost curve; express adoption timing as a range with explicit falsifiers (see cost-curve-tipping-point-to-s-curve-adoption).
- "Probability tech A beats tech B by year T" is a usable framing for relative-value technology bets.
Contradictions / tensions
- Aggregate forecastability (Nagy/Farmer) sits against per-technology instability (Carlino/Massiani) — reconciled by separating "cost direction" (forecastable) from "adoption timing/ceiling" (not).