How Historacle works

Historacle answers a single question: if your macro view plays out, what does history say happens next? You set a scenario — moves in rates, oil, inflation, the dollar, and unemployment, plus context flags like recession or stagflation — and the engine finds the historical episodes that most resemble it, then projects nine markets forward.

Crucially, it shows its work. Every projection is backed by named historical episodes, a regime read, and an explicit uncertainty band. It is an analysis instrument, not a prediction engine.

Historacle runs two independent projection engines and shows you both:

• Analog cone. Replays the actual trajectories of matched historical episodes, anchored to today's level and shifted (dampened) toward your specific inputs. This is “what actually happened in similar conditions.”
• Decomposed cone. Uses Gaussian kernel regression to estimate each macro driver's independent effect, then sums the regime-weighted contributions. This is “what the relationships imply.”
• Blended cone. A weighted combination — analog-dominant when you have a direct historical match, decomposed-dominant when your scenario is novel.

A Bayesian Markov regime-switching model (a Hamilton filter calibrated on ~39 years of data) infers a probability distribution over six cycle phases: crash, rebound, recovery, expansion, late-cycle, and stable. The same shock plays out very differently across regimes, so the engine weights historical response curves by the current posterior — and forward-simulates regime probabilities up to 24 months out.

The episode library contains 54+ historical periods (1980–present), each tagged with its macro deltas, condition flags, and cycle phase. When you run a scenario, the matcher scores every episode for similarity across all dimensions, keeps the closest, and weights them. Each match is shown with a quality badge and a plain-English explanation of why it qualified — so you can accept, question, or click through to overlay its real path on the chart.

Two kinds of controls shape a scenario:

• Six macro sliders. Fed policy (bps), oil (%), natural gas (%), CPI (pp), the US Dollar Index (pts), and unemployment (pp). Channels you don't pin are imputed from the others and the regime.
• Condition flags. Thirteen binary contexts across market stress (bank/credit stress, recession, housing bust), external shocks (geopolitical, tariff, EM crisis, energy crisis, pandemic), and macro regime (stagflation, deflation, dollar surge, yield inversion).
• Presets. One-click scenarios — bullish (soft landing, 1995 Goldilocks), bearish (2008 GFC, 1970s stagflation, 2020 pandemic), and neutral base cases.

The chart layers historical reality, published consensus, and your scenario. The shaded cone is the 10th–90th percentile range from 300 bootstrap resamples; the solid line is the scenario median. A thin dashed line shows published consensus where one exists, and a distinct futures-implied curve appears for markets like gold and copper. Below the chart, a stacked strip shows forward regime probabilities.

The stock panel projects any ticker 12 months out by layering three effects: a market layer (your macro scenario via the S&P projection), a sector layer (sector-specific sensitivities), and an optional event layer (earnings surprises, launches, regulatory shocks). You get an expected return, a 10–90 price range, probability of loss, and CAPM metrics (beta, volatility, R²).

Historacle is built on authoritative, live data with transparent fallbacks:

• FRED (Federal Reserve Bank of St. Louis) — historical series for all nine indicators.
• EIA STEO — official monthly forecasts for crude oil and natural gas (~24-month horizon).
• Fed SEP — central-bank projections for unemployment and inflation.
• Philadelphia Fed SPF — professional forecasters' median for the 10Y Treasury yield.
• Wall Street aggregate — curated median of major firms' S&P 500 targets.
• Yahoo Finance — equity prices and futures-implied forward curves.

Analog cone

Projection built by replaying the actual paths of matched historical episodes, anchored to today's level and delta-corrected toward your inputs.

Decomposed cone

Projection built from kernel-regression estimates of each macro driver's independent marginal effect on the indicator.

Blended cone

A weighted mix of the analog and decomposed cones — analog-heavy when you have direct historical precedent, decomposed-heavy in novel scenarios.

Regime posterior

The probability distribution over six cycle phases (crash, rebound, recovery, expansion, late-cycle, stable) inferred by the Bayesian filter.

Percentile band

The 10th–90th percentile range of outcomes from 300 bootstrap resamples across matched episodes — a measure of scenario uncertainty.

Consensus forecast

Published professional forecasts (Fed SEP, EIA STEO, Philly Fed SPF, Wall Street strategist aggregate), shown for comparison against your scenario.

Futures-implied curve

The market's risk-neutral forward expectation from futures prices — bundles consensus with risk premia and carry; shown distinctly from surveyed forecasts.

Condition flag

A binary scenario context — e.g. recession, stagflation, energy crisis — used to refine which historical episodes are eligible to match.

Historical analog forecasting and regime-switching models are respected analytical techniques, but they are not predictions. The future can break with the past; matched episodes are approximations; and the data carries revisions and gaps. Treat every projection as a structured way to reason about possibilities — not as investment advice. See the full disclaimer.