The GES Methodology

Step 1 - signal firing

Each signal is a standardized z-score. A signal "fires" when |z| ≥ 1.2, a threshold calibrated against eleven years of real signal history rather than assumed.

Step 2 - blend into an event probability

Five components, computed daily and blended by weight:

Each component is bucketed against its own empirical probability map, then combined by weight. Worked example: breadth 50%, persistence 100% (every day of the trailing month had a firing signal), cross-category spread 100% - blending to a probability of 56.5%.

Each fired event is separately scored for severity - Low, Medium, High, or Critical, from the magnitude of the underlying move against eleven years of historical norms for that regime. Severity doesn't gate the status ladder below; it travels downstream into the evidence-tier framing and the client factsheet.

Step 3 - regime status, with memory

Probability alone doesn't set status - each regime has its own thresholds and its own breadth/persistence confirmation, and status has memory rather than resetting daily:

Worked example: the regime escalated to Active on 30 Apr 2026 and, evaluated a month later, hasn't faded since - the memory rule is what keeps status Active through a full month of daily readings rather than resetting each day.

Step 1 - company-level score

Company GES = 100 × avg over 4 dimensions of (Exposure × Beta × Confidence × e⁻λ·age)

Each dimension comes from company fundamentals, real-filing-anchored to a named 10-K rather than a synthetic proxy; Beta sets how sensitive that dimension is to the live regime, confidence reflects data quality, and decay ages out stale filing data (λ=0.0067/day - a figure from a 45-day-old filing still carries ~74% weight, ~30% at 180 days). Four dimensions currently have filing-derived company-level data coverage; the remaining dimensions (cost, regulatory, financial, strategic) are withheld from company-level scoring until the required source data and extraction methodology are implemented, rather than filled with a synthetic placeholder:

Country risk isn't a separate dimension - it's carried inside Revenue and Supply Chain, which is where geographic concentration actually shows up in a company's numbers. In the worked example, NVDA scores 25.9 (Very High tier), anchored to a named 10-K across all four dimensions; AFL scores 16.77 (Very Low tier) on the same beta table, but falls back to a universe-average proxy rather than a named filing - the fallback chain in practice, not just described in the abstract.

Step 2 - sector aggregation and relevance

Company scores roll up to a sector median first. Sector vulnerability is then a separate, empirical measurement layered on top: for every past instance of this regime, we look at how each sector's price actually moved relative to the broader market, and bucket that historical relative impact into ±1 (clearly vulnerable/beneficiary), ±0.5 (moderately so), or 0 (no consistent relationship) - refreshed quarterly as more history accumulates. This is what turns "Technology has a high company GES" into "Technology has historically sold off under this regime, so that GES should count against the portfolio, not for it." Here: Financials, Industrials, Health Care, and Energy score vulnerable; Information Technology - the single largest sector at 31.5% of the book - scores beneficiary; Consumer Staples, Real Estate, Utilities, Consumer Discretionary, and Communication Services score neutral.

Step 3 - portfolio aggregation and direction

Portfolio GES = Σ (holding weight × company GES)

Vulnerable- and beneficiary-sector weight are summed directly; a 5% materiality floor on either side sets the direction label - Vulnerable, Beneficiary, Mixed, or Neutral. A single concentrated name can also trip it on its own: a top holding contributing ≥2.0 points to Portfolio GES counts as material even in a sector that's under 5% of the book, so a large, singularly exposed position can't hide inside an otherwise-diversified sector weight. Worked example: GES 24.28, vulnerable 29.0%, beneficiary 31.5% → Mixed.

Step 1 - GES band and base action

Portfolio GES is bucketed into five empirically-set bands (Very Low through Very High), calibrated separately per regime and per weighting scheme (cap- and equal-weighted portfolios have genuinely different achievable GES ranges, so they don't share one cutoff table). The full regime status ladder has seven values - Watch, Emerging, Active, Confirmed, Severe/Critical, Fading, Resolved; Fading and Resolved unwind whatever action was in place rather than opening a new one. Regime Status × GES Band looks up one of 7 × 5 = 35 pre-defined base actions internally.

That internal 35-cell matrix collapses to four client-facing actions, since that's what actually differs mechanically: Monitor (no trade), Reduce (bidirectional sector tilt), Hedge (the same tilt plus a hedge-ratio blend), and Rotate (full beneficiary rotation, tightly gated - see Step 2). Watch is always Monitor; Active + Very High is a Reduce.

Step 2 - action intensity

Intensity = Base Intensity(status) × (Portfolio GES / 100) × Client mandate multiplier

Base intensity is fixed per status (Active = 48%, Confirmed = 95%, Severe/Critical = 175% of the affected sleeve), then scaled by the portfolio's own GES. Recalibrated up from a far more conservative starting point after backtesting showed the earlier values under-executed so badly the system barely traded until a crisis had already been priced in.

Client mandate - the multiplier and the overlay

The same regime, portfolio, and base action produce four different outcomes depending on mandate:

Rotate's real gate is deliberately narrow: constraint = Regime Rotation Allowed, direction must be Beneficiary specifically(a Mixed book with some vulnerable exposure alongside doesn't qualify), status ≥ Active, and the historical evidence tier for that exact regime/action pairing must not be Weak or Insufficient. All three overlay rules were added after backtest evidence: the earlier Low Turnover reduce paid transaction cost for a 23% historical win rate; Rotation used to fire even at Very Low band (52% win rate there vs. 60–75% at every other band) and regardless of whether that regime/action combination had ever been seen enough times to trust - so the mandate with the most latitude doesn't also get the least discipline.

Step 3 - multi-regime overlay

Regimes rarely run alone. Every live regime is ranked and votes sector-by-sector on direction alongside the others:

Worked example: base intensity at Active status is 48% of the affected sleeve; against this portfolio's GES of 24.28 and the Balanced multiplier (1.00×), that's 48% × 0.2428 × 1.00 = 11.65% before the overlay. Six regimes are live; Earnings leads the priority ranking (45.17 vs. Geopolitical's own 42.98) even though Geopolitical is the one being evaluated here, but every live regime agrees on direction sector-by-sector (alignment 1.00, multiplier 1.00×) - so the overlay changes nothing this time, and intensity stays at 11.65%. The discount only bites when regimes actually disagree on a sector.

Step 1 - trim candidate scoring

Trim Score = Position weight % × Company GES × Vulnerability mult. × Liquidity mult.

Every holding in a vulnerable sector is scored, not just the largest name; the adjustment budget is split pro-rata by score share - a diversified trim, not a single stock call.

Step 2 - replacement scoring

Replacement Score = .35×GES benefit + .20×Sector fit + .10×Liquidity + .10×Confidence
Reserved, not yet active: Factor Similarity · Valuation

Candidates are drawn from the client's own uploaded universe, or the S&P 500 universe when none is supplied, excluding current holdings; ranked and pooled across every trimmed name. Factor similarity and a valuation penalty are reserved extensions in the methodology - no factor model or valuation-multiple pipeline exists yet, so they don't contribute to the score shown above until they do.

Step 3 - execution style by mandate

The same sizing executes differently by style: replace (sell + buy) for Balanced/Defensive, rotate (sell vulnerable, buy existing beneficiary holdings) for Regime Rotation, or a stand-down to Monitor below conviction.