Methodology and Change Record
How the portfolio signal is calculated, which evidence can affect allocation, and what changed when the active history was rebuilt.
How the current allocation is produced
- Start with the reference maximum-risk portfolio: 60% global stocks, 30% gold, and 10% bitcoin. It is illustrative, not a universal definition of investor risk.
- Classify the market regime: confirmations are assigned once to participation and risk appetite; credit and refinancing; rates, inflation pricing, and duration; or funding, the broad dollar, and financial conditions. Signals are normalized inside each domain, and each domain receives 25% of the combined score.
- Handle uncertainty explicitly: a tie, winner margin below 2 points, coverage below 70%, or stale required input produces Mixed with low confidence. The model holds its last confirmed top-down target; if none exists, it uses the defensive Deflation mapping.
- Apply VAMS by sleeve: multi-horizon volatility-adjusted momentum and five moving-average/trend votes classify stocks, gold, and bitcoin as Bullish, Neutral, or Bearish. Those states apply 100%, 50%, or 0% of the top-down sleeve target. RSI is display-only.
- Confirm changes asymmetrically: a lower total-risk target must persist for two consecutive closes. Adding total risk, or rotating between risk assets without lowering total risk, requires five consecutive closes. The confirmed allocation is actionable for the next trading session. Unused exposure goes to cash.
What affects allocation
| Role | Current contents | Effect |
|---|---|---|
| Allocation inputs | Four equal-weight market-regime domains—including one commercial-paper credit premium—plus each sleeve’s confirmed VAMS state. | Can change the top-down target or the 100% / 50% / 0% sleeve multiplier. |
| Confirmation and context | NFCI, UUP, the U.S. reserve-liquidity proxy, fast financial conditions, global-liquidity context, macro GRID, and Gavekal. | Explains or cross-checks the environment; does not directly change KISS allocation. |
| Diagnostics | Coverage, winner margin, confidence, domain agreement, sensitivity results, volatility, correlations, risk contribution, turnover, and drawdown. | Describes confidence, robustness, or portfolio behavior; does not set the target. |
| Display-only | RSI 14 and 12-month macro inflation rates. | Shown for interpretation; unused by the active allocation rule. |
Input and calculation conventions
- Prices: adjusted closes are used for investable ETFs and portfolio returns. Gold uses adjusted-total-return GLD before GLDM and GLDM afterward, chained on the first common trading session. Raw levels are used when the economic level is intended, including indexes, yields, spreads, volatility indexes, currencies, commodity-level proxies, bitcoin, and individual stocks.
- Rates and spreads: Treasury yields, breakevens, the commercial-paper credit premium, SOFR-IORB, and curve slopes use arithmetic basis-point changes. Level, direction, and acceleration remain separate features.
- Signal timing: a history row uses observations available at that date’s close. The performance engine executes a confirmed allocation on the next session with 5 basis points of transaction costs and 5 basis points of slippage.
- Cash: residual cash earns the FRED 3-month Treasury constant-maturity yield, accrued over calendar days. Taxes are excluded.
- Macro vintages: GRID uses revised FRED history and is context-only. It cannot enter an allocation backtest unless point-in-time ALFRED vintages are first fetched and stored.
- Gavekal timing: only completed monthly observations classify the active quadrant. The boom/bust axis uses raw FRED S&P 500 / WTI spot. The inflation axis uses adjusted IEF total return / raw GLD in the published Treasury-return-over-gold orientation; IEF remains a public investable approximation of a proprietary constant-duration Treasury total-return index. Partial current-month readings are context-only.
Sources, cadence, and proxy boundaries
Market prices come from public Yahoo Finance chart data. Macro, rates, credit, the broad dollar, and reserve-plumbing series come from FRED and retain their published daily, weekly, monthly, or quarterly cadence and observation dates. The full KISS snapshot, liquidity snapshot, and Gavekal snapshot publish the exact source, cadence, unit, observation date, coverage, and methodology notes used by the current build.
The credit allocation input is the spread between the Federal Reserve Board’s 30-day A2/P2 and AA nonfinancial commercial-paper rates, published through FRED as RIFSPPNA2P2D30NB and RIFSPPNAAD30NB. The model measures its 63-calendar-day arithmetic change: widening confirms risk-off conditions and narrowing confirms risk-on conditions. It is public-domain, same-maturity, daily, and reproducible back through the reconstruction warmup. It is a short-term refinancing-stress proxy—not option-adjusted bond spread, duration-matched corporate credit, or a full high-yield-market measure.
The gold series uses GLD adjusted closes before the first common GLD/GLDM session, then rescales GLDM to GLD on that overlapping session. The overlap date, source symbols, scale factor, and observed transition return are published in the KISS methodology and rebuild report. This preserves total-return continuity but does not make the two funds economically identical; their fees, liquidity, and tracking can differ.
RSP/SPY, small caps, cyclicals, high beta, sector ETFs, and country/region ETFs are investable participation proxies. They are not constituent-level advance/decline, new-high/new-low, or percentage-above-moving-average breadth. NFCI remains context-only because treating a composite financial-conditions index as another vote would double count credit, funding, volatility, rates, and equity conditions already represented elsewhere.
Methodology changelog
The confirmation-policy revision is kiss-2.1.0 because it changes when allocations become actionable and when the performance simulation trades. It retains the kiss-2.0.0 signal inputs, credit proxy, gold chain, VAMS formula, and 2020 reconstruction start. The 2-out/5-in rule was selected from a prespecified neighborhood after reviewing multiple market windows and both portfolio variants; it was not selected solely for the highest full-period return.
| Confirmation area | kiss-2.0.0 | kiss-2.1.0 |
|---|---|---|
| Risk reductions | Two consecutive closes | Two consecutive closes |
| Risk increases | Two consecutive closes | Five consecutive closes |
| Equal-total-risk rotations | Two consecutive closes | Five consecutive closes |
| Execution | Next trading session after confirmation | Next trading session after confirmation |
The July 2026 long-history revision was kiss-2.0.0 because it materially changed the credit input, gold history construction, and active sample. Material input, transformation, classification, allocation, confirmation/execution, or return-semantics changes require another version increment; copy-only and diagnostic-only changes do not.
| Area | kiss-1.9.0 | kiss-2.0.0 |
|---|---|---|
| Identity and range | kiss-1.9.0 history beginning in 2025 | kiss-2.0.0 reconstruction from the first trading session of 2020, with inputs fetched from 2016 for warmup |
| Credit | ICE BofA HY and IG OAS inputs without usable pre-July-2023 provider history | One public-domain 30-day A2/P2-minus-AA nonfinancial commercial-paper premium; a same-maturity refinancing-stress proxy with no redundant credit vote |
| Gold | GLDM alone constrained the common sample | Adjusted-total-return GLD before GLDM, chained on their first overlapping trading session so the handoff adds no artificial return or price jump |
| Availability | A recent observation could make a source appear historically available | Every signal is date-aware and requires its full transformation lookback; unavailable history stays unavailable rather than becoming neutral |
| History and performance | A short recalculated history described broadly as archived signals | Retrospective reconstruction is labeled separately from the archived-record period; both portfolio variants share one methodology and next-session execution |
Known limitations
- The active common portfolio sample begins on the first available trading session of 2020. Bitcoin uses its public spot-price history, while investable ETFs use adjusted returns; the comparison remains a model simulation rather than an implementable account record.
- Public data can be revised, delayed, missing, or restated. Macro GRID is deliberately excluded from allocation because its historical releases are not yet stored point in time.
- Participation measures are investable proxies, not full constituent breadth. Quarterly global-liquidity data are slow context, not a timing signal.
- Performance is a current-methodology model simulation. The pre-2025 portion is a retrospective reconstruction; the later portion is a recalculation in the archived-record period. Neither is audited live performance, and both exclude taxes, account constraints, market impact, and investor-specific implementation.
- The VAMS formula was retained because a simpler candidate was not clearly more robust across walk-forward windows, parameter neighborhoods, turnover, drawdown, and regime coverage. It was not selected to maximize the displayed historical result.
- The asymmetric confirmation policy is supported by retrospective, in-sample evidence. It reduces turnover and delays risk additions, but it can lag sharp recoveries and does not guarantee better future returns or drawdowns.