Personal quantitative research / ETF allocation experiment
Loading research snapshot…
PROSPECTIVE OPERATIONS

A frozen strategy. A separate forward record.

Export operations snapshot

Loading account snapshot…

Shares and cash

Observed execution costs

Modeled paper fills, imported broker-paper fills and real executions are separate evidence categories. No broker is connected.

Next actions and operating rules

The historical V2 report below remains unchanged and quarantined. Two 2008 EEM exceptions have a documented disposition, not invented replacement prices. Forward inputs use a separately audited 400-session window.

Download historical exception disposition · Scheduled runs (private GitHub)
DECISION BRIEF

Validate the edge before taking risk.

A 20% drawdown research gate. Failed results remain visible, and every simulation has traceable execution assumptions.

Export research record ↓
STATUS

Research only · Not cleared for live trading

Verified forward observations0 / 126 trading sessions
PREDECLARED HYPOTHESIS

Ensemble trend + portfolio risk budget

63 / 126 / 252-session trend votes relative to short-term Treasuries. Monthly reviews with a 2% rebalance band. Covariance estimates set portfolio risk; the remaining allocation goes to BIL.

10% volatility budget40% trend target cap per risk assetSimulated next-close fills

Returns in the context of risk

Historical portfolio value

● Selected strategy● Benchmark

Drawdown from peak

Includes initial capital · No guaranteed loss ceiling

All results, in one record

StrategyCAGRExcess Sharpe¹Max drawdownVolatilityAnnual turnover

¹ Calculated from daily returns less contemporaneous BIL total returns. Strategies differ in risk and invested exposure; higher returns alone do not imply a better strategy. All historical results are retrospective, not untouched out-of-sample evidence.

RESEARCH → REVIEW → FORWARD EVIDENCE

A profitable backtest is only a starting point.

Historical pass ≠ Live trading authorization

Candidate validation gates

Applies to the predeclared trend candidate and its core allocation variant.

Stress the execution assumptions

How certain is the edge over 60/40?

Paired 21-session block bootstrap with 1,000 resamples: an interval for annualized arithmetic active return. An interval spanning zero does not establish an edge. This interval does not correct for all historical research choices.

Performance across market eras

Predeclared eraCAGRMax drawdownVolatilitySessions

Era analysis is a robustness diagnostic. Previously observed history is not genuine forward evidence.

PORTFOLIO STATE

Track signals and holdings separately.

Latest target allocation

Historical model targets, not live orders. Dollar amounts scale proportionally; changing capital does not rerun market impact, taxes or fill constraints.

What drives the trend allocation?

Votes compare each asset’s price trend against BIL. A positive signal does not guarantee a gain next month. Simulated holdings drift from targets because of returns, fees and unfilled orders.

EXECUTION LEDGER

Execution constraints affect returns.

Fund purchases with available cash

Mark holdings to market, sell, deduct costs, then scale purchases to available cash. No borrowing or negative cash balances.

Participation limits

Fills in each asset are capped at 0.5% of its preceding 20-session average dollar volume. Unfilled amounts are canceled and recorded; liquidity is finite.

Risk pauses require manual review

Latest 15 simulated order records

Signal / fill dateAssetRequestedSimulated fillFeesUnfilled

Positive amounts are buys; negative amounts are sells. Accounting uses adjusted total-return dollar balances, not broker-ready share quantities or actual fill confirmations.

DATA CONTRACT

Data exceptions remain on the record.

Download audit & source hashes ↗

Market-data records awaiting verification

Two large EEM gains in 2008 retain their original flags. Issuer NAV is a different valuation and cannot verify exchange closing prices. The default pipeline rejects unresolved records; diagnostic mode permits inspection but keeps the data gate blocked.

What changed in V2

  • Holdings drift naturally; monthly reviews and a rebalance band reduce unnecessary turnover.
  • Covariance-based risk budgeting replaces a simple sum of individual asset volatilities.
  • Defensive allocations use BIL’s actual historical returns; uninvested cash earns zero.
  • History includes the 2008 crisis, with exceptions retained.
  • Capital, fill amounts, fees and unfilled quantities are reconciled for each order.
  • The original underperforming ML experiment remains in V1 for inspection.

Limitations