One mechanical execution on the first trading day of each month, under rules fixed
in advance — no discretion, ever. Refreshed every trading day · data through ·
how the strategy was chosen: research archive.
By default the page shows a full-history dry run of the same plan spec, monthly ledger included;
drag the slider to the far right (2026-07, the actual start) for the live record.
The plan
Spec: total budget $1,000,000, first decision day 2026-07-01.
The first month deploys $200,000; from 2026-08, $13,300 goes in on the first trading
day of each month for 60 months through 2031-07 (the $2,000 remainder folds into the final month,
so contributions total exactly $1,000,000). After that, contributions stop and the strategy runs
on its own. Undeployed budget sits in an off-portfolio cash pool compounding at the 3-month T-bill rate.
One row per decision day (normal DCA months included), newest first. “Branch amount” is
the cash that month’s branch actually moved: deep bottom = total TQQQ purchase; mild bottom / normal =
reserve deployed into QQQ; crash / overheat = sleeve proceeds; expensive = cash banked into the reserve.
Cap trims are badged separately.
The rules — all fixed in advance, no discretion
Three signals (first trading day of each month, using data through the prior close — no lookahead):
Cheap valuation: Shiller CAPE (S&P 500 proxy) below the 25th percentile of its rolling 30-year window
Deep drawdown: NDX (5-day avg) more than 20% below its rolling 2-year high
Panic: 5-day average VIX above 40
Six-branch decision tree (n = signals lit; first match from the top wins):
Condition
Action
n ≥ 2 (deep bottom)
Contribution + the full reserve buys the TQQQ sleeve
n = 1 (mild bottom)
Contribution buys QQQ, plus ⅓ of the reserve
25-day crash > 12%
Sell half the sleeve into the reserve; contribution buys QQQ
Expensive (CAPE pct > 70% and near the 2-year high)
Trim the sleeve 1/12 into the reserve; contribution buys QQQ
Otherwise
Normal DCA: contribution buys QQQ + 1/6 of the reserve drips into QQQ
Two hard caps: reserve ≤ 30% of the portfolio
(overflow force-buys QQQ); TQQQ sleeve ≤ 40% (checked every 6 months; excess trimmed into QQQ).
Full design rationale, five-start backtests and robustness checks:
research archive.
The honest caveats
The backtest’s excess return comes almost entirely from one crisis — the deep-bottom detector
fired only from 2008-11 to 2009-04 across 2000–2026. Effectively n = 1.
On this plan’s basis (61 contribution months + cash pool counted in total wealth), the same
spec run from the 2000 / 2005 / 2010 / 2015 / 2020 starts ends at
1.04× / 3.30× / 1.00× / 0.92× / 0.99× vs same-cash-flow QQQ — crisis-free windows slightly lose.
Once contributions end, the “expensive” branch has no new cash to bank and the drip drains the
reserve (by the 2008 bottom, the 2000 start’s reserve was 0.15% of the portfolio) — a crisis that
arrives too late can’t be caught. The plan’s crisis-alpha window is roughly 2026–2033.
The Calmar edge is threshold-sensitive: loosening −20% to −18% catches April 2020 but rides
TQQQ into the 2022 bear — Calmar falls from 0.68 to 0.58, below DCA QQQ’s 0.61.
Exit rules are weaker than entry rules: in a multi-year grind, n ≥ 1 locks out both sell
branches (“crash”, “overheated”), so the sleeve just rides it down — exactly the 2000–2002 shape.
A known, unresolved design tension.
Curves and max drawdown here are monthly (one point per decision day); true intraday and
interday drawdowns run deeper.
A backtest is not the future. This page is the public tracking of personal research —
not investment advice. Leveraged-ETF drawdowns are real.