ndxlab
Rules-based investing · public research model

One trend rule. Zero discretion.

10-Month SMA: if the signal is on at your start date—or turns on later—enter at your chosen QQQ / T-bill mix and hold; when it turns off, exit to 0% QQQ / 100% T-bills. The slider sets risk preference at entry, not a monthly rebalance target. Data through , refreshed daily.

Long-history check · broad U.S. equities 1963-05 to 2026-06 · 758 months · 100% equities above the 10-month average, 100% T-bills below it · next-month execution · 10bp turnover cost.
Max drawdown
50.31% → 24.48%
buy & hold → trend rule
Annualized return
10.85% → 10.04%
lower return, much lower drawdown
Calmar ratio
0.216 → 0.410
return per unit of max drawdown
Explore a scenario

Hypothetical capital, start & entry mix

$

Public market signal

Current model action

The outcome

Strategy return vs QQQ

10-Month SMA 100% QQQ benchmark

Every modeled move

Simulated monthly ledger

Decision dayActionTradeCostQQQCashTotal wealth

One row per decision day, newest first.

How the rule worksone monthly signal, one entry preference, zero rebalancing while on

10-Month SMA: compare the previous completed month's dividend-adjusted QQQ close with its trailing 10-month simple average. When the signal switches on, buy at the selected QQQ / T-bill mix on the next month's first trading day. While it stays on, make no trade and let both weights drift. When it switches off, sell QQQ and hold 100% 3-month T-bills.

The reference entry mix is 75% QQQ / 25% T-bills. The slider is the trader's risk preference for initial entry and re-entry; it is neither a maintained cap nor a monthly target. The risk-off allocation remains 0% QQQ / 100% T-bills. Trades are charged 10 basis points per 100% of portfolio turnover. No forecasting optimization is applied to the slider setting.

Why the model specification changed (2026-07): deploying the whole hypothetical capital at once beat spreading it over 60 months in about 4 of 5 historical windows, and at matched average exposure a mechanical cash buffer protected the worst cases better than the signal-driven reserve it replaces — the signals fired at a real bottom exactly once in 27 years (2008-11 … 2009-04) and spent the reserve mid-crash in 2000–2002.

The honest caveatswhat this model does not protect
  • Single-index concentration is the unhedged tail. The equity side is 100% Nasdaq-100. A Japan-1989-style multi-decade stagnation is not fixed by rebalancing, by schedules, or by signals — only by diversification, which this model deliberately does not do. That concentration is an explicit research assumption.
  • Risk control has a return cost. In the 1963–2026 broad-market check, the trend rule reduced annualized return from 10.85% to 10.04% and terminal wealth from 668.8× to 421.0×. It nearly doubled the Calmar ratio, not terminal wealth.
  • The website and the long-history check use different equity series. The research panel uses a broad U.S. equity total-return proxy. The interactive scenario uses QQQ from 2000 onward and defaults to a 75/25 entry mix that can drift. Its full available history produced a −27.6% monthly drawdown versus −79.6% for 100% QQQ, with terminal wealth about 13% lower. Neither path establishes a future outcome.
  • Turnover and tax matter. The simulations include 10bp per 100% of portfolio turnover, while taxes, bid-ask spread, fund fees and execution delay remain unmodeled. The broad-market trend rule turned over about 146% of portfolio value per year on average.
  • The trend rule can whipsaw. The 8–12 month stability study found similar historical drawdown control, while the best-returning lookback changed across eras. Ten months is a fixed research choice, not a uniquely optimal parameter.
  • Monthly granularity. Curves and drawdowns use one point per decision day; true intraday drawdowns run deeper.
  • Specification history. The original three-signal model, TQQQ sleeve, 60-month deployment schedule and monthly Constant-Mix variant were retired after independent re-tests. The current page uses 75/25 only as the default entry preference and applies the 10-month trend exit studied on 1963–2026 broad-market data. Every version remains in the repository history.
  • A backtest is not the future. This is public quantitative research. All capital, holdings and trades shown are hypothetical — not investment advice.