# Apex++GAF-DC2

> Machine-readable companion to https://alphy.live/ . The human page (https://alphy.live/) is canonical; if this file and the page disagree, the page wins. Spec as of 2026-10-11. Backtest numbers are in-sample (rules chosen on the same data) unless marked walk-forward. No prices, no account amounts.

## Summary
- Strategy: momentum rotation in S&P 500 and Nasdaq-100 stocks with accelerating revenue, plus a gated 3x Nasdaq-100 ETF (TQQQ) sleeve, in a US margin account.
- Backtest period: 2010-02-11 (first TQQQ session) to 2026-09-18.
- Headline (concentration-stress margin model): CAGR 97.8%, MaxDD -59.2% (2011-02-14 to 2011-11-25), 1 margin breach (daily low 2021-03-05).
- Robinhood-style margin: CAGR 93.8%, MaxDD -57.3% (2024-12-24 to 2025-04-08), 0 margin breaches.
- Walk-forward (rules re-chosen each January using only earlier years, 2014-2025): 77.0% a year; current rules held fixed: 80.2% (in-sample).

## Universe and selection
- universe: point-in-time S&P 500 or Nasdaq-100 members on the decision date, including later-delisted names; one share class per company (the higher-ranked class). No market-cap screen.
- trend filter: close > SMA200 and SMA200 > SMA200 twenty sessions earlier.
- growth filter (growth_accel > 0): growth(q) = revenue(q)/revenue(q-4) - 1; growth_accel(q) = growth(q) - growth(q-1); quarterly figures as originally reported, usable from the first trading session after the filing date, carried for at most 315 sessions.
- cash filter: cash/assets above the 50th percentile of reports filed in the prior 90 calendar days by companies already in the universe at filing.
- distress filter: (debt/assets - net income/assets - current ratio) below the 90th percentile of the same pool.
- rank: EMA50 / EMA200 of the close, highest first, eligible names only.
- picks: top_n = 4 per sleeve; buffer = 7 (a held name is kept while eligible and ranked 7 or better); empty slots filled by the highest-ranked eligible names not held.

## Weighting and leverage
- weights: w_i = cap_i^3 / sum(cap^3) over the picks (weighting power 3, cubed; 2 before 2026-10-11); cap = daily point-in-time company market cap (all share classes). Market cap is used only for weighting, never for selection.
- stock vol target: 0.70 annualized; basket_vol = sample stdev (ddof 1) of the last 20 daily returns of sum(w_i * r_i) x sqrt(252); scale = clip(0.70 / basket_vol, 0, 1.5).
- sleeve target_i = (equity at the open / 4) x w_i x scale.
- gross at a rebalance: at most 2.0x equity (1.5 stocks + 0.5 TQQQ); positions drift between rebalances.

## Schedule
- stock book: 4 sleeves of 1/4 equity; sleeve k (0..3) rebalances on sessions 5k, 5k+20, 5k+40, ... counted from 2010-02-11; decisions on the prior close.
- TQQQ sleeve: k = 0.5 x equity while the gate is on; resized to 0.5 on the 11th session of each month.
- TQQQ gate: qqq_vol = sample stdev (ddof 1) of the last 20 daily QQQ total returns x sqrt(252) on the prior close; threshold 30%, hysteresis 2% (turns OFF above 32%: sell all TQQQ at the next session; turns ON below 28%: buy to 0.5x equity at the next session; keeps its state between 28% and 32%). Gate starts ON.
- delisting: a held stock whose series ends is cashed at its last close if the delisting is an acquisition or merger; any other cause counts as 0.

## Each trading day (live)
- ~09:00 ET: update data to the prior close and validate; compute the TQQQ gate and today's sleeve rebalances; read the account; build the order list before the 09:30 open. Stale or invalid data, an order list not ready by 09:30, or any alert: no trading that day.
- 09:30-10:00 ET: buys and sells are worked during the first 30 minutes to beat the opening price, within fixed price and size limits (price band 3%, buy notional cap 1.02x planned); sells before buys; remainder completed at 10:00.
- ~10:30 ET: fills compared with the official open; trade log updated.
- A missed day does not change the rules; the next run works from the account as it stands.
- Backtest assumption: all trades at the official open.

## Costs and margin (backtest)
- trading cost: 5 bps per side on traded notional; no taxes.
- borrowing: fed funds plus a tiered spread (1.50% / 1.00% / 0.75% by loan size), 360-day year; cash credit fed funds - 0.50% above a threshold.
- headline margin model: maintenance = max(0.25 x stock value + 0.75 x TQQQ value, div(n) x sum_i stress_i x stock_value_i); div(n) = 1.00 (n<=2), 0.95 (3-5), 0.75 (6-9), 0.60 (10+); stress_i = min(1, max(0.30, 0.30 + 0.065 x max(0, ret1y_i - 0.50), 0.50 x vol30_i)).
- Robinhood-style model: 25% maintenance on stocks, 75% on TQQQ, 50% initial; tiered borrowing at the fed funds upper bound plus 1.25% / 1.05% / 0.75%.
- maintenance checked at every close and at the daily lows; a deficit is a breach: sell all positions pro rata at the next open until equity >= 1.10 x maintenance.

## Data and testing protocol
- Data: survivorship-bias-free, point-in-time prices (split and dividend adjusted, total return), index membership, company market cap, and quarterly fundamentals as originally reported; effective fed funds rate.
- Period: post-2010 only (TQQQ inception; the post-2008 rule and rate regime). Pre-2010 used only to check the TQQQ gate on a simulated 3x series.
- Filings used only from the first trading session after filing.
- Delistings in the backtest: five held names, all acquisitions, cashed at their last close.
- Train/check: rules tuned on 2010-2021, checked on 2022+ (CAGR since 2022-01-03: 115.9% headline, 111.5% Robinhood-style; fresh account from 2022-01-03: 115.0% / 111.1%). Rank buffer 7, vol target 0.70 and weighting power 3 were later chosen on the full history, so 2022+ is not a clean test.
- Rolling windows, path slices (151 four-year windows, monthly starts 2010-03 to 2022-09): headline median 85.0%, 10th pct 61.1%, worst +40.3%; Robinhood-style median 81.7%, 10th pct 57.2%, worst +38.5%.
- Rolling windows, fresh accounts (same 151 starts): headline median 84.0%, 10th pct 60.4%, worst +39.2%, 47 of 151 windows with at least one margin breach; Robinhood-style median 81.6%, 10th pct 56.5%, worst +37.3%, 0 windows with a breach.
- Multiple testing: about 380 variants tested in the latest research rounds; 377 rejected. Deflated Sharpe ratio across trials: no new signal passed. Adopted: rank buffer 7 (holding rule), vol target 0.70 (leverage setting) and weighting power 3 (weighting setting; did not pass the deflated test either; supported by a smooth monotone neighbourhood of powers and the walk-forward).

## Walk-forward (January, 2014-2025, fresh one-year accounts, Robinhood-style margin)
| Rule sets compared | Picked | Compounded yearly return |
|---|---|---|
| Apex++GAF (no cash/distress filters, buffer 12), fixed | - | 62.7% |
| Apex++GAF vs DC2 buffer 12 | DC2 12 of 12 | 68.7% |
| + DC2 buffer 7 (vol target 0.60) | buffer 7 in 2014, 2019-2025 | 69.6% |
| + DC2 buffer 7, vol target 0.70 | 0.70 in 2019-2025 | 72.5% |
| + DC2 buffer 7, vol target 0.70, cubed weights (all five) | cubed in 2014, 2019-2025 | 77.0% |
| DC2 buffer 7: vol target 0.60 vs 0.70 | 0.70 in 11 of 12 | 75.5% |
| DC2 buffer 7, vol target 0.70: squared vs cubed weights | cubed 12 of 12 | 80.2% |
| Current rules, fixed (in-sample) | - | 80.2% |
| Current rules without the TQQQ sleeve, fixed | - | 61.9% |
| No TQQQ sleeve vs gated sleeve | gated 12 of 12 | 80.2% |
Rows without a vol target use 0.60 (buffer 12 rows: 0.60 for DC2, 0.50 for Apex++GAF). Rows without cubed weights use squared weights for DC2 and power 1.5 for Apex++GAF.

## TQQQ gate: four-year folds (fresh account per fold; current rules; Robinhood-style margin)
| Fold | No sleeve | Gated 30% +/- 2 | Always on |
|---|---|---|---|
| 2007-2010 | 19.3% | 29.1% | 19.0% |
| 2011-2014 | 53.3% | 68.5% | 72.5% |
| 2015-2018 | 33.6% | 44.7% | 40.0% |
| 2019-2022 | 72.0% | 91.7% | 83.3% |
| 2023-2026-09 | 156.8% | 195.4% | 212.3% |
Chosen on training-period CAGR: gate picked in 5 of 5 folds (training from 1999) and 4 of 4 (training from 2010-02-11). Before 2010-02-11 TQQQ is a simulated 3x daily series.

## TQQQ gate threshold neighbourhood (full period CAGR, Robinhood-style margin; 0 breaches everywhere)
| Threshold | hyst 0 | hyst 2 | hyst 4 |
|---|---|---|---|
| 26% | 94.5% | 97.3% | 95.9% |
| 28% | 93.4% | 96.0% | 95.8% |
| 30% | 94.9% | 93.8% (live) | 93.6% |
| 32% | 94.3% | 92.5% | 93.2% |
| 34% | 94.5% | 93.8% | 93.1% |

## Sensitivity (independent implementation, headline model, base 98.4%)
- trading costs 10 bps per side: 97.0%; 25 bps: 92.9%
- all borrowing at fed funds + 3.00%, no interest on cash: 95.0%
- NVDA excluded: 86.6%; five largest contributors (NVDA, MU, PLTR, TSLA, MRNA) excluded: 48.0%
- half target exposure: 51.2% (MaxDD -32.7%)
- vol target: each 0.05 step adds about 2 points of CAGR and about 1-2 points of drawdown (smooth).
- rank buffer: not smooth; buffers 6 and 7 are about 5 points above neighbours (full-period CAGR).
- weighting power: smooth and monotone; 2.5 / 3 / 4 add about 3 / 5 / 7 points of CAGR over squared (Robinhood-style margin); most of the gain comes from five years (2013, 2020, 2021, 2024, 2026).
- concentration: the five names above supply 55% of the stock book's positive log-return contribution; peak single-name weight 193% of equity (AMZN, 2015; 184% with squared weights).

## Calendar-year returns (headline model)
2010 (from 02-11) +144%; 2011 -35%; 2012 +123%; 2013 +404%; 2014 +33%; 2015 +89%; 2016 +7%; 2017 +170%; 2018 -11%; 2019 +86%; 2020 +348%; 2021 +138%; 2022 -40%; 2023 +114%; 2024 +439%; 2025 +44%; 2026 (to 09-18) +267%.

## Risks
- Drawdowns of -59% (backtest) and -40% (2022); expect deeper.
- Leverage up to 2x at rebalances, more after drift; gaps can force sales.
- Single-name concentration above 100% of equity; cubed weighting puts more in the largest pick than squared did.
- TQQQ daily reset: choppy declines while the gate is on.
- One regime (2010+); overfitting; non-acquisition delisting is a total loss.

## Rule changelog
- 2026-10-11: weighting power raised from squared to cubed (market cap^3); backtest CAGR 93.1% -> 97.8% (Robinhood-style margin 88.9% -> 93.8%).
- 2026-10-11: stock volatility target 60% -> 70%.
- 2026-10-10: rank buffer 12 to 7.
- 2026-10-10: delisting rule cause-aware (acquisitions cashed at the last close, other delistings a total loss).
- 2026-10-10: cash and distress filters compare each stock only with filings available at the time.

## Links
- Human page (canonical): https://alphy.live/
- Trade log: https://alphy.live/trades/