Thesis for critique: SMA leader rotation plus a margin funded TQQQ sleeve

Status: research note for adversarial review. It is not a deployment instruction and it is not financial advice.

Claim

A simple cross sectional momentum rule can be turned into a serious challenger to buy and hold TQQQ if three pieces are combined carefully:

1. Hold the strongest large cap leaders, not the index.
2. Use only enough leverage to lift the compounding rate, not enough to make the broker part of the strategy.
3. Judge the strategy by fresh start four year windows and by behavior at the drawdown depths, not by a single full window headline.

The strategy is not presented as a low risk product. It is a high drawdown compounding machine. The open question is whether the edge is real enough, and the leverage sizing disciplined enough, for the machine to stay intact when it is down 60 to 80 percent.

Strategy definition

Universe: point in time union of the S&P 500 and Nasdaq 100. No market cap ranking or filter is applied after index membership.

Eligibility at each monthly decision date:

• The stock is an index member at that date.
• Adjusted close is above its 200 day simple moving average.
• In the slope gated form, the 200 day average is higher than it was 20 trading sessions earlier.

Ranking: eligible stocks are ranked by SMA50 divided by SMA200, highest first.

Portfolio: hold five names at equal target weight.

Buffer rule: an incumbent is kept while it remains eligible and ranked inside the top 15. Open slots are filled from the best ranked eligible names not already held.

Timing: decisions use the prior close. Execution is at the next scheduled rebalance open. The preferred schedule in the current tests is offset 10, meaning the 11th trading day of the month. Costs are 5 bps per side.

Leverage forms tested:

• Rotation only at fixed gross exposure, reset monthly.
• Hybrid: rotation names at 1.0 times equity, funded with cash, plus a TQQQ sleeve bought with margin debt.
• Allocation frontier: rotation leverage a and TQQQ leverage b under Robinhood style margin constraints.

Robinhood margin assumptions used in the current model:

• Rotation names: 50 percent initial margin, 25 percent maintenance margin in the engine. Screenshots from the account showed MU and SNDK at 30 percent maintenance, with STX, WDC, and DELL at 25 percent. That per name refinement has not been rerun yet.
• TQQQ: 75 percent initial margin and 75 percent maintenance margin.

Margin borrowing cost in the tests: 6 percent annual. Initial margin 50 percent and maintenance margin 25 percent are used only where the per instrument model does not override them.

Evidence base

All main results below come from backtests over 2010-03-01 to 2026-09-18 unless stated otherwise. Prices end 2026-09-18 in the cached data. Rolling tests use 151 fresh start four year windows beginning monthly from 2010-03 through 2022-09. Each rolling window starts with new cash, new holdings, and a reset leverage state, while indicators keep legitimate pre window price history.

Unlevered rotation base

Read: the unlevered rotation is a credible standalone equity strategy against QQQ. It does not beat TQQQ on full window CAGR. Its value is that it reaches high compounding with much less benchmark leverage and a different return path.

Important caveat: the pre 2020 decade was much weaker. Earlier work found 2010 to 2019 CAGR around 12.7 percent for the strict form, behind QQQ. Much of the full window edge is concentrated in 2020 onward and in the recent momentum run.

Slope gate

Adding a rising SMA200 requirement improved the unlevered selection in the tests:

• Offset 0 slope gated Buffer15: about 30.7 percent CAGR, Sharpe about 0.74, max drawdown about -40.0 percent.
• The slope gate improved return and Sharpe in the tested sample. It did not reduce max drawdown.

Open issue: the gate may be a real trend quality filter, or it may be another fitted condition on the same sample.

Rebalance phase

Offset means trading days after the first trading day of the month. Offset 0 is the first trading day. Offset 10 is the 11th trading day.

Unlevered full window phase results:

Offset 0 was not the best phase for either selection, so the verified month start numbers were not cherry picked high. Offset 10 won in both sample halves for both selections. But offset 10 was still selected after looking at the same history, so its full headline should not be treated as an unbiased expectation.

Rolling four year check for slope gated Buffer15 at fixed 1.5 times margin:

The honest expected CAGR used for this form is 33.3 percent, the middle of the offset 0 and offset 10 rolling means. It is below the 41.1 percent full window headline on purpose.

Offset 10 beat offset 0 in only 57.6 percent of rolling windows for this slope gated levered form. The phase edge is therefore useful but fragile.

Hybrid: rotation on cash plus TQQQ on margin

Structure: hold the five rotation names at 1.0 times equity. Then borrow on margin to buy TQQQ at k times equity. Rebalance monthly at offset 10.

Under Robinhood's 75 percent initial and 75 percent maintenance requirement for TQQQ, the sleeve is capped by initial margin. With rotation at 1.0 times equity, the maximum feasible TQQQ sleeve is:

(1.00 - 0.50) / 0.75 = 0.667 times equity.

So requested k values of 0.75 and 1.00 collapse to the same effective 0.667 times TQQQ sleeve.

Rolling four year results under the 75 / 75 TQQQ rule:

Benchmark for the same rolling design: TQQQ buy and hold had a 40.3 percent mean four year CAGR, 44.7 percent median, +2.0 percent worst window, and -81.7 percent worst drawdown in the rolling lab.

The k = 0.5 hybrid is the cleanest aggressive form in the current tests. It beats TQQQ on rolling mean, beats TQQQ in 57.6 percent of windows, has a better worst window than TQQQ, and never breaches in the tested history.

The capped maximum sleeve earns more on average and wins more often, but it needs forced sales in about half of four year windows. Whether that is acceptable depends on execution discipline at the depths, not on the average.

Allocation frontier under Robinhood margin

The frontier varies rotation leverage a and TQQQ leverage b subject to 0.50 a + 0.75 b <= 1.00 at each rebalance.

Selected points:

Conclusion from the frontier: nothing dominates the 1.00 rotation plus 0.50 TQQQ point. It is the best zero breach point. Above it, each step buys a small amount of extra rolling CAGR and pays with materially more forced sales. Levering the rotation alone is less efficient than mixing in a TQQQ sleeve.

The actual thesis

The core claim is not that leverage is safe. The claim is narrower:

1. The leader rotation has a real standalone edge over QQQ in the tested sample.
2. The edge is strong enough that a moderate amount of well placed leverage can lift it into TQQQ beating territory on rolling windows.
3. The leverage should be sized at the frontier knee, where the account still compounds through the worst tested depths without forced sales.
4. Margin calls are not the main event. The main event is whether the account is still alive, liquid, and compounding after a 60 to 80 percent drawdown.

Under that framing, the current candidate is:

• Slope gated Buffer15 leader rotation at 1.0 times equity.
• TQQQ sleeve at 0.50 times equity, funded by margin.
• Monthly rebalance at offset 10.
• Self managed deleveraging before the broker is forced to act.

Honest expectation language: use the rolling numbers, not the full window headlines. For the hybrid at k = 0.50, the rolling four year mean is 43.7 percent. That is an in sample research estimate over a favorable sample, not a promise.

Known weaknesses and unresolved risks

1. Sample regime risk. The data begins in 2010. It does not include the 2000 to 2002 Nasdaq collapse or the 2009 momentum reversal. A strategy built on leaders may behave worse when leadership breaks violently.

1. Selection bias. The base rule, buffer, slope gate, offset, and leverage points were all examined on overlapping history. There is no untouched holdout yet.

1. Phase risk. Offset 10 was selected after seeing it win. Rolling tests soften the concern, but the slope gated levered form only beat offset 0 in 57.6 percent of windows.

1. Concentration. Five names is concentrated. The current example holdings from the latest cached decision date were DELL, SNDK, MU, STX, and WDC. That is a storage and memory heavy basket, not a diversified portfolio.

1. Broker discretion. The model assumes forced sales can be executed at the next open in an orderly way. A real broker can liquidate intraday, change house margin, or impose concentration charges. Self management reduces this risk but does not remove gap risk.

1. Per name margin refinement. The engine used 25 percent maintenance for all rotation names. The account screenshots showed 30 percent for MU and SNDK. This has not been rerun.

1. Interest and financing. The model charges 6 percent on margin debt. Real rates, specials, corporate actions, and account specific terms can differ.

1. Tax and turnover. Turnover is high by passive standards. The current tests charge 5 bps per side but do not model taxes. This may or may not matter depending on the account and loss carryforwards.

1. TQQQ product risk. TQQQ has daily reset leverage inside the product and broker margin outside it. The hybrid stacks product leverage on top of account leverage. The frontier treats that stack explicitly, but tail behavior in a true Nasdaq crash is still model dependent.

1. Behavior at the depths. The historical rolling windows say the account survives. They do not say the operator will keep following the rules when the account is down 70 percent and the monthly rebalance says buy more weakness.

Questions for critics

Please attack the weakest joints first.

1. Is SMA50 / SMA200 a defensible momentum ranking, or is it just a noisy proxy for the usual winners that will decay when crowded?

1. Is the rising SMA200 slope gate a genuine quality filter, or a fitted patch that happened to help after 2020?

1. Is offset 10 a real calendar effect, a sampling artifact, or a proxy for avoiding month end flows?

1. Does the 33.3 percent expected CAGR haircut go far enough for phase selection and variant selection on the same sample?

1. Is the rolling fresh start design the right way to judge start date risk, or does it still hide path dependence because all windows share the same broad 2010 to 2026 regime?

1. At the frontier knee, is 1.00 rotation plus 0.50 TQQQ actually smarter than simply holding more TQQQ and less rotation?

1. How should the model treat broker liquidation if the account gaps through maintenance overnight? What buffer above maintenance makes self management realistic?

1. What would falsify this thesis in forward paper trading within 6 to 12 months?

1. What is the cleanest null model? A QQQ and TQQQ band rebalanced blend has been proposed but not fully tested. Should that be the bar before any live capital discussion?

1. Are five names the right concentration, or does the edge survive at 8 to 10 names with lower single name blowup risk?

What would change the conclusion

The thesis weakens materially if any of the following happen:

• The QQQ / TQQQ passive blend matches the hybrid on rolling CAGR with lower drawdown.
• Parameter stability tests show the result exists only near SMA50 / SMA200, top five, Buffer15, and offset 10.
• Pre 2010 or synthetic crash tests show the rotation plus leverage stack fails to recover within a four year window.
• Broker exact margin accounting removes the k = 0.50 zero breach result.
• Forward paper trading shows slippage, gap behavior, or missed rebalance execution that the monthly open model does not capture.

Reproducibility pointers

Local research directories used for the current numbers:

• ~/workspace/backtests/market_leader_sma/
• ~/workspace/backtests/market_leader_sma/rolling4y/
• ~/workspace/backtests/market_leader_sma/phase_offset/
• ~/workspace/backtests/market_leader_sma/phase_rolling/
• ~/workspace/backtests/market_leader_sma/hybrid_margin_tqqq75/
• ~/workspace/backtests/market_leader_sma/allocation_frontier/

These paths are internal working references. A public version should attach the code, data dictionary, and result tables before asking for serious external replication.