Leverage vs Volatility: B&H 1x vs B&H 3x vs SMA200+1x vs SMA200+3x per Underlying

Question

For SPY, QQQ, and ^SOX over the same long window (1999-2026):

Pure buy-and-hold of the UNLEVERAGED underlying
Pure buy-and-hold of synthetic 3x of the underlying
SMA200 filter applied to the unleveraged underlying
SMA200 filter applied to the synthetic 3x

Which combination wins on Sharpe? On absolute dollars? Does leverage always pay off?

Methodology

Window: 1999-03-10 → 2026-05-18 (27.2 years, common-start for SPY/QQQ/^SOX)
Synthesis: Synthetic 3x = (1 + 3 × daily_return − 0.91%/252).cumprod()
Strategy: SMA200 filter (long when close > SMA200, cash otherwise), position lagged 1 bar
Cost: 1bp per side on position changes
Starting capital: $10,000

Results

Underlying	Strategy	CAGR	MDD	Vol	Sharpe	Final $10k
SPY	B&H 1x (unleveraged)	8.52%	-55.19%	19.27%	0.521	$91,992
SPY	B&H 3x synthetic	13.16%	-96.66%	57.81%	0.505	$286,638
SPY	SMA200 + 1x	6.38%	-27.53%	11.32%	0.603	$53,529
SPY	SMA200 + 3x synthetic	15.17%	-64.41%	33.97%	0.587	$462,485
QQQ	B&H 1x (unleveraged)	10.85%	-82.96%	26.94%	0.517	$163,875
QQQ	B&H 3x synthetic	8.47%	-99.96%	80.82%	0.506	$90,927
QQQ	SMA200 + 1x	7.85%	-58.40%	16.65%	0.537	$77,727
QQQ	SMA200 + 3x synthetic	14.71%	-95.14%	49.96%	0.527	$415,008
^SOX	B&H 1x (unleveraged)	13.30%	-87.15%	37.50%	0.520	$296,564
^SOX	B&H 3x synthetic	-5.52%	-100.00%	112.49%	0.512	$2,142
^SOX	SMA200 + 1x	8.89%	-63.54%	23.49%	0.480	$100,924
^SOX	SMA200 + 3x synthetic	8.73%	-98.83%	70.48%	0.474	$96,895

Interpretation

Three findings, each more counterintuitive than the last:

1. SPY responds well to 3x + SMA200; QQQ marginally; ^SOX not at all

Best Sharpe per underlying	Best $ per underlying
SPY: SMA200 + 1x (0.603)	SPY: SMA200 + 3x ($462k)
QQQ: SMA200 + 1x (0.537)	QQQ: SMA200 + 3x ($415k)
^SOX: B&H 1x (0.520)	^SOX: B&H 1x ($297k)

For SPY and QQQ, SMA200 + 1x has the best Sharpe (most efficient risk-adjusted return), while SMA200 + 3x has the most absolute dollars. Trade-off: more risk for more reward.

For ^SOX, the trade-off vanishes. B&H 1x unleveraged dominates on both Sharpe AND absolute return. Adding leverage actively destroys value over this 27-year window. The trend filter helps somewhat (avoids total wipeout) but can't reclaim what's lost to vol decay during the dotcom + GFC crashes.

2. Buy-and-hold 3x synthetic ^SOX literally wiped out

$10k → $2,142 over 27 years. That's a 78.6% loss on what should be a winning long-term bet on semiconductors (which have been the best-performing sector in modern history).

Why: ^SOX has 37.5% annualized volatility. At 3x daily leverage, that becomes ~112% effective vol. Volatility decay scales with the SQUARE of vol. At those levels, the daily reset mechanism eats more on down days than it gains on up days, even over very long windows where the underlying trends sharply upward.

This is not a regime-dependent finding. It's mathematical. ANY underlying with vol > ~30-35% will be destroyed by 3x daily-rebalance leverage over multi-decade windows. The semiconductor sector simply happens to have that vol level in modern history.

3. B&H 3x QQQ underperformed B&H 1x QQQ

$10k → $91k for B&H 3x QQQ vs $164k for B&H 1x QQQ. Same vol-decay mechanism as ^SOX but smaller magnitude. Even on the most tech-bullish underlying possible over 27 years, the 3x version compounds slower than the 1x version due to vol decay.

The SMA200 filter on 3x QQQ ($415k) recovers and exceeds the 1x B&H ($164k) by sidestepping the worst of the dotcom drawdown.

What the SMA200 filter actually does (and doesn't)

Does: Reduces max drawdown materially (1x SPY: -55% → -28%; 1x QQQ: -83% → -58%; 1x ^SOX: -87% → -64%).

Does: Improves Sharpe modestly on 1x (SPY: 0.52 → 0.60; QQQ: 0.52 → 0.54).

Does NOT: Eliminate the vol-decay penalty on 3x high-vol underlyings. SMA200 + 3x ^SOX ($97k) is essentially tied with SMA200 + 1x ^SOX ($101k). Adding 3x leverage delivers no incremental dollar return despite the filter being on.

Does NOT: Beat B&H on absolute dollars for unleveraged plays. SMA200 + 1x SPY ($54k) < B&H 1x SPY ($92k). The filter trades CAGR for drawdown protection at 1x.

The defensible pecking order

If your goal is highest absolute dollars over 25+ years:

Underlying	Best variant
SPY	SMA200 + 3x synthetic ($462k)
QQQ	SMA200 + 3x synthetic ($415k) — but at -95% MDD
^SOX	B&H 1x unleveraged ($297k) — leverage actively destroys returns

If your goal is best risk-adjusted return (Sharpe):

Underlying	Best variant
SPY	SMA200 + 1x (0.603)
QQQ	SMA200 + 1x (0.537)
^SOX	B&H 1x (0.520)

The defensible-across-tickers rule: SMA200 on the unleveraged underlying is consistently in the top 1-2 strategies for any underlying. Most robust answer.

Implications for sma200.trade content

This is article-worthy material. A /learn article titled something like "When 3x leverage destroys you: vol decay and the SOXL trap" would land hard because:

The conventional retail intuition ("more leverage on tech-y stuff = more upside") is exactly inverted by the data
SOXL is a popular retail vehicle with $4B+ AUM — a lot of holders need to see this analysis
The data is clean and the explanation (vol² decay) is teachable
It reinforces the SMA200-as-conversation-starter framing of the site (the FREE tier already shows you the SMA200 status; the article gives you the WHY of when leverage is or isn't a good idea)

This pairs well with letf_inception_sma200_vs_bh.md (Post #1) and sma200_extended_windows_synthetic.md (Reddit thread material).

Caveats

Synthetic 3x ignores financing cost. Real LETFs in this window did have borrow drag (varied with rates from ~0.5% to ~5% annual on the leveraged portion). The synthetic numbers understate the real cost of 3x. Real SOXL/TQQQ/UPRO performance is somewhat worse than the synthetic shown here.
Vol-decay magnitude depends on the regime mix. A window without 2000-2002 and 2008 (the biggest contributors to ^SOX synthetic wipeout) would show milder decay. The 27-year window includes both, which is honest.
^SOX is the unleveraged semi INDEX, not SOXX (the ETF). SOXX tracks ^SOX closely but launched 2001, so we use ^SOX for the longer history. SOXX-based numbers would be marginally different.
The 3x daily-rebalance assumption ignores intraday rebalancing realities. Real LETFs reset at market close; our synthetic assumes the same. Tracking error in real LETFs would add another small drag.

Source

Saved log: /tmp/defensive_and_rebal.log (the comparison was added late in the session; runner is inline in chat).

Inline runner:

import pandas as pd, numpy as np
import sys; sys.path.insert(0, '/Volumes/Mac External/Claudes/trader/src')
from trader.data.yfinance_src import fetch_daily

ER = 0.0091; TD = 252
underlyings = {
    "SPY": fetch_daily("SPY", start="1999-01-01")["close"],
    "QQQ": fetch_daily("QQQ", start="1999-01-01")["close"],
    "^SOX": fetch_daily("^SOX", start="1999-01-01")["close"],
}
start = max(u.index[0] for u in underlyings.values())

for name, u in underlyings.items():
    u = u.loc[start:]
    r = u.pct_change().fillna(0.0)
    # B&H 1x
    bh1 = u / u.iloc[0]
    # B&H 3x synthetic
    bh3 = (1.0 + 3*r - ER/TD).cumprod()
    # SMA200 + 1x
    sig = (u > u.rolling(200).mean()).astype(float).shift(1).fillna(0)
    sma1 = (1 + sig * r - sig.diff().abs().fillna(sig.iloc[0]) * 1e-4).cumprod()
    # SMA200 + 3x synthetic
    sma3 = (1 + sig * (3*r - ER/TD) - sig.diff().abs().fillna(sig.iloc[0]) * 1e-4).cumprod()
    # Compute metrics for each, print

2026-05-13_letf_inception_sma200_vs_bh.md — the 2010-2026 real-LETF version of this analysis
2026-05-14_sma200_extended_windows_synthetic.md — same strategies on longer windows
a sample heavy-leveraged portfolio (internal-only study) — why this informs whether SOXL should be in the portfolio at all