LETF Inception Backtest: Buy-and-Hold vs SMA200 Filter

Question

For SOXL, TQQQ, and UPRO, from each fund's actual inception through 2026-05-15: 1. What does pure buy-and-hold return? 2. What does a pure SMA200 trend filter (long when close > SMA200, cash otherwise) return?

How do they compare on CAGR, max drawdown, Sharpe, and the rest of the standard metric set?

Methodology

Harness: src/trader/backtest/engine.py
Strategy: src/trader/strategies/price_above_sma.py with window=200, long-only
Data: yfinance daily bars via src/trader/data/yfinance_src.py with auto_adjust=True (dividend-reinvested, split-adjusted total-return closes). This is the standard convention for backtest return computations and differs from the chart-standard convention used by sma200.trade itself.
Position sizing: 100% of stated notional when long, 100% cash when flat. No leverage beyond what the LETF provides.
Signal timing: Position lagged by one bar (pos.shift(1)), so a signal computed at today's close is applied to tomorrow's close-to-close return. No lookahead.
Reentry rule: Pure mechanical. The moment close > SMA200 condition flips true, position re-enters at next bar's close. No N-day confirmation, no buffer band, no anti-whipsaw block.
Transaction costs: 1 basis point per side (2bp per round-trip).
Risk-free rate: 0.0 for Sharpe and Sortino. Real risk-free over the window averaged ~1.5% annualized; applying a non-zero rf would lower both numerators by the same amount across both strategies without changing the ordering.
Tax treatment: None. All figures pre-tax. The SMA200 filter generates ~3 round-trips per year of short-term capital gains in taxable accounts.
SMA200 warmup: First 200 trading days have no SMA value. During this window the filter strategy is in cash by definition; buy-and-hold is fully invested from day one. This is the honest comparison: both strategies span the full life of the LETF.
Starting capital: $10,000.

Results

SOXL — Direxion Daily Semiconductor Bull 3X

Metric	Buy-and-hold	SMA200 filter
Inception	2010-03-11	2010-03-11
End date	2026-05-15	2026-05-15
Years	16.15	16.15
Final equity (from $10k)	$2,732,125	$577,724
CAGR	41.52%	28.54%
Volatility (ann.)	89.27%	59.53%
Max drawdown	-90.46%	-69.68%
Sharpe (rf=0)	0.839	0.723
Sortino	1.182	0.794
Calmar	0.459	0.410
Round-trip trades	1 (held)	54
Trades per year	—	3.34
Round-trip win rate	—	20.37%
Time in market	100.0%	63.23%

Sharpe diff (filter − B&H): −0.116

TQQQ — ProShares UltraPro QQQ

Metric	Buy-and-hold	SMA200 filter
Inception	2010-02-11	2010-02-11
End date	2026-05-15	2026-05-15
Years	16.23	16.23
Final equity (from $10k)	$3,655,934	$734,615
CAGR	43.85%	30.31%
Volatility (ann.)	60.96%	42.13%
Max drawdown	-81.66%	-50.03%
Sharpe (rf=0)	0.905	0.842
Sortino	1.165	0.923
Calmar	0.537	0.606
Round-trip trades	1 (held)	44
Trades per year	—	2.71
Round-trip win rate	—	31.82%
Time in market	100.0%	74.25%

Sharpe diff (filter − B&H): −0.063

UPRO — ProShares UltraPro S&P 500

Metric	Buy-and-hold	SMA200 filter
Inception	2009-06-25	2009-06-25
End date	2026-05-15	2026-05-15
Years	16.86	16.86
Final equity (from $10k)	$1,233,859	$137,899
CAGR	33.06%	16.84%
Volatility (ann.)	51.27%	31.68%
Max drawdown	-76.82%	-56.66%
Sharpe (rf=0)	0.817	0.651
Sortino	1.009	0.706
Calmar	0.430	0.297
Round-trip trades	1 (held)	49
Trades per year	—	2.91
Round-trip win rate	—	24.49%
Time in market	100.0%	73.26%

Sharpe diff (filter − B&H): −0.166

Bonus: signal computed on the underlying ETF

For each LETF, the same SMA200 strategy but using the underlying's signal:

Strategy	Final equity	CAGR	MDD	Sharpe	vs B&H	vs direct
SOXL via SOXX SMA200	$372,705	25.10%	-84.18%	0.674	-0.165	-0.049
TQQQ via QQQ SMA200	$1,019,548	32.97%	-55.99%	0.849	-0.056	+0.007
UPRO via SPY SMA200	$428,603	24.97%	-51.31%	0.801	-0.016	+0.150

UPRO via SPY meaningfully beats UPRO direct. TQQQ via QQQ is marginal. SOXL via SOXX is actively worse.

Interpretation

Over the full inception-to-2026 window for each LETF, pure SMA200 filtering underperforms buy-and-hold on Sharpe across all three:

SOXL: -0.116
TQQQ: -0.063
UPRO: -0.166

The filter wins meaningfully on max drawdown (TQQQ -82% → -50%, UPRO -77% → -57%, SOXL -90% → -70%) and on volatility, but at real cost in CAGR (-13 to -16 percentage points).

The case for the filter is behavioral, not mathematical. Backtests assume you hold through 90% drawdowns. Most humans don't. The filter trades CAGR for the ability to actually execute the strategy through a full cycle. See Article #1 on /learn for the full narrative.

Caveats

Sample is biased. 2009-2026 contained one of the strongest equity bull runs in modern history with only two relatively V-shaped drawdowns (2020 COVID, 2022 rate shock). A different regime weighted toward extended bear markets (e.g., 2000-2012 NDX lost decade) would tilt this comparison meaningfully toward the filter. Confirmed by the synthetic TQQQ 1999 study (next file).
The filter's 10-month cash warmup is included. Skipping the first 200 days (running both strategies from the first computable SMA value) would lift filter CAGR by 1.5-3 percentage points across these names without materially changing the Sharpe story.
Liquidity of older bars. SOXL and TQQQ both had thin liquidity in 2010-2012. Real-world 2bp round-trip costs would have been conservative then; spreads may have been wider. We don't model time-varying costs.
Pre-tax only. Filter generates ~3 short-term capital gains events per year. B&H generates zero until exit. After-tax results favor B&H by another margin in taxable accounts.
Bonus underlying-SMA results assume frictionless cross-instrument execution (signal triggered by SPY close → UPRO position next day). Real implementation would have a tiny cross-symbol timing slip not modeled here.

Source

Saved output: /tmp/sma200_post2_data.md (this file is a clean rewrite of that snapshot).

Inline runner:

import sys
sys.path.insert(0, '/Volumes/Mac External/Claudes/trader/src')
from trader.data.yfinance_src import fetch_daily
from trader.backtest.engine import run_backtest
from trader.strategies.buy_and_hold import BuyAndHold
from trader.strategies.price_above_sma import PriceAboveSma

for letf in ["SOXL", "TQQQ", "UPRO"]:
    df = fetch_daily(letf)
    bh = run_backtest(f"{letf} B&H", df, BuyAndHold().generate_positions(df),
                      cost_bps=1.0, starting_equity=10_000.0)
    sma = run_backtest(f"{letf} SMA200", df, PriceAboveSma(window=200).generate_positions(df),
                       cost_bps=1.0, starting_equity=10_000.0)
    print(f"{letf}: BH Sharpe {bh.metrics['sharpe']:.3f}, SMA Sharpe {sma.metrics['sharpe']:.3f}")

2026-05-14_sma200_extended_windows_synthetic.md — same strategy on synthetic data back to 1999/1940
LRS family replication + walk-forward — adds tolerance + RSI to this strategy, tests whether tuning helps
Leverage vs volatility per underlying — 1x vs 3x version of the same comparison