SMA200 on Extended/Synthetic Long Windows

Question

Reddit commenter on Post #2 challenged the regime bias of the 2010-2026 LETF inception data, arguing that extending the backtest through dotcom + GFC would flip the filter-vs-B&H comparison.

This study tests that claim by extending the SMA200 vs buy-and-hold analysis on three longer windows:

Synthetic TQQQ since 1999 (~27 years, includes dotcom + GFC + COVID + 2022)
Unleveraged S&P 500 since 1940 (~86 years, includes WWII, 1970s stagflation, dotcom, GFC, COVID, 2022)
Synthetic 3x S&P 500 since 1940 (same window, leveraged version, illustrative not realistic)

Methodology

Harness: src/trader/backtest/engine.py, same as the inception study
Data: yfinance for QQQ (back to 1999-03-10), ^GSPC (back to 1927)
Synthesis: Synthetic LETFs constructed as (1 + leverage × daily_return − expense_ratio/252).cumprod(). TQQQ expense modeled at 0.86% annual; UPRO/SPXL at 0.91%. No financing cost modeled (this is the biggest pre-2009 unrealism).
Strategy: Pure SMA200 (long when close > SMA200, cash otherwise). Same lag, costs, no-lookahead conventions as the inception study.
Cost: 1bp per side.
Starting capital: $10,000.

Results

Synthetic TQQQ since 1999-03-10 (27.1 years)

	Buy-and-hold	SMA200 filter
Final equity (from $10k)	$90,612	$1,605,618
CAGR	8.46%	20.58%
Max drawdown	-99.96%	-77.51%
Volatility	80.83%	44.11%
Sharpe (rf=0)	0.506	0.647
Sortino	0.669	0.667

Sharpe diff (filter − B&H): +0.141. Compare to real TQQQ 2010-2026: −0.063. Extending back through dotcom flips the sign by ~0.20 Sharpe points.

Unleveraged S&P 500 since 1940-01-02 (86.2 years)

	Buy-and-hold	SMA200 filter
Final equity (from $10k)	$5,830,383	$4,174,993
CAGR	7.67%	7.25%
Max drawdown	-56.78%	-28.56%
Volatility	15.73%	10.30%
Sharpe (rf=0)	0.549	0.731
Sortino	0.693	0.807
Calmar	0.135	0.254

Sharpe diff (filter − B&H): +0.183. The canonical academic result for trend-following on the broad market, reproduced over the longest credible window.

Decade-by-decade on unleveraged S&P 500 (the real story)

Decade	B&H CAGR	SMA CAGR	B&H MDD	SMA MDD	Sharpe diff
1940s (WWII)	+2.91%	+5.20%	-41.5%	-16.9%	+0.322
1950s (boom)	+13.70%	+11.19%	-21.5%	-10.6%	+0.024
1960s (sideways)	+4.45%	+6.55%	-28.0%	-7.3%	+0.575
1970s (stagflation)	+1.50%	+2.96%	-48.2%	-14.5%	+0.230
1980s (bull + 1987)	+12.78%	+10.37%	-33.5%	-18.0%	+0.122
1990s (dotcom run-up)	+15.06%	+12.68%	-19.9%	-15.3%	-0.000
2000s (lost decade)	-2.63%	+2.64%	-56.8%	-21.5%	+0.353
2010s (QE bull)	+11.07%	+6.06%	-19.8%	-15.5%	-0.170
2020s (so far)	+13.21%	+9.90%	-33.9%	-19.8%	+0.228

The 2010s is the ONLY decade in 9 where the filter underperforms on Sharpe. And that's the decade dominating the LETF inception data (since LETFs only existed from 2009 forward).

Synthetic 3x S&P 500 since 1940 (illustrative, not realistic)

	Buy-and-hold	SMA200 filter
Final equity (from $10k)	$1,202,884,233	$15,697,944,713
CAGR	14.54%	18.01%
Max drawdown	-97.84%	-65.06%
Sharpe	0.530	0.711

Sharpe diff: +0.182 — basically the same gap as unleveraged.

Interpretation

The Reddit commenter was right. The original Post #2's "filter underperforms on Sharpe" framing was a 2010-2026 sample artifact, not a general truth.

Across 86 years of S&P 500 data: - Filter wins meaningfully in 5 of 9 decades (1940s, 1960s, 1970s, 2000s, 2020s) - Ties or marginal in 2 decades (1950s, 1980s) - Marginal loss in 1990s (clean dotcom run-up) - Meaningful loss in 2010s (QE bull)

The filter is structurally a drawdown manager that wins in bear/sideways regimes and loses in clean bull regimes. LETF inception backtests are biased toward the 2010s because that's when the funds existed; the structural bias of the sample misrepresents the long-run case for the filter.

For the synthetic TQQQ comparison specifically: extending from 16 years to 27 years (adding dotcom) was sufficient to flip the Sharpe ordering. The 86-year synthetic 3x SP result of +0.182 Sharpe is consistent with the unleveraged result, suggesting the trend filter's edge is robust across the leverage dimension.

Caveats

Pre-2009 LETF synthesis ignores financing costs. Real borrow at 4-15% prime rates during 1940-1980 would have eaten meaningful CAGR from any 3x daily-rebalance strategy. The synthetic 3x numbers (especially the $1.2B and $15.7B headline final values) are directional, not what you'd have actually earned.
The $1.2B → $15.7B headline numbers are visually absurd because we're compounding 86 years at high CAGR without realistic frictions. Use them only to compare strategies against each other, not as predictions.
^GSPC is the S&P 500 PRICE INDEX (no dividends). The "real" S&P with dividends reinvested (^SP500TR) only goes back to 1988. Using ^GSPC understates absolute returns for both B&H and the filter, but the COMPARATIVE answer is approximately right since dividends accrue proportionally to time-in-market.
The synthetic TQQQ vs testfol.io discrepancy: our model gives $90k B&H ending value; testfol shows ~$19k. Both methodologies hit ~99.97% max drawdown. The difference is recovery math from near-wipeout being extremely sensitive to modeling choices (financing, rebalance, slippage). Qualitative conclusion (B&H got eaten in 2000-2002, filter survived) is robust either way.

Source

Inline runner (SP500 1940 version):

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

spx = fetch_daily("^GSPC", start="1940-01-01")
bh = run_backtest("SPX B&H", spx, BuyAndHold().generate_positions(spx), cost_bps=1.0)
sma = run_backtest("SPX SMA200", spx, PriceAboveSma(window=200).generate_positions(spx), cost_bps=1.0)
print(f"BH Sharpe {bh.metrics['sharpe']:.3f}, SMA Sharpe {sma.metrics['sharpe']:.3f}")

Saved logs: /tmp/sma200_1940.log, /tmp/synthetic_tqqq.log.

2026-05-13_letf_inception_sma200_vs_bh.md — the original 2010-2026 inception study this one corrects
Leverage vs volatility per underlying — 1x vs 3x per underlying on same long window