The 200-Day SMA Beat Buy-and-Hold in 8 of the Last 9 Decades
Almost every 200-day SMA backtest you've read has the same blind spot: it starts in 2010.
Not because 2010 is special, but because that's roughly when leveraged ETFs came into existence, and leveraged ETFs are what people want to test the rule on. So the sample everyone argues over (QQQ, TQQQ, SOXL, 2010 to today) happens to cover one long, mostly-uninterrupted QE bull market.
Here's the problem. A trend filter is a drawdown manager. It earns its keep in bears and sideways chop, and it gives a little back in clean bull runs. If you only test it across a decade that was mostly a clean bull run, you are grading a raincoat during a drought and concluding raincoats are useless.
So let's zoom out. Way out.
86 years, one broad index
We ran the same rule (long when the close is above its 200-day SMA, cash otherwise, trade the next open) on the unleveraged S&P 500 price index from 1940 through 2026. That is 86 years covering WWII, 1970s stagflation, the dotcom bubble, the GFC, COVID, and 2022. The longest credible window we could assemble.
Over that full stretch:
| Buy-and-hold | SMA200 filter | |
|---|---|---|
| CAGR | 7.67% | 7.25% |
| Max drawdown | -56.8% | -28.6% |
| Volatility | 15.7% | 10.3% |
| Sharpe (rf=0) | 0.549 | 0.731 |
The filter gives up about 0.4 points of annual return and in exchange cuts the worst drawdown roughly in half and lifts the Sharpe ratio by +0.18. That is the canonical trend-following result, reproduced over the longest window I could find data for. Nothing surprising yet.
The interesting part is what happens when you slice it by decade.
The decade-by-decade table
This is the whole article in one table. Sharpe diff is filter minus buy-and-hold, so positive means the filter won.
| Decade | B&H CAGR | SMA CAGR | B&H MaxDD | SMA MaxDD | 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 |
Count the Sharpe-diff column. The filter beats buy-and-hold in eight of the nine decades. It ties in the 1990s (a clean dotcom run-up, the kind of straight-line bull where a filter can only cost you). And it meaningfully loses in exactly one: the 2010s.
The 2010s. The QE bull. The decade every leveraged-ETF backtest is built on top of, because that's the only decade those funds existed for.
That is the entire story. The one regime where the filter looks bad is the one regime the popular backtests can see.
The 2000s are the tell
If you want to understand what the filter is actually for, look at the 2000s row.
Buy-and-hold the S&P 500 for that decade and you earned -2.63% per year. Ten years, negative. That is the "lost decade" people talk about, two 50%-class drawdowns (dotcom, then the GFC) with a fake recovery in between.
The SMA200 filter turned that same decade into +2.64% per year and held the max drawdown to -21.5% instead of -56.8%. Not because it found some secret alpha, but because it was sitting in cash for the ugliest stretches of both crashes.
That is the trade in one row. A filter does nothing for you when the market only goes up. It does a great deal for you when the market spends a decade going nowhere through two crashes. You do not get to know in advance which kind of decade you are about to live through.
Why this matters for the leverage crowd
Everything above is on the plain, unleveraged S&P 500. No decay, no daily rebalance, no financing cost. I kept it unleveraged on purpose, because it isolates the trend filter's behavior from every leveraged-ETF complication people usually argue about.
But it maps directly onto the leverage question. If a 0.18-Sharpe, drawdown-halving edge shows up on the boring index across 86 years and eight of nine decades, then the case for running the filter on a 3x product (where an unmanaged drawdown is a -80% or -90% event instead of a -50% one) is stronger, not weaker. The filter's whole job is capping the left tail, and leverage is what makes the left tail lethal.
I am deliberately not putting synthetic 3x numbers here. Modeling leveraged ETFs before 2009 requires accounting for the daily borrow cost on their swap exposure, and most retail backtests skip it and produce fantasy returns. If you want that rabbit hole done honestly, it is a separate piece: The Hidden Cost Every Leveraged-ETF Backtest Ignores.
What to take from this
- Beware any SMA200 verdict drawn only from 2010 onward. That window is the single most filter-hostile decade in modern history. Concluding the rule is worthless from that sample is a survivorship error, just about the sample instead of the strategy.
- The filter is regime-dependent drawdown insurance. It wins bears and sideways chop (1940s, 1960s, 1970s, 2000s), roughly ties clean bulls, and costs you a little in a QE melt-up. Know which one you are buying.
- The edge is real but modest and lumpy. +0.18 Sharpe over 86 years is a genuine, academically-familiar result. It is not a money printer, and any decade can go against you.
The unsexy summary: the 200-day SMA did not beat buy-and-hold because it earns more. It beat it because it lost less, in eight of the last nine decades, in the decades that actually hurt.
Caveats worth naming
- ^GSPC is the S&P 500 price index, no dividends. The dividend-reinvested series only goes back to 1988, so the absolute CAGRs above understate total return for both strategies. Read the numbers as a comparison between the two paths, not as a prediction of what you would have banked.
- The pre-2010 record is index-level, not something you could have traded frictionlessly for 86 years. Costs, taxes, and the practical reality of holding a mechanical rule through WWII are not modeled. Treat the decade table as regime evidence, not a brokerage statement.
- One decade going against the rule is normal, not a bug. The 1990s tie and the 2010s loss are part of the deal. A filter that never underperformed in a bull market would be a filter that wasn't actually sitting out anything.