Everyone watches the 200-day moving average.
Funds defend it. Chartists draw it. Television announces when the index closes below it. If any single number in technical analysis has earned its reputation, this is the one.
Is it really as magical as it is cracked up to be?
So I checked it.
I ran the moving-average strategy on SPY, the S&P 500 ETF, across a range of lengths, none shorter than 50 days. Anything shorter flips position hundreds of times, and the dealing costs make it a rule nobody can actually run.
Drumroll….
On the S&P 500, it actually works
The rule is simple. Hold SPY while it trades above its 200-day average. Sit in cash otherwise. One day of execution lag before dealing costs.
Here it is from March 2011 to 19 August 2026.
Sharpe ratio is return per unit of wobble. Higher is better. The rule wins, slightly.
It gave up 3.7 points of annual return. It removed 14 points of drawdown. And it beat buying and holding on a risk-adjusted basis while sitting in cash 16% of the time.
Now stretch it across windows.
Four windows out of five. There is more.
I tested every length from 50 days to 300 on SPY. The 200-day came first. Of the eleven candidates, the famous one was the best choice.
Then I asked whether the timing of those spells in cash matters. Same number of days out of the market. Same length of each spell. Different dates, picked at random. I built ten versions like that. The real 200-day had to beat all ten. That bar is deliberately hard. Luck alone clears it 9.1% of the time. SPY’s 200-day cleared it. Most rules do not.
On the S&P 500, the 200-day moving average is real. I went looking to debunk it, but it held.
Ok, now is where an honest article has to slow down.
One chart is one chart
Granted, the S&P 500 is the most watched index in the world. Almost every equity strategy is benchmarked against it. After all that, it is still one index.
If a rule survives only there, it is a discovery about SPY, but nothing wider.
So here is the same test on four more: the Nasdaq 100 (QQQ), long-end Treasuries (TLT), Gold (GLD), and Bitcoin (BTCUSD).
The Nasdaq QQQ 200-day MA strategy underperforms simply buying and holding, even before transaction costs. The 150-day would have done better than the 200.
Gold loses to simply holding too. The 200-day ranks tenth of the eleven lengths available, and the 150-day scored 0.59 against its 0.46.
Long-end Treasuries are worse. The 200-day rule on TLT returned nothing at all, compared to 2.2% a year for simply holding. Over the last two years, it scored minus 1.26.
Bitcoin is the interesting one. The 200-day MA rule does beat simply holding. It also ranks ninth of eleven. The 125-day would have returned a Sharpe ratio of 1.30, compared with the 200-day’s 1.12.
Of those five, SPY is the only one that beat its own shuffled timing.
Across everything I cover
Extending the earlier argument, five names are still just five names.
Here are all 193 tickers I track, every length, full history.
Eleven lengths qualify once anything under 50 days is removed. Spread evenly, each moving average would win about 17 names by random chance.
The 200-day is the best of the eleven MA lengths on 16 tickers out of 193. Its median rank among the tested moving averages is 5th out of 11.
16 wins vs 17 expected. The most-watched number in markets wins about as often as a number drawn from a hat, as the winners are scattered across the whole range.
Here is how the popular ones actually did, averaged across the fleet.
Read the bottom row. Buy-and-hold scores a mean Sharpe ratio of 0.605. Not one moving average strategy, applied across the universe of tickers, gets close.
Read the third column as well. Across all eleven lengths, it runs from 20.2% to 24.9%. The 175-day ties the 200-day exactly. Whichever length you pick, it beats buying and holding the asset on roughly a fifth to a quarter of names. The gap between the best moving-average choice and the worst is under five points.
So, should you pick the best length for each name?
This is the question the whole thing sets up. Gold wants the 150. Bitcoin wants the 125. Fine. Give every name the length that suits it and move on.
I tested that. It fails twice.
First, the winner does not persist. Split each name’s history in half. Rank the eleven lengths in the first half, rank them again in the second half, and see whether the order holds.
The average correlation is 0.03. That is a coin flip with extra steps. The length that won the first half beats the second half’s median on 45% of names, which is worse than picking at random.
Knowing what worked tells you almost nothing about what is coming.
Second, re-optimising costs money. I ran a six-monthly update. At each of the eight anniversaries, I picked the length that worked best. Trade it for next year. Repeat.
The picker loses. It beats the fixed 200-day on only 41.5% of names.
It also churns. Across eight anniversaries, it changes its mind 2.1 times and uses 2.4 different lengths. Only one name in five keeps the same setting throughout.
Cherry-picking does not work. Updating it every six months works less well than never touching the 200-day MA strategy at all.
What to actually do
Stop asking which moving average length is best. Ask whether you want the trade-off at all.
Here is what the filter charges and what it pays on the five names at the 200-day.
Two of the five above assets pay you to run the moving-average filter. Two names charge a small amount for a large reduction in pain. The remaining one takes a lot and gives back almost nothing.
TLT is the clear instruction. You surrender 0.17 of Sharpe to remove 2.6 points of drawdown. That is the worst trade on the board, and it is worth knowing before you apply a rule to bonds because it worked on equities.
Across all 193 tickers in my coverage universe, the pattern holds. The filter produced a shallower worst fall in 82% of cases, cutting the average from -62% to -50%. It beat buying and holding on risk-adjusted return for 25% of the tickers.
The moving-average filter is, therefore, a drawdown insurance.
Usually it costs a little. Occasionally it pays. The premium barely moves, whichever length you choose.
This reduces to one question per holding, and the question is about you.
Could you sit through the whole fall without selling near the bottom? Across these names that fall has averaged 62%. If you would hold, own the thing outright and skip the filter. If you know you would sell, run the moving average strategy.
That is the honest case for a moving average. It is the crudest drawdown tool there is.
One line, one decision, no judgement required on the morning it matters. Its real virtue is that a frightened investor will actually follow it.
Better methods exist, of course (I use them at YX Insights). Sizing rules, volatility targeting, exits written down before the money goes on. Every one of them asks more of you than a single line does. For anyone holding none of them, the 200-day is a floor to stand on.
What the 200-day actually is
Three things are true at once, although you may only ever hear one of them pushed in the media or on FinTwit.
The 200-day genuinely works on the S&P 500. It is the best length there. It beats holding on risk-adjusted return. That is not nothing.
It mostly does not work anywhere else I have looked. Not on the Nasdaq. Not on gold. Not on bonds. Not across the 193 names I cover, where it wins no more often than chance hands out.
But the moving average, as a trend filter, still does one job well. It cuts the worst fall on four names in five and charges you absolute returns for the service.
The verdict
So where does that leave the most famous line in technical analysis?
On the S&P 500 it is the real thing. Best of the eleven different moving-average lengths, beats holding on risk-adjusted return, and beats its own shuffled timing. I went in to knock it over and it stayed up.
Everywhere else, it is just a number. It is the best moving average on 16 names out of 193 (chance hands out about 17). Median rank 5th of 11.
And as a filter, it does the one job it never gets credit for. Shallower worst fall on 82% of names, the average cut from minus 62% to minus 50%. It beat holding on risk-adjusted return for a quarter of them.
The exact length? Barely matters. Pick the winner from the first half of history, but it tells you close to nothing about the second. Re-optimise every year, and you beat the fixed 200-day on 41.5% of names while changing your mind 2.1 times for pleasure.
The moving average filter is a drawdown insurance
The 200-day’s real advantage turns out to be the boring one.
What now?
That is one rule, tested once. There are a lot more of them.
I run four models across equities, rates, commodities and crypto. I publish the daily Systematic Portfolios and Multi-model Signals, including deep dives into the portfolio holdings, at YX Insights.
You can learn more about how they work now.
Appendix & Disclaimer: Where the numbers come from
193 tickers, each on its own full price history to 19 August 2026. A 300-bar warm-up ensures every length is scored on identical days. Long while price is above the average, flat otherwise. One-day execution lag, before dealing costs. Sharpe is daily mean over standard deviation, annualised.
The shuffle is a matched-exposure placebo. I take the real position series and permute the order of its in and out stretches at random. Days invested, days flat, and the number of trades stay exactly equal. Ten seeds. Clearing the best of ten happens 9.1% of the time by chance. That is the bar in the tables above.
One caveat. My coverage is deliberately weighted to megacap technology, semiconductors, commodities and crypto. It is more volatile than a broad index.
DISCLAIMER: This newsletter is strictly educational. Nothing here is an offer to sell or a solicitation of an offer to buy any security, and nothing here is investment advice or should be relied upon to make investment decisions. Any opinions or analyses are those of the author as of the date of publication and are subject to change without notice.