A rule as simple as "hold the index while it is above its ten-month average" cut the worst drawdown of the last century roughly in half. It also sat in cash through stretches it should not have, signalled 180 times in 25 years, and looked far better in the backtests than it did afterward. The evidence on both sides, with the numbers.
Trend following is the oldest idea in trading: hold what is rising, step aside from what is falling, and let the market tell you which is which. On stocks it is usually expressed as a moving-average rule, and it has a longer and better-documented record than almost any other systematic idea. It also has a longer record of being oversold. This page is the evidence on both sides, and it is written by a desk whose own systems are rule-based, so the second half is as much about us as about anyone.
The simplest version, and what it did
Mebane Faber's 2007 paper tested a rule a child could follow: at the end of each month, if the index closed above its ten-month simple moving average, hold it; if below, hold Treasury bills. In the 2013 update, applied to the S&P 500 from 1901 to 2012, the rule's compounded return was 10.18% a year against 9.32% for buying and holding — and its worst peak-to-trough fall was 42.24%, against 83.66% for the index. It was invested about 70% of the time and traded less than once a year. Applied across five asset classes the same way, the worst drawdown fell from 46% to under 10%. Faber's own summary was "equity-like returns with bond-like volatility and drawdown."
The longer evidence points the same way. Hurst, Ooi and Pedersen took a time-series momentum strategy — long a market that has been rising over the past one, three and twelve months, short one that has been falling — across 67 markets from 1880 to 2016, and found it "has performed well in each decade for more than a century." Of the ten largest drawdowns of a 60/40 stock-bond portfolio in that span, the strategy "experienced positive returns in 8 out of 10." Moskowitz, Ooi and Pedersen had already shown, on 58 instruments, that returns persist for one to twelve months and that the strategy "performs best during extreme markets." Brock, Lakonishok and LeBaron found in 1992, on the Dow from 1897 to 1986, that "buy signals consistently generate higher returns than sell signals."
So the finding is real and the mechanism is worth stating plainly: a trend rule does not predict returns, it cuts off the left tail. It gets you out partway down the big declines and back in partway up the recoveries, and it pays for that insurance in the stretches between.
What it did not do
It looked much better in the backtest than afterward. Valeriy Zakamulin re-ran moving-average and momentum rules on U.S. data from 1926 to 2012 with realistic costs, choosing the rule's parameters only from data available at the time rather than from the whole history. Judged that way, the best moving-average rule's monthly Sharpe ratio fell from 0.160 in the backtest to 0.120 out of sample, against 0.109 for buy-and-hold — an advantage of roughly ten percent, where the backtest had shown fifty. In about 80% of the months the rule sat in Treasury bills, bills failed to beat stocks. The timing rules were "inferior during a 25-year period from 1975 to 1999," which is to say through the great bull market of the 1980s and 1990s. His abstract calls the reported performance of these rules "highly overstated, to say the least."
The best rule in any search is partly luck. Sullivan, Timmermann and White tested 7,846 technical rules on the Dow. The best of them, a five-day moving average, returned 17.2% a year across 1897–1996 — and 2.8% a year, statistically indistinguishable from nothing, in the out-of-sample decade from 1987 to 1996. What a backtest can and cannot tell you is the arithmetic of why.
It signals constantly in a sideways market. A moving average is a lagging line, and when price wanders around it the rule whipsaws: out, back in, out again, each time at a slightly worse price. Arthur Hill counted the S&P 500 crossing its 200-day average "some 180 times" between 1994 and 2019. Of 90 completed buy signals, 20 were winners averaging +13.69% and 70 were losers averaging −1.32%. The arithmetic still works — the few large wins pay for the many small losses — but a holder has to sit through seventy small losses to collect twenty large wins, and most people stop at loss number fifteen.
It costs more than it looks for a taxable investor. Every exit is a sale. Newfound Research measured a simple monthly trend rule on a total-market fund with ordinary tax rates and found the gap between pre-tax and after-tax returns ran about 100 basis points a year — one percent, every year, that the buy-and-hold investor did not pay. Schwab's 2026 illustration of turnover generally: at top rates, fully turning a portfolio over once a year costs about 194 basis points of annual return over thirty years.
| What the evidence says | The number | Where |
|---|---|---|
| It cuts the worst drawdown roughly in half | 83.66% → 42.24% (S&P 500, 1901–2012) | Faber 2013 |
| It has worked across a century and most markets | positive in 8 of the 10 biggest 60/40 drawdowns, 1880–2016 | Hurst, Ooi, Pedersen 2017 |
| It gives back most of its backtested edge afterward | Sharpe 0.160 in sample → 0.120 out; buy-and-hold 0.109 | Zakamulin 2014 |
| Its time in cash is usually wasted | ~80% of sell periods, bills trailed stocks | Zakamulin 2014 |
| It whipsaws | 180 crossings in 25 years; 70 of 90 buys lost | Hill 2019 |
| It costs a taxable holder | about 100 bps a year | Newfound 2019 |
What a desk that runs rules makes of this
Everything above transfers to the futures systems on the desk's rack, and we would rather say so here than have a member find out. A trend rule's value is the left tail it cuts off, and the price of that is a long run of small losses that the backtest compresses into one tidy line. A rule's backtest flatters it, by the mechanism Zakamulin measured, and the desk's response is the one his method implies: judge a system only on data it did not get to see, charge it full costs, and treat its best-case drawdown as a floor (drawdown decides whether you survive has the arithmetic). And the cost side is why the desk trades futures rather than stocks for this kind of rule: regulated futures are taxed under one regime whatever the holding period, so a system that trades often is taxed the same as one that trades once — which is not true of a taxable stock account, where every quick exit is a short-term gain at ordinary rates. How a futures contract works has the detail, and a tax professional has your situation.
None of this makes a trend rule good or bad for you. It makes it a specific trade: a smaller worst case, bought with lower returns in long bull markets, many small losses, and taxes. Whether that trade is worth making depends on which of those you can live with, and the evidence here is only the price list.
- A quantitative approach to tactical asset allocation — Mebane Faber, Journal of Wealth Management, 2007; updated February 2013
- A century of evidence on trend-following investing — Hurst, Ooi and Pedersen, Journal of Portfolio Management 44(1), 2017
- Time series momentum — Moskowitz, Ooi and Pedersen, Journal of Financial Economics 104(2), 2012
- The real-life performance of market timing with moving average and time-series momentum rules — Valeriy Zakamulin, Journal of Asset Management 15(4), 2014 (working-paper version)
- Simple technical trading rules and the stochastic properties of stock returns — Brock, Lakonishok and LeBaron, Journal of Finance 47(5), 1992
- Data-snooping, technical trading rule performance, and the bootstrap — Sullivan, Timmermann and White, Journal of Finance 54(5), 1999
- Just how effective is the 200-day moving average? — Arthur Hill, StockCharts, February 2019
- Taxes and trend equity — Nathan Faber, Newfound Research, April 2019
- How overtrading can undercut after-tax returns — Hayden Adams, Charles Schwab, July 2026
Education, not advice. Nothing on this page recommends buying or selling anything, and nothing is a promise of results — every figure is cited above or hypothetical. TOTH does not provide investment advice. New to the vocabulary? The glossary has plain definitions.