Proven Trading Strategies: Three With Published Evidence, and Their Limits

No trading strategy is proven in the sense that it is sure to make money. What exists is evidence: rules that made money over decades, on many markets, in studies that other researchers could check. Three strategies have that kind of published record, and each comes with limits that its own authors describe. Here is what each one does, what the research found, and what it did not find.

What "proven" can mean for a strategy

A backtest that made money is not proof. Try enough rules on the same prices and some will look good by chance. The research standard is stricter: a rule tested over long periods, on markets other than the one it was found on, and published so that others can repeat it.

Even then, results shrink. McLean and Pontiff (2016) followed 97 variables that published studies had found to predict stock returns. Portfolio returns were "26% lower out-of-sample and 58% lower post-publication" (McLean and Pontiff, Journal of Finance, 2016). They read the first drop as an upper bound of data mining and the rest as investors trading on what they learned from the papers. Expect any published edge to be smaller in your account than in the paper.

1. Time-series momentum: follow each market's own trend

The rule looks at one market at a time: if its return over the past 12 months is positive, go long; if negative, go short. Moskowitz, Ooi and Pedersen (2012) found that the past 12 months of returns predicted the next month in each of the 58 liquid futures contracts they studied, across equity indexes, currencies, commodities and bonds. The effect lasted from 1 to 12 months and partly reversed over longer horizons (Moskowitz, Ooi and Pedersen, Journal of Financial Economics, 2012).

The evidence is for a portfolio of many futures traded together, not for one market on its own. A single market can go sideways for years and the rule will lose in small steps while it waits. The authors also write that a diversified time-series momentum portfolio "performs best during extreme markets."

The diagram shows the difference with the next strategy. Time-series momentum asks one question per market, is it above where it was a year ago, and can be long everything or short everything at once. Cross-sectional momentum only ranks: it buys the relative winners and sells the relative losers, even when all of them are falling.

2. Cross-sectional momentum: buy the winners, sell the losers

This rule compares stocks with each other instead of with their own past. Rank stocks by their return over the past months, buy the best performers and sell the worst. Jegadeesh and Titman (1993) documented that such strategies "generate significant positive returns over three- to twelve-month holding periods," that the profits were not explained by systematic risk, and that "part of the abnormal returns generated in the first year after portfolio formation dissipates in the following two years" (Jegadeesh and Titman, Journal of Finance, 1993).

Two limits follow from the design. It needs a large list of stocks and a short leg, which most individual accounts cannot run at low cost. And published stock-return predictors as a group lost much of their return after publication (McLean and Pontiff); there is no reason to expect momentum to be exempt.

3. Trend timing with a 10-month average

The simplest of the three: hold the S&P 500 while its monthly close is above its 10-month simple moving average, move to cash when it closes below, and check once a month. Faber chose the 10-month average because it is the closest match to the 200-day moving average. Faber published the rule in The Journal of Wealth Management in 2007 and extended the test in a February 2013 update, released as a working paper. In that update, on the S&P 500 from 1901 to 2012, total returns included, the rule compounded at 10.18% a year against 9.32% for buy and hold, with cash earning Treasury bill interest and with taxes, commissions and slippage left out (Faber, 2013 update of the 2007 Journal of Wealth Management paper).

The gain did not come from a higher average year. Faber reports an average annual return of 11.26% for the S&P 500 and 11.22% for the rule; the rule compounded faster because it was less volatile, and it cut the deepest drawdown, in the late 1920s and early 1930s, from 83.66% to 42.24%. It did this while invested about 70% of the time, with less than one round trip a year, and while trailing the index in roughly half of all years.

The chart shows the rule's best-known exit. On monthly closes of an S&P 500 ETF, the close went below its 10-month average in December 2007 at 103.88 and came back above it in June 2009 at 67.76, so the rule sat out most of the fall to the lowest monthly close, 53.79 in February 2009. It also shows the cost: the close crossed the average four times between June and September 2010 and again in August 2011 and January 2012, and each time the rule bought back higher than it had sold. Our own test of this rule on trend trading, from 1993 and without interest on cash, gained less than buy and hold but cut the deepest fall in half.

What about chart patterns?

A success rate quoted without its test is not evidence, and chart patterns are no exception. We measure each pattern ourselves, on crypto, forex, stocks, ETFs, indexes and commodities, with fixed detection rules and the full outcome after every signal. The results are on each pattern's page, for example head and shoulders, double top and flag; the method is in how we test patterns.

How to judge a strategy before you trade it

• Ask for the test. Which markets, which years, which costs. A strategy without a stated test has no evidence, whatever its claimed win rate.
• Look for other markets and other periods. Faber counts the years after 2005 as out of sample, since his paper first appeared in 2006, and writes that his models "have performed well in real-time"; time-series momentum held on each of 58 markets rather than on one. A rule that works on one chart may be fitted to it.
• Count the costs. Faber's figures exclude commissions, slippage and taxes. Short holding periods multiply them.
• Test it yourself. Write the rule down, run it on data it was not built on, and trade it small first. See backtesting and trading plan.

For other approaches and how they are traded, see day trading strategies and mean reversion; more strategy guides are in trading concepts.