Trend Following: The Long-Run Evidence

Trend following is the strategy with the strongest empirical support in the academic FX literature. It is also one of the most painful strategies in the world to actually trade. Both facts coexist, and the gap between them explains why a tradeable edge can sit in plain sight for decades without being arbitraged away. The marketing material almost always shows the first fact; the survivors only talk about the second.

This article walks the record: what the academic studies actually find, the structural reasons the effect persists, the drawdown cost of capturing it, and what a careful retail reader should take from all of it.

What trend following means in FX

In its simplest form: buy currencies that have been going up; sell currencies that have been going down. The horizon for “have been going up” varies by study, but the bulk of the literature finds tradeable signal at horizons from one to twelve months. At shorter horizons (intraday, days), the noise dominates; at longer horizons (multi-year), the effect reverses into mean reversion.

The implementation varies. A simple version ranks G10 currencies by their three-month return, longs the top quintile, and shorts the bottom quintile. A more elaborate version uses dynamic position sizing, multiple lookback windows, and volatility scaling. The broad finding survives most reasonable implementations of the same underlying idea.

Fig. 1 A clearly-trending series with sustained directional drift. Trend strategies attempt to identify the periods when this kind of behaviour dominates the noise. They do well in such regimes and badly when the market is mean-reverting or ranging. Periods of clean trend, like the one shown, are a minority of the calendar. Illustrative data: a synthetic series generated for teaching, not a real market quote.

What the academic record shows

A condensed summary of the published evidence:

Sweeney (1986) documented profitability of technical trading rules in FX, controlling for risk. Among the earliest formal papers to find evidence of edge from price-based rules in currencies.
Levich and Thomas (1993) extended the methodology, applied bootstrap testing, and confirmed positive returns from moving average crossovers and filter rules across the major pairs.
Neely, Weller, and Dittmar (1997) used genetic programming to search for profitable rules; the search converged on trend-following-style strategies that produced statistically significant excess returns.
Park and Irwin (2007) surveyed 95 modern studies of technical trading profitability. Their summary: most studies find positive returns from trend rules in FX, with effect sizes that were larger in the 1970s-1990s and have compressed in the 2000s.
Menkhoff, Sarno, Schmeling, and Schrimpf (2012) (“Currency Momentum Strategies”) is the most-cited recent paper. They construct cross-sectional currency momentum portfolios (buy recent winners, sell recent losers) across G10 and EM currencies. The portfolios produce annualised excess returns of 7-10% over multi-decade samples, with a Sharpe ratio comparable to equity momentum.
Pukthuanthong-Le, Levich, and Thomas (2007) find similar effects with a longer-horizon focus and document the gradual decay of returns as institutional capital has chased the trade.

The literature is unusually consistent. Across methodologies, samples, and decades, trend-following currency strategies produce statistically significant positive excess returns. In FX specifically, the trend effect has more empirical support than any other style strategy.

This is not a controversial finding among quantitative researchers. The careful version of the question is no longer “is there a trend effect in FX” but “how big is it, how stable is it, and what does it cost to capture”.

Why the effect persists

The standard finance objection to any documented edge is that arbitrage should eliminate it. The trend effect has been documented in print since 1986 and remains tradeable in 2026. Why?

The literature offers several non-exclusive explanations:

The risk explanation. Trend strategies have negative skewness and large drawdowns. They lose money in flight-to-safety regimes and in sharp reversals. The premium may compensate for genuine, undiversifiable risk.
The slow-information explanation. Currencies respond slowly to changes in fundamentals (policy paths, real-yield differentials). Trend captures the gradual repricing as participants update their views, and the slowness is structural because central banks themselves move slowly.
The capital-constraints explanation. Even if professional traders see the edge, they face funding constraints, mandate limits, and career risk that prevent them from scaling into the trade aggressively enough to arbitrage it away.
The behavioural-friction explanation. The drawdowns are bad enough that even professional managers running the strategy are regularly fired by investors who cannot tolerate the equity-curve pattern. Capital flows out exactly when the strategy would mean-revert to its long-run mean. The discipline required is rare.

The likely answer is all four, in proportions that vary by regime. The net result is that the edge has survived 40 years of publication without being eliminated, and that pattern is unusual enough to be informative.

What it costs to capture

The Menkhoff et al. paper documents drawdowns of 30% or more on their currency-momentum portfolio over the sample. The CTA industry, which has run various flavours of this strategy since the 1970s, regularly experiences flat or losing two-to-three-year windows. The crucial fact: the edge appears as a long-run average across decades, not as a smooth equity curve.

In practice this means:

A trader running a trend system should expect long flat periods, including ones that feel like the strategy has stopped working. The literature is clear: those periods are part of the strategy, not a sign it is broken.
Drawdowns of 20-30% are normal. Drawdowns of 40% are not unusual over a long enough window. (See the drawdown asymmetry arithmetic: recovering from a 30% drawdown requires 43% on the remaining capital.)
Position sizing should be aggressive enough to capture the effect when it works and small enough to survive the drawdowns it guarantees. The fixed-percent rule applies, sized conservatively against the realised volatility of the strategy.
The strategy needs to be diversified across many pairs to capture the cross-sectional edge. A trend bet on a single pair is mostly idiosyncratic noise around a small signal.

The combined arithmetic produces a strategy that is intellectually well-supported, statistically real, and psychologically very hard to actually trade. The CTAs that run it professionally have multi-decade track records that include investors fleeing during every major drawdown.

What the retail-marketing version misses

The retail-marketing version of trend following is a long way from the literature. A representative pattern:

Charts of one currency pair, showing a clean trending period with a moving-average crossover signal that “caught the move”.
No mention of the failed crossovers during the ranging periods that bracket the trend.
No mention of the cross-sectional, multi-currency portfolio structure that produces the published returns.
Position-sizing advice that bears no relationship to the realised drawdowns of the strategy.
Marketing copy that promises smooth returns rather than the characteristic CTA equity curve.

The strategy survives the marketing version’s selection bias the same way the literature finds: in the long run, the average catches the trend. The retail trader rarely lasts to the long run, because the experience of the strategy is much worse than the chart-pattern marketing suggests.

Can it be traded at retail

The honest answer is: yes, but the conditions are stricter than the marketing implies. A retail-tradeable version of the strategy requires:

A portfolio of at least 8-10 currency pairs (G10 plus selected EM crosses), not a single-pair signal. The cross-sectional structure is where most of the edge sits.
A lookback window of 3-12 months, not the intraday or weekly windows beginner content tends to focus on.
Volatility-scaled position sizing so that each pair contributes roughly equally to portfolio risk. (See Risk Management Basics.)
Acceptance of drawdowns that will at some point reach 25-30% of starting capital, with no panic adjustment to the system during the drawdown.
A multi-year time horizon. The strategy’s expected return in any given year is dominated by noise; the edge shows up across decades.

Most retail traders find at least one of these conditions unworkable, which is why most retail trend strategies fail. The ones that succeed look more like a quantitative micro-fund than like a discretionary trading style.

The takeaway

Trend following is the FX style strategy with the strongest empirical support in the academic literature. It produces annualised excess returns of roughly 7-10% across long samples, across implementations, across decades. It also produces drawdowns of 30% or more, long flat periods, and an equity-curve shape that punishes traders who lack the discipline to keep running it during the bad windows. The persistence of the edge across 40 years of publication suggests it is a real premium for genuine, undiversifiable risk rather than a market-inefficiency that arbitrage will close.

For retail traders, the honest read is that trend is the most defensible systematic strategy to consider, but only at portfolio scale (8-10 pairs minimum), only with proper risk sizing, and only with a long-enough horizon to let the edge show up across the noise. Anyone selling a single-pair trend system on a short lookback is selling the marketing version of the strategy, not the strategy.

For the other end of the academic FX literature, the carry trade is the close companion piece: a different style premium with a different and sharper risk profile.