Trend-following trading takes a directional position after price has already moved and keeps that position while the trend rule remains intact. It does not predict the turn. It accepts being late in exchange for a chance to stay with a move that persists.
That sounds almost too simple. The difficulty is not drawing a trend after it happened. It is defining the signal, exit, sizing and costs before the next move begins.
What trend following actually is
Trend following is a rules-based method that uses an asset's own price history to decide direction. A basic version is long after a positive trend signal and short after a negative one. In academic work this is often called time series momentum, which differs from cross-sectional momentum: the signal compares a market with its own past, not with the recent performance of other markets (Moskowitz, Ooi and Pedersen, Hurst, Ooi and Pedersen).
The method is reactive by design. A trend must exist before a price-only rule can detect it, so entries and exits lag the exact turning points. That is not automatically a flaw. It is the price paid for replacing a discretionary forecast with a repeatable condition.
Three rule families, one underlying bet
Moving-average and breakout models are common ways to turn the idea into a signal. Industry and research primers describe both as standard trend-following implementations, although their exact parameters and portfolio construction can differ materially (CME-hosted research, Graham Capital trend-following primer).
| Rule family | Positive trend condition | Negative trend condition | Main ambiguity |
|---|---|---|---|
| Return sign | Price return over a fixed lookback is positive | The same return is negative | Lookback and rebalance timing |
| Moving average | Price or a fast average is above a slower average | Price or a fast average is below a slower average | Average type and signal timing |
| Breakout | Price exceeds a prior range high | Price falls below a prior range low | Range length and intrabar versus close |
A minimal model can be written without chart interpretation:
direction = +1 when trend signal is positive
direction = -1 when trend signal is negative
position = direction × predefined risk allocation
exit = when the trend signal changes or a separate risk rule firesThis is only a skeleton. A tradable specification still needs a market universe, execution rule, position-sizing method, leverage limits and a treatment for missing or stale prices. A moving-average cross and a channel breakout may agree often, but they are not interchangeable. The Donchian channel explainer covers the breakout branch; the Ichimoku explainer shows how a richer trend framework adds filters and more discretion.
Why the evidence attracts attention
The serious case for trend following comes from diversified research, not from a perfect chart example. Moskowitz, Ooi and Pedersen report time series momentum across futures and forwards in several asset classes. Hurst, Ooi and Pedersen extend the historical test much further back and report positive average performance across varied economic environments (Time Series Momentum, A Century of Evidence).
Those studies test portfolios, not a single indicator dropped onto one chart. Diversification, volatility scaling, rebalancing and instrument selection are part of the result. Remove those pieces and the claim changes. Evidence for a broad, risk-managed futures portfolio does not prove that any one moving-average pair has an edge in one currency pair.
There is also an intuitive mechanism, though it is not settled fact. Prices may adjust gradually to information, and persistent flows can take time to complete. The time-series-momentum paper and the CME-hosted regime study discuss slow information diffusion, autocorrelated flows and later reversal as possible mechanisms; neither turns that story into a universal law (Moskowitz, Ooi and Pedersen, CME-hosted research). A useful trading rule can exist without a tidy story; a tidy story is not evidence that the rule works.
What the critics say
The evidence is disputed. Kim, Tse and Wald argue that much of the reported performance is tied to volatility scaling rather than the directional signal alone. Huang, Li, Wang and Zhou challenge the statistical reliability of pooled tests and find weak asset-by-asset return predictability (Kim, Tse and Wald, Huang, Li, Wang and Zhou).
That disagreement matters because "trend following works" can hide several different propositions:
- past returns predict future direction;
- a diversified long-short portfolio earns a return premium;
- volatility scaling improves the portfolio;
- the strategy helps in some large, sustained market moves;
- one retail implementation survives its own costs.
Those are not the same claim. A backtest must isolate which one it is testing.
Trend following also has an unfriendly payoff rhythm. Gradual, sustained moves are helpful; repeated reversals create false entries and exits. Research on trend-follower regimes and later performance analysis both show that results vary with the size and persistence of market moves rather than arriving smoothly through time (CME-hosted regime study, AQR performance analysis). A strategy can be logically consistent and still spend a long period being wrong in small pieces.
Where a clean backtest usually cheats
Trend-following rules are objective enough to test and easy enough to overfit. The usual defects are mundane:
- Hindsight in the signal: using a completed bar to fill at a price that occurred before that bar closed.
- Parameter mining: choosing the lookback, average type and exit after seeing which combination won.
- Missing friction: ignoring spread, slippage, commission, financing and futures rolls.
- Hidden concentration: treating several highly related positions as independent trends.
- Unbounded sizing: letting a quiet volatility estimate create impractical exposure.
- Survivorship: testing only markets or contracts that remained easy to trade.
None of these objections says the method cannot work. They say the pretty version is not yet evidence.
How you'd actually test it
A useful trend-following test begins with a frozen rule and a benchmark, then asks whether the result survives nearby choices and unseen data.
- Fix the eligible markets using a point-in-time universe.
- Choose one signal family and specify every input, including lookback and bar timing.
- Define long, short and flat states; write the exit separately from the entry.
- State how position risk is set and how total portfolio exposure is capped.
- Execute on the next available price after the signal can genuinely be known.
- Charge spread, slippage, commission, financing and contract-roll costs where applicable.
- Compare against cash, passive exposure and the same signal without volatility scaling.
- Reserve out-of-sample data and test adjacent lookbacks rather than one selected setting.
- Report expectancy, drawdown, turnover, exposure, concentration and performance by market regime.
The key comparison is attribution. If the scaled version wins, did direction add value or did the sizing rule do the work? If the portfolio wins, did most of the result come from one market? If the rule survives in-sample, does it still survive out-of-sample testing and parameter-sensitivity checks?
realbacktesting is a trading-software studio for cTrader built around reproducible tests. For trend following, reproducibility means that another trader can rebuild every signal, fill and position size from the written rules and the same data. That standard matters more than whether the final equity curve looks smooth (how to verify a cTrader backtest).
Frequently asked
Is trend following the same as momentum?
Trend following is closely related to time series momentum because both use an asset's own past movement. Cross-sectional momentum is different: it ranks assets against one another.
Does trend following require a high win rate?
No. A trend-following system can have many small losing trades if occasional persistent moves are large enough to outweigh them. Whether that happens after costs is an empirical question, not a property guaranteed by the label.
Do moving averages predict trends?
No. A moving average summarises past prices. A crossover can define a repeatable trend state, but only an out-of-sample test can show whether that state has useful expectancy.
Is trend following subjective?
It can be discretionary, but systematic versions need not be. Once the signal, timing, sizing and exit are explicit, two independent implementations should produce the same trades from the same data.
Takeaway
Trend following has a credible research record and credible critics. The name proves nothing; the frozen rule, honest costs and out-of-sample result carry the argument.