What Is Trend Following? The Complete Guide

Trend following is the practice of trading in the direction a market is already moving, and staying with the position until the move demonstrably ends. It makes no attempt to call tops or bottoms and no attempt to predict anything at all; it reacts to what price has already done, and that reactive character is precisely what makes it so suitable for fully mechanical rules. This guide covers where the style came from, how a trend is actually defined and measured, how entries and exits are built, what the statistics of the style look like, and where it breaks.

• Who created trend following
• Defining a trend by structure
• Moving averages: single lines, crossovers and ribbons
• ADX and measuring trend strength
• Pullback entries versus breakout entries
• Exits and trailing: where the money is made and given back
• The statistical character: fat right tail, low win rate
• How the engine mechanises it
• Strengths and failure modes
• Deeper reading
• Glossary
• FAQ
• In the Systems Library

Who created trend following

Trend following has a longer and better documented lineage than almost any other trading style. Its founding observation, that markets move in persistent directional waves rather than random jitters, was written down before the twentieth century began.

Charles Henry Dow. An anonymous photograph taken before 1931, now in the public domain.

Charles Dow (1851 to 1902) co-founded Dow Jones and Company and The Wall Street Journal, and created the stock averages that still bear his name. In his Journal editorials he described markets as moving in trends of different sizes, primary movements lasting months or years, secondary reactions against them, and daily noise on top, and argued that a trend should be presumed intact until it gives clear evidence of reversal. Dow never wrote a book; after his death, William Peter Hamilton and later Robert Rhea codified his editorials into what became known as Dow Theory: trends persist, they unfold in phases of accumulation, public participation and distribution, and averages should confirm one another before a change of trend is believed. Almost every trend rule in use today is a mechanised descendant of those three ideas.

Jesse Livermore (1877 to 1940), the most famous discretionary speculator of the early twentieth century, supplied the style's motto. His central dictum, paraphrased, was that the big money is never made in the daily fluctuations but in the big swing, in sitting with a position through the main movement rather than trading in and out. Livermore traded by judgement rather than mechanical rule, but his emphasis on riding the major move and pyramiding into strength is the psychological core that later system builders bolted rules onto.

Richard Donchian (1905 to 1993) turned observation into machinery. In 1949 he launched Futures, Inc., widely regarded as the first publicly managed futures fund, and he spent decades publishing weekly rule-based trading guidance. His signature tool, now called the Donchian channel, is brutally simple: track the highest high and lowest low of the last N periods, buy strength through the upper band, exit or reverse through the lower. No forecast, no opinion, just a rule any clerk could execute. Donchian is generally credited as the father of systematic trend following and of the managed futures industry.

The Turtles (1983 to 1988) settled an argument. Chicago trader Richard Dennis believed successful trading could be taught from a rulebook; his partner William Eckhardt suspected it took innate talent. Dennis recruited a group of novices, nicknamed the Turtles, taught them a Donchian-style channel breakout system with strict volatility-based position sizing, and gave them real capital. The experiment became the most famous demonstration in trading history that a written, mechanical trend-following process can be handed to a beginner and followed. The enduring lesson was not the specific rules but the discipline: the hard part was taking every signal, especially after a string of losses.

Ed Seykota extended the mechanical thread into the computer age, building some of the earliest punch-card tested trend systems in the early 1970s, and the same DNA runs through the large managed futures firms, often called CTAs, that today apply diversified trend rules across hundreds of futures markets. The tooling has changed beyond recognition; the underlying logic, cut losses, ride trends, size by volatility, has barely changed since Donchian.

Defining a trend by structure

Before any indicator is involved, a trend has a structural definition that a chartist from Dow's era would recognise: an uptrend is a sequence of higher highs and higher lows, a downtrend is a sequence of lower lows and lower highs. Each swing high is a place where buyers pushed and then paused; each swing low is where sellers tried and failed to reclaim ground. When the lows keep rising, dips are being bought earlier each time, which is the footprint of persistent demand.

Mapping swings sounds trivial and is not. A swing point only exists relative to a scale: a wiggle that is a full swing on a five-minute chart is invisible on a daily chart. Mechanical swing mapping usually defines a swing high as a bar whose high exceeds those of some number of bars on either side, then reads the trend from the sequence of confirmed swings. Two structural events matter most. A continuation is a break above the last swing high while the last swing low holds, the trend stepping forward. A change of character is the first failure, price making a lower low after a run of higher lows, which does not prove reversal but withdraws the presumption of trend. This is Dow's logic made checkable on any bar: the trend is innocent until a structural break makes it guilty.

Moving averages: single lines, crossovers and ribbons

Swing mapping requires structure detection; moving averages compress the same question into one number. A moving average is simply the mean of the last N closes, recomputed each bar, either equally weighted (simple) or weighted towards recent bars (exponential).

A single average gives the crudest trend read: close above the 50-period or 200-period average means the trend is up, by definition. It is blunt, but blunt is not worthless; a rule that keeps a system on the right side of the long average sidesteps a great deal of counter-trend grief.

A crossover uses two averages, fast and slow. When the fast one crosses above the slow one, recent prices have accelerated relative to the older baseline, a mechanical turn signal. Donchian himself popularised an early crossover method, and the famous golden cross and death cross on stock indices are just the 50 and 200 day version.

Stacks and ribbons use several averages at once. A stack demands strict ordering, fast above medium above slow, so every lookback horizon agrees on direction. A ribbon such as the Guppy arrangement plots two groups, several fast averages representing short-term traders and several slow ones representing longer-term money, and reads the trend from their separation: fanned wide apart and ordered means an established trend, tangled and interwoven means chop. The ribbon's real gift is that it makes disagreement visible.

Averages lag by design, and that is a feature. Because an average blends the last N bars, it always answers late; it cannot turn until enough new bars pull it. Critics call this a flaw, but for a trend follower the lag is the filter: the style does not want to react to every twitch, it wants to act only when the move has persisted long enough to drag the averages with it. The cost, paid knowingly, is that entries come after the turn and exits come after the top.

ADX and measuring trend strength

Direction alone is not enough. A market can sit above its average while drifting sideways, and a crossover system in that state will be whipped in and out repeatedly. Welles Wilder's directional movement system asks a separate question: not which way, but how directional is this market at all?

The system has three lines. DI+ measures how much of recent range expansion has been upward; DI- measures how much has been downward. Their relative position gives direction: DI+ above DI- says up-moves are dominating. ADX then smooths the size of the gap between them into a single strength reading from 0 to 100 that deliberately ignores direction. A rising ADX means one side is consistently winning, whichever side that is.

The conventional reading, and the one most mechanical systems encode, is that ADX above roughly 25 marks a trending market and below about 20 marks a ranging one. Two things this does not mean deserve emphasis. First, ADX above 25 does not mean the market is going up; a violent downtrend prints a high ADX just as readily, which is why the reading is paired with the DI lines for direction. Second, a high ADX is a description of the recent past, not a forecast; because ADX is doubly smoothed it lags heavily, and extremely high readings often arrive when a trend is mature. Used honestly, ADX is a regime filter: it tells a system when trend-style rules have been earning their keep and when the market has been the kind that punishes them.

Pullback entries versus breakout entries

Once a trend is established, there are two honest ways in, and they buy different things at different prices.

Breakout entries buy strength. The Donchian and Turtle tradition enters when price exceeds the highest high of the last N bars: the trend has just reasserted itself, so join it. The virtue is that a breakout entry can never miss a major trend, because every major trend must, by definition, break out. The cost is a bad average entry price, at the top of the recent range, and a high proportion of false breakouts that immediately fall back in. The style pairs naturally with breakout trading generally, where the same trigger is used without requiring an established trend behind it.

Pullback entries buy weakness inside strength. The trend is up by the chosen definition, price dips against it, towards a rising average, a prior breakout level, or an oversold oscillator reading, and the system buys the dip on evidence it is ending, such as a strong reversal bar. The entry price is better and the stop can sit closer, below the pullback low, so risk per trade is smaller. The cost is the mirror of the breakout's virtue: the strongest trends barely pull back, and a pullback system can sit out the very move it was built to catch, waiting for a discount that never comes.

Neither is the correct answer. Breakouts trade certainty of participation against price; pullbacks trade price against certainty of participation. Many mature systems blend them, using a pullback entry when one is offered and a breakout entry as the catch-all so no large trend escapes entirely. Both differ from momentum trading, which keys off the recent speed of the move rather than the persistence of the directional state, and both are the philosophical opposite of mean reversion, which sells the very strength a trend follower buys.

Exits and trailing: where the money is made and given back

Ask experienced trend traders what matters and few will talk about entries. The exit is the heart of the style, because the exit is what converts an open trend into banked profit, and every exit rule is a knife-edge trade-off between giving the trend room and giving the profit back.

Channel exits mirror the channel entry: long from a 20-bar high breakout, exit on a 10-bar low. The exit lookback is usually shorter than the entry lookback, so the system demands more evidence to get in than to get out.

Moving-average cross exits close the position when the fast average crosses back below the slow one, or when price closes back through the slow average. Smooth and simple, but the lag that filtered the entry now costs on the way out: by the time the averages cross back, a substantial slice of the peak profit has usually evaporated.

ATR trails scale the exit to the market's own volatility. The chandelier exit is the classic form: hang a stop a multiple of the Average True Range, commonly around three, below the highest high since entry. The stop ratchets up as the trend makes new highs and never moves down. In quiet markets it hugs price; in wild ones it stands back, so the same rule gives every market room proportional to its noise.

All three share one property that defines the style: none of them exits at the top. A trailing exit can only trigger after price has retreated from its peak, so every winning trend trade ends by handing back part of its best profit. That give-back is not a defect to be optimised away; a trail tight enough to keep the peak would be tight enough to get shaken out of the trend long before the peak existed. Which brings us to the statistics.

The statistical character: fat right tail, low win rate

Trend following has a distinctive statistical fingerprint, and understanding it is the difference between running such a system and abandoning it at exactly the wrong moment.

Cut every loss quickly and let every winner run, and the distribution of trade outcomes becomes deeply asymmetric. The left side is a tall stack of small losses and small scratches: failed breakouts, pullbacks that kept falling, trends that died young, each clipped by the initial stop. The right side is a thin, long tail: the rare trades that caught a durable trend and rode it for a multiple of the initial risk. This is the fat right tail, and it carries the entire engine. Remove a handful of the best trades from a typical trend record and the whole edge can vanish, which is exactly why the rules insist on taking every signal; nobody knows in advance which breakout becomes the monster.

Three consequences follow. First, a low win rate is acceptable and normal for the style; it is common for well-behaved trend systems to lose on more trades than they win, because the average winner dwarfs the average loser. A win rate, on its own, says nothing about whether a system is sound. Second, droughts are structural. Trends are not always on offer; when markets range for months, a trend system bleeds small losses the whole time, and long flat or losing stretches punctuated by bursts of profit are the expected shape of the equity curve, not a malfunction. Third, give-back is built in: open profit swells during a trend and part of it is surrendered at every trailing exit, so the account routinely retreats from its high-water mark even while the system works as designed.

Livermore's dictum, Donchian's channels, the Turtles' rulebook: all of it is scaffolding around one asymmetry. Cutting losses caps the left tail; letting winners run leaves the right tail uncapped. That single sentence is the whole engine, and everything else in this article is machinery for enforcing it against human instinct, which reliably wants to do the opposite.

How the engine mechanises it

Inside Nebula, trend following is not one strategy but a family of checkable conditions the evolutionary engine can combine, test and discard. Every condition is something a computer can answer with yes or no on any bar. A few representative examples, paraphrased from the playbooks that ship with library systems, with their honest caveats intact:

• Average stack: fast, medium and slow exponential averages must sit in strict uptrend order, for example the 8-period above the 21, and the 21 above the 55. Every lookback horizon agrees before the condition passes.
• Trend strength gate: the 14-period ADX is above 25 with the positive directional line above the negative one; the market is trending, and trending up. This is a regime filter, a read on the recent past, not a forecast of continuation.
• Ribbon fully fanned: all six fast trader averages sit above all six slow investor averages, the Guppy ribbon completely separated, a strict and unambiguous definition of an established trend that deliberately arrives late.
• Structural break: price closes above the last confirmed swing high on the bar the break is confirmed, the market stepping to a fresh higher high while the prior low holds. This is Dow's continuation event rendered as a rule.
• Chandelier trail: the position is deemed intact while the bar closes above the chandelier line, the highest recent high minus a multiple of ATR, so the trend gets room scaled to its own volatility and the exit only ever ratchets in the trade's favour.

The playbooks are deliberately candid about proxies and limits. A higher-timeframe bias condition, for instance, may be built from a scaled average on the chart being traded rather than genuine data from the slower chart, and the prose says so. None of these conditions predicts anything; each one describes a state of the recent past, and a system passes or fails on how those descriptions combined to behave on data the engine never trained on.

Strengths and failure modes

Strengths. The style is honest about its ignorance: it needs no forecast, no valuation model and no news feed, only price, which makes it unusually robust to being wrong about why a move is happening. It is fully mechanisable, so it can be tested on historical data, run without discretion and audited after the fact. Its losses are structurally capped by the initial stop while its winners are structurally uncapped, a shape most other styles cannot claim. And because trends occur across currencies, indices, metals and rates, the same logic diversifies across markets that often trend at different times.

Whipsaw in ranges. The mirror-image weakness: in a sideways market every breakout fails and every crossover reverses, and the system pays a small loss for each false signal. Death by a thousand cuts is the characteristic failure, which is why strength filters such as ADX exist, and why they help without curing; every filter that avoids more chop also joins real trends later.

Drawdown droughts. Because the profits arrive in rare bursts, the time between bursts is spent flat or bleeding. A trend system can spend long stretches below its previous high-water mark while operating exactly as designed, and no rule change can remove this without also removing the right tail that pays for everything.

Psychological difficulty. A style that loses more often than it wins, surrenders open profit at every exit, and goes quiet for months, is brutally hard for a human to follow by hand. The recurring pattern is abandonment at the bottom of a drought, just before the next burst, and the Turtle experiment's real finding was that following the rules was harder than learning them. This is much of the argument for automation: a machine takes the 12th signal after 11 losses with the same indifference as the 1st.

Regime dependence. Trend following earns when markets trend and pays rent when they do not, and nothing in the rules controls which regime shows up. Results across years are therefore lumpy and regime-driven, and a backtest, however long, shows how the rules behaved on past regimes; it is evidence about robustness, never a promise about the future. Timeframe choice interacts with all of this, since trends exist at every scale but noise does too; see trading timeframes explained for how the same rules read differently across charts.

Deeper reading

• Charles Dow (Wikipedia): the journalist whose editorials started it all.
• Dow theory (Wikipedia): the codified principles of trends, phases and confirmation.
• Trend following (Wikipedia): overview of the style and the managed futures industry built on it.
• Richard Donchian (Wikipedia): the father of systematic trend following and the 1949 managed futures fund.
• Richard Dennis (Wikipedia): the Turtle experiment and the case that rules can be taught.
• Jesse Livermore (Wikipedia): the discretionary speculator behind the big-swing philosophy.
• ADX (StockCharts ChartSchool): full walkthrough of DI+, DI- and the strength reading.
• Moving averages (StockCharts ChartSchool): simple versus exponential, construction and use.

Glossary

Trend

A persistent directional movement in price, defined structurally by successive higher highs and higher lows (up) or lower lows and lower highs (down).

Swing high / swing low

A local turning point in price, typically a bar whose high (or low) exceeds those of a set number of neighbouring bars, used to map trend structure.

Moving average

The mean of the last N closing prices, recomputed each bar; simple averages weight all bars equally, exponential averages weight recent bars more.

Crossover

The event of a faster moving average crossing a slower one, used as a mechanical signal that recent prices have accelerated relative to the older baseline.

Ribbon

Several moving averages of different lengths plotted together; their ordering and separation describe how established a trend is.

ADX

Average Directional Index, Wilder's smoothed measure of trend strength from 0 to 100, deliberately blind to direction; readings above roughly 25 conventionally mark a trending market.

DI+ / DI-

The directional indicator lines of Wilder's system, measuring how much recent range expansion has been upward versus downward; their relative position supplies direction.

Donchian channel

A band formed by the highest high and lowest low of the last N periods; breakouts through it are the classic mechanical trend entry.

Breakout entry

Entering when price exceeds a defined recent extreme, buying strength on the logic that a resuming trend must break out.

Pullback entry

Entering on a dip against an established trend, buying temporary weakness at a better price with a closer stop.

ATR

Average True Range, a rolling measure of how far price typically travels per bar, used to scale stops and position sizes to the market's own volatility.

Chandelier exit

A trailing stop hung a multiple of ATR below the highest high since entry, ratcheting up with the trend and never moving down.

Whipsaw

A quick reversal that triggers a signal and then immediately negates it, the characteristic cost of trend rules in a ranging market.

Fat right tail

The shape of a trend follower's trade distribution: many small capped losses and a thin tail of rare large winners that carry the whole result.

Drawdown

The decline from an equity high-water mark to a subsequent low, expected structurally in trend following during trendless stretches and after trailing exits.

FAQ

Does trend following predict where the market is going?

No, and it does not try to. Every trend rule describes the recent past, price above an average, a fresh high, a strong ADX reading, and takes a position conditional on that state. The bet is that persistence is worth being exposed to, not that any particular move will continue.

Why is a low win rate acceptable in trend following?

Because the style caps losses and leaves winners uncapped, the average winner is designed to be far larger than the average loser. A system can lose on most trades and still be sound if the rare large winners carry the record. Win rate alone says nothing about soundness either way.

What is the difference between trend following and momentum trading?

They overlap heavily. Trend following keys off the persistence of a directional state and stays in until it breaks; momentum trading keys off the recent speed of the move and often trades shorter bursts. Many practical systems blend elements of both.

How does a trend follower know when the trend has ended?

By rule, decided before entry: a close through a trailing line such as a chandelier stop, a moving-average cross back down, a break of the recent swing low, or a channel exit. All of these trigger after the peak, so surrendering part of the best profit is a designed-in cost of the style.

Does trend following work on forex?

Currency markets trend and range like any other market, and trend rules can be tested on them directly. Whether a specific rule set held up is an empirical question, which is why systems in the library are validated on data they never trained on; past behaviour is evidence, not a promise.

In the Systems Library

The Orion RFX Systems Library tags every system by style, including the trend and momentum families described here, and each one ships with a full playbook of its exact conditions, validated on data it never trained on. If you are new to mechanical trading, start with how automated trading systems work, then come back and compare trend following with its faster cousin, its opposite and its entry-timing sibling via the links above.