Trend Following

Trend following is the family of trading strategies built on one assumption: an instrument that is already moving in a direction is more likely to keep moving that way than to reverse at any given moment. A trend follower does not try to forecast turning points. It waits until direction becomes measurable, joins it, and stays in the position until the evidence of direction disappears. Everything else — the indicator, the timeframe, the market — is implementation detail wrapped around that single premise.

This reframing has consequences that shape every version of the concept. Entries are late by design, because confirmation costs you the beginning of the move. Most trades are unremarkable, and many are small losses taken when a move fails to develop. The results tend to be carried by a minority of trades that are allowed to run far beyond the average. That asymmetry is not a side effect; it is the mechanism the approach depends on, and it explains most of the design choices below.

## How it works

A complete trend-following system needs four parts, and different authors weight them very differently.

**Trend definition.** Some rule that converts price into a directional state: above or below a moving average, a positive regression slope, a new N-period high, a sequence of higher highs and higher lows. This rule also implicitly defines the horizon of the strategy — a 20-period filter and a 200-period filter describe different phenomena on the same chart.

**Entry trigger.** Once direction is established, the trigger decides *when* to act: immediately on the signal, on a breakout of a level, or on a pullback back toward the trend's reference line.

**Risk unit and sizing.** How much is committed per position, usually expressed as a fixed fraction of equity or normalized by recent volatility so that a quiet market and a violent one produce comparable exposure.

**Exit.** Trailing stops, volatility-based stops, opposite signals, or time-based exits. In trend following the exit does more work than the entry, because it is what determines whether the rare large winner is actually captured.

## Main variants

**Breakout systems** enter on new highs or lows over a lookback window — the lineage most associated with the Turtle rules and Donchian channels. **Moving-average systems** use crossovers, price-versus-average relationships, or slope, and are the most common starting point. **Momentum and rate-of-change systems** rank or filter by the size of recent movement rather than by a level being crossed. **Statistical variants** fit a line to price — linear regression slope and regression channels — to express trend as a measurable gradient with a dispersion band around it. **Volatility-adaptive variants** change their own sensitivity as conditions change, using ATR-scaled channels or adaptive averages. **Structure-based versions** define trend visually through swing highs and lows rather than through an indicator. Finally, **flow-confirmed versions** keep a conventional trend rule but add a second layer — volume behavior, absorption at levels, or open interest — to decide which signals to take.

Counter-trend material sits alongside this family rather than against it: much of it is really about identifying where a trend is exhausted, which is the same regime question from the other side.

## What typically differentiates implementations

Given the same core idea, versions diverge mainly on: the lookback length of the trend filter; the timeframe and intended holding period; breakout versus pullback entry, which trades participation for entry price; exit design, which reshapes the entire distribution of outcomes; whether sizing is flat or volatility-normalized, and whether positions are pyramided; whether the system runs on one instrument or a diversified basket, since much of classical trend following was portfolio-level by construction; and what confirmation layer, if any, filters signals. Two systems described with identical language can behave very differently once these are pinned down.

## Common mistakes

Selecting a lookback after looking at the chart, which produces a rule fitted to a story. Running trend logic in a ranging market with no regime filter and absorbing repeated whipsaws. Cutting winners early or moving stops to breakeven too quickly, which removes exactly the tail the approach relies on. Treating several correlated indicators as independent confirmation when they encode the same information. Sizing each position in isolation while holding several correlated markets. Expecting a high hit rate and abandoning a system during a losing streak that is statistically ordinary. Ignoring execution costs, which fall hardest on breakout entries. And validating on a single symbol over a single favorable period.

## How to evaluate and backtest a version

Start by writing the rules unambiguously, including what happens when signals conflict. Then test across multiple instruments and multi-year windows that deliberately include at least one extended range and one high-volatility period. Examine parameter *neighborhoods* rather than single best values: a robust rule sits on a plateau, not a spike. Reserve out-of-sample data or use walk-forward testing. Model costs, gaps, and fill assumptions honestly, and make sure signals are evaluated on data that would have been available at the time.

Look past headline aggregates to the shape of the results: the contribution of the largest trades, time spent in the market, excursion behavior of open positions, and the *duration* of drawdowns rather than only their depth. Finally, benchmark against something trivial — buy and hold, or a plain moving-average cross on the same data and costs. If the elaborate version does not clearly beat the simple one, the extra complexity is not being paid for.

The versions linked from this page span different markets, timeframes, and confirmation tools. Reading several side by side is the fastest way to separate the shared skeleton of the concept from each author's personal preferences.

Strategies in this concept (40)

Frequently asked questions

Is trend following the same as momentum trading?

They overlap but are not identical. Trend following reacts to a persistent direction and typically holds until that direction ends, with the exit doing most of the work. Momentum trading focuses on the speed and strength of a move, often over shorter horizons, and may exit on a target or on deceleration rather than on a trend reversal. Many practical systems blend the two by using a trend filter for direction and a momentum condition for timing.

What timeframe does trend following work on?

The logic is scale-agnostic: the same rules can be expressed on daily bars or on five-minute bars. What changes with timeframe is the cost structure and the noise level. Shorter timeframes produce more signals, more false starts, and a much higher sensitivity to spread and slippage, so a version that survives on daily data can fail intraday purely on execution costs. Choose the timeframe first, then set the trend-filter lookback to match it.

Why do trend-following systems often have a low proportion of winning trades?

Because they take many small probes to find the few sustained moves. A system that waits for confirmation will repeatedly enter moves that stall, and each of those is closed for a small loss. The design intent is that the trades which do continue are held long enough to more than offset those losses. Judging such a system by the share of winners alone measures the wrong thing — the size relationship between winners and losers matters more.

How do I know whether a market is trending or ranging?

There is no universal answer, only operational definitions. Common ones include the slope of a regression line or a long moving average, the ratio of net directional movement to total path travelled, comparisons between short and long volatility, or simple price structure. Any of these can be encoded as a filter. The important part is picking one definition in advance and testing the system with it, rather than deciding after the fact that a losing period 'was not a real trend'.

Do I need indicators to follow a trend?

No. Price structure — a sequence of higher highs and higher lows, or a break of a prior swing — is a valid trend definition and is used in several versions of this concept. Indicators mainly make the definition explicit and repeatable, which matters if you intend to backtest or automate. A discretionary structural read and an indicator rule can describe the same market state; the indicator version is simply easier to test.

Can trend following and counter-trend logic be combined?

Yes, and pullback entries are the most common form of it: the higher timeframe defines the trend, while a short-term counter-move provides the entry. The risk is that the same tools used to time a pullback can tempt you into fading the trend itself. If you combine them, keep the regime rule and the entry rule strictly separated, and be explicit about which one has authority when they disagree.

All strategy concepts · Explore all strategies