Why Trend Following Works in US Equities, and What It Costs

"Trend following works" has been repeated so widely that almost nobody asks why it works or where it stops working. This article tries to answer both. The price of that is that it does not read like an advertisement.

1. What the distribution of stock returns looks like

One set of facts about long-horizon single-stock returns has been reproduced many times: over a long enough window, the overwhelming majority of an index's gain comes from a very small minority of its stocks, while most individual stocks fail to beat cash. The distribution is strongly right-skewed — a large mass of names ends up flat or below, a handful multiply many times over, and that handful carries the average.

That shape determines what kind of strategy can match it. If returns were symmetric and bell-shaped, the optimal approach would look like mean reversion: buy what has fallen, sell what has risen. But in a right-skewed, fat-tailed distribution, mean reversion means selling every genuine multi-bagger shortly after it starts working, while adding steadily to every name on its way to zero.

Trend following's payoff structure is the mirror image of the distribution:

• Stops truncate the left tail — each loss is capped inside a budgetable range.
• No profit target, with the stop trailing upward — the right tail is preserved intact.

In other words, trend following has no predictive power at all. What it does is align the shape of your P&L with the shape of the market, without requiring a forecast. That is the first reason it works, and the most durable one.

2. Why prices persist

If prices followed a pure random walk, the logic above would still bound your losses but could not generate positive expectancy. Trend following needs some degree of price persistence — momentum. Explanations for momentum fall into roughly three groups, and they are not mutually exclusive.

Information diffuses gradually. A material piece of news is not understood and priced by every participant within a second. A change in fundamentals — a structural improvement in margins, say — has to pass through earnings confirmation, sell-side revisions and passive rebalancing. That process unfolds over quarters, and price shows directional continuation as a result.

Behavioural biases are stable. The disposition effect leads investors to sell winners too early and hold losers too long, creating persistent selling pressure in the early stages of an advance so that price takes longer to reach where it is going. From a trend follower's point of view, that delay is the entry window. Anchoring works in the same direction.

Flows create positive feedback. Index additions, rebalancing by risk-parity and volatility-target strategies, and options dealer hedging all generate mechanical "buy higher, sell lower" flow. None of it reflects a fundamental view, and all of it moves price.

An important qualification: none of these mechanisms holds at every time scale. Momentum is weak or reversed intraday, clearest over weeks to months, and gives way to mean reversion over multi-year horizons. The parameters typical of chandelier-style strategies — 22-day breakouts, a 100-day moving average — sit at that scale not for aesthetic reasons but because that is where the evidence for momentum lives.

3. Cost one: a low win rate and a long grind

Trend-following win rates typically run 35%–45%. You will experience consecutive failures — five or six losses in a row is statistically ordinary, not even rare.

Harder still are trendless periods. Historically the most painful stretches for trend following are not crashes but extended range markets: price breaks out and falls back, over and over, the strategy bleeds small amounts continuously, and the equity curve is sanded thin. Such phases can last more than a year, and nothing during them will tell you the strategy is still functioning correctly — because a steady drip of small losses is what functioning correctly looks like.

This is a real cost that cannot be optimised away: trend following demands that you keep executing precisely when there is no positive feedback. Most strategies do not fail in the backtest. They fail in month eight of that phase.

4. Cost two: underperforming buy-and-hold in a strong bull market

This point is usually avoided. In a sustained advance with shallow pullbacks, buy-and-hold is nearly unbeatable:

• Trend following needs a breakout before entering, so it always misses part of the initial move.
• Every shallow pullback can trigger the stop, and after exiting you must wait for the next breakout to return — with the flat period landing squarely inside the advance.
• Every round trip costs commission and slippage.

So on total return in a bull market, trend following usually loses. Its value appears once drawdown enters the comparison: over the same window its max drawdown is typically much smaller than buy-and-hold's. Whether that trade is worthwhile depends on whether you want the highest terminal value or a path you can actually walk to the end. There is no universal answer, but the question must be answered explicitly rather than glossed over by a single number in a report.

5. Cost three: gaps and liquidity

A stop is a price level, not insurance. Gaps driven by earnings, mergers or delisting announcements can pass straight through it, and the difference between the actual fill and the planned loss has no upper bound. These events are far more frequent on single stocks than on broad ETFs, which is why moving a parameter set from an ETF to a small-cap single name causes the backtest's risk metrics to systematically understate real risk.

6. When you should not use it

Stating the boundaries honestly is more useful than another success story:

• When your capital is small and costs dominate. Frequent small trades hand most of the expectancy to commissions.
• When you cannot tolerate the psychological load of consecutive losses. This is a self-knowledge problem, not a willpower problem, and no backtest can test it for you.
• When the instrument has no trend behaviour. Products anchored inside a range by arbitrage will only charge you stop-loss costs.
• When the test window does not contain at least one complete up-and-down cycle. On a one-directional sample you cannot measure what this method is actually for.

Conclusion

Trend following works over the long run because the shape of its P&L matches the shape of the equity return distribution, and that shape is maintained jointly by market structure and human behaviour — it is not going to disappear soon. But its effectiveness is not delivered as a higher win rate or a smoother curve. It is delivered as "a handful of trades produced all of the return." Understanding and accepting that matters far more than finding better parameters.

To see what this looks like on a specific symbol, run the backtest and inspect the trade list: look at the two or three trades that produced the largest gains, and their count relative to everything else.