Trend Following

Trend following buys markets that are already rising and sells markets that are already falling, and it stays in until the trend definition breaks. The classic tools are a moving-average cross, a Donchian breakout, a higher-high and higher-low count. The method does not try to sell the high. It sells lower, after the trend has bent, and it relies on the few large winners to pay for many small losses. That payoff shape is the whole design. A trend follower who takes small profits and lets losses run has inverted the method and will not get the method's historical character. You do not get to keep the label if you take the exits of a range trader.

Pick one definition of trend and one exit. A close above the 200-day average, a 20-day high, a weekly higher low. Enter when the definition turns true. Exit when it turns false, or when a trail of N-bar lows is hit. Do not add a discretionary "it looks extended" exit unless you are ready to track that version separately and accept that it is a different system. Trade a basket if you can. Trend following on one instrument is a long wait and a large path dependency. Across many liquid markets, the odds improve that some trend is paying while others chop you. Size each position from its own volatility so one wild market does not dominate the book.

Use trend following as a slow system with pre-committed exits, and judge it over a sample that includes a choppy period. The drawdown in chop is not a surprise. It is the invoice for the trends you want later. Do not pause the system after a drawdown and turn it back on after the next trend has started. That behaviour buys the chop and misses the trend, which is the only way to make a positive-expectancy trend system lose. If you cannot tolerate the drawdown, reduce the fraction. Do not edit the entries.

Taking profit at the first target because a winner feels fragile, then watching the one trend that was going to pay for the year leave without you. Also adding filters until the system only takes the obvious past winners. And running a trend system on a single illiquid name. The method wants markets that can trend and that you can exit. A thin stock can do neither reliably.

Trend following bleeds in ranges and gives back a large open profit at the end of every trend. Those are not implementation errors. A backtest that exits at the extreme is not trend following. Live results will be worse than a frictionless test by the spread, the gap through the stop, and the times you overrode the rule. If you override, your result is the overridden system. Write that down honestly or the method will be blamed for a trade it did not take.