Trend following is a family of methods that sets exposure using a rule about the market’s previous direction. A breakout, moving-average crossover and own past return are different signals, not one measured strategy. Each needs execution, exit, sizing and risk rules; recognizing direction does not identify the beginning or end of the next trend.

The philosophy of trend following — react, do not predict

A rule responds to information available at decision time. For example, a long-only model can buy after the Close exceeds an earlier channel high and exit after a specified decline. Other versions may short or use returns instead. “Reacting” does not eliminate the uncertainty of future prices or execution.

A few large gains can contribute much of a specified system’s result, but neither a trend-following label nor a market’s assumed time in consolidation establishes positive expectancy. Measure all wins, losses, costs and exposure rather than assigning one success rate to the family.

Historical context is separate from a testable rule

The Original Turtle archive describes Richard Dennis and Bill Eckhardt recruiting and training traders at the end of 1983, with small accounts beginning in January. Its system page lists markets, sizing, entries, stops, exits and tactics as separate components. A reported historical story is not an audited test of the model below or proof that a present implementation will profit.

Preserve the rule set and record deviations when evaluating a mechanical method. Compliance answers whether instructions were followed; net results and uncertainty answer different questions. Participant anecdotes cannot isolate discipline as the causal explanation of every success or failure.

Close breaks the threshold because an old high expires

Before bar 4, the three preceding Highs are 110, 103 and 104: threshold 110, so Close 4 = 105 does not exceed it. Before bar 5, the window moves to 103, 104 and 105: threshold 105. Close 5 = 105.1 passes the strict test while remaining 4.9 units below the old 110.

A signal after the old maximum expiresFictional Highs in bars 1–5: 110, 103, 104, 105, 105.1. Closes 4 and 5: 105 and 105.1. Before bar 4, the prior-three-bar maximum is 110; before bar 5 it is 105. Price rises 0.1 while threshold falls 5. The fifth Close exceeds the current window, not the old 110.The window changed, not just CloseOld High: 110Threshold: 105Close 4: 105Close 5: 105.112345Price: +0.1; threshold: −5Threshold excludes the current bar
Green points show fictional bar Highs. Blue spans mark thresholds available before bars 4 and 5; their step comes from the shifted window.

Record Close, threshold and window bar IDs together. Price rises only 0.1 here while the threshold falls 5. A breakout marker does not establish an all-history high. Do not interpret the window change as extra price movement; it follows from the selected algorithm.

Assumptions: Fictional Highs and Closes in arbitrary units. Our rule requires a completed current bar’s Close to exceed the maximum High of its three completed predecessors. Exclude the current bar from the threshold; each supplied bar has High no lower than Close. Official material: TradingView: Donchian Channels.

Classic entries: Donchian channels and moving-average crossovers

Donchian boundaries use the highest High and lowest Low of a specified window. For our Close-based signal define U(t) = max H(t−N … t−1), and D(t) = min L(t−M … t−1), explicitly excluding the evaluated bar. If H(t) were included, strict C(t) > U(t) would be impossible because C(t) ≤ H(t). Using final intrabar information before it exists is a separate lookahead error. See the Donchian channel strategy.

Our long-only illustration: N = 3, M = 2
Data and entryDaily EUR/USD Bid bars; after Close exceeds prior-three High, buy at the next available Ask.
ExitAfter Close falls below prior-two Low, sell at the next available Bid.
ProtectionAssigned SL 1.0980, with activation preceding a close-based exit; equality does not pass strict inequalities.
Exposure10,000 EUR, one position, no additions or fixed target. After closure, evaluate a new entry at a subsequent Close; no automatic short.

Moving-average crossings, such as chosen 50/200 periods, define a different signal set. An extra filter may change entries and timing without necessarily reducing false signals or improving net results. A combined breakout strategy needs its own evaluation rather than inheriting another method’s performance.

Two synthetic paths have Closes 100,102,103,105 and 100,115,104,105. Their preceding three Highs are respectively 101,103,104 and 101,120,106. Both have endpoint change 105/100 − 1 = +5%; only the first has C4 above max H1–H3,105 > 104 rather than 105 > 120. This distinguishes a simple sign rule from a channel without determining profitability. It omits income/funding and does not reconstruct the long-horizon excess-return study.

A valid signal can still end in a net loss

Hurst, Ooi and Pedersen's 2017 paper reconstructs historical time-series momentum using data extending back to 1880. Its findings concern that defined model across several asset classes. They do not establish the profitability of every breakout strategy or a retail forex account. Evaluate later periods and any different entry, sizing or exit rules separately.

Use entirely fictional Bid data, no earlier history and no starting position. The table supplies H/L/C needed by our decision rule; assume other opening prices stay within their bar ranges. After bar 4 closes, the entry signal is known; at bar 5’s beginning assign Bid 1.1031 and actual buy Ask 1.1033. Do not backdate the fill to channel High 1.1020.

BarHigh / Low / CloseAvailable threshold / evaluation
11.1010 / 1.0990 / 1.1000Insufficient prior-three history
21.1020 / 1.1000 / 1.1010Insufficient prior-three history
31.1015 / 1.0995 / 1.1005History building; no entry
41.1035 / 1.1000 / 1.1030U = max H1–H3 = 1.1020; entry signal
51.1060 / 1.1025 / 1.1050D = min L3–L4 = 1.0995; no exit
61.1065 / 1.1040 / 1.1045D = min L4–L5 = 1.1000; no exit
71.1050 / 1.1015 / 1.1020D = min L5–L6 = 1.1025; exit signal

Assign actual full sale at bar 8’s beginning, Bid 1.1018. SL 1.0980 was not activated on this path. A close-based exit is not an intrabar stop at the channel: the model waits for the Close. Gross P/L is 10,000 × (1.1018 − 1.1033) = −15 USD. Assigned total commission 4 and funding 1 leave −20 net, with a USD account and no conversion. Spread is already in Ask/Bid fills. This is a cost model, not a broker offer.

Backdated exit at H6 = 1.1065 would require a different rule and information unavailable at that high. No fixed target allows exposure to continue but does not assure a large winner.

Three pillars of risk management

For a hypothetical sizing rule risking 1% of current equity at each trade, ten losses of exactly that amount leave 0.99^10 of starting equity: a drawdown of about 9.5618%, before additional costs. At 0.5%, the corresponding calculation is 1 − 0.995^10, about 4.8890%. Ten and five percent are simple sums only when each loss uses a fixed starting-equity cash amount. State the base and remember actual stop losses can differ.

From buy 1.1033 to SL 1.0980, the assigned initial distance is 53 pips, or 53 USD for 10,000 EUR before costs. This is not a maximum loss; an ordinary stop may fill worse. Check quantity steps, margin, fees, funding and combined currency exposure, as covered in the risk-management section. Different symbols need not diversify a shared risk. A data fault or contract restriction can justify withholding an entry under a stated rule; log that separately from discretionary skipping.

What to do tomorrow if you want to implement trend following

Trend following has been studied over long historical samples, but a particular retail implementation needs its own validation. Neither a 20-30 percent drawdown nor a win rate below 40 percent is a universal property or an acceptable risk allowance. Before choosing a system, examine its tested drawdowns and loss sequences alongside the proposed leverage and costs.

  1. Record the exact preceding-bar window, price side, decision time and executable instruction. Reproduce signals by hand and compare them with simulator fills; a default next-tick emulator rule is not a broker guarantee.
  2. Evaluate net results, marked-equity drawdown, time below prior highs, loss sequences and contribution from large trades. Preserve all tried settings and an untouched period; no single best historical setting establishes future profitability.
  3. Specify data/contract/exposure fault conditions and their response. Historical anecdotes, ten years of data, a six-month trial or mechanical compliance do not guarantee a later winner will recover losses.