5.3 Trend Identification and Following

Defining a trend (5.1) and placing it in a Wyckoff phase (5.2) still leaves the operational question: how do you identify the movement in real time, confirm that a breakout is real, follow it, and size it so one wrong call cannot wreck the account? That process is a Classical Techniques skill on CMT Level I, and it is also the bridge into later risk-management units. This OpenExamPrep section is independent teaching on identification, moving-average methods, the trend-following loop, reward-to-risk, and general risk rules—especially position size, invalidation, and not averaging blindly against the identified trend.

Why identification comes first

Every one of the four primary trades is a conditional tactic. Breakout entries assume the boundary failed. Rangebound entries assume it held. Continuation assumes the trend is still the three-movement label you gave it. Termination assumes that label just broke. If identification is wrong, the rest of the playbook is a well-executed error.

Identification also sets invalidation. An uptrend identified from higher lows is wrong under the last meaningful HL, not under an arbitrary round number. A 50-day average filter is wrong on a close back through that average if that was your chosen mathematical definition. You cannot follow what you have not defined, and you cannot size a stop you have not named.

Exam trap: reading a pretty continuation pattern on a stock that is still a two-sided range on the chart in the stem. Identify first. Patterns later.

Ways to confirm breakouts

A breakout is a close (or a sequence of closes) beyond a stated support or resistance, not an intraday wick that finishes back inside the range. False breaks are common; confirmation filters exist to reduce, not eliminate, them.

You will not use every filter on every chart. A gap-up through a well-tested ceiling on huge volume may not wait for a three-day time filter. A sleepy range on a dull stock may need the close and the retest. Level I wants you to name the toolkit and to reject the idea that any poke through a line is a confirmed break.

Bull trap: upside poke that fails, trapping longs above resistance. Bear trap: downside poke that fails, trapping shorts below support. Springs and upthrusts from 5.2 are structural names for those traps inside Wyckoff ranges.

Worked confirmation

Resistance at 50. Day 1 high is 50.40, close 49.70—not confirmed. Day 2 closes 51.60 on volume about 1.8 times the 20-day average, then Day 3 closes 52.10. A later throwback holds 50.30. Closing filter, volume, follow-through, and polarity all agreed. A candidate who bought the Day 1 wick was trading an unconfirmed break.

Manual ways of identifying trend

Manual methods use your eye and a pencil (or equivalent on a screen) on the price path itself.

• Peak-and-trough analysis. Label HH/HL versus LH/LL versus a range, as in 5.1. This remains the definition. Averages and lines are helpers.
• Trendlines. An uptrend line connects rising swing lows (at least two points to draw; a third touch validates). A downtrend line connects falling swing highs. A decisive close through a well-touched line is structural damage, especially if it also breaks the last swing.
• Channels. A parallel line on the opposite side of the trendline frames typical retracement height. Channel overthrows can warn of exhaustion; they are not automatic termination without a swing break.
• Horizontal support and resistance. Prior swing clusters, range edges, and gap boundaries. Polarity lives here.

Manual methods fail when the analyst pick-a-line shopping: drawing a new trendline after every bar so that the trend is never wrong. Require well-spaced touches and accept that a break is information.

Mathematical ways of identifying trend

Mathematical methods replace (or corroborate) the eye with a formula. They lag—by construction—because they average or look back. Lag is the price of reducing argument about which wiggle mattered.

• Price versus a moving average (MA). Discussed next as the one-average method.
• Two moving averages. Fast versus slow, discussed as the two-average method.
• Donchian / channel breakout. An N-period high or low (a 20-day close above the prior 20-day high is a mechanical upside break). This is identification by new extreme, not by slope.
• Linear regression / least-squares line. A fitted slope over a window: positive slope as an uptrend statistic, negative as downtrend. Useful as a description; still needs an invalidation rule.
• Trend-strength overlays (for example Wilder's Average Directional Index (ADX)) answer how strong the movement is, not which way. ADX high plus falling prices is a strong markdown, not a buy signal.

Mathematical identification is at its worst in ranges, where averages flatten and whip, Donchian highs and lows are the range edges themselves, and regression slopes hover near zero. That is a feature: a whipped MA is often telling you the three-movement label is range, so the matching trade is rangebound, not continuation.

Identifying trends with one moving average

A moving average is a lagging summary of past closes (or other prices). For identification, the common one-average rules are:

1. Filter: treat the market as uptrend-eligible when price is above a rising average; downtrend-eligible when price is below a falling average.
2. Cross signal: a close through the average changes the identified state (with lag).
3. Slope: a rising average is itself a simple trend statistic; a flat average is a range warning.

Length chooses the fractal. Short averages (about 10–20 periods) track minor swings and whip more. Intermediate averages (about 50 periods) track the swing many daily traders call the trend. Long averages (about 150–200 periods) track the larger movement. The later moving-average unit covers simple versus exponential construction. This unit only needs the identification use: price relative to one average, plus that average's slope.

In practice: daily close 102, 50-day average 98 and rising → one-average uptrend filter. Close 96 with the average rolling over → the filter has failed; you are at least in a warning, and possibly in a range or a downtrend depending on swings. Do not keep calling it an uptrend solely because last month was strong.

One-average methods follow as well as identify: the average can be a trailing line under markup (see following, below). They do not forecast. A price that is extended far above a short average can still be in markup and still be a terrible entry on reward-to-risk.

Identifying trends with two moving averages

A two-average method uses a fast average and a slow average.

• Regime filter: uptrend identification when the fast average is above a rising slow average, especially if price is also above the fast average (a stack). Downtrend identification when the fast is below a falling slow average and price is below the fast.
• Crossover signal: the fast crossing above the slow is a bullish identification change (the popular golden cross is often a 50-period average crossing above a 200-period average). The fast crossing below the slow is a bearish change (the popular death cross is the 50 below the 200). These are lagging regime labels, not precise turning-point timers.
• Spread: a widening gap between fast and slow often accompanies a strong markup or markdown; a narrowing gap often accompanies a slowing trend or a range.

Typical pairs: 10 and 30, 20 and 50, 50 and 200. Shorter pairs identify faster fractals and whip more. The 50/200 pair identifies a large-scale movement and will stay bullish deep into a secondary reaction that a 20-day chart already calls a range or a downtrend. That is fractal structure again, not a broken indicator.

Exam trap: treating a 50/200 golden cross as a guarantee of immediate upside, or treating two averages as if they abolished ranges. Dual averages still whip when price is boxed. In a range, two-average crossovers are often the wrong primary trade (they manufacture fake breakouts). Use them as a regime filter and still require swing structure.

The trend-following process from beginning to end

A complete loop, in order:

1. Choose the instrument and the working timeframe, plus one higher timeframe for context (fractal).
2. Identify the state with manual swings and, if you use them, one- or two-average filters: uptrend, downtrend, or range; Wyckoff phase if the cycle is readable.
3. Select the matching primary trade (breakout, rangebound, continuation, or termination). If the state is a range, do not force a continuation long.
4. Wait for a trigger that belongs to that family: a confirmed close through a level, a rejection at a held edge, a dip to a still-valid HL, or structural failure.
5. Mark invalidation before entry: the price that proves the identification wrong (below the last HL, below the throwback low, below the average you used as the filter, above the last LH for shorts).
6. Mark a structural target (next resistance/support, opposite side of the range, prior swing, measured height). Fantasy targets that sit in empty space do not count.
7. Compute reward-to-risk. Skip the trade if the payoff is too small relative to the stop, even if the identification is pretty.
8. Size the position from the account risk budget and the stop distance.
9. Enter. Prefer the trigger you planned, not a market order after a 6% runaway bar that now offers 0.4R to the next ceiling.
10. Follow: trail invalidation with new HLs or a moving-average / ATR rule; scale only with the still-valid trend; exit on opposite identification or at the target.
11. Record the identification, the invalidation, the R multiple, and whether you followed the plan. The journal is how you stop negotiating with a lost identification.

That is trend-following as a process, not as a personality type. Discretion still lives in steps 2–4. Mathematics still lives in steps 5–8 and 10. Skipping 5–8 is how a correct trend call becomes a random bet size.

Reward-to-risk ratio

Reward-to-risk (R:R) is potential reward divided by the amount risked to invalidation.

For a long: R:R = (target − entry) / (entry − stop). For a short: R:R = (entry − target) / (stop − entry). If you buy 72, invalidation is 69, and the next structural resistance is 81, you risk 3 points to make 9. R:R = 3:1. The 3-point risk is 1R. A +9 outcome is +3R.

Trend-following systems often win less than half the time because many breakouts fail and many continuation entries stop at the last swing. They can still have positive expectancy if winners are larger than losers. Expectancy per trade ≈ (win rate × average win) − (loss rate × average loss). At a 40% win rate:

• 1:1 payoff → 0.40×1 − 0.60×1 = −0.20R (a leak)
• 2:1 payoff → 0.40×2 − 0.60×1 = +0.20R
• 3:1 payoff → 0.40×3 − 0.60×1 = +0.60R

Break-even win rate is 1 / (1 + R:R): 50% at 1:1, about 33% at 2:1, 25% at 3:1. You do not invent a 10:1 target in empty sky just to decorate the ratio. The target must be a real structure. A 3:1 sketch that assumes you will hold through the next three earnings prints without a stop is not a 3:1 trade; it is a hope.

Manual and mathematical methods of following trends

Following is what you do after entry so that a good identification can pay.

Manual following:

• Raise a long stop under each new higher low (or under the throwback low, then under later HLs). Lower a short stop above each new lower high.
• Redraw the trendline as new valid touches appear; a close through the line plus a swing break is an exit cue.
• Honor polarity failure: if the level that was supposed to be new support gives way, the breakout you followed is no longer confirmed.

Mathematical following:

• Average trail: exit a long on a close back under the same average you used to identify the uptrend (for example the 20-day or 50-day). This is slow in a crash and whippy in a range—know which state you are in.
• Volatility trail: a Chandelier-style stop a multiple of Average True Range (ATR) beneath the high watermark of the trade (or above the low watermark for shorts).
• Dual-average reverse: flatten or flip when the fast average crosses back through the slow average.
• Donchian reverse: exit a long if price closes back below an N-period low (a stop-and-reverse system).

Manual following is tighter when swings are clean. Mathematical following is more repeatable when you would otherwise argue about which HL counts. Many practitioners combine them: identify with swings, follow with an ATR or average trail that is beyond the last HL so ordinary noise does not shake you out—then still exit if the HL itself breaks.

Important general risk-management principles

Three principles belong in this unit because identification without them is sightseeing.

Position size from the stop, not from conviction

Decide how much of the account one idea may lose if invalidation hits (a common teaching band is a fraction of a percent to about 1–2% per trade; Level I wants the method, not a branded percentage). Then:

Position size = (account × risk fraction) / (price distance to invalidation, in currency per unit).

Example: account 40,000; risk 1% = 400. Long entry 50; invalidation 47; risk 3 per share. Size = 400 / 3 ≈ 133 shares. If you instead buy 800 shares because you are sure, you have replaced identification with a hidden 6% account bet. Conviction is not a sizing input.

The same math sets a maximum size when the stop is tight and a minimum skip when the stop is so wide that even a tiny share count still exceeds the risk budget. Wide invalidation is not solved by shrinking the stop to a place that is not actually invalidation.

Invalidation is a price, not a mood

Write the level before the fill. For an HL-defined long, a close under that HL is the thesis failing. For an average-defined long, a close under the average is the thesis failing. After it fails, you exit or fully re-identify. You do not wait to feel ready. A new trade requires a new structure, a new stop, and a new R:R—not a hope that the old one will come back.

Do not average blindly against the identified trend

Averaging down a long after the market has already printed a lower high and broken the last higher low is not trend-following. It is arguing with markdown (or with a failed breakout). The identified trend said demand lost. Adding size against that identification raises the loss if the new structure is correct.

Contrast that with pyramiding with the trend: adding on a later HL in still-intact markup, with the original invalidation (or a raised one) still defining total risk. Adding to winners that keep structure is a continuation skill. Adding to losers that have already invalidated structure is how accounts and careers end. The charter path is a multi-year professional project; one doubled-down fight with a markdown is not a study plan.

Related rules that sit in the same family: do not size as if five correlated longs were five independent ideas; do not skip the stop because leftover time on the Prometric clock is short; do not replace a structural target with a round number that sits inside the same bar's noise.

Worked loop

Daily chart: HH/HL from 60 to 72; 20-day average rising under price; last HL at 69. You identify markup. Continuation long on a dip to 70.80 that holds above 69. Invalidation: close under 69 (manual) or under the 20-day average if you chose that mathematical follow. Target: prior measured height projecting near 81, which is also visible resistance. Risk 1.80 if the stop is 69; reward about 10.20 if 81 is real structure; R:R about 5.7:1 before costs—if 81 is honest. Size from 1% of 40,000 and a 1.80 stop ≈ 222 shares. Price later closes 68.40. Identification has failed. You exit. You do not buy more at 67 to improve the average. If a new accumulation range forms, that is a new identification and a new trade.

Key Takeaways

• Identify state first; the four trades are conditional on that state
• Confirm breaks with closes, filters, volume, follow-through, and held retests
• One average: price versus a rising/falling MA; two averages: fast versus slow stack and cross
• R:R = structural reward / distance to invalidation; lagging MA methods need that math too
• Size from the stop; do not average blindly against a trend that has already invalidated