A Kagi chart builds vertical price legs connected by short horizontal lines. A stated retracement creates a leg in the opposite direction; another hour passing does not. In the classic thickness convention used here, breaking an earlier Kagi shoulder or waist changes yin/yang. Reversing direction and changing thickness are separate events: a thin line can rise and a thick line can fall.

What a Kagi chart is and where it came from

Kagi is a price-based Japanese charting technique discussed in Steve Nison’s Beyond Candlesticks. A continuing new extreme extends the current leg; a smaller counter-move leaves its endpoint unchanged. The minimum retracement threshold must be specified. Source prices, initial state and threshold convention matter, so two charts named Kagi need not reproduce the same history.

A percentage threshold needs a recorded base

Two separate fictional rising legs reached peaks of 100 or 200 units. The latest reading is 4 units below each peak.

Scroll sideways to see the full diagram

A percentage needs a baseTwo separate rising lines with latest peaks 100 or 200 pull back 4 units. Our 3%-of-peak threshold needs 3 or 6 units of retracement; a fixed 3-unit threshold is unchanged.3% OF PEAK ≠ 3 UNITSPeak 10096Peak 200196Red threshold: −3 units.Green threshold: −3% of known peak.100 → 96: both; 200 → 196: only −3.

Our two close-only models: a retracement of at least 3 units, or at least 3% of the latest known peak. We do not attribute this percentage-base convention to a particular platform. The document StockCharts — Kagi Charts confirms that reversal settings can use points, percentages or ATR.

Each panel’s vertical axis measures distance from its own peak; it is not a shared scale of price levels.

A fixed 3% does not mean a fixed 3 units. A peak of 100 needs a 3-unit retreat; a peak of 200 needs 6. Reading 196 passes the fixed threshold (4 ≥ 3), but not the percentage threshold (4 < 6). Without a specified base, two exports labelled “3%” lack a complete specification. This tests construction, not signal quality; yin/yang thickness remains a separate rule.

How the reversal amount and the line thickness work

Specify the reversal threshold and its units: an absolute price amount, a percentage with a stated base, or an ATR-derived value. These are different models. Our close-only EUR/USD example uses 0.0030, equal to 30 pips at 0.0001 per pip. A counter-move of at least that amount reverses the leg; a smaller move does not. Any time-varying threshold also needs a declared update rule.

In the StockCharts thickness convention, yin becomes yang strictly above the prior shoulder, and yang becomes yin strictly below the latest waist. Reaching a reversal threshold does not require crossing either level. A thickness change is a chart event; entry and exit rules, fill prices and costs are additional choices. Colours are display settings.

Hypothetical EUR/USD close-only model: reversal and thickness

Initial state: a thin descending line from a prior shoulder 1.1040 to 1.1000. With a 0.0030 reversal threshold, subsequent closes 1.1015,1.1030,1.1040,1.1050,1.1020,1.1000 and 1.0990 have the following effects:

  1. 1.1015 is a 15-pip bounce: the drawn endpoint remains 1.1000. At 1.1030 the 30-pip bounce reverses upward and establishes a waist 1.1000, but the line stays thin.
  2. 1.1040 only touches the prior shoulder. At 1.1050 the line becomes thick. This is known at the 1.1050 observation, even if the final drawing places the thickness boundary at 1.1040.
  3. 1.1020 reverses downward after a 30-pip fall from 1.1050, still thick. Touching the waist at 1.1000 leaves it thick; 1.0990 crosses below and changes it to thin.

Equality is sufficient for our reversal, but thickness needs a strict crossing. The horizontal connector does not represent a fill or a fixed number of hours.

How to read and use this chart step by step

Step 1 — set the chart and the reversal amount

Record the platform’s source field, data interval, threshold unit, session settings and start of history before comparing charts. TradingView offers Traditional, ATR and Percentage(LTP) methods; its projected lines can change before the source interval closes. The Pine documentation notes that non-standard charts use lower-timeframe data as an approximation of tick construction. The technical analysis foundations provide context for distinguishing this transformation from an ordinary time axis.

Step 2 — read thickness, not just direction

Read both state variables. After the model’s 1.1030 close the leg rises while still thin; after 1.1020 it falls while still thick. The labels encode earlier Kagi-level crossings under our convention, not measured demand, supply or who currently controls trading.

Step 3 — wait for a confirmed yin↔yang switch

Do not use a later close to date a switch earlier. In our chronology, the yang transition is first observed at 1.1050, not at the drawn boundary 1.1040. A projection before the source interval closes is provisional. State separately whether a completed transition is used for observation, an order trigger or no trade.

Entry, stop and targets — an illustrative example

Identical aggregate OHLC can also hide different Kagi paths. Start with a thin falling leg from shoulder 104 to 100, reversal amount 3, and strict thickness crossings. Path 100→105→99→102 makes three reversals and finishes rising but thin: 105 crosses 104, 99 crosses below the waist 100, and 102 rebounds without crossing the new shoulder 105. Path 100→99→105→102 makes two reversals and finishes falling but thick: its waist is 99 and the last 102 does not break it. Both paths have O/H/L/C=100/105/99/102. One OHLC bar alone cannot reveal their order.

Changing thresholds changes chart history and must be tested as a separate specification. StockCharts documents rebuilding with an updated ATR; TradingView’s Percentage(LTP) applies the latest closing-price-based size across bars and warns of repainting. Do not infer an unchanged historical drawing merely from the label “percentage”. Define entries, cancellation, exits, size and costs independently; a synthetic crossing is not an executable quote.

Strengths and weaknesses you need to know

Kagi can make selected price-level crossings easier to inspect by omitting counter-moves below its threshold. That is a display property, not evidence that thickness measures order flow, reduces emotions or produces profitable signals.

The horizontal axis is irregular rather than a fixed hourly scale. Observation times can still be linked to source data; Kagi does not erase them. TradingView warns that fills using non-standard synthetic OHLC can produce unrealistic backtests. Test decisions at the moment their inputs were available and model execution from the relevant market prices.

How it compares with Renko, Three Line Break and Heikin-Ashi

Renko charts represent price moves as bricks; specify brick size and reversal convention. Three Line Break charts use preceding line levels; early/two-line and established three-line cases need distinction. Heikin-Ashi keeps regular time periods while transforming all four OHLC values. Kagi adds a thickness state to price-leg reversal. None of these transformations independently measures participant strength.

What to do tomorrow to get comfortable with Kagi

  1. Open any major currency pair in TradingView on the daily timeframe, switch the chart type to "Kagi", and compare the same period with the candlestick chart to see how many small swings the line skipped before it even turned.
  2. Trace reversals and thickness changes separately. Record the first source observation that confirms each change rather than backdating it to the eventual drawn boundary.
  3. Compare declared threshold methods using the same source history, initial state and test protocol. Save parameters before examining later outcomes and include cases that do not trade.
  4. Record source timestamps, threshold and unit, initial state, observed shoulder/waist crossings and actual order fills separately. Add size, costs and exit rules only for a fully specified trade test.
  5. Compare the transformed drawing with ordinary source prices. At our model’s 1.1015 reading, the drawn endpoint still stands at 1.1000: omission from the chart does not remove the 15-pip market move.