Descending Triangle
A Descending Triangle is a bearish pattern with a flat horizontal support and a descending trendline of lower highs, converging until price breaks below the support.
What it measures
At least two touches on a horizontal support and at least two lower highs pressing into it. Confirmation is a close below the support; projected target = triangle height projected downward from the break point.
Parameters
| Name | Default | Range | Note |
|---|---|---|---|
| min_touches | 2 | 2–5 | Minimum touches on both boundaries. |
| max_width_pct | 10 | 2–20 | Maximum triangle width as % of price. |
How it behaves
The mirror of the ascending triangle: buyers keep meeting price at the same level, but sellers are willing to sell for less each attempt. Volume typically thins into the apex and expands on the breakdown.
When it misleads
A "descending triangle" at the bottom of an established downtrend is often a bear trap — the breakdown fails and price rallies. As with all triangles, volume-thin breaks are the primary failure mode. Require directional context (existing downtrend for continuation, or use as a reversal only after clear distribution) and a volume-expanding close beyond the boundary.
Backtest result
Sourced backtest in preparation — we're running the rule on our own historical candles before publishing numbers we can stand behind.
FAQ
- Descending triangle vs falling wedge?
- A falling wedge has two falling lines (converging downward) and is usually bullish. A descending triangle has a flat bottom and is usually bearish. Opposite biases, similar-looking to the untrained eye.
- Which timeframes work best?
- Daily and 4-hour tend to produce the cleanest triangles. Below 15 minutes, apparent triangles resolve within noise and their statistical edge disappears.
- What is the price target?
- The triangle's height at its widest point, projected down from the breakdown. Take partials — targets are hit maybe half the time.
Build a strategy with Descending Triangle
Drop the block into the Nodlow builder and backtest on real historical data.
Try it in the builder