TL;DR: A liquidation trading strategy reacts to bursts of forced closures: either fading them on the theory that forced sellers (or buyers) are exhausted, or following them on the theory that the cascade has further to run. Liquidation spikes work better as a timing trigger inside a larger framework (trend, open interest, volatility) than as a standalone signal. In Trigr you can use liquidations as a trigger or filter, but the real history is short, so treat every result as a hypothesis for forward testing.
What does liquidation data actually measure?
On a perpetual futures exchange, a position is liquidated when its margin falls below the maintenance requirement. The exchange closes it at market, regardless of whether that is a good price. Hyperliquid's liquidation documentation describes when and how that happens on its venue.
Liquidation data aggregates those forced closures over time, usually split by side:
- Long liquidations: leveraged longs forced to sell. They show up during sharp drops.
- Short liquidations: leveraged shorts forced to buy. They show up during sharp rallies, often called short squeezes.
The key property is that liquidations are involuntary. A trader choosing to sell carries information about their view. A liquidation carries information about leverage and positioning: somebody was overextended, and the market found them.
Why do liquidation cascades happen?
A cascade is a feedback loop. Price falls, the most leveraged longs are liquidated, their forced market sells push price lower, which liquidates the next tier of longs, and so on. The same happens in reverse for shorts.
Cascades are more likely when:
- Open interest is high and rising, meaning a lot of leveraged exposure is in the market.
- Funding is strongly positive (for long cascades), meaning longs are crowded and paying to stay in.
- Order books are thin, so each forced order moves price further.
That is why liquidation strategies rarely stand alone. The spike tells you something happened; open interest, funding and trend tell you what kind of market it happened in. The companion article on funding-rate and open-interest strategies covers those two inputs in depth.
Exhaustion or continuation: which way do you trade it?
There are two opposite hypotheses, and a good strategy commits to one.
| Approach | Hypothesis | Typical setup | Main risk |
|---|---|---|---|
| Exhaustion (fade) | Forced sellers are done; price overshot and will partly recover | Long after a long-liquidation spike, in an uptrend | The cascade continues for several more bars |
| Continuation (follow) | Liquidations signal a regime break; more forced flow is coming | Short after a long-liquidation spike, below a long-term trend | The spike was the low; you sell the bottom |
| Squeeze fade | Short squeeze overshoots; price gives some back | Short after a short-liquidation spike in a downtrend | Squeeze turns into a genuine trend reversal |
| Squeeze follow | Shorts are trapped and still covering | Long after a short-liquidation spike above trend | Late entry at the top of the squeeze |
The regime filter usually decides which of these makes sense. Fading long liquidations in a higher-timeframe uptrend is the classic "buy the flush" setup. Doing the same in a downtrend is catching a falling knife. Many traders find the fade more intuitive, but there is no general rule that it wins; test each as its own strategy rather than combining them with direction set to both.
How do you build a liquidation strategy in Trigr?
In Trigr's Studio, liquidations are an API data node inside the usual graph: one trigger, optional filters, a signal, and a risk node. The no-code strategy builder guide covers that structure. The liquidations node takes:
- Side: long, short or both.
- Comparator: greater than, less than, at least, at most.
- Multiple of average: for example 2x, meaning liquidation volume is at least twice its recent average. The range runs from 0.5x to 20x.
- Lookback: the window the average is computed over, from 1 hour to 90 days.
Expressing the trigger as a multiple of a rolling average, rather than a fixed dollar amount, matters. Liquidation volume scales with price, open interest and market size, so a fixed threshold that fits BTC means nothing on a smaller asset, and a threshold that fit 2024 may not fit today.
Useful filters to pair with it:
- Higher-timeframe trend: price above or below a long EMA on 4H or 1D.
- Open interest change: open interest falling over the lookback suggests positions were flushed rather than added. In Trigr, open interest is a filter-only node reading a cross-exchange aggregate.
- Funding: negative or falling funding after a long flush suggests the crowd has already de-levered.
- Volatility: ATR or, for BTC and ETH, DVOL to skip entries in the most chaotic conditions.
An illustrative example
An idea to test, not a recommendation:
- Trigger: SOL long liquidations > 3x their 7-day average.
- Filter: SOL 4H price above its 200 EMA.
- Filter: BTC 4H price above its 200 EMA (cross-asset).
- Signal: long when all three hold.
- Risk: 5% size, 2x leverage, long only, max 1 position, 4% stop-loss, ATR trailing stop, time stop at 24 bars.
Keep leverage modest. You are entering right after a violent move, precisely when a second leg down would hurt most. Hyperliquid's margining documentation explains how leverage and maintenance margin set your own liquidation price.
What are the data pitfalls with liquidation series?
Liquidation data is noisier and shallower than price data. Four issues to know before you trust any result.
Short history. Trigr's liquidation series, like its long/short ratio and open-interest series, is hourly, and real history for this family is currently about three months. A liquidation spike 3x above average might occur a handful of times in that window. A backtest with eight trades says almost nothing about the future, however good it looks.
Aggregation. The series is aggregated by a derivatives data provider rather than read only from Hyperliquid's own liquidation engine. That is usually what you want (market-wide positioning), but it means the signal is not the same as "Hyperliquid liquidations," and the size of a move on your venue can differ. The live data catalog shows which assets have the feed.
Timing. The spike is measured over a completed bar. Trigr evaluates every node point-in-time, using only information available at the bar's close, and fills at the next bar's open. If you run a 15-minute strategy on an hourly liquidation series, the last completed hourly value is carried forward and never read ahead. Many hand-built backtests quietly use the liquidation total for the hour that is still in progress, which is exactly the move you are trying to predict. The look-ahead bias guide shows how easily that happens.
Crypto only. Liquidations, funding and open interest are crypto derivatives series. TradFi markets on Hyperliquid HIP-3, such as GOLD or SPX, use OHLCV-only inputs in the current beta.
How should you test and validate it?
Given the short history, validation needs to lean on forward evidence more than for a price-only strategy.
- Run net. Fees and the builder fee are always applied. Turn on slippage (flat basis points per fill) and funding; entries right after a cascade often pay the widest spreads, so be conservative with the slippage number.
- Read the trade log, trade by trade. With few trades, each one matters. Check whether wins came from one or two unusually large rebounds.
- Count your variants. Trying 2x, 3x and 5x thresholds across four lookbacks is twelve trials on a tiny sample. Iterate inside one strategy as labeled experiments so the count stays visible; the best of many tries is usually luck, as the selection bias guide explains.
- Forward test on paper. Freeze the version you like and run it on a paper agent. Paper agents fill at the current Hyperliquid price with fees deducted but no slippage or funding, so they check signal timing and frequency rather than exact costs. The paper trading guide covers what to watch.
Remember that a plain backtest does not report the Deflated Sharpe Ratio or the Probability of Backtest Overfitting, and the shaded trailing 25% in Studio is a recent-period diagnostic, not an out-of-sample holdout.
What this means for you
The benefit of building this in Trigr is that the hard parts are handled consistently: the liquidation node is point-in-time, the fill is next-bar-open, the intrabar rule is conservative, and the same graph that you backtested is the one a paper or live agent runs. What no platform can give you is more history. Accept that, size small, and let the forward record accumulate before committing real capital.
Backtests are not guarantees; perps are leveraged and can lose more than expected, especially in the fast markets where liquidation strategies trade.
Next steps
Check which assets carry the liquidations feed in the capability catalog, then build one version of the idea and test it net of costs. If you prefer to work through an AI assistant, connect ChatGPT, Claude or Codex over MCP and have it draft and backtest the graph for you.