---
title: "Backtest a Hyperliquid Perp Strategy"
description: >-
  A complete Hyperliquid perps backtest: leverage, maker/taker fees, and
  funding settled from recorded venue data, in twenty lines of kScript.
verified:
  engine: 3.0.64
---

Hyperliquid trades as perpetuals: leveraged, funded, and liquidatable. A backtest that ignores those three facts will happily approve a strategy the venue would have destroyed. This walkthrough runs a trend strategy under real perps accounting: isolated margin at 10x, maker/taker fees, and funding settled from the recorded funding rates of the chart's venue.

## The strategy

Open any Hyperliquid perp chart (`HYPERLIQUID_FUTURES`, e.g. `BTC`, on 1h) and paste:

```javascript title="hyperliquid-funding-trend.ks"
//@version=2
strategy(title="HL Funding-Aware Trend", position="onchart", axis=true, initialCapital=10000, instrument="perps", leverage=10, qtyType="percentOfEquity", qtyValue=10, makerFeePercent=0.015, takerFeePercent=0.045, funding="data", slippageBps=2);

timeseries data = ohlcv(symbol=currentSymbol, exchange=currentExchange);
timeseries fast = ema(source=data.close, period=21);
timeseries slow = ema(source=data.close, period=55);

if (crossover(fast, slow)) {
  strategy.entry("Trend", "long");
}
if (crossunder(fast, slow)) {
  strategy.closeAll();
}
if (strategy.positionSize() > 0) {
  strategy.exit("Protect", fromEntry="Trend", stop=strategy.positionAvgPrice() * 0.97);
}

plotLine(value=fast, width=1, colors=["#16a34a"], label=["EMA 21"], desc=["Fast EMA"]);
plotLine(value=slow, width=1, colors=["#dc2626"], label=["EMA 55"], desc=["Slow EMA"]);
```

Reading the declaration, which is doing most of the perps work:

- `instrument="perps"` with `leverage=10`: sizing commits **margin**, not notional. `qtyValue=10` means each entry commits 10% of equity as isolated margin; the position's notional is that margin times leverage. The engine tracks the position's liquidation price from your leverage and maintenance margin, and closes you there if a bar proves or assumes the level traded. See [Perps and Leverage](perps-and-leverage.md).
- `makerFeePercent=0.015, takerFeePercent=0.045`: fees route by fill type, so market entries, stops, and `closeAll` pay taker while limit-bound fills pay maker. Set your own tier's numbers; the declaration ships with the script so a shared backtest carries its own cost assumptions.
- `funding="data"`: the engine settles the actual recorded funding events from the chart's venue against your open position, debiting cash and committed margin together. This is the default; it is written out here because it is the point.

## Reading the result

Three lines to check before believing the equity curve:

1. **`fundingPaid`** in the stats: the signed funding bill. A long that looked fine gross can bleed to death through settlements while it holds; the same rules on a coarser interval hold through more settlements per trade. If part of the window had no recorded funding data, `fundingUnavailableCount` says so rather than pretending.
2. **`liquidationCount`** and trades whose exit says `liquidation`: at 10x, a 3% protective stop and the liquidation level are uncomfortably close neighbors. If liquidations show up, the venue closed you before your stop did.
3. **The fill-precision line** in the run-details popover: whether contested fills were verified with finer recorded data or settled by assumption. See [Reading the Strategy Tester](reading-the-tester.md).

## Tune it like it's real

- **Leverage down first.** At 5x the liquidation level sits twice as far; watch `liquidationCount` go to zero before you tune anything else.
- **Widen the stop or drop the interval.** Brackets arm on the bar after entry, so a tight stop on a coarse chart is exposed for one full bar. [Fill Simulation](fill-simulation.md#choosing-an-interval-to-trust) covers this trade-off.
- **Run the spot twin.** Duplicate the script, delete the perps params, and put both on the chart: compare mode shows exactly what leverage, fees, and funding cost you. That difference is the part most backtests never model.