Before you trust a Hyperliquid trading bot, backtest its net result after fees, slippage and funding, then check whether that result survives on data the bot has never seen. If it does, run it in paper trading on live prices, because a backtest alone can't show how the bot handles real market conditions.

New bots and bold screenshots tend to appear wherever attention goes, so it helps to know how to test them. This guide makes no claim about where prices go next.

First, define what you are testing

"Hyperliquid trading bot" can mean two different things:

  • A bot that trades the HYPE token, on any venue that lists it.
  • A bot that executes on the Hyperliquid exchange, on any of its markets.

The two have different risks. The first depends on one asset's price history. The second depends on one venue's fees, funding, order books and technical limits. Write down the market, venue, timeframe and order type before you run any test, or your results may not apply to what you later run.

What to backtest first

1. Total trading costs

Start with costs, because they can end the test quickly. Model all three of these:

  • Fees: use the fee tier you would actually pay, and separate maker fees from taker fees. Many backtests assume limit orders always fill, which is rarely true.
  • Slippage: estimate this from order book depth at your intended order size, not from a flat guess. Depth can vary widely between markets and times of day.
  • Funding: perpetual futures charge or pay funding while a position is open. Funding intervals differ by venue, so check the current rules in the exchange docs, and don't reuse assumptions from another venue.

2. How much history you really have

HYPE has a relatively short price history, so it covers a limited number of market conditions. A short history creates two problems:

  • The bot may have learned one market phase rather than a repeatable pattern.
  • You have fewer independent periods to test against.

A strong result on one short period is weak evidence. Our guide on how many trades a backtest needs explains why sample size matters. Testing the same rules on other liquid markets can also help, though a good result there doesn't prove the bot will work on HYPE.

3. Out-of-sample and walk-forward results

Split your data before you tune anything. Optimise on one part and keep the later part untouched. Better still, use walk-forward testing: tune on a window, test on the next window, roll forward and repeat.

Look for these warning signs:

  • Most of the result comes from a few trades or one short stretch.
  • Small parameter changes turn a gain into a loss.
  • Out-of-sample results are far weaker than in-sample results.

4. How many versions you tried

If you test hundreds of parameter combinations and keep the best, its score is inflated by selection. Record every variant you run. The deflated Sharpe ratio is one way to adjust for this, and it only works if you count the attempts honestly.

5. Market exposure versus edge

A long-biased bot can look clever simply because the market rose during the test. Compare the bot with buy-and-hold over the same dates, and split its results into rising, falling and sideways periods.

If most of the return appears only when the market rises, the bot may just be carrying market exposure. For a coin that tends to move with BTC, also measure the correlation with BTC, because many of your "different" trades may be one bet.

6. Drawdown and liquidation

Maximum drawdown should be measured on a mark-to-market basis, including open positions, not just on closed trades. If the bot uses leverage, replay the worst periods with liquidation rules included. A backtest that ignores liquidation can show a position recovering when a real account would already have been closed out.

Worked example: how costs change the picture

Think of a bot's result as two layers. The gross result is what the entry and exit prices alone would have produced. The net result is what remains after fees on both sides of each trade, slippage on both sides, and funding paid for the hours a position stays open.

For a bot that trades often and holds for a few hours, each of those costs is small on its own, but they are charged on every trade, so they add up in the same way the gross edge does. A gross edge that looks healthy per trade can shrink to almost nothing once all three are subtracted, and the summed result over a whole backtest can shrink by the same proportion.

Now stress the test. Assume slippage is worse than your first estimate, as it would be in a thin market or a fast move, and leave everything else unchanged. If the bot's gross edge per trade is only slightly larger than its total costs, that single change can turn a small net gain into a net loss over the whole test.

The strategy didn't change. Only one cost assumption did, and that was enough to flip the outcome. A bot whose result depends that heavily on a cost assumption has very little margin for error, and you should treat its backtest as fragile until you have tested it with pessimistic costs.

Move to paper trading, and measure the gap

A backtest is a controlled replay. Paper trading uses live prices, simulated money and rules fixed in advance, so it can reveal problems a replay hides, such as signal delays, data gaps and bugs in how the bot runs.

Paper trading still has limits. Your simulated orders don't move the market, and fills can look better than they would in reality. Our guide to honest paper trading for crypto futures covers how to reduce this gap.

Set these rules before you start:

  1. Fix the rules and parameters, with no tuning during the test.
  2. Decide the minimum number of trades and the test length in advance.
  3. Compare results with the backtest's trade frequency, average trade and drawdown.
  4. Stop and investigate if the differences are larger than you expected.

What a high win rate can hide

Our own published data shows why a single number isn't enough. BOTLAB tracks bots against 6 statistical gates (G1–G6), defined on the site. All figures below are paper trading, meaning simulated money at real prices:

  • Combo4h-paper/blocked: 64 trades from 12 October to 3 November 2026, with a 75.0% win rate. It passed 5 of 6 gates, has the status "not ready", and its verdict is "market beta, no proven edge".
  • mr4h/long: 143 trades from 16 August to 4 November 2026, with a 72.7% win rate. It passed all 6 gates, and its verdict is still "sample too small to conclude".

Neither result says anything about how a bot would behave on Hyperliquid or HYPE, and neither predicts future results. They simply show that a high win rate doesn't settle the question.

To run a basic review of your own bot, try Check your bot. The BOTLAB platform page explains how we monitor and report on bots.

Practical checks beyond the backtest

Some risks won't appear in a price-based test:

  • Permissions: API keys should be trade-only, with withdrawals off and IP restrictions on where supported. On wallet-based venues, never give anyone your seed phrase or main private key. Use a delegated trading wallet if the venue offers one, and read its documentation first.
  • Technical failure: test what happens if the connection drops, an order is rejected or a rate limit is hit. Know how open positions get closed if the bot stops.
  • Venue risk: exchanges can change fees, margin rules and listings, and they can suffer outages. No backtest captures all of this.
  • Local rules: check whether a venue is available where you live, and follow the rules that apply to you.

FAQ

Can I backtest a Hyperliquid trading bot with free data?

Often, yes, at least for candle data, but candles alone are usually not enough. For realistic results, you also need historical funding rates and some estimate of order book depth. Check the data for gaps before you trust the output.

Are HYPE backtest results reliable?

They are limited by a relatively short price history. Treat them as one piece of evidence, and add out-of-sample tests, cost stress tests and a paper trading period.

How long should I paper trade a bot?

There is no fixed length. Set a minimum trade count in advance and try to include different market conditions. A slow bot needs far more calendar time than a fast one to produce the same sample.

Does a good backtest mean the bot will work live?

No. It only means the rules would have produced that result on past data under your assumptions. Live trading adds real fills, delays, outages and changing markets, and it can lose money.

Key takeaways

  • Define whether you are testing the HYPE token, the Hyperliquid venue, or both.
  • Model fees, slippage and funding first, then stress them.
  • Treat short history and heavy tuning as reasons for extra caution.
  • Compare results with buy-and-hold to separate market exposure from edge.
  • Paper trade with fixed rules before any live decision, and measure the gap against your backtest.

Not available in restricted jurisdictions (see the Service Agreement). You trade with your own exchange account; BOTLAB never holds funds.

Paper trading results are simulated. Past results don't predict future results. Crypto trading can lose money, including all capital. Not financial advice.

Want these checks run on your own strategy? See the strategy and backtest audit.

Originally published on BOTLAB. Score your own bot's trade log free: Check your bot.