Backtest 2020–2026: 10,000 simulated wallets
Past results do not guarantee future results. The backtest was built using 1-minute candle data from the Binance exchange.
We simulated the strategy across 10,000 wallets with random initial capital between US$ 15 and US$ 1,000, births spread from 2020 to 2026, and life until October 2026 — 6.8 years covering two bear markets (2021–22 and 2024–25).
The simulation charged real execution costs (0.10% per position, related to Hyperliquid fees and estimated slippage) and the app's 2% fee on each deposit. No wallet got special treatment: same strategy, same rules.
Read the table honestly: most numbers look good, but 46.7% of wallets ended at a loss and the worst ones lost everything. That is what showing both sides means.
| +63.3% | Average return — arithmetic mean across all 10,000 wallets over initial capital. It sits well above the median because few wallets had extreme returns: the mean is pulled up by the right tail. |
| +7.9% | Median return — half the wallets ended below this value, half above. It is the portrait of the typical case, without the effect of outliers. |
| 53.3% | Wallets in profit — 53.3% of the 10,000 ended above initial capital; 46.7% ended at a loss. A strategy with a real edge can be profitable with less than half the wallets winning because average gains exceed average losses. |
| −80.4% | P05 (5th percentile) — 1 in 20 wallets lost 80% or more of capital. The left tail is real and belongs in your decision. |
| +376.9% | P95 (95th percentile) — 1 in 20 wallets ended above +376.9%. The asymmetry of extremes: the mirror of the left tail. (P99: +873.8%.) |
| 44.4% | Doubled their capital — 44.4% of wallets reached, at some point, twice their initial capital. |
| +144.5% | Total auto-withdrawn (average) — sum of the 50%-of-profit withdrawals per event, on average +144.5% of initial capital over the wallet's life. It is money that left the exchange for the user's wallet and stayed out of subsequent risk. |
| 5.1% | Dead wallets — 5.1% lost 100% of capital before the end of the simulation (first death: 7.3 days of life). |
The whole distribution, band by band
| Return band | Final exchange balance | Total withdrawn (% of initial capital) | Final result (balance + withdrawn) | Equity peak (band average) |
|---|---|---|---|---|
| P01–05 | −95.9% | +8.8% | −87.1% | 1.20x |
| P06–10 | −93.8% | +19.5% | −74.3% | 1.28x |
| P11–15 | −92.7% | +28.6% | −64.1% | 1.31x |
| P16–20 | −90.6% | +35.4% | −55.2% | 1.34x |
| P21–25 | −88.8% | +42.8% | −46.0% | 1.39x |
| P26–30 | −87.6% | +50.1% | −37.5% | 1.42x |
| P31–35 | −88.6% | +60.3% | −28.3% | 1.47x |
| P36–40 | −87.3% | +68.2% | −19.1% | 1.59x |
| P41–45 | −86.7% | +77.5% | −9.2% | 1.63x |
| P46–50 | −85.4% | +87.4% | +2.0% | 1.75x |
| P51–55 | −85.9% | +99.3% | +13.4% | 1.88x |
| P56–60 | −84.9% | +111.5% | +26.6% | 2.03x |
| P61–65 | −85.2% | +126.4% | +41.2% | 2.22x |
| P66–70 | −84.3% | +143.8% | +59.5% | 2.41x |
| P71–75 | −84.6% | +164.3% | +79.7% | 2.60x |
| P76–80 | −83.4% | +190.0% | +106.6% | 2.97x |
| P81–85 | −81.3% | +226.5% | +145.2% | 3.38x |
| P86–90 | −81.1% | +279.8% | +198.7% | 4.04x |
| P91–95 | −80.6% | +376.0% | +295.4% | 5.26x |
| P96–100 | −71.9% | +731.3% | +659.4% | 10.35x |
Each row groups 5% of the 10,000 wallets (500 wallets), ordered by final result. Exchange balance, total withdrawn and final result are weighted by the summed initial capital of the wallets in the band — that is why the first two columns add up exactly to the third (rounded values). The equity peak is the average of the wallets in the band.
Important reading: throughout the simulation, the wallets kept running the strategy continuously, each until the end of its period — none was stopped early. And none of the withdrawn amount was reinvested: withdrawals stayed in your wallet, outside the subsequent risk. That is why the final exchange balance is low in every band — the accumulated wealth sits in the total withdrawn, and the final result adds both.
Methodology
Audited simulation engine: margin accounting faithful to the app, conservation checks on 100% of wallets (0 failures), and bit-identical A/B execution between the engine and the code that runs in the app. Costs calibrated on Hyperliquid's real order book (taker 0.045%, slippage 0.01%).
Raw numbers, per-wallet JSON and the full report are archived; this page publishes the same content unedited.
Honest limitations
The price data comes from Binance (correlated crypto assets). Real Hyperliquid also trades uncorrelated assets — tokenized stocks, indices and commodities — that the simulation does not contain. 2024 and 2025 were bear markets: wallets born in those years had negative medians.
A backtest is evidence of robustness under real costs, not a forecast. Past results do not guarantee future results; most recent years had a negative median. Only trade capital you can lose entirely.
Summary chart: the 10,000 wallets
A) Result distribution per wallet
Reading the last group in the chart: the rightmost bar groups ALL wallets above 500% — 308 wallets (3.1% of the 10,000), including the best one (+5,577%) — and that is why it is taller than its neighbours. Their median initial capital is US$ 302.60 (whole set: US$ 509.90): they are 2.4% of the simulated capital and 14.0% of the final result, with a median peak of 10.2x and a median withdrawal of 772% of the initial capital.
B) Wallets sorted from worst to best
C) Capital invested and final result
All panels are in % of the capital invested, except panel C, in US dollars.
Second backtest 2020–2026: 1,000 users with 10 wallets each
Past results do not guarantee future results. This second backtest assumes that each user of the app will create 10 wallets and deposit the same amount in each one — randomly between US$ 15 and US$ 100 — so that each user places many buy orders and, potentially, many sell orders every day, providing liquidity to the market like a market maker, with the intention of profiting over the long run. What follows is the final result after the app has been running for 1 year.
1,000 simulated users, each with 10 wallets: 10,000 wallets, each with 365 days of life. The capital per wallet was drawn between US$ 15 and US$ 100 and repeated across the 10 wallets of the same user, so each user invested between US$ 150 and US$ 1,000 in total. Each user's start date was spread uniformly from January 2020 to October 2026.
The price base is the same as in the first backtest: 1-minute Binance candles from 2020-01-01 to 2026-10-09. Costs are real as well: 2% activation on each deposit and, per position, 0.045% taker plus 0.01% slippage per side — about 0.11% per round trip. The automatic withdrawal is the app's official rule: 50% of realized profit per event from US$ 22.
The picture is better than the single-wallet one because ten wallets dilute one wallet's bad luck: 68.7% of users ended in profit, the median was +30.7% and the total invested turned into US$ 819,341.30 on US$ 563,495.02 invested (+45.4% weighted). Even so, 31.3% ended at a loss and the worst user lost 78.4%: both sides are here.
| +44.3% | Average return per user — arithmetic mean of the 1,000 users over the capital invested (their 10 wallets combined). It sits above the median because a few users had extreme results. |
| +30.7% | Median per user — half of the users ended below this value, half above. It is the typical case, already with the effect of 10 wallets per user. |
| 68.7% | Users in profit — 68.7% of the 1,000 ended above what they invested; 31.3% ended at a loss. |
| −45.4% | P05 — 1 in every 20 users lost 45.4% or more of the capital. Even with 10 wallets, the left tail is real. |
| +188.7% | P95 — 1 in every 20 users ended above +188.7% (P99: +300.2%). The right tail mirrors the left one. |
| 40.0% | Reached 2x at the peak — 40.0% of users saw the combined equity of their 10 wallets reach twice the amount invested at some point; 18.7% finished with 2x or more. |
| 1.77x | Median peak — the median of the highest combined value of the 10 wallets was 1.77x the amount invested (maximum: 6.03x). |
| 48% | Withdrawn automatically — on average, 48% of the capital invested left the exchange for the user's wallet during the year, outside the following risk. Total: US$ 315,652.04. |
The whole distribution, band by band
| Return band | Final exchange balance | Total withdrawn (% of initial capital) | Final result (balance + withdrawn) | Equity peak (band average) |
|---|---|---|---|---|
| P01–05 | −59.8% | +3.6% | −56.2% | 1.11x |
| P06–10 | −51.1% | +11.8% | −39.3% | 1.22x |
| P11–15 | −45.4% | +15.2% | −30.2% | 1.35x |
| P16–20 | −39.4% | +17.8% | −21.6% | 1.39x |
| P21–25 | −37.1% | +23.9% | −13.2% | 1.43x |
| P26–30 | −34.0% | +29.0% | −5.0% | 1.43x |
| P31–35 | −32.9% | +34.5% | +1.6% | 1.55x |
| P36–40 | −16.2% | +26.3% | +10.1% | 1.61x |
| P41–45 | −23.0% | +40.6% | +17.6% | 1.64x |
| P46–50 | −21.0% | +46.6% | +25.6% | 1.77x |
| P51–55 | −14.0% | +48.1% | +34.1% | 1.85x |
| P56–60 | −17.5% | +59.8% | +42.3% | 1.91x |
| P61–65 | −4.3% | +55.7% | +51.4% | 1.99x |
| P66–70 | +0.2% | +62.4% | +62.6% | 2.15x |
| P71–75 | +7.1% | +68.5% | +75.6% | 2.24x |
| P76–80 | +5.1% | +83.4% | +88.5% | 2.44x |
| P81–85 | +11.8% | +92.5% | +104.3% | 2.60x |
| P86–90 | +20.2% | +106.3% | +126.5% | 2.81x |
| P91–95 | +45.9% | +118.6% | +164.5% | 3.15x |
| P96–100 | +89.3% | +159.6% | +248.9% | 4.15x |
Each row groups 5% of the 1,000 users (50 users, with their 10 wallets combined), ordered by final result. Balance in the exchange, total withdrawn and final result are weighted by the capital invested in the band — the first two columns add up exactly to the third. The peak is the band average of the highest combined value of the 10 wallets, in multiples of the amount invested.
Important reading: the 10 wallets of a user traded for the whole year, each until the end of its 365-day life, and nothing withdrawn went back into risk. That is why the balance in the exchange is low in almost every band: the accumulated wealth sits in the total withdrawn, and the final result adds both.
Methodology
Same audited engine as the first backtest, with two additions that do not change its behaviour: end of life per wallet (365 days) and recording of the combined value over time, to measure the peak of the 10 wallets together. It passed 31 checks: engine bit-identity on 10/10 seeds, accounting conservation in 10,000/10,000 wallets, final result equal to balance plus withdrawn in 10,000/10,000, peak confirmed by a second independent implementation and a price base without gaps (2,473 days).
The JSON with the 1,000 users and 10,000 wallets, the full report and the simulation code are archived; this page publishes the same numbers, without edits.
Honest limitations
The price base is Binance and the crypto assets are correlated with each other; the real Hyperliquid also trades uncorrelated assets that the simulation does not contain. One year of life per wallet is shorter than the 6.8 years of the first backtest, and users who started in 2024 and 2025 hit bear markets — the median in those periods is lower.
Ten wallets reduce the bad luck of a single wallet, but they do not create an edge: the return per USDC invested is the same, with less variance. A backtest is evidence of robustness under real costs, not a forecast. Past results do not guarantee future results — only trade capital you can lose entirely.
Summary chart: the 1,000 users
A) Result distribution per user
B) Users sorted from worst to best
C) Capital invested and final result
All panels are in % of the capital invested, except panel C, in US dollars.
Version requires Android 7.0 or later