10 เดือนดูดเงิน 10 ล้านดอลลาร์ เซียนอาร์บิทราจเปิดเผยกลยุทธ์ใหม่

Original author: CBB (@Cbb0fe)

Compiled by Odaily Planet Daily (@OdailyChina); Translator: Azuma (@azuma_eth)

Background

• Odaily note: For the full story behind this section, please refer to "HyperEVM's Top Arbitrage Team Reveals Its Strategy: How They Moved $5 Million in Six Months".

Let's rewind to October 2025.

For the eight months prior, we had been running the top arbitrage bot on HyperEVM.

But that grind was coming to an end. Over the past few months, we had been competing with Wintermute, and now, new players were entering the fray, significantly compressing our profits.

That's fine. My buddy and I were used to this by now. We never tried to compete head-to-head long-term with institutions and their legions of employees. We couldn't; there are only two of us.

Our edge has always been — deploying a strategy as fast as possible and extracting as much profit from it as we can before the big players arrive.

They can't deploy a strategy within 48 hours. They have regulatory constraints, internal processes, approval procedures, and so on. We have none of that. We just need to be as fast as possible.

So, it was time to find the next grind.

We started thinking, where is the next opportunity?

That event had just happened on October 10th. The crypto market looked completely hopeless, everyone had been beaten down hard, and there were no exciting new opportunities left to uncover.

So, we began scouting for a new battlefield. On October 13th, HIP-3 went live on Hyperliquid. Three days later, TradeXYZ officially launched its first equity perpetuals market — XYZ100.

Considering that Hyperliquid still had over 40% of its token supply undistributed to the community, we figured generating some volume on HIP-3 might be a good idea.

In fact, this was exactly the same thought process that led us to discover the HyperEVM arbitrage opportunity eight months prior: back then, we were just trying to farm some spot volume on Hyperliquid.

We had no idea if this would work, but we decided to try — to develop and operate an equity perpetuals bot that arbitrages between HIP-3 and traditional financial markets.

Step One: Entering the World of Traditional Finance

One thing I need to clarify upfront: before this, we had never traded stocks in our entire lives.

We didn't even really know what futures were. Basically, we knew nothing about traditional finance.

The only thing we did know was that IBKR was a highly competitive platform for what we wanted to do, so we decided to research it first.

For the first few days, I was mostly just trying to figure out how to use the IBKR platform at all. I basically screenshotted everything and sent it to Claude, asking: "What is this?" "What does this mean?" "How do I do this here?" "How are we supposed to hedge against XYZ100?"

Essentially, that's how we started learning TradFi. Meanwhile, my buddy began researching IBKR's API, trying to figure out what was possible and what wasn't.

Coming from the crypto world, he was used to plugging into an exchange API and getting a program up and running very quickly.

But IBKR was a whole different world. Market data subscriptions, contract specifications, order types, permissions, API limits, TWS, IB Gateway... We had a lot to figure out, and we weren't even sure at first if this could actually be done.

However, after a week of wrestling with IBKR, we started to find our footing.

Building the Arbitrage Bot

The strategy was actually quite simple.

We treated IBKR's price data as the "true price" and continuously checked HIP-3 for arbitrage opportunities.

If a market on HIP-3 was trading at a discount relative to IBKR, we would go long on HIP-3 — and only after our order on Hyperliquid was filled would we open a corresponding short position on IBKR.

If a market on HIP-3 was trading at a premium relative to IBKR, we would do the opposite — short on HIP-3, and once filled, go long on IBKR.

In theory, it's very simple, but in practice, we needed to configure a large number of parameters for every single HIP-3 market.

Let's take the NVDA arbitrage strategy on the IBKR side as an example:

["NVDA", 55, 400, { maxDelta: 800, slippage: 0.1 }]

• "55 shares" is our minimum hedge size. Since IBKR's commission has a minimum of $1, we don't want to execute a ton of micro trades. So, we let the delta exposure accumulate until it reaches 55 shares of NVDA, then hedge on IBKR.
• "400 shares" is the maximum hedge size for a single IBKR order, to avoid incurring excessive slippage.
• "maxDelta: 800" is our safety valve. If, for some reason, our trades on IBKR keep failing, causing the position difference between the two markets to reach 800 shares of NVDA, the bot will halt trading on that market entirely.
• "slippage: 0.1" is the maximum slippage we allow when hedging on IBKR.

Now for the HIP-3 side:

NVDA: pair("NVDA", "xyz:NVDA", {makerSize: 400, makerOffsetBuy: 0.12, makerOffsetSell: 0.12, cancelDelta: 0.02, takerRatioBuy: 0.05, takerRatioSell: 0.1, takerMin: 1, takerMax: 2000, limit: 110000, makerEnabled: true, preMarketOffset: 0.04 })

Looks complex, but the logic is actually quite simple.

• "makerSize" defines the size of our resting orders, while "makerOffsetBuy / makerOffset Sell" define the spread we want relative to the fair price.
• "cancelDelta" tells the bot how much the price needs to move before we cancel and re-place our resting orders.
• For taker trades, "takerRatioBuy / takerRatioSell" define how large the spread needs to be for us to aggressively take; "takerMin / takerMax" control the size of the trades we're willing to execute.
• "limit" is the maximum total position size we allow ourselves to hold on that market, and "makerEnabled" simply lets us toggle maker orders on or off.
• Finally, "preMarketOffset" adds extra spread during the pre-market session, because liquidity on the TradFi side is much thinner during those hours.

The First Trades

At the end of October, we were finally ready to start trying.

The first few days were chaotic. We kept fighting with the IBKR API, sometimes losing connections, and my buddy had to figure out how to keep everything connected and running smoothly.

But we quickly realized that opportunities were plentiful. Honestly, it felt like picking up money.

Throughout November, we executed about $850 million in volume on HIP-3, making over $500,000 in profit. Not bad.

December was a bit quieter. We did about $550 million in volume, and profits were still decent, but we actually started thinking about whether we should shift our focus to other things.

It was profitable, sure, but it didn't seem like a goldmine either.

Still, we decided to keep going. As always, if there's profit to be mined, we usually find it hard to stop.

The Precious Metals Frenzy

The real explosion came this January.

Gold and silver started surging wildly, and the demand on Hyperliquid was absolutely insane. Making money started to become almost too easy, and the preparation we had done over the past two months was exactly what was needed for a market like this.

One issue we faced was liquidity. Basically, everyone wanted to go long commodities on Hyperliquid, which meant we constantly needed to pour more capital into the IBKR side to hedge.

We kept adding funds to IBKR, but moving such large sums also brought banking headaches. EtherFi was an absolute MVP in this regard; they allowed us to execute large withdrawals very quickly.

Throughout January, we did $1.7 billion in volume on Hyperliquid, earning over $600,000 from funding fees alone.

But as I said before, there are only two of us. We have no internal processes; we move very fast. Essentially, everything is tested directly in production, and sometimes, that comes at a cost.

On January 27th, I had just landed in Dubai and was about to grab a coffee with my buddy to chat about the bot. Suddenly, I received a margin call warning from IBKR on my phone.

I had absolutely no idea what was happening. It was still early, and the market wasn't experiencing any major volatility.

I logged into IBKR. It turned out we were net short $120 million worth of gold futures. And gold was in the middle of a violent rally.

We immediately shut down the bot. At that moment, I was shaking all over. I was genuinely terrified of getting liquidated because virtually all trades on IBKR were executed by the bot, and I wasn't that familiar with IBKR myself.

Over the next 15 to 30 minutes, I manually closed out $120 million worth of short gold positions.

Later in the afternoon, when the market opened [in the US], we were finally able to calculate the damage — we had lost $1.1 million.

That was painful. But we didn't have time to cry about it. We had to figure out what went wrong and fix it as quickly as possible.

Eventually, we found that the cause was ridiculously stupid: a data refresh issue with the IBKR API.

The bot thought there was a Delta difference between our positions on Hyperliquid and IBKR, so it kept shorting gold on IBKR, trying to correct a Delta exposure that didn't actually exist.

Time and time again; time and time again; time and time again...

Until it had accumulated a $120 million short position in gold, and we started receiving margin call warnings.

Clearly, we needed more safeguards.

We needed to ensure the data we were getting from IBKR was actually up-to-date. We also needed to add extra checks before allowing the bot to expand its positions further. More importantly, the bot was originally never designed for such high volumes and such dense opportunity flows.

We spent the whole day fixing everything. The new version of the bot was back online the next day, but our confidence was shaken. Maybe we didn't really know what we were doing. Maybe the risk-reward ratio just wasn't worth it.

We had just lost $1.1 million due to a ridiculously stupid problem, and now we started feeling like the bot could malfunction again at any moment. This was the first time since the bot went live that we seriously considered — should we just stop here?

But by now, you probably know us well enough... We're just two greedy bastards. After losing seven figures, we don't quit. We only get more determined.

And I think this might actually be something we're good at. We've built many bots together in the past, and almost every time, we've gone through phases of crazy losses, but somehow, we always manage to claw our way back.

We don't sit around crying for hours. We try to figure out what went wrong, fix it, and move on.

Until the loss is earned back, it basically becomes a taboo subject between us brothers.

The next day, silver experienced a wild pullback after hitting an all-time high, and at one point, there was nearly a 3% price difference between Hyperliquid and IBKR.

We made about $600,000 in profit from that.

We were f*cking back!

Liquidity Management and Speed Upgrades

By this stage, we had already made a significant amount of money. And we knew how this game worked.

If there's this much money to be made here, then more people, and their legions of employees, are bound to come and fight for a piece of the pie. So, we had to get stronger, and fast.

The first issue was capital. This was completely different from pure crypto arbitrage — in crypto, rebalancing funds between venues can take less than five minutes; but here, we had to move funds in and out of IBKR through banks.

So, we designed a dynamic system that adjusts the strategy based on the liquidity available on the IBKR side. When liquidity on IBKR is low, we're willing to pay a certain cost to close out existing trades, thereby freeing up capital.

At the same time, we would demand a larger spread before opening new positions. When liquidity on IBKR is high, we do the opposite — we accept smaller spreads to open new positions and deploy capital more aggressively.

The second area for improvement was speed. Before this, we had been using IBKR's price data as our source of truth. It worked, of course, but this data feed was relatively slow.

As more participants entered the game, we knew it would eventually become a speed race. Relying on IBKR's data was no longer enough. So, we started looking for alternatives and found Databento.

With Databento and a Nasdaq data license, we could get a much faster, direct market data feed.

We applied for access in January and were finally approved at the end of the month.

From Precious Metals to Oil

By February, the metals market was still hot, and we did around $1.5 billion in volume. And as if that wasn't enough, at the end of February, Trump decided to bomb Iran, causing wild market volatility and pushing oil prices above $100.

At this point, we were earning roughly $60,000 to $120,000 per day from arbitrage spreads and funding fees. Except for Saturdays and Sundays, of course, when traditional financial markets are closed and we were bored out of our minds.

If we had "only" made $40,000 in the past 24 hours, we would start thinking something must be wrong.

So, we'd go check the bot, tweak parameters, and try to figure out what was happening and how we could optimize further.

Claude was a huge help in this process. We could feed it all our Hyperliquid and IBKR trade data and ask it to analyze: Where are we losing the most money? What exactly is going wrong? What can be improved?

This was actually our first time using AI to analyze our trading, and it made a significant difference. Even when things were going well, we maintained this almost obsessive state.

Basically, it's the only way we know how to stay ahead.

My buddy and I talk about the bot all day long. He pushes code updates almost daily, while I'm constantly tweaking parameters based on what's happening in the market.

The Semiconductor Frenzy

At the end of April, as the Iran conflict started to cool down, we thought this crazy profitability was finally going to end.

Over the past few months, we were making around $500,000 a week, and we honestly couldn't think of anything that could continue to create arbitrage opportunities on this scale. Then, semiconductors and all the "bottleneck trades" suddenly started exploding across the board.

Stocks like SNDK and MU started trading like pure meme coins. It was absolutely insane.

We started building this bot in October when basically nothing was happening in the market. Since then, we've seen the precious metals rally, the oil rally, and now semiconductors are being hyped like shitcoins on BSC.

What is happening to this world?

Clearly, there was a lot of luck involved. We happened to be in the right place at the right time, and we happened to have a product that was already running and specifically suited for this kind of market environment.

But I do think there was some good judgment in believing in HIP-3,