How Manifold got started
Manifold founder Jae Chung said in a republished essay that when he decided to start a quantitative trading firm in 2021, the idea was simple: build something that could make money while he slept.
He wrote that he was 21 at the time, had almost no direct experience in quant trading, and did not even know what alpha meant. Advisors told him the plan was foolish. He did not understand who he was up against, and the field was already dominated by firms such as Jump, Tower and Jane Street. He said those warnings were correct.
Before launching Manifold, Chung said he entered crypto in 2016 and 2017 as a white-hat hacker, ran validator nodes for several years, and made some fortunate trades and investments during DeFi Summer and the NFT boom. None of that, he said, had much to do with quant trading itself.
Early iterations failed
Chung said many people think quant trading works like this: find a pattern, then make money. He used to think the same way. In practice, it was much harder.
Manifold tried almost everything in its early phase, including arbitrage, spread capture, basis trades, statistical arbitrage, yield farming and even deep learning. At first, none of it worked. Chung described trading as a zero-sum game. If enough players at the same table are better than you, there is little left to capture.
That forced the team to ask a different question. The issue was not only how to find alpha, but where they could build an edge at all.
Why the firm moved toward DeFi and DEXs
Chung said top trading firms usually dominate in one or more of three areas: speed and latency through trading infrastructure, proprietary data or order flow, and research or alpha generation.
On centralized exchanges, he said, firms such as Jump and Tower had already built advanced low-latency infrastructure across major venues including Binance, OKX, Bybit and Coinbase. They also had fee tiers Manifold could not match at the time because it did not have comparable volume, and they had access to richer order flow and data. Manifold's own trading and research stack was, in his words, entry level, and it did not have a large army of engineers and researchers.
Instead of fighting on that ground, the firm looked for a different game. Chung said the answer was DeFi, especially decentralized exchanges.
At the time, many blockchain foundations were spending significant capital to incentivize passive liquidity on DEXs. Most DEXs did not use order books and instead relied on automated market maker models, where trades execute along a pricing curve. That meant a DEX trading system had to handle both market data, such as price, liquidity ranges and slippage, and on-chain order submission, execution and confirmation.
He said leading firms were not especially aggressive there, possibly because DEX trading sat outside their established infrastructure, the market was smaller than classic CEX trading, regulation around DEXs was uncertain, and those firms were already making substantial money elsewhere.
Chung also argued that Manifold's talent mix fit the opportunity. The team combined crypto-native engineers familiar with smart contracts and on-chain execution with quant researchers from more traditional backgrounds, including Citadel and Tower.
He gave one example from hiring. One of the firm's later top employees had been a data engineer at an insurance company while privately operating as an individual MEV searcher on Polygon, running a small atomic arbitrage strategy. Before the person joined, Chung communicated through the anonymous email address 0xaddress@protonmail and reviewed the bot's address history on a block explorer, which convinced him the candidate had potential.
Testing the CEX-DEX arbitrage thesis
Before shifting the whole team toward decentralized exchanges, Chung said he tested the idea himself. He built a simple on-chain trading system entirely in TypeScript and used it to run arbitrage between centralized exchanges and decentralized exchanges, including Binance and FTX against top-liquidity blockchains. The team called it CEX-DEX arbitrage.
He said the opportunity existed because blockchain transactions are delayed by block times. On some chains that delay was around 200 milliseconds, on others 1 to 2 seconds, and on some even longer. That is structurally slower than quote updates on centralized exchanges, so a lead-lag relationship naturally appears.
At the same time, much of the liquidity on DEXs was passive or stale and was not actively managed by bots, while price discovery mostly happened off-chain. That created price gaps across venues.
Even though the prototype was rough and not sensitive to latency, Chung said it still captured real arbitrage opportunities and remained profitable after fees. After months of research that had gone nowhere, this was the first sign that the team had found something real.
Once the edge appeared, execution became the focus
Chung said it is unusual for an arbitrage strategy to work at the prototype stage. By definition, arbitrage is risk-free profit lying on the ground, and someone faster is usually ready to pick it up. If a prototype works anyway, there is usually room to improve it.
He listed several paths the team identified early on:
- rewrite the TypeScript system in a faster language;
- add more DEXs per chain and expand to more chains;
- improve math around sizing and slippage;
- optimize fees to raise profit per trade;
- lower CEX fees as volume increased;
- reduce on-chain gas costs through technical changes;
- improve on-chain execution to raise fill rates;
- optimize inventory to improve returns, capital efficiency and uptime.
From that point, he said, the problem was execution. One advantage in high-frequency trading is that the market gives immediate feedback. Build something, run the strategy, and the result is quickly visible: either profitability improves or it does not. That created a short loop between engineering work and dollars earned.
Chung also recalled a moment from Dec. 24, 2021. He and co-founder Sid were coding through the night in their apartment office, with no weekends and no holidays. Around 1 a.m., Chung noticed the date and said, 「Sid, today is Christmas.」 About 30 seconds later, Sid replied, 「Oh. Merry Christmas.」 Then they kept coding. Chung said that, looking back, it was probably his favorite office because everything started there.
The strategy began making more than $10,000 a day
After a series of broad improvements, Chung said CEX-DEX arbitrage started generating more than $10,000 a day while using only a few million dollars of deployed capital.
The team's early priority was to make a new system, written in Go and Solidity, good enough to scale across multiple chains and 10 to 20 DEXs on each chain. It also began integrating newer decentralized exchanges that used the UniV3 tick pricing model. Those venues were more efficient because of concentrated liquidity and had lower slippage.
As the strategy scaled, Manifold's trading volume on centralized exchanges increased, which lowered its fee tier. As capital grew, the team also added more trading pairs to the pool.
Competition changed the game
As the easy opportunities were harvested, Chung said competition rose quickly. On some chains, Manifold could no longer wait for a 10 basis point spread to trigger an arbitrage because other bots were willing to trade at lower spreads.
He gave a simple example. If ETH was trading at $2,000 on Binance and $2,001 on Quickswap on Polygon, the spread was 0.05%, or 5 basis points. A bot configured to trigger at 5 basis points, after fees and slippage, would sell ETH on Polygon and buy ETH on Binance, pushing prices back toward parity.
That meant Manifold's bot would never see the trade if it was waiting for a 10 basis point spread, because another bot would close the gap before ETH on Polygon reached $2,002.
Chung said execution costs differ across firms. The main variables include fee levels on centralized exchanges, whether the CEX leg is taker or maker, and gas costs on-chain.
He said the team eventually ran into firms including Wintermute and had to participate in priority gas auctions, or PGAs, against sets of bot addresses. The technique comes from MEV. Traders repeatedly resubmit the same transaction with higher priority fees to improve its position within a given block.
To compete in those auctions, Chung said a firm needs to know almost exactly how much a trade will make, have infrastructure reliable enough to land the transaction within one block after spotting the opportunity, and know which addresses it is bidding against while quickly determining whether it won or lost.
He added that different blockchains have related but distinct transaction ordering and submission systems, and those rules keep changing as chains try to direct more value to users, protocols or the chain itself. Arbitrageurs have to keep adapting.
Gas optimization widened the set of profitable trades
Chung wrote that total gas cost for an on-chain transaction is roughly a function of gas used multiplied by gas price, with gas price depending on network congestion.
In PGA-style auctions, gas price and priority fee matter, but once a trader can reliably land a transaction within one block, there is not much more to optimize there. Reducing gas usage itself, however, can materially lower costs.
One technique the team used was efficient bit packing. The Ethereum Virtual Machine stores data in 32-byte, or 256-bit, slots. By placing smaller data types next to each other, Solidity can pack them into a single slot, which cuts the cost of read and write operations such as SSTORE and SLOAD.
Chung said that reduced zero padding in transactions and lowered total transaction size, which in turn reduced total fees.
The practical effect was important. Manifold could take trades that were invisible to competitors with higher gas costs. A trade that netted only $1 after fees could still be worth doing for an optimized system, while producing $0 for a less efficient one. That let the firm close spreads before other bots even saw a profitable opportunity.
Gas optimization also allowed the team to bid more aggressively to win arbitrage while staying profitable. Over time, winning more trades meant more profit and more volume, which then fed back into better CEX fee tiers and larger arbitrage capture per trade.
Capital efficiency became the next bottleneck
As more optimizations were added, Manifold expanded the strategy across more chains, DEXs and trading pairs. Chung said the broader the trading universe, the greater the capacity, so it made sense to allocate most of the firm's capital to CEX-DEX arbitrage.
He said the strategy was generating an annualized return above 100% and had no down days. Even so, the team believed it still had more capacity, which raised a new question: how to earn more with the capital already in place.
The problem was that CEX-DEX arbitrage requires capital to be spread across venues and inventory to be pre-positioned across several trading pairs. Chung used ETH/USDT between Binance and Arbitrum as an example. If the firm started with $500,000 of ETH and USDT on each venue and Binance traded at a premium, it would sell ETH on Binance and buy ETH on Arbitrum. If the premium persisted, Binance could quickly run out of ETH to sell, while Arbitrum could run out of USDT as it kept buying ETH.
Centralized exchanges offer margin services, which let traders use some leverage and keep trading through inventory imbalances. On-chain, once inventory runs out, the bot stops. To keep it alive, the firm had to rebalance by moving excess USDT from Binance to Arbitrum and excess ETH from Arbitrum back to Binance.
Even with round-the-clock rebalancing, Chung said the team still missed a lot of profit. Arbitrage PnL is concentrated during volatile periods because spreads widen then, and a $500,000 inventory can be exhausted in seconds. During those same periods, exchange withdrawal times become slower than the already slow baseline of about five minutes. That meant the bot could miss the most profitable windows.
Hydra and Hydra v2
To solve the inventory problem, the team first built an automated system called Hydra. The idea was to move funds continuously using bridges between blockchains and CEX withdrawal APIs, identifying excess assets in one place and sending them where inventory was short.
Chung said it did not work as planned. Bridges were unreliable. Assets got lost or disappeared, and the team had to manually track which bridge failed and when. Even when bridging worked, it often took too long, so the system still missed volatile markets.
The next version, Hydra v2, dropped the dependence on bridging. Instead, the team integrated Aave, or similar protocols where Aave was unavailable, to create its own on-chain spot margin system. Chung described Aave as a lending protocol natively deployed on most major chains.
The logic was straightforward. In the earlier example, if the firm kept buying ETH on Arbitrum, it would end up with too much ETH and too little USDT. The system would identify the excess ETH, lend part of it on Aave, and borrow USDT against it as collateral.
Chung said the implementation was simple, but Hydra v2 ended up being one of the biggest contributors to the firm's PnL at the time.
What he says the team learned
Looking back, Chung said none of the individual improvements were especially exotic, but the team could not have invented all of them at once. If the first prototype had failed badly, it might not have kept pushing for long. Because the initial live strategy worked, the firm kept iterating, monitoring fill rates and profitability, and using fast market feedback to identify what could be improved next.
His conclusion was that when a market is full of unknown obstacles, it is hard to crack the game. Once a firm reaches the frontier and understands the basics, it becomes easier to stay ahead. In trading, he said, table selection matters as much as execution.
The lifecycle of alpha
Chung said CEX-DEX arbitrage remained Manifold's main profit engine for a long time. In its strongest period, during high volatility, the strategy could generate more than six figures in a single day.
He also said the firm's infrastructure became generic enough to integrate a new chain within one hour, which often made Manifold one of the earliest arbitrageurs on new chains, capturing large spreads when those markets were least efficient.
Still, he said, like almost every alpha, the edge decayed over time. Arbitrage in particular attracts competition that compresses spreads until profit margins shrink sharply. He gave one example: if two firms in a PGA setting have identical gas optimization and identical CEX fee tiers, their per-trade costs are the same, so bidding will eventually rise to each firm's minimum profitable level.
By the end of 2025, Chung said, CEX-DEX arbitrage was no longer the extremely high-return strategy it once was.
He added that Manifold had other stable, high-return strategies running in the background, which allowed the firm over time to reallocate both time and capital toward more profitable approaches. Even so, he said CEX-DEX arbitrage still holds a special place for him because it gave the company a legitimate starting point and an anchor from which to build.
