In this episode of Professional Punters, we sit down with Michael Feng, co-founder of Hummingbot. His first finance job put him inside the machinery of subprime CDOs during 2008; today he builds standardized open-source software that lets traders connect to exchanges, which has facilitated more than $50 Billion of trades.
Michael joined Citigroup in 2001 and built models used to rate early subprime-mortgage CDOs, then moved to Hong Kong to sell the structures across Asia. He went to Stanford for engineering, started a PDF-data company that Nitro acquired, and eventually found his way back to markets through crypto.
CoinAlpha, his first crypto startup, tried to run a quantitative fund through smart contracts. The fund was unwound after SEC contact and the matter later settled, but its code did not disappear. Two exchange integrations, each painful to build, became the seed of Hummingbot: an open-source framework for market making and algorithmic trading across centralized and decentralized venues.
CDOs and Correlation
A Collateralized Debt Obligation (CDO) pools debt, slices the cash flows into tranches, and sells each layer to an investor with a different appetite for risk. One buyer takes the first losses for a higher return; another accepts a lower yield for a senior claim meant to survive all but an extreme default cycle.
Michael joined the industry as a junior analyst, building models that helped obtain ratings for early subprime-mortgage CDOs. He later sold the structures to life insurers and banks across Asia. By 2007, he and colleagues could see a bubble in their corner of the market; though they did not understand how far the failure would actually travel when everything eventually did blow up.
The structures had become recursive. Banks were taking senior tranches from earlier CDOs and packaging them again into CDO-squared vehicles. Some paired long-dated mortgage exposure with short-term funding or guarantees. Synthetic credit derivatives created still more exposure without requiring mortgage to even exist.
Rating models had data on the chance that two corporate borrowers such as Ford and GM would default together. They had far less evidence for how two diversified pools, each holding claims on thousands of mortgages exposed to the same housing cycle, would behave together. Diversification reduced the noise inside each pool; it did not remove housing as the common factor. When house prices fell nationally, the supposedly independent pools moved together.
Michael remembers telling his VP, “I think we might be in a bubble.” The answer was, “Yeah, we probably are.” Their mistake was thinking the damage would stay in their corner of the floor, among the desks trading CDOs, mortgage-backed securities, and credit-default swaps. It did not.
The Problem With PDFs
After the Financial Crisis, Michael left finance for a master’s program in engineering at Stanford. His first startup used computer vision to pull structured information out of PDFs. PDF looks like a document to a person but may look like drawing instructions to a machine: put this glyph at these coordinates, draw that line over there, etc... Recovering a table from an old filing meant using OCR, bounding boxes, and heuristics to infer words, rows, and columns that were never labeled.
The difficulty depends on the file’s history. A modern PDF exported from a word processor may retain useful internal structure. A decades-old filing may be little more than a scanned image, forcing the parser to reconstruct the document from visual clues. Michael thinks probabilistic AI is unusually well suited to that work: it can infer the likely relationship among glyphs, boxes, and tables where a deterministic parser needs another hand-built rule.
Michael raised roughly $240,000, much of it from friends and family, and persuaded former colleagues to become customers. Nitro eventually acquired doxIQ, returning the investors’ original capital. He and his co-founder then joined Nitro, a roughly 200-person company. As head of product, Michael learned to manage a team, write product-requirement documents, and operate inside a larger organization.
Crypto Hedge Fund
Michael returned to finance during the 2017 crypto boom. The team created a quantitative digital-asset fund and used a smart contract to manage investor flows, calculate each holder’s share, and process minting and redemption through Ethereum.
Because Chainlink was not yet live on mainnet, the team used an early oracle service to bring the value of a portfolio held at Coinbase onchain. The experiment worked technically, but the regulatory structure did not work. The SEC found that CoinAlpha publicly discussed the offering without an available registration exemption and had not taken the required steps to verify accredited status for every investor.
CoinAlpha halted the offering, reimbursed collected fees, surrendered future fees, unwound the fund, and made payments so that no investor suffered a loss. It settled without admitting or denying the findings and paid a $50,000 civil penalty.
Michael says CoinAlpha spent roughly $300,000 on lawyers before agreeing to the $50,000 penalty. CoinAlpha’s investors continued to back the team, according to Michael. The fund was gone, but its exchange integrations still worked. That code became the pivot.
One API, Many Exchanges
Building the first Coinbase integration took the team about a month. Gemini consumed another month. Across hundreds of centralized and decentralized exchanges, teams trading on multiple venues were repeating versions of the same tedious job. Hummingbot turned the integrations into a common open-source interface. A strategy could check balances, read an order book, and place an order through a standardized connector rather than learning every venue from scratch.
A user can do this in crypto because many venues expose direct APIs. In U.S. equities, a broker and its execution layer generally sit between the trader and the exchange. Hummingbot connects through the user’s own credentials and puts the venue-specific work into open connectors that everyone can reuse.
Michael calls his core user the “indie algo trader”: someone who did not go to Stanford or an Ivy League school, perhaps living in Azerbaijan, Turkey, or China and trading with $1000. The templates do not hand that person a finished strategy; they provide a place to start and a way to build a track record.
The same mission extends to education. Botcamp teaches market making and algorithmic trading through the Hummingbot framework. He also sees AI as a way to help users reach more strategies and venues through that framework. The important distinction is that easier access is not a finished edge: the user still has to choose, test, and supervise the strategy.
API-Broker Business Model
Giving the code away did not eliminate the business model. Michael says large exchanges operate API-broker or referral programs: when software routes a user’s trade, the venue can return part of the fee to the software provider. That fee share gave Hummingbot a way to keep connectors working as APIs changed.
Michael says the team raised a Series A in 2021 around a marketplace where token issuers would deposit rewards and independent makers would earn them for posting useful liquidity. The alternative was the traditional arrangement in which an issuer hires one specialist firm and supplies it with capital. Hummingbot’s wager was that a network of independent makers could compete for the same job.
The liquidity-mining model resembled the now-common practice of venues subsidizing market makers. “It didn’t really work,” Michael said of the standalone marketplace. Exchanges absorbed the feature and ran their own reward programs. A market can validate the behavior you predicted without validating the company you built around it.
TradFi, DeFi, & Systemic Risk
Asked whether the system got better at measuring systemic risk after 2008, Michael’s short answer was no. Dodd-Frank and higher capital requirements constrained what banks could warehouse, he said, but much of the structuring migrated from dealer balance sheets to the buy side.
Private credit is his clearest example. Michael describes it as CDO-like structuring moved to the buy side: the loans and structuring sit in funds rather than on bank balance sheets. The structures are less baroque than a synthetic CDO-squared, but he is not convinced the system carries less risk. “If Citadel goes bust, is that systemic?” he asked. “I don’t know.”
In crypto, Michael sees the old layering instinct in restaking, leverage loops, and emerging tranched-credit products. He was not calling every protocol another 2008; he was pointing to the rhyme. Michael has deliberately stayed out of that business: “I felt like I’ve already done it before.”
Perps, AMMs, & Prediction Markets
Michael’s case for crypto goes beyond direct trading access: it also lowers the floor for launching a financial product. At JPMorgan, he recalls, his desk could not take on an issuer selling less than $500 million of securities because compliance and distribution costs would overwhelm the deal. Onchain markets can launch products without recreating that entire stack. He points to three designs that benefited: perpetual futures, automated market makers, and prediction markets.
Michael remembers mornings spent quoting a family of Ford bonds: one maturity, another coupon, a convertible here or there, each requiring its own two-way market. Derivatives fragment for a different reason, with expiries dividing similar exposure across a calendar. A perpetual future removes the expiry and uses funding payments to pull one continuing long-or-short contract toward spot. Michael’s bet is that this simplicity will let perps displace some futures and options over time.
An automated market maker replaces the live order book with a pool and a pricing rule. Liquidity providers deposit assets, and takers trade against a curve rather than waiting for a specific counterparty’s order. Michael says newer Prop AMMs aim for a hybrid: customizable bonding curves with resting liquidity on the book.
Prediction markets turn a future event into a contract with defined resolution criteria and let both sides make or take prices. Versions of all three ideas existed before crypto. What changed was the cost of trying them: an open venue could launch, fail, and change the design without waiting for permission.
Trading vs Gambling
Trading and finance have always been, in Michael’s words, “kind of degen.” He points to Ed Thorp and to the long crossover between trading and gambling.
Cash flows may be absent from a meme coin, but data is not. A trader can watch which narratives spread, who controls the supply, how liquidity changes, and whether insiders can sell into everyone else. For Michael, a tokenized equity and a meme coin are radically different claims but part of the same quantitative problem: observable behavior becomes an input to price.
We described an event-driven tactic: watch Robinhood Crypto’s listing notifications and buy as a newly listed token became available, on the theory that other users would arrive later. The signal was not a view of intrinsic value. It was a discrete expansion in distribution. Like any reaction trade, it depended on acting before that expected demand was already in the price.
Creator behavior is part of the trade. Someone who controls supply, collects fees, or lends a name to a token is asking holders to believe those privileges will not be used against them. Trading activity can, in principle, produce a fee stream that funds development, rewards participation, or gives a creator a reason to keep building.
Michael’s broader definition of finance is useful here. A company, DAO, nonprofit, or creator community is an abstract organization; an instrument gives people a way to express and exchange a view of its value. For a creator-linked token to persist, some of the activity around it has to accrue to work or benefits that holders recognize, while the people with privileged control have to exercise it predictably.
That is why his ending comes back to trust rather than tokenomics. Some level of trust has to exist between the people trading and the people creating. Poison that well often enough and the traders leave.
Professional Punters is presented by Freeport Markets: Freeport lets users trade 24/7, weekends included, with no KYC and up to 200x leverage across stocks, indices, commodities, crypto, rates, and more, with access to pre-IPO exposure in companies like OpenAI & Anthropic. For traders who already follow sharp sources across Twitter, Substack, hedge fund filings, corporate insider activity, and political trading disclosures, Freeport uses AI to read the sources they trust, connect the dots across markets, and surface trade ideas as narratives develop.
About the Host: Lihong Wang is the founder and CEO of Freeport Markets. Before starting Freeport, he traded discretionary semiconductor names at quantitative market-making firms including Jane Street and IMC Trading, building both systematic and discretionary strategies. Lihong graduated from Duke with a degree in mathematics and statistics, and has raised over $2.5M to build Freeport from investors including Y Combinator, Alliance DAO, and Informed Ventures. For more of his long-form research, visit freeportlogbook.substack.com.
Disclaimer: The information provided on TheLogbook (the “Substack”) is strictly for informational and educational purposes only and should not be considered as investment or financial advice. The author is not a licensed financial advisor or tax professional and is not offering any professional services through this Substack. Investing in financial markets involves substantial risk, including possible loss of principal. Past performance is not indicative of future results. The author makes no representations or warranties about the completeness, accuracy, reliability, suitability, or availability of the information provided._
This Substack may contain links to external websites not affiliated with the author, and the accuracy of information on these sites is not guaranteed. Nothing contained in this Substack constitutes a solicitation, recommendation, endorsement, or offer to buy or sell any securities or other financial instruments. Always seek the advice of a qualified financial advisor before making any investment decisions.