Brett Harrison — FTX US former president speaks out

Patel opens with the claim that passive investing dilutes smart money, since active investors pay fees that subsidize price discovery. Harrison, formerly head of ETF technology at Citadel, disagrees. Most investors should not pick individual stocks, he says, and ETFs have made diversified index investing far cheaper than before.

His one worry is correlation. When Tesla joins the S&P 500 the company is unchanged, yet its beta to other members shifts because passive flows move all constituents together. He calls this potentially detrimental but admits it is hard to know what markets would look like with less passive money.

Patel notes that efficient markets imply traders would short an irrationally rising stock. Harrison agrees this mostly happens, but says liquidity differences cause temporary dislocations. An ADR should track its foreign share closely yet can diverge if one market is illiquid, and firms like Jane Street arbitrage such gaps away.

Asked whether the big market makers run similar strategies, Harrison says they differ far more than outsiders realize. Jane Street's early niche was mid-frequency ETFs, Citadel Securities' was lower-latency options market making, and horizons range from milliseconds to months. What they share is infrastructure: exchange connectivity, market data, simulation platforms and tools to visualize traders' models.

Asked for a finance equivalent of Carmack's fast square root, Harrison describes a trick he heard about from developers at other firms, not one he saw at his own employers. A trader who will buy Apple at an unknown price pre-builds the whole TCP/IP message to Nasdaq, leaving only the price bytes blank, and preloads it into the network card's send buffer.

the goal is to get to the approximately right trade as fast as you can

Harrison compares this to video-game graphics, where output only needs to look good enough and run fast enough. Patel asks what society gains from nanosecond advantages, microwave towers and co-located servers.

Harrison argues that if liquid, tight markets matter, sophisticated technologies will compete at the margins, and no one can design a market that tells people to stop innovating at a given point. He adds that the US microwave network, paid for by quantitative finance, might find other uses. Patel calls that a contingent spillover, like the NASA story, not a direct benefit of fast fills.

On direct value, he says news in one part of the world should reach prices elsewhere with as small a gap as possible, and nobody can say a second is too slow but a microsecond too fast.

Patel tests the limit with a hypothetical $5 trillion a year to shave a nanosecond. Harrison does not know prop trading's share of GDP but suspects it is small, and reasons that firms spend only because they profit, so the market is paying for a service. He doubts any network could attract $5 trillion, since there is not that much value to create.

Does market making change how Harrison thinks about civilizational risk? He says it improves estimates of rare-event probabilities but suggests no different solutions. A financial book can be hedged as a closed system, while pandemic-style risks involve collective action and domain unknowns that belong with experts. He adds that the trader mindset may tempt people to think they can do something about such risks.

Finally, Patel asks whether one matching engine could serve an interplanetary market. Harrison says nothing about the engine forbids it, but a distant sender's market order arrives at a different price than expected, which means far more adverse selection. An interstellar exchange is technically possible, he says, but might not be good for a person on the moon. The segment ends before Patel's question about fairness is answered.

Harrison answers the fairness question by pointing to a real analogue: automated market makers on slow blockchains. Nasdaq handles a message in tens to hundreds of nanoseconds, while Ethereum handles roughly 15 to 50 messages per second, yet those chains built fairly mature marketplaces. An automated function moves the price at the matching engine, so an arriving order always meets a predetermined fair price based on prevailing liquidity.