Ahead of the (Yield) Curve (summary)

Last updated: 29 September 2026

Summary of an essay by Abraham Thomas, published in Pivotal on 4 December 2024. Read the full essay on Pivotal →

Abraham Thomas was one of the lead builders, and the main day-to-day trader, of an early automated, proto-high-frequency trading system for US Treasury bonds at Simplex Asset Management, a Tokyo-based quantitative hedge fund. Starting after the 1998 collapse of Long-Term Capital Management, a five-person team built a system that tracked prices, ran yield-curve models, identified opportunities, designed and executed trades, and managed hedges “all in a matter of seconds”, at a time when most Treasury trading was done by phone. Its core idea was to profit from short-term noise in mean-reverting spreads rather than waiting for them to converge. In the US market, high-frequency curve trading grew from 10% to 50% of Simplex’s P&L, and to 80% in some months.

How did it start?

Thomas joined Simplex as a programmer-analyst in August 1998. A month later LTCM, “the world’s largest and most celebrated hedge fund”, blew up. Simplex, newly launched and trading many of the same quantitative convergence strategies, briefly had those opportunities to itself (“better lucky than smart”). But LTCM’s collapse raised doubts about the whole strategy class, “picking up nickels in front of a steamroller”. Simplex needed a strategy that didn’t rely on market-to-model convergence, didn’t use excessive leverage, and didn’t correlate with other investors’ positions.

What was the big idea?

Monetizing noise. Convergence trading bets that a spread between related securities will revert to zero, and blows up when it diverges instead. But even while a spread fails to converge, it oscillates. Selling every interim peak and buying every interim trough makes money “despite the lack of convergence.” It seems obvious now, but it wasn’t then: colleagues, the investment committee, risk managers, LPs and the prime broker all took convincing.

The obstacle was bid-ask cost. The techniques that overcame it included:

Why US Treasuries?

The market was efficient enough for quant models, volatile enough for noise-trading, and liquid enough for near-zero bid-ask. It was also technologically “stuck in the 1980s”: most trades were by phone, electronic execution was under 10% of the market, and few participants priced the curve well intraday. “A perfect market for us!”

How was it built?

Component How Simplex did it
Data Brute force: morning, noon and closing price runs from 5 major Treasury dealers (later from their Tokyo, London and New York desks); manual entry by back offices in Tokyo and Hong Kong; automated screen-grabs of trading venues, Bloomberg and Reuters; prime-broker prices. Cleaned and merged into one “golden” time-stamped price per bond, plus synthetic constant-maturity “virtual” bonds, whose maths Thomas worked on
Model “N3”, a non-linear yield-curve model with 12 parameters: 8 structural constants calibrated about once a year, and 4 daily factors (overnight funding rate, expected real growth, expected inflation, and risk premium). Letting the risk premium vary was one of Thomas’s modelling decisions after LTCM
Speed The breakthrough: solve N3 fully about every 60 minutes, and between solves use linear approximations, so each market move needed only matrix multiplications. “Matrices all the way down”
Execution Direct access to inter-dealer brokers’ electronic platforms, plus an automated voice-trading workflow with dealers: generated quote requests, parsed dealer quotes and trade tickets, and a human saying “done” or “nope”
Team Five people, all under 30; Thomas was 24. He “bridged trading and tech”: model, data and strategy R&D, many prototypes, and the most consistent day-to-day trading. He set up a satellite trading office in New Jersey
Timeline About a year of research, a year building the core system, and a year ramping up trading

Why did it work so well?

Beyond the P&L, the system exploited an industry blind spot: risk, credit limits and margin were computed on end-of-day positions, and the intraday noise-trading book was usually flat by the close. The result was little margin to post, little overnight leverage or financing, excellent return on balance sheet, and P&L rarely correlated with other market participants. “No requirement of model convergence; limited leverage or financing needs; minuscule margins to post; and low correlation with the rest of the market: this was the holy grail.” The proprietary strategy behaved like a successful market-making desk. Thomas’s aphorism: “Any sufficiently advanced form of prop trading is indistinguishable from market-making.” The team extended it to multiple models and to swaps, eurodollars, futures, options, and cross-currency basis.

What happened next?

Alpha decay. Competitors converged on similar approaches, and the 1–2 basis-point round-trip opportunities that used to appear daily disappeared. The high-frequency book’s share of P&L, having peaked at up to 80%, fell back to around 20%. Simplex recognized the decay early and moved to other strategies rather than scaling up positions. Thomas left Simplex in 2006 and later co-founded Quandl.