Quantitative research and the systems our clients trade on.

A backtest reports a level. The curvature around that level is what says whether it is stable, whether the parameters were ever identified, and how the risk behaves when two things move at once.

The same object, the matrix of second derivatives the firm is named for, is the gamma matrix on a derivatives book, the step in Newton's method, the observed Fisher information on a fit, and the test that separates a genuine optimum from a ridge. Most work stops at first order. That is where the failures come from.

It is also a disposition rather than a technique. First-order questions ask which way; second-order questions ask what shape the answer sits in, where it is fragile, and what it rests on that has not moved yet. That is what the firm is organised around, and it decides several things that would otherwise be matters of taste: method over outcome, negative results reported as results, and no track record offered.

Engagements are taken on globally, across equities, futures, options, fixed income, FX, commodities, credit and digital assets: listed and over the counter, linear and non-linear, vanilla through to structured and path-dependent payoffs.

Strategy work spans the frequency spectrum, from low-frequency systematic through intraday to latency-sensitive execution. Methods are chosen by the shape of the problem: classical statistics and econometrics, numerical optimisation, machine learning where the interactions are real and the evaluation can be made honest, and prediction markets as both a research object and a source of event probabilities.

This is research and engineering performed for clients; the firm does not trade its own book.

A delivered system spans the whole chain: normalised market data, a research environment, an order path that speaks FIX or a native binary protocol, and the reconciliation, risk controls and cost measurement around it. Work is scoped against whichever links a client already operates.

• What a backtest gets wrong about fills A backtest reports the price you asked for. A market reports the price you got. Almost every way that gap opens up is systematic, and almost every one of them flatters the result.
• Point-in-time data and the shapes lookahead takes Lookahead is rarely a line of code that reads tomorrow's price. It is usually a dataset that quietly knows how the story ended, and the correction is a property of the data rather than of the model.
• Gamma exposure: signal or description? A dealer gamma estimate is a restatement of positioning, computed from a chain everyone can see. Whether it predicts anything is a separate question from whether it describes something, and the two are routinely merged.

Three ways in: a research engagement answering a defined question, a scoped build delivered end to end, or an independent review of a system that already exists. Each is scoped before work starts and handed over in a form your team can operate.

Work begins with a short scoping conversation about what exists now, what it needs to do, and whether this is the right fit. If it is not, you will be told so.