What a Quantitative Researcher Does All Day

A quantitative researcher's day moves from a morning data pull through hypothesis testing to a meeting that decides an idea's fate.

A quantitative researcher spends the day testing whether an idea survives contact with data, moving from a data pull in the morning through a hypothesis test in the afternoon to a research meeting where the desk decides whether the idea is worth building into a live strategy. Unlike a trader, a researcher's day is not bound to the exchange clock, and the output is a document or a piece of tested code rather than a stream of executed trades.

The morning usually starts with checking that the previous night's data pipeline ran cleanly: tick data, fundamental data, alternative data sets depending on the desk. A researcher at a firm such as Two Sigma or AQR Capital will spend real time here, since a data set with a survivorship bias or a misaligned timestamp can produce a strategy that looks profitable in a backtest and loses money the moment it goes live. Cleaning and joining data sets correctly is unglamorous and takes up a meaningful share of the working week.

Once the data is confirmed clean, the researcher checks the results of any test that was left running overnight; a full backtest across ten years of tick data on a reasonable universe of instruments can easily take that long, and a researcher will queue up several variations before leaving the previous evening specifically so the results are ready by morning.

Before running new tests, the researcher writes down the hypothesis in specific terms: what economic or behavioral mechanism should produce the effect, over what horizon, and what would disprove it. This step is where a research career diverges sharply from an academic one, since a hypothesis at a trading firm has to be tradable at a realistic size and net of the transaction costs the desk actually pays; statistical significance on a chart alone is never enough to justify building it.

A common failure mode at this stage is testing so many variations of an idea that one of them passes a significance test by chance alone. Desks manage this with formal multiple-testing corrections and, more practically, by requiring that a result be explainable through a mechanism before it is taken seriously. The backtesting article covers the specific biases this process is designed to catch.