What is quantitative trading?

The pipeline, stage by stage

Almost every systematic trading firm, whatever its holding period, runs some version of the same five-stage pipeline. The differences between firms are differences of emphasis within it.

1. Data

Everything begins with a dataset and a view about what it says. That might be exchange market data, fundamental filings, reference data, or an alternative source such as shipping movements or card transactions. The unglamorous work dominates: aligning timestamps, handling corporate actions, identifying survivorship bias, and building a point-in-time view so that a backtest only ever sees what was actually knowable at the time.

A great deal of apparent alpha is a data error. Teams that treat data engineering as a junior function discover this expensively.

2. Signal research

A signal is a number that carries some predictive information about future returns. Research is the process of proposing signals, testing them, and discarding the large majority. The central discipline is avoiding overfitting: with enough parameters and enough attempts, any dataset will yield a beautiful backtest that means nothing.

Good research practice involves holding out data, limiting the number of hypotheses tested, preferring economically motivated signals over discovered ones, and treating a strategy's decay as expected rather than exceptional.

3. Portfolio construction

Individual signals are rarely traded directly. They are combined, weighted by confidence, and sized against risk constraints, exposure limits and correlation with existing positions. This is where a collection of ideas becomes a portfolio with a target volatility and a defined set of factor exposures.

4. Execution

A target portfolio has to be reached through actual orders in actual markets, and the cost of doing so determines whether the strategy is viable. Execution research covers order placement, venue selection, scheduling and the measurement of implementation shortfall. For a strategy with a holding period of weeks, execution cost is a drag to be minimised. For a strategy with a holding period of seconds, execution is the strategy.

5. Risk and controls

Live strategies run inside a system of pre-trade checks, position limits, loss limits and kill switches. This layer is what stops a software fault or a market dislocation from becoming an existential event.

Strategy families

Holding period is the most useful axis for classifying systematic strategies, because it determines the technology, the capacity and the team.

A firm's position on this spectrum tells you more about its hiring than its name does. A market-making firm competes on C++, kernel bypass networking and FPGA work. A mid-frequency equity fund competes on data breadth and research throughput, and may run entirely in Python. Both call themselves quantitative.

What quantitative trading is not

Three persistent misconceptions are worth clearing up.

It is not prediction with high accuracy. Profitable strategies are routinely right slightly more often than they are wrong, or right about the same number of times with a better payoff. A signal with a 52% hit rate, traded at scale with disciplined sizing, is a serious business.

It is not primarily machine learning. Modern methods are widely used, particularly on large cross-sectional datasets and in execution. But the binding constraint is usually the signal-to-noise ratio of financial data, not model capacity. Simple models, carefully validated, remain the backbone of most live books.

It is not a solved problem. Edges decay as they are discovered by others and as market structure changes. A firm's durable advantage is the speed and rigour of its research loop, not any individual strategy.

Talent implications

The pipeline above is also an organisation chart, and understanding it is how a hiring manager avoids searching for a person who does not exist.

• Quantitative researchers own hypotheses, data and signal validation. They are hired on research judgement and statistical discipline, usually from PhD or strong postgraduate backgrounds in mathematics, statistics, physics or computer science.
• Quantitative developers build the research platform, the backtester and the production implementation of strategies. They are hired on software engineering quality; the best ones understand the research well enough to challenge it.
• Quantitative traders own live risk, calibration and the day-to-day behaviour of a book. At a market-making firm this role is closer to systems operation; at a discretionary-overlay fund it retains real judgement.
• Core infrastructure and low-latency engineers own the path from exchange to decision and back. In high-frequency firms, this group frequently commands compensation equal to or above research.
• Risk and quantitative risk own the limit framework and independent valuation. This function is routinely under-hired at growing firms and is usually the first gap a regulator or an investor finds.

Two practical consequences follow for anyone hiring into this space. First, a job title tells you very little across firms: "quant" at one firm means signal research and at another means production C++. Define the stage of the pipeline before you define the role. Second, a candidate's holding-period background matters more than their asset class. Moving a researcher from equities to futures is routine; moving one from a monthly to a microsecond horizon rarely works, because the skills that matter are different.