Quantitative trading has long been built on correlation. Factor models, regression, and machine learning all assume that statistical relationships in historical data will persist. When X and Y moved together in the past, the logic goes, they will move together in the future. This assumption is often wrong. Correlations collapse during regime shifts. Spurious relationships vanish when the hidden confounder changes. The models that looked robust in backtests fail in live markets because they never modeled why the relationships existed — only that they did.
A causal diagram — a directed acyclic graph (DAG) — encodes structure that correlation cannot. It answers: Does Fed policy cause treasury yields, or do yields cause policy expectations? Does oil drive inflation, or does inflation drive oil demand? Correlation gives you a number. Causal structure gives you a direction. In trading, direction is everything. When you know A causes B, you can predict how B will respond when A changes. When you only know A and B are correlated, you cannot distinguish cause from effect — or from a third factor driving both.
Traditional quant models suffer from another weakness: they treat all relationships as symmetric. A Granger test or a correlation matrix doesn't tell you who causes whom. It tells you who moves first, or who moves with whom. Causal discovery algorithms — PCMCI, FCI, and dozens of variants — infer directed edges from data. They separate direct causes from indirect effects, identify confounders, and produce a graph that supports intervention queries: P(Y | do(X)), not just P(Y | X). For a trader, that means answering "If the Fed hikes 50bp, what happens to my portfolio?" instead of "When Fed and yields moved together before, what was the coefficient?"
The structural advantage compounds over time. Correlation-based models degrade when regimes shift because the correlation structure changes. Causal models can detect regime shifts — when the graph structure changes, you know the world has changed. They can also adapt: the same do-calculus applies to the new graph. Causal trading infrastructure doesn't just fit the past; it reasons about the future under interventions. That's the difference between curve-fitting and structural understanding.
Abel is built on this premise. Market data is the signal layer — prices, volumes, macroeconomic series — but the engine is causal. Graph discovery learns the DAG from data. Do-calculus computes intervention effects. Propagation delays (tau) between nodes capture how fast information flows. The result is a quantitative framework that doesn't just predict; it understands. As more capital moves into systematic strategies, the edge will belong to those who model causation, not just correlation. Causal diagrams aren't a nice-to-have. They're the future of quantitative trading.