Topic · Quant Research
← All topicsAlpha Research Methods
43 articles · 7 checkpoints · 26 deeper reads · 10 reference notes
A standalone topic: it is on no roadmap, so read it on its own terms.
Every article, in reading order
plant a flag as you finish eachRead these first
A column of numbers is not a signal. Getting from a raw data field to something you can rank names on and trade means a specific chain of decisions, timing, transform, scale, neutralise, each of which can quietly break the result if done in the wrong order.
A signal that back-tests well still needs an answer to one question: who is losing money to it, and why do they keep doing it? If you can't name a counterparty, you probably don't have an edge, you have a coincidence.
Four families of edge, risk premia, structural constraints, behavioural error, and informational or analytical advantage, sorted by who is on the other side and why they keep paying. Which family a signal belongs to predicts how long it lasts, how big it gets, and what evidence should convince you.
No single signal predicts returns well. The entire business case for a systematic research team is that many weak, mostly-independent signals combine into something much stronger than any one of them, but only if they're genuinely independent.
A signal researched on frictionless returns and only later checked against costs is a signal designed to fail that check. Building costs into the objective from day one changes which version of the signal you end up with.
A signal that earns 3% a year on $50 million and 0.2% on $2 billion isn't two different strategies, it's one strategy with a capacity limit. Estimating that limit before you scale is what keeps a good idea from being traded to death.
Before you can claim credit for a new signal, you have to prove it isn't just value or momentum wearing a disguise. Orthogonalisation strips out the part of a signal explained by known factors, leaving only the piece that is genuinely new.
Then the rest