Data-Over-Opinion Research
Approach markets with no preset answer; let data tell you what's real.
Most investors decide their conclusion first and then hunt for data that confirms it. Boyle's approach inverts this: start with no opinion, get historical data, and run tests to see whether a claim survives. The 90% of research that finds nothing is the point — it's how you isolate the rare signal that's real.
The framework treats investing claims (cheap stocks beat expensive, value beats growth, momentum works) as hypotheses to falsify rather than truths to defend. Statistics matters because it tells you whether a finding is signal or random artifact of the data. The discipline protects you from your own biases and from finance gurus who pound the table.
Applied consistently, it builds a small set of robust ideas you actually trust — and a healthy scepticism toward any prediction, including your own.
- Approach every market claim with no predetermined answer.
- Most research finds nothing — that's the cost of finding the rare real signal.
- Statistics exists to tell you whether a result is real or random.
- Results-driven research means torturing data until it agrees with you.
- The future isn't predictable, but data lets you make the decisions a wise person would make.
- Frame the claim as a falsifiable questionTake a market belief ('value beats growth', 'cheap stocks outperform') and rewrite it as something testable with data. Be precise about the time period, universe, and metric.Pro tipWrite the question down before you look at any data — it stops you reverse-engineering the test to fit a hunch.
- Pull long-history dataGet the longest clean dataset you can. Boyle uses 100 years of price-earnings ratios as an example. Short windows mistake regime for rule.WarningA signal that only works in the last 10 years may just be a regime, not a rule.
- Run the simplest possible test firstSort the universe in half (cheap vs expensive, growth vs value) and compare returns. Don't add complexity until the simple version is understood.Pro tipIf a basic split-sort doesn't show anything, fancy machine learning won't save it.
- Apply statistical tests for significanceUse stats to ask 'is this real or random?' rather than just eyeballing returns. A finding without significance testing is a story, not a signal.
- Discard 90% of findings without regretMost ideas you test won't hold up. Throwing them away is doing the job correctly. Keep only the 10% with both economic logic and statistical strength.Pro tipTrack what you killed and why — sometimes a dead idea becomes useful when conditions change.
- Hold confidence looselyEven your survivors decay as others discover them and arbitrage them away. Treat every signal as having a finite shelf life and keep researching.WarningIf you find cheap stocks and others copy you, they stop being cheap — the market changes as you interact with it.
Boyle describes the simplest possible quant test: take 100 years of P/E ratios, split the universe in half, and compare returns of the cheap basket to the expensive one. No theory required — the data either supports value or it doesn't.
Boyle recounts a story from Jim Simons' biography: a Renaissance trader explained their approach using Chrysler stock as an example — a company that had stopped existing five years earlier. Quant investors don't care about the name; they care about statistical properties.
Boyle started in markets in 1997 looking 21, struggling to be taken seriously next to senior traders. He couldn't buy 20 years of experience, so he bought books, built spreadsheets, and tested every claim he read. Moving from Ireland to the US with few friends and a new computer, he spent nights running tests like 'are cheap stocks really better than expensive ones?'
That habit of testing instead of believing pulled him into statistics — not the abstract calculations he'd been taught at university, but as a tool to answer 'is this signal real or noise?' That bottom-up curiosity became the basis of his quant career and his YouTube channel.