How a rule earns capital
Walk-forward windows, placebo tests and full trading costs: the three hurdles every idea must clear before it trades.
Most trading ideas look good on the data they were built from. That is not evidence; it is a description of the past. The question that matters is whether a rule keeps working on data it has never seen, after the costs of actually trading it. Our research process is built around making that question hard to dodge.
1. Walk forward, never backward
We fit a candidate rule on one stretch of history and test it on the stretch that follows, then roll both windows forward and repeat. The rule only ever sees the past when it makes a decision about the future, exactly as it would in live trading. Results are judged on the stitched-together out-of-sample periods alone.
We also split the full history in half and require a rule to hold up in both. A rule that made all of its money in one regime is a rule about that regime, and we treat it that way.
2. Beat a placebo
Markets drift, and many rules look profitable simply because they were long during a rising market. To separate skill from exposure, every candidate is compared with placebo versions of itself: the same holding period and the same number of trades, but fired at random times. If the real rule cannot clearly beat its own placebos, it has not found anything.
Where the sample allows, we bootstrap the results to see how much of an apparent edge could plausibly be luck, and we prefer rules whose evidence is broad (many trades, many instruments) over rules that hinge on a handful of outliers.
3. Pay full price
Every test charges realistic spreads, commissions and slippage, using the actual bid and ask where we have it. For options this matters enormously: a rule that looks excellent at the midpoint price can be worthless at the price you can really trade. Costs are not an afterthought applied at the end; a rule is designed around them.
What survives
Few ideas clear all three hurdles. Those that do enter the book small, are watched against their own test record, and are retired if live behaviour drifts away from what the evidence said to expect. Nothing is permanent; everything is re-earned.
This note describes research methods. It is not investment advice and does not describe the performance of any account or fund.