Research
Research Insights
What the screening record taught us, one idea at a time. Each entry carries the hypothesis as it was pre-registered, what the data showed, the number that settled it, and the exact condition that would turn it into a tradeable edge.
It is published because the lessons compound. Knowing which tests decide an idea is the fastest way to design the next one better, and it is the part of this work that cannot be reproduced by asking an AI to write you a backtest.
The insights
- Adding contracts to a position that is already winningEvery add earns less per unit of risk than the first lot and is correlated with it. Across forty-two schemes, flat sizing came out ahead.
- An edge that only exists in the last three yearsNinety-eight percent of a six-stream book's lifetime profit sat after 2022, which is why every result is now split by era before it ships.
- Picking the strategy with the best return-to-drawdown ratioReturn persists weakly out of sample and the drawdown ratio does not, so rank on return and correlation and let the portfolio shape the drawdown.
- Adding a filter to a strategy that already worksA filter that keeps half the trades needs a 41% lift in average R to draw level, which is the bar to design against.
- A strategy that only works on the 30-minute chartHold the entry instant and the stop distance fixed and an 18x span of bar sizes pays the same, which makes the strategy far more robust than it first appeared.
- Filtering a signal by a state variable from the same barsRegime conditioning works when the state is exogenous and lagged; measured on the same bars it becomes arithmetic.
- Waiting for a pullback before entering a breakoutThe better price is real, and the sessions that never pull back are the biggest winners, so the entry has to be conditioned on something else.
- Running a strategy on volume bars instead of time barsIt replicates cleanly and matches the time chart. The mechanism is real and monotone; it simply lands in the bracket, not in the signal.
- Range bars improve a breakout strategyThey lift the backtest by tens of thousands a year, and the lift is a bar-construction effect the time chart makes visible.
- Trading strategies on thin futures contractsStrong measured edges on thin contracts need a capacity curve beside them, because a standard gate stack never looks at volume.
- Order-flow signals ported from futures to single stocksEquity-tape delta carries roughly half the information of the futures tape, which sets the bar a single-stock signal has to clear.
- Averaging into a position on a price gridA constant position at matched risk beat every scale-in scheme tested, and the shuffled-bar placebo shows why.
- Fading the edge of the volume-profile value areaThe reversion premise did not hold at any zone definition, and the break needs an independent tape to confirm on.
- Per-bar order-flow delta predicts the next barThe delta bar is the price bar restated, so the information lives in cumulative flow measured against price, not in the per-bar figure.
- Running ES and NQ strategies for diversificationThey move as one factor at +0.92, so the second index contract works best as a confirming tape rather than a second bet.
- Trading the opening-range breakout before 10:00 ETThe same signal carries information after 10:00 that it does not before, so the clock is the lever.
- Daily-bar momentum on index futuresReturn persistence at the daily horizon reads zero, and most of the index drift accrues overnight.
- Letting walk-forward optimization choose your stopIt gravitates to the tightest stop whenever the objective restates the parameter, so score in dollars at constant risk instead.
It is published because the lessons compound. Knowing which tests decide an idea is the fastest way to design the next one better, and it is the part of this work that cannot be reproduced by asking an AI to write you a backtest.
What the tests reveal
Almost every finding here is about measurement, not about the market being efficient. The test was reading something other than what the trader thought it was reading, and once that is named, the idea can be rebuilt on firmer ground.
That is the good news: a pattern you can name is a pattern you can design around next time.
Deciding tests on record
18- Matched-risk comparison
- Era split
- No out-of-sample content
- Sub-sampling deficit
- Confounded comparison
- Circular conditioning
- Adverse selection
- No improvement over the control
- Bar-construction artefact
- Liquidity / capacity
- Signal strength / benchmark
- Placebo / matched-risk comparison
- Trade population / no confirmation
- Redundant signal
- Portfolio overlap
- Session timing
- Drift null
- Objective restates the parameter
- Settled
- 16
- Conditional
- 2
- Markets
- 13
The insights
18 publishedScreen your own idea — the first verdict is free
Describe it in plain English. The screener turns it into an exact specification, runs it through seven gates written down before the run, and answers in seconds.