90 / 97
The Two Lenses
Working definition
A framework in quantitative trading where every metric is evaluated through both its theoretical statistical properties and its physical microstructure execution constraints.
Two questions have to be answered about any quantitative result before it means anything, and answering only one of them is the most common way research goes wrong.
The first lens is statistical. Is the effect distinguishable from noise, given everything that was tried in order to find it? That question cannot be answered without a trial count, which is why data snooping, the deflated Sharpe ratio and the kill ledger all belong to this lens. A result that survives it has earned exactly one claim: that it is probably not an artefact of searching.
The second lens is physical. Could the result have been obtained by an order actually sent to an actual venue? The constraints here are the ones a spreadsheet never enforces — the spread that had to be paid, the depth that existed at the price, the latency between decision and arrival, the requote or rejection that turns a fill into a non-fill, and the financing charged while the position was held. A result that survives this lens has earned a different claim: that it is executable.
The two failure modes are symmetric. A finding that passes only the statistical lens is a real pattern that costs more to harvest than it returns — the familiar short-horizon reversal that evaporates the moment execution cost is charged honestly. A finding that passes only the execution lens is cheap to trade and never had an edge to begin with; it will trade smoothly all the way down.
Neither lens outranks the other and neither substitutes for the other, which is also why the two must not be collapsed into a single headline number. A composite score hides which lens the result failed, and the two failures call for opposite responses: one is fixed by more evidence, the other by a cheaper way to trade — or by not trading it at all.
Commonly confused with
Neighbouring concepts that get used interchangeably, and the distinction that actually separates them.
- A composite score
Collapsing both lenses into one number hides which one the result failed, and the two failures call for opposite responses. One is answered with more evidence; the other with a cheaper way to trade, or with not trading it at all. A single score cannot tell you which conversation to have.
- Statistical significance
Significance is the first lens alone. A result that survives it has earned exactly one claim — that it is probably not an artefact of searching — and that claim says nothing about whether an order could have captured it.
- Backtesting with costs
Charging costs is part of the second lens rather than the whole of it. Depth at the price, latency between decision and arrival, and the requote or rejection that turns a fill into a non-fill are constraints a cost assumption does not represent.
- A ranking of the two
Neither lens outranks the other and neither substitutes for the other. A real pattern that cannot be harvested and a harvestable pattern that was never real are both worthless, and they are worthless for unrelated reasons.
How to measure it in your own data
A definition you cannot test is a definition you have to take on trust. This is the shortest honest route from the concept to a number you computed yourself.
- Records you need
For the statistical lens, the trial count and the out-of-sample structure. For the physical lens, the spread paid, the depth that existed at the price, the latency between decision and arrival, the rejections and requotes, and the financing charged while the position was held.
- What you compute
Answer both questions separately and report them separately. Is the effect distinguishable from noise given everything tried to find it; and could the result have been obtained by an order actually sent to an actual venue.
- What the answer tells you
The two failure modes are symmetric and diagnostic. A finding that passes only the statistical lens is a real pattern that costs more to harvest than it returns — the familiar short-horizon reversal that evaporates once execution cost is charged honestly. A finding that passes only the physical lens is cheap to trade and never had an edge; it will trade smoothly all the way down.
Questions and answers
Why not combine both lenses into one score?
Because a composite hides which lens the result failed, and the two failures need opposite responses. A statistically weak result wants more evidence. An unexecutable one wants a cheaper way to trade — or abandoning. A single number tells you something is wrong and withholds the only part that would help.
What does passing the statistical lens actually establish?
One narrow claim: that the effect is probably not an artefact of the searching that found it. That is worth having and it is not a claim about profitability, capacity, or whether any order could have been filled at the prices the study assumed.
What does passing only the execution lens look like?
A strategy that trades smoothly, fills as expected, costs what it was modelled to cost — and loses. Cheap execution of a nonexistent edge is still a loss, and the smoothness makes it harder to diagnose because nothing in the execution record looks wrong.
Is this just "account for costs"?
Costs are part of it and not the whole. The physical lens also asks whether the depth existed at that price, what the latency between decision and arrival was, and whether the fill would have been a fill at all rather than a requote or a rejection. Those constraints are not representable as a cost assumption.
Related terms
Derived from the links this entry makes and the entries that link back to it.
In the research
The Two Lenses comes up in three research notes on this site.