Estimating How Long a Research Project Will Take
Quant research timelines are notoriously hard to predict because most of the time goes into unknown-unknowns — data problems and dead ends that only surface once you start — so estimates should be built from past project histories, not from task lists.
Ask a researcher how long it will take to test a new signal, and the honest answer is usually "longer than I think," because the estimate is built by imagining the happy path — pull the data, compute the feature, run the backtest — while the actual time sink is almost always something unplanned: the vendor's data has a silent schema change halfway through history, the feature needs three redefinitions before it's stable, or the promising early result turns out to be a lookahead bug that takes a week to track down.
This is the same planning fallacy documented across many kinds of project work: people estimate based on the specific plan they have in mind, not on how similar projects have actually gone in the past. The fix is to estimate from the outside in rather than the inside out — instead of decomposing a specific new idea into steps and timing each one, ask how long the last five similarly-scoped ideas actually took from first data pull to a stage-gate decision, and use that distribution as the estimate, adjusting only slightly for anything genuinely unusual about this one.
In practice this means research teams that track project durations end up with much better forecasts than those that re-estimate from scratch each time, because the historical distribution already encodes the base rate of things going wrong that any single plan will miss.
Estimate a new research project's duration from how long comparable past projects actually took (an outside view), not from timing the individual steps of this specific plan (an inside view) — the inside view systematically ignores the unknown-unknowns that dominate actual quant research timelines.
Further reading
- Kahneman, Thinking, Fast and Slow, ch. 23 (the planning fallacy)