Incremental vs Moonshot Research Bets

Research managers face the same allocation problem venture capitalists do: put every dollar into safe, incremental bets and you compound steadily but eventually plateau, because the low-hanging fruit runs out. Put too much into unproven long shots and most of your budget disappears with nothing to show for it. Most quant research teams handle this the same way a VC fund does — by consciously running a portfolio of both, rather than picking one style and committing entirely.

The idea

An incremental bet extends or refines something already known to work: adding a new sector-neutral variant of an existing factor, re-fitting a signal's decay parameters on more recent data, or extending a proven signal to a new geography. These projects have a high probability of producing a usable result, because the underlying edge is already validated, but the improvement each one delivers is typically small — a modest bump in Sharpe ratio or capacity, not a new source of return.

A moonshot bet explores something genuinely new: a novel data source nobody has used before, an entirely different modeling approach, or a hypothesis about market structure that hasn't been tested by anyone on the team. Moonshots fail far more often than incremental projects — most exploratory data sources turn out to carry no signal, most novel model architectures underperform simpler ones once transaction costs are included — but the rare success can open an entirely new, uncorrelated source of return that incremental work on existing signals could never produce, because it isn't just refining an existing edge, it's finding one that didn't exist in the portfolio before.

A concrete example

A ten-person research team sets an informal target: roughly 70% of research hours go to incremental work — extending existing factors, improving execution on live signals, refreshing stale models — because that work reliably keeps the current book healthy and funds the team's near-term numbers. The remaining 30% is protected for moonshots: this quarter, two researchers spend it testing whether a newly available satellite-imagery dataset predicts retail sales before official releases. Nine times out of ten, a moonshot like this ends in a null result and gets shelved, which the team treats as a normal, budgeted outcome rather than a failure — the 30% allocation was priced assuming most of it wouldn't pay off directly. When one does work, as this one eventually does, it's often worth more than a year of incremental refinements combined, because it adds a genuinely new, low-correlation return stream rather than squeezing more out of an existing one.

What this means in practice

The right split between incremental and moonshot work depends on a firm's stage and risk appetite — a young fund still building its first few strategies may lean almost entirely incremental just to survive, while an established, well-capitalized fund can afford a larger moonshot allocation because a string of failed long shots doesn't threaten the business. The mistake to avoid is not choosing a split deliberately at all, and then discovering a year later that every research hour went to incremental work by default, because moonshots have no natural deadline forcing anyone to start one.