A while ago, someone asked me to help come up with a financial structure that could help fund basic research. There's clearly a social benefit to it, but this benefit is hard to capture, so the private sector under-invests. The government certainly allocates plenty of money to it, but they need legible signs of progress, so 1) the more fundamental it is, the harder it is to get funded, and 2) imposing that legibility means that many of the most brilliant researchers spend a plurality of their working hours writing grant applications.¹ Independent researchers solve the incentive problem nicely, but they need to get somewhere between well-off and rich before they can afford to dedicate their life to study.

All of this looks like a case where there might be some clever contract that can divide the upfront costs and slice up the benefits in a way that leaves everyone better-off. But it's surprisingly hard to think of one. And one reason for that is that there's fundamental uncertainty about what the upside from the research might be. There are four main patterns to consider:

1. Fully privatizing the gains, monetizing with equity. This is the Genentech story: they were working on a speculative technology with direct commercial applications (insulin was harvested from pig pancreases, growth hormone from human corpses; both of these could be created by splicing the relevant genetic code into bacteria).

2. In-house research for a company vast enough that it'll almost certainly find a way to monetize. Google's moonshots are a good example of this: some projects that didn't go anywhere, some that other companies took advantage of at a wider scale first, and projects like Waymo and Wing, which give Google new directions in which to expand. Bell Labs also fit this profile. AT&T didn't capture most of the upside (the computer industry did), but they got a good return on investment.

3. Publicly-funded R&D projects that create private sector benefits: DARPA was willing to network together computers when there wasn't much of a reason to do this, and their efforts certainly paid off for the rest of the economy.

4. The most important category to consider, from the perspective of inventing a financial instrument, is the most common case: someone attempting approaches one through three and not producing anything of value.

We don't generally know what distribution we're sampling from with this kind of open-ended research project. Is it even meaningful to talk about the odds of discovering a new general-purpose technology, given how rare they are? But, by the same token: the more general a technology is, the harder it is to predict applications, which means it's even harder to predict whether the upside will be captured by the company that discovers the technology or by someone else entirely. The transistor made more fortunes in software than in chips themselves; air conditioning's biggest winners were property owners in Singapore, Dubai, and Atlanta; and one of the highest-margin businesses created by residential electrification was owning broadcast TV networks and local TV stations.

The most successful ways to fund basic research that's meant to generate an ROI are either equity or retained earnings, two of the oldest financial instruments around: equity is what you have if you start a business and don't borrow; retained earnings are what you get when that business was worth starting. In a way it's surprising that these incredibly basic financial instruments would still be the optimal way to fund complicated projects. But there's a conservation of complexity at work: financial instruments get more complex when the underlying assets get simpler; you can trade more elaborate exotic derivatives in rates and currencies than you can with single-name equities, because the underlying products are so liquid and the factors affecting their performance are so well-known. For a product like mortgages, or more recently loans for PE deals, the individual products are idiosyncratic but their collective performance is more predictable, so they end up in structured products.

Financial complexity is generally good at scaling something that works well, by turning statistical abstractions into realities with prices. It's one thing to note that the law of large numbers implies that a diversified portfolio of mortgages or commercial paper is a better thing to own than investing a similar amount in something from a single issuer. But it's a business to put these assets together into a mortgage-backed security or a money-market fund. Options are also a way to price a distribution of future price movements. That distribution isn't perfectly predictable, hence the existence of a market for options that often features wide bid/ask spreads. But it's a distribution that can be talked about, where distinctions like "realized vol will be closer to 55% than 50%" have predictable impacts on the prices market participants will pay.

The more early and speculative a given research project is, the more the distribution is some unknown-but-high chance of failure, and an incredibly wide range of outcomes for success. Occasionally, one big discovery can be turned into a company, but more often that discovery happens in a context where it will spawn multiple companies, and a whole supply chain of use cases.

Financial complexity can and does solve a lot of problems. Mortgages are cheaper now that they're packaged together, and passive investors are compounding a point or two faster than they otherwise would have. Options have simultaneously increased the returns for thinking statistically about how the world will change and made it possible for people who don't want to think about those distributions to pay a fair price to truncate them. All this activity does make the economy more efficient in the end, but to the extent that it helps to finance innovative deep research, it's through the side effect of making all of us richer.

We've looked at the question of how to finance big bets, and what complex financial systems get right, from many angles. Some Diff pieces covering this include:

• How bubbles and megaprojects parallelize innovation: if you have a list of dependencies for some project, it helps if you know someone's taking care of all of them for you.

• We've examined what looks like a paradox in volatility-reducing policy.

• Parametric insurance is financial engineering in the service of making it cheaper to insure against natural disasters ($).

• For more on the point on air conditioners and value capture, see this piece for how climate control led to consolidation among financial centers ($).

• In the earliest stages, you can't financially engineer your way to more investment in R&D. But once the business is scaling, as AI is, there are lots of opportunities to get creative ($).

Join the discussion!

1  AI may eventually contribute directly to scientific discoveries, but for now its biggest benefit in that regard is that it can speed up some of the bureaucratic processes.