MechanizedIntelligence

MIT’s Andy Lo has published another article about the securitization of biomedical research, this time proposing a private-public partnership specifically to create therapies for Alzheimer’s disease.  Pharmalot interviewed Andy and Derek Lowe has some comments. I heard Andy talk about his proposal at a seminar last winter.  The idea of securitizing a highly parallel research effort is compelling, but my take on it is that there are several factors that make the whole thing unworkable:

(1) Given expected failure rates of 95-98% or higher (rates which are actually estimated from historical non-Alzheimer’s research) there would be a need for 50 or more independent projects to get one success.  While there may be that many potential protein targets related to some aspect of Alzheimer’s (see Andy’s article for a list), there aren’t nearly enough independent disease mechanism hypotheses to provide the required “shots on goal”.

(2) Securitization works when the assets making up the portfolio are uncorrelated.  That will not be the case with Alzheimer’s research.  Outcomes will be highly correlated.  Either one or two mechanisms will work well, or none will.  There will not be multiple independent mechanisms that work to varying extents and bring in a range of revenue outcomes.

(3) Andy’s securitization plan includes low risk bond tranches and a high risk equity trance, together totaling $38B.  The bond tranche might find buyers, especially if collateralized or guaranteed.  But who will buy an equity tranche where you don’t even know in advance which specific projects the money will be invested in? Andy proposes that government might be that buyer, but in a world with many competing budget needs, I think that is unlikely (keep in mind that the entire NIH annual budget is only about $30B).  Admittedly, Andy and his co-authors make the point that the U.S. pays $200B in Alzheimer’s costs, and that cutting that expense would be a huge benefit to society.

(4) Unlike typical securitization where the revenue stream and lifetime of the security or investment are fairly well understood, there’s no guarantee that the required Alzheimer’s research can actually be completed in the 20-30 year horizon of the fund.

(5) Massively parallel research means that there would be little opportunity to learn from failures to increase later probability of success.  But often the limiting rate of scientific progress depends on multistep learning, not merely how many resources you throw at a problem simultaneously.

(6) There are potentially serious agency and incentive alignment problems with running a mega fund where the payoffs are so far in the future, while the downside to administrators is low/non-existant.

(7) Finally, Andy proposes government support because of the low expected returns of a privately funded mega portfolio (particularly in scenarios where portfolio project correlations are higher), but the high cost to society of the disease.  However, it’s highly probable that the fund will not be successful.  Given that, there’s potentially a large opportunity cost — after all , the mega fund money could be spent on other, in the end more valuable research efforts.

In contrast to a gigantic parallel research effort, the current paradigm of government-sponsored individual investigator-driven research is likely to pay off more efficiently.  Incremental research that builds knowledge along the way — and as a side benefit trains students and postdocs at the same time — is the way to go.  And if or when a promising disease hypothesis emerges, then companies can fund those large scale efforts.  And if good hypotheses don’t emerge, well, then those investments won’t be made.  And that’s the way it should be.