Monday: Billionaire Biologists

It is by will alone I set my mind in motion.

The NYTimes has an article about the world of D. E. Shaw quants and an Larry Summers.  Louise Story, who wrote the piece, doesn’t go into details about what it is that Shaw’s group does exactly and, to be honest, this is really a job for someone like David Kestenbaum, the Harvard trained physicist turned financial guru at Planet Money (Hint, hint).  But I do know a smidgeon about what Shaw’s group does in the pharma field and at least one of my former colleagues was approached by someone in finance like Shaw more than a decade ago.  She was a physical chemist working on a program that did sensitivity analysis.  The math would make your eyes glaze over because it’s all eigen this and Green’s functions that and stuff I’ve forgotten long ago.  Basically, her program would be able to track an amino acid in a protein and determine how sensitive it is to its environment.

Now, in the pharma industry, this stuff belongs to a type of computation called molecular dynamics.  We start with a 3D representation of a protein in a solvent, usually water.  The model is really just a set of coordinates for each atom in the protein.  There are parameters for each atom that account for bond length between atoms and springiness and intermittant bonds between atoms called hydrogen bonds.   A simulation can be run in several ways.  One of the most common is something called simulated annealing.  That is, heat is added to the system, the protein absorbs this heat and starts to move.  The system is allowed to equilibrate and a trajectory is calculated for a series of time steps.  After the simulation is run, you can concatenate all of the time steps together and run them like a movie.  What you get is something like this (actual simulation starts at about 30 seconds in):

In the animation, the purple and red springs are called α-helices and the yellow ribbons make up a β-sheet.  These are relatively stable secondary structures of the protein that come together to form the tertiary structure of the protein.  They wiggle and shimmy but don’t move much.  The most interesting part of the dynamics run was at the right of the screen where there is a “loop” that has the greatest movement.  It is the action of the loop that is of primary interest to the researcher.  What does it do?  What amino acids around it does it impact and can we tell from its movement what its function is in the protein?

Now, these are all cool and groovy models but there are some things about them that make them tricky.  First, the system is only as good as the parameters you give it.  Some simulations run in cellular membranes that add an additional level of complexity.  If you don’t account for absolutely everything, the result can be waaaaay off.  Second, the simulations suck up a lot of memory and I/O and, up til now, it’s been hard to find systems that will let you do more than a very short span of time.  Luckily for D. E. Shaw, Moore’s law has allowed the area to grow a lot lately.  Also, in the bio end of his business, he has contracted with one of the chip makers for ASICs, or custom made processors, that he is using to build a massive cluster with more than 1000 nodes.  Molecular dynamics has been around for a long time but until Shaw’s latest programs came out, it was a pain in the ass to set up and run more than a few time steps.  His latest molecular dynamics code is called DESMOND and it is commercially available. My site is just beginning to use it, albeit without the fancy cluster.

So, what does all of this have to do with financial markets?  Well, I *guess* you could think of the financial world as a giant protein and if you can figure out what the parameters are and how much heat (money?) you are putting into the system, you could calculate which parts of it are most sensitive to change and then place some bets on that portion of the system.  For instance, you might be able to predict what effect the stimulus package is going to make on the economy and which industries will be most sensitive to that stimulus.  Or, maybe you could calculate the TED spread or LIBOR or a zillion other indicators.   As I said, this is really a job for David Kestenbaum so let’s hope he has a Planet Money segment on it soon.

---