Wall Street is no stranger to computer models, which have been used for more than a decade by fund managers and traders to beat the market. But many of the models developed by the so-called “quants” have proved insufficiently prescient—they didn’t do a very good job, for example, of predicting the subprime mortgage crisis. Using computer modeling software initially developed to help pharmaceutical firms find biological targets for new drugs, Cambridge, MA-based Fina Technologies is hoping to recharge the troubled financial sector.
Fina has licensed its so-called “reverse engineering/forward simulation” (REFS) technology from Cambridge’s Gene Network Sciences, which has been developing and applying REFS for drug companies for several years. Under the REFS approach, software analyzes massive amounts of historical data to divine causal links, then makes forward projections based on the connections it’s discovered.
Fina boosted its prospects last month by closing a $4.5 million first-round financing from a pool of investors led by Reed Elsevier Ventures, the venture arm of the media and publishing conglomerate Reed Elsevier. Josh Holden, the CEO of Fina (and an early angel investor in Gene Network Sciences), talked to me recently about how the startup plans to use the cash to fund efforts to promote the REFS technology in the financial world.
Investment computer models—which are basically algorithms based on investment hypotheses—have had varied success. Holden, an MIT-trained engineer with more than a decade of experience in the investment world at Deutsche Bank and other financial institutions, says that a chronic problem is trying to jam too many variables or parameters into these algorithms. Fina aims to overcome this over-modeling problem with the REFS platform, which has shown that it can handle the massive amounts of data about biochemical signaling pathways in the human body needed to make predictions about how various drug candidates will affect the system.
“What we’re trying to do, at the most fundamental [level], is to automate the scientific method,” Holden says, “which is basically to propose a hypothesis, test whether that hypothesis is true in the presence of experimental data, [and] compare it to other hypotheses out there all with an eye toward controlling for over-fitting and complexity.”
The firm’s technology integrates input from multiple models before making predictions about the market, Holden explains, rather than using one algorithm overloaded with parameters. The upshot is that traders at hedge funds could predict changes in the market 5 minutes to 30 minutes before they happen, then buy or sell ahead of time to capitalize on the upward or downward shift in price, he says.
The startup has built sample models with the platform that make Holden confident that the technology makes the correct predictions about 55 to
