This is a guest post by Brent M. Eastwood, PhD. Brent is a former military officer, political scientist, and economic forecaster. He also has experience leading tech start-ups in sectors such as immersive video and biometrics. He is the Founder of GovBrain, a firm in Washington, DC that uses large amounts of government information from around the world to predict financial market moves. You can follow the firm on Twitter @GovBrain.
I was very excited to find out about Data Science DC and the Forecasting International Events meetup. Prof. Naren Ramakrishnan and Dr. Jay Ulfelder are known for their pioneering forecasting work. Their presentation will touch on how to tackle the problem of noisy data sets and to forecast major international events with statistical and machine learning tools.
At GovBrain, we utilize predictive analytics and machine learning techniques, so that our financial services clients can achieve a unique, profitable, and legal trading “edge.” The GovBrain patent-pending system of web applications searches federal, state, local, and international government databases, including political and financial news sources, to provide real-time insights for Wall Street. GovBrain links this information to individual stocks, bonds, commodities, or currencies and processes it through an artificial intelligence/machine learning engine that predicts the price change for each stock or exchange traded fund.
I like to say we are a “little data” company. Little data is when small firms or individuals produce and analyze their own unique and proprietary data at a lower order of magnitude than large organizations.
An example would be one of our products in development that we call our Doomsday App, with the official and unwieldy name of “Global Asset Bubble and Financial Crisis Prediction Application.”
We’re still forging the user experience, but basically it breaks down big government data into digestible snippets or events. Then it runs the event through a prediction engine. That aggregated output gets a wash through R Studio for analysis and visualization. (I can’t say enough about how much I love R Studio). Users will be able to plot financial contagion or bubbles on Google Earth or other mapping API in real time. We think it’s more forward-looking than similar products that rely on dated or backward-looking economic indicators.
At this point you should be thinking:
a) I’m skeptical of GovBrain and their voodoo claims.
b) If Eastwood ever publishes an academic article on this, I’m going to replicate and find flaws in his methodology.
And that is exactly what I want you to do if I can get this published some day in an academic journal once my patent attorney and board of directors allow it. If all goes bad with GovBrain, at least I can share our specific research methods with the data science community someday.
As theorists, we want to give others a chance to prove us wrong. I think our methodology on this may be parsimonious and adhere to Occam’s Razor, but others will probably disagree with some of our assumptions, rules, results, or conclusions. They may even say the whole thing is trivial.
I think this discourse and critical analysis among peers in data science is very important, especially when it comes to model building and testing theory and hypotheses. Now that may be difficult if you are working on a commercial endeavor. Your boss or CEO probably did not hire you to produce scholarship, but you can still try to get a peer review in other ways.
At GovBrain, we wanted to get “prove us wrong” feedback from not only PhDs but also recent grads. We talked to a young computer scientist whose father helped develop the R Project and a newly-minted data scientist who had a technical business major with plenty of instruction through the Johns Hopkins data instructional track from Coursera.
They had the right amount of skepticism, but were interested in learning more, so we encouraged additional tough–love critique. Many of you are already doing this type of peer review, but start-ups or big data companies may not be. I hope we can continue this knowledge discovery with the appropriate “prove me wrong” approach throughout the data science community, so we can all produce the best commercial product or scholarship possible.
Forecasting International Events will be held at George Washington University’s Elliot School of International Affairs on April 30th at 6:30p.