Forecasting Congressional Votes Could Yield Juicy Returns

STOCK traders hang on central bankers’ every utterance. They scan news sites for market-moving events, such as terrorist attacks, and monitor President Donald Trump’s tweets for hostility towards publicly traded firms. Curiously, though, few analyse goings-on in Congress, which can shift the course of the world’s largest economy. Jonathan Strong, a former reporter (including at Roll Call, a sister publication of The Economist), hopes to change that.

With the help of Øptimus, a firm of Republican data wonks, he has spent three years building Legis, an algorithm powered by vast quantities of data and a neural network (a computer system modelled on the human brain), which predicts the outcome of congressional votes. Each of the 44 votes it has forecast so far has been correct. Last year a hedge fund (which does not want to be named) began trading derivatives using its predictions.

According to legend, carrier pigeons brought news of the Duke of Wellington’s triumph at Waterloo to Nathan Rothschild in London. He promptly invested in British bonds and earned a handsome reward. On the day of the Brexit referendum in 2016, hedge funds commissioned private exit polls so they could predict which way sterling would move once the result was known. Lobbying firms in Washington earn large sums for political insights gleaned from personal connections. But most information today flows too freely for mere speed to give an investor an edge. And not everyone has access to inside information. Predictive-analytics firms such as Legis seek to gain insights, instead, by finding signals within the noise.

Congress is certainly noisy. It is a source of information not on a single subject, but on hundreds, each with their own quirks, donor bases and pet causes. Some representatives toe the party line, others are mavericks. As elections approach, political alliances are redrawn. Much of Mr. Strong’s years of toil involved collecting and cleaning 100m publicly available data points, including an “ideology score” that places each politician on the liberal-conservative spectrum; detailed voting histories; and characteristics of each congressional member’s district, such as major industries. These are crunched, along with live news, by a neural network christened The Underwood, in homage to the devious anti-hero of “House of Cards”, a popular political drama series.

Accurately predicting the outcome of votes can be useful, even for some time after they have been cast. Congressional leaders find it hard enough to work out how many votes they have in the hurly-burly of a legislative session. For journalists on deadline, it can be impossible. With breaking news, outcomes are not always immediately clear.

In March, just before two failed attempts at repealing Obamacare in the House of Representatives, Legis had assigned just a 7% probability to success. That allowed its hedge-fund customer to bet profitably on the S&P 500 and Tenet Healthcare, a hospital firm whose share price seemed especially correlated with the bill’s prospects. Another well-placed bet, on the S&P and Russell 2000 indices, returned 35% in 50 minutes. That was before a critical vote on tax reform, which Mr Strong gave an 84% chance of success.

Building a long-term business out of such successes will not be easy. Only a few congressional votes each year, on issues such as Puerto Rican debt and health-care reform, obviously present trading opportunities. But with ingenuity, others could, too. In January, after Betsy DeVos, a proponent of school choice, won a contentious confirmation fight to become secretary of education, shares of K12 Inc, which helps manage virtual charter schools, jumped.