The Need for Anticipatory Analytics to Predict the Future
Like many Americans, I didn’t get much sleep last night. I was glued to my television into the wee hours of the night watching the results of what has been a volatile presidential election. Finally, around 2:30 am Eastern time, it was made official; Donald Trump was named the 45th President of the United States. Regardless of party affiliation, many people did not accurately anticipate this outcome. In fact, some say Trump himself was expecting to lose even hours before the final day of the election. So what happened? Why did so many of us not see this coming?
It turns out we were focused on models that only looked at the past and did not take into account future variables. Moody’s Analytics Presidential Election Model, which has correctly predicted the winner of every U.S. presidential race since Ronald Reagan in 1980, predicted a huge victory for Hillary Clinton. According to its model, Clinton was expected to get 332 electoral votes, while Trump was predicted to get just 206. How did Moody’s get its projection? The Moody’s model is based on various economic and political factors, including looking at recent trends in gas prices, the housing market, fluctuating HHI and President Obama’s approval rating.
Moody’s was solely focused on looking at the past to predict the future. And there’s certainly nothing wrong with that. While predictive analytics has become a fairly common practice, it is not meant to tell you definitively what will happen in the future, particularly if you have unknown variables that may arise. And in this case, the 2016 election is unlike anything we’ve ever seen.
Therefore, predictive analytics can only tell you what might happen given the same set of circumstances – and it is often very good at doing just that. But at the end of the day, predictive analytics are still probabilistic in nature. In dealing with something as substantial as this election, it is important to factor scenarios you may not even know could exist – blue states turning red for the first time in 50 years, anyone?
“It is very important in historically unprecedented times to remember to be humble about the limitations of traditional approaches to predicting the future,” said Dun & Bradstreet’s Chief Data Scientist, Anthony Scriffignano. “Only through advanced methods and careful learning can we respond appropriately with data in such situations.”
What Scriffignano is referring to is the emerging practice of anticipatory analytics. Anticipatory analytics builds on the foundation of predictive analytics, where it can identify and adjust predictions based on inflection points such as the acceleration and deceleration of certain behaviors, or a sudden change in direction. In this case, voters who may have disregarded some of the positive economic and political trends that the projections were based upon; essentially, Moody’s accounted for the data on hand, and not variable human behavior.
Could anticipatory analytics have been used to help determine the winner of the election? That’s hard to say. But it could have looked at different variables aside from what has previously occurred to help draw a conclusion. There’s no doubt elections will endure and candidates, pollsters and the media will continue to try and predict who will win. It will be important for them seamlessly combine data and behavior trends from past risk/opportunity model outputs with current real-time data.
Gaining the ability to foresee and plan for risk and opportunity before they occur is incredibly powerful. In business especially, we can’t be caught off guard because we expect a specific outcome. As more companies focus on analytics to inform growth, it’s imperative to understand all the complexities and potential interactions that could unfold in the future, not just make sense of the past.