March 2019 is the current target date for the next US recession, says a machine-learning “forecasting engine” developed by San Diego-based Intensity Corporation. Intensity boasts a number of very large tech firms as clients—Apple, IBM, Microsoft, and others—and is itself comprised by a team of data scientists, statisticians, and econometrically-minded PhDs.
In a recent interview with Financial Sense Insider, two of Intensity’s Vice Presidents, Ray Bamford and Irina Telyukova, discussed their real-time forecasting model, how it compares to professional human forecasters, and how, as you’ll hear in the following clip, AI is being applied to forecasting and large-scale investing.
Q2 Hedge Funds Portable Database Now LIVE!!! Letters, Conferences, Slides And More [UPDATED 7/13 14:48 EST]
Simply click the menu below to perform sorting functions. This page was just created on 7/1/2020 we will be updating it on a very frequent basis over the next three months (usually at LEAST daily), please come back or bookmark the page. As always we REALLY really appreciate legal letters and tips on hedge funds Read More
“Our forecasting platform, broadly defined … is an engine that relies on continual model updating,” Telyukova said, as well as continual “model selection and model combination.”
The engine is designed to deliver forecasts that are based on large sets of data of different kinds and be responsive to the economy and market conditions in real time, she added.
Basically, the company’s engine takes real-time data, various selected models, and combines these to figure out the most likely outcome in a given situation. The business cycle forecast engine is a combination of multiple models and it produces a forecast that goes out to 5 years, Telyukova stated.
“We create what we call ‘alpha,’ or both attractive and uncorrelated return,” Bamford said. “We’ve applied this to economic forecasting, and that’s where our recession timing forecasts come in.”
As of May 11th, Intensity’s forecast, which is publicly available on their website, was signaling March 2019 as the expected start date for a US recession.
But this isn’t a fixed date. The forecast will change based on changing economic conditions (see here for their most current forecast).
“The way to think about this is, our forecasting model is relying on the best information that we have as of the day that we’re making this forecast,” Telyukova said.
For example, Donald Trump’s election was an important economic event that moved the model forecast, she noted. The same was true for the Brexit vote last year.
As of November 6, 2016, the model was forecasting a start to the next recession in December 2017, Bamford noted. By November 20, 2016, the model had pushed that date back to November 2018. Now, it sits at March 2019.
Ultimately, the company seeks to combine human intelligence with machine intelligence. It leverages academic research, formulates hypotheses based on a theoretical foundation, and then validates those hypotheses using advanced statistical methods, Bamford stated.
The company also focuses on checking its model’s accuracy.
“We do very extensive testing of our models, and in particular we focus on what is known as ‘out of sample’ performance,” Telyukova said.
This is important, particularly with machine learning, she noted, because machine learning can be subject to pitfalls such as overfitting, where variables appear to be highly predictive but end up as noise.
There is a lot of human intervention involved, and some of these processes are less automated, Telyukova stated.
The company has applied its engine to a wide variety of problems, and is generating results for Fortune 500 companies, e-commerce websites, professional sports teams, and a number of different firms in the financial and investment space, Bamford stated.
The goal is to leverage data and a strong understanding of economics to produce predictive models that improve performance in each of these domains, Bamford noted.
“We’re at a really interesting time in history,” he said. “The combination of the drop in computing costs, improvements in computing performance, the pervasiveness of data and improvements in modeling techniques … generate some really profound results.”
Listen to this full interview with Ray Bamford and Irina Telyukova on our website by logging in and clicking here. Become a subscriber and gain full access to our premium weekday interviews with leading guest experts by clicking here.
Article by Financial Sense