Investment Firm Tackles the Reproducibility Crisis

Investment Firm Tackles the Reproducibility Crisis

Live platform for prediction model analysis makes so-called p-hacking impossible.

Overfitting is rewarded in academia, but not on the microprediction platform”

— Peter Cotton

WEST PALM BEACH, FLORIDA, UNITED STATES, January 31, 2023 / — Microprediction LLC has announced that prizes for algorithm creators at Microprediction.Org will be doubled in the month of February.

The novel platform, sponsored by Intech Investments, matches predictive algorithms authored in Python, R and Julia against live streams of data. The platform is open to anyone needing live prediction of any measured quantity of interest to them, such as electricity usage, volatility of a stock, or even emoji counts.

Conversely, the platform allows any researcher or student to “recycle” a predictive model that might not have worked well on the original problem of interest. Predictive algorithms are encapsulated in so-called “crawlers” that travel from one data stream to the next looking for places where they have a competitive advantage.

The alternatives to this live battle include the academic empirical literature, which has come under heavy fire in recent years. Findings about the efficacy of some types of predictive models that have been published have not been replicated. Models which performed well in the past have failed on a forward basis, leading to accusations of “data snooping”, “data dredging” and overfitting on backtests.

Applications of the approach are discussed in a recent book published by MIT Press titled Microprediction:Building An Open AI Network.

Peter Cotton
Microprediction LLC
+1 646-637-1338
email us here
Visit us on social media:

Investment Firm Tackles the Reproducibility Crisis

Article originally published on as Investment Firm Tackles the Reproducibility Crisis

This story originally appeared at Tech - Social Gov