Alternative data might be the key to unlocking alpha but IBM CTO Tom Eck sees three challenges that must be overcome.
Dustin Blitchok filed this report for Benzinga:
Model bias, data lineage and governance and model explainability are three problems that come with alternative data, said Tom Eck, chief technology officer at IBM Watson.
Giving the example of models that predict credit risk for loans, Eck said they can be biased toward or against certain demographics. “It’s a side effect of the data,” he said.
A population can be underrepresented in data because they don’t apply for a lot of loans — and models can interpret this as a population that’s rarely approved for a loan, Eck said. “But the model doesn’t know any differently.”
Data lineage must be tracked just like source code, the CTO said. And model explainability “is a really tough problem that the industry as a whole is trying to deal with right now.”