I'd assume for machine learning developments, specifically as training data and back testing. Build a new model with 3x the data you had before or be able to re

I'd assume for machine learning developments, specifically as training data and back testing. Build a new model with 3x the data you had before or be able to retrospectively see how a model would have performed over 3 years rather than 1.