Research ideas lurking in the classroom

I heard the idea of conformal forecasting from a classmate before it was published.

In 2018, I was an undergraduate student at Monash studying actuarial science. I took Rob’s third-year course on forecasting. In one of the lectures, we were talking about forecast intervals. Rob made a comment that forecast intervals are usually too narrow because they do not consider the uncertainty due to model misspecification. One classmate asked a question – I remember him being quite chatty and always asking questions - is it possible to adjust the forecast interval based on simulations or test sets to find the right coverage? Rob’s response was that it was possible, but no one had done that.

When I was listening to Xiaoqian’s talk on conformal forecasting today at ISF2026, it suddenly dawned on me: wasn’t that idea proposed in the classroom eight years ago, essentially the core idea of conformal prediction?

That was 2018, just a few months before the first time series conformal forecasting paper (Chernozhukov, Kaspar and Zhu, 2018) was published.

References

Chernozhukov, Victor, Kaspar Wüthrich, and Zhu Yinchu. “Exact and robust conformal inference methods for predictive machine learning with dependent data.” In Conference On learning theory, pp. 732-749. PMLR, 2018.