Questions / Quant research, Machine learning judgment
Which point forecast does the loss prefer?
A conditional response takes values 1, 2 and 10 with equal probability. Which point prediction minimizes expected absolute error?
Answer it in the app It's free. You get hints, a second try if you miss, and a follow-up question.
Show the answer and worked solution
Answer: 2 response units
- Absolute error is minimized by the median.
- The median of {1, 2, 10} is 2.
- The mean, 13/3, minimizes squared error instead.
Interview shortcut. Absolute error → median of {1, 2, 10}.
Takeaway. Absolute loss targets the median; squared loss the mean.
In the app this has a follow-up where one thing about the setup changes. Try the follow-up.