Ronald Fisher

In 1922 Ronald Fisher introduced maximum likelihood: given the data we saw, choose the parameter value that explains it best. Probability describes the data; likelihood scores the hypothesis.

For a volatility forecast f, the Gaussian likelihood of tomorrow’s return leads directly to the QLIKE loss, QLIKE(y, f) = y/f − ln(y/f) − 1, used to train and compare GARCH and AI volatility forecasts.

Explore next

• Maximum likelihood: which model explains the clues?
• Learning from forecasts in markets (ML016)
• Volatility: the human-scale measure of risk (MF007)

Stories, illustrations, and interactive mini lessons. Ask a question.