Skip to content

CFA Level I · CFA Level I Exam · Introduction to Financial Data Science

A team builds a model to forecast bond returns using 200 features and only 150 observations. To reduce overfitting, the team adds a penalty that shrinks the coefficients of weak features, with some coefficients set exactly to zero. Compared with an unpenalized regression, this approach will most likely:

The penalty will most likely increase bias slightly while reducing variance, which usually improves out-of-sample performance. Shrinking or zeroing weak coefficients simplifies the model, which suits many features and few observations. It cannot remove both error sources or cut bias while leaving variance unchanged.

  1. Aincrease bias slightly while reducing variance, improving out-of-sample performanceCorrect
  2. Beliminate both bias and variance by removing all irrelevant features
  3. Creduce bias substantially while leaving variance unchanged

Explanation

Penalized regression such as LASSO shrinks coefficients and sets some to zero, adding a small amount of bias but lowering variance and complexity. This tradeoff typically improves out-of-sample fit when features are numerous relative to observations. It cannot eliminate both bias and variance, and it does not lower bias while leaving variance unchanged.

Did you get it right without looking?

One question tells you little. A timed set on Introduction to Financial Data Science shows your real accuracy, how long you take and where you lose marks.

Practise Introduction to Financial Data ScienceFree diagnostic mock

More Introduction to Financial Data Science questions