Machine Learning

Question 1

What is Artificial Intelligence (AI)?

Answer 1

AI involves machines perceiving their environment and taking actions to achieve human-set goals.

Question 2

Give examples of AI applications mentioned in the lecture.

Answer 2

Facebook’s Deepface, Uber’s self-driving cars, Apple Siri, and Google Translate.

Question 3

What distinguishes machine learning from traditional econometrics?

Answer 3

ML focuses on outcome prediction, while econometrics emphasizes parameter estimation and hypothesis testing.

Question 4

Name three Machine Learning methods mentioned.

Answer 4

Ridge regression, LASSO, and decision trees.

Question 5

What is the bias in Machine Learning models?

Answer 5

The average error between the true function and the model’s prediction.

Question 6

What does the variance measure in Machine Learning models?

Answer 6

The variability of the model’s predictions across different datasets.

Question 7

How does model complexity affect bias and variance?

Answer 7

High complexity reduces bias but increases variance, leading to overfitting.

Question 8

What is overfitting?

Answer 8

When a model fits the training data well but performs poorly on new, unseen data.

Question 9

What is R-Squared, and how is it calculated?

Answer 9

R-Squared measures the goodness-of-fit and is calculated as 1 - RSS/TSS.

Question 10

Why is adjusted R-Squared preferred over R-Squared?

Answer 10

Adjusted R-Squared accounts for the number of predictors, penalizing overly complex models.

Question 11

What are AIC and BIC?

Answer 11

Model selection criteria that penalize model complexity. BIC imposes a stronger penalty.

Question 12

What do machine learning models penalize besides complexity?

Answer 12

Large coefficient estimates to reduce out-of-sample prediction variance.

Question 13

What does OLS minimize?

Answer 13

The sum of squared residuals.

Question 14

How does ridge regression modify OLS?

Answer 14

It adds a penalty proportional to the sum of squared coefficients.

Question 15

How does the LASSO estimator differ from ridge regression?

Answer 15

LASSO uses the absolute values of coefficients as a penalty, which can shrink some coefficients to zero.

Question 16

What is the penalty parameter Lambda in machine learning?

Answer 16

A hyperparameter that controls the tradeoff between model fit and complexity.

Question 17

Name one method to choose Lambda.

Answer 17

Cross-validation.

Question 18

What is cross-validation?

Answer 18

A method to assess model performance by dividing the dataset into training and validation subsets.

Question 19

What are the steps in K-fold cross-validation?

Answer 19

Split the dataset into K subsets, train on K - 1 folds, test on the hold-out fold, and repeat.

Question 20

Why is cross-validation used?

Answer 20

To prevent overfitting and select optimal model parameters.

Question 21

What is a decision tree in machine learning?

Answer 21

A non-linear model that splits data into subsets based on predictor values.

Question 22

How does a decision tree make predictions?

Answer 22

By averaging outcomes within each terminal node.

Question 23

What is recursive partitioning?

Answer 23

Splitting the data repeatedly to form a tree structure.

Question 24

How is the depth of a decision tree selected?

Answer 24

Using a hold-out sample or cross validation.

Question 25

Name two financial applications of machine learning.

Answer 25

Predicting borrower default probability and selecting assets for investment.

Question 26

What is an example dataset for machine learning in housing markets?

Answer 26

The American Housing Survey with 150 variables on house and location characteristics.

Question 27

How do machine learning models compare to OLS in out-of-sample performance?

Answer 27

Machine learning models often outperform OLS due to better handling of complex relationships.

Question 28

What is the focus of OLS?

Answer 28

Estimation of parameters with desirable statistical properties.

Question 29

What is the focus of machine learning?

Answer 29

Accurate prediction of outcomes, often at the expense of interpretability.

Question 30

Can machine learning be used for causal inference?

Answer 30

No, machine learning is designed for prediction, not establishing causation.

Question 31

How does LASSO handle irrelevant predictors?

Answer 31

By shrinking their coefficients to zero, effectively excluding them from the model.

Question 32

What is the main difference between ridge regression and LASSO?

Answer 32

Ridge regression includes all predictors but shrinks coefficients, while LASSO can exclude predictors.

Question 33

Why is machine learning increasingly important in finance?

Answer 33

For applications like trading algorithms, asset management, and credit risk assessment.

Question 34

What advantage does machine learning provide in predictive modeling?

Answer 34

It can handle high-dimensional data and complex relationships better than traditional methods.

Question 35

Why do we still need traditional econometric methods?

Answer 35

To estimate causal relationships and understand underlying mechanisms.