Financial Machine Learning Week 1 Problem Set — Fat Tails, Overfitting, Credit Default

2 Просмотры

Спасибо! Поделитесь с друзьями!

Вам не понравилось видео. Спасибо за то что поделились своим мнением!

Издатель Sep 10, 2026

Three finance problems from Week 1, each worked all the way through by hand before
a single line of code is written. Every number on screen is one you can reproduce
on paper.

Problem 1 · Fat tails — are daily returns a nice bell curve? Skewness and kurtosis
from five returns, one of them a crash day.
Problem 2 · Overfitting — two models, four training days, one test day. Perfect in
training, nine times worse in testing.
Problem 3 · Credit default — score eight borrowers with a bank's one-line rule, then
count the damage with accuracy, precision and recall.

Each problem closes by scaling the hand calculation up to real data, so the numbers
you computed and the code you write afterwards agree exactly.

Unit 1 · ML and the Finance Problem — Part I, Foundations.

── COURSE MATERIALS ──────────────────────────
All problem sets and their data ship together — no download step, no vendor account:
https://keerhee.github.io/financial-machine-learning/

git clone https://github.com/keerhee/financial-machine-learning.git

── CHAPTERS ──────────────────────────────────
0:00 Three hands-on problems
0:40 Every problem follows the same four steps
1:34 Where the data comes from
2:23 Problem 1 · Fat tails
2:41 One crash day in five
3:34 The tools: third and fourth moments
4:33 The moment table — every number by hand
5:39 Verdict: left-skewed, slightly fat-tailed
6:41 Scale it: from five returns to two years
7:44 Problem 2 · Overfitting
8:03 Two models, four training days, one test day
8:55 The tools: MSE and its decomposition
9:48 Two small tables settle it
10:58 Perfect in training, nine times worse in testing
12:02 Scale it: the full U-shaped test curve
12:56 Problem 3 · Credit default
13:14 A bank's one-line scoring rule
14:15 The tools: a linear score and three metrics
15:06 Scoring all eight applicants
16:27 Counting the damage: 75% across the board
17:33 Scale it: from eight rows to ten thousand
18:36 Three things your hands now know

── THE WEEK ──────────────────────────────────
1. Primer — the vocabulary the lecture assumes
2. Lecture — the world of ML and the finance problem
3. Problem Set (this video)

A full-semester course in financial machine learning for third- and fourth-year
undergraduates at Yonsei University, taught in English. Built on Hull's Machine
Learning in Business, López de Prado's Advances in Financial Machine Learning, and
the causal factor investing literature.

Course materials CC BY-NC-SA 4.0.