A Financial Approach to Machine Learning with Applications to Credit Risk

"Joint work with Sven Sandow"

"Introduction"

Introduction
2 Credit Modeling Problems

Introduction
2 Credit Modeling Problems

Introduction
Our Main Goal

Performance Measures

Performance Measures
Our Paradigm

Performance Measures
Our Paradigm

Performance Measures
Our Paradigm

Performance Measures
Our Paradigm

Performance Measures
Our Paradigm

Performance Measures
Our Paradigm

Performance Measures
    Information Theoretic Interpretation

Performance Measures
    Information Theoretic Interpretation

Performance Measures
    Information Theoretic Interpretation

Performance Measures
    Information Theoretic Interpretation

Performance Measures
    Information Theoretic Interpretation

Performance Measures
    Important Class of Utility Functions

Performance Measures
    Important Class of Utility Functions

Performance Measures
    Important Class of Utility Functions

Performance Measures
    Important Class of Utility Functions

Performance Measures
    Important Class of Utility Functions

Performance Measures:
An Important Class of Utilities

Performance Measures
    pdf(y),prob(Y=y|x),pdf(y|x)

Maximum Expected Utility Models Introduction

Maximum Expected Utility Models Introduction

Maximum Expected Utility Models Formulation

Maximum Expected Utility Models Formulation

Maximum Expected Utility Models Formulation

Maximum Expected Utility Models
Formulation

Maximum Expected Utility Models
Formulation

Maximum Expected Utility Models
Formulation

Maximum Expected Utility Models
Formulation

Maximum Expected Utility Models
Formulation

Maximum Expected Utility Models
Dual Problem

Maximum Expected Utility Models
Dual Problem

Maximum Expected Utility Models
Summary of Approach

Maximum Expected Utility Models
Losing the O’s

Maximum Expected Utility Models
More General Context

Maximum Expected Utility Models Applications and Performance

US Public Firm Model Variables
in order of significance

US Public Firm Model Variables
in order of significance

Performance Measurement
Exploring the Model

Model Use
Surveillance Alerts

Model Use
Surveillance Alerts

Model Use
Surveillance Alerts

Recovery Model
Motivation

Recovery Modeling Approach
Conditional probabilities

Recovery Model
Approach

Recovery Model
Performance

Recovery Model Results:
Probability density versus RGD and collateral

Recovery Model Results:
Probability density versus RGD and collateral

Recovery Model Results:
Point probabilities versus collateral

Model Results:
Probability density versus Spread and Maturity

Model Results:
Probability density versus Spread and Rating

GUI Interface Screen Shots

GUI Interface Screen Shots

Procedure

Plot of 1-year PD transition matrix
conditioned on survival

"Next:"

Screenshot 1

Screenshot 2

Screenshot 3

Screenshot 4

Demo Program Interface

Model Results:
Default Time  Density when all explanatory variables are set to the medium value.

Model Results:
Probability density, given default within 3 yrs versus Net Income/Total Assets

Model Results:
Probability density, given default within 3 yrs versus Total Liabilities/Total Assets

Model Results:
Probability density, given default within 3 yrs versus Relative

Model Results:
Probability density, given default within 3 yrs versus Excess Return

Model Results:
Probability density, given default within 3 yrs versus Volatility

Slide 73