Credit scoring, response modelling and insurance rating a practical guide to forecasting consumer behaviour Steven Finlay.

List of Tables p. xii
List of Figures p. xiii
Acknowledgements p. xiv
Introduction p. 1
Scope and content p. 3
Model applications p. 5
The nature and form of consumer behaviour models p. 8
Linear models p. 9
Classification and regression trees (CART) p. 12
Artificial neural networks p. 14
Model construction p. 18
Measures of performance p. 20
The stages of a model development project p. 22
Chapter summary p. 28
Project Planning p. 30
Roles and responsibilities p. 32
Business objectives and project scope p. 34
Project scope p. 36
Cheap, quick or optimal? p. 38
Modelling objectives p. 39
Modelling objectives for classification models p. 40
Roll rate analysis p. 43
Profit based good/bad definitions p. 45
Continuous modelling objectives p. 46
Product level or customer level forecasting? p. 48
Forecast horizon (outcome period) p. 50
Bad rate (emergence) curves p. 52
Revenue/loss/value curves p. 53
Legal and ethical issues p. 54
Data sources and predictor variables p. 55
Resource planning p. 58
Costs p. 59
Project plan p. 59
Risks and issues p. 61
Documentation and reporting p. 62
Project requirements document p. 62
Interim documentation p. 63
Final project documentation (documentation manual) p. 64
Chapter summary p. 64
Sample Selection p. 66
Sample window (sample period) p. 66
Sample size p. 68
Stratified random sampling p. 70
Adaptive sampling p. 71
Development and holdout samples p. 73
Out-of-time and recent samples p. 73
Multi-segment (sub-population) sampling p. 75
Balancing p. 77
Non-performance p. 83
Exclusions p. 83
Population flow (waterfall) diagram p. 85
Chapter summary p. 87
Gathering and Preparing Data p. 89
Gathering data p. 90
Mismatches p. 95
Sample first or gather first? p. 97
Basic data checks p. 98
Cleaning and preparing data p. 102
Dealing with missing, corrupt and invalid data p. 102
Creating derived variables p. 104
Outliers p. 106
Inconsistent coding schema p. 106
Coding of the dependent variable (modelling objective) p. 107
The final data set p. 108
Familiarization with the data p. 110
Chapter summary p. 111
Understanding Relationships in Data p. 113
Fine classed univariate (characteristic) analysis p. 114
Measures of association p. 123
Information value p. 123
Chi-squared statistic p. 125
Efficiency (GINI coefficient) p. 125
Correlation p. 126
Alternative methods for classing interval variables p. 129
Automated segmentation procedures p. 129
The application of expert opinion to interval definitions p. 129
Correlation between predictor variables p. 131
Interaction variables p. 134
Preliminary variable selection p. 138
Chapter summary p. 142
Data Pre-processing p. 144
Dummy variable transformed variables p. 145
Weights of evidence transformed variables p. 146
Coarse classing p. 146
Coarse classing categorical variables p. 148
Coarse classing ordinal and interval variables p. 150
How many coarse classed intervals should there be? p. 153
Balancing issues p. 154
Pre-processing holdout, out-of-time and recent samples p. 154
Which is best - weight of evidence or dummy variables? p. 155
Linear models p. 155
CART and neural network models p. 158
Chapter summary p. 159
Model Construction (Parameter Estimation) p. 160
Linear regression p. 162
Linear regression for regression p. 162
Linear regression for classification p. 164
Stepwise linear regression p. 165
Model generation p. 166
Interpreting the output of the modelling process p. 168
Measures of model fit p. 170
Are the assumptions for linear regression important? p. 173
Stakeholder expectations and business requirements p. 174
Logistic regression p. 175
Producing the model p. 177
Interpreting the output p. 177
Neural network models p. 180
Number of neurons in the hidden layer p. 181
Objective function p. 182
Combination and activation function p. 182
Training algorithm p. 183
Stopping criteria and model selection p. 184
Classification and regression trees (CART) p. 184
Growing and pruning the tree p. 185
Survival analysis p. 186
Computation issues p. 188
Calibration p. 190
Presenting linear models as scorecards p. 192
The prospects of further advances in model construction techniques p. 193
Chapter summary p. 195
Validation, Model Performance and Cut-off Strategy p. 197
Preparing for validation p. 198
Preliminary validation p. 201
Comparison of development and holdout samples p. 202
Score alignment p. 203
Attribute alignment p. 207
What if a model fails to validate? p. 210
Generic measures of performance p. 211
Percentage correctly classified (PCC) p. 212
ROC curves and the GINI coefficient p. 213
KS-statistic p. 216
Out-of time sample validation p. 216
Business measures of performance p. 218
Marginal odds based cut-off with average revenue/loss figures p. 219
Constraint based cut-offs p. 220
What-if analysis p. 221
Swap set analysis p. 222
Presenting models to stakeholders p. 223
Chapter summary p. 224
Sample Bias and Reject Inference p. 226
Data methods p. 231
Reject acceptance p. 231
Data surrogacy p. 232
Inference methods p. 235
Augmentation p. 235
Extrapolation p. 236
Iterative reclassification p. 240
Does reject inference work? p. 241
Chapter summary p. 242
Implementation and Monitoring p. 244
Implementation p. 245
Implementation platform p. 245
Scoring (coding) instructions p. 246
Test plan p. 247
Post implementation checks p. 248
Monitoring p. 248
Model performance p. 249
Policy rules and override analysis p. 250
Monitoring cases that would previously have been rejected p. 252
Portfolio monitoring p. 252
Chapter summary p. 253
Further Topics p. 254
Model development and evaluation with small samples p. 254
Leave-one-out cross validation p. 255
Bootstrapping p. 255
Multi-sample evaluation procedures for large populations p. 256
k-fold cross validation p. 256
kj-fold cross validation p. 257
Multi-model (fusion) systems p. 257
Static parallel systems p. 258
Multi-stage models p. 259
Dynamic model selection p. 262
Chapter summary p. 262
Notes p. 264
Bibliography p. 273
Index p. 278

Within the financial services industry today, most decisions on how to deal with consumers are made automatically by computerized decision making systems. At the heart of these systems lie mathematically derived forecasting models. These use information about people and their past behavior, to predict how people are likely to behave in the future. For example, who is likely to repay a loan, who will respond to a mail shot and the likelihood that someone will claim on their household insurance policy. Decisions about how to treat people are then made on the basis of the predictions calculated by the system.

This book provides a step-by-step guide to how the forecasting models used by the worlds leading financial institutions are developed and deployed. It covers all stages involved in the construction of such a model, including project management, data collection, sampling, data pre-processing, model construction, validation, implementation and post-implementation monitoring of the model's performance.

Includes bibliographical references and index.