S-PLUS Predictive Modeling and Computational Finance
S-PLUS Predictive Modeling and Computational Finance
event with abstracts.
Nov 2004 Finance Event Proceedings for LossCalc II: Dynamic Prediction of LGD.
Greg Gupton, Moody's KMV
We describe LossCalc(tm) version 2.0, the Moody's KMV model to predict
loss given default (LGD). LGD is of natural interest to lenders and
investors wishing to estimate future credit losses. LossCalc is a
robust and validated model of LGD for loans and bonds globally.
LossCalc is a statistical model that incorporates information at all levels:
collateral, instrument, firm, industry, country, and the macroeconomy
to predict LGD. Also, and what may be more interesting than merely
having a powerful predictive model, is to see and understand the
underlying drivers of default recovery/loss that we show.
Predictive Modeling for Property & Casualty Pricing Decisions
Jeremy Stanley, Ernst & Young
This presentation focuses on the application of predictive modeling
methodologies to pricing decisions for property & casualty insurance
lines. Predicting the probability of an insured having one or more
claims in a policy period is a key ingredient to determining the price
a carrier will charge. This presentation will compare and contrast
three types of models applied to this problem: generalized linear
models (GLMs), generalized additive models (GAMs) and neural networks.
GAMs allow for non-linearity in the additive terms and limited types
of specified interactions, requiring an intensive modeling effort to
determine the appropriate model structure. GAMs benefit from fast
model fitting performance, robust measures of in-sample error (such as
the Akaike Information Criterion) and can be easily translated into a
multiplicative rating plan. Neural networks, through the control of
the number of optimization iterations, the size of the hidden layer,
and the use of a weight decay parameter, allow for the near-automatic
selection of model architecture, simultaneously encompassing
interaction terms and complex non-linearities. The predictions of
neural networks are difficult to visualize in high dimensions or with
more than two continuous factors, and are not easily translated into a
multiplicative rating plan. Model performance will be compared in
S-PLUS via cross-validation and bootstrap methods, and visualized with
the use of ROC curves and lift charts. Model structure will be
visualized with S-PLUS Trellis Plots, leading to insights that can
improve the selected model structure.