Loan Default Prediction - Aster Analytics

Teradata AsterĀ® Analytics Foundation User GuideUpdate 2

Product
Aster Analytics
Release Number
7.00.02
Published
September 2017
Language
English (United States)
Last Update
2018-04-17
dita:mapPath
uce1497542673292.ditamap
dita:ditavalPath
AA-notempfilter_pdf_output.ditaval
dita:id
B700-1022
lifecycle
previous
Product Category
Software

Financial institutions try to predict defaulting customers by analyzing previous transaction data. To reduce the number of defaulting customers, the financial institution may suggest default prevention initiatives to customers.

Loan status updates depend on customer payments. You can use loan status updates to build a Hidden Markov Model (HMM) to predict loan defaults. Assume that the statuses of the loans are the following:
  • Current
  • Late
  • One month late
  • Two months late
  • Three months late
  • Four months late
  • Defaulted
  • Paid

Use all transaction sequences ending with Paid to train one model, and all sequences ending with Defaulted to train a second model. Evaluate the current sequence of statuses for a customer against both models. Base the default prediction on the model that gives a higher probability.

The hidden states of the HMM inherently indicate the financial health of the customer. A hidden state with a high probability of emission of defaulted status indicates poor financial health, while a hidden state with high probability of emission of paid status indicates good financial health.

The observation symbols to build the HMM are the statuses referenced by number in the order shown in the following table.

HMMUnsupervisedLearner Example Observation Symbols
status symbols
current 1
late 2
one month late 3
two months late 4
three months late 5
four months late 6
defaulted 7
paid 8