Deeploans

Early Detection Module

By automating periodic reviews of retail exposures, this module enables the early detection of possible defaults and pre-payments up to 24 months in advance.

State transition model inventories for all European countries are available and ready to be deployed.

  • bridge-browser-top
  • bridge-browser-top
  • bridge-browser-top
  • bridge-browser-top
  • bridge-browser-top
    Phone

    Early Detection Module is powered by a model inventory based on neural network paradigms. These models are non-parametric extensions of the familiar logistic regression model, which is widely used by practitioners, and are originally conceived to overcome the explainability issues.

    +30 validated state transition models for various credit products and jurisdictions

    Models are validated using +100 million performing and non-performing loans periodically reported since 2013

    Each loan is described by more than 500 standard variables

    12-month forecast accuracy

    DeepLoans Deep Learning Models vs Standard Logistic Regression Models*

    *(trained as per Sirignano et al)

     

    This is the improvement we observed by credit product:

    16%

    Residential

    Residential mortgages, including different type of products

    21%

    Auto

    Auto loans, both used and new cars

    24%

    Consumer

    Consumer credit loans (Cessione del Quinto included)

    21%

    Credit Card

    Credit Card loans

    20%

    SME

    SME loans (secured and unsecured)