Optimizing portfolios of financial instruments
A decision model is applied to a set of current data from a mortgage portfolio to generate a first portfolio metric. The decision model is formed from a network of a plurality of predictive models that are trained on historical data derived from a plurality of mortgage account profiles for a plurality of mortgages within the mortgage portfolio. One or more of the predictive models is an action-based predictive model allowing modification of at least one attribute affecting performance of the mortgage portfolio. One or more of the attributes can be modified and a second portfolio metric can then be optimized, using the decision model, in light of at least one constraint (which can be user-defined). Related techniques, apparatus, systems, and articles are described.
1. A method for implementation by one or more data processors, the method comprising:
first applying, by at least one data processor, a decision model to a set of current data from a mortgage portfolio to generate a first portfolio metric, the decision model being formed from a network of a plurality of predictive models, the predictive models being trained on historical data derived from a plurality of mortgage account profiles for a plurality of mortgages within the mortgage portfolio, the historical data characterizing actions taken on mortgages in the mortgage portfolio, one of the predictive models being an action-based predictive model allowing modification of at least one attribute affecting performance of the mortgage portfolio;
receiving, by the at least one data processor, data modifying at least one modifiable attribute from an action-based predictive model to form a modified attribute;
second applying, by at least one data processor, the decision model to the set of current data from the mortgage portfolio with the modified attribute to generate a second portfolio metric;
optimizing, by the at least one data processor, a selection of the at least one modifiable attribute from the action-based predictive models to maximize the second portfolio metric taking into account a constraint to the optimization to thereby form an optimized second portfolio metric; and
providing, by the at least one data processor, data characterizing the optimized second portfolio metric;
varying, by the at least one data processor, values of the at least one modifiable attribute;
evaluating a sensitivity of the decision model based on varying values of the at least one modifiable attribute;
testing an effect of varying the values of the at least one modifiable attribute; and
determining an effect on the optimized portfolio based on testing an effect of varying the values of the at least one modifiable attribute.
2. A method as in claim 1 , wherein the data modifying at least one attribute from one of the action-based predictive models is user-generated and obtained via a graphical user interface.
3. A method as in claim 1 , wherein the at least one attribute is derived from the historical data that acts as a decision key.
4. A method as in claim 1 , wherein the constraint is applied by applying, by at least one data processor, a global constraint that applies to substantially all the mortgages in the mortgage portfolio.
5. A method as in claim I, wherein the constraint is applied by applying, by the at least one data processor, a local constraint that applies to a subset of the mortgages in the mortgage portfolio.
6. A method as in claim 1 , further comprising:
assigning, by the at least one data processor, optimal actions to apply to a plurality of mortgages of the mortgage portfolio for carrying out an optimized strategy.
7. A method as in claim 1 , wherein the predictive models comprise global predictive models built on all of the mortgages in the mortgage portfolio.
8. A method as in claim 1 , wherein the predictive models comprise segmented predictive models built on a subset of mortgages in the mortgage portfolio.