Main Content

Cox

Create Cox model object for lifetime probability of default

Since R2021b

Description

Create and analyze a Cox model object to calculate lifetime probability of default (PD) using this workflow:

  1. Use fitLifetimePDModel to create a Cox model object.

  2. Optionally, use discardResiduals to remove residual information from the Cox model object.

  3. Use predict to predict the conditional PD and predictLifetime to predict the lifetime PD.

  4. Use modelDiscrimination to return AUROC and ROC data. You can plot the results using modelDiscriminationPlot.

  5. Use modelCalibration to return the root mean square error (RMSE) of observed and predicted PD data. You can plot the results using modelCalibrationPlot.

Creation

Description

example

CoxPDModel = fitLifetimePDModel(data,ModelType,AgeVar=agevar_value) creates a Cox PD model object.

If you do not specify variable information for IDVar, LoanVars, MacroVars, and ResponseVar, then:

  • IDVar is set to the first column in the data input.

  • LoanVars is set to include all columns from the second to the second-to-last columns of the data input.

  • ResponseVar is set to the last column in the data input.

example

CoxPDModel = fitLifetimePDModel(___,Name=Value) sets optional properties using additional name-value arguments in addition to the required arguments in the previous syntax. For example, CoxPDModel = fitLifetimePDModel(data(TrainDataInd,:),"Cox",ModelID="Cox_A",Descripion="Cox_model",AgeVar="YOB",IDVar="ID",LoanVars="ScoreGroup",MacroVars={'GDP','Market'},ResponseVar="Default",TimeInterval=1,TieBreakMethod='Efron',WeightsVar="Weights") creates a CoxPDModel using a Cox model type. You can specify multiple name-value arguments.

Input Arguments

expand all

Data, specified as a table, in panel data form. The data must contain an ID column and an Age column. The response variable must be a binary variable with the value 0 or 1, with 1 indicating default.

Data Types: table

Model type, specified as a string with the value "Cox" or a character vector with the value 'Cox'.

Data Types: char | string

Name-Value Arguments

Specify required and optional pairs of arguments as Name1=Value1,...,NameN=ValueN, where Name is the argument name and Value is the corresponding value. Name-value arguments must appear after other arguments, but the order of the pairs does not matter.

Example: CoxPDModel = fitLifetimePDModel(data(TrainDataInd,:),"Cox",ModelID="Cox_A",Descripion="Cox_model",AgeVar="YOB",IDVar="ID",LoanVars="ScoreGroup",MacroVars={'GDP','Market'},ResponseVar="Default",TimeInterval=1,WeightsVar="Weights")

Required Cox Name-Value Argument

expand all

Age variable indicating which column in data contains the loan age information, specified as AgeVar and a string or character vector.

Note

The required name-value argument AgeVar is not treated as a predictor in the Cox lifetime PD model. When using a Cox model, you must specify predictor variables using LoanVars or MacroVars. The AgeVar values are the event times for the underlying Cox proportional hazards model.

AgeVar values for each ID should be increasing. If there are nonpositive age increments, fitLifetimePDModel warns when you create a Cox model and removes the IDs with nonpositive age increments. By default, the TimeInterval value is set to the most common age increment in the training data.

Data Types: string | char

Optional Cox Name-Value Arguments

expand all

User-defined model ID, specified as ModelID and a string or character vector. The software uses the ModelID to format outputs and is expected to be short.

Data Types: string | char

User-defined description for model, specified as Description and a string or character vector.

Data Types: string | char

ID variable indicating which column in data contains the loan or borrower ID, specified as IDVar and a string or character vector.

Data Types: string | char

Loan variables indicating which column in data contains the loan-specific information, such as origination score or loan-to-value ratio, specified as LoanVars and a string array or cell array of character vectors.

Data Types: string | cell

Macro variables indicating which column in data contains the macroeconomic information, such as gross domestic product (GDP) growth or unemployment rate, specified as MacroVars and a string array or cell array of character vectors.

Data Types: string | cell

Variable indicating which column in data contains the response variable, specified as ResponseVar and a logical value.

Note

The response variable values in the data must be a binary variable with 0 or 1 values, with 1 indicating default.

In Cox lifetime PD models, the ResponseVar values define the censoring information for the underlying Cox proportional hazards model.

Data Types: string | char

Variable indicating which column in data contains the observation weights, specified as WeightsVar and a string array.

Note

The default value ("") results in a weight of 1 for each row in the data.

For an example using WeightsVar, see Create Weighted Lifetime PD Model.

Data Types: string

Distance between age values in training data in the panel data input, specified as TimeInterval and a positive numeric scalar.

Use the TimeInterval name-value argument to fit time-dependent models and also as the time interval for the PD computation when you use the predict function. For example, if the age data (AgeVar) is 1, 2, 3, ..., then the TimeInterval is 1; if the age data is 0.25, 0.5, 0.75,..., then the TimeInterval is 0.25. For more information, see Time Interval for Cox Models and Lifetime Prediction and Time Interval.

Note

Unlike Logistic and Probit models, a Cox model requires an AgeVar variable. By default, if you do not specify a TimeInterval when creating a Cox model, the TimeInterval is inferred from the increments in the AgeVar values in the training data.

Data Types: double

Since R2023a

Method to handle tied default times, specified as a string or character vector with one of the following tie-break methods:

  • breslow — Breslow's approximation to the partial likelihood

  • efron — Efron's approximation to the partial likelihood

For credit applications, the time to default comes discretized and there are many "ties." This means that are multiple borrowers that may default at the same (discretized) time (such as, in the second year of their loan). TieBreakMethod supports the breslow or efron methods to handle this scenario.

Data Types: string | char

Properties

expand all

User-defined model ID, returned as a string.

Data Types: string

User-defined description, returned as a string.

Data Types: string

Underlying statistical model, returned as a returned as a Cox proportional hazards model object. For more information, see fitcox and CoxModel.

Data Types: CoxModel

ID variable indicating which column in data contains the loan or borrower ID, returned as a string.

Data Types: string

Age variable indicating which column in data contains the loan age information, returned as a string.

Data Types: string

Loan variables indicating which column in data contains the loan-specific information, returned as a string array.

Data Types: string

Macro variables indicating which column in data contains the macroeconomic information, returned as a string array.

Data Types: string

Variable indicating which column in data contains the response variable, returned as a string.

Data Types: string

Variable indicating which column in data contains the observation weights, returned as a string array.

Data Types: string

This property is read-only.

Distance between age values in panel data input, returned as a scalar positive numeric.

Data Types: double

Extrapolation factor, returned as a positive numeric scalar between 0 and 1.

By default, the ExtrapolationFactor is set to 1. For age values (AgeVar) greater than the maximum age observed in the training data, the conditional PD, computed with predict, uses the maximum age observed in the training data. In particular, the predicted PD value is constant if the predictor values do not change and only the age values change when the ExtrapolationFactor is 1. For more information, see Extrapolation for Cox Models, Extrapolation Factor for Cox Models, and Use Cox Lifetime PD Model to Predict Conditional PD.

Data Types: double

Method to handle tied default times, returned as a string.

Data Types: string

Object Functions

predictCompute conditional PD
predictLifetimeCompute cumulative lifetime PD, marginal PD, and survival probability
modelDiscriminationCompute AUROC and ROC data
modelCalibrationCompute RMSE of predicted and observed PDs on grouped data
modelDiscriminationPlotPlot ROC curve
modelCalibrationPlotPlot observed default rates compared to predicted PDs on grouped data
discardResidualsDiscard residual information of underlying Cox model

Examples

collapse all

This example shows how to use fitLifetimePDModel to create a Cox model using credit and macroeconomic data.

Load Data

Load the credit portfolio data.

load RetailCreditPanelData.mat
disp(head(data))
    ID    ScoreGroup    YOB    Default    Year
    __    __________    ___    _______    ____

    1      Low Risk      1        0       1997
    1      Low Risk      2        0       1998
    1      Low Risk      3        0       1999
    1      Low Risk      4        0       2000
    1      Low Risk      5        0       2001
    1      Low Risk      6        0       2002
    1      Low Risk      7        0       2003
    1      Low Risk      8        0       2004
disp(head(dataMacro))
    Year     GDP     Market
    ____    _____    ______

    1997     2.72      7.61
    1998     3.57     26.24
    1999     2.86      18.1
    2000     2.43      3.19
    2001     1.26    -10.51
    2002    -0.59    -22.95
    2003     0.63      2.78
    2004     1.85      9.48

Join the two data components into a single data set.

data = join(data,dataMacro);
disp(head(data))
    ID    ScoreGroup    YOB    Default    Year     GDP     Market
    __    __________    ___    _______    ____    _____    ______

    1      Low Risk      1        0       1997     2.72      7.61
    1      Low Risk      2        0       1998     3.57     26.24
    1      Low Risk      3        0       1999     2.86      18.1
    1      Low Risk      4        0       2000     2.43      3.19
    1      Low Risk      5        0       2001     1.26    -10.51
    1      Low Risk      6        0       2002    -0.59    -22.95
    1      Low Risk      7        0       2003     0.63      2.78
    1      Low Risk      8        0       2004     1.85      9.48

Partition Data

Separate the data into training and test partitions.

nIDs = max(data.ID);
uniqueIDs = unique(data.ID);

rng('default'); % For reproducibility
c = cvpartition(nIDs,'HoldOut',0.4);

TrainIDInd = training(c);
TestIDInd = test(c);

TrainDataInd = ismember(data.ID,uniqueIDs(TrainIDInd));
TestDataInd = ismember(data.ID,uniqueIDs(TestIDInd));

Create a Cox Lifetime PD Model

Use fitLifetimePDModel to create a Cox model using the training data.

pdModel = fitLifetimePDModel(data(TrainDataInd,:),"Cox",...
    AgeVar="YOB", ...
    IDVar="ID", ...
    LoanVars="ScoreGroup", ...
    MacroVars={'GDP','Market'}, ...
    ResponseVar="Default");
disp(pdModel)
  Cox with properties:

           TimeInterval: 1
    ExtrapolationFactor: 1
                ModelID: "Cox"
            Description: ""
        UnderlyingModel: [1x1 CoxModel]
                  IDVar: "ID"
                 AgeVar: "YOB"
               LoanVars: "ScoreGroup"
              MacroVars: ["GDP"    "Market"]
            ResponseVar: "Default"
             WeightsVar: ""

Display the underlying model.

disp(pdModel.UnderlyingModel)
Cox Proportional Hazards regression model

                                 Beta          SE         zStat       pValue   
                              __________    _________    _______    ___________

    ScoreGroup_Medium Risk       -0.6794     0.037029    -18.348     3.4442e-75
    ScoreGroup_Low Risk          -1.2442     0.045244    -27.501    1.7116e-166
    GDP                        -0.084533     0.043687     -1.935       0.052995
    Market                    -0.0084411    0.0032221    -2.6198      0.0087991


Log-likelihood: -41742.871

Validate Model

Use modelDiscrimination to measure the ranking of customers by PD.

DataSetChoice = "Testing";
if DataSetChoice=="Training"
    Ind = TrainDataInd;
else
    Ind = TestDataInd;
end

DiscMeasure = modelDiscrimination(pdModel,data(Ind,:),SegmentBy="ScoreGroup")
DiscMeasure=3×1 table
                                    AUROC 
                                   _______

    Cox, ScoreGroup=High Risk      0.64112
    Cox, ScoreGroup=Medium Risk    0.61989
    Cox, ScoreGroup=Low Risk        0.6314

disp(DiscMeasure)
                                    AUROC 
                                   _______

    Cox, ScoreGroup=High Risk      0.64112
    Cox, ScoreGroup=Medium Risk    0.61989
    Cox, ScoreGroup=Low Risk        0.6314

Use modelDiscriminationPlot to visualize the ROC curve.

modelDiscriminationPlot(pdModel,data(Ind,:),SegmentBy="ScoreGroup")

Figure contains an axes object. The axes object with title ROC Segmented by ScoreGroup, xlabel Fraction of Non-Defaulters, ylabel Fraction of Defaulters contains 3 objects of type line. These objects represent Cox, High Risk, AUROC = 0.64112, Cox, Medium Risk, AUROC = 0.61989, Cox, Low Risk, AUROC = 0.6314.

Use modelCalibration to measure the calibration of the predicted PD values. The modelCalibration function requires a grouping variable and compares the accuracy of the observed default rate in the group with the average predicted PD for the group.

CalMeasure = modelCalibration(pdModel,data(Ind,:),{'YOB','ScoreGroup'})
CalMeasure=table
                                         RMSE   
                                       _________

    Cox, grouped by YOB, ScoreGroup    0.0012471

disp(CalMeasure)
                                         RMSE   
                                       _________

    Cox, grouped by YOB, ScoreGroup    0.0012471

Use modelCalibrationPlot to visualize the observed default rates compared to the predicted PD.

modelCalibrationPlot(pdModel,data(Ind,:),{'YOB','ScoreGroup'})

Figure contains an axes object. The axes object with title Scatter Grouped by YOB and ScoreGroup Cox, RMSE = 0.0012471, xlabel YOB, ylabel PD contains 6 objects of type line. One or more of the lines displays its values using only markers These objects represent High Risk, Observed, Medium Risk, Observed, Low Risk, Observed, High Risk, Cox, Medium Risk, Cox, Low Risk, Cox.

Predict Conditional and Lifetime PD

Use the predict function to predict conditional PD values. The prediction is a row-by-row prediction.

%dataCustomer1 = data(1:8,:);
CondPD = predict(pdModel,data(Ind,:));

Use predictLifetime to predict the lifetime cumulative PD values (computing marginal and survival PD values is also supported).

LifetimePD = predictLifetime(pdModel,data(Ind,:));

Since R2023a

This example shows how to create a Cox model and select the tie-break method while fitting a Cox lifetime PD model.

Load Data

Load the credit portfolio data.

load RetailCreditPanelData.mat
disp(head(data))
    ID    ScoreGroup    YOB    Default    Year
    __    __________    ___    _______    ____

    1      Low Risk      1        0       1997
    1      Low Risk      2        0       1998
    1      Low Risk      3        0       1999
    1      Low Risk      4        0       2000
    1      Low Risk      5        0       2001
    1      Low Risk      6        0       2002
    1      Low Risk      7        0       2003
    1      Low Risk      8        0       2004
disp(head(dataMacro))
    Year     GDP     Market
    ____    _____    ______

    1997     2.72      7.61
    1998     3.57     26.24
    1999     2.86      18.1
    2000     2.43      3.19
    2001     1.26    -10.51
    2002    -0.59    -22.95
    2003     0.63      2.78
    2004     1.85      9.48

Join the two data components into a single data set.

data = join(data,dataMacro);
disp(head(data))
    ID    ScoreGroup    YOB    Default    Year     GDP     Market
    __    __________    ___    _______    ____    _____    ______

    1      Low Risk      1        0       1997     2.72      7.61
    1      Low Risk      2        0       1998     3.57     26.24
    1      Low Risk      3        0       1999     2.86      18.1
    1      Low Risk      4        0       2000     2.43      3.19
    1      Low Risk      5        0       2001     1.26    -10.51
    1      Low Risk      6        0       2002    -0.59    -22.95
    1      Low Risk      7        0       2003     0.63      2.78
    1      Low Risk      8        0       2004     1.85      9.48

Join the Data

Join the two data components into a single data set.

data = join(data,dataMacro);
disp(head(data))
    ID    ScoreGroup    YOB    Default    Year     GDP     Market
    __    __________    ___    _______    ____    _____    ______

    1      Low Risk      1        0       1997     2.72      7.61
    1      Low Risk      2        0       1998     3.57     26.24
    1      Low Risk      3        0       1999     2.86      18.1
    1      Low Risk      4        0       2000     2.43      3.19
    1      Low Risk      5        0       2001     1.26    -10.51
    1      Low Risk      6        0       2002    -0.59    -22.95
    1      Low Risk      7        0       2003     0.63      2.78
    1      Low Risk      8        0       2004     1.85      9.48

Partition the Data

Separate the data into training and test partitions.

nIDs = max(data.ID);
uniqueIDs = unique(data.ID);

rng('default'); % for reproducibility
c = cvpartition(nIDs,'HoldOut',0.4);

TrainIDInd = training(c);
TestIDInd = test(c);

TrainDataInd = ismember(data.ID,uniqueIDs(TrainIDInd));
TestDataInd = ismember(data.ID,uniqueIDs(TestIDInd));

Create a Cox Lifetime PD Model with Breslow's Method

Use fitLifetimePDModel to create a Cox model using the training data. Use the name-value argument TieBreakMethod to set tie-break method to 'breslow'. This is the default choice for this argument.

pdModel1 = fitLifetimePDModel(data(TrainDataInd,:),"Cox",...
ModelID="Cox-Breslow", IDVar="ID", AgeVar="YOB", ...
LoanVars="ScoreGroup", MacroVars={'GDP','Market'}, ...
ResponseVar="Default",TieBreakMethod='breslow');

Display the underlying model.

disp(pdModel1.Model)
Cox Proportional Hazards regression model

                                 Beta          SE         zStat       pValue   
                              __________    _________    _______    ___________

    ScoreGroup_Medium Risk       -0.6794     0.037029    -18.348     3.4442e-75
    ScoreGroup_Low Risk          -1.2442     0.045244    -27.501    1.7116e-166
    GDP                        -0.084533     0.043687     -1.935       0.052995
    Market                    -0.0084411    0.0032221    -2.6198      0.0087991


Log-likelihood: -41742.871

Use predict to predict the conditional PD.

pd1 = predict(pdModel1,data(TestDataInd,:));

Create a Cox Lifetime PD Model with Efron's Method

Use fitLifetimePDModel to create a Cox model using the training data. Use the name-value argument TieBreakMethod to set tie-break method to 'Efron'. This is the default choice for this argument.

pdModel2 = fitLifetimePDModel(data(TrainDataInd,:),"Cox",...
ModelID="Cox-Efron", IDVar="ID", AgeVar="YOB", ...
LoanVars="ScoreGroup", MacroVars={'GDP','Market'}, ...
ResponseVar="Default",TieBreakMethod='efron');

Display the underlying model. The coefficients are only slightly different for this data set.

disp(pdModel2.Model)
Cox Proportional Hazards regression model

                                 Beta          SE         zStat       pValue  
                              __________    _________    _______    __________

    ScoreGroup_Medium Risk       -0.6844     0.037029    -18.483    2.8461e-76
    ScoreGroup_Low Risk          -1.2515     0.045243    -27.662    2.006e-168
    GDP                        -0.084985     0.043691    -1.9452      0.051756
    Market                    -0.0085126    0.0032223    -2.6418     0.0082469


Log-likelihood: -41713.445

Use predict to predict the conditional PD for the second Cox model.

pd2 = predict(pdModel2,data(TestDataInd,:));

Compare Cox Models

The predictions for the two Cox models are almost the same for this data set.

[pd1(1:10) pd2(1:10)]
ans = 10×2

    0.0162    0.0161
    0.0091    0.0090
    0.0081    0.0081
    0.0073    0.0072
    0.0064    0.0064
    0.0072    0.0072
    0.0030    0.0030
    0.0016    0.0016
    0.0162    0.0161
    0.0091    0.0090

For this data set, the model discrimination (modelDiscrimination) does not seem to change with the TieBreakMethod method and the model accuracy (modelCalibration) shows only a negligible difference in RMSE.

modelDiscriminationPlot(pdModel1,data(TestDataInd,:),ReferencePD=pd2,ReferenceID=pdModel2.ModelID)

Figure contains an axes object. The axes object with title ROC Cox-Breslow, AUROC = 0.70048 Cox-Efron, AUROC = 0.70048, xlabel Fraction of Non-Defaulters, ylabel Fraction of Defaulters contains 2 objects of type line. These objects represent Cox-Breslow, Cox-Efron.

modelCalibrationPlot(pdModel1,data(TestDataInd,:),'Year',ReferencePD=pd2,ReferenceID=pdModel2.ModelID)

Figure contains an axes object. The axes object with title Scatter Grouped by Year Cox-Breslow, RMSE = 0.00047088 Cox-Efron, RMSE = 0.00047474, xlabel Year, ylabel PD contains 3 objects of type line. One or more of the lines displays its values using only markers These objects represent Observed, Cox-Breslow, Cox-Efron.

More About

expand all

References

[1] Baesens, Bart, Daniel Roesch, and Harald Scheule. Credit Risk Analytics: Measurement Techniques, Applications, and Examples in SAS. Wiley, 2016.

[2] Bellini, Tiziano. IFRS 9 and CECL Credit Risk Modelling and Validation: A Practical Guide with Examples Worked in R and SAS. San Diego, CA: Elsevier, 2019.

[3] Breeden, Joseph. Living with CECL: The Modeling Dictionary. Santa Fe, NM: Prescient Models LLC, 2018.

[4] Roesch, Daniel and Harald Scheule. Deep Credit Risk: Machine Learning with Python. Independently published, 2020.

Version History

Introduced in R2021b

expand all