Current situation: An established pattern in SAS Enterprise Miner is to use the IGN-node with binary transformation of a linear target variable to go on and make use of the scorecard node to make a logistic regression model based on the binary transformed new target variable. In Model Studio this pattern is only partly implemented as it is only the binary transformation in the IGN node that is possible to use. Due to the current limitation of Model Studio that prevents the user to switch target variable within a project, it is impossible to add a scorecard node to the IGN node as the project is initiated with a linear target variable. This behavior is confirmed in support case CS0221275.    The result is that the user has to make use of multiple Model Studio projects to be able to build a LGD model as proposed in this paper 141-2012: Building WOE Binned LGD Scorecards using SAS® Enterprise Miner™. In addition, if the user wants to build a scorecard model as a challenger model to other kinds of LGD models that make use of a linear target you end up with a scattered process with models developed here and there.   Desired outcome: Make it possible to create a scorecard model within the same project as where the binary transformation is performed.    Added benefit: Easier to document the model development and easier to maintain the model development project (mind you that these models and highly regulated and tend to live for years ). Easier to utilise link to Model Manager from Model Studio. When pushing a model from Model Studio to Model Manager, the default behavior is to create a new Model Manager project pr Model Studio project and this would then possibly result in multiple Model Manager projects if not resolved.    Negative effect of change: None.   ... View more

[DISCLAIMER: I am the author of the referenced documents] This post proposes incorporating a generalization of ROC curve analysis into SAS statistical subroutines (e.g., PROCS LOGISTIC and GENMOD). This generalized method is straightforward to implement and a git repository(1) is linked below with an implementation using PROC LOGISTIC.    Classical ROC curve analysis is defined for binary outcomes comprising two levels, but extension to discrete outcomes comprising three or more levels would generalize its scope of applications. Although a variety of methods have been proposed to accomplish this, they tend to entail significant limitations, such as: restriction of the outcome to relatively few levels; reliance on valid, but somewhat abstruse theory; implementation only with specialized software.    Another method has been described that extends ROC curve analysis to ordinal outcomes with three or more levels(2). The theoretical justification for this method -- "cumulative ROC curves" -- relies on the ROC curve's essential representation of a rank-based association between the discrete outcome and continuous predictor, as well as a straightforward interpretation of multinomial regression. Implementation is likewise straightforward, requiring only software for running multinomial regression models, specifically, but not necessarily limited to the cumulative logit model. Consequently, cumulative ROC curve analysis is suitable for a general scope of applications while also being accessible to quantitative tools already established for characterizing binary ROC curves, such as sensitivity, specificity, area-under-the-curve, Total Accuracy, and Youden's Index. In particular, these established tools can be directly applied to identifying cutpoints on the continuous measurement scale that discriminate between successive levels of the ordinal outcome. Cumulative ROC curves have been shown in simulations to perform as expected.   References 1. [REPOSITORY] deCastro, B.R. (2019). %cumRoc3 --- Cumulative ROC curve analysis of three-level ordinal outcomes (1.0.1). https://doi.org/10.5281/zenodo.3364094. 2. deCastro, B.R. (2019). Cumulative ROC curves for discriminating three or more ordinal outcomes with cutpoints on a shared continuous measurement scale. PLoS ONE, 14(8). https://doi.org/10.1371/journal.pone.0221433. ... View more

Dear all   I am pleased to inform you that I have developed two distinct SAS macros designed to explain the Low Default Portfolio modeling approach suggested by Pluto &Tasche  https://arxiv.org/pdf/cond-mat/0411699v3     These macros account for both correlated and independent default events. I believe these tools will be highly valuable for the SAS community, particularly in the area of Credit Risk Modeling for Low Default or Zero Default portfolios.   From my point of view, these ones are great examples of built-in functions to be deployed as part of an particular SAS Proc Statement, such as: PROC LDPT   Best regards, Raphael Chaves ... View more

It is not possible to cancel a program that has been submitted using the Background Submit feature in SAS Viya. There is no Cancel button for individual programs that are submitted using Background Submit. Only an admin can cancel the jobs using workload orchestrator.  ... View more