SlideShare a Scribd company logo
1 of 2
Download to read offline
APPLIED MULTIVARIATE ALGORITHMS INC
Daniel Kocis, Ph.D. – President help@multivariate.com Work: 631 871 3348 www.multivariate.com Twitter-MultiAlgo
PROFESSIONAL SUMMARY:
As Senior SAS Developer supporting a G-SIFI CCAR mandate for Consumer Credit Cards constructed an end to end
Model Risk Management MRM Tool (Dynamic Scoring Engine) which utilized SAS Software tools and SAS Macro
language providing documentation, validation, automation and inventory control. Continued with support, design and
development of a "Glass Box" front end driver that bundle streamlined source programs and model parameter inputs that
tracked a transition matrix of consumer credit behavior. It automated formula builds, compound and dynamic
transformations of variables, scoring of these formulas and the calculation of conditional probabilities across multinomial
and binomial distributions. Models focused on PPNR, PD, RWA and TDR
Developed modeling and reporting processes that migrated multiple LOB risk-reporting into a single group which
oversaw acquisition, default analysis, auditing and regulatory data governance issues. Provided key components in the
risk control production for this major money center bank, across several SAS 9.2 metadata information portals
connected to Teradata, DB2, Oracle enterprise repositories.
Supporting Regulatory Enterprise Credit Risk Reporting
Created several modeling and reporting systems across all consumer credit products (Card, UNS, Auto, Small Business,
International, MTG, and HE). Defined key metrics (outstanding balance, active/open/defaulted account exposures, credit
utilization, and OCC compliant indicators) were reported monthly across all consumer credit product origination and
portfolio tables and compared against base line predictive multivariate models. Data-sourced all tables and produced
dashboards for the BOD of the total risk exposure and credit utilization reporting enterprise wide geographic
concentrations. Created several production run model libraries that migrated all processes into automated projects.
Supporting portfolio credit performance trend tracking by LOB
Used credit bureau samples to define peer and total market segments of dealer based auto and specialty brokered loans.
These were profiled by geographic concentration risk, calculating share of business with current estimated losses and
volatility adjusted losses reported at an origination and portfolios level.
Supporting Bank Risk Policies monitoring
Produced delinquency rates wedges for all enterprise credit risk policy and geographic concentration limits by tracking
actual performance against portfolio target levels within domestic and international markets. Rolling historical MTD-QTD-
YTD by account open date with total outstanding available and delinquent balances with reporting delivered on a monthly
SLA to all LOB management with focus on PD, EAD and LGD. Used EM6.2 “Rule-Based Technique” and other SAS
techniques to investigate drivers of impaired accounts.
“Stresstest” Data Transformations
The service provided is lazer focused on the analysis of transformations; specifically how to introduce externally provided
fed “stresstest” variables into an existing asset class PD model.
Transformations - How to add HPI – home price index to a backtesting retail mortgage PD model.
In this example we begin with quarterly data using internal data sources (e.g. LTV, loan term) to define a back testing
environment without using external sources data. The goal is introduce external supplied data (HPI) and then assess its
impact on model performance.
http://support.sas.com/documentation/cdl/en/mdsug/65072/HTML/default/viewer.htm#n194xndt3b3y1pn1ufc0mqbsmht4.h
tm
Starting with an existing PD model, contemporaneous quarterly level HPI is added and its impact on model fit calculated.
The exact process provided explores several basic TRANSFORMATION_FUNCTIONS ( ie LAG, QOQ) and then creates
COMPOUND_TRANSFORMATION_FUNCTIONS.
A key consideration is the time span of the existing data. I am differentiating the reporting requirements of the scenario
(2014:Q4 – 2017:Q4) and focused on a back testing time span assumed to start 2006:Q1.
While each portfolio is different, I want to point out that complex transformations shorten available records for model
building.
The TRANSFORMATION_FUNCTIONS shown here include LAG, CUM with
COMPOUND_TRANFORMATION_FUNCTION of L1_CUM1 (Combined two or more TRANSFORMATION_FUNCTIONS
(LAG and CUM) with a maximum dimension of 5.
The idea is to demonstrate the ability to create and combine standard transformations. In this example we
created 220 transformations and then applied them to a back testing data base ready for reporting.
L1_HPI = lag1(HPI);
CUM1_HPI = HPI;
L1_CUM1_HPI = lag1(CUM1_HPI);

More Related Content

Similar to CCAR - Kocis

Risk dk
Risk dkRisk dk
Risk dkdkocis
 
Utsav sarkar senior consultant
Utsav sarkar senior consultant Utsav sarkar senior consultant
Utsav sarkar senior consultant Utsav Sarkar
 
Utsav sarkar senior consultant
Utsav sarkar senior consultant Utsav sarkar senior consultant
Utsav sarkar senior consultant Utsav Sarkar
 
Utsav sarkar senior consultant
Utsav sarkar senior consultant Utsav sarkar senior consultant
Utsav sarkar senior consultant Utsav Sarkar
 
VSI Client Engagements
VSI Client EngagementsVSI Client Engagements
VSI Client EngagementsChris Phillips
 
SD Basel process automation seminar presentation
SD Basel process automation seminar presentationSD Basel process automation seminar presentation
SD Basel process automation seminar presentationsarojkdas
 
Quant Foundry Labs - Low Probability Defaults
Quant Foundry Labs - Low Probability DefaultsQuant Foundry Labs - Low Probability Defaults
Quant Foundry Labs - Low Probability DefaultsDavidkerrkelly
 
Resume_Nidhi Malhotra_BA_shared
Resume_Nidhi Malhotra_BA_sharedResume_Nidhi Malhotra_BA_shared
Resume_Nidhi Malhotra_BA_sharedNidhi Malhotra
 
Resume_John Huff
Resume_John HuffResume_John Huff
Resume_John HuffJohn Huff
 
ORIGINATIONNEXT- Risk Assessment Model
ORIGINATIONNEXT- Risk Assessment ModelORIGINATIONNEXT- Risk Assessment Model
ORIGINATIONNEXT- Risk Assessment ModelCRMNEXT
 
Resume_Partha_Data Consultant_23_July_2016
Resume_Partha_Data Consultant_23_July_2016Resume_Partha_Data Consultant_23_July_2016
Resume_Partha_Data Consultant_23_July_2016Partha Sarathi Pattnaik
 
Chris Moore_Resume_Aug_2016
Chris Moore_Resume_Aug_2016Chris Moore_Resume_Aug_2016
Chris Moore_Resume_Aug_2016Chris M. Moore
 
Adopting a Top-Down Approach to Model Risk Governance to Optimize Digital Tra...
Adopting a Top-Down Approach to Model Risk Governance to Optimize Digital Tra...Adopting a Top-Down Approach to Model Risk Governance to Optimize Digital Tra...
Adopting a Top-Down Approach to Model Risk Governance to Optimize Digital Tra...Jacob Kosoff
 
rahul cv modified
rahul cv modifiedrahul cv modified
rahul cv modifiedRahul Patil
 
KI_res_24_yrs_exp_big_6
KI_res_24_yrs_exp_big_6KI_res_24_yrs_exp_big_6
KI_res_24_yrs_exp_big_6keith inman
 
KI_res_24_yrs_exp_big_6
KI_res_24_yrs_exp_big_6KI_res_24_yrs_exp_big_6
KI_res_24_yrs_exp_big_6keith inman
 
C ganti pred_resume_032013
C ganti pred_resume_032013C ganti pred_resume_032013
C ganti pred_resume_032013C.S. Ganti
 

Similar to CCAR - Kocis (20)

Risk dk
Risk dkRisk dk
Risk dk
 
Resume ia
Resume iaResume ia
Resume ia
 
Utsav sarkar senior consultant
Utsav sarkar senior consultant Utsav sarkar senior consultant
Utsav sarkar senior consultant
 
Utsav sarkar senior consultant
Utsav sarkar senior consultant Utsav sarkar senior consultant
Utsav sarkar senior consultant
 
Utsav sarkar senior consultant
Utsav sarkar senior consultant Utsav sarkar senior consultant
Utsav sarkar senior consultant
 
VSI Client Engagements
VSI Client EngagementsVSI Client Engagements
VSI Client Engagements
 
SD Basel process automation seminar presentation
SD Basel process automation seminar presentationSD Basel process automation seminar presentation
SD Basel process automation seminar presentation
 
Surender Reddy
Surender ReddySurender Reddy
Surender Reddy
 
Quant Foundry Labs - Low Probability Defaults
Quant Foundry Labs - Low Probability DefaultsQuant Foundry Labs - Low Probability Defaults
Quant Foundry Labs - Low Probability Defaults
 
Resume_Nidhi Malhotra_BA_shared
Resume_Nidhi Malhotra_BA_sharedResume_Nidhi Malhotra_BA_shared
Resume_Nidhi Malhotra_BA_shared
 
Resume_John Huff
Resume_John HuffResume_John Huff
Resume_John Huff
 
ORIGINATIONNEXT- Risk Assessment Model
ORIGINATIONNEXT- Risk Assessment ModelORIGINATIONNEXT- Risk Assessment Model
ORIGINATIONNEXT- Risk Assessment Model
 
Resume_Partha_Data Consultant_23_July_2016
Resume_Partha_Data Consultant_23_July_2016Resume_Partha_Data Consultant_23_July_2016
Resume_Partha_Data Consultant_23_July_2016
 
Chris Moore_Resume_Aug_2016
Chris Moore_Resume_Aug_2016Chris Moore_Resume_Aug_2016
Chris Moore_Resume_Aug_2016
 
Adopting a Top-Down Approach to Model Risk Governance to Optimize Digital Tra...
Adopting a Top-Down Approach to Model Risk Governance to Optimize Digital Tra...Adopting a Top-Down Approach to Model Risk Governance to Optimize Digital Tra...
Adopting a Top-Down Approach to Model Risk Governance to Optimize Digital Tra...
 
Flyer ProTrack
Flyer ProTrackFlyer ProTrack
Flyer ProTrack
 
rahul cv modified
rahul cv modifiedrahul cv modified
rahul cv modified
 
KI_res_24_yrs_exp_big_6
KI_res_24_yrs_exp_big_6KI_res_24_yrs_exp_big_6
KI_res_24_yrs_exp_big_6
 
KI_res_24_yrs_exp_big_6
KI_res_24_yrs_exp_big_6KI_res_24_yrs_exp_big_6
KI_res_24_yrs_exp_big_6
 
C ganti pred_resume_032013
C ganti pred_resume_032013C ganti pred_resume_032013
C ganti pred_resume_032013
 

More from Daniel Kocis Ph.D. - Chair (13)

Overview
OverviewOverview
Overview
 
Channel Strategy1
Channel Strategy1Channel Strategy1
Channel Strategy1
 
Results for Keith
Results for KeithResults for Keith
Results for Keith
 
Case Conditions
Case ConditionsCase Conditions
Case Conditions
 
How_to_maximize_the_number_of_spots
How_to_maximize_the_number_of_spotsHow_to_maximize_the_number_of_spots
How_to_maximize_the_number_of_spots
 
AD1
AD1AD1
AD1
 
Consumer Models
Consumer ModelsConsumer Models
Consumer Models
 
BuildHistoryfinal
BuildHistoryfinalBuildHistoryfinal
BuildHistoryfinal
 
Weekly Forecasts
Weekly ForecastsWeekly Forecasts
Weekly Forecasts
 
QTMS-EM-Combinatorical Model
QTMS-EM-Combinatorical ModelQTMS-EM-Combinatorical Model
QTMS-EM-Combinatorical Model
 
Brands Analysis
Brands AnalysisBrands Analysis
Brands Analysis
 
Pharma
PharmaPharma
Pharma
 
Media
MediaMedia
Media
 

CCAR - Kocis

  • 1. APPLIED MULTIVARIATE ALGORITHMS INC Daniel Kocis, Ph.D. – President help@multivariate.com Work: 631 871 3348 www.multivariate.com Twitter-MultiAlgo PROFESSIONAL SUMMARY: As Senior SAS Developer supporting a G-SIFI CCAR mandate for Consumer Credit Cards constructed an end to end Model Risk Management MRM Tool (Dynamic Scoring Engine) which utilized SAS Software tools and SAS Macro language providing documentation, validation, automation and inventory control. Continued with support, design and development of a "Glass Box" front end driver that bundle streamlined source programs and model parameter inputs that tracked a transition matrix of consumer credit behavior. It automated formula builds, compound and dynamic transformations of variables, scoring of these formulas and the calculation of conditional probabilities across multinomial and binomial distributions. Models focused on PPNR, PD, RWA and TDR Developed modeling and reporting processes that migrated multiple LOB risk-reporting into a single group which oversaw acquisition, default analysis, auditing and regulatory data governance issues. Provided key components in the risk control production for this major money center bank, across several SAS 9.2 metadata information portals connected to Teradata, DB2, Oracle enterprise repositories. Supporting Regulatory Enterprise Credit Risk Reporting Created several modeling and reporting systems across all consumer credit products (Card, UNS, Auto, Small Business, International, MTG, and HE). Defined key metrics (outstanding balance, active/open/defaulted account exposures, credit utilization, and OCC compliant indicators) were reported monthly across all consumer credit product origination and portfolio tables and compared against base line predictive multivariate models. Data-sourced all tables and produced dashboards for the BOD of the total risk exposure and credit utilization reporting enterprise wide geographic concentrations. Created several production run model libraries that migrated all processes into automated projects. Supporting portfolio credit performance trend tracking by LOB Used credit bureau samples to define peer and total market segments of dealer based auto and specialty brokered loans. These were profiled by geographic concentration risk, calculating share of business with current estimated losses and volatility adjusted losses reported at an origination and portfolios level. Supporting Bank Risk Policies monitoring Produced delinquency rates wedges for all enterprise credit risk policy and geographic concentration limits by tracking actual performance against portfolio target levels within domestic and international markets. Rolling historical MTD-QTD- YTD by account open date with total outstanding available and delinquent balances with reporting delivered on a monthly SLA to all LOB management with focus on PD, EAD and LGD. Used EM6.2 “Rule-Based Technique” and other SAS techniques to investigate drivers of impaired accounts. “Stresstest” Data Transformations The service provided is lazer focused on the analysis of transformations; specifically how to introduce externally provided fed “stresstest” variables into an existing asset class PD model. Transformations - How to add HPI – home price index to a backtesting retail mortgage PD model. In this example we begin with quarterly data using internal data sources (e.g. LTV, loan term) to define a back testing environment without using external sources data. The goal is introduce external supplied data (HPI) and then assess its impact on model performance. http://support.sas.com/documentation/cdl/en/mdsug/65072/HTML/default/viewer.htm#n194xndt3b3y1pn1ufc0mqbsmht4.h tm
  • 2. Starting with an existing PD model, contemporaneous quarterly level HPI is added and its impact on model fit calculated. The exact process provided explores several basic TRANSFORMATION_FUNCTIONS ( ie LAG, QOQ) and then creates COMPOUND_TRANSFORMATION_FUNCTIONS. A key consideration is the time span of the existing data. I am differentiating the reporting requirements of the scenario (2014:Q4 – 2017:Q4) and focused on a back testing time span assumed to start 2006:Q1. While each portfolio is different, I want to point out that complex transformations shorten available records for model building. The TRANSFORMATION_FUNCTIONS shown here include LAG, CUM with COMPOUND_TRANFORMATION_FUNCTION of L1_CUM1 (Combined two or more TRANSFORMATION_FUNCTIONS (LAG and CUM) with a maximum dimension of 5. The idea is to demonstrate the ability to create and combine standard transformations. In this example we created 220 transformations and then applied them to a back testing data base ready for reporting. L1_HPI = lag1(HPI); CUM1_HPI = HPI; L1_CUM1_HPI = lag1(CUM1_HPI);