Predicting Credit Score Calibrations through Economic Events

Size: px

Start display at page:

Download "Predicting Credit Score Calibrations through Economic Events"

Gilbert Reynolds
8 years ago
Views:

1 Predicting Credit Score Calibrations through Economic Events Joseph L. Breeden, Ph.D. President, COO, & Chief Scientist Michael A. Smith Chief Software Architect Copyright 2002 Strategic Analytics Inc. May 29, 2002

2 Overview Review of Score-Odds Calibrations Understanding Portfolio Dynamics Dual-time Dynamics Scenario-based Forecasting Score-Odds Predictions 2

4 Score-Odds Analysis Bureau scores have been shown to maintain effective rank ordering of consumers through a variety of environments. Score-odds calibrations calibrate credit scores to the specific odds of delinquency. For purposes of managing new originations, a common approach is to plot the odds of ever being 60+ DPD during the first 2 years of product ownership versus the score at origination. Re-estimation of score-odds calibrations are used as a guide in adjusting cut-off scores for originations. 4

For purposes of managing new originations, a common approach is to plot the odds of ever being 60+ DPD during the first 2 years

5 Score-Odds Calibrations Over Time Each vintage experiences a potentially different environment. The 1997 vintage performed better in its first 2 years than 1998 and Score-Odds by Vintage Good / Bad Odds (Ever 60+ in First 2 Years) Origination Score H H H

/ Bad Odds (Ever 60+ in First 2 Years) 10 1 500 520 540 560 580 600 620 640 660 680 700 Origination Score

6 Environmental Trends An independent measure of environmental impacts (via Dual-time Dynamics) confirms that the 1997 vintage lived through a better environment. 0.4 Environmental Trends 0.3 Relative Impact Jan-1997 Jan-1998 Jan-1999 Jan-2000 Jan-2001 Date 6

a better environment. 0.4 Environmental Trends 0.3 Relative Impact 0.

7 Score-Odds on Maturing Accounts Including recent vintages is problematic because they have been observed for a shorter period of time. 100 Score-Odds by Vintage Good / Bad Odds (Ever 60+ in First 2 Years) Origination Score H H H2 2000H H1 2000H2 1999H H1 2000H2 2001H1 1999H H2 2001H2 2000H

Origination Score 1997 0.158 1998 0.195 1999H1 0.186 1999H2 0.187 1997 0.158 1999H2 2000H1 19971998 0.158 0.187 0.1570.

8 Account Maturation Dynamics Peak delinquency risk may not occur for several years after booking. This portfolio shows peak delinquency risk between 24 and 36 months-onbooks. Vintage Maturation Process Monthly Change in Accts Ever % 0.6% 0.5% 0.4% 0.3% 0.2% 0.1% 0.0% Months-on-Books 8

This portfolio shows peak delinquency risk between 24 and 36 months-onbooks.

9 Originations Timeline Trailing score-odds calibrations have several problems: Cannot use the most recent vintages because of immaturity of the accounts. Accounts booked today will have peak delinquency up to several years in the future. Setting cut-off scores based upon a calibration to the previous 2 years data implicitly assumes that the macroeconomic environment will remain unchanged for 4 to 5 years. New accounts originated Calibration on previous 2 years of data Peak delinquency risk 9

Setting cut-off scores based upon a calibration to the previous 2 years data implicitly assumes that the macroeconomic

10 Creating an Ideal Approach The technique just illustrated is simplistic. We need a score-odds calibration methodology that: Uses the full history of both old and new vintages. Normalizes for changes in business practices, e.g. improved collections scores or systems. Can be projected under changing macroeconomic environments. 10

old and new vintages. Normalizes for changes in business practices, e.g. improved collections scores or systems.

12 Components of Portfolio Performance Vintage Lifecycle 3.0% Maturation 2.5% 2.0% 1.5% 1.0% 0.5% Months-on-Book (Age) 12

13 Components of Portfolio Performance Vintage Lifecycle Seasonality 3.0% Maturation + Seasonality 2.5% 2.0% 1.5% 1.0% 0.5% Months-on-Book (Age) 13

14 Components of Portfolio Performance Vintage Lifecycle Seasonality Management Actions 3.0% Maturation + Seasonality + Policies 2.5% 2.0% 1.5% 1.0% 0.5% Months-on-Book (Age) 14

15 Components of Portfolio Performance Vintage Lifecycle Seasonality Management Actions Competitive & Economic Environment Maturation (age-based) Exogenous (time-based) 3.0% Maturation + Seasonality + Policies + Economics 2.5% 2.0% 1.5% 1.0% 0.5% Months-on-Book (Age) 15

17 Dual-time Dynamics Modeling Behavior is decomposed into natural dynamics and environmental response. Captures the full nonlinear dynamics of the components. All historical data is made relevant for the present. Past environment may be replaced with a future scenario. Lifecycle Dynamics Raw Data Scenario-based Forecast Environmental Response Scenario 17

All historical data is made relevant for the present.

18 Dual-time Dynamics (DtD) A unique technology for modeling portfolio dynamics Unparalleled accuracy for quantifying maturation and management controls Patent-pending technology derived from techniques in nonlinear dynamics OTB Utilization Rate 30% 26% 22% 18% 14% 10% 6% 2% Open-to-Buy Purchase Utilization Rate Vintage Curves Jan-99 Jul-99 Jan-00 Jul-00 Jan-01 Jul-01 Date 18% 16% 14% 12% 10% 8% 6% 4% Maturation Curve 50% Vintage Sensitivities 40% % % % % % % % Jan-97 Jan-98 Jan-99 Jan-00 Months-on-Books Vintage Exogenous Curve Jan-99 Jul-99 Dec-99 Jun-00 Dec-00 Jun-01 Date 18

Vintage Curves Jan-99 Jul-99 Jan-00 Jul-00 Jan-01 Jul-01 Date 18% 16% 14% 12% 10% 8% 6% 4% Maturation Curve 50% Vintage Sensitivities 40% 0.03 0.02 30% 0.01 20% 0.

19 Decomposing the Exogenous Curve The exogenous curve measures the relative impact of external factors upon the intrinsic consumer dynamics e.g. 20% higher utilization of unused credit line than would have been expected from the maturation process To ascertain cause-and-effect, the exogenous curve is further decomposed into seasonality, environmental trends, management actions, and intrinsic volatility 50% 40% 30% 20% 10% 0% -10% -20% -30% OTB Purchase Util Rate Exogenous Curve Jan-99 Jul-99 Dec-99 Jun-00 Dec-00 Jun-01 Date 20% Seasonality 20% Unexplained Structure Seasonal Impact 10% 0% -10% 40% 30% 20% 10% Environmental Trend 10% 0% -10% -20% Jan Feb Mar Apr May Jun Jul Month Aug Sep Oct Nov 0% Dec -10% -20% -20% Jan-99 Jul-99 Dec-99 Jun-00 Dec-00 Jun-01 Date Jan-99 Jul-99 Dec-99 Jun-00 Dec-00 Jun-01 Date 19

21 Exogenous Scenarios Decomposition generates elements that are familiar and visual Scenario s for the future environment are best managed by adjusting seasonality and economic response separately Cumulative effects then drive the forecast Seasonality 20% 15% 10% 5% 0% -5% -10% -15% Jan-99 Jul-99 Dec-99 Jun-00 Dec-00 Jun-01 Dec-01 Jun-02 Date 35% 30% 25% 20% 15% 10% 5% 0% -5% -10% -15% Environmental Trend Jan-99 Jul-99 Dec-99 Jun-00 Dec-00 Jun-01 Dec-01 Jun-02 Date 2001 holiday spending is expected and spending should grow more to be weaker than prior years slowly while in recession 21

22 Vintage Forecasting Forecasts are built up from the vintage/segment level The DtD engine extrapolates the maturation curve and calibrates to the vintage using the vintage sensitivities Utilization Rate 7% 6% 5% 4% 3% 2% 1% Maturation Extrapolation Nominal Calibrated 0% Jan-00 Jul-00 Jan-01 Jul-01 Jan-02 Jul-02 Date Open-to-Buy Purchase Utilization Rate 22

23 Vintage Forecasting Economic and management actions are then introduced The DtD engine overlays the chosen exogenous scenario and calibrates the exogenous scenario to the vintage using the vintage sensitivities 7% Exogenous Scenario Overlay Utilization Rate 6% 5% 4% 3% 2% 1% 0% Jan-00 Jul-00 Jan-01 Jul-01 Jan-02 Jul-02 Date Calibrated Nominal Open-to-Buy Purchase Utilization Rate 23

25 Predicting Score-Odds Calibrations Renormalize the vintages as if they were all the same maturity (months-on-books) during the scenario. Project the vintages under a scenario for the future environment. Using scenarios means that the predictions depend upon management assumptions, but those assumptions are made explicit and available for validation. An audit trail is maintained. Compute the score-odds calibration under this scenario and maturity. Management should consider a range of possible future scenarios to determine the portfolio s sensitivity to environmental change. 25

26 Create the Scenarios Shows three possible scenarios for the next two years. At the time, Gradual Recovery was considered the most likely Scenarios for the Environment Relative Impact Jan-1997 Jan-1998 Jan-1999 Jan-2000 Jan-2001 Jan-2002 Jan-2003 Jan-2004 Date History History Gradual History Recovery Gradual History Recovery Gradual Double Recovery dip Double Strong dip Recovery 26

27 Optimal Cut-off Scores In this example, to maintain odds of 4:1 would require raising the cut-off score by 32 (Gradual Recovery), 41 (Double-dip), or 28 (Strong Recovery). 100 Predictive Score-Odds Good / Bad Odds (Ever 60+ DPD in First 2 yrs) Originations Score Gradual Recovery Gradual Gradual Recovery Double Recovery 1999H2 dipdouble Actuals Strong dip 1999H2 Recovery Actuals 1999H2 Actuals 1999H2 Actuals 27

28 Summary Score-Odds calibrations can and should account for future changes in the environment. Predictive Score-Odds are probably most important for short term loans (less than one full business cycle) that will not experience a broad range of environments. The Predictive Score-Odds technique leads naturally to optimizing profitability. Including revenue analysis in the Score-Odds calibration makes setting cut-off scores an act of optimizing profit rather than minimizing losses. 28

AT&T Global Network Client for Windows Product Support Matrix January 29, 2015

AT&T Global Network Client for Windows Product Support Matrix January 29, 2015 Product Support Matrix Following is the Product Support Matrix for the AT&T Global Network Client. See the AT&T Global Network