Ayasdi - Creating Effective Revenue Forecast Models for CCAR-PDF ...
Creating Effective Revenue Forecast Models for CCAR Michael Woods, Principal Data Scientist March 25, 2015
Complexity is the Challenge
# of possible models created from the Dataset exceeds two trillion
# of possible models created from the Dataset next year after it grows another 40% exceeds eight trillion
Dataset Company Confidential & Proprietary
2
Inadequate Response to Complexity Quants
Conventional Analytical Approaches
Overfitting y
x
Company Confidential & Proprietary
3
Clean Slate Requirements
Speed
Company Confidential & Proprietary
Accuracy
Defensibility
4
A Man-Machine Workflow
Data
Algorithms + Compute
Variable Identification
Group of Variables
Business Input
Variable Selection
Statistical Tests
Models
Model Identification
Business Validation
Model Selection
Machines
Company Confidential & Proprietary
5
Machine Intelligence
Scalable Compute
Machine Learning, Geometric + Statistical Algorithms
+
Company Confidential & Proprietary
Topological Data Analysis
+
6
Transforming Data to Insight
Fed Macro Variables ~300 Company Confidential & Proprietary
Transforms ~900
Lagged ~2700
Distillable Model
7
Correlate Variables with Business Performance
Correlation with Revenue High Company Confidential & Proprietary
Low 8
EurozoneUnemployment 20-Bond GO Index_Lag_1
UKGB10Yr
Corr AbsCorr 0.78 0.78 0.71 0.71 0.67 0.67 -0.66 0.66 0.62 0.62 0.51 0.51
B Group B
Eurozone Interest Rates Eurozonei1MLibor UKi1MLibor Eurozonei5YrSwap
Macro Growth EurozoneGDP_Index UKGDP_Index USUnemployment UKHousing_Lag_2
Corr AbsCorr 0.66 0.66 0.59 0.59 -0.57 0.57 0.47 0.47
A
Group C
C
E Funding Stress TED LIBOR less OIS LIBOR_Less_FED Eurozonei1MLibor_delta
Corr AbsCorr 0.85 0.85 0.76 0.76 0.67 0.67 -0.34 0.34
Company Confidential & Proprietary
Emerging Risk ASIAHangseng MexicoFxrate EMEAMXEMEA_Lag_2
UKFxrate
Corr AbsCorr -0.67 0.67 -0.51 0.51 0.49 0.49 -0.48 0.48
D Group D
Group E
Group A
Identify Potentially Explanatory Variables
Credit Spreads USCDS_US10YR_BBB_pctChange USCDXnonIG_pctChange USCDS_US10YR_A_pctChange USCDS_US25YR_BBB_pctChange USCDXIG_pctChange
Corr AbsCorr 0.70 0.70 0.70 0.70 0.68 0.68 0.65 0.65 0.58 0.58
9
EurozoneUnemployment 20-Bond GO Index_Lag_1
UKGB10Yr
Corr AbsCorr 0.78 0.78 0.71 0.71 0.67 0.67 -0.66 0.66 0.62 0.62 0.51 0.51
B Group B
Eurozone Interest Rates Eurozonei1MLibor UKi1MLibor Eurozonei5YrSwap
Macro Growth EurozoneGDP_Index UKGDP_Index USUnemployment UKHousing_Lag_2
Corr AbsCorr 0.66 0.66 0.59 0.59 -0.57 0.57 0.47 0.47
A
E Funding Stress TED LIBOR less OIS LIBOR_Less_FED Eurozonei1MLibor_delta
Corr AbsCorr 0.85 0.85 0.76 0.76 0.67 0.67 -0.34 0.34
Company Confidential & Proprietary
D Group D
Group E
Group A
Business Selects Variables
AbsCorr Credit Spreads Corr AbsCorr 0.70 USCDS_US10YR_BBB_pctChange 0.70 0.70 0.70 USCDXnonIG_pctChange 0.70 0.70 0.68 USCDS_US10YR_A_pctChange 0.68 0.68 0.65 USCDS_US25YR_BBB_pctChange 0.65 0.65 0.58 USCDXIG_pctChange 0.58 0.58
10
Generate Permutation Group of Models
Company Confidential & Proprietary
11
Select for Empirical Validity
Company Confidential & Proprietary
12
Business Selects Final Model
Company Confidential & Proprietary
13
A Man-Machine Workflow
Data
Algorithms + Compute
Variable Identification
Group of Variables
Business Input
Variable Selection
Statistical Tests
Models
Model Identification
Business Validation
Model Selection
Machines
Company Confidential & Proprietary
14
Corporate Headquarters 4400 Bohannon Drive Suite #200 Menlo Park, CA 94025 ayasdi.com
Complexity is the Challenge
# of possible models created from the Dataset exceeds two trillion
# of possible models created from the Dataset next year after it grows another 40% exceeds eight trillion
Dataset Company Confidential & Proprietary
2
Inadequate Response to Complexity Quants
Conventional Analytical Approaches
Overfitting y
x
Company Confidential & Proprietary
3
Clean Slate Requirements
Speed
Company Confidential & Proprietary
Accuracy
Defensibility
4
A Man-Machine Workflow
Data
Algorithms + Compute
Variable Identification
Group of Variables
Business Input
Variable Selection
Statistical Tests
Models
Model Identification
Business Validation
Model Selection
Machines
Company Confidential & Proprietary
5
Machine Intelligence
Scalable Compute
Machine Learning, Geometric + Statistical Algorithms
+
Company Confidential & Proprietary
Topological Data Analysis
+
6
Transforming Data to Insight
Fed Macro Variables ~300 Company Confidential & Proprietary
Transforms ~900
Lagged ~2700
Distillable Model
7
Correlate Variables with Business Performance
Correlation with Revenue High Company Confidential & Proprietary
Low 8
EurozoneUnemployment 20-Bond GO Index_Lag_1
UKGB10Yr
Corr AbsCorr 0.78 0.78 0.71 0.71 0.67 0.67 -0.66 0.66 0.62 0.62 0.51 0.51
B Group B
Eurozone Interest Rates Eurozonei1MLibor UKi1MLibor Eurozonei5YrSwap
Macro Growth EurozoneGDP_Index UKGDP_Index USUnemployment UKHousing_Lag_2
Corr AbsCorr 0.66 0.66 0.59 0.59 -0.57 0.57 0.47 0.47
A
Group C
C
E Funding Stress TED LIBOR less OIS LIBOR_Less_FED Eurozonei1MLibor_delta
Corr AbsCorr 0.85 0.85 0.76 0.76 0.67 0.67 -0.34 0.34
Company Confidential & Proprietary
Emerging Risk ASIAHangseng MexicoFxrate EMEAMXEMEA_Lag_2
UKFxrate
Corr AbsCorr -0.67 0.67 -0.51 0.51 0.49 0.49 -0.48 0.48
D Group D
Group E
Group A
Identify Potentially Explanatory Variables
Credit Spreads USCDS_US10YR_BBB_pctChange USCDXnonIG_pctChange USCDS_US10YR_A_pctChange USCDS_US25YR_BBB_pctChange USCDXIG_pctChange
Corr AbsCorr 0.70 0.70 0.70 0.70 0.68 0.68 0.65 0.65 0.58 0.58
9
EurozoneUnemployment 20-Bond GO Index_Lag_1
UKGB10Yr
Corr AbsCorr 0.78 0.78 0.71 0.71 0.67 0.67 -0.66 0.66 0.62 0.62 0.51 0.51
B Group B
Eurozone Interest Rates Eurozonei1MLibor UKi1MLibor Eurozonei5YrSwap
Macro Growth EurozoneGDP_Index UKGDP_Index USUnemployment UKHousing_Lag_2
Corr AbsCorr 0.66 0.66 0.59 0.59 -0.57 0.57 0.47 0.47
A
E Funding Stress TED LIBOR less OIS LIBOR_Less_FED Eurozonei1MLibor_delta
Corr AbsCorr 0.85 0.85 0.76 0.76 0.67 0.67 -0.34 0.34
Company Confidential & Proprietary
D Group D
Group E
Group A
Business Selects Variables
AbsCorr Credit Spreads Corr AbsCorr 0.70 USCDS_US10YR_BBB_pctChange 0.70 0.70 0.70 USCDXnonIG_pctChange 0.70 0.70 0.68 USCDS_US10YR_A_pctChange 0.68 0.68 0.65 USCDS_US25YR_BBB_pctChange 0.65 0.65 0.58 USCDXIG_pctChange 0.58 0.58
10
Generate Permutation Group of Models
Company Confidential & Proprietary
11
Select for Empirical Validity
Company Confidential & Proprietary
12
Business Selects Final Model
Company Confidential & Proprietary
13
A Man-Machine Workflow
Data
Algorithms + Compute
Variable Identification
Group of Variables
Business Input
Variable Selection
Statistical Tests
Models
Model Identification
Business Validation
Model Selection
Machines
Company Confidential & Proprietary
14
Corporate Headquarters 4400 Bohannon Drive Suite #200 Menlo Park, CA 94025 ayasdi.com
