Consistent projections of balance
sheet, risk-weighted assets, and
income
Model Symposium
June 2013
Anna Kovner
Federal Reserve Bank of NY
The views expressed in this presentation are those of the author and do not necessarily reflect the position of the
Federal Reserve Bank of New York or the Federal Reserve System.
Consistency is hard to achieve
Change
in asset
mix
Change
in asset
quality
Net
income
Changes
to
Changes
capital
to capital
Provisions
Change
in asset
amount
Changes
to RWA
Capital
Ratio
2
Even before projections need to vary with
changes in macroeconomic conditions
GDP
Changes
to Assets
Unemploy
ment
Stock
market
Revenue
Changes
to RWA
Provisions
Changes
to capital
Housing
3
But consistency is critical
• If there is no consistency across models, the sum
of the parts may be very different from what
would occur in an adverse macroeconomic
environment
– Do riskiest borrowers refinance? Are credit standards
tightening?
– Does reducing risk of trading assets impair market
making revenues?
– Is the expectation for interest income in the
AFS/Securities portfolio similar to the expectation for
interest income when the same assets are directly
held?
• If model errors are correlated, the capital forecast
will be biased
4
Federal Reserve Approach I
Revenue
Models
Capital
Calculations
Model
Oversight
Group
Asset
Models
Loss Models
5
Federal Reserve Approach II
Model Oversight
Group
Modeling team
Model Validation
Unit
Benchmark Models
External Model
Validation Council
•Single group reviews all models
•Encourages consistency
•Eg. Connects modeling of MBS exposures in securities book to on
balance real estate exposures
•Content experts
•Encourages robust processes
•Alternate approaches to same question (eg. top down vs. bottom up,
different specifications) can reveal inconsistencies
Consistent,
robust
models of
capital
•Insight from outside of Federal Reserve System on modeling
approaches
6
Consistency across multiple
dimensions
Consistent data:
• 18 CCAR BHCs, next including the CAPR firms
• Pro forma for acquisitions where practical
Consistent principles:
• Flight to quality is not consistent with the spirit of
a stress test
• Revenues, expenses, assets, and risk weighted
assets are linked
– higher revenues come at a cost; expense cuts may not
materialize as planned; growth strategies may not be
deployed as planned
• Unexpected negative events may occur
7
Adapting a budgeting methodology to
forecasting a stress case can be difficult
A
B
C
Balance Reference rate Spread
Borrower A
100
Borrower B
125
Borrower C
100
200
LIBOR
Borrower D
500
PRIME
Borrower E
300
Borrower F
100
PRIME
Borrower G
20
Borrower H
30
Borrower I
Total
100
$1,475
175
PRIME
50
D
Rate
Interest
income
B+C
AXD
900
AXD
50
75
B+C
AXD
50
B+C
AXD
800
60
AXD
50
B+C
AXD
EURIBOR
75
B+C
AXD
PRIME
50
B+C
AXD
800
AXD
8
Balances / Revenues / Provisions
• Severely adverse scenario: Increase in price of risk (BBB
spreads), Decrease in LT treasury rates, ST rates remain at
lower bound
• What might happen to loans?
– Interest income:
• Constant / lower interest income from OLD floating rate loans
• Assuming constant portfolio risk (including new originations and
existing borrower renegotiations)  Higher interest income
• Assuming risk decreases  Higher or lower interest income
– Loan balances:
• Are same borrowers more or less risky? (given adverse macro
environment)
• How are maturing and defaulting loans replaced? (if at all)
• How do new originations compare to the existing portfolio?
• While next slides show what happened historically, just
because we see it 2007-8, doesn’t mean it is in the spirit of
a stress test
9
Balances / Revenues / Provisions
• In aggregate, when price of risk increases, loan balances stop growing and
change in price of risk of loan portfolio dominates de-risking in the loan
portfolio (coefficient on BBB spread is positive for interest income)
5
9
4.5
8
4
7
3.5
6
3
5
2.5
4
2
3
1.5
2
1
1
0.5
0
0
Interest Income on Loans / Total Loans (Annualized %)
LT Treasury (%)
Total Loans ($T)
Dollars (Trillions)
Percent
10
BBB Bond Yield - LT Treasury (%)
ST Treasury (%)
Source: Pro forma adjusted Y-9C data for sum of 18 CCAR firms. Includes firms only as they become Y-9C filers.
10
Compensation
• Compensation to assets has been trending down,
but is not nearly as variable as the stock market
9.8
25
20
9.6
15
10
9.2
5
Log Value
Percent
9.4
9
0
8.8
-5
-10
8.6
Growth rate of Assets (%)
Compensation/Assets (Annualized %)
Log Dow
Source: Pro forma adjusted Y-9C data for sum of 18 CCAR firms. Includes firms only as they become Y-9C filers.
Log Real GDP
11
Deposits
• Deposit rates vary negatively with balances –
even after controlling for the fall in ST rates and
even in crisis period
4.5
6
4
5
3
Percent
4
2.5
3
2
1.5
2
Dollars (Trillions)
3.5
1
1
0.5
0
0
Interest expense on deposits/Total Interest Bearing deposits (Annualized %)
Service charges on deposits / Total domestic deposits (Annualized %)
ST Treasury (%)
Total Interest Bearing Deposits ($T)
12
Models vs. Expert Judgment
• Hard to evaluate if expert judgment is consistent
across models (even if the same expert is doing
everything)
• But, unvalidated models may be no better than
expert judgment
– Would we rather be right or consistently, replicably
wrong?
BUT
• Experts can be wrong and have been
• Understand how and when different methods
produce different answers
13
Challenges and trade-offs
• Consistency requires additional reviews of all
models
Pros
• Deeper understanding of implicit and explicit
assumptions
• Challenges and testing of assumptions across
models
Cons
• Benefits of expert judgment likely diminished
• Hard to translate some implicit assumptions
• Time and resource intensive
14
A foolish consistency is the hobgoblin of little
minds, adored by little statesmen and
philosophers and divines.
Ralph Waldo Emerson
15