Monitoring Systemic Risk with V-Lab - Robert Engle - June 25 2013 - First International Conference on Syrto Project:

1. Monitoring Systemic
Risk with V-LAB
SYstemic Risk TOmography:
Signals, Measurements, Transmission Channels,
and Policy Interventions
Robert Engle
Director of Volatility Institute
At NYU Stern
Brescia, 25 June 2013

2.  The Volatility Laboratory or V-LAB is a public website
with a wide range of risk measures that are updated
daily.
 The site is a laboratory to determine the accuracy and
reliability of cutting edge statistical models of financial
risk.
 The site is designed to bring new academic
developments to practitioners and regulators in a
timely way.
 The Volatility Institute is supported by generous
funding from the Sloan Foundation, Banque de France,
Armellino Foundation, Deutsche Bank, BlackRock, and
our collaborators UNSW and Universite de Lausanne

3.  Or Google vlab
 There are four main regions on V-LAB
 Volatility
 Correlation
 Systemic Risk
 Long Run Risk
 Lets take a look at each before turning to
systemic risk.

14.  VaR is the 1% quantile of returns for some asset
one day in the future. It is a very widely used
measure of risk but is clearly a short run
measure of risk.
 What is the VaR one month or one year from
now? How much can some asset decline in a
year?
 Measuring “the risk that the risk can change”
 Use GARCH and options to forecast quantiles.

17.  In the global financial crisis of 2007-2009, we
learned that failure of the financial system
could have massive effects on the real
economy, potentially for years to come.
 The sovereign debt crisis in Europe is a
reminder that even though the causes may
appear different, financial failure has real
effects.
 How can we forsee and prevent these events?

18. 6/26/2013 VOLATILITY INSTITUTE 18

19.  We saw that the Lehman bankruptcy began the
worst episode of the global financial crisis. But
it coincided with deep distress at FANNY and
FREDDIE, Citi, Bank of America, Merrill
Lynch, Goldman Sachs, Morgan Stanley, AIG,
WAMU, Wachovia, and many more.
 Lehman was not a domino – this was a tsunami.
Weakness of the financial sector as a whole is
the most important determinant of a financial
crisis. It is not enough that one bank is
undercapitalized.

20.  How much capital would a financial institution need to raise
in order to function normally if we have another financial
crisis?
 We measure this econometrically based on market data
on equities and balance sheet data on liabilities. We
update weekly on V-LAB for US and Global financial
firms. We call this SRISK.
 Principle investigators: Viral Acharya, Matt Richardson and me at the
Volatility Institute at NYU’s Stern School. Collaboration with HEC Lausanne
and the Institute for Global Finance at University of New South Wales.
Contributions by Christian Brownlees, Rob Capellini, Diane Perriet, Emil
Siriwardane.
 References: Acharya, Pedersen, Phillipon, Richardson “Measuring Systemic
Risk (2010); Acharya, Engle, Richardson “Capital Shortfall, A New Approach
to Ranking and Regulating Systemic Risks, AEAPP (2012), Brownlees and
Engle, “Volatilities, Correlations and Tails for Systemic Risk
Measurement”,2010

21.  Regulators measure this based on supervisory data
and stress scenarios.
 Many measures based on public data are being
developed. See the surveys by Brunnermeier and
Oehmke and by Bisias, Flood, Lo and Valvanis
 Some of these measures are firm specific such as
CoVaR and SRISK. Others are financial industry
quality measures such as volatility or liquidity.
 Some measures are based on a network model and
others such as SRISK are based on a Tsunami
model.
6/26/2013 VOLATILITY INSTITUTE 21

22.  SRISK is computed from:
 Where k is a prudential level of equity relative to
assets taken to be 8% (and 5.5% for IFRS firms)
and LRMES is the decline in equity values to be
expected if there is another financial crisis.
 SRISK depends upon size, leverage and risk.
 
  
  
,
, ,1 1
i t t i
t t n t n t n t n
t i t i t
SRISK E Capital Shortfall Crisis
E k Debt Equity Equity Crisis
kDebt k LRMES Equity
   

  
   

23.  Bank of America has a market cap of $141billion. Its
accounting liabilities are $2.0 trillion for a leverage
ratio of 14.9
 If we have another financial crisis which is assumed to
be a fall of 40% in broad US equities over six months,
then we estimate shares in BAC will fall by 50%.
 This reflects a Dynamic Conditional Beta of 1.0 today
that will move in the future due to mean reversion in
volatilities and correlations and also will rise with
downside returns.
 SRISK = $91 billion.
 It is undercapitalized somewhat today and this will be more
severe under the stress of an equity decline.

24.  Credit Agricole has a market cap of $23 billion
 It has liabilities of $2.2 trillion for a leverage
ratio of 102.
 Any fluctuation in asset or liability valuations
can easily move the firm into bankruptcy.
 Most of the capital shortfall is needed to bring
the leverage down now. The risk is only a
small part of the capital shortfall calculation.
 Most likely, Credit Agricole is no longer
making loans except possibly the most secure.

25.  If any firms have high SRISK, they will
recognize their vulnerability and will begin to
delever and derisk by selling assets and
making fewer loans.
 If many firms have high SRISK, this will
impact the real economy.
 As the macro economy slows, stock prices will
fall, volatility will rise, and SRISK will go up
more.

26.  Investors recognize financial institution
weakness and lower valuations, increasing
SRISK
 Forward looking investors could make this
happen in one step.
 Bankruptcies and other failures will occur until
eventually, the return to capital is high enough
to bring new capital to the industry.

27.  The spiral can be arrested before the bottom
with taxpayer money.
 However, this will erode market discipline and
may impose huge regulatory costs on the
financial sector going forward.
 Thus regulation is needed in advance. Ideally
it would be countercyclical.

28. 6/26/2013 VOLATILITY INSTITUTE 28

29.  Externalities – if only one firm has high SRISK,
there is no spiral. Competition!
 Implicit and Explicit government guarantees
such as deposit insurance or “too big to fail”
assurances
 Regulatory incentives that encourage
investments that might not be prudent.
 Miscalculation

30. 6/26/2013 VOLATILITY INSTITUTE 30
DYNAMIC CONDITIONAL BETA

31. DYNAMIC CONDITIONAL BETA

32. ARE BETAS CONSTANT?
 LEAST SQUARES MODELS ARE USED IN
COUNTLESS EMPIRICAL STUDIES IN FINANCE
AND ECONOMICS
 RARELY IS THE HYPOTHESIS THAT BETAS ARE
CONSTANT GIVEN CAREFUL SCRUTINY
 WHAT TOOLS DO WE HAVE?

33. MODELLING TIME VARYING BETA
 ROLLING REGRESSION
 INTERACTING VARIABLES WITH TRENDS, SPLINES
OR OTHER OBSERVABLES
 TIME VARYING PARAMETER MODELS BASED ON
KALMAN FILTER
 STRUCTURAL BREAK AND REGIME SWITCHING
MODELS
 EACH OF THESE SPECIFIES CLASSES OF
PARAMETER EVOLUTION THAT MAY NOT BE
CONSISTENT WITH ECONOMIC THINKING OR DATA.

34. THE BASIC IDEA
 IF is a collection of k+1
random variables that are distributed as
 Then
 Hence:
 , , 1,...,t ty x t T
  , ,,
1
, ,,
~ , ,
yy t yx ty tt
t t t
xy t xx tx tt
H Hy
N H N
H Hx



    
             
F
  1 1
1 , , , , , , , ,, ~ ,t t t y t yx t xx t t x t yy t yx t xx t xy ty x N H H x H H H H  
   F
1
, ,t xx t xy tH H 


35. HOW TO ESTIMATE H
 Econometricians have developed a wide range of
approaches to estimating large covariance
matrices. These include
 Multivariate GARCH models such as VEC and BEKK
 Constant Conditional Correlation models
 Dynamic Conditional Correlation models
 Dynamic Equicorrelation models
 Multivariate Stochastic Volatility Models
 Many many more
 Exponential Smoothing with prespecified
smoothing parameter.

36. IS BETA CONSTANT?
 For none of these methods will beta appear
constant.
 In the one regressor case this requires the ratio
of to be constant.
 This is a non-nested hypothesis (or more
technically a partially nested hypothesis)
, ,/yx t xx th h

37. ARTIFICIAL NESTING
 Consider the model where means element by
element multiplication or Hadamard product:
 If lambda is zero, the parameters are constant
 If phi is zero, the parameters are time varying.
 If both are non-zero, the nested model may be
entertained.
 ' 't t t t ty x x v    
o

39. A. CALDER 1973, LA PORTE DE L’ESPACE

41.  Combining the constant beta and dynamic
conditional beta into one regression:
 Where u will be an MA(1) GARCH
 Updated weekly for 1200 Global financial
institutions
 Results on V-LAB
   , 1 1 , , 2 2 , , 1i t i t m t i t m t tR R R u          

42. 6/26/2013 VOLATILITY INSTITUTE 42

43. 6/26/2013 VOLATILITY INSTITUTE 43

44. 6/26/2013 VOLATILITY INSTITUTE 44

45. 6/26/2013 VOLATILITY INSTITUTE 45

46. 6/26/2013 VOLATILITY INSTITUTE 46

50. 6/26/2013 VOLATILITY INSTITUTE 50

51. 6/26/2013 VOLATILITY INSTITUTE 51

52. 6/26/2013 VOLATILITY INSTITUTE 52

55. How can you validate a
systemic risk model?
Do we get advanced warning
of systemic events?
6/26/2013 VOLATILITY INSTITUTE 55

56. “A Look Back”

58. 6/26/2013 VOLATILITY INSTITUTE 58

62.  A small piece of evidence.
 Monthly SRISK, calculated recursively at the
end of each month and summed over all US
financial institutions.
 Tested with monthly industrial production and
unemployment.
 All variables log differenced, 3 lags of all
variables, OLS estimation

63. SRISK INDPRD URATE
SRISK 4.31 0.02
INDPRD 20.65*** 8.99**
URATE 0.01 7.39***

64.  November 4, 2011 BIS with FSB of the G-20
released its list of Global Systemically
Important Financial Institutions GSIFIs.
 They listed 17 European Banks
 November, our list of the top 17 banks is
identical with one exception:
 We have Intesa Sanpaolo instead of Dexia
 Furthermore, we have ranked these
 It took BIS two years and many meetings. We
have now updated many times.

65.  See Acharya, Engle, Pierret ”Testing
Macroprudential Stress Tests: The Risk of
Regulatory Risk Weights” , To be presented at
Carnegie Rochester Conference in October.
 Findings: both European and US FED stress tests
have rankings of firms similar to VLAB when
measured relative to total assets. When measured
relative to risk weighted assets, they are not
similar.
 VLAB has slightly better forecast performance.
 Risk weighted measures are negatively correlated
with outcomes and risk.

66. 6/26/2013 VOLATILITY INSTITUTE 66

67. This project is funded by
the European Union under the
7th Framework Programme (FP7-SSH/2007-2013)
Grant Agreement n°320270
!
!
!
!
!
!
!
www.syrtoproject.eu
This document reflects only the author’s views. The European Union is not liable for any use that may be made of the information contained therein.