Predictive Intelligence for Hunting Down Financial Crisis Tipping Points
The New Risk Management Framework after the 2008 Financial Crisis
1. Prepared for Presentation 13 July 2010 GARP, London Chapter by Barry Schachter The New Risk Management Framework After the 2008 Financial Crisis
2. “Those who cannot remember the past are condemned to repeat it.” Santayana, The Life of Reason (vol. 1) “What experience and history teaches us is that nations and governments have never learned anything from history, or acted on principles deduced from it.” Hegel, Lectures on the Philosophy of History Learning Lessons
4. These aren’t bugs, they’re features e.g., mis-measurement of known risks – correlation failure to take risks into account – liquidity A case of 20-20 hindsight bias e.g., over-reliance on rating agencies failure to analyze crisis scenario risks The questions asked are too superficial – failure to question the appropriateness of the paradigm What’s Wrong with Received Wisdom
5. I agree with Hoyle, “…human behavior is controlled by an interlocking system of nonlinear feedback loops... It is far beyond present-day resources to take full and accurate account of all the feedback effects that occur in human society. (Man and Materialism (1956)) Rational economic behavior is not what is assumed in neo-classical economics. Stable and static equilibria do not describe the world in which we live. Risk Management can move forward only by adopting a network view of markets with fallible, but strategically behaving risk takers. A Personal View of Lessons Learned
6. Attribution of effects to causes is intrinsic to us; it is how we build our model of the environment, and how we devise rules-of-thumb behavioral responses Assigning causes is also how we maintain a sense of control of what is really a very complex and confusing world There is a dark side of this in which we grasp at causes, even if they are fictions, and proceed to “address” them We have to guard against simple responses to complex problems, because they may be ineffective and have unintended negative consequences This applies to how we reinvent risk management in light of the crisis A Caution on Causation
7. Weare (daily) data junkies – Can’t get enough Financial data isn’t IID, requires transformation or estimation of conditional moments of the generating process We make up data for regulatory historical stress scenarios – we construct shocks for instruments that did not exist (ERM, Oct 1987) We extrapolate even though we know extrapolation is Evil The Data “Problem” Has No Fix
8. “scenarios most likely to cause their current business model to become unviable” define all “ruin scenarios”, pick plausible ones; or define plausible scenarios, find the “ruin scenarios” as a subset Maximum Loss first developed by Studer (1997) Problem may be ill-posed and ill-conditioned, depending on the actual distributions of risk factor and portfolio returns Estimation may be computationally daunting Innovation: Reverse Stress Testing
9. Finding portfolio vulnerabilities through stress testing is an ill-conditioned problem (small changes in the shocks may result in large changes in the stress loss) It is also an ill-posed problem (small changes in the stress loss may be associated with very different shock scenarios) Whereas standard stress tests are arbitrary, in reverse stress tests results dependent on data (period) and fitted distribution. But Reverse Stress Tests No Panacea
10. Stressed VaR: 10-day, 99th percentile...calibrated to...a period of significant financial stress relevant to the firm’s portfolio. For most portfolios, the [Basel] Committee would consider a 12-month period...in 2007/2008 to be a period of such stress.”http://bit.ly/cQbmoZ Chichilnisky (2010) proposes a subjective probability measure to increase the probability weight of rare events. (Also, Algorithmics approach)http://bit.ly/dfyEYx What to do When Events are too Rare
11. George Buffon corresponded (1760) with Daniel Bernoulli on event probabilities small enough that could be disregarded in evaluating decisions. Buffon called events with probability less than 1/10,000 a “moral certainty”. Bernoulli suggested 1/100,000. The History of Statistics in the 17th and 18th Centuries, E. S. Pearson, ed. (New York: MacMillan Publishing Co.), 1978, p. 193. So, 1 time in 27.4 years or perhaps 1 time in 274 years. How to Decide the Smallest Probability Worth Measuring for Risk?
12. In the US, something called the “Kanjorski Amendment” gives Systemic Risk Council power to break up banks that pose “a grave threat to the financial stability or economy” (http://bit.ly/9ex0ai). Create need to monitor own firm systemic risk contribution (as perceived by the regulator), e.g., using CoVaR model (http://bit.ly/cTqJRv). Possibly alter business decisions which could result in adverse impact on systemic risk contribution. New Need for Systemically Risky Risk Measuranagement
13. Persaud (2002) and Morris and Shin (2003) argue that market-sensitive risk regulation might reduce stability by reducing diversity. In an evolutionary context, diversity is key to species survival (I’ll return to this) Some new regulatory changes might affect diversity, e.g., the widespread use of Stressed VaR, based on identical 12-month period...in 2007/2008. Aside: Convergence as a Systemic Risk
14. Financial Risk Management is atomistic Possible Exceptions to this statement Liquidity risk Stress testing counterparty default Positive Feedback and Contagion are aspects of a networked system Positive Feedback is outside the Risk Management Paradigm
15. In an evolutionary biology view, risk management as an adaptation for enhancing survival of an “agent”. Agents are risk taking entities, competing for success in financial markets by adopting different trading and risk management strategies Market shocks threaten survival of agents and stability of the system (via contagion) Risk Management adaptations that increase survivability are selected and propagate through subsequent “generations” The Networked Economy and Risk Management
16. Consider N Hedge Funds, with capital allocated in each fund equally between 2 Traders, each taking long or short positions in 2 of 4 possible markets. Traders submit demand schedules, markets must clear Prices are then shocked Traders cheer/cry, then re-compute their wealth Assume that a trader “dies” if wealth falls from peak by more than 20% or if combined Fund resources fall by 15% In each case there is a liquidation of positions A Financial Market Network Model
17. In liquidation the demand schedule submitted has 0 elasticity. Liquidation affects prices in every market in which a position is held. Depending on the demand schedules and wealth, the liquidation may have spillover effects. Note that the liquidation is driven not by bankruptcy, but by a risk control mechanism - funding. Further connections may arise via other mechanisms – e.g., VaR limits, the key is feedback and contagion. Fragility in the Network
18. Illiquidity Crowded Trades Hidden Correlations New Approaches to Risk Control New Risk Measurement and Management Directions
19. Concepts “Small” positions, when traded, induce “large” market impact costs (total market impact affected by aid/ask spread, return volatility, volume) Liquidity black hole – where price movements trigger additional trading in a positive feedback loop resulting in large price changes Liquidity black holes are not specifically about calculating liquidity-adjusted VaR (see L-VaR survey by Ernst, Stange and Kaserer (2009))http://bit.ly/9dRpDy Illiquidity
20. Qualitative detection Quantitative detection – Normal illiquidity Exchange-traded Position as % of ADV Total cost to exit (e.g., Almgren and Chriss (2000))http://bit.ly/acx9hi OTC Quantitative detection – Black Holes Brunnermeier and Pedersen (2008) generate black holes from the feedback between funding and market liquidity through margin requirementshttp://bit.ly/9oHisw Illiquidity
21. Definition by CRMPG-II (2005): Multiple parties entering into correlated trading strategies...where the aggregate volume of trades is sufficient to constrain the ability of traders to exit from the position on a simultaneous basis; A crowded trade is characterized by dampening volatilities, increased measured liquidity, and decreased “spreads”. (http://bt.ly/9S7HRU) Also seen in RV type crowded trades (e.g., DAX v. STOXX) higher than normal correlations between instruments/markets Crowded Trades
22. Qualitative detection Quantitative detection Pojarliev and Levich (2009) use returns-based style analysis for a (22-60) hedge fund sample and 4 currency trading styles (trend, carry, volatility, and value), and compute the difference between the number of funds with positive and negative statistically significant style exposures. Pericoli and Sbracia (2010) examine the pairwise median correlation among hedge fund returns after extracting common macro factors. Crowded Trades
23. The network model suggests that a “phase transition” reveals hidden correlations Risk management should seek out signs of hidden correlations (e.g., crowded trades, or risk control mechanisms) No current methods for systematically investigating this at the individual firm (risk manager) level Hidden Correlations
24. It seems likely that the new risk management framework will be defined by the old risk management paradigm Improvements within the old paradigm are possible, but represent a missed opportunity to re-think risk management Within a view of markets as a dynamic network and risk takers as fallible but strategic agents the focus of risk management shifts and the design of risk control mechanisms may take a new direction. Summary
25. Golub, B. and Crum, C., Risk Management Lessons Worth Remembering from the Credit Crisis of 2007–2009, The Journal of Portfolio Management, Spring 2010, Vol. 36, No. 3: pp. 21-44 http://dx.doi.org/10.3905/JPM.2010.36.3.021 Stulz, R., Risk Management Failures: What Are They and When Do They Happen? (October 2008)http://bit.ly/b9r3CP Senior Supervisors Group, Observations on Risk Management Practices during the Recent Market Turbulence (March 6, 2008)http://bit.ly/agzD9x Rosengren, E., Risk-Management Lessons from Recent Financial Turmoil (May 14, 2008)http://bit.ly/c8H1Og References
26. Jorion, P., Risk Management Lessons from the Crisis (June 2008)http://bit.ly/azljSx President's Working Group on Financial Markets, Policy Statement on Financial Market Developments (March 2008)http://bit.ly/aI9FNG Alexander, C., The Present and Future of Risk Management (February 2004)http://bit.ly/a3Iy2R McNeil, A. and Smith, A., Multivariate Stress Testing for Solvency II (March 2010)http://bit.ly/cgMz5R Studer, G. Maximum Loss for Measurement of Market Risk (1997)http://bit.ly/deZ5Sa (More) References
27. Pojarliev, M. and Levich, R., Detecting Crowded Trades in Currency Funds (December 2009).http://bit.ly/cp1TNr Pericoli, M. and Sbracia, M., Crowded Trades among Hedge Funds (May 2010).http://bit.ly/balwis (Still More) References
Notas do Editor
Choquet Expected Utility and Prospect Theory models also give higher weight to small probabilities. Decisions from experience – underweight rare events; from description (Kahnemann & Tversky) overweight rare events.