Showing posts with label God. Show all posts
Showing posts with label God. Show all posts

Jun 28, 2013

Confidence Level Crisis

When you're - like me - a born professional optimist, but nevertheless sometimes worry about the unavoidable misery in the world, you ask yourself this question:

Why does God not act? 

Think about this question and try to answer it, before reading any further..

The answer to this question is very simple:

God does not act because he's conscious of everything  

The moral of this anecdote is that when you're fully aware of all the risks and their possible impact, chances are high you'll not be able to take any well-argued decision at all, as any decision will eventually fail when your objective is to rule out all possible risks.

You see, a question has come up that we can't agree on,
perhaps because we've read too many books.

Bertolt Brecht, Life of Galileo (Leben des Galilei)

On the other hand, if you're not risk-conscious at all regarding a decision to be taken, most probably you'll take the wrong decision.

'Mathematical Confident'
So this leaves us with the inevitable conclusion that in our eager to take risk-based decisions, a reasoned decision is nothing more than the somehow optimized outcome of a weighted sum of a limited number of subjective perceived risks. 'Perceived' and 'Weighted', thanks to the fact that we're unaware of certain risks, or 'filter', 'manipulate' or 'model' risks in such a way that we can be 'mathematical confident'. In other words, we've become victims of the "My calculator tells me I'm right! - Effect".

Risk Consciousness Fallacy
This way of taking risk based decisions has the 'advantage' that practice will prove it's never quite right. Implying you can gradually 'adjust' and 'improve' or 'optimize' your decision model endlessly.
Endlessly, up to the point where you've included so much new or adjusted risk sources and possible impacts, that the degrees in freedom of being able to take a 'confident' decision have become zero.

Risk & Investment Management Crisis
After a number of crises - in particular the 2008 systemic crisis - we've come to the point that we realize:
  • There are much more types of risk than we thought there would be
  • Most type of risks are nonlinear instead of linear
  • New risks are constantly 'born'
  • We'll not ever be able to identify or significantly control every possible kind of risk
  • Our current (outdated) investment model can't capture nonlinear risk
  • Most (investment) risks depend heavily on political measures and policy
  • Investment risks are more artificial and political based and driven, than statistical
  • Market Values are 'manipulable' and therefore 'artificial'
  • Risk free rates are volatile, unsure and decreasing
  • Traditional mathematical calculated 'confidence levels' fall short (model risk)
  • As Confidence Levels rise, Confidence Intervals and Value at Risk increase

One of the most basic implicit fallacies in investment modeling, is that mathematical confidence levels based on historical data are seen as 'trusted' confidence levels regarding future projections. Key point is that a confidence level (itself) is a conditional (Bayesian) probability .

Let's illustrate this in short.
A calculated model confidence level (CL) is only valid under the 'condition' that the 'Risk Structure' (e.g. mean, standard deviation, moments, etc.) of our analysed historical data set (H) that is used for modeling, is also valid in the future (F). This implies that our traditional confidence level is in fact a conditional probability : P(confidence level = x% | F=H ).

  • The (increasing) Basel III confidence level is set at P( x ∈ VaR-Confidence-Interval | F=H) = 99.9% in accordance with a one year default level of 0.1% (= 1-99,9%).
  • Now please estimate roughly the probability P(F=H), that the risk structure of the historical (asset classes and obligations) data set (H) that is used for Basel III calculations, will also be 100% valid in the near future (F).
  • Let's assume you rate this probability based on the enormous economic shifts in our economy (optimistic and independent) at P(F=H)=95% for the next year.
  • The actual unconditional confidence level now becomes P( x ∈ VaR-Confidence-Interval) = P( x ∈ VaR-Confidence-Interval | F=H) × P(F=H) = 99.9% × 95% = 94.905%
Although a lot of remarks could be made whether the above method is scientifically 100% correct, one thing is sure: traditional risk methods in combination with sky high confidence levels fall short in times of economic shifts (currency wars, economic stagnation, etc). Or in other words:

Unconditional Financial Institutions Confidence Levels will be in line with our own poor economic forecast confidence levels. 

A detailed Societe Generale (SG) report tells us that not only economic forecasts like GDP growth, but also stocks can not be forecasted by analysts.

Over the period 2000-2006 the US average 24-month forecast error is 93% (12-month: 47%). With an average 24-month forecast error of 95% (12-month: 43%), Europe doesn't do any better. Forecasts with this kind of scale of error are totally worthless.

Confidence Level Crisis
Just focusing on sky high risk confidence levels of 99.9% or more is prohibiting financial institutions to take risks that are fundamental to their existence. 'Taking Risk' is part of the core business of a financial institution. Elimination of risk will therefore kill financial institutions on the long run. One way or the other, we have to deal with this Confidence Level Crisis.

The way out
The way for financial institutions to get out of this risk paradox is to recognize, identify and examine nonlinear and systemic risks and to structure not only capital, but also assets and obligations in such a (dynamic) way that they are financial and economic 'crisis proof'. All this without being blinded by a 'one point' theoretical Confidence Level..

Actuaries, econometricians and economists can help by developing nonlinear interactive asset models that demonstrate how (much) returns and risks and strategies are interrelated in a dynamic economic environment of continuing crises.

This way boards, management and investment advisory committees are supported in their continuous decision process to add value to all stakeholders and across all assets, obligations and capital.

Calculating small default probabilities in the order of the Planck Constant (6.626 069 57 x 10-34 J.s) are useless. Only creating strategies that prevent defaults, make sense.

Let's get more confident! ;-)

- SG-Report: Mind Matters (Forecasting fails)
Are Men Overconfident Users?

Dec 3, 2010

God’s Definition of Risk

To snap things in the right perspective, now and then it's good practice to consider how actuarial science really started:

Yes, like Laplace stated in his masterwork 'Théorie Analytique des Probabilités', it all began with 'games of chance'... and - today -  perhaps it still is.....

From 'gaming', probability theory developed to 'actuarial science' and finally to 'risk management'.

Risk Levels
Today we distinguish three main types of risk levels:

Risk Level 1
In fact what we are modeling mostly, are the risks we know, the 'known risks'... These risks are the familiar operational, financial and compliance risks

Risk Level 2
These are the strategic risks. Risks related to new markets, mergers and acquisitions, investments, but also business development, brand and reputation risks.

Risk Level 3
These are the unpredictable, the so called 'unknown, unknown risks'.

The Rumsfeld definitions of risk levels
A similar more humorous, but also interesting definition of risk levels, has been given by the United States Secretary of Defense  Donald Rumsfeld  during the Iraq War:
  1. Known Knowns
    There are known knowns; there are things we know that we know
  2. Known Unknowns
    There are known unknowns; that is to say, there are things that we now know we don’t know
  3. Unknown Unknowns
    But there are also unknown unknowns; there are things we do not know we don’t know."

If we're honest, we'll have to admit that even our 'known known' and 'known unknown' risks in our models in reality have a high 'unknown unknown' origin.

Or, as Barry du Toit at Riskworx shows in an excellent paper called 'Risk, theory, reflection: Limitations of the stochastic model of uncertainty in financial risk analysis' : our stochastic model of uncertainty is powerful but limited.

It's (p.e.) an illusion to use 'standard deviation' as a stand alone measure for risk. We must be aware to apply our models without a healthy portion of 'common sense'. Or, to put it in air-plane words:

The danger inherent in 'altimeter usage' is that its unquestioning use will stop pilots from using a range of more intuitive risk measures, such as looking out of the window!

God’s definition of risk
There is no ultimate "God’s definition of risk", we'll have to manage with our limited models as a help to our Risk Insight. Success!

Sources and related links:
- Limitations of the stochastic model of uncertainty in financial risk analysis
- Laplace: analytic theory of probabilities (English)
- Strategic Management of Three Critical Levels of Risk
- Managing Projects in the Presence of Unknown Unknowns