putting the power of modern applied stochastics into dfa
DESCRIPTION
Putting the Power of Modern Applied Stochastics into DFA. Peter Blum 1)2) , Michel Dacorogna 2) , Paul Embrechts 1). 1) ETH Zurich Department of Mathematics CH-8092 Zurich (Switzerland) www.math.ethz.ch/finance. 2) Zurich Insurance Company Reinsurance CH-8022 Zurich (Switzerland) - PowerPoint PPT PresentationTRANSCRIPT
Putting the Power of Modern Applied
Stochastics into DFAPeter Blum 1)2) , Michel Dacorogna 2), Paul
Embrechts 1)
1)
ETH Zurich
Department of
Mathematics
CH-8092 Zurich
(Switzerland)
www.math.ethz.ch/
finance
2)
Zurich Insurance
Company
Reinsurance
CH-8022 Zurich
(Switzerland)
www.zurichre.com
Situation and intention
Applied stochastics provide lots of models that lend
themselves to use in DFA scenario generation:
=> Opportunity to take profit of advanced research.
However, DFA poses some very specific
requirements that are not necessarily met by a
given model.
=> Risk when using models uncritically.
Goal: provide some guidance on how to (re)use
stochastic models in DFA.
Topics
Observations on the use of models from
mathematical finance (one discipline of
applied stochastics) in DFA
Updates on the modelling of rare and
extreme events (multivariate data and time
series)
Annotated bibliography
DFA & Mathematical Finance: Situation DFA scenario generation requires models for
economy and assets: interest rates, stock
markets, inflation, etc.
Mathematical finance provides many such
models that can be used in DFA.
However, care must be taken because of
some particularities related to DFA.
Hereafter: some reflections...
Mathematical Finance: Background Most models in mathematical finance were
developed for derivatives valuation. Fundamental paradigms here:
No – arbitrageRisk – neutral valuation
Most models apply to one single risk factor; truly multivariate asset models are rare.
Most models are based on Gaussian distribution or Brownian Motion for the sake of tractability. (However: upcoming trend towards more advanced concepts.)
Excursion: the principle of no-arbitrage „In an efficient, liquid financial market, it is not
possible to make a profit without risk.“
No-arbitrage can be given a rigorous mathematical
formulation (assuming efficient markets).
Asset models for derivative valuation are such that
they are formally arbitrage-free.
However, real markets have imperfections; i.e.
formally arbitrage-free models are often hard to fit
to real-world data.
Excursion: risk-neutral valuation In a no-arbitrage environment, the price of a
derivative security is the conditional expectation of its terminal value under the risk-neutral probability measure.
Risk neutral measure: probability measure under which the asset price process is a martingale.
Risk-neutral measure is different from the real-world probability measure: different probabilities for events.
Many models designed such that they yield explicit option prices under risk neutral measure.
Implications on models Many models in mathematical finance are designed such
thatThey are formally aribtrage-free.They allow for explicit solutions for option prices. i.e. model structure often driven by mathematical
convenience.
Examples: Black-Scholes, but also Cox-Ingersoll-Ross,
HJM.
These technical restrictions can often not be reconciled
with the observed statistical properties of real-world data.classical example: volatility smile in the Black-Scholes
model.
Consequences for DFA
Most important for DFA: Models must faithfully
reproduce the observable real-world behaviour of
the modelled assets.
Therefore: fundamental differences in paradigms
underlying the selection or construction of models.
Hence: take care when using models in DFA that
were mainly constructed for derivative pricing.
A little case study for illustration...
A little case study: CIR Cox-Ingersoll-Ross model for short-term interest
rate r(t) and zero-coupon yields R(t,T).
( ) ( ( )) ( ) ( )dr t a b r t dt s r t dZ t
1, log R t T r t B T A T
T
2
2( ) / 22
( )( )( 1) 2
aba G T s
GT
GeA T
a G e G
2( 1)( )
( )( 1) 2
GT
GT
eB T
a G e G
2 22G a s
where:
CIR: Properties
One-factor model: only one source of randomness.
Nice analytical properties: explicit formulae forZero-coupon yields,Bond prices, Interest rate option prices.
(Fairly) easy to calibrate (Generalized Method of Moments).
But: How well does CIR reproduce the behaviour of the real-world interest rate data?
CIR: Yield Curves
True Yield Curves (CHF)
0.02
0.025
0.03
0.035
0.04
0.045
1 2 3 4 5 6 7
Time to maturity [Y]
Rat
e
Simulated yield curves using CIR (CHF)
0.0200
0.0250
0.0300
0.0350
0.0400
0.0450
1 2 3 4 5 6 7
Time to maturity [Y]
Rat
e
CIR Yield Curves: Remarks
CIR: yield curve fully determined by the short-term
rate!
Simulated curves always tend from the short-term
rate towards the long-term mean.
Hence: Insufficient reproduction of empirical
caracteristics of yield curves: e.g. humped and
inverted shapes.
From this point of view: CIR is not suitable for DFA!
But: What about the short-term rate?
CIR: Short-term Rate (I)
Classical source: the paper by Chan, Karolyi,
Longstaff, and Sanders („CKLS“).
Evaluation based on T-Bill data from 1964 to
1989: involving the high-rate period 1979-1982 involving possible regime switches in 1971
(Bretton-Woods) and 1979 (change of Fed policy).
Parameter estimation by classical GMM.
CKLS‘s conclusion: CIR performs poorly for short-
rate!
CIR: Short-term rate (II) More recent study: Dell‘Aquila, Ronchetti, and
Trojani
Evaluation on different data sets: Same as CKLS Euro-mark and euro-dollar series 1975-2000
Parameter estimation by Robust GMM.
Conclusions: classical GMM leads to unreliable estimates; CIR with parameters estimated by robust GMM describes fairly well the data after 1982.
Hence: CIR can be a good model for the short-term rate!
Methodological conclusions Thorough statistical analysis of historical data is
crucial! Alternative estimation methods (e.g. robust statistics) may bring better results than classical methods.
Models may need modification to fit needs of DFA.
Careful model validation must be done in each case.
Models that are good for other tasks are not necessarily good for DFA (due to different requirements).
Residual uncertainty must be taken into account when evaluating final DFA results.
Excursion: Robust Statistics Methods for data analysis and inference on data of
poor quality (satisfying only weak assumptions).Relaxed assumptions on normality.Tolerance against outliers.
Theoretically well founded; practically well
introduced in natural and life sciences.
Not yet very popular in finance, however: emerging
use.
Especially interesting for DFA: Small Sample
Asymptotics.Relevance of estimates based on little data...
An alternative model for interest rates (I) Due to Cont; based on a careful statistical study of
yield curves by Bouchaud et al. (nice methodological
reference)
Consequently designed for reproducing real-world
statistical behaviour of yield curves.
Can be linked to inflation and stock index models.
Theoretically not arbitrage-free. However – if well
fitted:
„as arbitrage-free as the real world...“
An alternative model for interest rates (II)
1 21 11 12( , ) ( , ) ( , )t t t t t t t t tdr r s dt r s dW r s dW
1 22 21 22( , ) ( , ) ( , )t t t t t t t t tds r s dt r s dW r s dW
( ) ( ) ( )t t t tf r s Y X
( ) 0, ( ) 1MIN MAXY Y ( ) 0, ( ) 0t MIN t MAXX X
( ) ( )t t t tdX X dt X d B
Short-rate
Spread
Forward rate
Average yield curve shape
Stochastic deformation
Time evolution of deformation
In principle, is an infinite-dimensional process. However, it can be
boiled down to an easily tractable finite dimensional one.tX
Multivariate Models: Problem Statement Models for single risk factors (underwriting and
financial) are available from actuarial and financial
science.
However: „The whole is more than the sum of its
parts.“ Dependences must be duly modelled.
Not modelling dependences suggests
diversification possibilities where none are present.
Significant dependences are present on the
financial and on the underwriting side.
Dependences: Example
02468
1012141618
1981
1983
1985
1987
1989
1991
1993
1995
1997
1999
[%]
US Interest Rate
US Inflation
02468
1012141618
1981
1983
1985
1987
1989
1991
1993
1995
1997
1999
[%]
CH Interest Rate
CH Inflation
00.10.20.30.40.50.60.70.80.9
1
1981
1983
1985
1987
1989
1991
1993
1995
1997
1999
US
D
USD per CHF
Particlular problem: integrated asset model An economic and investment scenario generator
for DFA (involving inflation, interest rates, stock prices, etc.) must reflect various aspects:
marginal behaviour of the variables over time in particular: long-term aspects (many years ahead)dependences between the different variables„unusual“ and „extreme“ outcomeseconomic stylized facts
Hence: need for an integrated model, not just a collection of univariate models for single risk factors.
General modelling approaches Statistical: by using multivariate time series models
established standard methods, nice quantitative propertiespractical interpretation of model elements often difficult
Fundamental: by using formulae from economic theoryexplains well the „usual“ behaviour of the variablesoften suboptimal quantitative properties
Phenomenological: compromise between the twomodels designed for reflecting statistical behaviour of dataallowing nevertheless for practical interpretation
Phenomenological approach most promising for DFA.
Economic and investment models „CIR + CAPM“ as in Dynamo
Wilkie Model in different variants (widespread in UK)
Continuous-time models by Cairns, Chan, Smith
Random walk models with Gaussian or - stable innovations
Etc.: see bibliography.
None of the models outperforms the others.
Investment models: open issues Exploration of alternative model structures Model selection and calibration Long-term behaviour: stability, convergence,
regime switches, drifts in parameters, etc. Choice of initial conditions Inclusion of rare and extremal events Inclusion of exogeneous forecasts Time scaling and aggregation properties Framework for model risk management
Excursion: Model Risk Management Qualify and (as far as possible) quantify uncertainty as
to the appropriateness of the model in use. Which relevant dangers are (not) reflected by the
model? Interpretation of simulation results given model
uncertainty Particularly important in DFA: long-term issues. Little done on MRM in quatitative finance up to now
(exception: pure parameter risk). Sources of inspiration: statistics (frequentist and
Bayesian), economics, information theory (Akaike...),
etc.
Rare & extreme events: problem statement Rare but extreme events are one particular
danger for an insurance company.
Hence, DFA scenarios must reflect such
events.
Extreme Value Theory (EVT) is a useful tool.
C.f. Paul Embrechts‘ presentation last year.
Some complements of interest for DFA:Time series with heavy-tailed residualsMultivariate extensions
The classical case
X1, ... , Xn ~ iid (or stationary with additional
assumptions)
Xi : univariate observations
Investigation of max {X1, ... , Xn}
=> Generalized Extreme Value Distribution (GEV)
Investigation of P (Xi – u x | x > u)
(excess distribution of Xi over some threshold u)
=> Generalized Pareto Distribution (GPD)
The classical case: applications
Well established in the actuarial and financial
field:Description of high quantiles and tailsComputation of risk measures such as VaR or
Conditional VaR (= Expected Shortfall Expected
Policyholder Deficit)Scenario generation for simulation studiesEtc.
In general: consistent language for describing
extreme risks across various risk factors.
Multivariate extremes: setup and context As before: X1, ... , Xn ~ iid, but now: Xi n (multivariate)
Relevant for insurance and DFA? Yes, in some cases, e.g.Correlated natural perils (in the absence of suitable CAT
modelling tool coverage).Presence of multi-trigger products in R/I
Area of active research; however, still in its infancy:Some publications on workable theoretical foundationsFew (pre-industrial) applied studies (FX data, flood, etc.)Considerable progress expected for the next years.
Multivariate extremes: problems (I) No natural order in multidimensional space:
=> no „natural“ notion of extremes
Different conceptual approaches present:Spectral measure + tail index (think of a
transformation into polar coordinates)Tail dependence function (= Copula transform
of joint distribution)Both approaches are practically workable.
Generally established workable theory not
yet present.
Multivariate extremes: problems (II) In the multivariate setup:„The Curse of
Dimensionality“Number of data points required for obtaining „well
determined“ parameter estimates increases
dramatically with the dimension.However, extreme events are rare by definition...
Problem perceived as tractable in „low“ dimesion
(2,3,4)Most published studies in two dimensionsHigher-dimensional problems beyond the scope of
current methods
Time series with heay-tailed residuals Given some time series model (e.g. AR(p)):
Xt = f (Xt-1 , Xt-2 , ... ) + t | 1 , 2 , ... ~ iid, E (i) = 0
Usually: t ~ N(0, 2) (Gaussian)
However: there are time series that cannot be
reconciled with the assumption of Gaussian
residuals (even on such high levels of time
aggregation as in DFA).
Therefore: think of heavier-tailed – also skewed –
distributions for the residuals! (Various
approaches present.)
Heavy-tailed residuals: example
QQ normal plots of yearly inflation (Switzerland and
USA)
Straight line indicates theoretical quantiles of
Gaussian distribution.
Heavy-tailed residuals: direct approach Linear time series model (e.g. AR(p)), with residuals
having symmetric--stable (ss) distribution.ss: general class of more or less heavy-tailed
distributions; = characteristic exponent; can be estimated from data. = 2 Gaussian; = 1 Cauchy.Disadvantage: ss RV‘s in general difficult to simulate.
Take care with other heavy-tailed distributions (e.g
Student‘s t): multiperiod simulations may become
uncontrollable.
Superposition of shocks
Normal model with superimposed rare, but extreme
shocks:
Xt = f (Xt-1 , Xt-2 , ... ) + t + t t
1 , 2 , ... ~ iid Bernoulli variables (occurrence of shock)
1 , 2, ... the actual shock events
Problem: recovery of model from the shock!Shock itself is realistic as compared to data.But model recovers much faster/slower than actual data.
Hence: Care must be taken.
Continuous-time approaches „Alternatives to Brownian Motion“ (i.e. Gaussian
processes) General Lévy processes Continuous-time - stable processes Jump – diffusion processes (e.g. Brownian motion
with superimposed Poisson shock process) Theory well understood in the univariate case. Emerging use in finance (e.g. Morgan-Stanley) Mutivariate case more difficult: difficulties with
correlation because second moment is infinite.
Further approaches
Heavy-tailed random walks (ss – innovations);
possibly corrected by expected forward
premiums (where available).
Regime-switching time series models, e.g.
Threshold Autoregressive (TAR or SETAR = Self-
Excited TAR).
Non-linear time series models: ARCH or GARCH
(however: more suitable for higher-frequency
data).
Conclusions (I)
Applied stochastics and, in particular, mathematical finance offer many models that are useful for DFA.
However, before using a model, careful analysis must be made in order to assess the appropriateness of the model under the specific conditions of DFA. Modifications may be necessary.
The quality of a calibrated model crucially depends on sensible choices of historical data and methods for parameter estimation.
Conclusions (II)
Time dependence of and correlation between
risk factors are crucial in the multivariate and
multiperiod setup of DFA. When particularly
confronted with rare and extreme events:Time series models with heavy tails are well
understood and lend themselves to the use in
DFA.Multivariate extreme value theory is still in its
infancy, but workable approaches can be expected
to emerge within the next few years.