Optimizing S-shaped utility and risk management: ineffectiveness of VaR and ES constraints

Similar documents
Optimizing S-shaped utility and risk management

LECTURE 4: BID AND ASK HEDGING

Statistical Methods in Financial Risk Management

References. H. Föllmer, A. Schied, Stochastic Finance (3rd Ed.) de Gruyter 2011 (chapters 4 and 11)

IEOR E4602: Quantitative Risk Management

Characterization of the Optimum

How do Variance Swaps Shape the Smile?

All Investors are Risk-averse Expected Utility Maximizers

All Investors are Risk-averse Expected Utility Maximizers. Carole Bernard (UW), Jit Seng Chen (GGY) and Steven Vanduffel (Vrije Universiteit Brussel)

Umut Çetin and L. C. G. Rogers Modelling liquidity effects in discrete time

Optimal investment and contingent claim valuation in illiquid markets

SOLVENCY AND CAPITAL ALLOCATION

MATH3075/3975 FINANCIAL MATHEMATICS TUTORIAL PROBLEMS

A lower bound on seller revenue in single buyer monopoly auctions

An Application of Extreme Value Theory for Measuring Financial Risk in the Uruguayan Pension Fund 1

Prudence, risk measures and the Optimized Certainty Equivalent: a note

On the Lower Arbitrage Bound of American Contingent Claims

Lecture 8: Introduction to asset pricing

Risk, Coherency and Cooperative Game

Basic Arbitrage Theory KTH Tomas Björk

Ph.D. Preliminary Examination MICROECONOMIC THEORY Applied Economics Graduate Program June 2017

Distortion operator of uncertainty claim pricing using weibull distortion operator

Department of Mathematics. Mathematics of Financial Derivatives

Risk measures: Yet another search of a holy grail

Hedging under Arbitrage

Short Course Theory and Practice of Risk Measurement

Optimal Allocation of Policy Limits and Deductibles

Math489/889 Stochastic Processes and Advanced Mathematical Finance Homework 5

Approximate Revenue Maximization with Multiple Items

Mathematics in Finance

Short Course Theory and Practice of Risk Measurement

Risk Measures and Optimal Risk Transfers

Lecture 8: Asset pricing

based on two joint papers with Sara Biagini Scuola Normale Superiore di Pisa, Università degli Studi di Perugia

CONSISTENCY AMONG TRADING DESKS

Consistent Measures of Risk

3 Arbitrage pricing theory in discrete time.

Guarantee valuation in Notional Defined Contribution pension systems

Hedging under arbitrage

Optimal Investment with Deferred Capital Gains Taxes

Rho-Works Advanced Analytical Systems. CVaR E pert. Product information

Martingales. by D. Cox December 2, 2009

Basic Concepts in Mathematical Finance

Lecture 1 Definitions from finance

A model for a large investor trading at market indifference prices

4: SINGLE-PERIOD MARKET MODELS

Time Resolution of the St. Petersburg Paradox: A Rebuttal

Pricing in markets modeled by general processes with independent increments

Pricing theory of financial derivatives

Chapter 23: Choice under Risk

Risk Measures for Derivative Securities: From a Yin-Yang Approach to Aerospace Space

Estimation of Value at Risk and ruin probability for diffusion processes with jumps

Portfolio Choice via Quantiles

Value at Risk, Expected Shortfall, and Marginal Risk Contribution, in: Szego, G. (ed.): Risk Measures for the 21st Century, p , Wiley 2004.

6: MULTI-PERIOD MARKET MODELS

Microeconomics of Banking: Lecture 2

Local vs Non-local Forward Equations for Option Pricing

Choice under Uncertainty

Lecture Note 8 of Bus 41202, Spring 2017: Stochastic Diffusion Equation & Option Pricing

Introduction to Probability Theory and Stochastic Processes for Finance Lecture Notes

Copyright (C) 2001 David K. Levine This document is an open textbook; you can redistribute it and/or modify it under the terms of version 1 of the

Lecture 6: Option Pricing Using a One-step Binomial Tree. Thursday, September 12, 13

Pricing and hedging in incomplete markets

6.254 : Game Theory with Engineering Applications Lecture 3: Strategic Form Games - Solution Concepts

Capital Allocation Principles

Week 2 Quantitative Analysis of Financial Markets Hypothesis Testing and Confidence Intervals

Portfolio Optimization using Conditional Sharpe Ratio

CHOICE THEORY, UTILITY FUNCTIONS AND RISK AVERSION

On Utility Based Pricing of Contingent Claims in Incomplete Markets

Optimal stopping problems for a Brownian motion with a disorder on a finite interval

Econ 101A Final exam Mo 18 May, 2009.

Comparative Analyses of Expected Shortfall and Value-at-Risk (2): Expected Utility Maximization and Tail Risk

Lecture Quantitative Finance Spring Term 2015

Utility Indifference Pricing and Dynamic Programming Algorithm

Model-independent bounds for Asian options

Risk-Averse Decision Making and Control

A class of coherent risk measures based on one-sided moments

Financial Mathematics III Theory summary

Optimal Investment for Generalized Utility Functions

Probability. An intro for calculus students P= Figure 1: A normal integral

Prospect Theory: A New Paradigm for Portfolio Choice

Risk Minimization Control for Beating the Market Strategies

Class Notes on Financial Mathematics. No-Arbitrage Pricing Model

Hedging Credit Derivatives in Intensity Based Models

Risk based capital allocation

Uncertainty in Equilibrium

IEOR E4602: Quantitative Risk Management

Hedging volumetric risks using put options in commodity markets

Optimal Investment for Worst-Case Crash Scenarios

Exercises for Chapter 8

Portfolio rankings with skewness and kurtosis

LECTURE 2: MULTIPERIOD MODELS AND TREES

Viability, Arbitrage and Preferences

Rohini Kumar. Statistics and Applied Probability, UCSB (Joint work with J. Feng and J.-P. Fouque)

Utility Maximization for an Investor with Asymmetric Attitude to Gains and Losses over the Mean Variance Efficient Frontier

Allocation of Risk Capital via Intra-Firm Trading

A generalized coherent risk measure: The firm s perspective

Model-independent bounds for Asian options

STOCHASTIC CALCULUS AND BLACK-SCHOLES MODEL

UNIFORM BOUNDS FOR BLACK SCHOLES IMPLIED VOLATILITY

Transcription:

Optimizing S-shaped utility and risk management: ineffectiveness of VaR and ES constraints John Armstrong Dept. of Mathematics King s College London Joint work with Damiano Brigo Dept. of Mathematics, Imperial College London Risk Minds, 7 Dec 2017

Agenda Motivation S-shaped utility and tail-risk-seeking behaviour Law invariant portfolio optimization and rearrangements Expected shortfall constraints Utility Constraints J. Armstrong & D. Brigo S-shaped utility vs VaR and ES Risk Minds, 7 Dec 2017 2 / 36

Motivation Standard XVA methodology uses risk-neutral pricing, assumes all risk can be hedged. This is a questionable assumption, e.g. for KVA (see [7]). We wish to consider indifference pricing. Indifference price P I of liability L solves: sup E(u(X)) = sup E(u(X + L + P I )) strategies strategies where X is the payoff achieved by following a strategy. We will focus on the question of how to compute sup E(u(X)) strategies Challenge: shareholders have limited liability so their utility function will not be concave. Our results will have implications that go beyond XVA calculation. J. Armstrong & D. Brigo S-shaped utility vs VaR and ES Risk Minds, 7 Dec 2017 3 / 36

S-shaped utility I In [14], Kahneman & Tversky observed that individuals appear to have preferences governed by an S-shaped utility function u. (i) u is increasing (ii) strictly convex on the left (iii) strictly concave on the right (iv) non-differentiable at the origin (v) asymmetrical: negative events are considered worse than positive events are considered good. Utility Terminal Wealth J. Armstrong & D. Brigo S-shaped utility vs VaR and ES Risk Minds, 7 Dec 2017 4 / 36

S-shaped utility II An increasing function u : R R (to be thought of as a utility function) is said to be risk-seeking in the left tail if there exist constants N 0, η (0, 1) and c > 0 such that: u(x) > c x η x N. (1) Similarly u is said to be risk-averse in the right tail if there exists N 0, η (0, 1) and c > 0 such that u(x) < c x η x N. (2) The standard pictures of S-shaped utility functions in the literature appear to have these properties. Furthermore the S-shaped utility functions that would arise due to a limited liability would be bounded below and so would certainly be risk-seeking in the left tail. J. Armstrong & D. Brigo S-shaped utility vs VaR and ES Risk Minds, 7 Dec 2017 5 / 36

S-shaped utility III We give a formal definition of S-shaped for the purposes of this work. u is S-shaped if 1. u is increasing 2. u(x) 0 for x 0 3. u(x) 0 for x 0 4. u(x) concave for x 0. 5. u risk-seeking in the left tail. 6. u risk-averse in the right tail. NL Utility NR Terminal Wealth J. Armstrong & D. Brigo S-shaped utility vs VaR and ES Risk Minds, 7 Dec 2017 6 / 36

S-shaped utility IV Example Limited liability: u + (x) = max{u(x), 0} Utility Terminal Wealth Solving optimization problems for expected concave utility typically yields convex optimization problems. Our optimal utility problem will be non-convex, but we can still solve it. J. Armstrong & D. Brigo S-shaped utility vs VaR and ES Risk Minds, 7 Dec 2017 7 / 36

Law invariant portfolio optimization Let (Ω, F, P) be a probability space and let dq dp be a positive random variable with dq Ω dp dp(ω) = 1. We will use this model to represent a complete financial market as follows: (i) We assume there is a fixed deterministic risk free interest rate r. (ii) Given a random variable f, one can purchase a derivative security with payoff at maturity T (simple claim) given by f(ω) for the price E Q [e rt f] := assuming that this integral exists. Ω e rt f(ω) dq (ω) dp (3) dp J. Armstrong & D. Brigo S-shaped utility vs VaR and ES Risk Minds, 7 Dec 2017 8 / 36

Law invariant portfolio optimization Investor preferences are law-invariant. i.e. The investor s preferences are encoded by some function v : M 1 (R) R where M 1 (R) is the space of probability measures on R, so that an investor will prefer a security with payoff f over a security with payoff g iff v(f f ) > v(f g ) (F is the cdf). Risk constraints are law-invarant. i.e. whether a payoff f is acceptable depends only on F f. Examples include VaR and CVaR constraints. We have a cost constraint. J. Armstrong & D. Brigo S-shaped utility vs VaR and ES Risk Minds, 7 Dec 2017 9 / 36

The optimization problem In summary, our investor wishes to solve the optimization problem: sup f L 0 (Ω,P) subject to a price constraint risk management constraints v(f f ) Ω e rt f(ω) dq (ω) dp(ω) C dp F f A M 1 (R). (4) J. Armstrong & D. Brigo S-shaped utility vs VaR and ES Risk Minds, 7 Dec 2017 10 / 36

Rearrangement Theorem Theorem (Rearrangement) Assume Ω is non-atomic. Then there exists a standard uniformly distributed random variable U such that: = (1 F dq ) 1 U almost surely. dp (ii) If f satisfies the price and risk management constraints of our (i) dq dp problem then ϕ(u) = F 1 f U also satisfies the constraints of our problem and is equal to f in distribution, and hence has the same objective value as f. The implication of this is that we can simplify our market model so that we are simply betting on the final value of a single uniform variable U. We may assume that the payoff of our investment is an increasing function of U, while dq dp is a decreasing function of U. J. Armstrong & D. Brigo S-shaped utility vs VaR and ES Risk Minds, 7 Dec 2017 11 / 36

Example: The Black-Scholes case Consider derivatives in a Black Scholes market with (deterministic) r, we can write the log-stock price as s T = s 0 + (µ 1 2 σ2 )T + σ T N 1 (U) The P and Q measure densities are ( p{q} BS 1 (s T ) = σ 2πT exp (s ) T (s 0 + (µ{r} 1 2 σ2 )T )) 2 2σ 2. T 3.5 dq d 3.0 2.5 2.0 1.5 1.0 0.5 0.2 0.4 0.6 0.8 1.0 U J. Armstrong & D. Brigo S-shaped utility vs VaR and ES Risk Minds, 7 Dec 2017 12 / 36

Equivalent problem Let q(u) = dq dp (U) then we can reduce our optimization problem to solving sup ϕ:[0,1] R,ϕ increasing subject to the price constraint F(ϕ) := 1 0 1 0 u(ϕ(x)) dx (5) ϕ(x)q(x) dx C (6) and risk management constraints F ϕ A M 1 (R). (7) The only feature of the market that is relevant is the decreasing function q(u). Intuition: dp dq is a measure of how good value you think a bet is compared to the market. You should place your bets on events where dp dq is high. Nothing else matters. J. Armstrong & D. Brigo S-shaped utility vs VaR and ES Risk Minds, 7 Dec 2017 13 / 36

Example: expected shortfall Consider the optimization problem with expected shortfall constraints sup ϕ:[0,1] R,ϕ increasing subject to the price constraint and the expected shortfall constraint F(ϕ) := 1 1 p 0 p 0 1 0 u(ϕ(x)) dx (8) ϕ(x)q(x) dx C (9) ϕ(x) dx L. (10) Theorem If u is risk seeking in the tail and q(x) is essentially unbounded, then the supremum of this problem is equal to sup u. i.e. investors with S-shaped utility are untroubled by expected shortfall constraints. J. Armstrong & D. Brigo S-shaped utility vs VaR and ES Risk Minds, 7 Dec 2017 14 / 36

Sketch proof Consider the payoff φ(x) = { k 2 k 1 when x < α otherwise Since q(u) as U 0, the market contains events of arbitrarily good value: The price constraint requires roughly k 2 c 1 αq(α) k 1. The ES constraint requires roughly k 2 c α k 1. By taking α small enough we can find k 2 meeting our constraints whatever we choose for k 1. By the ES constraint ( ) c η E(u) = αu(k 1 ) + (1 α)u(k 2 ) α α k 1 + (1 α)u(k 1 ) u(k 1 ) as α 0. J. Armstrong & D. Brigo S-shaped utility vs VaR and ES Risk Minds, 7 Dec 2017 15 / 36

Implications In the Black Scholes model with µ r, an investor with S-shaped utility subject to just Expected Shortfall constraints will be unconcerned by these constrains. In the Black Scholes model µ r, an investor with S-shaped utility subject to just Value at Risk constraints will be unconcerned by these constrains. In fact, in any reasonable complete market model with non-zero market price of risk we expect that Expected Shortfall constraints and Value at Risk constraints will be ineffective. In general indifference prices cannot be defined for an investor with S-shaped utility subject only to price, expected shortfall constraints and value at risk constraints. J. Armstrong & D. Brigo S-shaped utility vs VaR and ES Risk Minds, 7 Dec 2017 16 / 36

Optimization with limited liability & utility constraints We now specialize to the case of an investor with limited liability u I & We suppose a regulator is indifferent if portfolio payoff is positive, and imposes risk constraint with 2nd utility u R on negative payoff part. Risk constraint is E(u R ) L Utility 1-2 -1 1 2-1 -2-3 -4 Payoff ui ur J. Armstrong & D. Brigo S-shaped utility vs VaR and ES Risk Minds, 7 Dec 2017 17 / 36

Solving the optimization problem We know the optimal payoff function φ is increasing, so it must be negative for values less than some p [0, 1] and positive for values greater than p. Given p [0, 1], define C 1 (p) and V (p) as p C 1 (p) = inf f 1 (x)q(x)dx V (p) = f 1 :[0,p] (,0), with f 1 increasing subject to sup f 2 :[p,1] [0, ), with f 2 increasing subject to 1 p 1 p 0 p 0 u R (f 1 (x)) dx L. u I (f 2 (x))dx (11) f 2 (x)q(x)dx e rt C C 1 (p) (12) J. Armstrong & D. Brigo S-shaped utility vs VaR and ES Risk Minds, 7 Dec 2017 18 / 36

Theorem The supremum of the optimization problem for an investor with limited liability subject to a concave utility constraint on the loss is sup p [0,1] V (p). The value of the theorem comes from the fact that the optimization problems to compute C 1 (p) and V (p) are convex problems and so are easy to solve by standard techniques. We may then compute sup p [0,1] V (p) by line search. Unlike the case of expected shortfall, these utility constraints are typically binding. J. Armstrong & D. Brigo S-shaped utility vs VaR and ES Risk Minds, 7 Dec 2017 19 / 36

Implications Utility constraints are typically effective in constraining investors with S-shaped utility. In typical cases, for example in the Black-Scholes model when u I is unbounded, an investor with S-shaped utility will choose investments with infinitely bad u R utility if they are not subject to u R constraints but only expected shortfall constraints. Indifference prices can be defined and calculated if we consider investors with limited liability under utility constraints. J. Armstrong & D. Brigo S-shaped utility vs VaR and ES Risk Minds, 7 Dec 2017 20 / 36

Example: the Black Scholes case The optimal payoff function is plotted against the stock price for the Black-Scholes model. We also show the P and Q measure density functions. Payoff 40 20 0.8 0.9 1.0 1.1 1.2 1.3 S T P Q -20 J. Armstrong & D. Brigo S-shaped utility vs VaR and ES Risk Minds, 7 Dec 2017 21 / 36

Example: the Black Scholes case This figure shows how the strategies vary as the risk limit is changed. Payoff 100 Out[604]= -100 0.7 0.8 0.9 1.0 1.1 1.2 1.3 ST ur -1 ur -10 ur -100 95%-ES Arbitrage P Q -200 In black we see a portfolio with positive payoff, negative expected shortfall and negative price. An investor with limited liability could purchase an arbitrary quantity of this asset to achieve any desired utility. J. Armstrong & D. Brigo S-shaped utility vs VaR and ES Risk Minds, 7 Dec 2017 22 / 36

Proof of Rearrangement Theorem I Consider a finite probability space where each atomic event has the same P measure. We show a graph of dq dp in purple and the payoff f of some option in orange (LHS). The x-axis corresponds to the different events. The choice of the x-axis completely arbitrary, so we might as well choose our plot so that dq dp is decreasing (RHS). J. Armstrong & D. Brigo S-shaped utility vs VaR and ES Risk Minds, 7 Dec 2017 23 / 36

Proof of Rearrangement Theorem II Now swap the order of the payoff columns so that we get taller bars on the right. Swapping bars doesn t change the P distribution of the payoff, but clearly lowers the price. We have a cheaper portfolio with identical P distribution. Our objective and risk-management constraints are law invariant, so the payoff on the right meets all our constraints. J. Armstrong & D. Brigo S-shaped utility vs VaR and ES Risk Minds, 7 Dec 2017 24 / 36

Proof of Rearrangement Theorem III To generalize this to the continuous case we need to define what we mean by rearrangement. Definition 1 Given random variables X, f L 0 (Ω, R) with X having a continuous distribution we define the X-rearrangement of f, denoted f X by: f X (ω) = F 1 f (P(X X(ω))) = F 1 (F X (X(ω))). We know that X = F 1 X (U) for some uniformly distributed U f X is equal to f in distribution f X depends on U alone f X is an increasing function of U f J. Armstrong & D. Brigo S-shaped utility vs VaR and ES Risk Minds, 7 Dec 2017 25 / 36

Proof of Rearrangement Theorem IV To prove our theorem we will take an arbitrary payoff f and replace it with the rearranged payoff f dq dp. (We assume for simplicity that dq dp has a continuous distribution. A minor technical lemma is needed to prove the general case.) Our rearrangement theorem will then follow from: Lemma 2 If f, g L 0 (Ω; R) and: (i) fg dp > ; (ii) g 0; (iii) Ω g dp exists; (iv) X has a continuous distribution; then < Ω fg dp Ω f X g X dp. J. Armstrong & D. Brigo S-shaped utility vs VaR and ES Risk Minds, 7 Dec 2017 26 / 36

Proof of Rearrangement Theorem V For simplicity assume that dq dp and payoff are defined on [0, 1], with f and g also taking values in [0, 1]. We consider the layer cake representations like f(u) = 1 this is depicted below (LHS). 0 1 f(u) l dl, g(u) = 1 0 1 g(u) l dl J. Armstrong & D. Brigo S-shaped utility vs VaR and ES Risk Minds, 7 Dec 2017 27 / 36

Proof of Rearrangement Theorem VI 1 f(u)g(u)du = 0 1 = 0 1 0 1 0 1 f(u) l 1 g(u) k dk dl du 1 f(u) l and g(u) k dk dl du So the integral aggregates the intersections of the layers. J. Armstrong & D. Brigo S-shaped utility vs VaR and ES Risk Minds, 7 Dec 2017 28 / 36

Proof of Rearrangement Theorem VII The layer cake representations for f U and g U look as shown This increases the integral of the product is because this rearrangement increases the amount any layers intersect as shown below Hence f U (U)g U (U)dU f(u)g(u)du. J. Armstrong & D. Brigo S-shaped utility vs VaR and ES Risk Minds, 7 Dec 2017 29 / 36

Proof of Rearrangement Theorem VIII This layer cake argument is due to Hardy and Littlewood who used it to prove their inequality on so-called symmetric decreasing rearrangements. We use exactly the same idea, but applied to a notion we call X-rearrangement. J. Armstrong & D. Brigo S-shaped utility vs VaR and ES Risk Minds, 7 Dec 2017 30 / 36

Summary Indifference pricing provides a pricing methodology that can be used when not all risks can be perfectly hedged. The key to indifference pricing is computing optimal investment strategies. Market players may have limited liability and hence non-concave utility functions, nevertheless we can apply rearrangement to obtain tractable optimization problems. Expected shortfall and value at risk constraints typically do not constrain investors with S-shaped utility functions. Utility constraints typically do constrain investors with S-shaped utility functions. Future Research 1: Investigate S-shaped utility optimization in incomplete market models. Future Research 2: Apply these techniques to indifference pricing of XVA type liabilities. See [7] for initial results in this area. J. Armstrong & D. Brigo S-shaped utility vs VaR and ES Risk Minds, 7 Dec 2017 31 / 36

References I [1] C. Acerbi and B. Szekely. Back-testing expected shortfall. Risk Magazine, 2014. [2] C. Acerbi and D. Tasche. On the coherence of expected shortfall. Journal of Banking and Finance, (26):1487 1503, 2002. [3] J. Armstrong and D. Brigo. Optimizing S-shaped utility and implications for risk management. Available at arxiv.org, report 1711.00443 [4] P. Artzner, F. Delbaen, J.-M. Eber, and D. Heath. Coherent measures of risk. Mathematical finance, 9(3):203 228, 1999. J. Armstrong & D. Brigo S-shaped utility vs VaR and ES Risk Minds, 7 Dec 2017 32 / 36

References II [5] Basel Committee on Banking Supervision. Fundamental review of the trading book:a revised market risk framework, second consultative paper. 2013. [6] Basel Committee on Banking Supervision. Minimum capital requirements for market risk. 2016. [7] D. Brigo, M. Francischello, and A. Pallavicini An indifference approach to the cost of capital constraints: KVA and beyond Available at arxiv.org, 2017. [8] J. Danielsson. Financial Risk Forecasting. Wiley, 2011. J. Armstrong & D. Brigo S-shaped utility vs VaR and ES Risk Minds, 7 Dec 2017 33 / 36

References III [9] M.H.A Davis. Verification of internal risk measure estimates. Statistics and Risk Modeling, 33:67 93, 2016. [10] T. Fisslet and J. Ziegel. Higher order elicitability and osband s principle. Annals of Statistics, 44(4):1680 1707, 2016. [11] H. Föllmer and A. Schied. Stochastic finance: an introduction in discrete time. Walter de Gruyter, 2011. [12] T. Gneiting. Making and evaluating point forecasts. J. Amer. Statist. Assoc., 106:746 762. J. Armstrong & D. Brigo S-shaped utility vs VaR and ES Risk Minds, 7 Dec 2017 34 / 36

References IV [13] E. P. Hsu. Stochastic analysis on manifolds, volume 38. American Mathematical Soc., 2002. [14] D. Kahneman and A. Tversky. Prospect theory: An analysis of decision under risk. Econometrica: Journal of the econometric society, pages 263 291, 1979. [15] A. J. McNeil, R. Frey, and P. Embrechts. Quantitative Risk Management: Concepts, Techniques and Tools. Princeton University Press, Princeton, NJ, USA, 2015. [16] R. T. Rockafellar. Convex analysis. Princeton university press, 2015. J. Armstrong & D. Brigo S-shaped utility vs VaR and ES Risk Minds, 7 Dec 2017 35 / 36

References V [17] W. Sierpiński. Sur les fonctions d ensemble additives et continues. Fundamenta Mathematicae, 1(3):240 246, 1922. [18] D. Tasche. Expected shortfall is not elicitable - so what? Seminar presented at Imperial College London, 20 November 2013. J. Armstrong & D. Brigo S-shaped utility vs VaR and ES Risk Minds, 7 Dec 2017 36 / 36