Portfolios with Trading Constraints and Payout Restrictions

Similar documents
Winter 2008 CS567 Stochastic Linear/Integer Programming Guest Lecturer: Xu, Huan

Tokyo Institute of Technology Periodic Sequencing Control over Multi Communication Channels with Packet Losses

Economics 8105 Macroeconomic Theory Recitation 1

Perfect Competition and the Nash Bargaining Solution

Economics 101. Lecture 4 - Equilibrium and Efficiency

EEL 6266 Power System Operation and Control. Chapter 3 Economic Dispatch Using Dynamic Programming

Lecture Notes on Linear Regression

CIS526: Machine Learning Lecture 3 (Sept 16, 2003) Linear Regression. Preparation help: Xiaoying Huang. x 1 θ 1 output... θ M x M

CIE4801 Transportation and spatial modelling Trip distribution

Donald J. Chmielewski and David Mendoza-Serrano Department of Chemical and Biological Engineering Illinois Institute of Technology

Assortment Optimization under MNL

Lecture 21: Numerical methods for pricing American type derivatives

Additional Codes using Finite Difference Method. 1 HJB Equation for Consumption-Saving Problem Without Uncertainty

Mathematical Economics MEMF e ME. Filomena Garcia. Topic 2 Calculus

Module 3 LOSSY IMAGE COMPRESSION SYSTEMS. Version 2 ECE IIT, Kharagpur

k t+1 + c t A t k t, t=0

Credit Card Pricing and Impact of Adverse Selection

Chapter Newton s Method

10) Activity analysis

Outline. Communication. Bellman Ford Algorithm. Bellman Ford Example. Bellman Ford Shortest Path [1]

Mean Field / Variational Approximations

CHAPTER 5 NUMERICAL EVALUATION OF DYNAMIC RESPONSE

MMA and GCMMA two methods for nonlinear optimization

Ryan (2009)- regulating a concentrated industry (cement) Firms play Cournot in the stage. Make lumpy investment decisions

ON A DETERMINATION OF THE INITIAL FUNCTIONS FROM THE OBSERVED VALUES OF THE BOUNDARY FUNCTIONS FOR THE SECOND-ORDER HYPERBOLIC EQUATION

Simultaneous Optimization of Berth Allocation, Quay Crane Assignment and Quay Crane Scheduling Problems in Container Terminals

1. Inference on Regression Parameters a. Finding Mean, s.d and covariance amongst estimates. 2. Confidence Intervals and Working Hotelling Bands

The Constrained Multinomial Logit: A semi compensatory choice model

VARIATION OF CONSTANT SUM CONSTRAINT FOR INTEGER MODEL WITH NON UNIFORM VARIABLES

Which Separator? Spring 1

18.1 Introduction and Recap

A FAST HEURISTIC FOR TASKS ASSIGNMENT IN MANYCORE SYSTEMS WITH VOLTAGE-FREQUENCY ISLANDS

Market structure and Innovation

1 GSW Iterative Techniques for y = Ax

Lecture 10 Support Vector Machines II

THERE ARE INFINITELY MANY FIBONACCI COMPOSITES WITH PRIME SUBSCRIPTS

The Gaussian classifier. Nuno Vasconcelos ECE Department, UCSD

Test code: ME I/ME II, 2007

Supporting Information for: Two Monetary Models with Alternating Markets

Lecture 14: Bandits with Budget Constraints

princeton univ. F 17 cos 521: Advanced Algorithm Design Lecture 7: LP Duality Lecturer: Matt Weinberg

A Simple Inventory System

Externalities in wireless communication: A public goods solution approach to power allocation. by Shrutivandana Sharma

Continuous Time Markov Chain

Lecture 10 Support Vector Machines. Oct

Expectation Maximization Mixture Models HMMs

CHAPTER 17 Amortized Analysis

Equilibrium with Complete Markets. Instructor: Dmytro Hryshko

OPTIMISATION. Introduction Single Variable Unconstrained Optimisation Multivariable Unconstrained Optimisation Linear Programming

Infinitely Split Nash Equilibrium Problems in Repeated Games

ECE559VV Project Report

Online Appendix. t=1 (p t w)q t. Then the first order condition shows that

Pattern Classification

Kernel Methods and SVMs Extension

Yong Joon Ryang. 1. Introduction Consider the multicommodity transportation problem with convex quadratic cost function. 1 2 (x x0 ) T Q(x x 0 )

CS294 Topics in Algorithmic Game Theory October 11, Lecture 7

Economics 2450A: Public Economics Section 10: Education Policies and Simpler Theory of Capital Taxation

University of California, Davis Date: June 22, 2009 Department of Agricultural and Resource Economics. PRELIMINARY EXAMINATION FOR THE Ph.D.

Some modelling aspects for the Matlab implementation of MMA

The Geometry of Logit and Probit

College of Computer & Information Science Fall 2009 Northeastern University 20 October 2009

ECTRI FEHRL FERSI Young Researchers Seminar 2015

8. Modelling Uncertainty

Welfare Properties of General Equilibrium. What can be said about optimality properties of resource allocation implied by general equilibrium?

Duality in linear programming

On the Multicriteria Integer Network Flow Problem

Statistical analysis using matlab. HY 439 Presented by: George Fortetsanakis

Supporting Materials for: Two Monetary Models with Alternating Markets

Limited Dependent Variables

See Book Chapter 11 2 nd Edition (Chapter 10 1 st Edition)

Polynomial Regression Models

MIMA Group. Chapter 2 Bayesian Decision Theory. School of Computer Science and Technology, Shandong University. Xin-Shun SDU

Inventory Model with Backorder Price Discount

LINEAR REGRESSION ANALYSIS. MODULE IX Lecture Multicollinearity

( ) [ ( k) ( k) ( x) ( ) ( ) ( ) [ ] ξ [ ] [ ] [ ] ( )( ) i ( ) ( )( ) 2! ( ) = ( ) 3 Interpolation. Polynomial Approximation.

Marginal Models for categorical data.

Hidden Markov Models & The Multivariate Gaussian (10/26/04)

A Cournot-Stackelberg Advertising Duopoly Derived From A Cobb-Douglas Utility Function

Interactive Bi-Level Multi-Objective Integer. Non-linear Programming Problem

Lagrange Multipliers Kernel Trick

Flexible Allocation of Capacity in Multi-Cell CDMA Networks

INF 5860 Machine learning for image classification. Lecture 3 : Image classification and regression part II Anne Solberg January 31, 2018

Convexity preserving interpolation by splines of arbitrary degree

Chapter 1. Probability

Markov Chain Monte Carlo (MCMC), Gibbs Sampling, Metropolis Algorithms, and Simulated Annealing Bioinformatics Course Supplement

Basic Statistical Analysis and Yield Calculations

APPROXIMATE PRICES OF BASKET AND ASIAN OPTIONS DUPONT OLIVIER. Premia 14

Tracking with Kalman Filter

Stochastic Optimal Controls for Parallel-Server Channels with Zero Waiting Buffer Capacity and Multi-Class Customers

Idiosyncratic Investment (or Entrepreneurial) Risk in a Neoclassical Growth Model. George-Marios Angeletos. MIT and NBER

1 Derivation of Rate Equations from Single-Cell Conductance (Hodgkin-Huxley-like) Equations

Finite State Equilibria in Dynamic Games

Module 9. Lecture 6. Duality in Assignment Problems

Dynamic Slope Scaling Procedure to solve. Stochastic Integer Programming Problem

Analysis of Discrete Time Queues (Section 4.6)

Resource Allocation with a Budget Constraint for Computing Independent Tasks in the Cloud

A Robust Method for Calculating the Correlation Coefficient

U.C. Berkeley CS294: Beyond Worst-Case Analysis Luca Trevisan September 5, 2017

Dynamic Programming. Lecture 13 (5/31/2017)

Engineering Risk Benefit Analysis

Transcription:

Portfolos wth Tradng Constrants and Payout Restrctons John R. Brge Northwestern Unversty (ont wor wth Chrs Donohue Xaodong Xu and Gongyun Zhao) 1 General Problem (Very) long-term nvestor (eample: unversty endowment) Payout from portfolo over tme (want to eep payout from declnng) Invest n varous asset categores Decsons: How much to payout (consume)? How to nvest n asset categores? Complcaton: restrctons on asset trades 2 1

Outlne Basc formulaton General nfnte horzon soluton method Smplfed problem and contnuous tme soluton Results for restrcted-tradng portfolo Future ssues 3 Problem Formulaton Notaton: current state ( X) u (or u ) current acton gven (u (or u ) U()) δ sngle perod dscount factor P u probablty measure on net perod state y dependng on and u c(u) obectve value for current perod gven and u V() value functon of optmal epected future rewards gven current state Problem: Fnd V such that V() ma u U() {c(u) + δ E Pu [V(y)] } for all X. 4 2

Approach Defne an upper bound on the value functon V 0 () V() X Iteraton : upper bound V Solve for some TV ( ) ma u c( u) + δ E Pu [V (y)] Update to a better upper bound V +1 Update uses an outer lnear appromaton on U 5 Successve Outer Appromaton V 0 V 1 TV 0 0 V* 6 3

Propertes of Appromaton V* TV V +1 V Contracton TV V * δ V V* Unque Fed Pont TV*V* f TV V then V V*. 7 Convergence Value Iteraton T V 0 V* Dstrbuted Value Iteraton If you choose every X nfntely often then V V*. (Here random choce of use concavty.) Deepest Cut Pc to mamze V ()-TV () DC problem to solve Convergence agan wth contnuty (cauton on boundary of doman of V*) 8 4

Detals for Random Choce Consder any Choose and s.t. < ε Suppose V K V () V*() V ()-V ( )+V ( )- V*( )+V*( )-V*() 2 ε K + δ V -1 ( )-V*( ) 2 ε K + δ V 0 ( 0 ) V*( 0 ) 9 Cuttng Plane Algorthm Intalzaton: Construct V 0 ()ma u c 0 (u) + δe Pu [V 0 (y)] where c 0 c and c 0 concave. V 0 s assumed pecewse lnear and equvalent to V 0 ()ma {θ θ E 0 + e 0 }. 0. Iteraton: Sample X (n any way such that the probablty of A s postve for any A X of postve measure) and solve TV ( ) ma u c( u) + δ E Pu [V (y)] where V (y) ma{θ θ E l y + e l l0.} Fnd supportng hyperplanes defned by E +1 and e +1 such that E +1 + e +1 TV (). +1. Repeat. 10 5

Specfyng Algorthm Feasblty: A + Bu b Transton: yf u for some realzaton wth probablty p Iteraton Problem: TV ( ) ma uθ c( u) + δ p θ s.t. A + B u b - E l (F u) - e l + θ 0 l. From dualty: TV ( ) nf µλl ma uθ c( u) -µ(a +Bu-b) + δ (p θ + l λ l (E l (F u) + e l - θ )) ma uθ c( u) -µ (A +Bu-b) + δ (p θ + l λ l (E l (F u) +e l - θ )) for optmal µ λ l for c( u ) + c( u ) T (- -u ) -µ A +µ b + ( l λ l e l ) Cuts: E +1 c( u ) T -µ A e +1 equal to the constant terms. 11 Investment Problem Determne asset allocaton and consumpton polcy to mamze the epected dscounted utlty of spendng State and Acton (cons rsy wealth) u(cons_newrsy_new) Two asset classes Rsy asset wth lognormal return dstrbuton Rsfree asset wth gven return r f Power utlty functon c ( cons _ new) cons _ new 1 γ 1 γ Consumpton rate constraned to be non-decreasng cons_new cons 12 6

Estng Research Dybvg 95* Contnuous-tme approach Soluton Analyss Consumpton rate remans constant untl wealth reaches a new mamum The rsy asset allocaton α s proportonal to w-c/r f whch s the ecess of wealth over the perpetuty value of current consumpton α decreases as wealth decreases approachng 0 as wealth approaches c/r f (whch s n absence of rsy nvestment suffcent to mantan consumpton ndefntely). Dybvg 01 Consdered smlar problem n whch consumpton rate can decrease but s penalzed (soft constraned problem) * Duesenberry's Ratchetng of Consumpton: Optmal Dynamc Consumpton and Investment Gven Intolerance for any Declne n Standard of Lvng Revew of Economc Studes 62 1995 287-313. 13 Obectves Replcate Dybvg contnuous tme results usng dscrete tme approach Evaluate the effect of tradng restrctons for certan asset classes (e.g. prvate equty) Consder addtonal problem features Transacton Costs Multple rsy assets 14 7

100 Results Non-decreasng Consumpton 90 80 As number of tme perods per year ncreases soluton converges to contnuous tme soluton Allocaton to Stoc 70 60 50 40 30 Dybvg N 1 N 2 N3 N4 N 6 N12 20 10 0 2 2.25 2.5 2.75 3 3.25 3.5 3.75 4 4.25 4.5 4.75 5 Consumpton Rate 15 Results Non-Decreasng Consumpton wth Transacton Costs 60 50 40 No Transacton Cost Transacton Cost (ntal stoc allocaton 0%) Transacton Cost (ntal stoc allocaton 100%) Stoc Allocaton 30 20 10 0 3.5 3.7 3.9 4.1 4.3 4.5 4.7 4.9 Consumpton Rate 16 8

Observatons Effect of Tradng Restrctons Contnuously traded rsy asset: 70% of portfolo for 4.2% payout rate Quarterly traded rsy asset: 32% of portfolo for same payout rate Transacton Cost Effect Small dfferences n overall portfolo allocatons Optmal m depends on ntal condtons 17 Etensons Soft constrant on decreasng consumpton Allow some decreases wth some penalty Lag on sales Watng perod on sale of rsy assets (e.g. 60- day perod) Multple assets Allocaton bounds 18 9

Conclusons Can formulate nfnte-horzon nvestment problem n stochastc programmng framewor Soluton wth cuttng plane method Convergence wth some condtons Results for trade-restrcted assets sgnfcantly dfferent from maret assets wth same rs characterstcs 19 Approach Applcaton of typcal stochastc programmng approach complcated by nfnte horzon Q δt ( ) ma p ( c + e Q( ) s. t. A b T Intalzaton. Defne a vald constrant on Q() 0 0 ( ) E e Q + Requres problem nowledge. For optmal consumpton problem assume etremely hgh rate of consumpton forever 20 10

11 21 Approach (cont.) Iteraton ( ) ( ) ( ) where V U Fnd mn γ Epensve search over possble for the optmal consumpton problem because of small number of varables ( ) { } and e E V mn 0 + ( ) ( ) 1 0.. ma + Θ Θ + e E T b A s t e c p U t K δ ( ) + > e b p e T p E defne a new cut Else If ρ µ µ ε γ termnate.