This document presents a discrete valuation methodology for swing options using a forest model approach. It develops numerical implementations of swing options on one-factor and two-factor mean-reverting underlying processes using binomial trees. It establishes convergence via finite difference methods and considers qualitative properties and sensitivity analysis. The methodology values swing options as a system of coupled European options and allows for various discrete models of the underlying process.
This document discusses constraints on parity violating conformal field theories in 3 dimensions from the positivity of energy flux. It considers local perturbations created by the stress tensor or conserved current with specific polarizations. The energy measured in different directions is defined and required to be positive. This leads to constraints on the parameter space of 3-point functions involving the stress tensor and current. Explicit calculations are shown for the energy matrix corresponding to charge excitations, with off-diagonal elements arising from parity violating terms in 3-point functions. Positivity of energy flux implies an inequality constraint on the parameters.
This document discusses conformal field theories and three point functions. It begins by introducing conformal field theories and noting that three point functions of stress tensors are constrained by conformal invariance. For parity even CFTs in d=4 dimensions, the three point functions of the stress tensor T can be written in terms of coefficients that correspond to effective numbers of scalars, fermions, and vectors. Positivity of energy flux from collider experiments imposes constraints on these coefficients. The document then discusses extending such analyses to parity violating CFTs in d=3 dimensions, where the three point functions of the stress tensor T and conserved current j contain both parity even and odd terms. It proposes studying modifications to existing constraints from applying coll
IJRET : International Journal of Research in Engineering and Technology is an international peer reviewed, online journal published by eSAT Publishing House for the enhancement of research in various disciplines of Engineering and Technology. The aim and scope of the journal is to provide an academic medium and an important reference for the advancement and dissemination of research results that support high-level learning, teaching and research in the fields of Engineering and Technology. We bring together Scientists, Academician, Field Engineers, Scholars and Students of related fields of Engineering and Technology
This document proposes generalized additive models (GAMs) to model conditional dependence structures between random variables. Specifically, it develops a GAM framework where a dependence or concordance measure between two variables is modeled as a parametric, non-parametric, or semi-parametric function of explanatory variables. It derives the root-n consistency and asymptotic normality of the maximum penalized log-likelihood estimator for the proposed GAMs. It also discusses details of the estimation procedure and selection of smoothing parameters.
Numerical method for pricing american options under regime Alexander Decker
This document presents a numerical method for pricing American options under regime-switching jump-diffusion models. It begins with an abstract that describes using a cubic spline collocation method to solve a set of coupled partial integro-differential equations (PIDEs) with the free boundary feature. The document then provides background on regime-switching Lévy processes and derives the PIDEs that describe the American option price under different regimes. It presents the time and spatial discretization methods, using Crank-Nicolson for time stepping and cubic spline collocation for the spatial variable. The method is shown to exhibit second order convergence in space and time.
Regularity and complexity in dynamical systemsSpringer
This chapter discusses how variational methods have been used to analyze three classes of snakelike robots: 1) hyper-redundant manipulators guided by backbone curves, 2) flexible steerable needles, and 3) concentric tube continuum robots. Variational methods provide a means to determine optimal backbone curves for manipulators, generate optimal plans for needle steering, and model equilibrium conformations for concentric tube robots based on elastic mechanics principles. The chapter reviews how variational formulations using Euler-Lagrange and Euler-Poincare equations are applied in each case.
Cointegration and Long-Horizon Forecastingمحمد إسماعيل
This document summarizes research on comparing the accuracy of long-horizon forecasts from multivariate cointegrated systems versus univariate models that ignore cointegration. The main findings are:
1) When accuracy is measured using standard trace mean squared error, imposing cointegration provides no benefit over univariate models at long horizons.
2) Both multivariate and univariate long-horizon forecasts satisfy the cointegrating relationships exactly.
3) The cointegrating combinations of forecast errors from both approaches have finite variance at long horizons.
This document discusses constraints on parity violating conformal field theories in 3 dimensions from the positivity of energy flux. It considers local perturbations created by the stress tensor or conserved current with specific polarizations. The energy measured in different directions is defined and required to be positive. This leads to constraints on the parameter space of 3-point functions involving the stress tensor and current. Explicit calculations are shown for the energy matrix corresponding to charge excitations, with off-diagonal elements arising from parity violating terms in 3-point functions. Positivity of energy flux implies an inequality constraint on the parameters.
This document discusses conformal field theories and three point functions. It begins by introducing conformal field theories and noting that three point functions of stress tensors are constrained by conformal invariance. For parity even CFTs in d=4 dimensions, the three point functions of the stress tensor T can be written in terms of coefficients that correspond to effective numbers of scalars, fermions, and vectors. Positivity of energy flux from collider experiments imposes constraints on these coefficients. The document then discusses extending such analyses to parity violating CFTs in d=3 dimensions, where the three point functions of the stress tensor T and conserved current j contain both parity even and odd terms. It proposes studying modifications to existing constraints from applying coll
IJRET : International Journal of Research in Engineering and Technology is an international peer reviewed, online journal published by eSAT Publishing House for the enhancement of research in various disciplines of Engineering and Technology. The aim and scope of the journal is to provide an academic medium and an important reference for the advancement and dissemination of research results that support high-level learning, teaching and research in the fields of Engineering and Technology. We bring together Scientists, Academician, Field Engineers, Scholars and Students of related fields of Engineering and Technology
This document proposes generalized additive models (GAMs) to model conditional dependence structures between random variables. Specifically, it develops a GAM framework where a dependence or concordance measure between two variables is modeled as a parametric, non-parametric, or semi-parametric function of explanatory variables. It derives the root-n consistency and asymptotic normality of the maximum penalized log-likelihood estimator for the proposed GAMs. It also discusses details of the estimation procedure and selection of smoothing parameters.
Numerical method for pricing american options under regime Alexander Decker
This document presents a numerical method for pricing American options under regime-switching jump-diffusion models. It begins with an abstract that describes using a cubic spline collocation method to solve a set of coupled partial integro-differential equations (PIDEs) with the free boundary feature. The document then provides background on regime-switching Lévy processes and derives the PIDEs that describe the American option price under different regimes. It presents the time and spatial discretization methods, using Crank-Nicolson for time stepping and cubic spline collocation for the spatial variable. The method is shown to exhibit second order convergence in space and time.
Regularity and complexity in dynamical systemsSpringer
This chapter discusses how variational methods have been used to analyze three classes of snakelike robots: 1) hyper-redundant manipulators guided by backbone curves, 2) flexible steerable needles, and 3) concentric tube continuum robots. Variational methods provide a means to determine optimal backbone curves for manipulators, generate optimal plans for needle steering, and model equilibrium conformations for concentric tube robots based on elastic mechanics principles. The chapter reviews how variational formulations using Euler-Lagrange and Euler-Poincare equations are applied in each case.
Cointegration and Long-Horizon Forecastingمحمد إسماعيل
This document summarizes research on comparing the accuracy of long-horizon forecasts from multivariate cointegrated systems versus univariate models that ignore cointegration. The main findings are:
1) When accuracy is measured using standard trace mean squared error, imposing cointegration provides no benefit over univariate models at long horizons.
2) Both multivariate and univariate long-horizon forecasts satisfy the cointegrating relationships exactly.
3) The cointegrating combinations of forecast errors from both approaches have finite variance at long horizons.
ESTIMATION OF
NONLINEAR
TIMESERIES WITH
MATHEMATICAL
MODELS-1, How to estimate with nonlinear time series?
45/5000
How to make a estimation with R programming language.
Clustering CDS: algorithms, distances, stability and convergence ratesGautier Marti
Talk given at CMStatistics 2016 (http://cmstatistics.org/CMStatistics2016/).
The standard methodology for clustering financial time series is quite brittle to outliers / heavy-tails for many reasons: Single Linkage / MST suffers from the chaining phenomenon; Pearson correlation coefficient is relevant for Gaussian distributions which is usually not the case for financial returns (especially for credit derivatives). At Hellebore Capital Ltd, we strive to improve the methodology and to ground it. We think that stability is a paramount property to verify, which is closely linked to statistical convergence rates of the methodologies (combination of clustering algorithms and dependence estimators). This gives us a model selection criterion: The best clustering methodology is the methodology that can reach a given 'accuracy' with the minimum sample size.
You may have already read many times that the job of a Data Scientist is to skim through a huge amount of data searching for correlations between some variables of interest. And also, that one of his worst enemies (besides correlation doesn't imply causation) is spurious correlation. But what really is correlation? Are there several types of correlations? Some "good", some "bad"? What about their estimation? This talk will be a very visual presentation around the notion of correlation and dependence. I will first illustrate how the standard linear correlation is estimated (Pearson coefficient), then some more robust alternative: the Spearman coefficient. Building on the geometric understanding of their nature, I will present a generalization that can help Data Scientists to explore, interpret, and measure the dependence (not necessarily linear or comonotonic) between the variables of a given dataset. Financial time series (stocks, credit default swaps, fx rates), and features from the UCI datasets are considered as use cases.
This document summarizes a research paper that proposes using symbolic determinants of adjacency matrices to solve the Hamiltonian Cycle Problem (HCP). The HCP involves finding a cycle in a graph that visits each vertex exactly once. The paper shows that the HCP can be reduced to solving a system of polynomial equations related to the graph's adjacency matrix. Specifically, it represents the matrix symbolically and derives equations constraining the symbols to represent a Hamiltonian cycle. Solving this system of equations determines if the original graph has a Hamiltonian cycle.
The document describes a three-parameter generalized inverse Weibull (GIW) distribution that can model failure rates. Key properties of the GIW distribution include:
- It reduces to the inverse Weibull distribution when the shape parameter γ equals 1.
- Its probability density function, survival function, and hazard function are defined.
- Formulas are provided for the moments, moment generating function, and Shannon entropy of the GIW distribution.
- Methods are described for maximum likelihood estimation of the GIW distribution parameters from censored lifetime data.
The document discusses distributivity of fuzzy implications over t-norms and t-conorms. Specifically, it aims to solve the open problem of characterizing functions I that satisfy the equation I(x, S1(y, z)) = S2(I(x, y), I(x, z)), where S1 and S2 are strict or nilpotent t-conorms and I is an R-implication from a strict t-norm. It provides background on t-norms, t-conorms, and fuzzy implications. It then characterizes solutions I when S1 and S2 are strict or nilpotent t-conorms, showing there are no
Effective properties of heterogeneous materialsSpringer
1) The document discusses the multipole expansion method (MEM) for analyzing the microstructure and effective properties of composite materials.
2) MEM reduces boundary value problems for heterogeneous materials to systems of linear algebraic equations. It expresses fields like temperature and stress as expansions of basis functions related to inclusion geometry.
3) MEM has been applied to analyze conductivity and elasticity problems in composites with spherical, spheroidal, circular, and elliptic inclusions. It provides analytical solutions for local fields and exact expressions for effective properties involving only dipole moments.
This document presents a novel Copula based approach to generate critical sea states given a target reliability index based on the return period of the extreme event. Copula based approach is much more flexible and powerful when compared to conventional approaches using linear correlation coefficient.
This paper proposes testing for integration and threshold integration between interest rates and inflation rates. It examines whether there is a cointegrating relationship between the variables and addresses issues of structural breaks. The paper analyzes inflation and interest rates in Canada using cointegration, threshold autoregressive (TAR), and momentum threshold autoregressive (MTAR) models to test for nonlinear relationships. The results show the variables are integrated at level one, there is cointegration between interest rates and inflation, and the TAR model best captures the adjustment process. No asymmetry is found, indicating inflation increases and decreases have the same effect on interest rates.
The document provides an overview of dimensional analysis and its applications in fluid mechanics. Dimensional analysis is a tool used to correlate analytical and experimental results and predict prototype behavior from model measurements. It involves identifying the fundamental dimensions of variables (e.g. mass, length, time) and establishing dimensionless relationships between variables using methods like Rayleigh's or Buckingham's π-theorem. Dimensional analysis is important for model analysis where dynamic similarity between a model and prototype must be achieved. The document discusses different model laws used to design models for similarity based on dominant forces like viscosity, gravity, etc. It also provides scale ratios required for geometric, kinematic and dynamic similarity.
Stress-Strength Reliability of type II compound Laplace distributionIRJET Journal
The document discusses estimating the reliability (R) of a stress-strength model where the strength (Y) and stress (X) variables follow a type II compound Laplace distribution. R is defined as the probability that stress is less than strength (Pr(X<Y)).
It first provides background on stress-strength models and reliability (R) in engineering. It then defines the type II compound Laplace distribution and derives an expression for R when X and Y follow this distribution. Maximum likelihood estimation is used to estimate the three parameters of the distribution. Simulation studies validate the estimation algorithm. Applications of stress-strength models are discussed in fields like engineering, medicine, and genetics.
- The document describes a study that uses a modified Kolmogorov-Smirnov (KS) test to test if the innovations of a GARCH model come from a mixture of normal distributions rather than a standard normal distribution.
- It establishes critical values for the KS test and modified KS (MKS) test through simulation under the null hypothesis. It then uses simulation to calculate the size and power of both tests when the innovations come from alternative distributions like the normal, Student's t, and generalized error distributions.
- The results show that the KS and MKS tests maintain the correct size when the innovations are actually from the mixture of normals. The power of both tests is greater than the nominal level when the innovations come
International Journal of Engineering Research and DevelopmentIJERD Editor
This document analyzes a discrete-time host-parasitoid model that incorporates both Allee effects for the host population and aggregation effects for the parasitoid population. The model is described using difference equations. Fixed points and local stability are analyzed mathematically. It is found that adding Allee effects can simplify the population dynamics by making the system less sensitive to initial conditions and compressing the dynamics. The inclusion of Allee effects and aggregation effects provides a more biologically realistic model of host-parasitoid interactions compared to previous models.
An improved demspter shafer algorithm for resolving conflicting eventsGauravv Prabhu
This document proposes an improved Dempster-Shafer algorithm to resolve conflicting evidence. It begins with an overview of Dempster-Shafer theory and identifies shortcomings when evidence conflicts. It then presents a new method to verify and modify conflicting evidence before combination. Experiments show the new method improves reliability by modifying conflicting evidence and producing more intuitive combination results, even when evidence highly conflicts.
Optimal Transport between Copulas for Clustering Time SeriesGautier Marti
Presentation slides of our ICASSP 2016 conference paper in Shanghai. They describe the motivation and design of the Target Dependence Coefficient, a coefficient which can target or forget specific dependence relationships between the variables. This coefficient can be useful for clustering financial time series. Several of such use-cases are described on our Tech Blog https://www.datagrapple.com/Tech/optimal-copula-transport.html
This document introduces a new class of information-theoretic divergence measures called the K and L divergences. Unlike existing measures like Kullback-Leibler divergence, the new measures do not require probability distributions to be absolutely continuous. The K and L divergences also have desirable properties like being bounded above by the variational distance. A generalized form, the Jensen-Shannon divergence, can measure differences between multiple distributions and provides both upper and lower bounds for classification error probability.
This document discusses soliton solutions in neurological systems. It first provides background on solitons and their properties. It then summarizes a recent study that found soliton-like solutions in modeling neuron signaling pathways using partial differential equations. The document numerically confirms that the proposed neuron signaling solutions exhibit characteristics of soliton waves, including maintaining shape and velocity during propagation and collisions.
Dimensional analysis is a mathematical technique used to establish relationships between physical quantities in fluid phenomena. It involves considering the dimensions of quantities and grouping dimensionless parameters to better understand flows. Dimensional analysis provides guidance for experimental work by indicating important influencing factors. It is applied to develop equations, convert between units, reduce experimental variables, and enable model studies through similitude. The key concepts are that theoretical equations must be dimensionally homogeneous and empirical equations have limited applications. Dimensional analysis methods include Rayleigh's method of exponential relationships and Buckingham's Π-method of grouping variables into dimensionless terms.
Proportional and decentralized rule mcst gamesvinnief
This document presents two cost allocation rules - the proportional rule and the decentralized rule - for minimum cost spanning extension problems. The proportional rule allocates costs proportionally based on an algorithm that constructs a minimum cost spanning extension while simultaneously allocating link costs. The decentralized rule is based on an algorithm that constructs the extension in a decentralized manner, with links being added one by one and their costs immediately allocated. Both rules are shown to be refinements of the irreducible core defined for these problems. The proportional rule is then characterized axiomatically.
IJERA (International journal of Engineering Research and Applications) is International online, ... peer reviewed journal. For more detail or submit your article, please visit www.ijera.com
ESTIMATION OF
NONLINEAR
TIMESERIES WITH
MATHEMATICAL
MODELS-1, How to estimate with nonlinear time series?
45/5000
How to make a estimation with R programming language.
Clustering CDS: algorithms, distances, stability and convergence ratesGautier Marti
Talk given at CMStatistics 2016 (http://cmstatistics.org/CMStatistics2016/).
The standard methodology for clustering financial time series is quite brittle to outliers / heavy-tails for many reasons: Single Linkage / MST suffers from the chaining phenomenon; Pearson correlation coefficient is relevant for Gaussian distributions which is usually not the case for financial returns (especially for credit derivatives). At Hellebore Capital Ltd, we strive to improve the methodology and to ground it. We think that stability is a paramount property to verify, which is closely linked to statistical convergence rates of the methodologies (combination of clustering algorithms and dependence estimators). This gives us a model selection criterion: The best clustering methodology is the methodology that can reach a given 'accuracy' with the minimum sample size.
You may have already read many times that the job of a Data Scientist is to skim through a huge amount of data searching for correlations between some variables of interest. And also, that one of his worst enemies (besides correlation doesn't imply causation) is spurious correlation. But what really is correlation? Are there several types of correlations? Some "good", some "bad"? What about their estimation? This talk will be a very visual presentation around the notion of correlation and dependence. I will first illustrate how the standard linear correlation is estimated (Pearson coefficient), then some more robust alternative: the Spearman coefficient. Building on the geometric understanding of their nature, I will present a generalization that can help Data Scientists to explore, interpret, and measure the dependence (not necessarily linear or comonotonic) between the variables of a given dataset. Financial time series (stocks, credit default swaps, fx rates), and features from the UCI datasets are considered as use cases.
This document summarizes a research paper that proposes using symbolic determinants of adjacency matrices to solve the Hamiltonian Cycle Problem (HCP). The HCP involves finding a cycle in a graph that visits each vertex exactly once. The paper shows that the HCP can be reduced to solving a system of polynomial equations related to the graph's adjacency matrix. Specifically, it represents the matrix symbolically and derives equations constraining the symbols to represent a Hamiltonian cycle. Solving this system of equations determines if the original graph has a Hamiltonian cycle.
The document describes a three-parameter generalized inverse Weibull (GIW) distribution that can model failure rates. Key properties of the GIW distribution include:
- It reduces to the inverse Weibull distribution when the shape parameter γ equals 1.
- Its probability density function, survival function, and hazard function are defined.
- Formulas are provided for the moments, moment generating function, and Shannon entropy of the GIW distribution.
- Methods are described for maximum likelihood estimation of the GIW distribution parameters from censored lifetime data.
The document discusses distributivity of fuzzy implications over t-norms and t-conorms. Specifically, it aims to solve the open problem of characterizing functions I that satisfy the equation I(x, S1(y, z)) = S2(I(x, y), I(x, z)), where S1 and S2 are strict or nilpotent t-conorms and I is an R-implication from a strict t-norm. It provides background on t-norms, t-conorms, and fuzzy implications. It then characterizes solutions I when S1 and S2 are strict or nilpotent t-conorms, showing there are no
Effective properties of heterogeneous materialsSpringer
1) The document discusses the multipole expansion method (MEM) for analyzing the microstructure and effective properties of composite materials.
2) MEM reduces boundary value problems for heterogeneous materials to systems of linear algebraic equations. It expresses fields like temperature and stress as expansions of basis functions related to inclusion geometry.
3) MEM has been applied to analyze conductivity and elasticity problems in composites with spherical, spheroidal, circular, and elliptic inclusions. It provides analytical solutions for local fields and exact expressions for effective properties involving only dipole moments.
This document presents a novel Copula based approach to generate critical sea states given a target reliability index based on the return period of the extreme event. Copula based approach is much more flexible and powerful when compared to conventional approaches using linear correlation coefficient.
This paper proposes testing for integration and threshold integration between interest rates and inflation rates. It examines whether there is a cointegrating relationship between the variables and addresses issues of structural breaks. The paper analyzes inflation and interest rates in Canada using cointegration, threshold autoregressive (TAR), and momentum threshold autoregressive (MTAR) models to test for nonlinear relationships. The results show the variables are integrated at level one, there is cointegration between interest rates and inflation, and the TAR model best captures the adjustment process. No asymmetry is found, indicating inflation increases and decreases have the same effect on interest rates.
The document provides an overview of dimensional analysis and its applications in fluid mechanics. Dimensional analysis is a tool used to correlate analytical and experimental results and predict prototype behavior from model measurements. It involves identifying the fundamental dimensions of variables (e.g. mass, length, time) and establishing dimensionless relationships between variables using methods like Rayleigh's or Buckingham's π-theorem. Dimensional analysis is important for model analysis where dynamic similarity between a model and prototype must be achieved. The document discusses different model laws used to design models for similarity based on dominant forces like viscosity, gravity, etc. It also provides scale ratios required for geometric, kinematic and dynamic similarity.
Stress-Strength Reliability of type II compound Laplace distributionIRJET Journal
The document discusses estimating the reliability (R) of a stress-strength model where the strength (Y) and stress (X) variables follow a type II compound Laplace distribution. R is defined as the probability that stress is less than strength (Pr(X<Y)).
It first provides background on stress-strength models and reliability (R) in engineering. It then defines the type II compound Laplace distribution and derives an expression for R when X and Y follow this distribution. Maximum likelihood estimation is used to estimate the three parameters of the distribution. Simulation studies validate the estimation algorithm. Applications of stress-strength models are discussed in fields like engineering, medicine, and genetics.
- The document describes a study that uses a modified Kolmogorov-Smirnov (KS) test to test if the innovations of a GARCH model come from a mixture of normal distributions rather than a standard normal distribution.
- It establishes critical values for the KS test and modified KS (MKS) test through simulation under the null hypothesis. It then uses simulation to calculate the size and power of both tests when the innovations come from alternative distributions like the normal, Student's t, and generalized error distributions.
- The results show that the KS and MKS tests maintain the correct size when the innovations are actually from the mixture of normals. The power of both tests is greater than the nominal level when the innovations come
International Journal of Engineering Research and DevelopmentIJERD Editor
This document analyzes a discrete-time host-parasitoid model that incorporates both Allee effects for the host population and aggregation effects for the parasitoid population. The model is described using difference equations. Fixed points and local stability are analyzed mathematically. It is found that adding Allee effects can simplify the population dynamics by making the system less sensitive to initial conditions and compressing the dynamics. The inclusion of Allee effects and aggregation effects provides a more biologically realistic model of host-parasitoid interactions compared to previous models.
An improved demspter shafer algorithm for resolving conflicting eventsGauravv Prabhu
This document proposes an improved Dempster-Shafer algorithm to resolve conflicting evidence. It begins with an overview of Dempster-Shafer theory and identifies shortcomings when evidence conflicts. It then presents a new method to verify and modify conflicting evidence before combination. Experiments show the new method improves reliability by modifying conflicting evidence and producing more intuitive combination results, even when evidence highly conflicts.
Optimal Transport between Copulas for Clustering Time SeriesGautier Marti
Presentation slides of our ICASSP 2016 conference paper in Shanghai. They describe the motivation and design of the Target Dependence Coefficient, a coefficient which can target or forget specific dependence relationships between the variables. This coefficient can be useful for clustering financial time series. Several of such use-cases are described on our Tech Blog https://www.datagrapple.com/Tech/optimal-copula-transport.html
This document introduces a new class of information-theoretic divergence measures called the K and L divergences. Unlike existing measures like Kullback-Leibler divergence, the new measures do not require probability distributions to be absolutely continuous. The K and L divergences also have desirable properties like being bounded above by the variational distance. A generalized form, the Jensen-Shannon divergence, can measure differences between multiple distributions and provides both upper and lower bounds for classification error probability.
This document discusses soliton solutions in neurological systems. It first provides background on solitons and their properties. It then summarizes a recent study that found soliton-like solutions in modeling neuron signaling pathways using partial differential equations. The document numerically confirms that the proposed neuron signaling solutions exhibit characteristics of soliton waves, including maintaining shape and velocity during propagation and collisions.
Dimensional analysis is a mathematical technique used to establish relationships between physical quantities in fluid phenomena. It involves considering the dimensions of quantities and grouping dimensionless parameters to better understand flows. Dimensional analysis provides guidance for experimental work by indicating important influencing factors. It is applied to develop equations, convert between units, reduce experimental variables, and enable model studies through similitude. The key concepts are that theoretical equations must be dimensionally homogeneous and empirical equations have limited applications. Dimensional analysis methods include Rayleigh's method of exponential relationships and Buckingham's Π-method of grouping variables into dimensionless terms.
Proportional and decentralized rule mcst gamesvinnief
This document presents two cost allocation rules - the proportional rule and the decentralized rule - for minimum cost spanning extension problems. The proportional rule allocates costs proportionally based on an algorithm that constructs a minimum cost spanning extension while simultaneously allocating link costs. The decentralized rule is based on an algorithm that constructs the extension in a decentralized manner, with links being added one by one and their costs immediately allocated. Both rules are shown to be refinements of the irreducible core defined for these problems. The proportional rule is then characterized axiomatically.
IJERA (International journal of Engineering Research and Applications) is International online, ... peer reviewed journal. For more detail or submit your article, please visit www.ijera.com
This document discusses open educational resources (OER) and open education practices. It reports on surveys and focus groups conducted with academics to understand their awareness and use of OER. The results found that while about a third of respondents were aware of OER and their uses, only 9% shared resources through repositories. It also describes a research project aimed at transforming university teachers into "agents of change" for openness. Finally, it outlines roles and a procedure for a debate on whether third level educators should aspire to be "open" educators.
El documento describe tres sistemas principales de una computadora: el sistema operativo, el sistema de aplicaciones y el sistema de programación. El sistema operativo permite la interacción entre el usuario y la computadora y administra los dispositivos. El sistema de aplicaciones incluye programas diseñados para realizar tareas específicas como hojas de cálculo y bases de datos. El sistema de programación proporciona herramientas para que los programadores desarrollen programas como JavaScript, Ruby y Perl.
El documento es un extracto de la novela corta "El pozo y el péndulo" de Edgar Allan Poe. Narra la historia de un hombre que es juzgado y condenado a muerte por la Inquisición española. Después de desmayarse durante la sentencia, despierta en un lugar completamente oscuro sin saber dónde está ni cuál será su destino. Comienza a explorar tanteando las paredes con sus manos, presa del terror ante la posibilidad de haber sido enterrado vivo.
Este documento presenta un tutorial básico sobre Java EE. Explica conceptos como HTTP, HTML, hojas de estilo CSS, servlets, sesiones, cookies, compartir información entre servlets, manejo de errores, problemas de concurrencia, bases de datos, listeners, filtros y páginas JSP. El tutorial contiene 228 páginas e incluye ejemplos de código para cada tema.
Eliza Boon is an 18-year-old university student studying environmental design who is looking for part-time work in Launceston, where she will be relocating. She has over 3 years of experience in retail and customer service roles at various stores. She has completed high school and is currently studying at university. Her resume lists relevant skills such as communication, organization, working under pressure, and food safety certifications. She provides two referees from past employers who can speak to her work experience.
'Troca de Família': espiando as famílias, enxergando a sociedade.Jefferson Sampaio
1) O documento analisa o programa de TV "Troca de Família" exibido no Brasil, que segue o formato do reality show norte-americano "Trading Spouses".
2) O programa coloca mães de famílias com estilos de vida diferentes para viverem 5 dias na casa uma da outra.
3) A análise discute como o programa evidencia diferenças culturais e valores entre as famílias, além de explorar conflitos de uma forma que atrai a audiência.
La refracción es el cambio de dirección que experimenta una onda al pasar de un medio a otro debido a los cambios en la velocidad de propagación. Cuando un rayo de luz pasa de un medio a otro con diferente densidad, cambia su velocidad lo que genera un cambio en su dirección formando un ángulo de refracción que depende del índice de refracción de cada medio.
La canción habla sobre la atracción intensa que siente el cantante hacia una mujer cuando bailan juntos, describiendo cómo su cuerpo y el de ella se mueven al ritmo de la música, y cómo esto enciende una pasión que ya no puede controlar y que lo lleva a desear estar con ella bailando y besándose toda la noche.
Este blog tratará sobre diversos temas presentados generalmente en listas de "Top 10", también incluirá un "Personaje de la Semana" ocasionalmente. Habrá votaciones y otras actividades para los lectores.
El día de San Valentín es una celebración tradicional de países anglosajones que se ha extendido a otros lugares en el siglo XX para que las parejas expresen su amor mutuamente el 14 de febrero. En Colombia se celebra el tercer sábado de septiembre como el Día del Amor y la Amistad, con tradiciones como el "amigo secreto" y el "endulzar". Aunque algunos creen que es una celebración reciente, su origen se remonta a la época del Imperio Romano cuando el sacerdote San Valent
Este documento describe cómo insertar y reproducir un video en una aplicación de Android. Primero se crea un proyecto llamado ReproducirVideo con una actividad MainActivity. Luego se agrega un archivo de video a la carpeta raw y se vincula a un VideoView en el layout. Finalmente, el código de la actividad establece la ruta del video y reproduce el VideoView. Se agregan controles de medios para controlar la reproducción del video.
Este documento presenta un mapa conceptual sobre un plan de negocios desarrollado por Hernaly Ramos, estudiante de la Universidad Fermín Toro en Cabudare, Venezuela. El mapa conceptual detalla los componentes clave de un plan de negocios como la misión, visión, objetivos, análisis FODA y estrategias.
Trabajo en slideshare will polanco coronadowillpolanco
Este documento presenta una actividad realizada por Will Polanco Coronado para su clase sobre cultura ciudadana. Consiste en 10 preguntas y respuestas sobre temas como la universidad soñada, formas de aprendizaje, y soluciones a problemas de tráfico cerca de la Universidad del Atlántico. El estudiante propone mejorar la seguridad vial a través de semáforos y más presencia policial para regular el tráfico.
El documento compara aspectos de la comunicación, transporte, iluminación, electrónica, entretenimiento y medicina en el pasado y en la actualidad, describiendo algunas de las tecnologías y métodos utilizados anteriormente como humo, cartas y palomas mensajeras para la comunicación; bicicleta y barco de vapor para el transporte; velas y mecheros de gas para la iluminación; y ausencia de anestesia y odontología en medicina, contrastando con los avances actuales como internet, celulares, lámparas LED, pantallas plan
El documento describe los pasos para crear y ejecutar un programa en NetBeans. Primero se abre NetBeans y se selecciona "archivo" para crear un nuevo proyecto Java. Luego se edita el nombre del proyecto a "decimo" y se termina el proceso. Finalmente se muestra la pantalla donde se puede empezar a crear el décimo programa.
Este documento resume la sinopsis de la película Scary Movie. Cuenta que un grupo de estudiantes recuerdan haber atropellado accidentalmente a un pescador mientras conducían borrachos en Halloween, y creen que esto puede estar relacionado con el asesinato de una compañera de clases. El documento también incluye la ficha técnica de la película, sus protagonistas y una breve crítica que señala que la película se burla de clichés de películas de terror pero contiene demasiado humor negro.
El documento presenta un resumen de las funciones de Microsoft Excel, incluyendo funciones de fecha y hora, de texto, de búsqueda y financieras. Define cada función y proporciona ejemplos de su uso.
Metaheuristic Optimization: Algorithm Analysis and Open ProblemsXin-She Yang
This document analyzes metaheuristic optimization algorithms and discusses open problems in their analysis. It reviews convergence analyses that have been done for simulated annealing and particle swarm optimization. It also provides a novel convergence analysis for the firefly algorithm, showing that it can converge for certain parameter values but also exhibit chaos which can be advantageous for exploration. The document outlines the need for further mathematical analysis of convergence and efficiency in metaheuristics.
This document discusses time series models and their application to unemployment forecasting. It summarizes that:
- Time series models like autoregressive (AR) and moving average (MA) models are commonly used to understand and predict economic data over time.
- The document will apply an ARMA(3,3) model to seasonally adjusted unemployment data and use a Kalman filter on non-seasonally adjusted data to generate forecasts.
- It will compare forecasts from seasonally adjusted vs. non-adjusted data at different time intervals.
Algorithms Behind Term Structure Models II Hull-White ModelJeff Brooks
This document summarizes the Hull-White trinomial tree model for pricing interest rate derivatives. It describes:
1) The Hull-White model uses a trinomial tree to model the short-term interest rate as following either a Vasicek or Black-Karasinski process over discrete time steps.
2) Negative interest rates are avoided by manipulating the tree's branching probabilities or geometry. Three alternative branching processes are introduced to control rates.
3) The tree is constructed in two steps - first a "level tree" approximates the short rate process, then the tree probabilities are set to match the process expectations and variance.
This document describes a stock price model based on the voter model from statistical physics. The model represents stock investors on a lattice who can be in a buying, selling, or neutral position based on interactions with neighboring investors. The stock price is then derived from the positions of investors. Computer simulations of the 1D model are presented, showing the fluctuations of generated stock prices and returns. The document also discusses properties found in real stock market data, such as power law distributions and volatility clustering, that the model aims to replicate.
- Dimensional analysis is a technique used to determine the relationship between variables in a physical phenomenon based on their dimensions and units.
- It allows reducing the number of variables needed to describe a phenomenon through the use of dimensionless parameters known as π terms.
- Lord Rayleigh and Buckingham developed systematic methods for dimensional analysis. Buckingham's π-method involves identifying all variables, their dimensions, and grouping them into as many dimensionless π terms as needed to describe the phenomenon.
1. The document discusses scaling hypothesis and statistical mechanics in the context of critical phenomena during phase transitions. It introduces concepts like critical exponents, universality, fractals, and ergodicity breaking.
2. It presents Rushbrooke's and Widom's identities which relate critical exponents to each other based on the assumption of homogeneity. This allows reducing the number of independent exponents needed to describe critical behavior.
3. The identities are consistent with experimental data and exact solutions, supporting the scaling hypothesis over saddle-point approximations for describing critical phenomena.
This document describes specification tests that can be used after estimating dynamic panel data models using the generalized method of moments (GMM) estimator. It presents GMM estimators for first-order autoregressive models with individual fixed effects that exploit moment restrictions from assuming serially uncorrelated errors. Monte Carlo simulations are used to evaluate the small-sample performance of tests of serial correlation based on GMM residuals, Sargan tests, and Hausman tests. The tests are also applied to estimated employment equations using an unbalanced panel of UK firms.
Monotonicity of Phaselocked Solutions in Chains and Arrays of Nearest-Neighbo...Liwei Ren任力偉
This document summarizes a paper on phaselocked solutions in chains and arrays of coupled oscillators. It introduces the topic of coupled oscillators and their importance in modeling neural activity. It describes previous work analyzing one-dimensional chains of oscillators using continuum approximations. The current paper aims to rigorously prove the existence of phaselocked solutions in chains without requiring a continuum limit. It also analyzes two-dimensional arrays by decomposing them into independent one-dimensional problems under certain frequency distributions. Key results include proving monotonicity of phaselocked solutions and spontaneous formation of target patterns in two-dimensional arrays with isotropic synaptic coupling.
This study deals with the active control of the dynamic response of a string with fixed ends and mass
loaded by a point mass. It has been controlled actively by means of a feed forward control method. A point mass of a
string is considered as a vibrating receiver which be forced to vibrate by a vibrating source being positioned on the
string. By analyzing the motion of a string, the equation of motion for a string was derived by using a method of
variation of parameters. To define the optimal conditions of a controller, the cost function, which denotes the dynamic
response at the point mass of a string was evaluated numerically. The possibility of reduction of a dynamic response
was found to depend on the location of a control force, the magnitude of a point mass and a forcing frequency
Spillover Dynamics for Systemic Risk Measurement Using Spatial Financial Time...SYRTO Project
Spillover Dynamics for Systemic Risk Measurement Using Spatial Financial Time Series Models. Andre Lucas. Amsterdam - June, 25 2015. European Financial Management Association 2015 Annual Meetings.
MULTI RESOLUTION LATTICE DISCRETE FOURIER TRANSFORM (MRL-DFT)csandit
In many imaging applications, including sensor arrays, MRI and CT,data is often sampled on
non-rectangular point sets with non-uniform density. Moreover, in image and video processing,
a mix of non-rectangular sampling structures naturally arise. Multirate processing typically
utilizes a normalized integer indexing scheme, which masks the true physical dimensions of the
points. However, the spatial correlation of such signals often contains important
information.This paper presents a theory of signals defined on regular discrete sets called
lattices, and presents an associated form of a finite Fourier transform denoted here as
multiresolution lattice discrete Fourier transform (MRL-DFT). Multirate processing techniques
such as decimation, interpolation and polyphase representations are presented in a context
which preserves the true spatial dimensions of the sampling structure.Moreover, the polyphase
formulation enables systematic representation and processing for sampling patterns with
variable spatial density, and provides a framework for developing generalized FFT and
regridding algorithms.
The document applies the variational iteration method (VIM) to solve linear and nonlinear ordinary differential equations (ODEs) with variable coefficients. It emphasizes the power of the method by using it to solve a variety of ODE models of different orders and coefficients. The document also uses VIM to solve four scientific models - the hybrid selection model, Thomas-Fermi equation, Kidder equation for unsteady gas flow through porous media, and the Riccati equation. The VIM provides efficient iterative approximations for both analytic solutions and numeric simulations of real-world applications in science and engineering.
Application of smooth transition autoregressive (star) models for exchange rateAlexander Decker
This document discusses applying Smooth Transition Autoregressive (STAR) models to analyze exchange rate data. Specifically, it evaluates the suitability of Logistic STAR (LSTAR) and Exponential STAR (ESTAR) models. The document outlines the methodology for testing whether exchange rate data exhibits linear or nonlinear properties. It describes testing procedures that involve estimating an auxiliary regression model based on a Taylor series expansion to detect nonlinear behavior. The study applies these LSTAR and ESTAR models to exchange rate data to determine which model provides a better fit and explanation of deviations from the mean exchange rate. The document concludes that an ESTAR model provides the best adjustment for the analyzed exchange rate data series.
The document discusses various econometric modeling techniques including regression equations, cointegration, error correction models, vector autoregressive (VAR) modeling, and vector error correction models (VECM). It explains that regression equations can produce spurious results if the data is non-stationary, and that cointegration exists if the residuals from a regression equation are stationary. Error correction models specify the short-run relationship that maintains the long-run equilibrium between cointegrated variables. VAR models express current values of variables as functions of past values, while VECMs are VARs in first differences that incorporate the long-run cointegrating relationships between variables.
FDM is an older method than FEM that requires less computational power but is also less accurate in some cases where higher-order accuracy is required. FEM permit to get a higher order of accuracy, but requires more computational power and is also more exigent on the quality of the mesh.29-Jun-2017
https://feaforall.com › difference-bet...
What's the difference between FEM and FDM? - FEA for All
Feedback
About featured snippets
International Journal of Engineering Research and Applications (IJERA) is an open access online peer reviewed international journal that publishes research and review articles in the fields of Computer Science, Neural Networks, Electrical Engineering, Software Engineering, Information Technology, Mechanical Engineering, Chemical Engineering, Plastic Engineering, Food Technology, Textile Engineering, Nano Technology & science, Power Electronics, Electronics & Communication Engineering, Computational mathematics, Image processing, Civil Engineering, Structural Engineering, Environmental Engineering, VLSI Testing & Low Power VLSI Design etc.
The document discusses several real-world applications of differential and integral calculus. It provides examples of first-order differential equations being used to model jumping motions in video games and the cooling of objects. Surface and volume integrals are applied in fields like electrostatics, fluid dynamics, and continuity equations. Matrix determinants can estimate areas like that of the Bermuda Triangle. Overall, calculus has wide applications in science, engineering, economics and other domains.
This document discusses dimensional analysis and model analysis. It begins by introducing dimensional analysis as a technique that uses the dimensions of physical quantities to understand phenomena. It then describes the two types of dimensions: fundamental dimensions like length, time, and mass, and secondary dimensions that are combinations of fundamental ones, like velocity.
It presents the methodology of dimensional analysis, which requires equations to be dimensionally homogeneous. Two common methods are described: Rayleigh's method and Buckingham's pi-theorem. Rayleigh's method can be used for up to 4 variables while Buckingham's pi-theorem groups variables into dimensionless pi-terms. Model analysis is introduced as an experimental method using scale models, and the importance of similitude or
This document provides an introduction to Markov jump linear systems (MJLS). It begins by discussing how dynamical systems are increasingly complex due to technological advances, requiring more efficient control systems. MJLS are introduced as a class of models that can account for abrupt changes in system dynamics by allowing the system to switch among a set of linear models based on a Markov process. Key characteristics of MJLS are described, including that the overall system behavior depends on the switching rate between models. Examples are provided to illustrate how MJLS can exhibit different properties than traditional linear systems due to switching dynamics.
The document describes a study that investigated parameter behaviors related to stationarity in autoregressive (AR) and moving average (MA) models through simulations. The study conducted simulations using R at different parameter orders (p for AR, q for MA) and sample sizes. It analyzed properties like the mean, variance, autocorrelation function (ACF), and partial autocorrelation function (PACF). The study found that the ACF and PACF values increased with higher parameter values but decreased with higher orders, eventually tending to zero at larger lags. It also observed that means and variances declined with increasing sample size.
1. A DISCRETE VALUATION OF SWING OPTIONS
ALI LARI-LAVASSANI, MOHAMADREZA SIMCHI, AND ANTONY WARE
Abstract. A discrete forest methodology is developed for swing options as
a dynamically coupled system of European options. Numerical implementa-
tion is fully developed for one- and two-factor, mean-reverting, underlying
processes, with application to energy markets. Convergence is established via
finite difference methods. Qualitative properties and sensitivity analysis are
considered.
1. Introduction
Swing contracts have traditionally been used in the natural gas industry to pro-
vide hedging via the introduction of limited dynamic flexibility in the quantity of
gas acquired. More generally this type of option is of value in any market where the
physical transfer of the underlying asset must take place through interconnected
networks, and is thus subject to volume constraints. This is the case for natural
gas and pipelines, electricity and cable based telecommunications and their trans-
mission lines, wireless telecommunications and their bandwidths. Going beyond
issues of trading, dynamic risk management of real assets such as these complex
interconnected networks could benefit from the methodology presented here.
Our aim in writing this paper is two-fold. First, in spite of their important appli-
cations, little fundamental work seems to have been done in modeling and valuing
swing options; the few available published papers are qualitative in nature and do
not provide a mathematical model or a solid ground for numerical implementation.
Secondly, mean-reverting processes are increasingly being used to model the price
processes of commodities and energy; we provide here a unified treatment of the
numerical implementation of discrete mean-reverting processes.
The swing option can be viewed as a generalization of a Bermudan option, due
to its multiple exercise feature. We model it as a dynamically coupled system of
European options, and opt for a discrete forest model, which results in a realistic
and practical numerical algorithm. We limit ourselves to modeling a swing op-
tion on one given underlying asset. Generalization to the case of multiple assets
Date: July 11, 2000.
1991 Mathematics Subject Classification. Primary 91B28, 91-08; Secondary 65C.
Key words and phrases. Option Pricing, Swing Option, Energy Commodity Modeling and
Derivatives, Mean-reverting Models, Two-Factor Model, Forest, Numerical Methods, Finite
Difference.
A. Lari-Lavassani is partially supported by a grant from the University of Calgary and NSERC.
M. Simchi is supported by grants from the University of Calgary and TransAlta Energy Marketing
Corp (TEM). A. Ware is supported by MITACS and a grant from TEM. All three authors are
grateful to Dan Calistrate for many constructive discussions, to Shaun Hatch of TEM for intro-
ducing them to this problem and providing valuable insight and support, and to Tom Coyle from
TEM for supporting the entire project.
This paper is in final form and no version of it will be submitted for publication elsewhere.
1
2. 2 ALI LARI-LAVASSANI, MOHAMADREZA SIMCHI, AND ANTONY WARE
is straightforward but leads to challenging computational problems due to high
dimensionality. One promising avenue would be to investigate numerical imple-
mentation on parallel computer. We do however allow the single underlying asset
to follow a multi-factor stochastic process. A distinctive characterization of the
methodology developed here is its abstract and general nature, which makes it ap-
plicable to any discrete modeling of the underlying asset, in particular binomial or
trinomial trees, or indeed trees with any number of branching jumps. This abstract
setting lends itself to a European style of exercise of the swing rights. We fully
develop numerical implementations of swing options on mean-reverting underlying
processes via binomial trees.
The paper is written in a self contained manner and full arguments are presented
to the extent possible. It is organized as follows: Section 2 discusses some classical
examples of one- and two-factor mean-reverting linear stochastic differential equa-
tions. These model energy price or interest rate processes, and throughout this
work, they will be used to illustrate the methods and results. Section 3 presents a
general abstract look at discrete modeling of stochastic processes. Section 4 inves-
tigates numerical implementations of mean-reverting processes on binomial trees.
Section 5 describes a general swing option contract. Section 6 develops a mathe-
matical model for the swing option via a forest methodology. Section 7 is devoted
to numerical investigation of convergence, hedge parameters and comparison with
a basket of American options. Section 8 proves the convergence of the binomial
tree swing option valuation, showing that the upper bound on the error is inversely
proportional to the number of time steps in the tree.
2. Underlying Processes and Applications to Energy Models
The general setting for swing options developed in Section 6 is quite abstract and
applies to all discrete processes. To make ideas more transparent we first introduce
in this section one- and two-factor models historically used for short term interest
rates, and which are now finding applications in the energy commodity markets
as in [S, 1998], [P, 1997], [HR, 1998], [JRT, 1998]. When using these models to
price options, it is implicitly assumed that they reflect the risk neutralized price
processes, perhaps via the incorporation of a model of the cost-of-carry or the
market-price-of-risk functions, see [H, 1999]. This will be the working assumption
for the remainder of this paper.
A general diffusion one-factor model is:
dSt = µ(St, t)dt + σ(St, t)dZt,(2.1)
where µ and σ are C2
differentiable and satisfy the usual linear growth conditions
to guarantee the existence of solutions of the above stochastic differential equation
(see [KP, 1999] Theorem 4.5.3, p. 131), and dZt is a standard Brownian motion,
E(dZt) = 0, and E((dZt)2
) = dt. We will consider two classical particular cases:
dSt = µSt dt + σSt dZt(2.2)
dSt = α(L − St) dt + σSt dZt,(2.3)
where µ, α, σ and L are constants. The following system of stochastic linear
differential equations is a possible two-factor model for energy and commodity spot
3. A DISCRETE VALUATION OF SWING OPTION 3
price processes:
dSt = α(Lt − St) dt + σSt dZt
dLt = µ Lt dt + ξLt dWt,
(2.4)
where α, µ, σ and ξ are constants and dZt and dWt are uncorrelated standard
Brownian motions.
Ignoring the stochastic parts in the mean-reverting equations (2.3) and (2.4)
yields the deterministic differential equations governing the time evolution of the
mean of each process:
dE(St)
dt
= α(L − E(St))(2.5)
d
dt
E(St)
E(Lt)
=
−α α
0 µ
E(St)
E(Lt)
.(2.6)
We shall establish that the orbits associated with the E(Lt) equations are at-
tracting invariant spaces for the flows of these deterministic stochastic differential
equations. This justifies the use of the terms long-run mean and mean-reverting.
For more on the notions of dynamical systems introduced in the next Proposition
and its proof we refer to [R, 1995] Chapter 5.
Proposition 1. For the process St, given by (2.3), L is the stable equilibrium of
the hyperbolic differential equation governing (2.5) if α > 0. For the process St,
given by (2.4), the orbit of E(Lt) is the unstable or stable manifold of the hyperbolic
differential equation (2.6). This is the unstable manifold when α > 0, and µ > 0,
which is the case for the financial applications we are considering.
Proof. Let S0 be the value of St at t = t0. E(St) = L is the equilibrium of (2.5).
Since the eigenvalue −α < 0, stability follows. Therefore as t −→ ∞, the solution
curves of E(St) decay exponentially to L. Denote the two-by-two matrix of (2.6) by
A, and the value of Lt at t = t0 by L0. This system is hyperbolic since the real part
of the eigenvalues −α and µ of A are nonzero. Furthermore, the equilibrium at the
origin is a saddle since −α < 0, and µ > 0. The stable and unstable manifolds of
this hyperbolic equilibrium are the corresponding eigenspaces and, since the system
is linear, they are global [R, 1995] Theorem 6.1, p. 111. They are respectively the
lines, {(1, 0) e−α (t−t0)
, t ∈ R} and {( α
µ+α , 1) L0 eµ (t−t0)
, t ∈ R} in the phase space
(E(St), E(Lt)). The solution to the above linear system is given by (see [R, 1995]
Proposition 3.1, p. 97)
E(St)
E(Lt)
= e(t−t0)A S0
L0
=
S0e−α(t−t0)
+ α
α+µ L0(eµ(t−t0)
− e−α(t−t0)
)
L0eµ(t−t0)
which can be decomposed in term of the direct sum of the stable and unstable
manifolds as
S0 −
α
α + µ
L0 e−α(t−t0) 1
0
+ L0eµ(t−t0)
α
α+µ
1
.
Dynamically speaking, in the phase space, any trajectory t −→ (E(St), E(Lt)), off
the stable or unstable manifolds, tends in forward time as t −→ ∞, to L0eµ(t−t0) α
µ+α , 1
which is the unstable manifold.
4. 4 ALI LARI-LAVASSANI, MOHAMADREZA SIMCHI, AND ANTONY WARE
Note that we used the notation E(St) to denote the conditional expectation
E(St|St0
)|t. We next gather and present results on the first moments of the above
processes:
Proposition 2. a) Consider the process, dSt = (µSt + b)dt + σSt dZt over the
time interval [ti, tk], where µ, b and σ are constants and let Sti = Si. We make the
generic non-degeneracy assumptions:
(H1) σ2
+ µ = 0, 2µ + σ2
= 0 and µ = 0.
Then the first and second moments of St are given by:
E(St|Si)|tk
= −
b
µ
+ (Si +
b
µ
)eµ(tk−ti)
,
E(S2
t |Si)|tk
= c e(2µ+σ2
)(tk−ti)
−
2b(Si + b
µ )
σ2 + µ
eµ(tk−ti)
+
2b2
µ(2µ + σ2)
,
where
c = Si
2
−
2b(Si + b
µ )
σ2 + µ
+
2b2
µ(2µ + σ2)
.
b) Consider the process given by (2.4) over the time interval [ti, tk], and let Sti
= Si,
Lti
= Li. We make the generic non-degeneracy assumptions:
(H2) µ + α − σ2
= 0 = (2µ + ξ2
− σ2
+ 2α) = 0 = (µ + ξ2
+ α) = 0
Then the first and second moments are given by:
E(St|Si)|tk
= Sie−α(tk−ti)
+
α
α + µ
Li(eµ(tk−ti)
− e−α(tk−ti)
),
E(Lt|Li)|tk
= Lieµ(tk−ti)
,
E((St, St)|Si)|tk
= S2
i ϕ1(tk − ti) +
2α
Ψ
SiLi(ϕ1(tk − ti) − ϕ2(tk − ti))
+2α2
L2
i (
−ϕ1(tk − ti)
ΨΓ
+
ϕ2(tk − ti)
ΨΘ
+
ϕ3(tk − ti)
ΘΓ
),
E((St, Lt)|(Si, Li))|tk
= ϕ2(tk − ti)SiLi +
α
Θ
L2
i (ϕ3(tk − ti) − ϕ2(tk − ti))
E((Lt, Lt)|Li)|tk
= L2
i ϕ3(tk − ti),
where ϕ1(t) = e(σ2
−2α)(tk−ti)
, ϕ2(t) = e(µ−α)(tk−ti)
, ϕ3(t) = e(2µ+ξ2
)(tk−ti)
, Ψ =
−µ − α + σ2
, Θ = µ + ξ2
+ α, and Γ = 2µ + ξ2
− σ2
+ 2α.
Remark 2.1. By generic is meant that in the parameter space R{σ, µ} or
R{σ, µ, α}, the set of values satisfying (H1) or (H2) is open and dense. See [HS,
1974] p 154. From a mathematical modeling point of view, it suffices to only study
generic models since any model can be made generic via a small perturbation.
Proof. The differential equations governing time evolution of E(St) and E(S2
t ) are
give in [KP, 1999] p. 113, see also, [G, 1997] p. 113. For the case at hand, they
become,
d
dt
(E(St)) = µE(St) + b
d
dt
(E(S2
t ) = (2µ + σ2
) E(S2
t ) + 2bE(St).
Then a simple integration yields the result. The system of ordinary differential
equations governing the time evolution of the first and second moments of St and
5. A DISCRETE VALUATION OF SWING OPTION 5
Lt are quoted in general in [KP, 1999] p. 152. For the first moments E(St) and
E(Lt), these equations are (2.6), and they were solved in the proof of Proposition
1. The equations for the second moments become:
d
dt
E(St, St)
E(St, Lt)
E(Lt, Lt)
=
(σ2
− 2α) 2α 0
0 (µ − α) α
0 0 (ξ2
+ 2µ)
E(St, St)
E(St, Lt)
E(Lt, Lt)
.
Denoting the above three-by-three matrix by B, the solution to this system is
given by
E(St, St)
E(St, Lt)
E(Lt, Lt)
= e(tk−ti)B
S2
i
SiLi
L2
i
.
The computation of e(tk−ti)B
is routine in an eigenbasis, see [HS, 1974] chapter 7,
or can be carried out in one operation in the symbolic language Maple.
3. An Abstract Approach to Discrete Processes
Traditionally in mathematical finance literature various discrete modeling of sto-
chastic processes under the guise of binomial or trinomial trees (and beyond) have
been introduced. A single abstract unifying framework can be developed for all
these trees. Consider a single or multi-factor stochastic process St modeled dis-
cretely over some time interval. At each time step i, the possible states of the
world are given by finitely many vectors S(i)
= (S
(i)
j ), where j is in a finite index
set J(i)
. Note that we are not restricting ourselves to one-dimensional models by
this notation. The state vector S(i)
can incorporate the values of the underlying
asset as well as secondary factors in the multi-factor case. As time flows from the
time step i to i + 1, the discrete node S
(i)
j can go to any S
(i+1)
j
, j ∈ J(i+1)
. What
determines the discrete process is a transition probability matrix P(i)
= (P
(i)
j,j ),
so that the element P
(i)
j,j represents the probability of S
(i)
j going to S
(i+1)
j
. The
number of rows in P(i)
is the cardinal of J(i)
, and the number of columns is the
cardinal of J(i+1)
; and for all j ∈ J(i)
,
j ∈J(i+1)
P
(i)
j,j = 1.
Example 3.1. A one-factor non-recombining binomial discrete stochastic pro-
cess is S(i)
= (S
(i)
j ) with j ∈ J(i)
= 1, . . . , 2i
. The transition probabilities are
given for every i by a vector (pi
j), j ∈ J(i)
, such that P
(i)
j,j is defined to be pi
j
if j = 2j, 1 − pi
j if j = 2j − 1, or 0 otherwise, as depicted in Figure 1. The
dashed lines in that figure represent prohibited connections which are ruled out by
setting their corresponding probabilities equal to zero. In a multiplicative process
one defines the up and down jump sizes u
(i)
j and d
(i)
j such that, S
(i+1)
2j = S
(i)
j u
(i)
j
and S
(i+1)
2j−1 = S
(i)
j d
(i)
j . We note that this tree is not numerically efficient since the
number of nodes grows exponentially with time.
Example 3.2. A one-factor recombining binomial discrete stochastic process is
S(i)
= (S
(i)
j ) with j ∈ J(i)
= {1, . . . , i + 1}. The transition probabilities, are given
for every i by a vector (pi
j), j ∈ J(i)
, such that P
(i)
j,j is defined to be pi
j if j = j +1,
6. 6 ALI LARI-LAVASSANI, MOHAMADREZA SIMCHI, AND ANTONY WARE
Figure 1. A non-recombining binomial tree, showing the label-
ing of the asset prices at the nodes, and of selected elements of
the transition probability matrix P(1)
. Dashed lines indicate zero
values for the associated probabilities. The non-zero transition
probabilities are indicated by solid lines joining the nodes.
1−pi
j if j = j, or 0 otherwise, see Figure 2. In a multiplicative process one defines
the up and down jump sizes u
(i)
j and d
(i)
j so that, S
(i+1)
j+1 = S
(i)
j u
(i)
j and S
(i+1)
j =
S
(i)
j d
(i)
j . By definition of recombination u
(i)
j d
(i+1)
j+1 = d
(i)
j u
(i+1)
j . A standard choice
is to set the up and down jumps constant across the tree: u
(i)
j = u and d
(i)
j = d,
and to have a single probability p such that p
(i)
jj = pδjj −1 + (1 − p)δjj , where δjj
is equal to 1 if j = j and 0 otherwise. Then S
(i)
j = S0uj
di−j
, with S0 = S
(0)
1 . A
corresponding additive process might be defined via S
(i)
j = S0 + ju + (i − j)d.
7. A DISCRETE VALUATION OF SWING OPTION 7
Figure 2. A recombining binomial tree, illustrating the labeling
of the asset-prices at the nodes, as well as the links associated with
non-zero probabilities.
Example 3.3. A one-factor recombining trinomial discrete stochastic pro-
cess is S(i)
= (S
(i)
j ) with the index set J(i)
= {1, . . . , 2i + 1} and the transi-
tion probabilities, given for every i by two vectors (pi
j), (qi
j), j ∈ J(i)
, such that
P
(i)
j,j isdefinedtobepi
j if j = j +2, qi
j if j = j +1, 1−pi
j −qi
j if j = j, or 0 otherwise.
See Figure 3.
Example 3.4. In some cases, two-factor stochastic processes can be modeled
as a direct product of two trees, see Theorem 2 below. Consider two recombining
binomial trees S(i)
= (S
(i)
j ), j ∈ J(i)
and L(i)
= (L
(i)
k ), k ∈ K(i)
as in Example 3.2,
possessing the same number of nodes along their time axes, that is J(i)
= K(i)
for
all i. Denote the transition probabilities of S(i)
by pi
j with up and down jumps ui
j,
di
j and let those of L(i)
be qi
k, wi
k and hi
k respectively. The direct product S(i)
×L(i)
8. 8 ALI LARI-LAVASSANI, MOHAMADREZA SIMCHI, AND ANTONY WARE
Figure 3. A recombining trinomial tree, illustrating the labeling
of the asset-prices at the nodes, as well as the links associated with
non-zero probabilities.
is naturally defined to be a tree T(i)
with a typical node T
(i)
j,k = (S
(i)
j , L
(i)
k ) at time
i. In the next time step, T
(i)
j,k branches into four nodes T
(i+1)
j+1,k+1, T
(i+1)
j,k+1, T
(i+1)
j+1,k and
T
(i+1)
j,k , defined along with their probabilities by
Nodes Probabilities
T
(i+1)
j+1,k+1 = S
(i)
j ui
j, L
(i)
k wi
k pi
jqi
k
T
(i+1)
j,k+1 = S
(i)
j di
j, L
(i)
k wi
k (1 − pi
j)qi
k
T
(i+1)
j+1,k = S
(i)
j ui
j, L
(i)
k hi
k pi
j(1 − qi
k)
T
(i+1)
j,k = S
(i)
j di
j, L
(i)
k hi
k (1 − pi
j)(1 − qi
k).
9. A DISCRETE VALUATION OF SWING OPTION 9
4. Numerical Implementations
In this section we derive formulae for numerical implementations of various under-
lying processes discussed in Section 2 in terms of the multiplicative recombining
binomial trees of Section 3. We first discuss the general, equation (2.1) and deduce
the cases of log-normal and one factor mean-reverting processes as an application.
We next show that up to the first order approximation, the two-factor model (2.4)
can be discretized as a direct product of the two one-factor trees above.
Consider the discrete binomial process S(i)
wherein the value of the process at
the time step i, given by S
(i)
j , can undergo an up jump to S
(i+1)
j +1 = S
(i)
j ui
j with
the probability pi
j, or a down jump to S
(i+1)
j = S
(i)
j di
j. Note that this situation
is general enough to encompass both Example 1 and 2 of Section 3, and to also
include the possibilities of multiple jumps. To approximate a continuous process,
we match the first and the second moments of the discontinuous process on the tree
with those of the continuous process over time intervals of length ∆t.
Proposition 3. Consider the continuous process dSt = µ(St, t)dt + σ(St, t)dZt as
in (2.1), and the recombining binomial tree (S
(i)
j ) = S(i)
described above. Assume
that at t = ti, Sti = S
(i)
j . Over the time interval [ti, ti + ∆t], we denote the
conditional expectations of the continuous process by E(St|Sti )|ti+∆t, and that of
the discrete process by Ed(S(i)
|S
(i)
j )|ti+∆t.
Matching the first and second moments of these two processes results in the
binomial tree specified by:
pi
j =
A(Sti , ∆t) − di
j
ui
j − di
j
, ui
j = ecosh−1
(θ)
, di
j = 1/ui
j(4.1)
where,
A(Sti , ∆t) := E(St|Sti )|ti+∆t/S
(i)
j ,
B(Sti
, ∆t) := E(S2
t |Sti
)|ti+∆t/(S
(i)
j )2
,
θ(Sti
, ∆t) :=
1 + B(Sti
, ∆t)
2A(Sti , ∆t)
Remark 2.2. This general result applies to all binomial approximation schemes.
Some classical particular cases are discussed in [K,1998] Chapter 5.1.
Proof. Using the definition of Ed(.|.), the matching equations become
Ed(S(i)
|S
(i)
j )|ti+∆t : = pi
jS
(i+1)
j +1 + (1 − pi
j)S
(i+1)
j = E(St|Sti
)|ti+∆t
Ed((S(i)
)2
|S
(i)
j )|ti+∆t : = pi
j(S
(i+1)
j +1 )2
+ (1 − pi
j)(S
(i+1)
j )2
= E(S2
t |Sti
)|ti+∆t
Since S
(i+1)
j +1 = S
(i)
j ui
j and S
(i+1)
j = S
(i)
j di
j, the above equations reduce to
pi
jui
j + (1 − pi
j)di
j = A(Sti
, ∆t)(4.2)
pi
jui2
j + (1 − pi
j)di2
j = B(Sti
, ∆t)
10. 10 ALI LARI-LAVASSANI, MOHAMADREZA SIMCHI, AND ANTONY WARE
where A(Sti
, ∆t) := E(St|Sti
)|ti+∆t/Si
j and B(Sti
, ∆t) := E(S2
t |Sti
)|ti+∆t/Si2
j .
Solving for pi
j results in pi
j =
A(Sti
,∆t)−di
j
ui
j −di
j
. To remove an extra degree of free-
dom in these equations we assume that di
j = 1/ui
j, then substituting pi
j into the
second equation and solving gives ui
j +di
j =
1+B(Sti
,∆t)
A(Sti
,∆t) , which yields the result.
Remark 2.3. One should note1
that 0 ≤ pi
j ≤ 1. Moreover, although the
above values for ui
j and di
j match (4.2) precisely, in general they will not result in
a recombining tree. In practice, for every particular choice of µ(St, t) and σ(St, t)
in Equation (2.1) appropriate approximations must be chosen to ensure numerical
feasibility as in the following theorem.
Theorem 1. a) Consider the linear stochastic differential equation
dSt = (µSt + b)dt + σStdZt
with constant coefficients over the time interval [t0, T]. Let S(i)
= (S
(i)
j ) be a
recombining binomial tree as described in Example 3.2, with time step ∆t. Suppose
there exists a lower bound Smin = 0 such that S
(i)
j > Smin for all i, j. Matching
the first and second moments of S(i)
= (S
(i)
j ) with those of St, up to order ∆t and
for Si
j above Smin, results in
pi
j =
1
2
+
µ + b/S
(i)
j − σ2
/2
2σ
√
∆t , ui
j = eσ
√
∆t
, di
j = 1/ui
j.
b) Without assuming a lower bound Smin, the log-normal process dSt = µSt dt+σS
dZt can be discretized by a recombining binomial tree with
pi
j =
1
2
+
µ − σ2
/2
2σ
√
∆t , ui
j = eσ
√
∆t
, di
j = 1/ui
j.
1To see why this is so, note first that the definitions of A(Sti , ∆t) and B(Sti , ∆t) as normalized
expectations of S(t) and (S(t))2 respectively imply (dropping extraneous notation) that
B ≥ A2
.
One immediate consequence of this is that
θ =
1 + B
2A
≥
1 + A2
2A
= 1 +
(A − 1)2
2A
≥ 1.
A second consequence is that
2Aθ = 1 + B ≥ 1 + A2
,
so that
1 + A2
− 2Aθ ≤ 0.
By adding θ2 − 1 to both sides we obtain
(A − θ)2
≤ θ2
− 1.
It follows now that
θ − θ2 − 1 ≤ A ≤ θ + θ2 − 1.
Noting that cosh−1
θ = ln(θ +
√
θ2 − 1) we see that we have
d ≤ A ≤ u.
The required result now follows directly from the definition of p.
11. A DISCRETE VALUATION OF SWING OPTION 11
c) The mean-reverting process dSt = α(L − St) dt + σSt dZt, can be discretized by
a recombining binomial tree with
pi
j =
1
2
+ (α
L/S
(i)
j − 1
2σ
−
σ
4
)
√
∆t , ui
j = eσ
√
∆t
and di
j = 1/ui
j.
Remark 2.4. The above approximations are valid for a ∆t which is indepen-
dent of the node. Consequently, the up and down jumps ui
j and di
j are constant
throughout the tree and do not depend on the position of the node, forcing this
tree to recombine. Furthermore, for ∆t small enough, 0 ≤ pi
j ≤ 1. Therefore the
above trees are numerically efficient.
Proof. a) Using the expressions for E(St|Sti )|ti+∆t and E(S2
t |Sti )|ti+∆t from Propo-
sition 2 in Proposition 3, leads to:
A(Sti , ∆t) = −
b
µS
(i)
j
+ (1 +
b
µS
(i)
j
) eµ∆t
B(Sti
, ∆t) =
c
(S
(i)
j )2
e(2µ+σ2
)∆t
−
2b(Si
j + b
µ )
(σ2 + µ)(S
(i)
j )2
eµ∆t
+
2b2
µ(2µ + σ2)(S
(i)
j )2
.
In the interval (Smin, ∞), A(Sti
, ∆t) and B(Sti
, ∆t) can be expanded in Taylor
series in ∆t uniformly with respect to Si. Indeed, A(Sti , ∆t) = 1 + (µ + b
S
(i)
j
)∆t +
R(∆t, S
(i)
j ) where R(∆t, S
(i)
j ) = (1 + b
µS
(i)
j
) r(∆t), with lim
∆t→0
r(∆t)
∆t
= 0. Since
b
µS
(i)
j
≤ b
µSmin
for all i, j, lim
∆t→0
R(∆t, S
(i)
j )
∆t
= 0. A similar argument holds for
B(Sti
, ∆t). Therefore,
A(Sti
, ∆t) = 1 + (µ +
b
S
(i)
j
)∆t + O(∆t2
)(4.3)
B(Sti
, ∆t) = 1 + (
2b
S
(i)
j
+ 2µ + σ2
)∆t + O(∆t2
)
and hence using (4.1) one has θ(Sti
, ∆t) = 1 + σ2
2 ∆t+O(∆t2
). On the other hand,
cosh(σ
√
∆t) = 1 + σ2
2 ∆t + O(∆t)2
, therefore
θ(Sti
, ∆t) = cosh(σ
√
∆t) modO(∆t2
).
Substituting this value for θ in (4.1) yields:
ui
j = eσ
√
∆t
mod(∆t)2
and pi
j =
1
2
+
µ + b/S
(i)
j − σ2
/2
2σ
√
∆t + O(∆t3/2
)
The notation ui
j = eσ
√
∆t
mod(∆t)2
signifies that ui
j and eσ
√
∆t
agree up to and
including terms of degree (∆t). Cases b) and c) are particular cases of a). If b = 0,
the uniform expansion of A and B will be valid on (0, ∞). Part b) is a special case
of a) with b = 0, and part c) is a special case of a) with µ = −α and b = αL.
We now turn to the two-factor model and give
12. 12 ALI LARI-LAVASSANI, MOHAMADREZA SIMCHI, AND ANTONY WARE
Theorem 2. Consider the system of linear stochastic differential equations (2.4)
over the time interval [t0, T],
dSt = α(Lt − St) dt + σSt dZt
dLt = µLt dt + ξLt dWt,
a) Over any interval [ti, ti + ∆t] ⊂ [t0, T], the processes St and Lt will remain
uncorrelated up to the order O(∆t, Sti , Lti ).
b) Let S(i)
= (S
(i)
j ) and L(i)
= (L
(i)
k ) be recombining binomial trees as described
in Example 3.2, with time step ∆t. Suppose there exist lower and upper bounds
Smin = 0, respectively Lmax < ∞ such that S
(i)
j > Smin and L
(i)
k < Lmax for all
i, j, k. Then, away from Smin and Lmax, matching the first and second moments
of S(i)
, respectively L(i)
with those of St, respectively Lt, and up to order ∆t results
in
L(i)
: qi
k =
1
2
+
µ − ξ2
2ξ
√
∆t , wi
k = eξ
√
∆t
, hi
k = 1/wi
k.
S(i)
: pi
j =
1
2
+ (α
L
(i)
k /S
(i)
j − 1
2σ
−
σ
4
)
√
∆t , ui
j = eσ
√
∆t
, di
j = 1/ui
j.
The above system can be discretized as a direct product of the two trees S(i)
×L(i)
described in Example 3.4.
Proof. a) Assume that at t = ti the value of St is S
(i)
j and that the value of Lt is
L
(i)
k . A first order approximation of the results obtained in Proposition 2 part b)
yields
E(St|S
(i)
j )|ti+∆t = S
(i)
j + α(L
(i)
k − S
(i)
j )∆t + O(∆t2
, Sti , Lti )
E(Lt|L
(i)
k )|ti+∆t = L
(i)
k + µL
(i)
k ∆t + O(∆t2
, Lti
)
and
E((St, St)|S
(i)
j )|ti = (S
(i)
j )2
+ ((σ2
− 2α)(S
(i)
j )2
+ 2αS
(i)
j L
(i)
k )∆t + O(∆t2
, Sti , Lti )
E((St, Lt)|(S
(i)
j , L
(i)
k ))|ti
= S
(i)
j L
(i)
k + ((µ − α)S
(i)
j L
(i)
k + α(L
(i)
k )2
)∆t + O(∆t2
, Sti
, Lti
)
E((Lt, Lt)|L
(i)
k )|ti
= (L
(i)
k )2
+ (ξ2
+ 2µ)∆t + O(∆t2
, Lti
).
Therefore, up to O(∆t, Sti , Lti ),
Cov(St, Lt)|ti
= E((St, Lt)|(S
(i)
j , L
(i)
k ))|ti
− E(St|S
(i)
j )|ti+∆t E(Lt|L
(i)
k )|ti+∆t = 0
and hence at (S
(i)
j , L
(i)
k ) and up to O(∆t, Sti , Lti ) the processes St and Lt will
remain uncorrelated over [ti, ti + ∆t].
b) By part a) at each (S
(i)
j , L
(i)
k ) the two processes St and Lt remain uncorrelated
over [ti, ti + ∆t]. Therefore at such a point and over [ti, ti + ∆t], St and Lt can
be approximated by independent binomial trees, and hence (2.4) can in turn be
approximated by the direct product of these two trees over the same time interval.
Since L is a log-normal process, part b) of Theorem 1 yields the tree L(i)
. Note that,
at (S
(i)
j , L
(i)
k ), St follows a one-factor mean-reverting process (2.3) with L = L
(i)
k .
13. A DISCRETE VALUATION OF SWING OPTION 13
Therefore from Theorem 1 we have
A(Sti
, ∆t) =
L
(i)
k
S
(i)
j
+ (1 −
L
(i)
k
S
(i)
j
) e−α∆t
B(Sti
, ∆t) =
c
(S
(i)
j )2
e(σ2
−2α)∆t
−
2αL
(i)
k (Si
j − L
(i)
k )
(σ2 − α)(S
(i)
j )2
e−α∆t
−
2αL
(i)
k
2
(σ2 − 2α)(S
(i)
j )
2 .
Note that since Smin and Lmax exist
L
(i)
k
S
(i)
j
is bounded. Therefore A and B can be
uniformly expanded in Taylor series, as in Theorem 1. Hence the tree of S follows
from part c) of Theorem 1.
5. A General Swing Contract
We begin by introducing the parameters necessary for modeling a fairly general
swing option contract which accomodates a wide range of applications.
The economy considered has two assets, a risk free interest rate, and a risky asset
S exchanged in units, and two players, a buyer and a seller. Consider three time
values T0 ≤ T1 < T2 where T0 is the time when the swing option is priced, and the
interval [T1, T2] is the swing contract period during which the buyer is assumed to
be purchasing from the seller a determined amount of S, called the base load. Base
loads can be easily priced, we will therefore only focus on modeling and pricing the
swing. Beyond the base load, the swing option contract provides the possibility
of exchanging the asset S in fixed quantities and at determined strike prices. By
definition, an up swing consists in the buyer acquiring Vu units of S immediately
upon request at a strike price of Ku per unit, and a down swing consists in the buyer
delivering Vd units of S to the seller at a strike price of Kd per unit, immediately
after notification of the latter. The cost of a swing is to be settled immediately.
The swing option entitles the buyer to exercise, during the time interval [T1, T2], up
to Nu up swings and Nd down swings. An exercise can only occur at a discrete set
of times {τ1,..., τe} ⊂ [T1, T2], and consists of at most one up or down swing at each
time. A penalty is to be imposed if the net amount of S acquired by the buyer via
swing exercises is not bounded by Vmin and Vmaxat expiry. More precisely let nu
and nd denote respectively the actual number of up and down swing exercises that
occurred during [T1, T2], then to avoid a penalty one must have
Vmin ≤ nuVu − ndVd ≤ Vmax.(5.1)
If not, there are cash penalties of A1 (respectively A2) per unit of the net amount
of units acquired, nuVu −ndVd, short of Vmin, (respectively in excess of Vmax). Then
the resulting penalties are A1(Vmin −nuVu +ndVd) (respectively, A2(nuVu −ndVd −
Vmax)).
6. Discrete Forest Methodology
We now turn to modeling the swing option contract described in the above
section via a forest. It is assumed that the underlying asset S, follows a one- or
multi-factor risk-neutralized stochastic process, expressed in terms of a single or
a system of stochastic differential equations. We use the notation of the previous
section and recall that a swing exercise can only occur at discrete times {τ1, ..., τe} ⊂
14. 14 ALI LARI-LAVASSANI, MOHAMADREZA SIMCHI, AND ANTONY WARE
[T1, T2]. We may without loss of generality assume τ1 = T1 and τe = T2. To simplify
the presentation, we assume that the rights to exercise occur at equally spaced time
intervals, or ticks, of length ∆te. We also assume that there are integers N1 and
N2 such that T1 = N1∆te and T2 = N2∆te. To obtain an accurate valuation of the
stochastic process for the underlying asset S we model it on a discrete tree with
a finer time-scale obtained, by dividing each tick into n computational timesteps
∆t = ∆te/n. Therefore S follows a process discretized by a tree S(i)
, P(i)
, with nN2
time grid nodes, that is i ∈ I = {0, . . . , nN2}, whereas option exercise is permitted
at nodes whose indices belong to the set Ie = {N1n, (N1 +1)n, . . . , N2n}. The local
value of the swing option, denoted by V i
nu,nd
(j), depends on the generalized index
set (i, j, nu, nd), where nu and nd denote the number of remaining up and down
swings respectively, i ∈ I and j ∈ J(i)
. The forest terminology finds its origin in this
setting: for each (nu, nd), the tree inhabited by the values Vnu,nd
= {V i
nu,nd
(j), j ∈
J(i)
, i ∈ I} may be viewed as a tree isomorphic to the tree {S(i)
, i ∈ I}; thus we
have not one tree but (Nu + 1) × (Nd + 1) trees.
Define the discounted expected value
Wi
nu,nd
(j) = Ri
j ∈J(i+1)
P
(i)
j,j V i+1
nu,nd
(j ),(6.1)
where Ri = exp(−r∆t) is the present value of 1 unit at a time ∆t in the future.
When not at an exercise node (i ∈ Ie), each tree Vnu,nd
in the forest is back-folded
independently of the others, and V i
nu,nd
(j) = Wi
nu,nd
(j).
At a point on the tree when exercise is permitted (i ∈ Ie), one can still use (6.1)
to calculate a discounted expected value on each tree, but the exercise of a swing
(up or down) will cause a shift from the tree Vnu,nd
to one of the trees Vnu−1,nd
or
Vnu,nd−1. Exercising the swing will result in an immediate cash flow corresponding
to the relevant payoff formula. However, the option now has one fewer exercise
rights, and a jump to a neighboring tree has taken place. Then the option value
V i
nu,nd
(j) is given as the maximum of three possibilities:
Wi
nu,nd
(j) (no swing)
Wi
nu−1,nd
(j) + (Si
j − Ku) (swing up)
Wi
nu,nd−1(j) + (Kd − Si
j) (swing down).
This forest methodology is depicted in Figure 4 for a swing option with just
one up- and one down-swing exercise right using 4 computational timesteps. Each
tree carries the values of an option with a particular combination of up- and down-
swing rights used up, over the course of 4 time steps. The calculation proceeds
by stepping backwards in time down through the trees. At the top time level, the
option values are determined solely by the penalties imposed. Then, at each state
of the world, a decision is made as to whether it is optimal to exercise an up- or
down-swing right. The choices made at the final time are shown in (a), where the
links indicate the transfer of information from tree-to-tree as a result of an exercise
of a swing right. The resulting values are then transfered to the next time level by
a discounted expectation, and a set of decisions made as to whether or not to swing
once more. Those decisions are shown in (b), and also in (c) at the next time level.
The final two time levels are not shown because in this calculation, at no point was
it found to be optimal to exercise the swing rights. The total history of all swing
choices in all states of the world is shown in (d). The final option value is to be
15. A DISCRETE VALUATION OF SWING OPTION 15
(a) (b)
(c) (d)
Figure 4. A 2 × 2 forest for pricing a swing option with one up-
and one down-swing exercise right using 4 computational
timesteps.
found at the bottom node of the front tree. One feature to note is that the trees
have all been truncated. This is because of the imposition of an Smin as described
in Theorem 1.
7. Sensitivity Analysis
Pricing a swing contract involves a considerable number of input parameters and
as such a detailed sensitivity analysis of such a contract would be the subject of
a publication on its own. To exhibit a fair amount of information while keeping
the presentation of different scenarios tractable, we consider two swing contracts
with their parameters defined below. These parameters are fixed, unless otherwise
stated. We follow the notation of Section 6.
For the one-factor mean-reverting model dSt = α(L − St)dt + σStdWt we label
the following setting by (A) and implement it according to the binomial tree given
in Theorem 1, (c).
16. 16 ALI LARI-LAVASSANI, MOHAMADREZA SIMCHI, AND ANTONY WARE
n α = 0.2, L = 30 α = 1.2, L=30
Smin = 5 Smin = 15 Smin = 25 Smin = 5 Smin = 15 Smin = 25
80 196.2270 196.2270 197.2613 183.7410 183.7411 184.7600
160 196.2245 196.2246 197.2006 183.7372 183.7373 184.6986
320 196.2208 196.2208 197.1931 183.7338 183.7339 184.6915
Table 1. Convergence behaviour and its dependence on α and Smin.
(A): S0 = 45, Smin = 5, r = 0.5, α = 10, L = 40, σ = 1.5,
T2 − T0 = 30 days,
∆te = τi+1 − τi = 1 in units of day, N1 = 1, N2 = 30, n = 80,
Nu = 1, Nd = 0, Vu = 20, Vd = 20, Ku = 40, Kd = 40,
Au = 3, Ad = 3, Vmax = 150, Vmin = −150.
For the two-factor mean-reverting model dSt = α(Lt −St) dt+σSt dZt, dLt = µ
Lt dt + Lt ξ dWt, we label the following setting by (B) and implement it according
to Theorem 2.
(B):S0 = 1.5, Smin = 0.1, r = 0.05, α = 0.2, L0 = 1, σ = 1.5, µ = 1,
ξ = 1,
T2 − T0 = 30 days,
∆te = τi+1 − τi = 1 in units of day, N1 = 1, N2 = 30, n = 80,
Nu = 3, Nd = 3, Vu = 1, Vd = 1, Ku = 1, Kd = 1,
Au = 0, Ad = 0, Vmax = 15, Vmin = −15.
7.1. Numerical Convergence. We compute a sequence of the values of the swing
option by doubling the number of computational time steps n. The convergence
pattern of the value of the swing as n increases is illustrated by verifying that the
error goes to zero. More specifically, the value of the swing option is calculated with
n = 109
for the one-factor scenario A, and with n = 27
for the two-factor scenario
B. These values are then subtracted from the approximate values of the swing
option obtained with n = 10 2i
, i = 0, ...5 for A, respectively n = 2i
, i = 1...6
for B. These yield estimates for the error, which are plotted versus n, in Figure 5,
respectively Figure 6 in a base 2 logarithmic scale. Note that in Figure 5 all of the
values agree to 3 significant figures. A fourth digit of accuracy is added with more
than 640 computational timesteps (see Section 6).
We finally note that the convergence pattern is a function of the various param-
eters of the swing contract as discussed in the next section.
7.2. Parameter Sensitivity. Table 1 is generated for scenario (A) wherein the
values of α, Smin and n are varied as prescribed, and L = 30.
First note that the above setting (A) with long-run mean L = 30 only involves
one up-swing with an strike price, Ku = 40, which is above L and below the initial
asset price. As a result, the swing option behaves as an in the money Bermudan
call option. Next observe that the closer Smin is to L, the higher is the value of the
option. This is to be expected for, setting Smin closer to L increases the probability
of the asset prices being closer to the strike price, resulting in a higher value for
17. A DISCRETE VALUATION OF SWING OPTION 17
Figure 5. Option values for the one-factor scenario (A), calcu-
lated with increasing numbers of computational timesteps.
the option since it is a call option. This behaviour is presented in Figure 7. One
may also note that the value of the swing option is not sensitive to changes in Smin
for Smin less than 20, which is about half the long-run mean L. As the process is
mean-reverting, asset values far from its long-run mean are less probable to occur
and as such excluding them does not effect the price of the option. Finally, the
stronger the strength of mean reversion is, the less valuable is the option. Indeed, as
the strength of mean reversion increases, the process is pulled to its mean, which is
below the strike price in this scenario, making the call option less valuable. Figure
8 captures this behaviour.
7.3. Hedging Parameters. Generally speaking, there are at least as many hedge
parameters for a contract as input parameters. As such the hedge parameters of
a swing contract constitute a long list. For the sake of exposition we provide two
conventional hedge parameters, ∆ = ∂V
∂S |S0
and Λ = ∂V
∂σ |S0
, and a less conventional
18. 18 ALI LARI-LAVASSANI, MOHAMADREZA SIMCHI, AND ANTONY WARE
Figure 6. Option values for the two-factor scenario (B), calcu-
lated with increasing numbers of computational timesteps.
∂V
∂L |S0
. Consider the following setting for a swing contract C, written on a one-
factor mean-reverting process:
S0 = 1.5, Smin = .01, r = 0.05, α = 2, L = .8, σ = 1.5,
T2 − T0 = 30 days,
∆te = τi+1 − τi = 1 in units of day, N1 = 2, N2 = 30, n = 30,
Nu = 4, Nd = 3, Vu = 1, Vd = 1, Ku = .4, Kd = .4,
Au = .1, Ad = .1, Vmax = 3, Vmin = −2.
We compute the above Greeks for values of S0 varying in a range which covers
the cases in the money, at the money and out of the money. Table 2 introduces a
few values from the set of data which is used to generate Figures 9, 10 and 11.
19. A DISCRETE VALUATION OF SWING OPTION 19
Figure 7. The effect of varying Smin on the option price for the
one-factor scenario A.
Hedge parameter In the money At the money Out of the Money
S0 = .2 S0 = .4 S0 = .6
∆ -1.95 1.35 2.94
Λ 0.08 0.26 0.19
∂V
∂L |S0
0.11 0.29 0.41
Table 2. Sample values of some hedge parameters for various
values of S0.
Note that contract C is a combination of Bermudan put and call options. For
asset values S below the strike price Ku = Kd = 0.4, the swing contract is out of
the money and behaves as a put option. On the contrary, asset values above the
strike price result in the contract being similar to a call option. This observation
20. 20 ALI LARI-LAVASSANI, MOHAMADREZA SIMCHI, AND ANTONY WARE
Figure 8. The effect of varying strengths of mean-reversion on
the option value. As one might expect, higher values for α mean
lower option values.
explains the concavity and curvature in the ∆ and Λ curves above. As for the ∂V
∂L
curve the change occurs at S = L0, which is predictable.
7.4. Swings versus American Options. We compare a swing contract with a
basket of American options. Having a swing contract in hand, the buyer of the
contract will have the opportunity of exercising his swing rights at particular dates
at his leisure. Another way of replicating such a right is to purchase the same
number of rights via American options having the same expiry date as that of the
swing contract. Of course a buyer with a basket of American options in hand is
able to exercise his rights anytime. He will also be able to exercise all or some of
those rights at a single moment of time. However such extra freedom comes with
an associated cost. Figure 12 captures this cost. Each point on the curve represents
the value of a swing option with Nu upswing rights in a month, Nu = 1, 2, ..., 30,
21. A DISCRETE VALUATION OF SWING OPTION 21
Figure 9. ∆ as a function of the initial asset price S0.
whereas each point on the line indicates the value of Nu individual American call
options expiring within the same period. As seen from the figure, due to the
flattening out of the graph, the buyer of the swing option does not have to pay
much for the extra rights purchased. He is losing some flexibility but at the same
time he is saving money. The buyer of the American basket has more flexibility at
a much higher cost. Therefore the choice between a swing contract and a basket of
American options depends on the flexibility required and the price attached to it.
8. Proof of Convergence
In this section we demonstrate the convergence of the value of the swing option
calculated on a forest of recombining binomial trees to the solution of a partial
differential equation model of the swing option. The error is shown to decrease
linearly with the number of computational time steps.
22. 22 ALI LARI-LAVASSANI, MOHAMADREZA SIMCHI, AND ANTONY WARE
Figure 10. Λ as a function of the initial asset price S0.
We limit the exposition to the proof of convergence for the one-factor mean-
reverting binomial model described in Theorem 1, and note that the result could
be readily adapted to cover other one-factor trees. The proof for the two-factor
case would be similar in structure but rather more involved.
8.1. The Swing Option as a Coupled System of European Options. Con-
sider the option described in Section 5, and write ν = (nu, nd). Let Vν(S, t) be
the value at time t ∈ [0, T2] of an option with nu up-swing rights and nd down-
swing rights remaining, given an asset price of S at time t. Noting that the holder
of such an option may choose to exercise at the exercise times t ∈ {τ1, . . . , τe},
we see that Vν(S, t) may have a jump discontinuity at those times. We write
Vν(S, τ+
i ) = limt→τ+
i
Vν(S, t).
As described in Section 6, the exercise of a swing right brings with it an associ-
ated cash flow, but also means that the option value is exchanged for the value of
23. A DISCRETE VALUATION OF SWING OPTION 23
Figure 11. ∂V
∂L as a function of the initial asset price S0.
an option with one fewer right. At each exercise time, for each value of the under-
lying asset S, it is assumed that the optimal choice is made among the available
possibilities. Thus, for ν ∈ (1, . . . , Nu) × (1, . . . , Nd), and for i = 1, . . . , e − 1, the
(payoff) functions can be defined by
Pν(S, τi) = max
Vν(S, τ+
i )
Vνu (S, τ+
i ) + Payu(S)
Vνd (S, τ+
i ) + Payd(S),
(8.1)
where νu
= (nu − 1, nd), νd
= (nu, nd − 1), and
Payu(S) = Vu(S − Ku), Payd(S) = Vd(Kd − S),
with obvious changes to the definition for the cases nu = 0 or nd = 0. The payoff
function at the final exercise time is determined by the penalties being applied.
24. 24 ALI LARI-LAVASSANI, MOHAMADREZA SIMCHI, AND ANTONY WARE
Figure 12. The difference between a swing option with only up-
swing rights and an American option with an equivalent total vol-
ume. The option values are plotted as a function of the number of
up-swing rights (each with unit volume) for the swing option, and
as a function of the total volume for the American option.
The function Penν(S) is defined by
Penν(S) = A1(nuVu − ndVd − Vmin)+ − A2(nuVu − ndVd − Vmax)+.(8.2)
The final exercise payoff function is then
Pν(S, τe) = max
−Penν(S)
−Penνu (S) + Payu(S)
−Penνd (S) + Payd(S),
(8.3)
again for nu = 1, . . . , Nu and nd = 1 . . . , Nd, and with obvious changes for the cases
nu = 0 or nd = 0.
25. A DISCRETE VALUATION OF SWING OPTION 25
For t in the interval (τi, τi+1), the option value Vk,l(S, t) is the value of a Euro-
pean contract at time t with expiry at time τi+1 and payoff function Pν(S, τi+1).
The value of this swing option in any given time interval thus depends (via the
payoff function) on the values of related swing options in subsequent time intervals.
In particular, the value of the original swing contract is Vν0 (S0, 0). This is the
value at time 0 and asset price S0 of a European option with expiry time τ1 and
payoff Pν0 (S, τ1), and thus depends on the values of all of the swing options in the
subsequent time intervals.
Obtaining a value for Vν0 (S0, 0) thus involves, for i running backwards from e−1
to 1, first calculating each Pν(S, τi+1) via (8.1) or (8.3), and then evaluating the
European options described above in the interval (τi, τi+1). These values are then
used to calculate the next set of payoff functions. Finally, the European option
with expiry time τ1 and payoff Pν0 (S, τ1) is used to provide the required value at
S = S0 and t = 0.
8.2. A Partial Differential Equation Model. The question remains of how
the European options described above are to be valued. Here we describe a partial
differential equation model based on a generalization of the Black-Scholes equation,
Vt +
σ2
2
S2
VSS + rSVS = rV.
This models, for example, the price V of a stock with volatility σ and risk-free
interest rate r contingent on a stock price S which follows the stochastic process
(2.2). The complete-market assumptions which underlies this model may not apply
in the context we are interested in. Relaxing these assumptions somewhat, we have
the equation
Vt +
σ2
2
S2
VSS + (µ − λσS)VS = rV,
where S is now assumed to follow the stochastic process (2.3), and λ is the market
price of risk function (See [H, 1999], [W, 1998]). We assume for ease of exposition
that λ = 0. The results obtained in this section would hold equally well for other
forms of λ satisfying quite mild continuity assumptions.
We make two concessions to our numerical schemes in the model we describe.
The first is that we impose a lower bound Smin > 0 on the asset price S. The
second is due to the fact that the payoff functions described above are continuous,
but are only piecewise differentiable. This lack of smoothness introduces theoretical
problems for the convergence of numerical schemes such as those described here.
We define a ‘mollified’ maximum function, max , by means of a formula such as
max (a, b) =
a + b + + (a − b)2)
2
,
with
max (a, b, c) = max (a, max (b, c)).
Note that when = 0 this definition agrees with the standard definition of the
maximum. However, defining the payoff functions using max instead of max means
that the payoff functions Pν(S, τi) in (8.1) are just as smooth as the functions
Vν(S, τ+
i ) used to define them. As long as is chosen sensibly, it is easy to verify
that the effect on the option price is negligible.
26. 26 ALI LARI-LAVASSANI, MOHAMADREZA SIMCHI, AND ANTONY WARE
We are now ready to describe our partial differential equation model for the value
of a swing option.
Let Smin > 0, σ > 0 and r ≥ 0 be constants. We seek a set of functions Vν(S, t)
satisfying
∂
∂t
Vν(S, t) +
σ2
S2
2
∂2
∂S2
Vν(S, t) + (AS + B)
∂
∂S
Vν(S, t) = rSVν(S, t),
for S > Smin, t ∈ [0, T]{τi}e
i=1, with
Vν(S, T) = max (−Penν(S), −Penνu (S) + Payu(S),(8.4)
− Penνd (S) + Payd(S)),
Vν(S, τi) = max (Vν(S, τ+
i ), Vνu (S, τ+
i ) + Payu(S),(8.5)
Vνd (S, τ+
i ) + Payd(S)), i = 1, . . . , e − 1
∂
∂S
Vν(Smin, t) = 0, t ∈ [0, T].(8.6)
The comparison with the binomial scheme will in fact be more straightforward if
we describe the equivalent log-transformed problem for wν(x, t) = e−r(T −t)
Vν(S, t):
∂
∂t
wν(x, t) +
σ2
2
∂2
∂x2
wν(x, t) + (a + be−x
)
∂
∂x
wν(x, t) = 0,(8.7)
for x > ln Smin, t ∈ [0, T]{τi}e
i=1, with
wν(x, T) = max (−Penν(ex
), −Penνu (ex
) + Payu(ex
),(8.8)
− Penνd (ex
) + Payd(ex
)),
wν(x, τi) = max (wν(x, τ+
i ), wνu (x, τ+
i ) + Payu(ex
),(8.9)
wνd (x, τ+
i ) + Payd(ex
)), i = 1, . . . , e − 1
∂
∂x
wν(ln Smin, t) = 0, t ∈ [0, T],(8.10)
where a = A − σ2
/2 and b = B.
8.3. The One-factor Binomial Scheme. The binomial approximation to the
above partial differential equation model consists in finding vectors Wi
ν(j) satisfy-
ing, for j ≥ jmin,
Wi,∗
ν (j) = pjWi+1
ν (j + 1) + (1 − pj)Wi+1
ν (j − 1), i < T/∆t = N2n,
WN2n,∗
ν (j) = −Penν(exj
),
Wi
ν(j) = Wi,∗
ν (j), i + 1 ∈ Ie
Wi
ν(j) = max Wi,∗
ν (j), Wi,∗
νu (j) + Payu(exj
), Wi,∗
νd (j) + Payd(exj
) , i + 1 ∈ Ie.
Here xj = ln(S0ej∆x
), with ∆x = σ
√
∆t, and jmin is the value of j for which
Sj = exj
= Smin. The above system is completed by setting
Wi+1
ν (jmin − 1) = Wi+1
ν (jmin + 1),
thus approximating the zero-derivative boundary condition at Smin. The required
option value is obtained once W0
ν0 (0) is obtained. Note that not all values of Wi
ν(j)
are required in order to achieve this. For instance, Wi
ν(j) with i+j odd never needs
to be calculated. Moreover, only values with |j| ≤ i can ever affect W0
ν0 (0).
27. A DISCRETE VALUATION OF SWING OPTION 27
8.4. Convergence of the Value of the Swing Option. Here we write wi
ν(j) =
wν(Sj, t+
i ), with ti = i∆t, and denote the difference between the result of the
binomial calculation and the true solution of the partial differential equation model
by ei
ν(j) = Wi
ν(j) − wi
ν(j). Then we have the following global convergence result
for the binomial forest method.
Theorem 3 (Convergence). Suppose that for each x ∈ [xmin, ∞), wν(x, t) is con-
tinuous on (τi, τi+1] for i = 1, . . . , e and on [0, τ1], and that ∂2
∂t2 wν(x, t) exists and
is bounded on each of these intervals. Suppose also that for each t ∈ [0, T2], wν(x, t)
is three times continuously differentiable, with bounded derivatives up to order four
on (xmin, ∞). Then there exists a constant C such that for each ν,
max
i
max
j
ei
ν(j) ≤ C∆t.
The remainder of this section will be taken up with the proof of Theorem 3.
Substituting the vectors ei
ν(j) into the binomial scheme yields, for j > jmin,
ei,∗
ν (j) = pjei+1
ν (j + 1) + (1 − pj)ei+1
ν (j − 1) + Ti
ν(j), i < N2n(8.11)
ei,∗
ν (j) = 0, i = N2n
ei
ν(j) = ei,∗
ν (j), i + 1 ∈ Ie,(8.12)
ei
ν(j) = max Wi,∗
ν (j), Wi,∗
νu (j) + Payu(exj
), Wi,∗
νd (j) + Payd(exj
) −(8.13)
max wi,∗
ν (j) + Ti
ν(j), wi,∗
νu (j) + Ti
νu (j) + Payu(exj
),
wi,∗
νd (j) + Ti
νd (j) + Payd(exj
) i + 1 ∈ Ie,
where the truncation error Ti
ν(j), which will be the amount by which wi
ν(j) fails
to satisfy the equations defining the binomial approximation, is given by
Ti
ν(j) = wi,∗
ν (j) − wi
ν(j),(8.14)
and
wi,∗
ν (j) = pjwi+1
ν (j + 1) + (1 − pj)wi+1
ν (j − 1) = Wi,∗
ν (j) − ei,∗
ν (j)
for j > jmin. To completely specify the error, we need the boundary equation
ei
ν(jmin) = ei+1
ν (jmin + 1) + Ti
ν(jmin),(8.15)
where
Ti
ν(jmin) = wi+1
ν (jmin + 1) − wi
ν(jmin).(8.16)
We shall restrict our attention to relevant values of j: for each i we only consider
j ∈ Ji with
Ji = {j : j + i even, j ≥ jmin, |j| ≤ i}.
It will be shown inductively that the maximum error ei
= maxν,j∈Ji
ei
ν(j) is
O(∆t). In order to do this, is it necessary to get a handle on the truncation error
terms.
28. 28 ALI LARI-LAVASSANI, MOHAMADREZA SIMCHI, AND ANTONY WARE
¿From (8.16), using the zero-derivative boundary condition together with Taylor-
series expansions of wν(x, t), we have
Ti
ν(jmin) = wi+1
ν (jmin + 1) − wi
ν(jmin)
= (wi+1
ν (jmin + 1) − wi+1
ν (jmin)) + (wi+1
ν (jmin) − wi
ν(jmin))
=
∆x2
2
∂2
∂x2
wν(ξ1, ti+1) + ∆t
∂
∂t
wν(xmin, ξ2),
for some ξ1 ∈ (xjmin
, xjmin+2) and ξ2 ∈ (ti, ti+1). Given the bounded derivatives of
wν, and the fact that ∆x2
= ∆t/σ2
, one can conclude that there is a constant C1
such that
Ti
ν(jmin) ≤ C1∆t(8.17)
for i < N2n.
We next consider the generic case j > jmin, i + 1 ∈ Ie, for which the truncation
error involves the probabilities
pj =
1
2
+
(a + be−xj
)∆t
2∆x
=
1
2
+
µj∆t
∆x
.
We have, making use of the fact that wν(x, t, ) satisfies (8.7),
Ti
ν(j) = pjwi+1
ν (j + 1) + (1 − pj)wi+1
ν (j − 1) − wi
ν(j)
= wi+1
ν (j) − wi
ν(j) +
1
2
wi+1
ν (j + 1) − 2wi+1
ν (j) + wi+1
ν (j − 1)
+
µj∆t
2∆x
wi+1
ν (j + 1) − wi+1
ν (j − 1)
= ∆t
∂
∂t
wν(xj, ti) +
∆t2
2
∂2
∂t2
wν(xj, ξ3) +
∆x2
2
∂2
∂x2
wν(xj, ti)
+
∆x4
12
∂4
∂x4
wν(ξ4, ti) + µj∆t
∂
∂x
wν(xj, ti) + µj∆t∆x2 ∂2
∂x2
wν(ξ5, ti)
=
∆t2
2
∂2
∂t2
wν(xj, ξ3) +
∆x4
12
∂4
∂x4
wν(ξ4, ti) + µj∆t∆x2 ∂2
∂x2
wν(ξ5, ti)
for some ξ3 ∈ (ti, ti+1) and ξ4, ξ5 ∈ (xj, xj+1). It follows from the boundedness of
the derivatives of wi
ν that
max
ν
max
j>jmin
Ti
ν(j) ≤ C2∆t2
, i < N2n.(8.18)
Let
ei
= max
ν
max
j≥jmin
ei
ν(j) .
Then (8.18), (8.17) and (8.11) imply that for i + 1 ∈ Ie,
ei
≤ max ei+1
+ C2∆t2
, C1∆t) .(8.19)
In order to obtain a similar result for i + 1 ∈ Ie we bound the error ei
ν(j). By
the Lipschitz continuity of the maximum function, there is a constant C3 such that
|max (a, b, c) − max (a , b , c )| ≤ C3 max (|a − a | , |b − b | , |c − c |)
29. A DISCRETE VALUATION OF SWING OPTION 29
for all a, b, c, a , b , c ∈ R. Then it follows from (8.13) that, when i + 1 ∈ Ie and
j > jmin,
ei
ν(j) ≤ C3 max ei+1
ν (j) + Ti
ν(j) , ei+1
νu (j) + Ti
νu (j) , ei+1
νd (j) + Ti
νd (j) ,
with obvious changes for the cases nu = 0 and nd = 0, so that
ei
≤ C3 max ei+1
+ C2∆t2
, C1∆t) .(8.20)
Comparing (8.19) and (8.20), we find that they differ only by the extra factor of
C3 when i+1 ∈ Ie. For each i, define k(i) to be the size of the set {i ≥ k : i+1 ∈ Ie}.
Then we make the inductive claim
ei
≤ C
k(i)
3 C1∆t + C2∆t2
(nN2 − i) .
Since eN2n
= 0, the inductive hypothesis holds for i = N2n. Suppose that it holds
for i = i + 1. Then if i + 1 ∈ Ie,
ei
≤ C3 max ei +1
+ C2∆t2
, C1∆t)
≤ C3C
k(i +1)
3 C1∆t + C2∆t2
(nN2 − i − 1) + C2∆t2
= C
k(i )
3 C1∆t + C2∆t2
(nN2 − i )
so that the inductive step is complete in this case. The proof in the case i + 1 ∈ Ie
is exactly similar, and so the theorem is proved, with
C = C
k(0)
3 (C1 + C2T2).
REFERENCES
[BG, 1996] Berbieri, Angelo and Garman, Mark B. October 1996. Putting a Price on
Swings. Energy & Power Risk Management.
[G, 1997] Gardiner, Crispin W. 1997. Handbook of Stochastic Methods. Springer.
[JRT, 1998] Jaillet Patrick, Ronn Ehud I. and Tompaidis Stathis. 1998. Modeling Energy
Prices and Pricing and Hedging Energy Derivatives. Preprint.
[H, 1999] Hull, John. 1999. Options, Futures and Other Derivative Securities. Prentice
Hall.
[HR, 1998] Hilliard Jimmy E and Reis Jorge. 1998. Valuation of Commodity Futures
and Options under Stochastic Convenience Yields, Interest Rates, and Jump
Diffusions in the Spot. J. of Financial and Quantitative Analysis, Vol. 33,
No. 1, pp. 61-86.
[HS, 1974] Hirsch Morris and Smale Stephen. 1974. Differential Equations, Dynamical
Systems and Linear Algebra. Academic Press.
[KP, 1999] Kloeden Peter E. and Platen Eckhard. 1995. Numerical Solution of Stochas-
tic Differential Equations. Springer-Verlag.
30. 30 ALI LARI-LAVASSANI, MOHAMADREZA SIMCHI, AND ANTONY WARE
[K, 1998] Kwok Yue-Kuen. 1998. Mathematical Models of Financial Derivatives. Springer-
Verlag.
[P, 1997] Pilipovic Dragana. 1997. Valuing and Managing Energy Derivatives. McGraw-
Hill.
[R, 1995] Robinson Clark. 1995. Dynamical Systems. CRC Press.
[S, 1998] Schwartz Eduardo S. 1998.Valuing Long-Term Commodity Assets. Financial
Management, Vol. 27, No. 1, pp. 57-66.
[W, 1998] Wilmott, Paul. 1998. Derivatives: The Theory and Practice of Financial
Engineering. John Wiley & Sons.
The Mathematical and Computational Finance Laboratory. Department of Mathe-
matics and Statistics. University of Calgary. Calgary, Alberta N1G 4Z4 Canada
E-mail address: lavassan@math.ucalgary.ca
simchi@math.ucalgary.ca
ware@math.ucalgary.ca