Dynamic Programming Approaches for Estimating and Applying Large-scale Discrete Choice Models

People go through their life making all kinds of decisions, and some of these decisions affect their demand for transportation, for example, their choices of where to live and where to work, how and when to travel and which route to take. Transport related choices are typically time dependent and characterized by large number of alternatives that can be spatially correlated. This thesis deals with models that can be used to analyze and predict discrete choices in large-scale networks. The proposed models and methods are highly relevant for, but not limited to, transport applications. We model decisions as sequences of choices within the dynamic discrete choice framework, also known as parametric Markov decision processes. Such models are known to be difficult to estimate and to apply to make predictions because dynamic programming problems need to be solved in order to compute choice probabilities. In this thesis we show that it is possible to explore the network structure and the flexibility of dynamic programming so that the dynamic discrete choice modeling approach is not only useful to model time dependent choices, but also makes it easier to model large-scale static choices. The thesis consists of seven articles containing a number of models and methods for estimating, applying and testing large-scale discrete choice models. In the following we group the contributions under three themes: route choice modeling, large-scale multivariate extreme value (MEV) model estimation and nonlinear optimization algorithms. Five articles are related to route choice modeling. We propose different dynamic discrete choice models that allow paths to be correlated based on the MEV and mixed logit models. The resulting route choice models become expensive to estimate and we deal with this challenge by proposing innovative methods that allow to reduce the estimation cost. For example, we propose a decomposition method that not only opens up for possibility of mixing, but also speeds up the estimation for simple logit models, which has implications also for traffic simulation. Moreover, we compare the utility maximization and regret minimization decision rules, and we propose a misspecification test for logit-based route choice models. The second theme is related to the estimation of static discrete choice models with large choice sets. We establish that a class of MEV models can be reformulated as dynamic discrete choice models on the networks of correlation structures. These dynamic models can then be estimated quickly using dynamic programming techniques and an efficient nonlinear optimization algorithm. Finally, the third theme focuses on structured quasi-Newton techniques for estimating discrete choice models by maximum likelihood. We examine and adapt switching methods that can be easily integrated into usual optimization algorithms (line search and trust region) to accelerate the estimation process. The proposed dynamic discrete choice models and estimation methods can be used in various discrete choice applications. In the area of big data analytics, models that can deal with large choice sets and sequential choices are important. Our research can therefore be of interest in various demand analysis applications (predictive analytics) or can be integrated with optimization models (prescriptive analytics). Furthermore, our studies indicate the potential of dynamic programming techniques in this context, even for static models, which opens up a variety of future research directions.

Essays in dynamic optimal taxation: working experience, unemployment scarring and poverty

This thesis investigates the design of optimal tax systems in dynamic environments. The first essay characterizes the optimal tax system where wages depend on stochastic shocks and work experience. In addition to redistributive and efficiency motives, the taxation of inexperienced workers depends on a second-best requirement that encourages work experience, a social insurance motive and incentive effects. Calibrations using U.S. data yield higher expected optimal marginal income tax rates for experienced workers for most of the inexperienced workers. They confirm that the average marginal income tax rate increases (decreases) with age when shocks and work experience are substitutes (complements). Finally, more variability in experienced workers' earnings prospects leads to increasing tax rates since income taxation acts as a social insurance mechanism. In the second essay, the properties of an optimal tax system are investigated in a dynamic private information economy where labor market frictions create unemployment that destroys workers' human capital. A two-skill type model is considered where wages and employment are endogenous. I find that the optimal tax system distorts the first-period wages of all workers below their efficient levels which leads to more employment. The standard no-distortion-at-the-top result no longer holds due to the combination of private information and the destruction of human capital. I show this result analytically under the Maximin social welfare function and confirm it numerically for a general social welfare function. I also investigate the use of a training program and job creation subsidies. The final essay analyzes the optimal linear tax system when there is a population of individuals whose perceptions of savings are linked to their disposable income and their family background through family cultural transmission. Aside from the standard equity/efficiency trade-off, taxes account for the endogeneity of perceptions through two channels. First, taxing labor decreases income, which decreases the perception of savings through time. Second, taxation on savings corrects for the misperceptions of workers and thus savings and labor decisions. Numerical simulations confirm that behavioral issues push labor income taxes upward to finance saving subsidies. Government transfers to individuals are also decreased to finance those same subsidies.

Parallel dynamic load-balancing for adaptive unstructured meshes

This chapter describes a parallel optimization technique that incorporates a distributed load-balancing algorithm and provides an extremely fast solution to the problem of load-balancing adaptive unstructured meshes. Moreover, a parallel graph contraction technique can be employed to enhance the partition quality and the resulting strategy outperforms or matches results from existing state-of-the-art static mesh partitioning algorithms. The strategy can also be applied to static partitioning problems. Dynamic procedures have been found to be much faster than static techniques, to provide partitions of similar or higher quality and, in comparison, involve the migration of a fraction of the data. The method employs a new iterative optimization technique that balances the workload and attempts to minimize the interprocessor communications overhead. Experiments on a series of adaptively refined meshes indicate that the algorithm provides partitions of an equivalent or higher quality to static partitioners (which do not reuse the existing partition) and much more quickly. The dynamic evolution of load has three major influences on possible partitioning techniques; cost, reuse, and parallelism. The unstructured mesh may be modified every few time-steps and so the load-balancing must have a low cost relative to that of the solution algorithm in between remeshing.

Dynamic multi-partitioning for parallel finite element applications

The central product of the DRAMA (Dynamic Re-Allocation of Meshes for parallel Finite Element Applications) project is a library comprising a variety of tools for dynamic re-partitioning of unstructured Finite Element (FE) applications. The input to the DRAMA library is the computational mesh, and corresponding costs, partitioned into sub-domains. The core library functions then perform a parallel computation of a mesh re-allocation that will re-balance the costs based on the DRAMA cost model. We discuss the basic features of this cost model, which allows a general approach to load identification, modelling and imbalance minimisation. Results from crash simulations are presented which show the necessity for multi-phase/multi-constraint partitioning components.

Asymptotic preserving and time diminishing schemes for rarefied gas dynamic

In this work, we introduce a new class of numerical schemes for rarefied gas dynamic problems described by collisional kinetic equations. The idea consists in reformulating the problem using a micro-macro decomposition and successively in solving the microscopic part by using asymptotic preserving Monte Carlo methods. We consider two types of decompositions, the first leading to the Euler system of gas dynamics while the second to the Navier-Stokes equations for the macroscopic part. In addition, the particle method which solves the microscopic part is designed in such a way that the global scheme becomes computationally less expensive as the solution approaches the equilibrium state as opposite to standard methods for kinetic equations which computational cost increases with the number of interactions. At the same time, the statistical error due to the particle part of the solution decreases as the system approach the equilibrium state. This causes the method to degenerate to the sole solution of the macroscopic hydrodynamic equations (Euler or Navier-Stokes) in the limit of infinite number of collisions. In a last part, we will show the behaviors of this new approach in comparisons to standard Monte Carlo techniques for solving the kinetic equation by testing it on different problems which typically arise in rarefied gas dynamic simulations.

Dynamic Programming Approaches for Estimating and Applying Large-scale Discrete Choice Models

People go through their life making all kinds of decisions, and some of these decisions affect their demand for transportation, for example, their choices of where to live and where to work, how and when to travel and which route to take. Transport related choices are typically time dependent and characterized by large number of alternatives that can be spatially correlated. This thesis deals with models that can be used to analyze and predict discrete choices in large-scale networks. The proposed models and methods are highly relevant for, but not limited to, transport applications. We model decisions as sequences of choices within the dynamic discrete choice framework, also known as parametric Markov decision processes. Such models are known to be difficult to estimate and to apply to make predictions because dynamic programming problems need to be solved in order to compute choice probabilities. In this thesis we show that it is possible to explore the network structure and the flexibility of dynamic programming so that the dynamic discrete choice modeling approach is not only useful to model time dependent choices, but also makes it easier to model large-scale static choices. The thesis consists of seven articles containing a number of models and methods for estimating, applying and testing large-scale discrete choice models. In the following we group the contributions under three themes: route choice modeling, large-scale multivariate extreme value (MEV) model estimation and nonlinear optimization algorithms. Five articles are related to route choice modeling. We propose different dynamic discrete choice models that allow paths to be correlated based on the MEV and mixed logit models. The resulting route choice models become expensive to estimate and we deal with this challenge by proposing innovative methods that allow to reduce the estimation cost. For example, we propose a decomposition method that not only opens up for possibility of mixing, but also speeds up the estimation for simple logit models, which has implications also for traffic simulation. Moreover, we compare the utility maximization and regret minimization decision rules, and we propose a misspecification test for logit-based route choice models. The second theme is related to the estimation of static discrete choice models with large choice sets. We establish that a class of MEV models can be reformulated as dynamic discrete choice models on the networks of correlation structures. These dynamic models can then be estimated quickly using dynamic programming techniques and an efficient nonlinear optimization algorithm. Finally, the third theme focuses on structured quasi-Newton techniques for estimating discrete choice models by maximum likelihood. We examine and adapt switching methods that can be easily integrated into usual optimization algorithms (line search and trust region) to accelerate the estimation process. The proposed dynamic discrete choice models and estimation methods can be used in various discrete choice applications. In the area of big data analytics, models that can deal with large choice sets and sequential choices are important. Our research can therefore be of interest in various demand analysis applications (predictive analytics) or can be integrated with optimization models (prescriptive analytics). Furthermore, our studies indicate the potential of dynamic programming techniques in this context, even for static models, which opens up a variety of future research directions.

On the study of train-track dynamic interactions caused by rail welds on discrete supported rails

The paper studies the influence of rail weld dip on wheel-rail contact dynamics, with particular reference to freight trains where it is important to increase the operating speed and also the load transported. This has produced a very precise model, albeit simple and cost-effective, which has enabled train-track dynamic interactions over rail welds to be studied to make it possible to quantify the influence on dynamic forces and displacements of the welding geometry; of the position of the weld relative to the sleeper; of the vehicle's speed; and of the axle load and wheelset unsprung mass. It is a vertical model on the spatial domain and is drawn up in a simple fashion from vertical track receptances. For the type of track and vehicle used, the results obtained enable the quantification of increases in wheel-rail contact forces due to the new speed and load conditions.

On the Cost of Using Happy Eyeballs for Transport Protocol Selection

Concerns have been raised in the past several years that introducing new transport protocols on the Internet has be- come increasingly difficult, not least because there is no agreed-upon way for a source end host to find out if a trans- port protocol is supported all the way to a destination peer. A solution to a similar problem—finding out support for IPv6—has been proposed and is currently being deployed: the Happy Eyeballs (HE) mechanism. HE has also been proposed as an efficient way for an application to select an appropriate transport protocol. Still, there are few, if any, performance evaluations of transport HE. This paper demonstrates that transport HE could indeed be a feasible solution to the transport support problem. The paper evaluates HE between TCP and SCTP using TLS encrypted and unencrypted traffic, and shows that although there is indeed a cost in terms of CPU load to introduce HE, the cost is rel- atively small, especially in comparison with the cost of using TLS encryption. Moreover, our results suggest that HE has a marginal impact on memory usage. Finally, by introduc- ing caching of previous connection attempts, the additional cost of transport HE could be significantly reduced. 

Assessing mackerel scad, Decapterus macarellus, in Cape Verde: using a Bayesian approach to biomass dynamic modelling in a data-limited situation

The fisheries for mackerel scad, Decapterus macarellus, are particularly important in Cape Verde, constituting almost 40% of total catches at the peak of the fishery in 1997 and 1998 ( 3700 tonnes). Catches have been stable at a much lower level of about 2 100 tonnes in recent years. Given the importance of mackerel scad in terms of catch weight and local food security, there is an urgent need for updated assessment. Stock assessment was carried out using a Bayesian approach to biomass dynamic modelling. In order to tackle the problem of a non-informative CPUE series, the intrinsic rate of increase, r, was estimated separately, and the ratio B-0/X, initial biomass relative to carrying capacity, was assumed based on available information. The results indicated that the current level of fishing is sustainable. The probability of collapse is low, particularly in the short-term, and it is likely that biomass may increase further above B-msy, indicating a healthy stock level. It would appear that it is relatively safe to increase catches even up to 4000 tonnes. However, the marginal posterior of r was almost identical to the prior, indicating that there is relatively low information content in CPUE. This was also the case in relation to B-0/X There have been substantial increases in fishing efficiency, which have not been adequately captured by the measure used for effort (days or trips), implying that the results may be overly optimistic and should be considered preliminary. (c) 2006 Elsevier B.V. All rights reserved.

Reducing the global burden of depression : population level analysis of intervention cost-effectiveness in 14 world regions

International evidence on the cost and effects of interventions for reducing the global burden of depression remain scarce. Aims: To estimate the population-level cost-effectiveness of evidence-based depression interventions and their contribution towards reducing current burden. Method: Primary-care-based depression interventions were modelled at the level of whole populations in 14 epidemiological subregions of the world. Total population-level costs (in international dollars or I$) and effectiveness (disability adjusted life years (DALYs) averted) were combined to form average and incremental cost-effectiveness ratios. Results: Evaluated interventions have the potential to reduce the current burden of depression by 10–30%. Pharmacotherapy with older antidepressant drugs, with or without proactive collaborative care, are currently more cost-effective strategies than those using newer antidepressants, particularly in lower-income subregions. Conclusions: Even in resource-poor regions, each DALYaverted by efficient depression treatments in primary care costs less than 1 year of average per capita income, making such interventions a cost-effective use of health resources. However, current levels of burden can only be reduced significantlyif there is a substantialincrease substantial increase intreatment coverage.