In the field of choice modeling, the availability of ever-larger datasets has the potential to significantly expand our understanding of human behavior, but this prospect is limited by the poor scalability of discrete choice models: as sample sizes increase, the computational cost of maximum likelihood estimation quickly becomes intractable for anything but trivial model structures. To mitigate this issue, we investigate the use of a dataset reduction technique to generate weighted batches that better represent the whole dataset and, as a result, lead the optimization algorithm to faster convergence.

This study delves into the determinants of commuting and residential migration decisions among French-born workers exposed to Luxembourg's economy. We rely on micro-data from two different sources to study two choices of the population: that of residing in France or Luxembourg, and that of working in France or Luxembourg. The study investigates the interactions of house prices and wages with education levels, shedding light on the complex dynamics driving cross-border labor supply. Our research explores the implications of these decisions on the size and selection of individuals on both sides of the borders. As each micro-dataset only allows us to observe partial segments of the population, we propose a method to derive joint decisions, based on the combination of estimation results from the conditional datasets. Initial results reveal significant influences of wage differentials and housing market disparities on migration patterns, that vary by skill group. Further analyses explore counterfactual scenarios to understand how changes in economic conditions impact the actual selection and distribution of cross-border workers and foreign residents in Luxembourg.

We introduce the Breakpoint Exact Algorithm with Capacity (BEAC), based on the state-of-the-art Breakpoint Exact Algorithm (BEA) to address the choice-based pricing problem (CPP) with capacity constraints, together with the Breakpoint Heuristic Algorithm (BHA) for both uncapacitated and capacitated instances. We furthermore develop valid inequalities for the MILP formulation of the CPP, allowing us to use the heuristic solution to speed up the exact Branch \& Benders Decomposition (B\&BD) approach. When including capacity, an approach based on an exogenous priority queue, as well as a supplier-controlled queuing strategy to generate maximal or minimal profit for robust optimization is developed. The BHA leverages a coordinate descent method, which produces high-quality solutions in a short time. Results show that when optimizing two prices simultaneously, in the capacitated case, the BEAC reports runtimes up to 20 times faster than the state-of-the-art mixed-integer linear programming (MILP) approach, while the BHA performs from 100 to 5000 times faster than the MILP. For the uncapacitated case, the BHA outspeeds the BEA as well as the B\&BD approach by multiple orders of magnitude, especially for high-dimensional instances. Our results show significant improvements in computational time for the exact method (B\&BD) when using the heuristic solution to guide the algorithm.

In this paper, we extend the hybrid simulator from the individual to the household level by including a broader set of simulated demographic events affecting households and redefining a resampling procedure using the Gibbs Sampler. Usually, projection methods use historical demographic rates that may not account for sudden events like COVID-19, potentially hindering the accuracy of transportation models that rely on these projections. To test the resilience of projection methods to unforeseen events, we project synthetic samples from 2010 to 2021 using dynamic projection and a hybrid simulator. We test two scenarios based on pre-pandemic and post-pandemic demographic rates using Swiss Mobility and Transport Microcensus data. The results show that the hybrid simulator is more robust and less dependent on rates when it comes to unforeseen events than dynamic projection as it includes an intermediate resampling update that helps reduce the errors of dynamic projection.

TBA

We consider the decision problem under uncertainty, where the decision-maker can sequentially select information about the relevant state of the world. We show that the optimal behaviour mirrors the mechanics of the classical nested-logit model and it allows for rich substitution patterns.

Traditional Activity-based models (ABMs) treat individuals as isolated entities, limiting behavioural representation. Econometric ABMs assume agents schedule activities to maximise utility, explained through discrete choices. Using discrete choice models implies the need for calibration of maximum likelihood estimators of the parameters of the utility functions. However, classical data sources like travel diaries only contain chosen alternatives, not the full choice set, making parameter estimation challenging due to unobservable, and combinatorial activity spatio-temporal sequence. To address this, we propose a choice set generation algorithm for household activity scheduling, to estimate significant and meaningful parameters. Using a Metropolis- Hastings sampling approach, we sample an ensemble containing clusters of schedules for all agents in a household. Alternatives for all household agents are generated in parallel, encompassing household-level choices, and time arrangements. Utilising this approach, we then estimate the parameters of a household-level scheduling model presented in (Rezvany et al. 2023). This approach aims to generate behaviourally sensible parameter estimates, enhancing the model realism in capturing household dynamics.

This paper investigates the ride-acceptance behavior of drivers on ride-sourcing platforms, considering drivers’ freedom to accept or reject ride requests. Understanding drivers’ preferences is vital for ride-sourcing services to improve the matching of requests to drivers. To this end, we obtained a unique dataset from a reputable ride-sourcing platform in Iran. This dataset provides comprehensive details of driver and ride characteristics for both successful and unsuccessful matchings. We investigate the ride-acceptance behavior of drivers using a hierarchical drift-diffusion model, which captures the dependency between drivers’ choices and response times. This dependency implies that response time carries information about drivers’ preferences which allows us to better comprehend drivers’ ride-acceptance behaviors. Furthermore, we conduct a thorough comparison between the drift-diffusion model and the logit model, considering their predictive ability, parameter estimates, and elasticities. Within the drift-diffusion model framework, we also derive time-dependent elasticities of acceptance probability and elasticity of drivers’ response times. Our results demonstrate that ride fare, ride duration to request origin, and rainfall volume have the most impact on drivers’ ride-acceptance decisions. The insights derived from this study can be utilized to enhance platform matching algorithms and strategies, thereby improving the efficiency of ride-sourcing platforms.

In this work we investigate the implementation of sampling of alternatives in the context of migration aspiration models. The use of discrete choice models (DCMs) is a common approach to model migration aspirations. However, more complex specifications, such as the cross-nested logit model (CNL) are required to capture the complex substitution patterns between destinations. The estimation of CNL can be computationally expensive as the number of observations and alternatives increases. The estimation time can be reduced though via sampling of alternatives i.e. reducing the number of alternatives in the model specification. In the current work, we implement sampling of alternatives on migration aspiration choices using the Gallup World Poll data. We examine the impact of stratification and number of alternatives on the NL and CNL models with respect to quality of parameter estimates and estimation time.

We examine how Brexit-induced access restrictions have influenced migration patterns among European Union (EU) residents, particularly their destination choices. We analyze whether the change in migration policies has altered European Union residents' intentions to migrate to the UK and explore how these potential migrants are redirecting their destination choices. Using discrete choice modelling framework, we apply a logit model on the Gallup World Poll data set containing declared migration aspiration of European Union citizens. We show that Brexit has led to decreased migration intentions towards the UK among EU residents, with varying effects observed during the uncertainty period (June 2016 to 2021) and the post-Brexit period (after 2021). During the uncertainty period, access conditions remained unchanged, while post-Brexit measures restricting entry had a substantial impact on migration dynamics.

Assortment optimization problems are revenue-maximizing problems, which involve the selection of a subset of products to be offered to customers. We present the assortment optimization problem under the mixed logit model demand (AOP-MXL). To reduce the computational effort in the sample average approximation process due to large samples, we discuss multiple variance reduction techniques for this problem. We derive theoretical bounds of the predictive errors based on variance reduction methods, the random parameters distribution, and the number of draws (realizations). We conduct numerical experiments to investigate the impact of the mentioned factors on the computational effort to solve AOP-MXL.

A large literature has documented the incentive effect of emigration prospects in terms of human capital accumulation in origin countries. Much less attention has been paid to the impact on speci c educational choices. We provide some evidence from the behaviour of students of the University of Lorraine located in the North-East of France and close to Luxembourg, a booming economy with attractive work conditions. We  nd that students who paid attention to the foreign labour market at the time of enrolment tend to choose topics that lead to occupations that are highly valued in Luxembourg. These results hold when accounting for heterogeneous substitution patterns across study  elds through the estimation of advanced discrete choice models. Incentive e ects of emigra- tion prospects are also found when accounting for the potential endogeneity of the interest for the foreign labour market using a control function approach based on the initial locations of these students at the time of enrolment. Con- sistently, students showing no attention to the foreign labour market are not subject to the incentive e ect of emigration prospects.

The dualistic distinction between coercion and voluntarism in migration is controversial in both academic and practical circles. In this report, we construct a discrete choice model that takes into account life- and livelihood-threatening experiences and perceptions as a preliminary step toward clarifying the mix of coercive and voluntary factors in the desire to migrate. We estimate the model for Afghanistan, a long-term refugee departure country, and Pakistan, a country that is geographically and culturally close to Afghanistan and has high migration flows, and find that the stronger the exposure to threats, the stronger the desire to migrate.

This paper develops a fast algorithm for computing the equilibrium assignment with the perturbed utility route choice (PURC) model. Without compromise, this allows the significant advantages of the PURC model to be used in large-scale applications. We formulate the PURC equilibrium assignment problem as a convex minimization problem and find a closed-form stochastic network loading expression that allows us to formulate the Lagrangian dual of the assignment problem as an unconstrained optimization problem. To solve this dual problem, we formulate a quasi-Newton accelerated gradient descent algorithm (qN-AGD*). Our numerical evidence shows that qN-AGD* clearly outperforms a conventional primal algorithm as well as a plain accelerated gradient descent algorithm. qN-AGD* is fast with a runtime that scales about linearly with the problem size, indicating that solving the perturbed utility assignment problem is feasible also with very large networks.

Enrolment decisions at the university level are influenced by many factors, such as the number of credits, the expected difficulty and the schedule. Scheduling decisions made by universities typically take into account the availabilities (and preferences) of teachers, but disregard student preferences. Research has shown the impact of university course schedules in student absence, which in turn can negatively impact academic performance. Thus, student preferences should be taken into account to support enrolment decisions. To this end, we start this research by conducting an exploratory data analysis on historical data to assess the extent of such impact. This historical data contains information on the course offer of the last six academic years as well as information on the courses enrolled by students during those years.

This paper presents a framework for modeling epidemiological responses during a pandemic. The objective of the model is to include pandemic-related restrictions, such as imposed curfews or other activity-restriction policies, when computing activity schedules. Building upon the ABM developed by \cite{pougala_capturing_2022}, this study presents an updated formulation capable of including pandemic restriction and typical responses. In particular, we integrate two key aspects: first, we estimate latent factors that capture the psychological and emotional sensitivity of people to the pandemic's effects, integrating these into the optimization problem. Second, we account for the direct impact of restrictions on activity participation and the adaptive strategies individuals might employ, such as altering the time, location, or nature of their activities. This dual approach allows for a more comprehensive understanding of population behavior in response to public health policies. Moreover, we introduce a dynamic programming algorithm to efficiently solve the updated optimization problem. The use of dynamic programming allows for efficient handling of large-scale populations and numerous activities, a significant advancement over the limitations identified in Pougala et Al., 2022. This methodological improvement ensures accurate representation of all possible contacts, capturing the true dynamics of infection transmission within the population.

The global economy is strongly influenced by special issues, e.g., natural disasters, pandemics, wars and as well as mega trend such as digitalization or sustainability, which creates the need to adapt and optimize the commercial and operational business of companies in air cargo industry in terms of increasing yields and revenues as well as reducing cost. One of the most relevant success drivers are setting up the right price and shape of offered services according to the customers profile and needs (covering willingness to pay, price elasticity, offer/product features, mega trends etc.). There are some models used, based on the decision and choice-making theories to further support correct design of these success drivers in different industries, especially also in the aviation industry (example: dynamic pricing based on various parameters at passenger revenue management). Nevertheless, the impact of various parameters such as economic situations, mega trends (example: sustainability and regulation around the topic), relevance of industries, taxation and costs, diversity of cultures, needs and habits of the customers in air cargo industry and putting those parameters in the regional perspective is mostly unclear. To bridge this gap, the aim of this research project would be to investigate which parameters are valid globally, regionally, and what is their degree of impacting offer choice (to the level of the product feature) and pricing for spot offers algorithms of chosen company within air cargo industry for example Swiss WorldCargo. Once parameters would be defined, the goal would be to develop innovative models using tools like AI and machine learning to create valuable quantitative analyses and qualitative guidelines.

Welfare measurement in a discrete choice framework has a long record of theoretical investigations and applications and is a key ingredient in benefit-cost analysis. Since D. McFadden, the compensating variation (CV) has been used to evaluate welfare changes. CV is defined as the income variation an agent would need ex-post, after a change in the attributes of the alternatives to recover its ex-ante utility level. The literature has focused its attention on the expected CV. For the linear-in income logit, the expected CV is the monetized difference of logsums. This canonical formula plays a key role in the appraisal of transportation and urban projects and (environmental) policies, but it is unable to quantify the individuals gains and losses. For a given policy, some individuals may stick to their choice before an attributed change, while others may alter their choices after this change. In the spirit of discrete choice models, compensations are individual specific and described by the modeler as a random variable. The distribution of the CV for additive random utility model was derived by de Palma and Kilani (2011). Based on the distribution of the CV, we show that the analysis of inequality across the population is tractable. However, it requires to solve some technical issues concerning the definition of the inequality measurements. Methodologies for income inequality analysis have been extensively investigated in public economics. Different inequality measures have been proposed (see, e.g. Cowell, 2011). The Lorenz curve relates the cumulative proportion of individuals to the cumulative proportion of income when agents are reordered by ascending order of their income. The Gini coefficient builds on the Lorenz curve. The use of the CV per se for the analysis of welfare change is, however, problematic, since welfare change can be either positive or negative. By contrast, in their original definition, the Gini’s mean difference and the Lorenz curve refer to positive income. The few contributions that have relaxed the positive income assumption (Schutz, 1951; Chen et al., 1982) have been restricted to negative income values with positive expectation. The welfare impact of policies is measured in this paper using a Hicksian money-metric utility function, i.e. the income to be provided to the agent’s in the (ex-ante) state without the change to bring her to the level of utility in the (ex-post) state with the change. We derive new theoretical and exact formulas to characterize the distribution of the money-metric utility function. We then derive two theorems related to the Lorenz function and the Gini index for any discrete choice model. For the linear-in-income logit, and for the CES model (using its disaggregated version), it turns out that the Lorenz function and the Gini index have closed forms. We illustrate our theory using synthetic data and numerical examples, and discuss other measures of inequality that can be provided as outcomes of our theory. References de Palma, A. and K. Kilani, Transition Choice Probabilities and Welfare Analysis in Additive Random Utility Models, Economic Theory, 2011, 46 (3), 427–454. Chen, et al. “The Gini coefficient and negative income,” Oxford Economic Papers, 1982, 34 (3), 473–478. Cowell, F., Measuring inequality, Oxford University Press, 2011. Schutz, R, On the measurement of income inequality, The American Economic Review, 1951, 41 (1), 107–122.

Following several contributions in new empirical industrial organization, we propose a framework that integrates discrete game theory with statistical methods to account for strategic behaviors of people and the uncertainty stemming from imperfect information. By employing techniques from econometrics and game theory, our approach enables the estimation of demand parameters in presence of a large number of heterogenous players while considering the complex dynamics of decision-making and the influence of incomplete information. Once reviewed some key challenges, we analyse properties of some baseline specifications and we apply our approach to location choice behavior.