Archive Seminars and Workshops Quantitative Economics
Winter -Spring 2025
MLSE is a (mostly) bi-weekly seminar to foster cooperation between the Department of Microeconomics and Public Economics and the Department of Quantitative Economics. It aims to give researchers the opportunity to present their ongoing work and to facilitate cooperation
Tuesday 21 January 2025 13:15-14:15 TS53 A0.24
Authors: Marc Schröder (joint with Shaul Rosner and Laura Vargas Koch)
Title: Nash flows over time with tolls
Abstract:
Traffic congestion remains an important problem in our current society. There are multiple ways and approaches to model this. We consider a theoretic, but realistic model known as the deterministic fluid queuing model (or Vickrey's bottleneck model). We introduce (fixed) tolls to the model and test whether some of the known properties of dynamic equilibria carry over. Using three examples we show that (1) dynamic equilibria with tolls need not be unique, (2) particles might overtake in dynamic equilibria with tolls, (3) dynamic equilibria with tolls need not reach a steady state.
Tuesday 1 April 2025, 13.00 -14.00 PM TS53 H0.04
Authors: Dries Vermeulen and Dmitriy Kvasov
Title: Computation of the pre-nucleolus for non-negative monotonic games
Abstract: For a non-negative monotonic game, a coalition is called minimal if every strict subcoalition has a strictly lower value. Each non-negative monotonic game is characterized by its collection W of minimal coalitions, together with the associated value for each minimal coalition.
We show that the collection W* consisting of W, together with all coalitions that consist of a minimal coalition plus exactly one player, determines the pre-nucleolus.
We also identify a subcollection of W* that determines the pre-nucleolus on the set of all non-negative pre-imputations, and that moreover contains only a negligible fraction of all coalitions as the number of players in the game is large.
Wednesday 25 March 13:00 PM-14:00 PM TS53 A0.23
Author: Ákos Miklós Balázs, Péter Biró
Title: Comparing mechanisms for course allocation with contracts
Abstract: We study a course allocation problem with contracts which is unique in several aspects. Courses have lexicographic preferences that favour students from higher priority groups, and within these groups, those students who wish to take the course with higher-priority contract terms. Courses are also characterised by finite capacities. Students have preferences over sets of course-term pairs, which are their private information. However, they can send a signal that contains a ranking over singletons and a capacity for each contract term. It is also restricted that the same course cannot be listed with more than one contract term. We consider six different mechanisms for this course allocation problem: the HBS draft, its slight modification (referred to as SZISZ), the random serial dictatorship (RSD), the deferred acceptance with single (DASTB) and multiple tie break (DAMTB), and the latter followed by the stable improvement cycles algorithm (DAMTB+SIC). Our aim is to compare the performance of these mechanisms from several perspectives. First, we evaluate them by checking whether they satisfy certain desiderata (strategy-proofness, possible and necessary player- and student-efficiency, and pairwise stability). We also show that no mechanism can satisfy both strategy-proofness and pairwise stability, and the same is true for possible student-efficiency and pairwise stability. Next, we use a dataset containing the signalled preferences of students (as well as cardinal utilities) from 2023. We apply each mechanism to these signals several times and calculate some welfare indicators from the resulting matchings. Our findings indicate that although the RSD and DASTB mechanisms satisfy more theoretical desiderata, they are outperformed in most welfare indicators by the SZISZ and even more so by the HBS draft mechanism.
Yanru Sun will present her work at the MLSE Seminar:
Tuesday 22 April 2025, 13.00 -14.00 PM TS53 A0.23
Authors: Yanru Sun, Hao Sun, Panfei Sun, Xuanzhu Jin and Yimei Yang
Title: Elevating the corporate social responsibility level: A media supervision mechanism based on the Stackelberg-Evolutionary game model
Abstract: Environmental taxes alone may not solve the social dilemma posed by the conflict between the myopic pursuit of profit and the cost of corporate social responsibility (CSR). Designing a reasonable supervision mechanism is crucial to correcting market failures. We develop a media supervision mechanism through a Stackelberg-Evolutionary game model to study the impact of media supervision on the evolutionary behavior of the manufacturer population. Assuming the media is leader, manufacturers' demands are heterogeneous under different strategy profiles after the media determines the effort level of supervision. The best response of the manufacturer population is the evolutionarily stable strategy under supervision, where the percentage of CSR strategies is defined as CSR level. It is proved that the CSR level elevates with the increase of effort level. We analyze the existence and uniqueness of Stackelberg-Evolutionary equilibrium and a numerical algorithm to compute it. The results show that CSR level under Stackelberg-Evolutionary equilibrium is higher than that without supervision. Our research not only illustrates the effectiveness of media supervision in reducing environmental pollution but also provides suggestions for governments to formulate environmental policies and improve regulatory mechanisms.
Jing Ren will present her work at the MLSE Seminar:
Tuesday 20 May 2025, 13.00 -14.00 PM, TS 53 C-1.07
Authors: Jing Ren, Iwan Bos, Dries Vermeulen
Title: Myopic coalition formation
Abstract: This paper studies a dynamic coalition formation process, which generates a myopically rational coalition sequence. Our results show that such a dynamic coalition formation process does not have a cycle, which implies that a myopically rational coalition sequence converges to a stable coalition. Moreover, our results have a wide application in economic and social settings. In this paper, we examine the models with open membership and restricted membership, where the dynamic coalition formation process is acyclic. Additionally, our results apply to the setting with and without ranked admission.
Bas Dietzenbacher will present his work at the MLSE Seminar:
Tuesday 13 May 2025, 13:15-14:15PM, TS53 A0.23
Authors: Shasha Ding, Bas Dietzenbacher, Hans Peters
Title: Strategic cartel profit sharing
Abstract: This paper studies firms in a collusive oligopoly that divide the cartel profits based on their reported capacities. We model these situations as biform games, where the capacities are strategically reported in the form of threats, and these reported capacities result in a cooperative game. We define a family of allocation rules that divide the cartel profits among the collusive firms based on this cooperative game and study the corresponding equilibrium capacities reported when these allocation rules are implemented. We compare these equilibria with the Cournot equilibrium and particularly focus on the equal split rule, the Shapley value, and the dual equal split rule.
July 1st, 14:00-15:00
Authors: Pedro Gonzalez-Fernandez, Stefan Terstiege, and Elias Tsakas
Title: Regulating Information
Abstract: Economic agents often seek to acquire information before taking an important action, but in many domains, gathering this information requires approval from a regulator. This paper develops a model in which an agent designs an experiment to inform his decision, but can only implement it if a regulator authorizes it ex ante. We characterize the agent’s optimal experiment under this approval constraint and show that, whenever the regulator rejects full revelation, the agent strategically reduces informativeness in the states where their disagreement is strongest. We then extend the model to settings with multiple regulators, comparing sequential and collective approval mechanisms. The analysis yields predictions for how institutional structure shapes access to information, with applications to clinical trials, data privacy, and ethics boards.
Operations Research
Joint ORBEL - NGB conference on Operations Research at Maastricht University
29 January 2025 12:00 - 31 January 2025 16:15
Joint ORBEL - NGB Conference on Operations Research
The conference is intended as a meeting place for researchers, users and potential users of Operational Research, Statistics, Computer Science and related fields. It will provide managers, practitioners, and researchers with a unique opportunity to exchange information on quantitative techniques for decision-making.
The joint conference will take place at the School of Business and Economics of Maastricht University, and will be organized by the Department of Quantitative Economics, section Operations Research.
The objective of Operational Researchers is to work with clients to find practical and pragmatic solutions to operational or strategic problems, often working within tight timing constraints. Once a good or better way of proceeding has been identified, Operational Researchers can also be central to the management of implementing the proposed changes.
Organizations may seek an extensive range of operational improvements - for example, greater efficiency, better customer service, higher quality or lower cost. Whatever the business engineering aim, OR can offer the flexibility and adaptability to provide objective help. Most of the problems OR tackles are messy and complex, often entailing considerable uncertainty. OR can use advanced quantitative methods, modelling, problem structuring, simulation and other analytical techniques to examine assumptions, facilitate an in-depth understanding and decide on
Tuesday, March 11th, Janos will present his work at the MLSE Seminar: Games with a Finite Yet Arbitrarily Large Number of Active Players.
To secure a nice room in a convenient period of time that allows for questions after the presentation, our seminar will start 15 minutes later than usual.
Date and time: March 11th, 2025 (13:15-14:15), TS53 C-1.09
Authors: Janos Flesch, Miklós Pintér, Arkadi Predtetchinski, and William Sudderth
Title: Games with a Finite Yet Arbitrarily Large Number of Active Players
Abstract:
We propose a model of games with infinitely many players. Its key feature is that the set of active players is finite almost surely, and yet, every single player becomes active with probability 1. More precisely, in such a game:
1. A finite set of players, called active, is drawn from an infinite set of potential players, according to the so-called selection charge. The selection charge is a finitely additive probability measure on the collection of finite subsets of players with the property that, for every single player, the probability to be active is 1.
2. Each active player is informed that she is active, but not of who the other active players are.
3. Finally, the active players choose their actions simultaneously, and, depending on the set of active players and the action profile, they receive the payoffs.
We examine the Nash equilibria of these games. We show that these games admit a Nash equilibrium under various conditions, among others: (i) when the action sets are compact metric and for each given player, the collection of her payoff functions for the different possible sets of active players is equicontinuous, and (ii) when the players have a common finite action set and each player’s payoff depends on her own action and continuously on the frequency of the actions chosen by the other active players, and (iii) in minority games where there are only two actions and each player’s goal is to choose the action that is chosen by the minority of the active players. Some of these results depend heavily on whether iterated integrals, where one integral is taken with respect to a finitely additive probability measure, depend on the order of integration.
Abstract:
Traffic congestion remains an important problem in our current society. There are multiple ways and approaches to model this. We consider a theoretic, but realistic model known as the deterministic fluid queuing model (or Vickrey's bottleneck model). We introduce (fixed) tolls to the model and test whether some of the known properties of dynamic equilibria carry over. Using three examples we show that (1) dynamic equilibria with tolls need not be unique, (2) particles might overtake in dynamic equilibria with tolls, (3) dynamic equilibria with tolls need not reach a steady state.
Econometrics
Author: Rosnel Sessinou (Erasmus University Rotterdam)
Wednesday 12 February 2025 15:00 PM 16:00 PM TS53 Room info later
Title: Validating a Selected Model: Encompassing, Progression, and Redundancy Testing
Abstract: Testing for encompassing is a necessary preliminary step before conducting valid post-selection inference, such as efficient market hypothesis testing. However, no existing test accommodates high-dimensional stationary data, such as the US factor zoo dataset. This paper introduces a Subseries-based Cauchy Combination Test (SCT) that fills this gap. SCT is a (high-dimensional) score test that bypasses the need to estimate large covariance matrices; unlike the Wald or J tests, it outperforms in low and high dimensions. SCT is unbiased and has power at least equal to the minimum p-value test or sup-score test powers. When developing linear factor models, SCT enables reproducibility, redundancy, or progression testing. Applying SCT to US equity market data from 1964 to 2020 reveals that some prominent US factor models can span the mean-variance efficient frontier of US blue-chip stocks even if these models are all misspecified. In line with the recent literature, SCT fails to reject the null hypothesis of a replication crisis, suggesting no widespread breakdown in the factors' performance. However, SCT also fails to reject the null hypotheses of factor redundancy in the US factor zoo and, therefore, that of a progression crisis in the asset pricing literature during the same period.
Author: Barend Spanjers (VU Amsterdam)
Wednesday 26 March 15:00 PM-16:00 PM
Titel:
Increased persistence of warm and wet winter weather in north-western Europe due to trends towards strongly positive NAO
Abstract:
The winter North Atlantic Oscillation (NAO) has seen a long-term trend towards strongly positive values over the last decades. Although this shift aligns with climate model projections under scenarios of strong greenhouse gas forcing, the observed changes exceed the range predicted by these models. While the relationship between the NAO and temperature or precipitation patterns is well-established, the shift’s impact on weather persistence remains unclear and may influence different parts of the temperature distribution and the probability of precipitation in distinct ways. We introduce statistical models to capture this distributional heterogeneity. When comparing 1950–1980 with 1990–2020, we find that both warm temperature and precipitation persistence have increased significantly in north-western Europe in winter.
This project is joint work with Eric Beutner, Dim Coumou and Julia Schaumburg.
Author: Jad Beyhum (KU Leuven)
Wednesday 2 April 15:00 PM-16:00 PM
Title: Inference after discretizing unobserved heterogeneity
Paper: https://arxiv.org/abs/2412.07352
Details follow.
Author: Simon Freyaldenhoven (Federal Reserve Bank of Philadelphia)
Wednesday 16 April 15:00 PM- 16:00 PM
(Visualizing) Plausible Treatment Effect Paths –
joint with Christian Hansen
We consider point estimation and inference for the treatment effect path of a policy. Examples include dynamic treatment effects in microeconomics, impulse response functions in macroeconomics, and event study paths in finance. We present two sets of plausible bounds to quantify and visualize the uncertainty associated with this object. One set of bounds covers the average (or overall) effect rather than the entire treatment path. Our second set of bounds imposes data-driven smoothness restrictions on the treatment path. Post-selection Inference (Berk et al. [2013]) gives us formal coverage guarantees for these bounds. The chosen restrictions also imply novel point estimates that perform well across our simulations. Both plausible bounds are often substantially tighter than traditional confidence intervals, and can provide useful insights even when traditional (uniform) confidence bands appear uninformative.
Author: Sofia Borodich Suarez (University of Luxembourg and UM)
Wednesday 14 May 1500 PM -16:00 PM
Title: Robust Priors in Non-linear Panel Data Models - Estimating Average Marginal Effects.
Abstract: In a seminal paper, Arellano and Bonhomme (“Robust priors in nonlinear panel data models,” Econometrica, 2009) propose priors to simultaneously reduce the bias for estimating the common parameters (theta_0) and the average marginal effects (M) in non-linear panel data models with fixed effects. However, we show that the Arellano-Bonhomme (AB) priors are not simultaneously bias reducing by proving mathematically that although the AB priors always reduce the bias for estimating theta_0, they do not always reduce the bias for estimating M. E.g., in the static panel probit model or, generally, for dynamic panel data models, the AB priors only reduce the bias for estimating theta_0 but not for estimating M, whereas in the static panel logit and Poisson models the AB priors simultaneously reduce the bias for estimating theta_0 and M. Furthermore, we construct priors that we prove are simultaneously bias reducing for general non-linear panel data models including panel probit, and show numerically in a simulation study that the AB priors for a panel probit model do not reduce the bias for estimating M — the bias of the estimator of M based on the AB priors remains comparable to the bias of the maximum likelihood estimator of M even when the panel is long — whereas the priors that we construct do.
This is joint work with her supervisors, Martin Schumann and Gautam Tripathi. It appears in part in print in the forthcoming (Arellano, M., Bonhomme, S., Borodich Suarez, S., Schumann, M., Shi, X., and Tripathi, G. (2025): "Erratum to “Robust Priors in Nonlinear Panel Data Models”," Econometrica, 93, 3 , doi: 10.3982/ECTA23441, in production).
EPICENTER Summer Course on Epistemic Game Theory
Reasoning in Static and Dynamic Games
Maastricht University, June 30 – July 11, 2025
Website: Detailed information about the course, and how to register, can be found on the course website https://www.epicenter.name/summercourse/ .
About the course: Are you interested in game theory, and its relation to human reasoning and decision making? Then this is the perfect course for you.
In this course we explore game theoretic situations from an epistemic perspective, by zooming in on the reasoning of a player before he makes a decision in the game. This reasoning does not only concern the possible choices of his opponents, but also the beliefs that his opponents may have before they make a choice.
The course consists of three parts: standard reasoning in static games, cautious reasoning in static games, and reasoning in dynamic games.
In static games, the players only make one choice, and choose in complete ignorance of the other players' choices. For standard reasoning we focus on the central reasoning concept of common belief in rationality, and investigate what happens if we add a correct beliefs assumption.
Cautious reasoning means thar a player, before he makes a choice himself, does not completely discard any opponent's choice from consideration. For this part we will investigate different variants of common belief in rationality.
In dynamic games, a player may have to make more than one choice, and may fully or partially observe what other players have done before he makes a choice himself. We explore backward and forward induction reasoning, embodied by the reasoning concepts of common belief in future rationality and common strong belief in rationality.
The course is open to advanced bachelor students, master students, PhD students and researchers all over the world.
Book: This course is based on the textbook Epistemic Game Theory: Reasoning and Choice by Andrés Perea, which was published by Cambridge University Press in 2012.
Difference with previous edition: In 2024 we provided a course on incomplete information, unawareness and psychological games. To follow this year’s course, it is not necessary to have followed the previous edition. In fact, you should be able to follow this year’s course without having any prior knowledge about game theory. The course is completely self-contained.
Forward: Please forward this message to all people whom you think might be interested.
Questions? Please send an E-mail to Andrés Perea at: course@epicenter.name
Autumn 2025
MLSE Seminars Programme 2025 Autumn
Date and time: September 9th, 12:00-13:00 Room: A1.22
Authors: Galit Ashkenazi-Golan, János Flesch, Eilon Solan
Title: Games with an infinite past
Abstract: We study multi-player games with perfect information and general payoff functions, where the set of stages is the set of non-positive integers {...,−2,−1,0}. We define two related equilibrium notions: one only considering deviations at finitely many stages and another considering all deviations. We show that (i) The sets of equilibrium plays coincide for the two equilibrium notions, provided that at least two players are active along each infinite play. (ii) For win-lose games, the game has an equilibrium if the winning sets have Borel rank at most 2, and we provide a counter-example showing that this is no longer true for Borel rank 3. (iii) In general non-zerosum games, the game has an equilibrium if either (i) the payoff functions have finite range and Borel rank at most 2, or (ii) the payoff functions are continuous, for example, with reversed-time discounted payoffs. The challenge for all these results is that not all strategy profiles admit a consistent infinite play, hampering the use of backward induction arguments.
On 23 of September, Stefan Terstiege will present his work at the MLSE Seminar: Design and sale of market segments.
Date and time: 23 September, 13:00-14:00 TS53 A0.24
Authors: J. Bizzotto, J. Rüdiger
Title: Design and sale of market segments
Abstract: A platform segments a market and runs auctions in which firms bid for the
access to market segments. Segmenting the market more finely allows firms to
extract more profits from consumers but can decrease competition in the auctions.
We first characterize extremal markets: markets that cannot be segmented further
without reducing auction revenue. Second, we employ this result to characterize
optimal segmentations in a Hotelling environment. Our results give an insight
into the welfare losses and consumer surpluses resulting from for-profit market
segmentation.
Tuesday 4 November, 13:15-14:15
Room: A1.22
Author: Andrés Perea
Title: Pure backward induction reasoning in dynamic games
Abstract:
Pure backward induction reasoning means that the solution of a game, when applied to a subgame, should yield the same as applying the solution to the subgame in isolation. Remarkably, none of the existing backward induction concepts in the literature satisfies this criterion. The purpose of this paper is to present, and analyze, a reasoning concept that does meet the pure backward induction principle. In the concept, a player can only commit to his current action, not to his future actions. At every history, the player holds a probabilistic belief about the opponents' current and future actions, and his own future actions, and chooses his current action based on that belief. The key conditions are that a player believes that he, and his opponents, will choose rationally in the future, that he believes that every opponent believes that he, and the other players, will choose rationally in the future and so on. We show that the resulting actions can be obtained by an easy-to-use iterated elimination procedure, which may be viewed as a natural generalization of the backward induction procedure. In finitely repeated games, the procedure reduces to an even easier algorithm that can be used to compute the possible actions at every period.
Date and time: 18 November 18, 13:15-14:15
Author: Jean-Jacques Herings, Christian Seel, Arkadi Predtetchinski
Room A1.22
Title: Random networks
Abstract: Akin to the literature on random games, we analyze pairwise stability (Jackson and Wolinsky, 1996) for two models of socioeconomic networks in which the utilities of each individual are drawn at random. In the case of network-dependent utilities, the utility is drawn at random for each possible network, whereas in the case of neighbor-dependent utilities, the utility is required to be the same whenever the individual has the same set of links.
For each model, as the network gets large, the probability that at least one pairwise stable network exists approaches one, but the expected fraction of pairwise stable networks approaches zero. We also obtain lower and upper bounds on the expected number of pairwise stable networks as the network gets large. For the case of network-dependent utilities, we provide additional results on the distribution of pairwise stable networks.
QE Seminars Programme 2025 Autumn - Winter
Wednesday September 3, 12.30-13. 30 at TS53 A1.22.
The speaker of our first session is Kenneth Sörensen (University of Antwerp).
We invite you all to attend the seminars and hope to see you there. An overview of confirmed speakers during the upcoming weeks is included at the bottom of this email as well, so you might want to block your agenda weekly on Wednesdays already.
------------
Metaheuristics: too much empirical research or not enough?
World War II gave us mathematical programming, a discipline born from logistical necessity and built on rigorous mathematical foundations. Duality theory provided elegant economic interpretations, complexity theory classified tractability, and polyhedral theory offered deep geometric insights into solution spaces. As a result, the field of optimization became firmly embedded within applied mathematics.
Yet many practical optimization problems resisted exact solution methods, leading to the rise of heuristics and, later, metaheuristics. From the outset, efforts were made to ground these approaches in mathematical theory. Landscape analysis, convergence proofs, and runtime analyses aimed to inform heuristic design and provide theoretical insight. However, such results often relied on strong assumptions, and their practical relevance has been limited. Unlike in linear programming, where theory directly informs both modeling and algorithmic choices, the connection between theory and practice in metaheuristics has remained weak.
As a result, most impactful metaheuristics research, especially that which drives progress in optimization practice, has remained empirical. Researchers build algorithms that outperform others, including exact methods, but typically without formal guarantees or theoretical justification. Many therefore argue that the field of metaheuristics is too empirical, lacking the theoretical rigor expected of a mature scientific discipline.
This talk takes a different view: the problem is not that we do too much empirical research, but that we do too little of the right kind. We must move beyond superficial performance comparisons and toward structured empirical work that generates cumulative knowledge. An evidence-based and methodologically sound empirical culture is not a weakness, but can form the foundation of a mature and credible scientific discipline.
Speaker: Marc Schröder
Wednesday 10 September, 12:30-13:30
Room: A1.22
Title: Symmetric Nash equilibria in symmetric congestion games
Abstract: We study the efficiency of symmetric Nash equilibria in symmetric congestion games. These games are motivated by for example traffic networks, scheduling problems or oligopoly markets. We show that in order to solve this game-theoretic problem, we need tools from other fields like duality of linear programs and the convergence of Binomial distributions to Poisson distributions.
Speaker: Garth Tarr (The University of Sydney)
Wednesday 17 September , 12:30-13:30
Room: A1.22
Title: Outlier-robust estimation of state-space models using a penalised approach
Abstract: State-space models are a broad class of statistical models for time-varying data. The usual Gaussian assumption on the observation noise can lead to poor parameter estimates in the presence of observational outliers. We address this issue using a penalised approach that introduces a shift parameter at each timepoint: outliers are expected to receive non-zero shifts, while clean observations retain zero shifts after estimation. A sparsity-inducing penalty is applied to prevent all shift parameters from being non-zero. In addition to enabling robust and reliable estimation in the presence of outliers, this approach provides automatic outlier detection and visual diagnostics to help researchers and practitioners better understand the outlier structure in their data. The utility of this method will be demonstrated on animal tracking data.
Wednesday 24 September. Exceptionally, this week’s seminar will start at 12:00, instead of 12:30.
Room: A1.22
Speaker: Tom Demeulemeester (Maastricht University)
Title: Smart Lotteries in School Choice: Ex-ante Pareto Improvement with Ex-post Stability
Abstract: We study the school choice problem, where school seats are assigned to prospective students who have preferences over them. Because schools are often indifferent between large subsets of the students, it is common practice to use random tie-breaking techniques. In this paper, we study how we can improve the students’ assigned schools, in comparison to applying the celebrated deferred acceptance (DA) algorithm with random tie-breaking, without violating the stability of the solution. We show that both this problem and computing the expected outcome of DA with random tie-breaking are NP-hard. We also propose a heuristic, and illustrate how it improves upon DA with random tie-breaking, but also upon other existing mechanisms from the literature that Pareto-dominate DA. Along the way, I aim to give a general overview of the literature on applying mathematical programming techniques to school choice problems.
Wednesday 1 October, 12:30-13:30.
Speaker: Daniele Girolimetto
Title: Forecast Combination and Reconciliation
Abstract:
Forecasts play a central role in decision-making across economics, business, energy, and policy. So far, two recurring challenges arise in practice. First, forecasts often need to satisfy logical or accounting constraints: for example, national GDP must equal the sum of its income, expenditure, and output measures; or total electricity demand must equal the sum of demand across regions and fuel types. Producing forecasts that are accurate and also coherent can be challenging. Second, forecasts may come from multiple sources: different models, experts, or institutions frequently produce competing forecasts, each capturing different aspects of the underlying system. How can we combine these multiple forecasts, while also satisfying the necessary constraints?
https://www.maastrichtuniversity.nl/archive-seminars-and-workshops-quantitative-economics
This talk introduces coherent forecast combination, a framework that addresses these two challenges simultaneously. It extends the well-established idea of forecast reconciliation by integrating it with the broader task of forecast combination. In doing so, it leverages the strengths of both approaches: forecast combination improves accuracy by pooling diverse sources of information, while reconciliation ensures that forecasts satisfy the required constraints or accounting identities. We consider different linear approaches. First, combination and reconciliation are solved together in a single optimization step, producing forecasts that are both accurate and coherent by design. Then, we present a sequential approach, where the two steps are performed one after the other: we can first combine forecasts for each individual variable and then reconcile them, or vice versa. Both strategies have practical advantages depending on the application, the data structure, and the availability of forecasts.
Finally, we show that coherent forecast combination delivers 15min forecasts of Italian energy load disaggregated by geographical bidding zones, that are both more accurate and aligned with the geographical constraints. This illustrates the value of this framework: making better use of the forecasts we already have, while ensuring results remain consistent and trustworthy. The open-source R package FoCo2 makes these approaches directly available for applied work.
Wednesday 8 October, 12:30-13:30
Room A1.22
Speaker: Dries Vermeulen (Department of Quantitative Economics, Maastricht University)
Title: The pre-nucleolus for monotonic games
Joint work with Dmitriy Kvasov (Waseda University)
Abstract: For a monotonic game, a coalition is called minimal if every strict subcoalition has a strictly lower value. Each monotonic game is characterized by its collection W of minimal coalitions, together with the associated value for each minimal coalition.
We show that the collection W* consisting of W, together with all coalitions that consist of a minimal coalition plus exactly one player, determines the pre-nucleolus. We also identify a subcollection of W* that determines the pre-nucleolus on the set of all non-negative pre-imputations, and that moreover contains only a negligible fraction of all coalitions as the number of players in the game is large.
In the proofs, we use results from combinatorial mathematics and probability theory, notably Sperner’s Theorem, which provides a bound on the number of elements in an anti-chain, and a bound on binomial coefficients, closely related to Stirling’s formula.
Wednesday 15 October, 12:30-13:30.
Room TS 53 A1.22
Speakers: Frits Spieksma (TU Eindhoven)
Title: Price of Diversity: the case of the TSP
Abstract: We describe a framework for quantifying the trade-off between diversity and optimality in combinatorial optimization problems, which we refer to as the Price of Diversity (PoD).
We apply this concept to the Traveling Salesman Problem (TSP) with the triangle inequality, and show how the demand for multiple diverse solutions influences tour quality. Specifically, we consider the problem of finding two edge-disjoint tours that minimize their bottleneck cost (the length of the longer tour). For this setting , we show that the PoD, is asymptotically 8/5 in a (special) one-dimensional case and exactly 2 in general. We also consider a stricter form of diversity for the TSP and connect the resulting problem to placing queens on a chess board.
Wednesday 22 October, 12:30-13:30
Room TS53 A1.22
Speakers: Ties Bos (QE), Jip de Kok (MUMC+), Frank van Rosmalen (MUMC+)
Title: Where KE meets medicine
The Intensive Care Unit (ICU) at MUMC+ collects vast amounts of data that capture complex and dynamic processes. Patient data include vital signs, treatments, outcomes, and resource use amongst others. Such large real-world databases provide great opportunities for research but also come with substantial data quality challenges. The data is noisy, sparse, unbalanced, inherently heterogeneous, and contains many endogenous relationships. In their “Table0” paper (
https://shorturl.at/JZfeI) the team of the ICU illustrates how a dataset containing observations of over 54,000 ICU admissions has been meticulously prepared for research. Such datasets provide rich opportunities for data-driven research but many methodological challenges remain to be tackled.
This talk introduces a growing collaboration between the Department of Quantitative Economics (KE) and the ICU. It starts by outlining how and what data are available at the ICU, and what practical and methodological challenges one might face when analysing them. Next, it highlights how the data are currently used in clinical research, with a few examples such as clustering patient populations, predicting health outcomes, and validating published models on real-world data.
Finally, the talk explores how econometric methods and insights can help address some of the ICU’s analytical challenges, using existing examples from past master and bachelor theses from collaborations between KE and MUMC+, and a glimpse into the future with a recently started joint PhD project, tackling some of these methodological problems by taking a panel data econometric approach.
Wednesday 29 October, 12:30-13:30
Room TS 53 A1.22
Speaker: Shahrezad Fahmy (QE)
Title: Joint Optimization of Fixed and On-Demand Public Transportation
Abstract: What if cities could redesign public transport networks to balance fixed and on-demand bus routes? Traditional fixed-route public transport is efficient in high-demand areas, but becomes less effective when demand is low or variable. Conversely, on-demand services provide flexibility but lack the scalability and cost-efficiency of fixed networks.
This ongoing research addresses that challenge by jointly optimizing both modes within a single, integrated framework. The proposed model integrates the design and operation of fixed-line and on-demand services. The framework combines line planning, frequency setting for fixed lines, hub location selection, and on-demand routing within a unified multilevel heuristic approach.
Wednesday 5 November, 12:30-13:30
Room TS 53 A1.22
Speaker: Etienne Wijler (Vrije Universiteit Amsterdam)
Title: Can I Depend on You? An Impartial Look at Asset Correlation Stability and Market Structure
Abstract: We develop a data-driven procedure to identify which correlations in high-dimensional dynamic systems should be time-varying, constant, or zero. The method integrates a vine-based multivariate partial correlation model with sequential penalized estimation. Applied to 50 US equities and systematic risk factors, results indicate that asset-level correlation dynamics are primarily induced by time-varying exposures to systematic factors. We further uncover persistent, non-zero, and occasionally time-varying partial correlations within industries, even after controlling for standard risk and industry factors. Finally, we show how the new methodology may be used to explore the relevance of systematic risk factors in an impartial way.
Wednesday 12 November 12:30-13:30.
Room TS 53 A1.22
Speaker: Xuanzhu Jin
Title: Competitive Facility Location Games with Congestion Effects
Abstract: We consider a competitive facility location game in a market with finitely many firms and a unit mass of consumers. First, firms simultaneously choose where to set up a facility, with the objective of attracting as many consumers as possible. Second, based on the firms’ location strategy profile, consumers travel to a firm’s location, with the aim of minimizing their own travel cost. We start with two firms and prove the existence of pure Nash equilibria (PNE) by introducing a binary relation over the set of locations. For finitely many firms, we show that every instance of the competitive location game with linear travel cost functions admits a PNE and the price of anarchy equals 2. In contrast, when extending the analysis to cases with affine cost functions, a PNE may not exist.
Wednesday 19 November 12:30-13:30.
Room TS 53 A1.22
Speaker: Yordi van Kruchten
Title:
Optimizing Capacity Management for the Treatment Daycare Center for Oncology and Hematology: From Thesis Research to Practical Application
Abstract:
This seminar will present the development of a heuristic scheduling framework for the Treatment Daycare Center for Oncology and Hematology at Maastricht UMC+ (MUMC+), from the initial research phase to its ongoing application in practice. The primary goal of the research was to design an efficient scheduling template that optimizes chair allocation and nurse workload. However, before full implementation, efforts are focused on reducing variability at the front-end of the process to ensure smoother operations. This work involves aligning patient inflow from outpatient clinics with available treatment capacity, thus minimizing inefficiencies.
The talk will outline the methodological approach, including clustering, optimization techniques such as Variable Neighborhood Search (VNS), and simulation-based evaluations. Furthermore, I will discuss the ongoing work on real-time data visualizations in Power BI, which are being used to enhance decision-making and capacity planning at the center. The seminar will conclude with an overview of future steps, including the implementation of the template scheduling, once the current variability reduction efforts and other planned improvements are completed.
This presentation will be of interest to students and faculty involved in operations research, healthcare optimization, and data-driven decision-making.
Wednesday 3 December, 12:30-13:30.
Speaker: Stijn Vansteelandt
Title: Assumption-Lean (Causal) Modeling
Abstract: Traditional inference in (semi-)parametric models, such as generalized linear models, assumes that models are correctly specified and pre-determined. However, this approach is increasingly inadequate because models are often adaptively selected based on the data, introducing unacknowledged uncertainty. Furthermore, since models rarely represent a true underlying mechanism, standard inference is prone to bias from model misspecification; this is especially a concern in causal modeling, where even small degrees of misspecification in the range of the observed data can give rise to large biases. Recent advances in debiased machine learning and targeted learning have addressed these issues by reducing reliance on correct model specification. However, their model-free nature can limit their applicability and the insight they can deliver in complex settings.
Assumption-lean modeling rethinks the trade-off between model correctness, parsimony, and interpretability. It begins with data-adaptive outcome predictions, which are then projected onto specific model parameters. This projection is designed to ensure that the parameters remain interpretable or meaningful, even under model misspecification. By incorporating debiased machine learning techniques, assumption-lean modeling minimizes bias, maximizes interpretability, and provides valid confidence intervals that account for both model uncertainty and model misspecification.
In this talk, I will introduce the core principles of assumption-lean modeling, focusing on its application to generalized linear models for accessibility. The presentation will draw on the work of Vansteelandt and Dukes (2022) that was presented in a discussion paper for the Journal of the Royal Statistical Society: Series B. I will also showcase recent advancements aimed at balancing efficiency with interpretability.
References:
Vansteelandt, S. (2021). Statistical Modelling in the Age of Data Science. Observational Studies, 7(1), 217-228.
Vansteelandt, S and Dukes, O. (2022) Assumption-lean inference for generalised linear model parameters (with discussion). Journal of the Royal Statistical Society: Series B (Statistical Methodology), 84(3), 657– 685.
Vansteelandt, S., Dukes, O., Van Lancker, K., & Martinussen, T. (2024). Assumption-lean Cox regression. Journal of the American Statistical Association, 119(545), 475-484.
workshop in Econometrics
Thursday 4 December 2025,15:00-17:00
Room: TS53, C-1.03
Speakers: Antonio Cosma (University of Bergamo), Geert Dhaene (KU Leuven), Benjamin Holcblat (University of Luxembourg), Gautam Tripathi (University of Luxembourg)
15:00-15:30: Antonio Cosma (University of Bergamo)
Title: Missing endogenous variables in conditional moment restriction models (with Andreï V. Kostyrka and Gautam Tripathi)
Abstract:
We consider the estimation of finite dimensional parameters identified via a system of conditional moment equalities when at least one of the endogenous variables (outcomes and/or explanatory variables) is missing at random for some individuals in the sample. We derive the semiparametric efficiency bound for estimating the parameters and use it to demonstrate that efficiency gains occur only if there exists at least one endogenous variable that is nonmissing, i.e., observed for all individuals in the sample. We show how to construct “doubly robust” estimators and propose an estimator that achieves the efficiency bound. A simulation study reveals that our estimator works well in medium-sized samples for point estimation as well as for inference. To see what insights our estimator can deliver in empirical applications with very large sample sizes, we revisit the female labor supply model of Angrist and Evans (1998) and show that if there is even medium missingness in female labor income (the outcome variable), then having more than 200,000 observations is not enough for a researcher using inverse propensity score weighted GMM to find a statistically significant negative effect of having a 3rd child (the endogenous explanatory variable) on labor income. In contrast, our semiparametrically efficient estimator can deliver point estimates of this effect that are comparable to the GMM estimates as well as being statistically significant.
15:30-16:00: Geert Dhaene (KU Leuven)
Title: Approximate functional differencing estimation of average effects in panel models (with Jad Beyhum, Cavit Pakel, and Martin Weidner)
Abstract:
Average effects in nonlinear panel models with unobserved heterogeneity (e.g., fixed effects) are often not point-identified when T, the number of time periods, is finite. Yet the identified set may be small and shrink rapidly to the true average effect as T increases. In such cases, when T is only moderately large, one may settle for an almost consistent, easy-to-compute point estimator. We show that this is possible via an approximate version of the functional differencing approach of Bonhomme (Econometrica 2012). Approximate functional differencing can be viewed as an iterative bias correction scheme, applied to an initially biased estimator and implemented with a finite or infinite number of iterations. The key ingredient at each iteration is to replace, in the exact bias expression, the unknown heterogeneity distribution with the posterior estimate thereof, with the posterior based on a possibly misspecified prior. In models for discrete outcome variables (e.g., in panel logit or probit models with fixed effects), the implementation requires only elementary matrix computations. We show that, under suitable regularity conditions, infinitely-iterated approximate functional differencing yields average-effect estimates whose bias shrinks exponentially fast in T.
16:00-16:30: Benjamin Holcblat (University of Luxembourg)
Title: Generalized ESP estimator (with Ali Atabaigi and Fallaw Sowell)
Abstract:
Several studies have documented the instability of moment-based estimators such as the generalized method of moments (GMM) estimators, and the generalized empirical likelihood (GEL) estimators. We introduce a novel class of moment-based estimators. We show they correspond to GEL estimators shrunk toward parameter values with lower implied estimated variance, so they are more stable. We call them generalized empirical saddlepoint (GESP) estimators because they are based on generalizations of the ESP approximation. We prove existence, consistency and asymptotic normality of GESP estimators, and derive test statistics.
16:30-17:00: Gautam Tripathi (University of Luxembourg)
Title: Estimating parameters and marginal effects in nonlinear panel data models
Abstract:
We present some new developments in the estimation of parameters and marginal effects in nonlinear panel data models with fixed effects.
Wednesday 10 December, 12:30-13:30.
Room TS53 A1.22
Speaker: Heiko Röglin (University of Bonn)
Title: A Tour through Connected Clustering Problems
Abstract: Clustering problems arise in various application domains and have been studied extensively both in theory and in practice. In many models, data points are provided in a metric space, and a number k is specified. The objective is to partition the data points into k clusters to optimize a specific objective function, such as k-center, k-median, or k-means. Clustering problems in applications are, however, often more complex due to additional constraints the clustering needs to satisfy. We focus on connected clustering problems, where an additional undirected graph on the data points is given and each cluster needs to induce a connected subgraph of this connectivity graph. Such problems naturally arise in various applications, including geodesy, cartography, and community detection. In this presentation, we will provide an overview of the existing results for the connected k-center and the connected k-median problem. Given that these problems are NP-hard, we will primarily present approximation algorithms for various versions of these problems.
Wednesday 17 December, 12:30-13:30.
Room TS53 A1.22
Speaker: Jannis Kurtz (University of Amsterdam)
Title: Towards Explainable (Integer) Optimization
Abstract: In recent years, there has been a rising demand for transparent and explainable AI models. Although significant progress has been made in providing explanations for machine learning (ML) models, this topic has not received the same attention in the Operations Research (OR) community. However, algorithmic decisions in OR are made by complex algorithms which cannot be considered to be explainable or transparent as we will argue in this talk. To tackle this issue we present two promising concepts to provide explanations for (integer) optimization problems. In the first part we introduce the concept of counterfactual explanations and show how it can be used to calculate explanations for linear integer optimization problems. We show that the resulting problem is Sigma_2^p-hard but can be solved in reasonable time for certain special cases. In the second part we present a model-agnostic method called CLEMO which can provide explanations for any type of optimization algorithms (exact or heuristic). The idea is to approximate the input-output relationship of an optimization algorithm by an interpretable ML model.
Schedule Overview:
- 10/09: Marc Schröder (Maastricht University)
- 17/09: Garth Tarr (The University of Sydney)
- 24/09: Tom Demeulemeester (Maastricht University)
- 1/10: Daniele Girolimetto (University of Padova)
- 8/10: Dries Vermeulen (Maastricht University)
- 15/10: Frits Spieksma (TU Eindhoven)
- 22/10: Ties Bos (QE), Jip de Kok (MUMC+), Frank van Rosmalen (MUMC+
- 29/10: Shahrezad Fahmy (Maastricht University)
- 5/11: Etienne Wijler (Vrije Universiteit Amsterdam)
- 12/11: Xuanzhu Jin (Maastricht University)
- 19/11: Yordi van Kruchten and Flip van Kasteren (MUMC+)
- 26/11: Philipp Ketz (Paris School of Economics)
- 3/12: Stijn Vansteelandt (Ghent University)
- 10/12: Heiko Röglin (University of Bonn)
- 17/12: Jannis Kurtz (UVA)
Autumn 2024
MLSE is a (mostly) bi-weekly seminar to foster cooperation between the Department of Microeconomics and Public Economics and the Department of Quantitative Economics. It aims to give researchers the opportunity to present their ongoing work and to facilitate cooperation
Tuesday 15 Oktober 2024 - 13.00-14.00 PM TS53 C.1-05
Speakers: Kristof Bosmans (joint work with Koen Decancq (Universitity of Antwerp), Erwin Ooghe (KU Leuven))
Title: Multidimensional welfare and inequality: we need to talk about efficiency
Abstract:
We strengthen the theory of multidimensional welfare and inequality measurement by explicitly acknowledging considerations of allocative efficiency. Our approach combines the axiomatic rigor of more recent studies (Tsui, 1995, 1999, Gajdos and Weymark, 2005, Weymark, 2006, Seth, 2013) with the utility-based foundations of earlier studies (Kolm, 1977, Atkinson and Bourguignon, 1982, 1987, Maasoumi, 1986). Our axioms sharply distinguish efficiency considerations, captured by the standard Pareto axiom, from inequality considerations, represented by a new transfer axiom that only considers commodity transfers neutral in their effect on efficiency. These axioms reappear, suitably adapted, in our characterizations of welfare, inequality, and efficiency criteria. The analysis yields new characterizations of a well-established class of welfare criteria, a “dissident” class of inequality criteria, and new efficiency criteria.
Tuesday 12 November 2024 - 13:00-14:00 PM - TS53 A0.24
Author: Jana Gieselmann (Duesseldorf Institute for Competition Economics)
Title: (Mis-)Matchmaker
Abstract:
On matching platforms, users (implicitly) pay for the platform's services, but the platform makes money as long as it does not match them. This paper analyzes the matching rule of a profit-maximizing monopoly platform when the incentives between users and the platform are misaligned --- search is costly for users, but the platform either commits to display advertisements or charges a search fee each period. I demonstrate that frequently studied matching rules, such as random matching, are strictly suboptimal for the platform. Instead, the platform strategically lowers match quality to prolong search time and increase revenue, leading to unnecessary delay and potentially inefficient matches. Finally, I provide two explanations for why platforms adopt business models with misaligned incentives: targeted advertising and the presence of overconfident users.
Jana Gieselmann (Duesseldorf Institute for Competition Economics) is on the job market this year and will present her work at the MLSE Seminar on (Mis-)Matchmaker. She is primarily interested in behavioral economics, industrial organization and microeconomic theory
You can find more information about Jana here: https://www.jana-gieselmann.com/
Tuesday 26 November 2024 - 13.00 -14.00 PM TS 53 A1.23
Author: Rastislav Rehák (joint with Maxim Senkov)
Title: Persuasive Pooling
Abstract:
We analyze a quadratic Bayesian persuasion model in which the sender and receiver have arbitrary misalignment regarding their bliss actions as functions of the state. Our focus is on the structure of the optimal signal, particularly how states are pooled in the supports of the induced posterior distributions. We apply our findings to settings where the sender and receiver share a common preference over the ordering of bliss actions but differ in the sensitivity of their responses to state changes. This framework captures scenarios such as an advisor and a CEO with aligned priorities but divergent risk attitudes.
Tuesday 10 December 2024 - 13:15-14:15, TS53 A1.23
Authors: Hannes Rusch (with Maximilian Schmitt, Gewei Cao and Thomas Meissner)
Title: On the Microeconomics of Exploitation
Abstract:
According to the WalkFree Foundation, 50 million people are currently victims of modern slavery, the ILO estimates that the annual profit from forced labor is $236 billion, and two of the UN's Sustainable Development Goals directly address the need to end exploitation. Surprisingly, despite the fact that exploitation has existed across time and space, the incentive structures and strategic logic of exploitative interactions remain poorly understood. Our research uses a simple principal-agent approach to examine exploitation from a microeconomic perspective. Comparing exploitative interactions to a ‘free labor’ benchmark, we show that exploitation is harmful for victims and society at large, while being beneficial for the exploiter. Contrary to the (sparse) existing literature, comparative statics with respect to victims’ outside options imply that victims with better alternatives will be coercived more. Correspondingly, the exploiter's profits decrease in victims’ outside options, making it more profitable to exploit poor victims. Against the backdrop of our model, finally, we examine some preliminary evidence drawn from accounts of trafficking victims in the United States and around the world.
Operations Research Seminars
Wednesday September 11 2024 - 13:00- 14.00 PM TS53 E3.09
Whiteboard talk by Leo Krull - UM. - "Minimizing Tardy Processing Time on a Single Machine in Near-Linear Time"
Abstract:
this work, we revisit the elementary scheduling problem $1||\sum p_j U_j$. The goal is to select, among $n$ jobs with processing times and due dates, a subset of jobs with maximum total processing time that can be scheduled in sequence without violating their due dates. This problem is NP-hard, but a classical algorithm by Lawler and Moore from the 60s solves this problem in pseudo-polynomial time $O(nP)$, where $P$ is the total processing time of all jobs. With the aim to develop best-possible pseudo-polynomial-time algorithms, a recent wave of results has improved Lawler and Moore's algorithm for $1||\sum p_j U_j$: First to time $\widetilde O(P^{7/4})$ [Bringmann, Fischer, Hermelin, Shabtay, Wellnitz; ICALP'20], then to time $\widetilde O(P^{5/3})$ [Klein, Polak, Rohwedder; SODA'23], and finally to time $\widetilde O(P^{7/5})$ [Schieber, Sitaraman; WADS'23]. It remained an exciting open question whether these works can be improved further.
In this work, we develop an algorithm in near-linear time $\widetilde O(P)$ for the $1||\sum p_j U_j$ problem. This running time not only significantly improves upon the previous results, but also matches conditional lower bounds based on the Strong Exponential Time Hypothesis or the Set Cover Hypothesis and is therefore likely optimal (up to subpolynomial factors). Our new algorithm also extends to the case of $m$ machines in time $\widetilde O(P^m)$. In contrast to the previous improvements, we take a different, more direct approach inspired by the recent reductions from Modular Subset Sum to dynamic string problems. We thereby arrive at a satisfyingly simple algorithm.
Joint work with Nick Fischer
Econometric Seminars
Thursday September 12 2024 - 16.00 -17.00 PM
Speaker: Enrico Wegner (UM) - “Transmission channel analysis in dynamic models”.
Abstract: We propose a framework for the analysis of transmission channels in a large class of dynamic models. To this end, we formulate our approach both using graph theory and potential outcomes, which we show to be equivalent. Our method, labelled Transmission Channel Analysis (TCA), allows for the decomposition of total effects captured by impulse response functions into the effects flowing along transmission channels, thereby providing a quantitative assessment of the strength of various transmission channels. We establish that this requires no additional identification assumptions beyond the identification of the structural shock whose effects the researcher wants to decompose. Additionally, we prove that impulse response functions are sufficient statistics for the computation of transmission effects. We demonstrate the empirical relevance of TCA for policy evaluation by decomposing the effects of policy shocks arising from a variety of popular macroeconomic models.
Link: https://doi.org/10.48550/arXiv.2405.18987
Wednesday 6 November 2024 - 11.00-12.00 PM TS53 room H0.06
Speaker: Lucas Harlaar (UM)
Title: Constrained Multivariate Unobserved Components Time Series Models for National Accounts Data
Abstract:
We develop a multivariate time series model for the complete National Accounts data set of a country. The dynamic features in the data are modeled by means of unobserved components that allow for stochastically evolving trend, cycle and seasonal effects. The components can be uniquely associated with a particular variable but can also impact multiple variables simultaneously. The model considers both expenditure and production variables and includes constraints for a synchronized and consistently-defined Gross Domestic Product variable. The model constraints are handled using restricted Kalman filtering and smoothing methods. Parameter estimation is based on exact maximum likelihood which we show is feasible despite the high-dimensional parameter vector. The proposed model-based framework can be used for synchronized nowcasting and forecasting of Gross Domestic Product, but also for its components on the expenditure and production sides. Given that the time series model includes seasonal components, we show that constraints in the model can be altered for the forecasting of seasonally adjusted variables in a model-consistent way. Finally, when the model includes a single cycle component that is shared by all variables in the National Accounts, we can naturally interpret it as a business cycle indicator. We provide evidence that the extracted business cycle is more accurate and more timely when compared to well-known filtering methods. We empirically illustrate our model-based methodology for National Accounts data from Germany, Italy and The Netherlands. We argue that our proposed analyses are of significance from both macroeconomic and statistical perspectives.
Tuesday 12 November 2024 - 12.00-13.00 PM TS53 room A0.23
Speaker: Dr. Edmondo Trentin, Dept. of Information Engineering and Mathematics, University of Siena
Dr. Trentin is visiting SBE also as a committee member for the PhD defense of Dewi Peerlings.
Title: Some Approaches to Density Estimation using Artificial Neural Networks
Abstract: The talk presents robust connectionist techniques for the empirical estimation of multivariate probability density functions (pdf) from unlabeled data samples (still an open yet crucial issue in pattern recognition and machine learning). Data may either be samples of random feature vectors or generalized random graphs. First, a soft-constrained unsupervised algorithm for training a (possibly deep) feed-forward neural net is discussed. A variant of the Metropolis--Hastings algorithm (exploiting the very probabilistic nature of the present deep network) is used to guarantee a model that satisfies numerically Kolmogorov's second axiom of probability. The approach overcomes the major limitations of the established statistical estimators. Graphical and quantitative experimental results show that the proposed technique can offer estimates that improve significantly over parametric and nonparametric approaches, regardless of (1) the complexity of the underlying pdf, (2) the dimensionality of the feature space, and (3) the amount of data available for training. Then, a hybrid machine (combining a graph neural network with a RBF-like network) is presented that can be trained via maximum-likelihood to estimate pdfs over structured (i.e, graphical) patterns.
Website: https://www3.diism.unisi.it/~trentin/HomePage.html
Wednesday 27 November 2024 - 15.00 -16.00 PM room A0.23
Speaker: Ivan Ricardo (UM)
Title: Reduced-Rank Matrix Autoregressive Models: A Medium N Approach
Abstract: Reduced-rank regressions are powerful tools used to identify co-movements within economic time series. However, this task becomes challenging when we observe matrix-valued time series, where each dimension may have a different co-movement structure. We propose reduced-rank regressions with a tensor structure for the coefficient matrix to provide new insights into co-movements within and between the dimensions of matrix-valued time series. Moreover, we relate the co-movement structures to two commonly used reduced-rank models, namely the serial correlation common feature and the index model. Two empirical applications involving U.S. states and economic indicators for the Eurozone and North American countries illustrate how our new tools identify co-movements.
Link to the paper: https://arxiv.org/abs/2407.07973
Thursday 5 December 2024 - 15.00 -16.00 PM TS 49A 0.008
Speaker: Luke Servat (UM)
Titel: Optimal Investment for Retirement with Intergenerational Benchmarking.
Abstract:
"The Law Future Pensions (Wet Toekomst Pensioenen), adopted in
2023, adds the Netherlands to the growing list of countries whose second pen-
sion pillar is changing to a defined contribution plan. Due to the cessation of
guarantees in this plan, pension funds have been given the novel task of finding
an optimal solidary strategy, that aims to prevent so-called ”fortunate” and
”unfortunate” generations. This paper investigates how differences in pensions
can be reduced by changing the investment strategy during the accumulation
phase. As the classic life-cycle does not have this solidarity feature, we adapt
the utility function such that a cohort not only looks at its own accumulated
wealth, but also compares this to the wealth of the preceding cohort. The op-
timal investment strategy that is found as a solution to this novel problem,
what we call the ”intergenerational” life-cycle, deviates significantly from the
classic life-cycle. We observe the ”intergenerational” strategy de-risks during
the periods in which consecutive cohorts are not both investing, but increases
exposure during periods in which they are both accumulating. Furthermore,
we observe that, due to the finite leverage and quasi-linearity, that our solution
bears a strong resemblance to the ”100 − age” rule, which is often used in prac-
tice in the Netherlands. Finally, we conduct multiple simulations which confirm
that, without any loss of the level of wealth, the pursued leveling effect of the
investment strategy is indeed realized."
Vectum Lectures
Tuesday 8 October 2024 - 19:00 - TS53 Room C-1.03.
Speaker: Prof Dr. Jan Christopher Kops.
Topic: (Mis)Interpreting the World
Abstract:
As humans, we do not perceive the world as it is. We interpret it, and even worse we often misinterpret it. There are many reasons why our interpretation can get distorted, ranging from logical mistakes to wrong convictions, measurement errors, language barriers, or even attempts to simplify reality. This research project takes a closer look at real-world instances of such (mis)interpretations as well as the mathematics behind it.