Articles (Departament d'Economia)
http://hdl.handle.net/10230/8584
2020-11-23T13:23:37ZUsing iterated local search for solving the flow-shop problem: parametrization, randomization and parallelization issues
http://hdl.handle.net/10230/45862
Using iterated local search for solving the flow-shop problem: parametrization, randomization and parallelization issues
Juan, Angel A.; Ramalhinho-Lourenço, Helena; Mateo, Manuel; Luo, Rachel; Castella, Quim
Iterated local search (ILS) is a powerful framework for developing efficient algorithms for the permutation flow‐shop problem (PFSP). These algorithms are relatively simple to implement and use very few parameters, which facilitates the associated fine‐tuning process. Therefore, they constitute an attractive solution for real‐life applications. In this paper, we discuss some parallelization, parametrization, and randomization issues related to ILS‐based algorithms for solving the PFSP. In particular, the following research questions are analyzed: (a) Is it possible to simplify even more the parameter setting in an ILS framework without affecting performance? (b) How do parallelized versions of these algorithms behave as we simultaneously vary the number of different runs and the computation time? (c) For a parallelized version of these algorithms, is it worthwhile to randomize the initial solution so that different starting points are considered? (d) Are these algorithms affected by the use of a “good‐quality” pseudorandom number generator? In this paper, we introduce the new ILS‐ESP (where ESP is efficient, simple, and parallelizable) algorithm that is specifically designed to take advantage of parallel computing, allowing us to obtain competitive results in “real time” for all tested instances. The ILS‐ESP also uses “natural” parameters, which simplifies the calibration process. An extensive set of computational experiments has been carried out in order to answer the aforementioned research questions.
2013-01-01T00:00:00ZMaximum likelihood estimation of the latent class model through model boundary decomposition
http://hdl.handle.net/10230/45558
Maximum likelihood estimation of the latent class model through model boundary decomposition
Allman, Elizabeth Spencer, 1965-; Baños Cervantes, Hector; Evans, Robin; Hoşten, Serkan; Kubjas, Kaie; Lemke, Daniel; Rhodes, John A. (John Anthony), 1960-; Zwiernik, Piotr
The Expectation-Maximization (EM) algorithm is routinely used for maximum likelihood estimation in latent class analysis. However, the EM algorithm comes with no global guarantees of reaching the global optimum. We study the geometry of the latent class model in order to understand the behavior of the maximum likelihood estimator. In particular, we characterize the boundary stratification of the binary latent class model with a binary hidden variable. For small models, such as for three binary observed variables, we show that this stratification allows exact computation of the maximum likelihood estimator. In this case we use simulations to study the maximum likelihood estimation attraction basins of the various strata and performance of the EM algorithm. Our theoretical study is complemented with a careful analysis of the EM fixed point ideal which provides an alternative method of studying the boundary stratification and maximizing the likelihood function. In particular, we compute the minimal primes of this ideal in the case of a binary latent class model with a binary or ternary hidden random variable.
2019-01-01T00:00:00ZReasoning about others' reasoning
http://hdl.handle.net/10230/45348
Reasoning about others' reasoning
Alaoui, Larbi; Janezic, Katharina A.; Penta, Antonio
Recent experiments suggest that level-k behavior is often driven by subjects' beliefs, rather than their binding cognitive bounds. But the extent to which this is true in general is not completely understood, mainly because disentangling ‘cognitive’ and ‘behavioral’ levels is challenging experimentally and theoretically. In this paper we provide a simple experimental design strategy (the ‘tutorial method’) to disentangle the two concepts purely based on subjects' choices. We also provide a ‘replacement method’ to assess whether the increased sophistication observed when stakes are higher is due to an increase in subjects' own understanding or to their beliefs over others' increased incentives to reason.
We find evidence that, in some of our treatments, the cognitive bound is indeed binding for a large fraction of subjects. Furthermore, a significant fraction of subjects do take into account others' incentives to reason. Our findings also suggest that, in general, level-k behavior should not be taken as driven either by cognitive limits alone or beliefs alone. Rather, there is an interaction between own cognitive bound and reasoning about the opponent's reasoning process. These findings provide support to more subtle implications of the EDR model (Alaoui and Penta, 2016a) than those which were previously tested, and show that the EDR framework is a useful tool for analyzing and understanding the complex interaction of cognitive abilities, incentives, and strategic reasoning.
From a broader methodological viewpoint, the tutorial and replacement methods have broader applicability, and can be used to study the beliefs-cognition dichotomy and higher order beliefs effects in non level-k settings as well.
2020-01-01T00:00:00ZCommon value experimentation
http://hdl.handle.net/10230/45344
Common value experimentation
Eeckhout, Jan; Weng, Xi
In many economic environments, agents often continue to learn about the same underlying state variable, even if they switch action. For example, a worker's ability revealed in one job or when unemployed is informative about her productivity in another job. We analyze a general setup of experimentation with common values, and show that in addition to the well-known conditions of value matching (level) and smooth pasting (first derivative), this implies that the second derivatives of the value function must be equal whenever the agent switches action. This condition holds generally whenever the stochastic process has continuous increments. The main appeal of our approach is its applicability, which is demonstrated with two applications featuring common value experimentation: strategic pricing, and job search with switching costs.
Includes supplementary material: online appendix for Common value experimentation, additional results
2015-01-01T00:00:00Z