Seminarios de Negocios
El propósito del seminario es convertirse en el lugar donde presentar nuevas investigaciones así como también ser un foro para aumentar el conocimiento mutuo entre los miembros del profesorado. Planeamos 1 hora de exposición, seguido por 30 minutos de preguntas y de una discusión más informal.Martes 23 de Diciembre - 10:00
- Joaquin Navajas – Institute of Cognitive Neuroscience, University College London, UK
ABSTRACT
Vivimos en un mundo en constante cambio, es decir, las decisiones que hoy nos resultan óptimas (e.g., dónde invertir nuestros recursos) mañana dejaran de serlo para ser reemplazadas por otras. Para sobrevivir en un ambiente tan incierto, debemos saber balancear dos tipos de estrategias conocidas en el campo de la inteligencia artificial como “explotación” (i.e., usar nuestro conocimiento sobre el mundo para obtener el mayor provecho posible) y “exploración” (i.e., buscar nuevas y mejores opciones). Afortunadamente, no estamos solos en el mundo y muchas de estas decisiones las podemos tomar en grupo, por ejemplo, a través de una votación simple y democrática. Esto nos permite, entre otras cosas, exprimir al máximo la “sabiduría de las masas”, es decir, el hecho de que las decisiones colectivas pueden ser (bajo ciertas circunstancias) mucho mejores que la mejor de las decisiones individuales. Pero, ¿cuáles son los límites de este efecto sinérgico? ¿Cuántas decisiones necesitamos combinar para que esto ocurra? ¿Qué condiciones del entorno hace que nos convenga tomar decisiones en grupo? ¿Qué tipo de decisiones contribuyen a la “sabiduría de las masas” y cuales derivan en "delirios populares"? En esta charla, voy a contarles un experimento con aprox. 1000 participantes que, combinado con simulaciones y modelado computacional de las decisiones, permite empezar a contestar algunas de estas preguntas.
Jueves 11 de Diciembre - 16:00
- Fernando Bernstein (Duke Univ.)
We consider a retailer with limited inventory of identically priced, substitutable products. The retailer faces a market with multiple segments of customers that are heterogeneous with respect to their product preferences. Customers arrive sequentially and the firm decides which subset of products to offer to each arriving customer depending on the customer’s preferences, the inventory levels, and the remaining time in the season. We show that it is optimal to limit the choice set of some customers (even when the products are in stock), reserving products with low inventory levels for future customers who may have a stronger preference for those products. In certain settings, we prove that it is optimal to follow a threshold policy under which a product is offered to a customer segment if its inventory level is higher than a threshold value. The thresholds are decreasing in time and increasing in the inventory levels of other products. We introduce two heuristics derived by approximating the future marginal expected revenue by the marginal value of a newsvendor function that captures the substitution dynamics between products. We test the impact of assortment customization using data from a fashion retailer. We find that the revenue impact of dynamic assortment customization can be significant, especially when customer heterogeneity is high and when the products’ inventory-to-demand ratios are asymmetric. Our findings suggest that assortment customization can be used as another lever for revenue maximization in addition to pricing.
- Marco Scarsini (Economics, LUISS, Roma)
Buying and Learning with Online Ratings
ABSTRACT
Consumer reviews and ratings of products and services have become ubiquitous on the internet. We analyze, given the sequential nature of reviews, the information content they communicate to future customers. In particular, we focus on two settings. A benchmark settings in which a customer observes all past reviews of consumers who purchased the product. In this setting, we show that social learning takes place in that customers end up learning the true quality of the product. In addition, the consumer surplus losses due to the initial lack of knowledge of the true underlying quality of the product are small in a sense that we will make precise. With this benchmark in hand, the main question we address in the present paper is what happens when customers do not have access to all past reviews but simply the average of all such reviews (or alternatively when these reviews are available but customers are not able to process all past reviews and base their purchasing decision only on the aggregate statistic corresponding to the mean of past reviews). In this setting, while fully Bayesian inference becomes intractable, we derive a simple inference that corrects for the selection bias due to observing only reviews of customers who purchase, and show that this rule leads to social learning with consumer surplus losses of the same order as in the full information setting.
- Lawrence Sherman, (Criminology, University of Cambridge)
- Flavia Cardoso (Marketing, UDESA)
ABSTRACT
Viernes 23 de mayo - 16.00
- (Universidad de Chile)
We consider a natural scheduling problem where we need to assign a set of jobs to machines in order to minimize the makespan, i.e., the time where the last job finishes. In our variant jobs may be split into parts, where each part can be (simultaneously) processed on different machines. Splitting jobs can help balancing the load of the machines, but this is not for free: each part requires a setup time which increases the processing requirement of the job. I will introduce the problem and present algorithms based on rounding and linear programming that yield solutions with guaranteed quality. Our main result is an algorithm whose solutions have makespan at 1 + φ times the optimum, where φ ≈ 1.618 is the golden ratio. This is joint work with J. R. Correa, A. Marchetti-Spaccamela, J. Matuschke, O. Svensson, L. Stougie and V. Verdugo.
Mario Bravo, "Reinforcement learning in games"
In this talk we will introduce the idea of reinforcement learning for repeated normal form games. This framework has been used to study how people learn the equilibria of repeated games where the available information is scarce. Time permitting, we will discuss the following result: Consider a 2-player normal-form game repeated over time. We introduce an adaptive learning procedure, where the players only observe their own realized payoff at each stage. We assume that agents do not know their own payoff function, and have no information on the other player. Furthermore, we assume that they have restrictions on their own actions such that, at each stage, their choice is limited to a subset of their action set. We prove that the empirical distributions of play converge to the set of Nash equilibria for zero-sum and potential games, and games where one player has two actions. This is joint work with Mathieu Faure.
- Andres Weintraub (Universidad de Chile)
"La Gestión de Operaciones Como Herramienta Estratégica en las Empresas"
ABSTRACT
En el marco de los Seminarios de la Escuela de Negocios se invitó al Prof. Andres Weintraub de la Universidad de Chile, quien habló sobre herramientas de modelamiento, algoritmos y sistemas computacionales que han sido utilizadas con éxito en diversas empresas y organizaciones en Chile. En esta charla se demostró proyectos concretos en áreas forestal, minera, transporte marítimo, programación del futbol profesional, y otras. Para estos casos se describió el problema, como se modelo, el método de solución, la implementación computacional y base de datos. A su vez se analizó el impacto logrado con la implementación de estos sistemas, que superan los 200 millones de dólares anuales.
- Wouter Dessein (Columbia Univ.)