Periodic Reporting for period 1 - PartialIO (Can less be more?: Semiparametric and partial identification in panel data discrete response models with an application to consumer demand)
Reporting period: 2021-09-01 to 2023-08-31
In many settings, individuals are observed making repeated discrete choices, for example, whether to work or not in a specific year, or which brand of cereals to buy on each shopping trip. Furthermore, it is often noted that these choices are intertemporally correlated.
Traditional panel data discrete response models often use distributional and functional form assumptions for identification and estimation. If misspecified, these models can lead to interpretation and inference problems. Developing models that are flexible is thus very important since they can be applied to many different settings and datasets, making clear what can be tested and what conclusions can be drawn from the analysis. Getting a better understanding of how individuals make decisions is particularly important for society since this affects the success and effectiveness of the economic and policy decisions of firms and governments.
PartialIO’s overall objectives were the development and examination of new and less restrictive dynamic panel data discrete response models. These models would be applicable to consumer demand, capturing for example the role of inertia, habits and lock-in in choices. The project developed a new model where individuals base their decision on whether they switch from the option they chose last period, without imposing assumptions on the unobserved individual heterogeneity. The identification power of this model was examined, and although the model did not provide a single solution for the parameters of interest, identification bounds were derived under different scenarios.
Traditional panel data discrete response models often use distributional and functional form assumptions for identification and estimation. If misspecified, these models can lead to interpretation and inference problems. Developing models that are flexible is thus very important since they can be applied to many different settings and datasets, making clear what can be tested and what conclusions can be drawn from the analysis. Getting a better understanding of how individuals make decisions is particularly important for society since this affects the success and effectiveness of the economic and policy decisions of firms and governments.
PartialIO’s overall objectives were the development and examination of new and less restrictive dynamic panel data discrete response models. These models would be applicable to consumer demand, capturing for example the role of inertia, habits and lock-in in choices. The project developed a new model where individuals base their decision on whether they switch from the option they chose last period, without imposing assumptions on the unobserved individual heterogeneity. The identification power of this model was examined, and although the model did not provide a single solution for the parameters of interest, identification bounds were derived under different scenarios.
PartialIO developed a new binary response dynamic panel data model with switching state dependence without imposing assumptions on the unobserved individual heterogeneity. Departing from the standard approach, where last period’s choice enters as an additional regressor, the project allowed for current period’s decision to depend on whether this period’s choice differs from last period’s choice. This form of correlation is suitable for modeling cases where individuals face some form of high “switching costs”. This new way of modelling contemporaneous effects in choices, lead to the model being simultaneously both incomplete and incoherent. These two properties impose additional challenges that were addressed by examining the identification power of the model under different approaches used in the literature. In all cases where identification bounds can be derived, identifying information comes from individuals who switch. Nevertheless, assumptions on what is observed in each approach differ. This highlights the importance of the assumptions imposed in evaluating individual decisions over time.
Dissemination and communication activities included several presentations at conferences and invited seminars, and the organization of one international conference and one international workshop. The project’s activities were promoted and a working paper with the project’s main results was made publicly available on the fellow’s personal webpage.
Dissemination and communication activities included several presentations at conferences and invited seminars, and the organization of one international conference and one international workshop. The project’s activities were promoted and a working paper with the project’s main results was made publicly available on the fellow’s personal webpage.
PartialIO’s approach goes beyond the state of art, since it studies a new way of modelling dynamics in individual decision processes, without imposing a-priory strong assumptions. This new approach contributes to the dynamic discrete response models’ literature, which has been growing in recent years. Furthermore, due to the nature of the model developed, the project contributes to the literature examining models that deliver no or multiple solutions. Such models often do not point-identify the parameters of interest. The project employs recent advanced methodological approaches to derive the identification bounds, contributing thus to the literature on partial identification.
The methodologies in the project can form the basis for a broad spectrum of theoretical and applied research in econometrics and microeconomics. Discrete response data is found in many applications including marketing, industrial organization, health and labour economics. Getting a better understanding of how individuals make decisions over time can have potential impact on the economic decision and policy making process of firms and governments.
The methodologies in the project can form the basis for a broad spectrum of theoretical and applied research in econometrics and microeconomics. Discrete response data is found in many applications including marketing, industrial organization, health and labour economics. Getting a better understanding of how individuals make decisions over time can have potential impact on the economic decision and policy making process of firms and governments.