Mathematical Sciences: Conference on Forecasting, Predictionand Modeling in Statistics and Econometrics: Bayesian and Non-Bayesian Approaches

在所有科学和应用领域，都迫切需要一个好的程序来制定在解释、预测和决策中有用的定量模型。目前，模型制定更多的是一门艺术，而不是一门科学，对于如何解决这个问题有很多分歧。有些人强调简单、简约和奥卡姆剃刀，而另一些人则强调复杂、细节和现实主义。已经提出了许多方法，这些方法已经在实践中使用，即矩量法、使用自协方差矩阵的时间序列“识别”程序等，包括方法、结构计量经济学建模-时间序列分析程序、最大熵、量子统计推断等，此外，模型制定中必须考虑测量、描述、异常和丑陋事实的作用。在Geisser会议上提交的论文将详细考虑这些困难的问题，介绍和比较贝叶斯和非贝叶斯方法，并结合数据分析来说明应用各种方法的结果。将编写一份会议研究结果摘要。预测、预测和建模是统计学和计量经济学的核心理论和应用课题。时间序列预测方法被广泛应用于工业和政府部门，对未来的发展做出了有用的预测，但在解释方面却很少。另一方面，因果模型可以预测和解释未来的发展，以及各种政策如何影响这些发展。改进制定、实施和使用预测和因果模型的程序是这次会议的一个主要目标。在这方面，为实现上述目标，将提供贝叶斯和非贝叶斯方法的比较评估，使用许多应用程序来说明一般要点和评估替代方法。因此，会议将评价和应用制定、实施和使用预测和因果模型的新旧程序。

Good procedures for formulating quantitative models that are useful in explanation, prediction and decision-making are sorely needed in all sciences and applied areas. Currently, model formulation is more of an art than a science with much disagreement about how to approach the problem. Some emphasize simplicity, parsimony, and Ockham's razor while others emphasize the need for complexity, detail and realism. A number of approaches have been put forward that have been used in practice, namely, method of moments, time-series "identification" procedures using autocovariance matrices, etc., encompassing methods, structural econometric modeling-time series analysis procedures, maximum entropy, quantum statistical inference and so on. Further, the roles of measurement, description, unusual and ugly facts in model formulation have to be considered. Papers presented at the Geisser Conference will consider these difficult issues in detail, present and compare Bayesian and non-Bayesian approaches, and incorporate analyses of data to illustrate the results of applying various methods. A summary of the Conference's research findings will be prepared. Forecasting, prediction and modeling are central theoretical and applied topics in Statistics and Econometrics. Time series forecasting methods that are widely employed in industry and government yield useful forecasts of future developments but little in the way of explanation. On the other hand, causal models can provide predictions and explanations of future developments and how they may be influenced by various policies. Improving procedures for developing, implementing and using forecasting and causal models is a major objective of this Conference. In this connection, a comparative evaluation of Bayesian and non-Bayesian methods for achieving the above objective will be provided using many applications to illustrate general points and evaluate alternative approaches. Thus the Conference will provide evaluations and applications of old and new procedures for developing, implementing and using forecasting and causal models.