• 主题
高级搜索  >
历史搜索 清空历史
首页> 外文会议>Annual conference on Computer Science;Conference on Computer Science >An expert system for tuning a macroeconometric model (abstract only)
【24h】

An expert system for tuning a macroeconometric model (abstract only)

机译：用于调整宏观经济计量模型的专家系统（仅摘要）

获取原文
页面导航
  • 摘要
  • 著录项
  • 相似文献
  • 相关主题

摘要

A "macroeconometric model" is a mathematical representation of a nation's economic behavior. Typically, such models are a collection of equations developed by using multiple regression to determine the coefficients of linear relationships hypothesized by economic theory. Macroeconometric models are used primarily for forecasting purposes. In practical application of the model for forecasting, the constant terms on the stochastic equations are adjusted by the forecaster on an ad hoc basis so as to compensate for perceived deficiencies in the model, e.g.,

inadequate theory (missing or wrong variables);

short-term exogenous disturbances;

data revisions.

The process of making such adjustments is sometimes referred to as "tuning" the model, and it is a good example of what often happens when experts and computer models interact, i.e., there is a need to combine qualitative heuristics and quantitative analysis. Tuning an econometric model is difficult and time-consuming for non-experts who would like to use macroeconometric models for preparing their own forecasts.

The current research being reported on here is an effort to develop an expert system that would guide users through the tuning process. The knowledge-base will contain rules by which expert macroeconometric model users adjust the constant terms on the stochastic equations. Many of the rules must, by their nature, be inexact. For example, If orders for producer durable goods increased strongly, then the impact on investment will be very positive. If building contracts increased very strongly, then the impact on investment will be positive. An effort is being made to treat "increased strongly", "very positive", etc., as linguistic variables in the fuzzy set theory sense. A major difficulty in doing this is the mutual dependence of the premises of the rules, e.g., "orders for producer durable goods increased strongly" and "building contracts increased very strongly" are not probabilistically independent. An alternate approach being studied is to translate the natural language variables into numerical estimates using statistical distributions of the corresponding quantitative variables. In this approach, Bayesian estimates would be utilized to deal with the problem of mutual dependence of the variables.
机译：

“宏观计量模型”是一个国家经济行为的数学表示。通常，此类模型是通过使用多元回归来确定经济学理论假设的线性关系的系数而开发的方程式的集合。宏观计量模型主要用于预测目的。在预测模型的实际应用中，随机方程的常数项由预测者临时调整，以补偿模型中的感知缺陷，例如，

理论不足（变量缺失或错误）;

短期外源性干扰;

数据修订。

进行这种调整的过程有时称为“调整”模型，它是专家和计算机模型交互时经常发生的一个很好的例子，即，需要将定性启发式方法和定量分析相结合。对于想使用宏观经济计量模型来准备自己的预测的非专家而言，调整经济计量模型既困难又耗时。

此处正在报告的当前研究是努力开发一种专家系统，该系统将指导用户完成调整过程。知识库将包含专家宏观经济计量模型用户用来调整随机方程上的常数项的规则。就其性质而言，许多规则必须不精确。例如，如果生产者耐用品的订单猛增，那么对投资的影响将是非常积极的。如果建筑合同增长非常强劲，那么对投资的影响将是积极的。在模糊集理论意义上，正在努力将“强烈增加”，“非常肯定”等作为语言变量。这样做的主要困难是规则前提的相互依赖性，例如，“生产者耐用品的订单大大增加”和“建筑合同大大增加”并不是概率独立的。正在研究的另一种方法是使用相应定量变量的统计分布将自然语言变量转换为数值估计。这种方法将利用贝叶斯估计来处理变量之间的相互依赖问题。

著录项

相似文献

  • 外文文献
  • 中文文献
  • 专利
获取原文

客服邮箱：kefu@zhangqiaokeyan.com

京公网安备：11010802029741号 ICP备案号：京ICP备15016152号-6 六维联合信息科技 (北京) 有限公司©版权所有

  • 客服微信

  • 服务号