UK Regional Forecasting using Mixed Frequency Big Data

拟议的研究将涉及在混合频率大数据时代改进英国区域经济即时预报和预测的模型开发。这些系统将通过提供季度即时预报和预测，最终协助政府和私营部门的经济学家执行政策决定。目前，英国13个地区的总增加值（GVA）每年只发布一次，而且延迟很长时间。因此，在很长一段时间内，区域决策者在没有准确了解当前全球增值是多少的情况下做出决策，更不用说在不久的将来会是多少了。然而，许多区域GVA的潜在预测因素被更频繁和及时地发布。这种频率不匹配提供了比目前做法更经常地更新即时预报或预报的可能性。拟议研究的主要影响将来自政策制定者希望得到我将提出的及时准确的预测。 我计划从一个相对较小的数据集开始，使用英国13个地区的GVA数据和整个英国的GVA（按季度发布，比地区GVA数据延迟少得多），但将扩展到包括其他预测因素。这些变量包括联合王国作为一个整体可用的变量（如工业生产、失业和金融变量）以及区域一级可用的变量（如劳动力市场变量和商业调查）。这意味着我将处理大量的变量，因此，将导致大数据问题。也就是说，即使只使用区域和英国的GVA数据，也将涉及一个包含14个变量的计量经济学模型。每个额外的英国变量将使模型中的变量数量增加1，每个额外的区域变量将增加13个变量。目标是使用涉及多达100个变量的预测模型。向量自回归（VAR）模型是用于宏观经济预测的黄金标准预测模型，从Banbura et al.（2010）的开创性论文开始，研究前沿已经向前发展，以处理涉及这种规模VAR的大型VAR。我计划使用这种大型VAR方法来制作区域预测和即时预报。 但区域预测涉及另一个问题，在大量的VAR文献中没有涉及。这就是数据具有混合频率的事实。也就是说，区域GVA按年度提供，联合王国GVA按季度提供，其他预测指标（如工业生产和许多商业调查）按月提供。有一个新的，前沿的，混合频率计量经济学的文献，我计划在这项工作中扩展。英国区域预测背景与现有文献在几个重要方面有所不同。首先，有许多较低的频率变量（区域GVA变量）比平常。其次，英国各地区的GVA加起来等于英国GVA，这一事实提供了一个额外的横截面限制，这应该有助于改善预测。我计划在我的研究中解决这些问题。

The proposed research will involve the development of models for improving regional UK economic nowcasting and forecasting in an era of mixed frequency Big Data. These will ultimately assist economists in government and the private sector in implementing policy decisions by providing quarterly nowcasts and forecasts. Currently Gross Value Added (GVA) for the 13 regions in the UK is released only once a year with a long delay. Thus, for long periods, regional policymakers are making decisions without an accurate knowledge of what current GVA is, much less what it will be in the near future. However, many potential predictors for regional GVA are released on a more frequent and timely basis. This frequency mis-match offers the potential for updating nowcasts or forecasts more regularly than is current practice. The main impact of the proposed research will arise from the desire by policymakers to have the timely and accurate forecasts that I will produce. I plan on beginning with a relatively small data set, working with GVA data for 13 UK regions and GVA for the UK as a whole (which is released on a quarterly basis and with much less delay than regional GVA data), but will expand to include other predictors. These include variables available for the UK as a whole (e.g. industrial production, unemployment and financial variables) and those available at the regional level (e.g. labour market variables and business surveys). This implies I will be working with a huge number of variables and, thus, will lead to Big Data issues. That is, even working with only regional and UK GVA data will involve an econometric model with 14 variables. Each extra UK variable will add one to the number of variables in the model and each extra regional variable will add 13 variables. The goal is to work with forecasting models involving up to 100 variables. Vector autoregressive (VAR) models are the gold standard forecasting model used for macroeconomic forecasting and, beginning with the pioneering paper of Banbura et al. (2010), the research frontier has moved forward to deal with large VARs involving VARs of this size. I plan on using such large VAR methods to produce regional forecasts and nowcasts. But regional forecasting involves another issue that is not addressed in the large VAR literature. This is the fact that the data is of mixed frequency. That is, regional GVA is available at an annual frequency, UK GVA at the quarterly frequency and other predictors (e.g. industrial production and many business surveys) are available at the monthly frequency. There is a new, cutting edge, literature on mixed frequency econometrics that I plan on extending in this work. The UK regional forecasting context differs from the existing literature in several important ways. First, there are many lower frequency variables (the regional GVA variables) than usual. Second, the fact that GVA for the UK regions adds up to UK GVA provides an extra cross-sectional restriction which should help improve forecasts. I plan on addressing these issues in my proposed research.