ABSTRACT

Her Majesty's Revenue and Customs (HMRC) has the ambitious target of making tax digital for all its customers and collecting tax in a more efficient, effective and accurate manner for both the government and UK taxpayers. Self-assessment tax returns, the biggest key business event for HMRC, is also one of the most popular digital services with over 90% of the approximately 12 million taxpayers in self assessment filing their return online each year. The majority of returns are filed in January immediately prior to the self-assessment deadline (31st January), putting significant pressure not only on the self-assessment digital service but also on all other HMRC digital services. Hence, understanding and predicting demand for the system is vital to provide a robust and responsive service. We therefore developed mathematical models with Bayesian inference techniques to forecast volumes of Self-assessment (SA) returns submitted online during January, providing accurate hourly predictions of traffic on the digital system in the run up to the SA deadline. Because none of the models being considered is believed to be the true model, we use an ensemble modelling technique that combines forecasts from different models to develop a less risky demand forecast.

摘要

英国税务海关总署 (HMRC) 的雄心勃勃的目标是为所有客户实现税收数字化，并以更高效、更有效和更准确的方式为政府和英国纳税人征税。自我评估纳税申报是 HMRC 最大的关键业务活动，也是最受欢迎的数字服务之一，每年约有 1200 万自我评估纳税人中有超过 90% 的人在网上提交纳税申报表。大多数申报表是在自我评估截止日期（1 月 31 日）之前的 1 月份提交的，这不仅对自我评估数字服务，而且对所有其他 HMRC 数字服务都造成了巨大压力。因此，了解和预测对系统的需求对于提供稳健且响应迅速的服务至关重要。因此，我们使用贝叶斯推理技术开发了数学模型，以预测 1 月份在线提交的自我评估 (SA) 回报的数量，从而在 SA 截止日期前对数字系统上的流量提供准确的每小时预测。因为没有一个被考虑的模型被认为是真正的模型，所以我们使用集成建模技术，结合来自不同模型的预测来开发风险较小的需求预测。