Modelling time series
Modelling time series
复制标题
建模时间序列
DOI：
--
复制
发表时间：
2011
期刊：
影响因子：
--
通讯作者：
P. Sweeting
中科院分区：
文献类型：
--
作者：
P. Sweeting
引言许多被衡量的风险是随着时间的推移而发展的。因此，必须正确模拟这些风险的发展方式。这意味着需要对时间序列分析有很好的理解。确定性建模有两大类模型：确定性模型和随机性模型。最基本的是，确定性建模涉及为每个预测变量商定一个单一假设。单一假设甚至可能仅限于数据历史，例如，过去20年中以前每月观测的平均值。采用确定性方法时，只能通过所用假设的边际或通过改变假设来增加谨慎性。第一阶段可能是考虑依次改变每一个基本假设，并注意其影响。这就是所谓的敏感性分析。它的帮助在于它给出了一组结果对每个基本因素变化的敏感性的概念，从而允许识别对特定风险的重大暴露。然而，在真实的世界中，变量很少单独变化。因此，需要采取一种考虑到所有假设变化的办法。这就引出了情景分析。这是确定性方法的扩展，其中使用不同的预先指定的假设来评估少量场景。所使用的情景可能是基于以前的情况，但重要的是，它们不限于过去的经验-一系列可能的未来被考虑在内。这是场景测试的关键优势：可以测试一系列“假设”场景，无论它们是否在过去发生过。然而，这并不意味着可以涵盖所有可能的场景-场景将始终受到建模者认为合理的限制。情景分析的另一个重要局限性是，它没有给出情景发生的可能性的指示。在考虑风险处理时，这一点很重要，将在风险的潜在影响及其可能性的背景下考虑成本。这些情景本身可能会以相当笼统的措辞给出，例如“高国内通胀率，高失业率”。这些情景需要转换为对相关变量的假设。重要的是，每一种设想都要保持内部一致，所依据的假设既要反映总体设想，也要相互反映。
Introduction Many risks that are measured develop over time. As such, it is important that the ways in which these risks develop are correctly modelled. This means that a good understanding of time series analysis is needed. Deterministic Modelling There are two broad types of model: deterministic and stochastic. At its most basic, deterministic modelling involves agreeing a single assumption for each variable for projection. The single assumption might even be limited to the data history, for example, the average of the previous monthly observations over the last twenty years. With deterministic approaches, prudence can be added only through margins in the assumptions used, or through changing the assumptions. A first stage might be to consider changing each underlying assumption in turn and noting the effect. This is known as sensitivity analysis. It is helpful in that it gives an idea of the sensitivity of a set of results to changes in each underlying factor, thus allowing significant exposures to particular risks to be recognised. However, variables rarely change individually in the real world. An approach that considers changes in all assumptions is therefore needed. This leads us to scenario analysis. This is an extension of the deterministic approach where a small number of scenarios are evaluated using different prespecified assumptions. The scenarios used might be based on previous situations, but it is important that they are not restricted to past experience – a range of possible futures is considered. This is the key advantage to scenario testing: a range of ‘what if’ scenarios can be tested, whether or not they have occurred in the past. However, this does not mean that all possible scenarios can be covered – the scenarios will always be limited by what is thought to be plausible by the modeller. Another important limitation of scenario analysis is that it gives no indication of how likely a scenario is to occur. This is important when risk treatments are being considered, the cost will be considered in the context of the potential impact of the risk but also its likelihood. The scenarios themselves might be given in quite general terms, such as ‘high domestic inflation, high unemployment’. These scenarios need to be converted into assumptions for the variables of interest. It is important that each scenario is internally consistent and that the underlying assumptions reflect both the overall scenario and each other.