In this study, we address three key goals. First, we investigate the existence of outliers in commodity futures price data. Second, using an innovative and robust unsupervised machine learning outlier identification algorithm (UMLA), we investigate intervention models to filter the outlier effects. Third, using the specified UMLA models, we assess the impact of outliers on risk metrics of commodities and the improvement of the inference capabilities of these models. Our results show the importance of investigating and controlling for potential outlier effects because of the impact on risk metrics. We illustrate how risk metrics based on raw data can lead to higher than expected actual losses. Our research demonstrates that it is crucial to include intervention parameters to address outlier impacts in order to obtain robust and coherent risk metrics from which more informed decisions are made in regards to risk and credit management, governance, and compliance activities.

在这项研究中，我们解决了三个关键目标。首先，我们调查商品期货价格数据中是否存在异常值。其次，我们使用创新且稳健的无监督机器学习异常值识别算法 (UMLA)，研究干预模型以过滤异常值影响。第三，使用指定的 UMLA 模型，我们评估异常值对商品风险指标的影响以及这些模型推理能力的改进。我们的结果表明，由于对风险指标的影响，调查和控制潜在异常值影响的重要性。我们说明了基于原始数据的风险指标如何导致高于预期的实际损失。