Creating ‘Optimally Complex’ Models for Forecasting
提出使用最小描述长度（MDL）等基于复杂度的建模准则，从数据中归纳提取模型，以应对金融市场建模的挑战，帮助研究者选择输入变量并定义模型。
Financial markets present perhaps the greatest challenge to data modeling. Attempts to model financial processes in fact persist only because of the rewards that accrue to the use of even a moderate amount of information. Recognition of the repeatable patterns in market behavior is facilitated by use of powerful modern statistical techniques. Foremost among these is inductive modeling. With inductive modeling, the model is extracted the data, rather than being imposed from above by an analyst. Definition of the best model is made possible by the use of a complexity-based modeling criterion such as Minimum Description Length. MDL evaluates a model's perceived accuracy in the context of its complexity (and, incidentally, does not break down in the face of chaotic problems). Whereas traditional statistics only finds the parameter values for a given model, MDL also allows one to discover which inputs to use and how to define them.