Model-based financial data interpretation

Analyzing the recent financial history of a firm is an important step in the annual audit cycle and in many other financial reasoning tasks. Auditors and other analysts performing this task have available to them a variety of mathematical models to fit to a set of historical and current period financial statement data. They also have expectations about the likelihood and magnitudes of errors in that current period financial statement. Given these models and expectations, they must find the combination of models and statement errors that is most consistent with the numbers in the current period statement. A crucial difficulty in this task is in keeping track of the various models being tried and coping with the multiplicity of consistent combinations. This is a hard and general problem, and an important one in any uncertain environments in which human analysts must juggle many incompatible possibilities to arrive at a consistent view of the world. The program discussed in this paper prototypes a solution mechanism with potential to deal with this class of problems. Novel aspects of the CROSBY program include the application of model-based diagnosis to a quantitative financial model