信用リスク記事の第一段落

Computing Counterparty Credit Risk via Machine Learning Algorithms
The computation of counterparty credit risk involves two fundamental loss possibilities. Firstly, there exists the risk that the counterparty will default on the settlement of a trade and cause the initial investment and any gains thereafter to be forfeited. Secondly, there exists the risk that the counterparty's credit rating will be downgraded, causing the prices of debt-related securities like bonds or loans to decrease in value. Both of these scenarios cause the institution to incur losses which would need to be covered either by capital reserves or by income from other lines of business in order for the firm to remain solvent. The Basel Committee on Banking Supervision has set capital requirements based on the riskiness of a trade which compliant financial institutions must adhere to. The capability to anticipate such credit rating transitions among current or future counterparties would enable better allocation of capital among business units, more useful value-at-risk (VaR) forecasts, and more profitable trading decisions. Such a capability might be provided by machine learning algorithms embedded in order management and risk profiling software. Such algorithms, trained to recognize a state vector of financial indicators for a firm (such as income, debt, and stock price moves) and interpret such a vector in terms of the likelihood of a credit transition, can provide numerical output indicating the probability of various transitions. This figure can be used to compute the valuation of securities whose price is affected by the credit worthiness of this counterparty.
Machine learning can be broadly classified into three methods: supervised learning, unsupervised learning, and reinforcement learning(footnote 1).

footnote:
1. Wikipedia also mentions transduction learning and "learning to learn", but these seem less emphasized in the literature the author has observed.
bio:
Patrick Toolis was extensive experience in the design, development, and deployment of software for financial trading systems, e-commerce, and operating systems. He was the manager in charge of all technology at JapanCross Securities, a joint venture between Nikko Citigroup and Instinet which was one of the first proprietary trading systems in Japan. He subsequently migrated the order management, matching engine, and reporting infrastructure to Instinet (now owned by Nomura Securities). Outside of the financial industry has held positions at Microsoft, IBM, and Hewlett-Packard. Mr. Toolis is fluent in Japanese.

Bibliography:
www.wikipedia.com Search "Machine Learning".