Bond Event Data

The project from which this data is taken involved "reading" several hundred billion words of text, to derive output for about 900 companies for the period 1996-2002. Accuracy was experimentally determined to be equivalent to that of skilled human readers.

The "periodical" reports a "bond_event" from a set of about a dozen such (e.g. planned_new_issuance, call_in_debt, worsened_outlook, upgrade, file_debt_shelf), at a "time" on a "date", with respect to the actions of the "source" (either a rating agency like Fitch, S&P, Moody's, or simply the company itself, in the case of new issuance of bonds, etc.).

The utility of this data is in the mathematical modelling of equity prices or the prices of other financial instruments. Time series such as this are blended with more traditional time series in a single model, whose accuracy is thereby enhanced.