This study proposes a binomial logit model to estimate the probability of an event of bankruptcy for Internet companies, i.e. companies dependent on e-business. Sixty-one American companies were selected and divided into two samples: 25, which had filed for petition under the US Bankruptcy Code, between 1999 and 2001, and 36, which had not, for the same period. A training sample was randomly selected to determine the model variables and parameters and a holdout sample has confirmed its high accuracy by correctly classifying 95.1% of the total sample of 61 companies, based on annual financial statements as of one year before the bankruptcy's occurrence; and 88.1%, as of two years prior to the same event. Then, the e-Score Model reveals to be very significant by using only three variables, of 63 first analyzed: (i) INT/TL: Interest Expense (Income) to Total Liabilities; (ii) R&D/EMP: Research & Development Expenses to Number of Employees; e (iii) OCF/CL: Operating Cash Flow to Current Liabilities. While (i) and (iii) can be obtained from conventional financial statements, (ii) is innovative, indicating the company which invests in R&D on a per employee basis has a greater chance of survival. In contrast, companies presenting relatively high Net Interest Expenses and low Operating Cash Flow have less chance for survival.
bankruptcy prediction models; internet; e-business; logit regression; statistical modeling
