CDS Evaluation Model with Neural Networks

This paper provides a methodology for valuing credit default swaps (CDS). In these financial instruments a sequence of payments is promised in return for protection against the credit losses in the event of default. Given the widespread use of credit default swaps, one major concern is whether the credit risk has been priced accurately. Credit risk assessment of counterparty is an area of renewed interest due to the present financial crises. This article proposes a non parametric model for estimating pricing of the CDS, using learning networks, based on the structural approach pioneered by Merton [1] as regards the independent variables; he proposed a model for assessing the credit risk of a company by characterizing the company’s equity as a call option on its assets. The model that we are introducing turns out peculiar not only for the use of the neural network, but also for the use of the implied volatility of one-year options written on the shares of the analyzed companies, instead of historical volatility: this leads to a higher capability of getting the signals launched by the market about the future creditworthiness of the firm (historic volatility, being a medium value, brings in temporal lags in the evaluation). Besides, our analysis differs from the structural approach for the fact that it considers the 30-month mean-reverting historical series for CDS spreads, and this turns out to be one of the main advantages of our forward-looking model.