Corporate rating forecasting using Artificial Intelligence statistical techniques

  • Received January 22, 2019;
    Accepted May 21, 2019;
    Published June 24, 2019
  • Author(s)
  • DOI
    http://dx.doi.org/10.21511/imfi.16(2).2019.25
  • Article Info
    Volume 16 2019, Issue #2, pp. 295-312
  • TO CITE АНОТАЦІЯ
  • Cited by
    9 articles
  • 970 Views
  • 138 Downloads

Creative Commons License
This work is licensed under a Creative Commons Attribution 4.0 International License

Forecasting companies long-term financial health is provided by Credit Rating Agencies (CRA) such as S&P, Moody’s, Fitch and others. Estimates of rates are based on publicly available data, and on the so-called ‘qualitative information’. Nowadays, it is possible to produce quite precise forecasts for these ratings using economic and financial information that is available in financial databases, utilizing statistical models or, alternatively, Artificial Intelligence techniques. Several approaches, both cross section and dynamic are proposed, using different methods. Artificial Neural Networks (ANN) provide better results than multivariate statistical methods and are used to estimate ratings within all the range provided by the CRAs, obtaining more desegregated results than several proposed models available for intervals of ratings. Two large samples of companies ‘public data’ obtained from Bloomberg are used to obtain forecasts of S&P and Moody’s ratings directly from these data with high level of accuracy. This also permits to check the published rating’s reliability provided by different CRAs.

view full abstract hide full abstract
  • JEL Classification (Paper profile tab)
    C45, G17
  • References
    32
  • Tables
    17
  • Figures
    6
    • Figure 1. ANN estimated with sample I for S&P’s ratings
    • Figure 2. ROC curve for the S&P prediction model with sample I
    • Figure 3. Network to estimate Moody’s ratings (Sample I)
    • Figure 4. ROC chart for the Moody’s model
    • Figure 5. Network MLP (47, 13, 1) to estimate S&P rating
    • Figure 6. ROC curve
    • Table 1. Total default probability for each rating level
    • Table 2. S&P and Moody’s rating distributions in samples I and II
    • Table 3. Rating’s codes
    • Table 4. Distribution of rates in sample I
    • Table 5. Sample I description of the X variables
    • Table 6. S&P and Moody’s distributions
    • Table 7. Areas under the ROC curves for the S&P ratings prediction (sample I)
    • Table 8. Relative importance of the exogenous variables
    • Table 9. Areas under ROC curves
    • Table 10. Relative importance of the exogenous variables of the network
    • Table 11. Classification with a network in three classes (sample I, S&P).
    • Table 12. Classification with a network in three classes (sample I, S&P)
    • Table 13. Descriptive statistics for basic variables
    • Table 14. Fisher’s discriminant functions classification
    • Table 15. Area under the ROC curve for each rating
    • Table 16. Importance and normalized importance of each explanatory variable
    • Table 17. Importance and relative importance of the input variables