Comparison of Empirical Methods for the Reproduction of Global Manufacturing Companies’ Credit Ratings

The quantitative assessment of the credit quality of manufacturing companies is a task of great interest to researchers and practitioners. This is underpinned by the elevated credit risk of these companies stemming from rapid technological changes. However, few studies have addressed this issue specifically for manufacturing companies. This study aimed to fill this research gap by comparing the predictive power of various methods in reproducing manufacturing companies’ public credit ratings from available financial and non-financial data. The sample included 109 manufacturing companies from developed and emerging markets over the period 2005–2016. The analysis included three methods: ordered logistic regression (OLR) and two machine learning techniques, random forest and gradient boosting. The results showed that machine learning techniques outperformed OLR in terms of predictive power. In the best specification model, random forest had an accuracy of 50%, followed by gradient boosting (47%) and OLR (25%). We also tested two types of sampling in the training and test sets: random and time-dependent. The results showed that the models’ predictive power was greater with random sampling. The inclusion of macroeconomic variables did not improve the models’ predictive power due to the rating agencies’ preferred through-the-cycle rating approach. The study’s findings have implications for the development of manufacturing firms’ internal credit ratings. They can also be useful for researchers exploring the accuracy of empirical models in predicting industrial firms’ insolvency and creditworthiness.

Credit rating; Machine learning; Manufacturing companies; Rating agencies; Risk management

Public credit ratings (PCRs) are the most common measure of creditworthiness. These ratings are assigned by domestic and international rating agencies, such as Moody’s Investors Service, Fitch Ratings, and Standard & Poor’s Financial Services (Karminsky and Peresetsky, 2009). The ratings provide a consistent global framework for accurate assessment and comparison of companies’ and countries’ credit quality (Karminsky and Polozov, 2016). However, ratings are often assigned to large, diversified manufacturers mainly from developed markets. This is due to the high cost of the rating, the need to provide agencies with a large amount of information, and the reluctance of small companies to publish low ratings. Other limitations of PCRs are their long update intervals, as well as errors and inefficiencies of rating agencies which were found by various studies (Langohr and Langohr, 2008). The errors included maintenance ratings at the investment grade level for companies with high credit risks or slow response to the credit crisis. To address these issues, investors develop internal credit rating models (ICRs). A common approach is to reproduce the PCRs of unrated companies from available financial and non-financial data using empirical methods. ICRs have proven to be objective and low-cost. However, their predictive power varies significantly depending on the underlying models (Karminsky and Peresetsky, 2007). The task of comparing various models and selecting the optimal one has therefore become extremely important.

The aims of this study were: (1) to compare the predictive power of various empirical methods in reproducing Moody’s ratings specifically for manufacturing firms; and (2) to identify the optimal model in terms of data availability, forecast accuracy, and outcome interpretability. We also examined whether the addition of macroeconomic factors to the set of explanatory variables increases the models’ prediction accuracy. This study contributes to the literature in several ways. First, it focuses on manufacturing companies, which are not the focus of most studies. Second, it applies explanatory variables that match those used in Moody’s methodology. Third, it uses data spanning a long period (from 2005 to 2016). The study’s findings have implications for the development of manufacturing firms’ internal credit ratings. The results can be useful for researchers exploring the ability of empirical models to predict industrial firms’ insolvency and creditworthiness.

In this study, we compared the accuracy of OLR, RF, and GB in reproducing global manufacturing companies’ PCRs. RF and GB outperformed OLR by a factor of almost 2. Random sampling yielded higher predictive power than time-dependent sampling. The inclusion of macroeconomic variables did not improve the models’ predictive power. The predictive power of our models is consistent with the literature. We conclude that ML techniques can be effective in reproducing manufacturing companies’ PCRs. Future research should examine how the addition of non-financial metrics can improve models’ predictive power. Such metrics include indicators of market and operational performance, corporate governance and quality of management, and companies’ intellectual capital. Future studies should also determine and compare sets of explanatory factors for reproducing credit ratings for diverse non-financial industries, such as oil and gas, metal and mining, chemical, automotive, and fast-moving consumer goods companies. Determining the best sets of explanatory factors for each industry will help improve the models’ predictive power.

This research was supported by the Academic Excellence Project 5-100 proposed by Peter the Great St. Petersburg Polytechnic University. 

Abdi, H., Williams, L.J., 2010. Principal Component Analysis. Wiley Interdisciplinary Reviews: Computational Statistics, Volume 2(4), pp. 433–459

Altman, E., Haldeman, R., Narayanan, P., 1977. ZETA^TM Analysis. A New Model to Identify Bankruptcy Risk of Corporations. Journal of Banking & Finance, Volume 1(1), pp. 29–54

Amato, J., Furfine, C., 2004. Are Credit Ratings Procyclical?. Journal of Banking & Finance, Volume 28(11), pp. 2641–2677

Basel Committee on Banking Supervision, 2017. Basel III: Finalising Post-Crisis Reforms. Bank for International Settlements. Available Online at https://www.bis.org/bcbs/publ/d424.pdf, Accessed on July 20, 2020

Bhushan, S., Reddy, C., 2016. A Four-Level Linear Discriminant Analysis-based Service Selection in the Cloud Environment. International Journal of Technology, Volume 7(5), pp. 859–870

Cuny, ?., Dube, S., 2017. When Transparency Pays: The Moderating Effect of Disclosure Quality on Changes in the Cost of Debt. Hutchins Center, Working Paper No. 34

Chopra, A., Bhilare, P., 2018. Application of Ensemble Models in Credit Scoring Models. Business Perspective and Research, Volume 6(2), pp. 129–141

Demeshev, B., Tikhonova, A., 2014. Default Prediction for Russian Companies: Intersectoral Comparison. Economic Journal of Higher School of Economics, Volume 18(3), pp. 359–386

Fachrurrazi, H.S., Munirwansyah, H., 2017. The Subcontractor Selection Practice using ANN-Multilayer. International Journal of Technology, Volume 8(4), pp. 761–772

Hossin, M., Sulaiman, M.N., 2015. A Review on Evaluation Metrics for Data Classification Evaluations. International Journal of Data Mining and Knowledge Management Process, Volume 5(2), pp. 1–11

Huang, Z., Chen, H., Hsu, C.J., Chen, W.H., Wu, S., 2004. Credit Rating Analysis with Support Vector Machines and Neural Networks: A Market Comparative Study. Decision Support Systems, Volume 37(4), pp. 543–558

Langohr, H., Langohr, P., 2008. The Rating Agencies and Their Credit Ratings: What They Are, How They Work, and Why They are Relevant. USA: John Wiley & Sons, Inc.

Liao, Y., Loures, E., Deschamps, F., Ramos, L.F., 2017. Past, Present and Future of Industry 4.0 – A Systematic Literature Review and Research Agenda Proposal. International Journal of Production Research, Volume 55(12), pp. 3609–3639

Lee, Y.C., 2007. Application of Support Vector Machines to Corporate Credit Rating Prediction. Expert Systems with Applications, Volume 33(1), pp. 67–74

Kamstra, M., Kennedy, P., Suan, T., 2001. Combining Bond Rating Forecasts using Logit. Financial Review, Volume 36(2), pp. 75–96

Karminsky, A., 2011. Corporate Rating Models for Emerging Markets. Journal of Corporate Finance Research, Volume 5(3), pp. 19–29

Karminsky, A., Peresetsky, A., 2007. Rating Models of International Agencies. Applied Econometrics, Volume 1(5), pp. 3–19

Karminsky, A., Peresetsky, A., 2009. Ratings as Measure of Financial Risk: Evolution, Function and Usage. Journal of the New Economic Association, Volumes 1–2, pp. 86–104

Karminsky, A., Polozov, A., 2016. Handbook of Ratings: Approaches to Ratings in the Economy, Sports, and Society, Switzerland: Springer

Kumar, K., Bhattacharya, S., 2006. Artificial Neural Network vs Linear Discriminant Analysis in Credit Ratings Forecast: A Comparative Study of Prediction Performances. Review of Accounting and Finance, Volume 5(3), pp. 216–227

Moody’s Investors Service, 2020. Manufacturing Methodology. Available Online at https://www.moodys.com/research/Manufacturing-Methodology--PBC_1206079, Accessed on July 20, 2020

Saitoh, F., 2016. Predictive Modeling of Corporate Credit Ratings using a Semi-Supervised Random Forest Regression. In: 2016 IEEE International Conference on Industrial Engineering and Engineering Management (IEEM), Bali 2016, pp. 429–433

Senaviratna, N.A.M.R., Cooray T.M.J.A., 2019. Diagnosing Multicollinearity of Logistic Regression Model. Asian Journal of Probability and Statistics, Volume 5(2), pp. 1–9

Sermpinis, G., Tsoukas, S., Zhang, P., 2018. Modelling Market Implied Ratings using LASSO Variable Selection Techniques. Journal of Empirical Finance, Volume 48, pp. 19–35

Tsai, C.F., Chen, M.L., 2010. Credit Rating by Hybrid Machine Learning Techniques. Applied Soft Computing, Volume 10(2), pp. 374–380

Ye, Y., Liu, S., Li, J., 2008. A Multiclass Machine Learning Approach to Credit Rating Prediction. In: 2008 International Symposiums on Information Processing, Moscow 2008, pp. 57–61

World Bank Open Data, 2020. Available Online at https://data.worldbank.org/, Accessed on July 20, 2020