Date Submitted: 17 Nov 2017
Predicting Credit Rating Changes Conditional on Economic Strength
Julia Sawicki, Jun Zhou and Yonggan Zhao
This paper addresses a fundamental credit analysis question: what is the probability that a firm's rating will be upgraded, downgraded or unchanged for a specific rating cohort and state of the economy? We develop a new structural model for predicting credit rating changes using both financial accounting variables and macroeconomic indicators. Economic strength (contraction / expansion) is predicted with a set of systematic factors in a Markov regime-switching model. Measures of firm-specific business and financial risk are used in a multinomial logistic regression model to estimate transition probabilities (upgrade, stay or downgrade) conditional on the predicted state of the economy. Tests of statistical differences in probabilities indicate that the likelihood of upgrade and downgrade are asymmetric across regimes; general probability of a downgrade during a recession is nearly twice the expansion downgrade probability; up-grade probabilities are relatively stable.
In addition to developing a sophisticated predictive model, our work sheds light on a major (and poorly-understood) concern related to broad economic ratings' determinants: the pro-cyclical nature of credit ratings. Ratings that are particularly harsh (lenient) in weak (strong) economic conditions can exacerbate tightening (loosening) of credit, contributing to volatility and uncertainty in financial markets. This potential destabilizing impact of ratings changes draws considerable attention of regulators and policy makers. Our estimates of upgrade and downgrade probabilities in expansions and recessions provide insight at the policy level and a valuable analytical tool for investors.
Our model is an important contribution to the literature. Research devoted to identifying the determinants of ratings has focussed primarily on entity-specific measures of operating performance and financial condition, such as profitability and leverage. While link between credit risk and macro-economic is clear (and empirical evidence points to the potential to improve ratings predictions by modelling systematic conditions), empirical work has proved wanting. The standard approach using linear regression analysis is not appropriate for predicting changes in credit ratings.
Our model captures macroeconomic states with regime-switching model that offers two important advantages: results are not specification dependent and the model is forward-looking in that ratings transitions are predicted contingent on a prediction about the state of the economy. Forward-looking investment decisions are based on future credit quality, including estimates of how likely a firm's rating will be upgraded or downgraded under various economic states. The ordered logistic regression analysis with the economic strength conducted in this paper provides this information.
Tests of statistical differences indicate that the transition probabilities in contractions are different from those in expansions. The nature of these differences largely depends upon whether the issue is investment or non-investment grade. In general, it is much easier for junk bonds to be upgraded than investment grade bonds, regardless of economic state. This is not the case with downgrades. Notwithstanding CCC and below, investment and non-investment grades share similar downgrade probabilities in expansion (ranging from 2 to 4 percent). The probability of being downgraded jumps for non-investment grade issues in contractions to a range of 5 to 11 percent. Comparing transition probabilities between states and within rating we find that investment grade issues tend to exhibit similar upgrade probabilities in both expansion and contraction, whereas downgrade probabilities are somewhat higher during contractions than during expansions. Regarding non-investment grade issues, it seems that the 'dogs' get a boost during expansions with a much higher probability of being upgraded. The differences become very pronounced as the rating falls (for, example CCC contraction upgrade probability of 3.8 percent triples to 14 percent in an expansion). The chance of a downgrade during contractions is much higher than in expansions, almost doubling in many cases.
While these tests provide strong evidence consistent with a procyclical nature of rating changes, an important question remains. Does the evidence point to harsher (more lenient) ratings criteria applied during weak (strong) economic conditions? Or is it indicative of permanent, rather than temporary, changes in firm-level creditworthiness? To develop some insight into this question, average changes in firm-level credit risk measures are calculated conditional on economic strength. Preliminary analysis points towards evidence of distinct ratings criteria changes across regimes, consistent with procyclicality. This is an empirical question and the subject of further inquiry.
By the very nature of the subjectivity involved in analyst-driven, forward-looking ratings, identifying the determinants of credit ratings will never be an exact science. However, our understanding credit ratings, in particular the factors that prompt rating changes, is vital to the well-functioning of capital markets. Our results provide insight into the stability of credit ratings. CRA publications and empirical evidence point to ratings that seek a balance between some continuum of stability (immunity from cycles) and accuracy (changing in response to changes in both firm and macroeconomic conditions). Can we be more precise in our understanding of where they might lie on a through-the-cycle and point-in-time continuum? In particular, to what extent does rating' determination reflect changes reflect the state of the maro-economy? This paper is a step towards answers to these questions which are of critical importance to financial markets, including investors, portfolio managers, corporations and regulators.
Empirical modelling is particularly challenging because the parameters or functional forms are unstable, making it difficult to capture the complex nature of the interaction between the state of the economy and credit risk. Figlewski et al (2013) demonstrate that results are extremely sensitive to model specification, choice of macro-variables, averages versus lagged versus contemporaneous, and time period. Estimating various specifications and they find
that significance and macro-variable coefficient signs depend upon which variables are included in the model.