Atypical behavior of credit: Evidence from a monetary VAR

Credit booms are often costly. After a credit boom, recessions and financial crises tend to be more painful than otherwise (Taylor (2012)). Timely identification of credit cycles as well as understanding their causes is crucial for developing appropriate policy responses.

 The identification of credit boom episodes in the literature is typically based on threshold methodology (Gourinchas et al. (2001)). According to this methodology, credit variable (often, credit-to-GDP ratio) is decomposed into trend and cyclical component with a filter (e.g. Hodrick-Prescott or its versions), and then particularly large deviations from trend are considered as excessive credit and are therefore indications of a credit boom. Mendoza and Terrones (2008) show that this methodology may be sensitive to the smoothness of the trend and therefore deliver controversial results. Furthermore, credit-to-GDP ratio by construction signals a credit "boom" in situations when GDP is falling, whereas credit is relatively stable or is falling at a slower pace than GDP. These episodes do not constitute a credit boom and have to be deleted manually (DellAriccia et al. (2012)). Finally, threshold methodology is univariate and cannot account for endogenous interactions between credit, real economy, asset prices, and policy. Credit booms, however, are general equilibrium phenomena; therefore it appears desirable to detect them from multivariate systems rather than single time series (Borio and Lowe (2002)). To address these drawbacks, I propose an indicator to detect atypical behavior of credit from a multivariate system - a monetary vector autoregression (VAR).

I apply a different operational definition of credit booms seeing them as departures from fundamentals (captured by business cycle variables) rather than from their own trend. When credit growth, justified by the current state of the business cycle, is substantially lower than the observed credit growth, this may signal of an ongoing credit boom in the economy (and vice versa for a credit bust).

The analysis is conducted for the U.S., Euro Area, and Japan. The VAR is estimated with Bayesian techniques under Sims and Zha (1998) prior. The estimation approach is Bayesian rather than "frequentist", as Bayesian methodology allows to deal with overfitting more efficiently. The baseline model dataset includes monthly data on industrial production, CPI, short-term interest rate, stock price index, monetary and credit aggregates. I use revised and real-time data. Sample size differs across countries: 1959-2010 for the U.S., 1970-2009 for Japan and 1994 -2009 for Euro Area.

The proposed methodology inter alia detects credit boom episodes in the U.S. and Euro Area prior to the Great Recession 2007-2008 and the credit boom in 1986-1989 in Japan, which preceded the bubble burst in early 1990s. Furthermore, I detect a link between atypically low short-term interest rates in the U.S. and Euro Area in early 2000s and the build-up of excessive credit growth prior to the Great Recession. The results are robust to inclusion of other variables and prove useful also when real-time data is used.