Scandal in the Portuguese banking sector – how a banking specific earnings management model predicted the fall of a family business group

We examined the external ability of the loan loss provision (LLP) model to detect extreme cases of earnings management (EM). According to the literature, the LLP model is the most useful in examining EM in banking institutions. We used it herein to explore the time-series behaviour of a fraudulent business group in the Portuguese banking sector between 1992 and 2013 − the Banco Espírito Santo Group (GBES). We conclude that GBES did not make discretionary use of LLP (i.e., DLLP) in the fraud period (2008 to 2013) when compared with the pre-fraud years (1992 to 2007). However, the level of LLP was significantly higher in the latter period; this was consistent with the procyclical nature of GBES's LLP. The results of a difference-in-difference approach did not reveal any significant differences between GBES's DLLP and non-fraudulent banks in the fraud period. Interestingly, the full bank sample (including GBES) provided evidence of the procyclical nature of LLP. Additional tests did not support the hypothesis of income smoothing via LLP, either amongst the bank sample as a whole or by GBES. The proven facts of the fraud indicated a significant undervaluation of loans and financial instruments and an underestimation of LLP. Thus, we expected to find negative DLLP in the fraud period and significantly different DLLP between the pre-fraud period and the fraud period itself. The DLLP of GBES should also have been significantly different from non-fraudulent banks in the fraud period. The LLP model proved ineffective in detecting GBES fraud and assessing the decisions of the bank's leader and his team, while the use of DLLP was effective. The evidence collected in our study will be of benefit to scholars and banking regulators.