This thesis aims to empirically measure biases in non-performing loan (NPL) data caused by differing country definitions. Building on the work of Barisitz (2011, 2013b), I identify that Italy has an unambiguous theoretical upwards bias and that the U.K. and U.S. have an unambiguous theoretical downward bias. This thesis adds to the literature by first trying to empirically measure the bias of differing NPL definitions. I attempt to capture these biases by dividing gross charge-offs by lagged NPLs and call this ratio the conversion ratio. I combine a legacy BankScope dataset (1989-2012) with a dataset from Orbis Bank Focus (2012-2019), totalling 35,739 observations for the conversion ratio. The results are in line with the expectations. First, the conversion ratio for Italy is lower than the conversion ratio for the U.K. and U.S. from 2006 to 2019. Second, the conversion ratios for France, Germany and Italy are around 70% lower than the conversion ratios for the U.K. and U.S. between 2012 and 2018, thereby providing evidence of an ongoing bias. This has important implications for the cross- country comparability of NPL data. Finally, this thesis shows that cross-country LLP timeliness models estimating country level timeliness are being impaired by biases in NLP data.

Additional Metadata
Keywords Non-performing loans, NLP, definitions, cross-country, bias, loan loss provision timeliness
Thesis Advisor Elfers, Dr. F.M. (Ferdinand), Yu, Dr. J. (Jaeyoon)
Persistent URL
Korpershoek, I.P.A.(Igor). (2020, April 28). An empirical investigation in NPL definitional biases in the EU & US between 1989 and 2019, especially regarding LLP timeliness models. Retrieved from