Page 419 - Special Topic Session (STS) - Volume 3

STS552 Natalia Nehrebecka
recoveries which yields to a strong bimodality. The mean is given by 37% and
the median by 24%, i.e., LGDs are highly skewed. Both properties of the
distribution may favour the application of quantile regression because most
standard methods do not adequately capture bimodality and skewness.
Furthermore, many LGDs are lower than 0 and higher than 1 due to
administrative, legal and liquidation expenses or financial penalties and high
collateral recoveries.
The regression model can be presented as follows (see Nehrebecka, 2019):

Recovery ratei,t = f (Debt Characteristics i, Bank Characteristics i,
Firm Characteristicsi,t-1, Macroeconomic Variablest-1)

The LGD coefficient - was obtained as a result of nonparametric method
of Quantile Regression and Bayesian Model Averaging. Two approaches are
presented in the paper: Point in Time and Through the Cycle. The historical
loan losses recorded by the National Bank of Poland were used for estimation.

3. Results
The empirical analysis was based on the individual data from different
sources (from the years 2007 to 2018), which are:
- Data on banking defaults are drawn from the Prudential Reporting
managed by Narodowy Bank Polski. Act of the Board of the Narodowy
Bank Polski no.53/2011 dated 22 September 2011 concerning the
procedure and detailed principles of handing over by banks to the
Narodowy Bank Polski data indispensable for monetary policy, for
periodical evaluation of monetary policy, evaluation of the financial
situation of banks and bank sector's risks. Large exposures – for a bank
that is a joint-stock company, state-run bank and a non-associated
cooperative bank – mean exposures towards one enterprise in excess
of 2,000,000 PLN.
- Data on insolvencies/bankruptcies come from a database managed by
The National Court Register, that is the national network of Business
Official Register.
- Financial statement data (source: AMADEUS, NOTORIA, BISNODE, F-
02).
- Data on external statistics of enterprises (source: Narodowy Bank
Polski).

408 | I S I W S C 2 0 1 9