**4.2. 2002 to 2012 LCRs for class I and II banks**

**4.1. Description of banking data**

*4.1.1. Class I and II banks*

In this subsection, we describe the banking data pertaining to LCRs.

82 Dynamic Programming and Bayesian Inference, Concepts and Applications

the BCBS and Macro-Economic Assessment Group (MAG).

implicit government protection schemes for banks.

*4.1.2. Banking data restrictions*

*4.1.3. Banking data computations*

published on Monday, 7 January 2013.

[17]).

We investigate liquidity for Class I banks that hold more than US \$ 4 billion in Tier 1 capital (T1K) and are internationally active. Moreover, we consider Class II banks that violate one or both of these conditions (see, for instance, [9] and [17]). In reality, some Class II banks considered could have been classified as Class I if they were internationally active. Nevertheless, these banks make a large contribution to the total assets of Class II banks. Invariably, all Class I banks can also be classified as large in that their gross total assets (GTA) exceed US \$ 3 billion. Many of the banks in our study come from jurisdictions affiliated to

Our investigation includes 157 Class I and 234 Class II LIBOR-based banks from 38 countries. These banks (with the number of Class I and Class II banks in parenthesis for each jurisdiction, as well as \* and ' denoting BCBS and MAG members, respectively) are located in Argentina\* (1,3), Australia\*' (5,2), Austria (2,5), Belgium\* (1,2), Botswana (1,1), Brazil\*' (3,1), Canada\*' (7,3), China\*' (7,1), Czech Republic (4,3), Finland (0,14), France\*' (5,5), Germany\*' (7,24), Hong Kong SAR\* (1,8), Hungary (1,2), India\* (6,6), Indonesia\* (1,3), Ireland (3,1), Italy\*' (2,11), Japan\*' (14,5), Korea\*' (6,4), Luxembourg\* (0,1), Malta (0,3), Mexico\*' (1,8), Namibia (0,1), the Netherlands\*' (3,13), Norway (1,6), Poland (0,5), Portugal (3,3), Russia\* (0,3), Saudi Arabia\* (4,1), Singapore\* (5,0), South Africa\* (4,5), Spain\*' (2,4), Sweden\* (4,0), Switzerland\*' (3,5), Turkey\* (7,1), United Kingdom\*' (8,5) and United States\*' (35,66). In order to limit depositor losses, all 38 jurisdictions have explicit deposit insurance schemes or

In our study, we did not consider Central Banks, subsidiaries, banks with incomplete (inconsistent or non-continuous) information nor observations with negative HQLA, NCO, ASF, RSF or other values (see, for instance, [9] and [17]). Furthermore, we use non-permanent samples that do not suffer from survivorship bias to study cross sectional patterns. For our sample, bank failure data for the period 2002 to 2012 was obtained from deposit insurance schemes or implicit government protection schemes. For instance, for the US, such data was obtained from the Federal Deposit Insurance Corporation (see [9] and [17] for more details). We choose the period 2002-2012 because available EMERG global liquidity data does not allow us to reliably determine the LCR and NSFR prior to 2002 (see, for instance, [9] and

Estimating the LCR and NSFR using available EMERG public data proved to be a challenge. Firstly, the prescripts for these risk standards are sometimes ambiguous and subject to frequent regulatory amendment. For instance, the final rules relating to the LCR were only In this subsection, we provide 2002 to 2012 LCRs for Class I and II banks.

Table 1 shows that the LCR has been in a downward trend from 2002 through 2007. The average LCR had risen sharply from 2007 to 2009 and peaked in 2009. The general impression from Figure 1 is that the LCR time series is non-stationary.
