• No results found

Specific country and system studies

2.5 Literature with a macro prudential and bank regulatory focus

2.5.2 Specific country and system studies

Specific country studies have a slightly longer history, pioneered by researchers in the US where large data sets have been available quite early. Seminal work in the US stems from Keeton

& Morris (1988) and again from Sinkey & Greenawalt (1991)11. This section will, however, not follow the literature chronologically through time but will rather concentrate on country studies of the recent past which analyse (1) similar time periods as in this thesis and (2) banking markets with a comparable institutional setting as the one in Australasia. The list of studies also includes one for Australia (Esho & Liaw, 2002) which is discussed at the end of the section.

The first two studies discussed are publications of the Central Bank & Financial Services Authority of Ireland (Kearns, 2004) and the Bank of England (Pain, 2003). This is followed by reviews of studies for the Austrian, Spanish and the Italian banking market. It concludes with some work by Keeton (1999) for the US market.

11 Sinkey & Greenawalt (1991) regress loan loss rates observed in 1987 with bank specific variables of the preceding three years

In his Irish study, Kearns (2004) uses provisioning panel data of a total of 14 Irish credit institutions which yields a total of 132 annual time series observations starting mostly from the early to mid 1990s. The author estimates a model with predominantly macro factors as

explanatory variables but with a fixed effect estimator to account for unobserved individual bank characteristics. He finds some evidence for Ireland that the level of loan losses, proxied by loan-loss provisions, rises when GDP growth declines but more significantly when unemployment rises. Kearns then applies the coefficient estimates to run stress tests for the Irish financial system and concludes that the simulated increase in provisioning would differ across lenders but that every institution could afford the increase in provisions out of a typical year’s profit.

Pain (2003) uses panel regression analysis to investigate factors that may lead to increases in loan-loss provisions for a sample of eleven major UK commercial and mortgage banks (1978-2000). He finds that there are indeed a number of macroeconomic variables that can provide information about banks’ provisioning requirements, in particular real GDP growth, real interest rates and lagged aggregate lending growth. Because data are sourced from information-rich bank annual accounts instead of from generic databases, the author is also able to shed light on bank specific factors affecting provisioning such as the composition of the lending portfolio, the rapid credit expansion to certain sectors (e.g. commercial property lending) or collateral. All in all, the Pain paper follows a methodology (type of data used, analysis), which is most comparable to the one used in this thesis.

Slightly earlier, a group of researchers associated with the Bank of Austria (Arpa, Giulini, Ittner, & Pauer, 2001) published some analysis of provisioning patterns of Austrian banks12. They confirm, again in line with global studies above, that Austrian banks increase risk provisions in times of falling real GDP growth rates (procyclical behaviour) and in times of rising bank operating income or operating results (income smoothing).

12 The authors also investigate macroeconomic factors that affect net interest, respectively operating incomes.

A study by Bank of Spain’s Fernández de Lis, Martínez, & Saurina (2000) reviews loan growth and provisioning in the Spanish market and at the same time presents the Central Bank of Spain’s approach to counteracting procyclical provisioning behaviour. In line with other banking supervisors (e.g. Banque de France, 2001), the Bank of Spain had at the time introduced a so-called statistical provisioning regime aimed at proper recognition of forward looking, i.e. ex ante credit risk. The results of the study provide support for the new regime in that lending in Spain was found to have been strongly procyclical while provisions had shown a similarly procyclical bias, being largely linked to the volume of contemporaneous problem assets. In line with the results of the global sample studies, the authors conclude that book profits have tended to

overstate true profits in periods of low non-performing loans and high credit growth (upturn) and understate them in periods of high problem loans and low credit growth (downturn).

Also for the Spanish banking market Salas & Saurina (2002) analyze the credit risk in two institutional regimes: the Spanish commercial and savings banks. In particular, they study the determinants of bank problem loans using panel data of both macroeconomic and bank specific variables in the period of 1985–1997. Drivers of problem loans are the GDP growth rate, levels of corporate and consumer indebtedness, rapid past credit or branch expansion, the portfolio composition, bank size, net interest margin, capital ratio, and market power. However, the authors find significant differences between commercial and savings banks, which confirm the relevance of the institutional form. Like this thesis, Salas & Saurina partially rely on original bank annual reports for their data.

The performance of Italian banks over a period of 1985 to 2002 is analysed in a paper by Quagliariello (2004). For a sample of 207 banks13, the author in particular investigates whether loan loss provisions and non-performing loans show a cyclical pattern. In line with research reviewed above, he finds that the flow of new bad debts and the provisions against loan losses

13 Bank accounting data are sourced from non-public reports to the Bank of Italy (for use in their supervisory statistics) for a period of 1985 to 2002.

tend to increase when economic conditions deteriorate. As an interesting twist, however, GDP growth turns out to be significant only when lagged by 1 and 2 years, implying that the cyclical impacts in Italy are not instantaneous, but delayed. Similar to Kearns (2004), the author then applies the sensitivity parameters found to stress test the impact of a recession. In line with results for the Irish market, he concludes that level of Italian banks’ earnings and capital buffers would be, on average, sufficient to absorb the effects of the shocks.

Determinants of loan losses have also been analysed by regulators in the US. In fact, studies on the US banking market, due to its size and regional differences, could well be considered as multi-country analysis. Keeton (1999) focuses on bank specific factors that help predict loan losses. He conducts his investigation with a sample of quarterly call reports filed by US commercial banks for the period 1982 to 1996. On balance, his data provide some support for the intuitive result that faster loan growth leads to higher loan losses. Specifically, US States experiencing unusually rapid loan growth over the period tended to experience unusually big increases in delinquencies several years later. Keeton nonetheless puts a question mark behind this seemingly clear-cut relationship since in his theory only increases in loan supply, associated with relaxation of underwriting standards, should lead to credit losses down the track.

Conversely, increases in demand for loans, leading to more stringent selection of credits, or a productivity shift should, in his view, not give rise to increased loss rates in the future.

Finally, there is one study for Australia. A 2002 APRA study (Esho & Liaw, 2002) investigates appropriate levels of risk weights on lending secured by residential mortgages. The sample includes up to 16 Australian banks for sample periods ranging from 1991 to 2001 (shorter period for some estimates) for which the authors in particular model the level of

impaired assets as a function of the relative proportion of loans in the various Basel I risk weight categories. They find no significant difference in risk associated with varying the proportion of assets held between the 20% and 50% risk weight buckets from which they conclude that the Basel I 50% risk weight on housing lending might be excessive.