Rapid growth of Croatian household debt in period prior to the outbreak of the financial crises
relaxed the financial constraint of the households, allowing them to frontload some of their
consumption on expectations of rapidly growing incomes. However, at the same time it raised
concerns about the potential implications of household over-indebtedness on stability of the
financial system as economic outlook hugely deteriorated. The goal of this paper is to explore
implications of the rapid debt accumulation by the households for financial stability1
.
Standard approach to this topic observes different macro-drivers of sharp household lending
with special attention given to the EU integration process and real convergence. More
recently, a consensus has started to form in the literature that the level of systemic risk in the
financial sector depends on the actual distribution of debt (and assets) amongst the households
rather than the aggregate level of indebtedness, prompting strong reliance of the field on
micro data sources (Beck et. al., 2010 and World Bank, 2009). Such an approach ignores
macroeconomic risks stemming from build up of external imbalances, but it is nevertheless an
important extension of macro-prudential tools, such as early-warning indicators for sudden
stops in capital flows or financial crises, in case those risks materialize.

This paper expands the range of the traditional micro-data analysis techniques as it aims to
account for the changes in the distribution of household debt. First step in this direction is
identification of the determinants of household debt over the observed period. However,
instead of looking the main determinants of the amount owned by households at the mean of
the distribution (estimated by the OLS), which is standard approach in the debt determinants
literature, this paper employs a quantile regression analysis. The quantile regression (QR)
allows for identification of the effect of different households' demographic and
socioeconomic characteristics upon the total amount of debt across the whole distribution of
the indebted households.
Since identification of the debt determinants is preformed on the non-random sample of only
indebted households, the problem of the sample selection bias is likely to emerge. This
problem could arise because some groups of households may be subject to financial constraint
due to banks' lending policy or subjective perception of those policies. In order to control for
the non-random bias, a modification of a two-stage Heckman model for the quantile
regression proposed by Buchinsky (1998) was employed. According to this methodology
unknown form of the sample selection bias term can be approximated following two-step
procedure: first, the credit market participation selection parameter is estimated using
distribution free semiparametric least squares estimation of single index (Ichimura, 1993),
followed by a power series approximation of the bias term.