People outside economics academia are always surprised to hear how long it takes to publish a paper. They often also think it is rather straightforward to write a paper – well, yes, but the tough part is to publish
it. This paper is certainly an extreme case. We started working on it almost 10 years ago, and had a first draft in June 2013 (background paper for a World Development Report). At some point I told my co-author Olivier De Jonghe: this paper will
be like a red Bordeaux, which will take a long time to mature – I did not imagine it would take that long. At some point, we invited a third co-author – Klaas Mulier – who gave us new impetus. We almost gave up last summer when we were
rejected in the third round. But here we are, the paper is now forthcoming in the Journal of Money, Credit and Banking.
The paper tries to make both a contribution to methodology and tests specific hypotheses. First, we construct measures of sectoral concentration along three dimensions: specialization (capturing high exposures),
differentiation (capturing deviation from peer banks), and financial sector exposure (capturing direct connectedness), using a method used in the mutual fund literature. Typically, researchers had to rely on credit registries for a specific country to compute
such measures. Our measures, on the other hand, we capture more than lending exposure of banks. Specifically, our measures also take into account banks' securities holdings and derivative positions through which banks might hedge excessive sectoral lending
exposures (or, alternatively, create such positions when there is no sectoral lending exposure). Furthermore, they also account for sectoral exposures at the liability side of banks' balance sheets (e.g. sectoral concentration in corporate deposits).
The underlying assumption of our methodology is that one can identify a bank's sectoral concentration choices from the covariation between its stock returns and
the returns on sectoral indices, and thus relies on market participants' information on bank choices. Going through earnings calls transcripts, for instance, we found indicate that analysts often ask questions in terms of actual exposure to and performance
across economic sectors but also in terms of hedging instruments used to hedge against sectoral concentration. We then define bank sectoral specialization as the percentage variation of the bank's stock returns that is incrementally explained by the sector-specific
indices over and above the variation explained by the other factors. Next, we define bank sectoral differentiation as the Euclidean distance between a bank's estimated sectoral exposures and the average sectoral exposures of all other banks in the same country
and year. Lastly, we define a bank's financial sector exposure as the estimated sensitivity of its stock returns to the returns on the financial sector index.
The advantage of our measures is that they can be computed for a large number of banks and countries without having to draw on proprietary information. On the downside, we capture only listed banks and inferring information on sectoral concentration
from banks' stock prices also hinges upon faith in weak form efficiency of the efficient market hypothesis.
We gauge the validity of our
market-based measures with hand-collect actual sectoral lending exposures from the notes to the annual statements for a small subsample covering the largest banks in our sample (with the help of several master students under the supervision of Olivier
almost ten years ago). We show a significant correlation between our market- and the accounting-based measures of sectoral specialisation.
the second part of the paper, we relate our three new measures to individual bank risk, proxied by the banks' distance to default, and to systemic bank risk, proxied by the banks' marginal expected shortfall, We find four relationships that exhibit a consistent
sign and significance over time and across countries and numerous robustness checks. First and second, bank specialization is negatively related to individual bank risk and to systemic bank risk, in line with theories advocating the benefits of specialization.
Third, differentiation from peer banks is positively associated with individual bank risk, consistent with theory that argues that differentiation may imply higher funding costs for banks, which --all else equal-- imply lower margins and higher risk. Fourth,
financial sector exposure is positively associated with systemic bank risk in line with theoretical models on financial contagion.
findings stress the importance of distinguishing between specialization on the bank-level and differentiation within the banking system and thus the distinction between micro- and macro-prudential regulation. Our approach is sufficiently flexible enabling
additional applications. First, it allows identifying which sectors contribute to aggregate specialization, as well as identifying whether this is due to an over- or underexposure. Likewise, our approach allows constructing semi-aggregate sectoral specialization
indices (cyclical versus defensive sectors, tradable versus non-tradable sectors). Such sector-specific specialization measures might be more useful for regulatory or early warning purposes than our aggregate measure of specialization as it allows assessing
which banks will be hit more when a specific shock hits the economy (e.g., an oil price shock, a pandemic). Second, our methodology can also be used for also be implemented to determine a bank's exposure to certain geographical areas, certain types of companies
(SMEs or large corporates), sovereign bond exposures or commodity prices. Finally, there is much more to explored. Why do banks specialize in certain sectors? Why do banks differentiate from or herd with other banks? Do regulatory variables influence the choice
of sectoral concentration? Certainly, lots more research to be done.