The Aggregate and Country-Specific Effectiveness of ECB Policy: Evidence from an External Instruments VAR Approach

The paper asks how ECB monetary policy transmits both to the euro area in aggregate and to individual member countries over 2002:M1-2016:M10, a sample spanning conventional policy and the unconventional-policy era (asset purchases, TLTROs) in which short rates were at the effective lower bound. In this environment the policy stance “is no longer appropriately summarized by the short-term policy rate,” and because the ECB uses several instruments simultaneously, standard recursive (Cholesky) or sign-restriction identification schemes are difficult to justify for variables – especially financial variables – that move quickly and jointly around announcements. This motivates identifying the policy shock instead with an external, high-frequency instrument.

The instrument is the daily change in the German 10-year government bond yield on ECB Governing Council meeting days, supplemented with three special-event days: the SMP announcements of 10 May 2010 and 7 August 2011, and Mario Draghi’s “whatever it takes” speech of 26 July 2012. The authors use a daily (not a tighter intraday) window because the ECB’s 13:45 policy statement is followed by a 14:30 press conference, so the market’s response to the news is spread across both, and a 30-minute window would miss part of it. German Bunds were chosen because they remained risk-free throughout the sample (avoiding a structural break the instrument would otherwise pick up from peripheral sovereign stress) and because the 10-year maturity is sensitive to the forward-guidance and unconventional-policy expectations that a short-rate-based instrument would miss. Identification rests on the efficient-markets assumption that yield changes on these specific days reflect policy news.

The instrument passes a standard weak-instrument check, with a first-stage F-statistic of 10.44 – above the Stock-Wright-Yogo threshold of 10 – indicating no weak-instrument problem. As an additional event-study validation (Table 1), the authors show a surprise tightening in the instrument appreciates the euro, raises the EURIBOR future (coefficient of 0.890) and raises the corporate bond spread (coefficient of 0.192), both statistically significant, which the authors interpret as confirming the instrument carries genuine monetary-policy information rather than noise.

In the baseline monthly four-variable VAR (industrial production, HICP, the corporate bond spread, and the Wu-Xia shadow short rate, with oil prices exogenous, 6 lags chosen by AIC/FPE), an expansionary shock normalized to a 25-basis-point drop in the shadow rate significantly raises industrial production and prices and significantly narrows the corporate bond spread, the last of which the authors read as external-finance-premium/credit-channel evidence (following Zhu 2013). These results are robust to using alternative shadow-rate measures (Krippner 2012; Lemke-Vladu 2017), the one-year German bond rate, and alternative lag lengths. By contrast, under a recursive Cholesky identification (ordering industrial production, prices, the shadow rate, then the spread), the corporate bond spread does not respond significantly – a difference the authors take as support for the external-instruments approach, while also noting it means the credit-channel finding is identification-dependent.

Adding one variable at a time to the baseline system shows the euro-area 10-year bond yield falling, the real effective exchange rate depreciating on impact (an exchange-rate channel), a house-price proxy (the rent component of HICP) rising, bank lending standards relaxing (a risk-taking channel), and credit demand rising in response to an expansionary shock. Unemployment falls but is statistically insignificant (cannot rule out a zero effect), and the Euro Stoxx 50 has the expected positive sign but is likewise insignificant over the full horizon, implying no significant stock-market transmission channel in this specification. Despite relaxed lending standards and higher credit demand, total loan volume to nonfinancial corporations actually falls in the post-2008 sample – a “credit puzzle” the authors describe as underscoring structural problems in euro-area credit markets, since “aggregate lending does not increase despite relaxed standards and higher credit demand.”

In the post-2008:M10-2016:M10 subsample (re-estimated with 3 lags), the four baseline responses (output, prices, spread, shadow rate) are qualitatively similar to the full-sample results, but the real-exchange-rate, unemployment, and stock-market responses are weaker or cross zero. The authors characterize this era – marked by sizable intra-euro-area bond spreads – as one in which “transmission through employment and the stock market is particularly impaired.”

Following Jarocinski and Karadi (2018), the authors decompose the announcement-day instrument via principal component analysis of standardized changes in German 2-, 3-, 5-, and 10-year yields, the Euro Stoxx 50, and the FTSE Euro 100 – the first two components together explain 92% of the variance. The component with all-positive loadings (PC1) is interpreted as a central-bank information shock, while the component with bond yields loading negatively and stocks loading positively (PC2) is interpreted as the pure policy shock. The pure policy shock reproduces the baseline aggregate results, while a negative information shock instead lowers prices, raises the corporate spread, raises unemployment, and lowers equity prices – a distinct pattern from a policy easing. The authors conclude the baseline findings are robust to purging information effects from the instrument.

The paper feeds the aggregate euro-area policy shock identified in the VAR into Jordà (2005) local projections estimated separately for ten member countries (Germany, France, Spain, Italy, Portugal, Greece, Ireland, the Netherlands, Finland, and Austria, together over 95% of euro-area GDP) plus a synthetic euro area, over 2002:M1-2016:M1, using Newey-West standard errors with a maximum lag of horizon+1. This approach assumes the ECB sets policy off the aggregate, so feeding the common shock into single-country equations abstracts from any country-to-ECB feedback. The results show substantial heterogeneity beneath an aggregate response that looked fairly uniform: unemployment falls significantly in core countries (Germany, France, the Netherlands) but not in periphery countries (Italy, Spain, Greece), consistent with the insignificant aggregate unemployment response found earlier; industrial production improves in all ten countries and is relatively homogeneous (Spain is the main outlier); consumer prices rise moderately and homogeneously, consistent with a single currency area; the real exchange rate depreciates on impact in all countries; stock prices rise for the euro area in the short run, but Spain, Greece, the Netherlands, Ireland, Portugal, and Austria deviate negatively after roughly ten months; and loans rise significantly only in Germany, Austria, and Greece, remaining insignificant elsewhere.

The authors conclude that prices and industrial production transmit fairly homogeneously across the euro area, while unemployment, stock prices, and bank lending are markedly heterogeneous, and that impaired transmission through the financial system (stock and credit markets) together with labor-market frictions explain the uneven country-level effectiveness of ECB policy. A single high-frequency instrument summarizing the ECB’s full toolkit works reasonably well on prices, output, the exchange rate, and the external finance premium, but is comparatively weak on bank lending and equity valuation – most visibly in periphery countries with stressed banking systems – so that a “one-size-fits-all” policy may fail to stimulate demand where national banking systems remain “blocked” by deleveraging and nonperforming loans, while core countries benefit disproportionately on unemployment. The authors explicitly flag several limitations: the daily (rather than intraday) instrument window may capture more than the pure announcement surprise, though this is partly addressed by the information-shock decomposition; the Bank Lending Survey and loan data begin only in 2003, shortening those samples; the counter-intuitive fall in loan volumes is flagged as inconsistent with theory yet unresolved (though consistent with Boeckx-Dossche-Peersman 2017); and, most importantly for the country-level results, the authors caution that their local-projection findings are “purely positive” and that “we should be careful not to overemphasize the normative implications.”