Panel dataset and replication package for "Smaller but Safer: How Quantitative Easing Reshapes Bank-Intermediated Financial Systems" — 33 economies, 2012–2020
Description
Empirical materials underlying "Smaller but Safer: How Quantitative Easing Reshapes Bank-Intermediated Financial Systems," a study of the structural, as distinct from the prudential, consequences of unconventional monetary policy for banking systems. PANEL. Balanced annual country-year panel of 33 advanced and emerging economies, 2012-2020: 297 observations, no duplicate country-year pairs. Coverage (ISO-3): AUS, AUT, BEL, BRA, CHL, CHN, COL, CZE, DEU, ESP, FIN, FRA, GRC, HUN, IDN, IND, ISR, ITA, JPN, MYS, NGA, NLD, NOR, PER, PHL, POL, PRT, RUS, SWE, THA, TUR, USA, ZAF. Two observations are missing on the bank Z-score and one on deposit intensity; all other series are complete. RAW VARIABLES. Country, ISO-3 code, economy type, year, bank Z-score, quantitative-easing shock (qe_shock, % of GDP), non-bank intermediation intensity, the product qe_x_nbfi, bank assets to GDP (bank_size), net interest margin, bank deposits to GDP (deposit_gdp), real GDP growth, inflation, capital adequacy (car). The non-bank and margin fields do not enter the reported specifications but are retained so that the risk-migration hypothesis reported as unsupported may be re-examined. CONSTRUCTED VARIABLES. The cleaning script derives the asset-to-deposit ratio (bank_size / deposit_gdp), a composite in which the common GDP denominator cancels and which is therefore insulated from the 2020 output collapse; its logarithm; the log Z-score; the policy variable as log(1 + qe_shock); mean-centred policy and capital terms with squares and interactions; a linear trend; a 2020 indicator; a within-country first lag of the policy variable; and a country-level classification of financial-system depth. Centring constants and the depth threshold are exported separately so that the quadratic vertex can be recovered in original log units. INTEGRITY. All transformation occurs on an ordinary data frame; the panel object is built last. The script halts unless it verifies 297 rows, zero duplicate country-year pairs, a balanced panel, and exactly 33 missing values in the lagged policy variable, one per country. No step is stochastic. VALUE. The denominator-free outcome design transfers to other questions vulnerable to common-denominator artefacts. Both components of the composite are retained separately, permitting the decomposition to be reconstructed or alternative composites substituted. The depth classification cuts across the advanced/emerging division and is reusable. CONSTRAINTS. T = 9 precludes credible dynamic panel estimation and weakens panel unit-root inference; an alternative-inference specification is included. The policy variable is a single scalar and does not distinguish purchase type, announcement from implementation, or maturity structure. SOURCES. Banking indicators from <source>; macroeconomic aggregates from <source>; capital adequacy from <source>; the policy shock constructed from <source>, as set out in the codebook.