Research

I use intraday asset price changes around 83 ECB Governing Council meetings between January 2015 and October 2025 to ask whether the transmission of monetary policy surprises to the German yield curve changed when the euro area moved from balance sheet expansion to normalisation. Decomposing each meeting into orthogonal target, path and balance-sheet surprises, I find that the transmission did not weaken uniformly. It rotated. The balance-sheet factor transmits with a stable coefficient near unity across both regimes. What changed is the long end of the curve: a hawkish path surprise lowered thirty-year Bund yields by 1.30 basis points per basis point before 2022 and has no detectable effect since, a shift of 1.35 basis points with a wild-bootstrap p value of 0.037. In parallel, target surprises acquired a slope effect they previously lacked, flattening the two-to-ten segment by 1.46 basis points per basis point against essentially zero before. The share of meetings on which yields and equities co-move, the signature of a central bank information shock, fell from 46 to 15 percent. In a four-country panel, target surprises transmit 2.40 basis points more strongly to periphery than to core ten-year yields, but that amplification is confined to the expansion regime and disappears after July 2022. Two caveats discipline the reading. Joint stability tests do not reject coefficient constancy in any single equation, and no individual interaction survives a family-wise correction across the twenty-one tests examined; the evidence for the rotation is a consistent pattern of point estimates across maturities, classifications and robustness variants rather than a decisive rejection. Taken at face value, the estimates imply that curve sensitivities calibrated on the quantitative easing period misstate the current reaction, and that the misstatement is concentrated in the long-duration positions for which it matters most.

The stock-bond correlation turned positive in 2022 and the balanced portfolio lost its diversification. I ask whether the correlation is the right measure of what was lost. Using 16,065 daily observations across 258 quarters of US data, I define the hedging benefit of a bond sleeve as the reduction in portfolio volatility relative to the equity sleeve alone, which depends on the correlation and on the ratio of the two sleeve volatilities and on nothing else. A Shapley decomposition attributes 99 percent of the 2015–19 to 2021–23 deterioration to the correlation channel: equity and bond volatility rose together and left their ratio nearly unchanged. The correlation focus is therefore justified for the average. It is not justified for the tail. Bonds returned +0.03 percent on the worst equity days of the post-2021 quarters against +0.33 percent before, but conditional correlations are mechanically attenuated by conditioning, and against a normality benchmark that removes this bias the post-2021 tail excess is −0.10, the most negative of any regime in the sample. The macro state that the literature uses to explain the unconditional correlation explains it (R² = 0.19) but not the tail (R² = 0.07), and in real time, with publication lags imposed, it is beaten decisively by last quarter’s realised correlation at a one-quarter horizon and only draws level at one year. Correlation-aware allocation improves the Sharpe ratio from 0.68 to 0.79, but the naive rule beats the macro model and the return difference is not distinguishable from zero.

A government that restricts civilian firearms to reduce crime and a government that disarms citizens ahead of coercion take the same observable action. I study what follows from that single fact. In a signalling model with a privately informed government, restriction cannot separate types over a wide region of the parameter space: the predatory type always has a mimicking incentive, and the informational content of firearms policy is therefore zero in equilibrium. Citizens respond to restriction by arming more, not less, which partially offsets the policy. The welfare-relevant object is the ex ante choice between constitutional entrenchment and political discretion. Entrenchment costs a known amount under a liberal government and insures against strategic disarmament under a predatory one, so it is optimal above a threshold prior π* that the calibration puts at 0.034 but that ranges from 0.017 to 0.19 across plausible parameters. The result is not that arms deter states: the model never requires civilians to win a fight. Two findings run against the motivating hypothesis. Credible universal disarmament remains first best. And the value of entrenchment is hump-shaped in institutional quality, negative in both consolidated democracies and failed states, because the probability that a constitution is needed and the probability that it binds move in opposite directions.