A Reassessment of Monetary Policy Surprises and High‑Frequency Identification
Interest-rate moves around FOMC announcements are the standard instrument for monetary policy shocks. The paper shows they are predictable from earlier data, and asks what happens to the estimated effects once that predictable part is removed.
BS2023_replication.ipynb
export_web.pyWhat we did
- Predictability
Table 1: the FOMC surprises regressed on six indicators known before each announcement, over the paper’s three samples.
- The instrument
The monthly surprise and its orthogonalized version, rebuilt from the authors’ announcement data and checked against their series.
- Proxy SVAR
Industrial production, CPI, the excess bond premium and the 2-year yield, 12 lags, 1973–2020, identified with each instrument: first-stage F statistics, 90% wild-bootstrap bands (Figure 3) and the three sample periods of Figure 2.
What we found
Once the surprises are purged of what was predictable, the effects of monetary policy are about four times larger, but the instrument becomes weak.
- −0.38 → −1.39%
- Trough of industrial production after a shock that raises the 2-year yield by 25bp, raw vs orthogonalized instrument (paper: −0.35 → −1.4%).
- 4×
- Roughly the same factor for prices and credit spreads: the CPI falls 4.6 times as much and the excess bond premium rises 5 times as much.
- 6.9 → 1.95
- First-stage F, raw vs orthogonalized (paper: 8.19 → 1.83). Below 10 the instrument counts as weak.
Table 1 matches the paper in every coefficient, t-statistic, R² and sample size, and the rebuilt instrument matches the authors’ to 10−4. The F statistic of the raw instrument is the one number we could not pin down: robust standard errors give 5.0–6.9, and only conventional ones come close to the paper (8.0). The conclusion is the same.
Not replicatedThe paper restores a strong instrument with surprises around the Fed Chair’s speeches, which are not public; Table 2 needs Greenbook and Blue Chip forecasts.