Demystifying monetary policy surprises: Fed response to financial conditions and wait and see for new economic data

High-frequency monetary policy surprises (MPS) – interest-rate futures changes measured in narrow windows around FOMC announcements – are theoretically supposed to be exogenous, yet documented predictors from before the meeting partially and significantly predict them; Chen proposes that this happens because the Fed responds mainly to financial conditions rather than directly to economic data, and adopts a “wait-and-see” posture toward new data that markets fail to anticipate. The paper’s alternative to the two standard explanations in the literature – that predictability reflects the Fed’s private information, or that it reflects the Fed’s aggressive real-time response to incoming news – is that the Fed treats broad financial conditions (proxied by the OFR Financial Stress Index) as a summary statistic of the economic outlook, and that documented MPS predictors are predictive mainly because they are themselves already reflected in financial conditions before the meeting, not because the Fed possesses private information markets lack.

The paper’s model has the Fed setting the policy rate as a Taylor-type reaction to pre-announcement financial conditions and other variables, where those financial conditions are themselves a market-formed expectation of the economic outlook. Formally, the policy rate responds to pre-announcement financial conditions S_t^pre and other variables (i_t^post = beta_1 S_t^pre + beta_2 X_{2,t}^pre + epsilon_t^MP), while S_t^pre is modeled as a time-varying-weighted market expectation of underlying economic conditions. The pure monetary policy shock, epsilon_t^MP, is unobserved by markets before the meeting. Motivated by Fed Chair Jerome Powell’s May 3, 2023 statement that “we look at financial conditions as an important piece of information regarding the likely path of the economy,” the model treats financial conditions as a summary statistic the Fed uses rather than a separate policy target, and it generates three testable predictions that distinguish this financial-conditions channel from the private-information hypothesis and from a “Fed aggressively responds to news” hypothesis.

Yes: all six of the Bauer-Swanson (2023b) documented MPS predictors also strongly predict the pre-meeting level of financial stress, with the regression of FSI on those six predictors achieving an adjusted R-squared of 0.57 across 169 FOMC announcements (an alternative financial-stress measure, the excess bond premium, achieves 0.47 over 146 observations). For example, a positive nonfarm-payroll surprise is associated with a fall in financial stress (coefficient on FSI of -0.12, t = -3.19). Chen reads this pattern as evidence that the information in documented MPS predictors originates in public, market-observable financial conditions rather than in information privately held by the Fed.

Yes: the OFR Financial Stress Index (FSI) level on the day before a meeting predicts a more dovish (rate-cutting-leaning) surprise in raw, unorthogonalized MPS measures – for the combined MPS measure, FSI enters with coefficient -0.25 (t = -3.18) and Treasury-yield skewness (TR_SKEW) with coefficient 0.21 (t = 3.74), for a regression R-squared of 0.11; for the Nakamura-Steinsson (NS) measure, FSI is -0.31 (t = -6.06) and TR_SKEW is 0.18 (t = 2.91), R-squared 0.12. Consistent with the financial-conditions story, MPS measures that have already been orthogonalized to economic/financial information before the announcement – MPS_orth and JK_MP – show insignificant FSI coefficients, since their predictable component with respect to pre-meeting information has already been purged by construction.

No: the paper finds that the FSI’s own change on the announcement day itself is statistically insignificant for every MPS measure, while its change on the day after the announcement is strongly positive and significant (for MPS, 0.37, t = 4.75; for NS, 0.32, t = 3.33), which is the pattern implied by FSI reacting to the monetary surprise rather than a pre-existing risk premium in financial contracts driving the surprise. This test rules out the alternative explanation that time-varying risk premia embedded in the underlying financial contracts, rather than a genuine Fed response to financial conditions, generate the observed MPS predictability.

After controlling for financial conditions, a more positive real-activity surprise (Scotti 2016’s index, aggregating GDP, industrial production, employment, retail sales, and PMI news) predicts a more dovish MPS (coefficient -0.19, t = -3.35 for MPS; -0.17, t = -2.37 for NS) rather than a more hawkish one, which is inconsistent with a Fed that reacts quickly and directly to incoming data. Breaking the real-activity surprise out by how many days before the meeting it was released, the negative, significant coefficient persists for data released 1, 3, and 14 days before the meeting (and is significant with a mixed sign pattern at 7 days), but becomes statistically insignificant at 21 and 28 days out. Chen interprets this as showing the Fed does not react to real-activity news within roughly the last two weeks before a meeting, while data three or more weeks old has already been fully absorbed into financial conditions and financial markets, and so loses independent predictive power for the surprise.

Nearly all of it, by the paper’s accounting: purging the Bauer-Swanson (2023b) six documented predictors of their financial-conditions component sharply reduces their explanatory power for MPS – adjusted R-squared falls from 12.3% to 3.4% for MPS, from 12.0% to 0.6% for NS, from 7.7% to 0.0% for JK_Info, from 3.4% to 0.7% for Target, and from 7.3% to 0.0% for Path. A parallel exercise on the underlying Eurodollar futures (ED1-ED4) finds the raw Bauer-Swanson predictors explain 9.2%-11.3% of their variance, which drops below 4.0% once the financial-conditions component is removed; for scheduled meetings only, the nearest-maturity ED1 contract retains some residual predictability, which the paper attributes to the wait-and-see effect for very recent economic news rather than to a channel the financial-conditions story fully captures.

Yes: in a monthly proxy SVAR for the U.S. economy (estimated in levels with 12 lags over January 1995-December 2023, using industrial production, unemployment, CPI, a commodity price index, and the Wu-Xia (2016) shadow rate, with the instrument-based identification itself restricted to January 2000-December 2019), using the raw Nakamura-Steinsson (2018) surprise as the external instrument produces distorted impulse responses with output- and price-puzzle features, whereas using a version of the same surprise orthogonalized to financial conditions (FSI and Treasury skewness) yields impulse responses free of those puzzles – industrial production and CPI decline (rather than rise) and unemployment does not fall following a contractionary policy shock, in line with standard theory. An intermediate instrument, orthogonalized only to the Bauer-Swanson (2023b) predictor set rather than to financial conditions directly, produces intermediate results. The paper reads this as corroborating evidence, from a different identification strategy, that the financial-conditions component of raw high-frequency surprises is contaminating standard proxy-SVAR estimates.

The authors flag that the premise – that the Fed should respond to financial conditions – is a behavioral assumption that “might not represent the optimal strategy for the central bank” (footnote 2), and separately note that the persistent gap between the Fed’s actual reaction function and market understanding of it could reflect a deliberate communication strategy by the Fed rather than markets slowly learning the true reaction function (footnote 1). Samples are also constrained by data availability: the FSI-based tests run on at most 169 FOMC announcements from January 2000, the excess-bond-premium robustness check on only 146 observations from August 2002, and the real-activity-surprise tests on 139 observations from June 2003; the COVID period is excluded from the main sample, though the paper states the results hold in an unreported post-pandemic robustness check. The paper also does not compare its results against internal Fed forecasts (Greenbook data), unlike some related private-information studies, so its test of the private-information channel is indirect, run through the financial-conditions channel rather than a direct Fed-forecast comparison.