Unconventional but Different After All? A Unified Series of Narrative Monetary Policy Shocks

The obstacle is that conventional and unconventional monetary policy shocks have been identified by different methods, so a measured difference in their effects cannot be cleanly attributed to the policy rather than to the identification. The authors put it as the lack of “a monetary policy shock series that spans both conventional and unconventional episodes and is constructed in a consistent way”, and note that without one “it is difficult to determine whether observed differences in the effects of unconventional policy reflect genuine differences in transmission or merely differences in identification.” Most of the conventional evidence is narrative, following Romer and Romer (2004), or high-frequency, following Kuttner (2001); the unconventional evidence is almost entirely high-frequency — Gürkaynak, Sack, and Swanson (2005) on forward guidance, Gertler and Karadi (2015), Swanson (2021) adding asset purchases, Inoue and Rossi (2021) using shifts in the whole term structure, and Jarociński (2024) using non-Gaussianity. Holding identification constant across regimes is what lets the paper run a like-for-like test rather than compare estimates across studies.

The authors regress the change in the Wu and Xia (2016) shadow federal funds rate at each FOMC meeting on the Federal Reserve’s own Greenbook forecasts of real GDP growth, inflation and unemployment, and treat the residual as the unconventional policy shock. The forecast terms enter at horizons from the previous quarter through two quarters ahead, in levels and in revisions from the previous meeting, alongside the lagged level of the shadow rate; unemployment enters only as the current-quarter forecast. This is the Romer and Romer (2004) construction with one substitution. The authors are explicit that the regression “is not a structural policy rule in the behavioral sense” but “a forecasting equation that partials out the systematic, information-driven component of policy changes by conditioning on the Fed’s own Greenbook/Tealbook forecasts”, so the residual is the unpredictable component of the stance whichever rate sits on the left-hand side.

The argument is that unconventional actions ultimately show up in the yield curve, and the shadow rate summarises the yield curve in a single implied short rate that behaves like the funds rate did before the bound. The authors note that Wu and Xia (2016) show their shadow rate “has similar dynamic correlations with standard macro-economic variables in the 2009–16 period as the federal funds rate had with the same variables in the earlier period”, and that because the funds rate was effectively constant from 2009 to 2016, shocks estimated over that window can be interpreted as resulting from unconventional policy. They list asset purchases, forward guidance and collateral-requirement changes as the actions the shadow rate synthesises. The paper also offers a validation from the results side: finding aggregate dynamics under the shadow-rate-based shocks comparable to conventional ones echoes Wu and Zhang (2019) and Sims and Wu (2020), who use the shadow rate directly rather than identified shocks.

On serial correlation and predictability the unconventional series passes where the conventional series does not, though the comparison depends on how many lags are used. With ten lags, the F-test for joint significance of own lags rejects for the conventional shocks (F = 2.92, p = 0.001) but not for the unconventional ones (F = 1.04, p = 0.425), and no unconventional autocorrelation exceeds the 95% bands. The authors flag a qualification themselves: under an AR(4) specification the conventional series would also fail to reject (F = 1.78, p = 0.131), “which is not aligned with the original conclusions in Miranda-Agrippino and Ricco”, so they present the extended-lag specifications as the more informative comparison. For predictability, Granger causality tests on industrial production, unemployment, the CPI, consumption and the S&P 500 return no significant p-values for the unconventional series, and regressing each shock on one lag of ten macrofinancial factors built by Miranda-Agrippino and Ricco (2021) rejects for the conventional shocks (F = 2.18, p = 0.019) but not the unconventional ones (F = 1.09, p = 0.385).

No — the correlations with the established measures are small and statistically indistinguishable from zero, which the authors read as their series being complementary rather than competing. Pairwise correlations with Swanson’s (2021, 2024) forward guidance and large-scale-asset-purchase factors, and with Jarociński’s (2024) Odyssean forward guidance, Delphic forward guidance and asset-purchase factors, run between −0.17 and 0.18, none significant. Correlations with financial shocks (the excess bond premium of Gilchrist and Zakrajšek 2012, the credit supply shock of Bassett et al. 2014), the four uncertainty measures of Baker, Bloom, and Davis (2016) and Jurado, Ludvigson, and Ng (2015), and the oil shocks of Känzig (2021) and Baumeister and Hamilton (2019) are likewise small. The authors note that the correlations with asset-purchase measures at least “move in the expected direction”, being positive though insignificant. They also treat the low correlation as only weak reassurance on two concerns they do not address directly — the Fed information effect and the Fed’s reaction to news — since Swanson’s and Jarociński’s factors are built to be robust to those, so agreement in the impulse responses is indirect evidence rather than a test.

Normalised to a 25 basis point fall in the shadow rate on impact, the effects are comparable in magnitude to conventional shocks but more persistent. The interest rate keeps falling slowly for about 10 months, reaching a trough of 41 basis points. Industrial production barely reacts on impact and builds up slowly, peaking at around 1% twelve months after the shock. Unemployment moves less cleanly — the authors call the response “more subdued and erratic” — falling by about 0.15 percentage points 24 months out. The CPI rises by 0.37% at its peak within the first twelve months and then falls back, becoming insignificant as the stimulus fades; the authors note there is no initial price puzzle in this specification. They report that an internal-instrument VAR ordering the shock first and a proxy VAR using it as an external instrument for the shadow rate give very similar magnitudes, with a less erratic unemployment path and a slightly more persistent rise in inflation.

No, and that is the finding: the peak responses are statistically indistinguishable across regimes. Testing peak-response equality variable by variable, the p-values are 0.43 and 0.56 for the interest rate, 0.09 and 0.20 for industrial production, 0.15 and 0.25 for unemployment, and 0.30 and 0.42 for the CPI, under the “simple break” and “full break” specifications respectively. The authors deliberately test peaks rather than whole impulse response paths for aggregates, on the reasoning that conventional and unconventional responses can differ in timing — the interest-rate effect is more persistent at the bound — so a test of equality at all horizons “can potentially reject the null even when the economic magnitudes are quantitatively similar.” They read the result as supporting the “irrelevance hypothesis” of Debortoli, Galí, and Gambetti (2020). Note that the industrial-production p-value of 0.09 under the simple break is a non-rejection at conventional levels but not a comfortable one.

The direction reverses: expansionary conventional policy compresses the wealth distribution and expansionary unconventional policy widens it. Using the Distributional Financial Accounts at quarterly frequency, and normalising both to an immediate 25 basis point decline, an expansionary conventional shock lowers the top 10% wealth share by about 0.3% at the trough while raising the middle 40% share by about 0.56%; an expansionary unconventional shock raises the top 10% share by a maximum of about 0.36% while the middle 40% share falls by about 0.7%. The conventional result is consistent with Coibion et al. (2017), who document that expansionary conventional shocks reduce income inequality. Joint Wald tests over the dynamic path reject equality across regimes for the stock-to-house price ratio (p = 0.0077 and 0.0303), the top 10% share (p = 0.0036 and 0.0004), the middle 40% share (p = 0.0002 and 0.0064) and the bottom 50% share (p = 0.006 and 0.0097); for the top 1% share the simple break rejects (p = 0.0039) but the full break does not at the 5% level (p = 0.0982). The paper’s conclusion characterises these as p < 0.001 in most cases, which is stronger than the reported table supports — most of the p-values sit between 0.003 and 0.03.

The authors attribute it to the relative movement of stock and house prices, summarised by the stock-to-house price ratio. Top-decile households hold a sizable fraction of their wealth in stocks while the typical middle-class portfolio is dominated by housing, so a policy that raises equities relative to housing widens the wealth distribution. In the estimates, a conventional rate cut produces a sustained decline in the relative price of stocks, with the ratio falling about 3.13% at the trough in a U-shape very similar to the top 10% wealth response, while an unconventional shock raises the ratio rapidly by over 3% with dynamics very similar to its own wealth-inequality response. The authors present this as the monetary-policy-conditional counterpart of Kuhn, Schularick, and Steins (2020), who show that the relative evolution of stock and house prices strongly influences US wealth inequality over the last seventy years. They contrast this with Lenza and Slacalek (2024), who argue the distributional effect of unconventional policy runs largely through employment.

It checks the asset-price channel directly by revaluing actual household portfolios, and reproduces the same pattern. Taking each household’s portfolio composition from the Survey of Consumer Finances and applying the estimated stock and house price responses at the four-quarter horizon, the authors mechanically revalue holdings for four wealth groups — bottom 10%, 10–50%, 50–90% and top 10% — and report the shift in wealth shares, the percentage change in net worth, the share of aggregate dollar gains, and mean gains split between equity and housing. Under an expansionary unconventional shock the top 10% capture most of the total revaluation gains, driven almost entirely by stock price increases; the 50–90% group benefits primarily through housing; and the bottom groups receive comparatively little from either channel. The authors describe households in the bottom 10% and bottom 50% as largely “asset excluded”, holding little to no equity and often no housing, which limits the direct revaluation channel for them. They are explicit about what the exercise omits: it does not account for return heterogeneity within groups or for portfolio rebalancing, “both of which could further amplify these effects.”

Pooling reintroduces the price puzzle that the regime-specific estimation avoids. The authors report that relative to the baseline zero-lower-bound estimates, both pooled-break specifications “tend to generate a more pronounced price puzzle”, which they describe as reintroducing issues present in the pre-bound Romer and Romer conventional series that the sample-specific estimation in the main text handles better. The two break specifications differ in how much is allowed to change: the “simple” break lets only the coefficient on the shock differ in the bound period, while the “full” break also lets the relationship between outcomes and the control vector differ. Both are reported throughout rather than one being chosen, and for the top 1% share and industrial production the two specifications give materially different p-values.

The unified series covers 1969–2008 for conventional policy and 2009–16 for unconventional policy, but the estimation windows are shorter. The conventional series is Romer and Romer (2004) re-estimated and extended to December 2008 building on Wieland and Yang’s (2020) corrections, supplemented with the most recently published Tealbook/Greenbook forecasts; the authors describe the result as the longest available narrative monetary policy shock series for the United States. For aggregate impulse responses the unconventional sample covers 2009–15, with pre-bound conventional results reported as a benchmark. The distributional analysis uses 1989–2008 for conventional and 2009–15 for unconventional shocks, converted to quarterly frequency because the Distributional Financial Accounts are quarterly; because of the smaller sample the authors include only one lag of the controls used in their aggregate specification rather than the four used there. The shock-correlation table rests on 84 monthly observations over 2009M1–2015M12.

They align with several existing estimates while adding a test none of those studies could run. The authors report agreement with Miranda-Agrippino and Ricco (2023), who use factors derived by Swanson (2021); note that Swanson (2024), which adds the Fed’s response to news, finds smaller and puzzling effects on output and inflation relative to theirs; and that Bundick and Smith (2020) and D’Amico and King (2023) likewise find more persistent effects from unconventional policy, with D’Amico and King also indicating larger effects than for conventional policy. On the distributional side, they note that Mangiante and Meichtry (2025) independently compare the two kinds of policy using an SVAR and Swanson’s (2021) factors and find similar results for wealth inequality. The authors state the value added as being able to test aggregate equivalence within a unified framework rather than comparing across studies using different identification strategies.

The distributional results rest on short quarterly samples, the identification does not directly address two known concerns, and the paper is a short communication rather than a full structural investigation. The unconventional distributional window is 2009–15 at quarterly frequency, a small number of observations, which is why the authors reduce the control lags. They do not directly address the Fed information effect or the Fed’s reaction to news, and assess these only indirectly through agreement with Swanson’s and Jarociński’s factors. The microsimulation is mechanical, holding portfolios fixed and ignoring return heterogeneity within groups. And the interpretation of the 2009–16 residuals as unconventional policy rests on the identifying assumption that the shadow rate captures the stance during a period when the funds rate was effectively constant. The paper’s own conclusion states the distributional Wald rejections more strongly than the reported table does.