03
From expected actions to global repricing.
Worcester College
Monetary Policy Surprises.
Expected actions, target news, path news, global spillovers, and synchronized impact prices.
The Lecture Moves from Actions to News
Identification begins by replacing the observed action with the expectation revision
01
Define the surprise
Policy news is the revision between what markets expected before an announcement and what they price after it.
After expectation − before expectation
02
Isolate current-target news
Kuttner uses federal funds futures to separate anticipated target actions from unexpected actions.
Current-target surprise
03
Recover future-path news
Gürkaynak, Sack, and Swanson show that one factor is not enough: statements also reshape expected future rates.
Target + path factors
Then Take the Surprise Across Borders
The object becomes richer, but it remains policy news
04
Measure global spillovers
Hausman and Wongswan trace U.S. target and path news into foreign equities, currencies, and interest rates.
Foreign-asset response
05
Align the traded asset and window
Kansoy uses U.S.-listed country ETFs to observe impact repricing while foreign cash markets are closed.
Same-window impact
06
Compare the estimands
Each paper answers a richer measurement question without abandoning the surprise principle.
Identification ladder
I
Why surprises, not changes?
Introduction
Policy News Revises Expectations.
The action matters only relative to what the market had already priced.
Information sets before and after
Target versus path news
High-frequency identification
The Policy Object Can Be a Target or a Path
Markets price an entire expected policy sequence—not only today’s rate
\[
g_t \equiv g\!\left(r_t^{*},r_{t+1}^{*},\ldots,r_{t+H}^{*}\right)
\]
Policy state
\(r_{t+h}^{*}\)
The central-bank target at date \(t+h\).
Horizon
\(H\)
How far ahead expected policy matters for the asset payoff.
Chosen object
\(g_t\)
Current target, future path, or an average expected policy rate.
Policy News Is a Filtration Revision
Compare the conditional expectation immediately after the announcement with the one immediately before
\[
S_t(g)\equiv
\mathbb E[g_t\mid\mathcal F_{t^+}]
-
\mathbb E[g_t\mid\mathcal F_{t^-}]
\]
Before
priced expectation
\(\mathcal F_{t^-}\)
\(\mathcal F_{t^-}\)
Announcement
new information arrives
After
revised expectation
\(\mathcal F_{t^+}\)
\(\mathcal F_{t^+}\)
Policy News Reorders Expectations
The new information changes the ordering of expected policy states and reshapes the path
State revision
The market already held a distribution
The announcement does not create expectations from nothing. It changes the probabilities and ordering of states that were already priced.
Event-study object
After minus before
Measure the same asset and the same policy object on both sides of the information arrival.
Interpretation: policy news is the revision, not the level of the post-announcement path.
The Surprise Is Unpredictable Before the Event
Unpredictable does not mean economically small—it means orthogonal to pre-announcement information
\[
\mathbb E\!\left[S_t(g)\mid\mathcal F_{t^-}\right]=0
\]
Definition
If the market could predict the revision using \(\mathcal F_{t^-}\), that predictable component should already be in the pre-announcement price.
Positive surprise
More hawkish than priced
For the chosen object \(g_t\), the post-announcement conditional expectation is higher.
Zero surprise
Confirmation
The announcement delivers the expectation already embedded in the price.
Only Surprises Move Impact Prices
Anticipated policy is already absorbed; the surprise branch creates a price revision
Anticipated branch
Action without news
When the decision equals the market’s conditional expectation, the announcement confirms the price already in place.
Unexpected branch
News without a predetermined sign
The revision can move the yield curve in either direction. Identification isolates the branch; the data determine the response.
Target and Path News Differ
A meeting can contain almost no target news and still deliver large path news
Current-target news
\[
S_t^{\mathrm{target}}
=
r_t^{*}
-
\mathbb E[r_t^{*}\mid\mathcal F_{t^-}]
\]
The unexpected part of today’s target decision.
Future-path news
\[
S_{t,h}^{\mathrm{path}}
=
\mathbb E[r_{t+h}^{*}\mid\mathcal F_{t^+}]
-
\mathbb E[r_{t+h}^{*}\mid\mathcal F_{t^-}],
\quad h>0
\]
The revision to a future policy rate.
One Announcement, Two Dimensions
The short lever moves the current target; the long lever reshapes the future path
Target factor
Current decision
Loads directly on the current-month policy-rate surprise.
Path factor
Expected future stance
Moves the expected rate path while remaining orthogonal to the current-target loading.
Why one factor fails: the announcement can move either lever without moving the other.
Policy-Path Lab: Which Horizon Moves?
Switch between current-target news, future-path news, and a combined announcement
Before
After
Announcement dimension
Current-target news shifts the near point and fades along the horizon.
Reading rule
Ask which horizon moved
A current-target surprise and a forward-guidance revision can occur together, separately, or with opposite signs.
Measure the Intrawindow Change
Measure the price immediately before and after the announcement
\[
\Delta_w x_t \equiv x_{t^+}-x_{t^-},
\qquad w=[t^-,t^+]
\]
Old expectation in price
t− · before
t+ · after
New expectation in price
Long Yields Price a Path Plus a Term Premium
A long yield can move even when the current target does not
\[
i_t^{(n)}
\approx
\frac{1}{n}\sum_{j=0}^{n-1}
\mathbb E[i_{t+j}^{(1)}\mid\mathcal F_t]
+
\operatorname{TP}_t^{(n)}
\]
Expected short-rate path
Average expected stance
Current and expected future one-period rates over the maturity.
Term premium
Compensation for risk
Duration, inflation, liquidity, and other risks embedded in the long bond.
A Yield Surprise Has Three Additive Pieces
The accounting identity is clear; identification requires policy-news measures
\[
\Delta_w i_t^{(n)}
\approx
\underbrace{\frac{1}{n}\Delta_w i_t^{(1)}}_{\text{current short rate}}
+
\underbrace{\frac{1}{n}\sum_{j=1}^{n-1}\Delta_w\mathbb E[i_{t+j}^{(1)}\mid\mathcal F_t]}_{\text{expected future path}}
+
\underbrace{\Delta_w\operatorname{TP}_t^{(n)}}_{\text{term premium}}
\]
Current rate
near-point revision
Future path
expected-rate revision
Term premium
risk compensation
Long yield
\(\Delta_w i_t^{(n)}\)
Path News Revises Expectations
Forward guidance changes conditional expectations of future short rates
\[
\Delta_w\mathbb E[i_{t+j}^{(1)}\mid\mathcal F_t]
=
\mathbb E[i_{t+j}^{(1)}\mid\mathcal F_{t^+}]
-
\mathbb E[i_{t+j}^{(1)}\mid\mathcal F_{t^-}]
\]
Predictable action
No revision
If already priced, the action creates no path news.
Statement language
Path revision
Guidance can move distant expected short rates immediately.
Risk channel
Premium revision
A term-premium change can move the same long yield.
Monthly Data Mix Policy and Macro News
The solution is a different regressor—not merely a different estimator
\[
\Delta y_t=\alpha+\beta\Delta r_t+\varepsilon_t
\]
| Failure | Why it enters | Consequence |
|---|---|---|
| Reverse causality | The bank reacts to financial conditions while markets react to the bank. | \(\operatorname{Cov}(\Delta r_t,\varepsilon_t)\neq0\) |
| Omitted macro news | Inflation, employment, output, and risk news move both the rate and the asset. | \(\beta\) absorbs non-policy information. |
| Anticipation | Expected policy is priced before the meeting. | The observed action is not the news. |
High Frequency Changes the Regressor
Use a policy-news vector extracted from futures prices inside a narrow event window
\[
\Delta_w y_t
=
\alpha+\theta^\prime S_t+\varepsilon_{t,w}
\]
\[
\mathbb E[\varepsilon_{t,w}\mid S_t]=0
\]
Window
Short enough to exclude unrelated news; long enough for markets to price the announcement.
Shock
\(S_t\) is an expectation revision, not a raw policy action.
Caution: a market surprise can still combine target news, path news, and central-bank information about the economy.
II
The target surprise from federal funds futures.
Kuttner (2001)
Separate Actions from News.
The expected component attenuates raw-action regressions; futures recover the unexpected current-target change.
Action = expected + surprise
Calendar-weighted futures
Yield responses reappear
Raw Target Changes Make Policy Look Weak
The puzzle is the regressor—not the disappearance of monetary-policy transmission
\[
\Delta R_{j,t}=\alpha_j+\beta_j\Delta r_t^{*}+\varepsilon_{j,t}
\]
Measurement problem
Raw changes mix old and new information
The post-1989 coefficient is weaker because target changes became more predictable, not because yields stopped responding to news.
| 100 bp action | 3m | 2y | 10y |
|---|---|---|---|
| Cook–Hahn, 1974–79 | 55 | — | ≈10 |
| Kuttner, 1989–2000 | 26.8 | 18.2 | 4.3 |
Cook and Hahn (1989); Kuttner (2001), Table 1.
Actions Split into Expected and Surprise
Only the unexpected component is new policy information at the announcement
\[
\Delta r_t^{*}
=
\underbrace{\Delta r_t^{e}}_{\text{anticipated}}
+
\underbrace{\Delta r_t^{u}}_{\text{surprise}}
\]
Observed action
\(\Delta r_t^*\)
Expected part
already priced
Surprise part
new information
Asset response
\(\Delta R_t=\beta\Delta r_t^u+\varepsilon_t\)
Expected Actions Attenuate the Estimate
Explore how the surprise share maps a true response into the raw-action estimate
\[
\operatorname*{plim}\widehat\beta_{\mathrm{raw}}
=
\beta
\frac{\operatorname{Var}(\Delta r_t^u)}
{\operatorname{Var}(\Delta r_t^u)+\operatorname{Var}(\Delta r_t^e)}
\]
80
True \(\beta\)
40
Raw estimate
50%
Attenuation
Numerator
True policy surprises
The variation that can create an announcement response.
Denominator
All target changes
Surprise variation plus predictable variation.
Predictability is good communication but bad measurement when expected actions are treated as shocks.
The Variance Ratio Is the Surprise Share
Actual policy can vary substantially while announcement surprises remain small
\[
\frac{\operatorname{Var}(\Delta r_t^u)}
{\operatorname{Var}(\Delta r_t^u)+\operatorname{Var}(\Delta r_t^e)}
=
\frac{\text{variation in surprises}}{\text{variation in actual target changes}}
\]
| Meeting | Actual \(\Delta r_t^*\) | Expected \(\Delta r_t^e\) | Surprise \(\Delta r_t^u\) |
|---|---|---|---|
| 1 | +25 | +25 | 0 |
| 2 | +25 | +20 | +5 |
| 3 | 0 | 0 | 0 |
| 4 | −50 | −45 | −5 |
Monthly Futures Price an Average
The target surprise affects only the remaining \(D-d\) calendar days
Contract object
Average effective federal funds rate
Let \(D\) be days in the month and \(d\) the days already fixed when the announcement arrives.
\[
\omega_{\mathrm{repriced}}=\frac{D-d}{D}
\]
Calendar exposure
The announcement changes the expected rate only for the red portion of the settlement month.
Before and After Share the Same Fixed Days
The historical part and the old target are common to both contract prices
\[
\begin{aligned}
f^0_{s,t^-}
&= \frac{d}{D}\bar r_{\le d}
+ \frac{D-d}{D}
\left(r^{old}+\mathbb E[\Delta r_t^*\mid\mathcal F_{t^-}]\right)
+\mu^0_{s,t^-},\\[0.45em]
f^0_{s,t^+}
&= \frac{d}{D}\bar r_{\le d}
+ \frac{D-d}{D}
\left(r^{old}+\Delta r_t^*\right)
+\mu^0_{s,t^+}.
\end{aligned}
\]
Before: expected action
The remaining-month block contains the action priced before the announcement.
After: realised action
The fixed days and old target are unchanged; the action is now known.
Differencing Removes Fixed Days
The old target and historical average cancel
\[
\Delta_w f^0_{s,t}
=
\frac{D-d}{D}
\left(
\Delta r_t^*
-
\mathbb E[\Delta r_t^*\mid\mathcal F_{t^-}]
\right)
+
\Delta_w\mu^0_{s,t}
\]
Cancels
Historical average
The \(d\) already-realised days are identical before and after.
Cancels
Old target
The pre-event rate is common to both contract valuations.
Remains
Diluted surprise
Plus any intrawindow futures-premium change.
Kuttner Rescales the Futures Movement
Undo the monthly-average attenuation to recover the unexpected current-target change
\[
\Delta_w\mu^0_{s,t}\approx0,
\qquad
\Delta r_t^u
\equiv
\Delta r_t^*-\mathbb E[\Delta r_t^*\mid\mathcal F_{t^-}]
\]
\[
\boxed{
\mathrm{mp1}_t
\equiv
\Delta r_t^u
=
\frac{D}{D-d}
\left(f^0_{s,t^+}-f^0_{s,t^-}\right)
}
\]
The Scale Factor Reverses Monthly Averaging
Move the announcement through a 30-day month and observe the gross-up
15
Repriced days
2.00
Scale \(D/(D-d)\)
10.0
mp1 from a 5 bp move
\[
\mathrm{mp1}
=
\frac{30}{30-d}\times 5\text{ bp}
\]
Original Kuttner window
Close to close
Previous-day futures close to announcement-day futures close.
Later implementations
Narrow intraday window
The economic object is unchanged; better data make the information window more local.
Near month-end, the same gross-up that recovers the signal also amplifies quote and premium noise.
Read the Implied Rate, Not the Futures Price
Reversing the price/rate sign reverses the economic interpretation
\[
f=100-\text{futures price}
\]
Futures price falls
\(\Delta F<0\)
Implied rate rises
\(\Delta f>0\)
mp1 > 0
hawkish surprise
Implementation check: perform all surprise algebra in implied-rate units, then translate back to the quoted futures price if needed.
Month-End Scaling Amplifies Noise
When \(D-d\) is small, the gross-up becomes fragile
Mechanical issue
\(\frac{D}{D-d}\)
Explodes as the remaining-day count approaches zero.
Noise channel
Bid–ask and premium moves
The same factor multiplies quote error and intrawindow futures-premium changes.
Operational response
Switch contracts
Use the next-month contract when too few current-month days remain exposed.
\[
\Delta r_t^e=\Delta r_t^*-\Delta r_t^u
\]
Estimate Expected and Surprise Effects
The regression asks whether yields respond to the action or to the news inside it
\[
\Delta R_{j,t}
=
\alpha_j
+
\beta_{e,j}\Delta r_t^e
+
\beta_{u,j}\Delta r_t^u
+
\varepsilon_{j,t}
\]
Expected coefficient
\(\beta_{e,j}\)
Response to a policy action already priced before the meeting.
Surprise coefficient
\(\beta_{u,j}\)
Basis-point yield response to a 100 bp unexpected tightening.
A Rate Cut Can Still Be Hawkish News
Hawkish and dovish describe the action relative to the expected counterfactual
\[
\Delta r_t^u=-25-(-50)=+25\text{ bp}
\]
Observed action
−25 bp
The Federal Reserve eases.
Unexpected component
+25 bp
Policy is 25 bp tighter than the market expected.
The Yield Response Reappears in the Surprise
Native chart: basis-point yield response to a 100 bp target component
Raw action
Surprise component
Two-year example
≈15 bp
A 25 bp surprise tightening implies \(0.25\times61.4\simeq15\) bp.
Result
Large across the curve
Surprise responses are 79.1, 61.4, 48.1, and 31.5 from 3 months to 10 years.
Kuttner (2001), Tables 1 and 3.
Target Surprises, 1990–2004
Target news varies sharply across meetings and is often small even when actions occur
Gürkaynak, Sack, and Swanson (2005); supplied course chart.