# The surprise in monetary surprises: a tale of two shocks

Silvia Miranda-Agrippino.

Empirical identification of the effects of monetary policy requires isolating exogenous shifts in the policy instrument that are distinct from the systematic response of the central bank to actual or foreseen changes in the economic outlook. Because the same tools are used to both induce changes in the economy, and to react to them, distinguishing between cause and effect is a far from trivial matter.  One popular way is to use surprises in financial markets to proxy for the shock. In a recent paper, we argue that markets are not able to distinguish the shocks from the systematic component of policy if their forecasts do not align with those of the central bank. We thus develop a new measure of monetary shocks, based on market surprises but free of anticipatory effects and unpredictable by past information.

A recently proposed way to identify monetary policy shocks in structural VARs is to use monetary surprises as noisy measures of the shock. These are defined as the movement in interest rate linked securities that are observed after a monetary policy announcement.

The econometrician’s hope (and assumption) is that a monetary policy shock is the only reason why markets move at announcements. However, because private sector forecasts may differ from central banks’ forecasts, markets may fail to correctly account for the systematic component of policy when they are surprised by an interest rate decision. In other words, they may be confounding cause and effect if their forecasts do not coincide with those of the monetary authority. If the two forecasts differ, the monetary surprises cannot be thought of as being exogenous, or assumed to be isolating the correct signal.

We use ‘expected’/’unexpected’ to refer to private-sector/market forecasts, and ‘anticipated’/’unanticipated’ for central banks’ forecasts. An event that is anticipated and unexpected is in the information set of the central bank but not in that of the public (e.g. news about future inflation). A monetary policy shock is a deviation from the monetary policy rule.

What is a monetary surprise?

Monetary surprises are the changes in the price of interest rate futures that are triggered by monetary policy announcements. Because markets tend to react quickly to new information, narrow (30-minutes) time windows are typically enough to measure these surprises. Also, a short window ensures that the policy announcement dominates or, better yet, is the sole event responsible for the move.

Interest rate futures allow translating price adjustments into changes in the expected path of policy. The price of a futures contract that pays the nominal interest rate it at date t + h can be written as the sum of two components. One relates to what investors expect the interest rate to be at t + h. The other, residual, is a risk compensation (premium) that they require to hold such a contract until maturity.

If the announcement is the only event in the window, and if the risk premium is constant in that time frame, the price change is a revision in the expectation about the policy rate that is triggered by the announcement.

Moreover, markets would want to react only to what they didn’t see coming, as everything else should have already be priced in. The surprises are thus a measure of the component of monetary policy unexpected by market participants.

Unexpected decisions and monetary policy shocks

To map the revision in market-based expectations about the policy rate into a monetary policy shock one needs to make another crucial assumption: the set of forecasts on which the central bank builds its decision and those of the private sector coincide.

Only if this holds what is unexpected by markets is also unanticipated by the central bank, in the sense that it is not part of the systematic reaction to current or expected macroeconomic conditions that make up the reaction function of the monetary authority.

The mapping between price updates and monetary policy shocks is then completed, and the surprises can be used for the identification of its effects on whichever macroeconomic or financial environment of interest. This holds true even if some random measurement error is present. But it does not if forecasters disagree.

To see this, consider the price of a futures contract a few minutes before the announcement, at ${t-\Delta t}$. Let’s also assume that we have access to the same data as the central bank, and that we understand the policy rule $f(\ \widehat{\Omega}_{t|t}\ )$. Given a set of economic forecasts $\ \widehat{\Omega}_{t|t}\$, we know what the response of the central bank will be.

Both the expected interest rate decision and the risk compensation are a function of our current assessment of the economy (or nowcast) $\ \widehat{\Omega}^M_{t|t}\$. The price is then

$p_{t-\Delta t} = f(\ \widehat{\Omega}^M_{t|t}\ ) + \zeta(\ \widehat{\Omega}^M_{t|t}\ )$

$f(\ \widehat{\Omega}^M_{t|t}\ )$ is the expected decision given our assessment of the economy. Based on that same assessment, the premium that we demand is  $\zeta(\ \widehat{\Omega}^M_{t|t}\ )$

After the decision is announced, the price is updated to

$p_t = \underbrace{f(\ \widehat{\Omega}^{CB}_{t|t}\ ) + e_t}_{\textstyle{i_t}} + \zeta(\ \widehat{\Omega}^{CB}_{t|t}\ )$

The new interest rate is a function of the central bank’s assessment of the economy $f(\ \widehat{\Omega}^{CB}_{t|t}\ )$, plus a potentially non-zero shock et.  Because the relevant forecast has changed, the risk premium is now $\zeta(\ \widehat{\Omega}^{CB}_{t|t}\ )$

The monetary surprise (mpst) is thus equal to

$mps_t \equiv p_t - p_{t-\Delta t} = e_t + \tilde{f}(\ \widehat{\Omega}^{CB}_{t|t} - \widehat{\Omega}^M_{t|t}\ ) + \tilde{\zeta}(\ \widehat{\Omega}^{CB}_{t|t} - \widehat{\Omega}^M_{t|t}\ )$

and it is thus only a function of et if $\ \widehat{\Omega}^{CB}_{t|t} = \widehat{\Omega}^M_{t|t}\$ . This assumption has been challenged in a number of papers, starting with Romer and Romer (2000).

A testable implication of this decomposition is that surprises should be predictable by the forecast that the central bank uses to inform its decision. On top of that, our theory implies that because of the forecast asymmetry, monetary surprises are contaminated by a time-varying risk premium. This can be tested by regressing the surprises on prior information, such as factors summarising the pre-existing economic and financial conditions in the economy. We find significant support for these predictions in the data: ‘surprises’ are predictable both in the US and in the UK.

Hence, when markets react to the announcement they can be surprised due to the monetary policy shock, the nowcast update, or a combination of the two.

Does it matter why markets move at announcements?

The short answer is “Yes, it does”.

To illustrate the point, Figure 1 depicts responses of US variables to a contractionary monetary policy shock that increases the policy rate by 1% within a fairly standard monetary VAR (e.g. Coibion, 2012, Ramey, 2016) where no controls for financial/credit markets are included. The shock is identified using average surprises in 3-month federal funds futures as in Gertler and Karadi (2015).

The response of both output and unemployment are opposite to what standard macroeconomic theory would predict. While the dynamic of the response functions is bound to depend on the VAR composition, if the monetary surprises were truly only a function of the shock, they should recover impact responses of the correct sign regardless of what other variables are included.

A better candidate for the job

According to our story, the key is to remove anticipatory effects. Or realign the two forecasts.

The way suggested is to project the raw surprises on the information likely to enter the central bank reaction function, and use the residuals to identify the monetary policy shock. The conditioning set, similar to the one in Romer and Romer (2004), includes central banks’ forecasts and revisions to forecasts for output, inflation and unemployment. Also, controls for systematic reactions to rate changes of either sign, and for contracts based on other interest rates (Libor), are included. As implied by the decomposition in the previous section, the conditioning removes the dependence on past information as well: the same lagged macro-financial factors that were predictive of the raw surprises are uncorrelated with the conditional ones.

The effect of this ‘cleaning’ is in the red lines in Figure 2, where two new sets of responses are superimposed to the ones in Figure 1 (dark blue). As a reference point, the light blue lines identify the monetary policy shock assuming that the central bank reacts only to the information in the VAR (standard Cholesky factorization with fed funds rate ordered last). The composition of the VAR is the same in the three cases.

The conditional surprises, free of anticipatory effects, recover responses of the expected sign even in this limited-information VAR. Analogous results are obtained for the UK.

Conclusions

The empirical identification of monetary policy shocks requires isolating exogenous shifts in the policy instrument that are not due to the endogenous response of policy to the economic outlook. We argue that while market surprises successfully capture the component of policy unexpected by market participants, they map into the shocks only under the (restrictive) assumption that markets can correctly and immediately disentangle the systematic component of policy from any observable policy action. We develop a new measure for monetary policy shocks that is orthogonal to the central banks own forecasts and unpredictable by past information. This measure allows us to recover impulse response functions with the ‘right’ signs, even in small and informationally deficient VARs.

Silvia Miranda-Agrippino works in the Bank’s Macro-Financial Analysis Division.

The underlying data for this post is available to download in .XLSX format.