RDP 2005-02: The Impact of Monetary Policy on the Exchange Rate: A Study Using Intraday Data 2. The Estimation Framework

2.1 Event Study Methodology

We use an event study approach, estimating the change in the exchange rate around the announcement of ‘monetary policy decisions’. Decisions include both announced changes to monetary policy, and announcements of decisions to not change policy (‘no-change’ decisions) so long as the market knew for certain that a policy announcement would take place. Further discussion of the events used can be found in Section 2.2. In many cases, monetary policy decisions are widely anticipated by the market and so their impact should already be incorporated into interest rates and exchange rates. In order to identify the impact of a monetary policy decision, we isolate the surprise component of the change in monetary policy by using changes in market interest rates.[2] This technique, developed in Kuttner (2001), is commonly used in the literature. Market interest rates incorporate a risk premium, but the change in the market interest rate is a good proxy for the policy surprise as the risk premium is unlikely to move in the short time periods used in the event study (Piazzesi and Swanson 2004).

For each of the events, we measure the movement in the exchange rate around the event using intraday data. We use a short, 70-minute, event window. This reduces the amount of information received by the market in the event window and so the number of events that would have to be discarded due to the exchange rate and interest rate jointly responding to other news, such as a macroeconomic data release. Because the interest rate surprise will be a more dominant piece of information in a short event window, it should also result in more accurate estimates.

One potential source of concern is that exchange rate movements could influence monetary policy decisions. However, this is not likely in this study because in each central bank the main deliberation on policy changes occurs the day before the announcement. Given that changes in exchange rates are not autocorrelated, this means that the event window will not contain exchange rate movements that influenced the policy decision. The daily market interest rates may be affected by other events or an endogenous response to exchange rate movements, but this is minimised by the fact that events occur on days for which monetary policy is likely to be the most important shock to interest rates. Unfortunately, intraday interest rate data are not available for our sample of countries and time.[3] The events that are excluded from our sample for reasons of contamination are outlined in Section 2.2.

2.2 Monetary Policy Operations

The monetary policy operations of the four economies used in this study have changed considerably over the past decade (Brown 1997; Archer, Brookes and Reddell 1999; Parent, Munro and Parker 2003; Zettelmeyer 2004). This section briefly outlines the current monetary policy regimes, how they have changed, and how these changes may impact on this study. Using this information, Section 2.3 explains how the set of events used in the analysis were determined.

The four countries currently have very similar monetary policy operations. In particular, they all have:

  • fixed announcement dates for monetary policy decisions, albeit with the option to make changes at other times in response to extreme events;
  • a short-term interest rate as the policy instrument;
  • a preference for not surprising the market; and
  • an inflation target.

While all four countries have been inflation targeters for the full sample considered in this paper, institutional aspects of monetary policy operations have changed since the early 1990s in important ways. Some of these changes have been gradual, while others have been more abrupt. In Australia, monetary policy operations have changed progressively since the early 1990s, to resemble the current operational framework by about 1998. Prior to 1998, though the dates of the 11 Board meetings were known (the first Tuesday of every month, with no meeting in January), monetary policy decisions were typically not announced or implemented immediately after a meeting. From 1998 onwards, all changes in monetary policy have been announced the day after a Board meeting. Only since September 2002 has there been a public announcement on the day after the Board meeting in the event that policy was not being changed. However, market commentary in the period 1998 to 2002 suggests that if policy was not changed the day after a Board meeting then no change was anticipated until the subsequent meeting. So for Australia we have included no-change decisions from the beginning of 1998, as well as all changes in monetary policy from mid 1993. Table 1 provides a summary of the sample periods and number of events for each country.

Table 1: The Set of Events
  Australia Canada NZ UK
Sample 30/07/1993–
Number of events used 79 33(b) 42 82
Number of changes 24 23 20 27
Number of no-changes 55 10 22 55
Regime change 1998 December 2000 March 1999 June 1997
Old regime Policy changes included Policy changes included None included None included
New regime Changes and no-changes included
No of meetings per year 11 8 8 12
Time of announcement(a) 9.30 am on the day after the Board meeting 9 am on fixed announcement days 9 am on fixed announcement days 12.30 pm on the second day of the MPC meeting
Notes: (a) Some events do occur at different times.
(b) Changes after the 11 September 2001 terrorist attacks and in response to the Russian crisis are excluded. Also, eight events that coincide with changes in the Fed funds rate are excluded.

In the other countries, changes in monetary policy operations have been more distinct and in some cases substantial. At the start of 1999, the Reserve Bank of New Zealand (RBNZ) moved from focusing on a monetary conditions index (a combination of the overnight interest rate and the exchange rate) as the main intermediate target of policy to using the overnight cash rate to implement monetary policy. Accordingly, we begin our New Zealand sample in 1999, as prior to this the motivation for interest rate changes were inextricably linked to exchange rate movements.

Like New Zealand, Canada has implemented a system of eight fixed announcement dates (starting December 2000). However, unlike New Zealand, there has been little change in the framework and focus of policy. We therefore include most changes to policy from 1996 onwards, when our intraday exchange rate data for the Canadian dollar begin. Eight changes are excluded, when they are on the day of, or the day after, a change in the Fed funds rate, reflecting the likelihood of contamination of the measure of the surprise in policy. Two further policy changes are excluded, the one following the August 1998 Russian crisis and the one after the 11 September 2001 terrorist attacks, again for reasons of possible contamination.

In the United Kingdom the operational responsibility for monetary policy passed from the Government to the Bank of England in June 1997, ensuring an independent monetary policy-maker. We exclude monetary policy decisions made prior to this, owing to the large shift in the monetary policy regime and uncertainty about the exact time at which changes were announced prior to 1997.

For each event we searched Bloomberg and other sources to ensure that there was no contaminating information in the event window. Because we use a narrow event window, we did not find cause to exclude any events other than those outlined for Canada.[4] We also record the exact time of the event in order to make them completely comparable.

Given the similarities of the current monetary policy regimes in the four countries, we also present results using a pooled sample. In order to keep the sample as homogenous as possible, only those events that are part of the current regimes, in which monetary policy is implemented according to fixed announcement dates, are included. This is the full sample for New Zealand and the UK, and from 1998 for Australia and the end of 2000 for Canada.

2.3 The Data

The two data series used in the event study are interest rates and exchange rates. We use bank bill interest rates (1-month and 3-month) and futures contracts on the 3-month bank bill interest rate. Most of the literature for the US has calculated monetary policy surprises using Fed funds futures contracts (see, for example, Bernanke and Kuttner forthcoming; Faust, Rogers, Swanson and Wright 2003; Kuttner 2001). However, futures contracts over the policy instrument interest rates are not available for the countries we use over our sample period, and so we use bank bill interest rates to calculate monetary policy surprises. One advantage of bank bill rates is that, unlike futures, the horizon of the instrument does not vary from one event to another, thereby simplifying the calculation of the surprise. Piazzesi and Swanson (2004) find that eurodollar interest rate futures provide good measures of interest rate surprises for the US, and are only marginally outperformed by Fed fund futures. The interest rate surprise is calculated as the change in the 1-month or 3-month bank bill interest rate from the close of the day prior, to the close of the day of the monetary surprise. The surprises are measured in percentage points (100 basis points). We use this unit of measurement because it is a convenient round number and other authors have done likewise, not because it is a plausible magnitude for the surprise. Presently, the central banks in our sample move their policy interest rates in 25 basis point increments, and so this is presumably the maximum surprise. The interest rates we use, and their sources, are described in Appendix A.

The bilateral exchange rates are the US dollar price of the domestic currency, from the Reuters electronic trading system, at 10-minute intervals. At each 10-minute interval, the exchange rate observation is the average of the closest active bid and ask quotes. Goodhart and Payne (1996) and Danielsson and Payne (2002) have found that at 10-minute intervals, quote data are good proxies for actual transaction prices in exchange rate markets.

Table 2 gives a brief summary of the data. The average absolute change in the policy rate, |Δi|, is based on change and no-change event days in the sample. The average absolute change is typically around twice as large as the average absolute surprise, |Δis|. Exchange rate volatility measured as the average absolute change in the exchange rate over 10 minute intervals, is higher on event days than on non-event days (from 2 hours before to 6 hours after the event), providing some initial evidence that monetary policy has an effect on the exchange rate.

Table 2: The Data
  Australia Canada NZ UK
Number of events used 79 33 42 82
Average |Δi| 0.13 0.21 0.15 0.09
Average |Δis| 0.07 0.06 0.05 0.07
Ratio of event to non-event day exchange rate volatility (a) 1.34 1.17 1.33 1.13
Average |Δe10m| 0.049 0.041 0.059 0.052
Notes: |Δi| is the absolute change in the official interest rate in percentage points. |Δis| is the absolute change in the 1-month interest rate in percentage points.
(a) The volatility is calculated as the average absolute change in the exchange rate over 10-minute intervals. Averages are taken over a window starting two hours before the event and ending six hours after. The sample of non-event days is constructed by taking the day exactly one week prior to each event.
e10m| is the per cent change over 10-minute intervals on event days.


This does not mean that anticipated changes in monetary policy have no effect on the exchange rate, but that the effect has been incorporated into the exchange rate at the same time as markets came to the conclusion that there would be a change in monetary policy. [2]

We investigated using intraday exchange rate forwards to derive a measure of the intraday interest rate shock based on covered interest rate parity. But quotes for exchange rate forwards are not updated frequently and so the length of the period used to measure the shock varied from one event to another, potentially in a way that correlates with the nature of the shock. This measure of the shock was then found to result in larger standard errors, though the point estimates were roughly equivalent. [3]

A few events are excluded due to missing intraday exchange rate data. [4]