Time Variation in Asset Price Responses to Macro Announcements - PDF

Please download to get full document.

View again

of 17
All materials on our website are shared by users. If you have any questions about copyright issues, please report us to resolve them. We are always happy to assist you.
Information Report
Category:

Documents

Published:

Views: 5 | Pages: 17

Extension: PDF | Download: 0

Share
Related documents
Description
Federal Reserve Bank of New York Staff Reports Time Variation in Asset Price Responses to Macro Announcements Linda S. Goldberg Christian Grisse Staff Report No. 626 August 213 This paper presents preliminary
Transcript
Federal Reserve Bank of New York Staff Reports Time Variation in Asset Price Responses to Macro Announcements Linda S. Goldberg Christian Grisse Staff Report No. 626 August 213 This paper presents preliminary findings and is being distributed to economists and other interested readers solely to stimulate discussion and elicit comments. The views expressed in this paper are those of the authors and are not necessarily reflective of views at the Federal Reserve Bank of New York or the Federal Reserve System. Any errors or omissions are the responsibility of the authors. Time Variation in Asset Price Responses to Macro Announcements Linda S. Goldberg and Christian Grisse Federal Reserve Bank of New York Staff Reports, no. 626 August 213 JEL classification: E43, E44, E52, F31, G12, G14, G15 Abstract Although the effects of economic news announcements on asset prices are well established, these relationships are unlikely to be stable. This paper documents the time variation in the responses of yield curves and exchange rates using high-frequency data from January 2 through August 211. Significant time variation in news effects is present for those announcements that have the largest effects on asset prices. The time variation in effects is explained by economic conditions, including the level of policy rates at the time of the news release, and risk conditions: Government bond yields increase in response to good news, but less so when risk is elevated. Risk conditions matter since they can capture the effects of uncertainty on the information content of news announcements, the interaction of monetary policy and financial stability objectives of central banks, and the effect of news announcements on the risk premium. Key words: macroeconomic news announcements, high-frequency data, bond yields, exchange rates, monetary policy, risk Goldberg: Federal Reserve Bank of New York ( Grisse: Swiss National Bank ( The authors thank Katrin Assenmacher and participants at the Swiss National Bank s brown bag seminar and the 212 annual meeting of the American Economic Association (Chicago) for comments, as well as Leslie Shen and Diego Gilsanz for excellent research assistance. The views expressed in this paper are those of the authors and do not necessarily reflect the position of the Swiss National Bank, the Federal Reserve Bank of New York, or the Federal Reserve System. 1 Introduction A rich literature explores the consequences of economic news announcements, such as inflation releases and employment payrolls reports, for asset prices, risk premia, and exchange rates. These consequences are measured within windows that cover minutes or hours after the economic data release, as in Andersen, Bollerslev, Diebold, and Vega (23), and sometimes are assessed in relation to the predictions of basic economic models containing interest parity conditions and Taylor-rules for monetary policy, as in Gürkaynak, Swanson and Sack (25) or Faust, Rogers, Wang, and Wright (27). When the economic news effects are assessed in light of these models, they are viewed as informing how market participants view future interest rate paths conditioned on updated views of trajectories of inflation and the output gap. In the international setting, the news inform the relative trajectories of yields across countries, as well as informing exchange rates and risk premia. The magnitudes of such effects of economic news are often discussed as if rules-of-thumb underlie the relationships. Yet, there is little reason to expect that the relationships between economic news and asset prices should be stable over time. Some studies provide relevant insights, for example showing that the effects of Federal Reserve policy announcements change in a zero lower bound environment (as in Kiley 213). Policy regimes also play a role as central banks convince markets of the relative importance of inflation and output priorities in a policy reaction function, as Goldberg and Klein (211) show: variation in economic news effects on European asset prices and on the euro/dollar exchange rate are indicative of market participants having evolving perceptions of the relative inflation aversion reflected in ECB policymaking. In this paper, we argue that time-variation in the effects of news on bond yields and exchange rates should be viewed as an empirical regularity. This time variation could have a number of sources, which we motivate in the context of Taylor-rule type models of policy reaction functions. We conjecture that time variation arises as the policy outcomes of news change due to a perceived reweighting of inflation and output preferences within reaction functions, due to changing implications of a unit of news for forecasts of output or inflation as the state of the economy shifts closer to or further from targets, due to changing risk preferences in the economy, or due to the importance of financial stability conditions leading to a (short run) shift of priorities of central banks. We document the time variation in consequences of US economic news on the interest rates and exchange rates of the US, UK, Germany, and France using high frequency data for the period from 2 to 211. Using econometric methods developed by Müller and Petalas (21) and Elliott and Müller (26), we show that persistent time variation is present to differing degrees in the high frequency data. We relate the observed time-variation patterns to macroeconomic conditions, the level of the Federal Funds rate, and to measures of risk. The level of interest rates and risk conditions have the greatest explanatory power for changes observed in asset price responsiveness to news. In particular, while US bond yields usually increase in response to positive US macroeconomic news, the increase is smaller when policy rates and risk conditions are elevated. The role of risk in explaining time variation in economic news effects likely reflects two 1 possibly complementary channels. First, markets may view the Federal Reserve as less likely to raise rates in times of increased financial turmoil, perhaps due to a latent financial stability objective. Second, markets may place less weight on news announcements when the relationship between these news and the economic outlook is more uncertain. The information content of the news may be diminished when overall risk is elevated. Quantitatively, we find that the responses of US 2-year bond yields to a one standard deviation surprise in non-farm payrolls vary between -2 and +13 basis points (measured over the window including 5 minutes before and after the release), compared with an average effect of 5 basis points between 2 and 211. The bulk of that variation is explained by the level of the policy rate and the VIX index. Section 2 provides a brief review of the related literature. Section 3 describes our data and empirical methods, and section 4 reports our baseline results for asset price responses to US data announcements, as well as tests for evidence of gradual time variation in these responses. Section 5 explores how asset price responses to news announcements vary with changes in macroeconomic and financial conditions. Section 6 concludes with a discussion of the economic relevance of time variation and open questions for research. 2 Relationship to the previous literature A large number of papers has established that asset prices respond to macroeconomic data announcements, and are thus directly linked to underlying economic fundamentals. Most papers find that economic news is incorporated quickly (within minutes) into asset prices, with some measurable persistence of these effects. Some types of news for example, US non-farm payrolls announcements generate larger asset price responses than others. News which are more timely (in the sense that the announcement date and the reference date are close together), more precise (in the sense of being subject to smaller revisions on average), and contain more information (in the sense of being better able to better forecast GDP growth, inflation or central bank policy decisions) have a larger effect on asset prices (Andersen et al. (23), Hautsch and Hess (27), Gilbert et al. (21)). Several studies have also considered time variation in the effect of a given type of announcement. In an early contribution, Cocco and Fischer (1989) find evidence that the response of US interest rates to money announcement surprises is stable over time within a linear model where the news response coefficient is assumed to follow an AR(1) process. 1 More recently, a number of papers have estimated the effect of news separately over different sample periods and tested for parameter constancy. Using a Nyblom (1989) test, Faust et al. (27) argue that the effects of news are mostly stable over time. However, they also find evidence that some news effects on asset prices have fallen over time in absolute magnitude. Fratzscher (29) finds that positive US macro announcements were associated with an appreciation of the US dollar between 1994 and 28, but with a depreciation of the US dollar between 28 and 29. Using rolling regressions and random effects models applied to data that span the period from 1993 to 28, 1 See also Fischer (1989). 2 Ehrmann et al. (211) find that the responses of euro area bond yields to data announcements became more similar across countries after the introduction of the EMU. A number of papers have gone beyond showing that time variation exists and have highlighted specific reasons for that variation. Three findings emerge. First, asset price responses to news often appear to be non-linear: negative surprises have larger absolute effects than positive surprises, and larger surprises generate a disproportionately larger response (Andersen et al. (23), Andersen et al. (27), Ehrmann and Fratzscher (25), Hautsch and Hess (27)). Second, the reaction may depend on the state of the economy with news announcements have a larger effects on bond yields during economic contractions (Andersen et al. (27). The sign of the response of stock prices to real announcements (unemployment) also depends on the state of the economy: higher than expected unemployment increases stock prices in expansions and reduces stock prices in recessions. This asymmetric response could reflect the effect of news on expected interest rates, expected cash flows or the risk premium. As argued by Boyd et al. (25), the discount rate effect dominates in expansions (higher unemployment implies lower expected interest rates), while the cash flow effect dominates in contractions (higher unemployment implies lower expected earnings). 2 Ehrmann and Fratzscher (27) find larger exchange rate responses to news following weeks of high FX volatility, following a string of news announcements that surprised markets in the same direction, and following a string of large surprises. They conclude that uncertainty matters for the news response. Third, market participants may change their view of central bank priorities. Goldberg and Klein (211) argue that time variation in euro area bond yield responses to news evolved in the years after the introduction of the euro. The pattern of evolution was consistent with the markets viewing the ECB as having established more inflation-fighting credibility after a few years of operation and responses to macroeconomic conditions. Building on these earlier papers, we focus squarely on time variation in the response of crosscountry bond yields and exchange rates to US macroeconomic announcements. Relative to the previous literature our paper makes three contributions. First, we provide a deeper evaluation of time variation in the effects of economic news on asset prices, applying the econometric techniques of Elliott and Müller (26) and Müller and Petalas (21). Second, we argue that time variation should be viewed as the default condition and that asset price responses to news should change with risk conditions and macroeconomic context, as well as with (likely less frequent) changes policy reaction functions. Third, we test these propositions using a rich set of data and over a relevant historic period. The high frequency asset price data covers the period from 2 to 211, which encompasses the global financial crisis and changes in the state of the macroeconomy economy and policy environment. The asset prices we examine in depth are bond yields and exchange rates for the United States, Germany, France, Switzerland, and Japan. 2 Conrad et al. (22) show that the response of stock prices to earnings announcements depends on the level of the overall stock market. 3 3 Data and methodology 3.1 Data The data releases we examine pertain to United States economic activity, including those indicators that have been previously established as important for generating price reactions, and are those for which market expectations are available. 3 We focus on only those data releases that have announcement times of 8:3am Eastern Standard Time (EST), a restriction that facilitates our work of collecting high frequency asset price data over an eleven year interval and still captures the majority of important US announcements. The data releases we include are: the consumer price index (CPI, total and excluding food and energy), the change in non-farm payrolls, the unemployment rate, GDP, housing starts, core inflation in personal consumption expenditures (PCE), personal income and spending, retail sales less autos, and the empire manufacturing survey. Data sources, frequency, and units are provided in Table 1. Most series have 14 observations for the 2 to 211 period, given that releases are typically monthly and the sample spans about eleven years. [Table 1 about here] The economic news that lead to asset price updating are constructed, following the convention in the literature, as the difference between the actual release value and the markets prior expectation of the contents of the release. The expectations data we use are median responses from weekly surveys of market participants conducted by Money Market Services, a division of Standard & Poor s, for the early part of the sample and more recently from Action Economics or Bloomberg News. 4 The bond yield and exchange rate series are constructed from high-frequency data drawn from transaction-level databases from Thomson-Reuters, supplemented by BrokerTec data for U.S. bond yields (Table 2). We focus mainly on 2, 5, and 1 year bond yields for the United States, United Kingdom, Germany, and France. 5 The exchange rates examined are euros (EUR), UK pounds (GBP), Japanese yen (JPY), and Swiss francs (CHF) all measured as US dollars (USD) per foreign currency. 3 Some examples are Andersen, Bollerslev, Diebold and Vega (23), Bartolini, Goldberg, and Sacarny (28), Ehrmann and Fratzscher (25), Faust, Rogers, Wang and Wright (27), Fleming and Remolona (1999), Goldberg and Leonard (23), and Gürkaynak, Sack and Swanson (25). 4 Money Market Services were the source of these data through December 23. Haver Analytics provided continuous expectations and announcement data through 25 using data from Action Economics. Gürkaynak and Wolfers (27) show that these data have been among the best performing expectations series for important macroeconomic variables over the sample period that we analyze. Later period data were drawn from Bloomberg. 5 The BrokerTec data had the most complete coverage of U.S. 2, 5, and 1 year transactions in Treasuries. However, these data report price information but not the yield. We compile the coupon rates for the 2-, 5-, 1 year treasury over the time period and use the price, settlement date, and maturity date to compute the yields. While we also have collected bond yields for Swiss and Japanese bonds, the particular high frequency platforms that serve as our data source do not have adequate representation of transactions across the range of news dates covered by our study. Most of the secondary market trading of Japanese government bonds is concentrated on OTC transactions through Japan Bond Trading Company 4 [Table 2 about here] From the transaction-level observations we build observed prices for each date and time window relevant for our analysis. The windows are chosen to allow for information diffusion and to generate sufficient transaction observations at each date. The price at a time stamp such as 8:25am in our analysis is constructed as the average of all transaction prices in the two minutes on either side of the indicated time (so 8:23-8:27am in this example). In the case of the spot exchange rate observations, the spot transaction prices are constructed as the average of bid and ask prices, or just the bid or ask price if information on only one of the two prices is reported for a transaction. Our empirical exposition presents results for asset price responses to news over the windows from 8:25am to 8:35am, and from 8:25am to 4pm. We also have run all specifications for the windows: 8-8:35am, 8-9am, 8am-4pm, 8:25-9am, and 8:25am-4pm. The exposition focuses only on the short window and the long window since these results appropriately reflect the trade-offs associated with window selection and implicit in prior studies. A tight time frame for market reactions as reflected in the 8:35 end time has the advantage of capturing a spontaneous market response. Yet the short window could be too abbreviated to capture analysis of news by market participants and thus may miss the full market reaction. The broader time frame, as reflected in a 4pm closing time, allows for a more thorough analysis of the information content of the announcement, but, as stressed in Andersen et al. (23), introduces the likelihood that additional information during the longer time frame could bias the coefficients (if correlated with the announcement surprise included in the regression) or cloud the significance of the estimated effects. 3.2 Empirical methods The empirical approach proceeds in three steps. First, the high-frequency asset price responses to economic data surprises are estimated in a setting where the effects of news surprises are assumed to be constant over time. This analysis complements earlier studies which have looked at the same types of effects of news on asset prices but over different sample periods and using different data sources. The results serve as an analytical benchmark for our subsequent analysis of time-varying coefficients. In the second step, we employ econometric methods developed by Elliott and Müller (26) and Müller and Petalas (21) to test for time variation in the effects html). Bond yields are constructed using the formula ( ) redemption + rate 1 freqency Y IELD = par 1 + A E rate freqency ( ) par + A rate 1 E freqency frequency E DSR where A denotes the number of days from the beginning of the coupon period to the settlement date (accrued days); DSR is the number of days from the settlement date to the redemption date; E is the number of days in the coupon period; and frequency is the number of coupon payments per year. For annual payments, frequency = 1; for semiannual, frequency = 2; for quarterly, frequency = 4; rate is the security s annual coupon rate; redemption is assumed to be $1, for every $1 of the bond; and par is the quoted transaction price in dollars for every $1 of the bond. 5 of data surprises on asset prices, and to estimate the parameter paths of these effects. Third, we explore the contributions of macroeconomic and financial conditions to the observed time variation in the effects of economic data surprises on financial markets. For the first step we estimate the linear model: q t + q t = K β k s k,t + ε t (1) k=1 where q t + q t is the change in asset price q over a time window from t to t + around t, s k,t is the surprise component of the kth data announcement released at time t, and β k are parameters assumed to be constant. 6 US announceme
Recommended
View more...
We Need Your Support
Thank you for visiting our website and your interest in our free products and services. We are nonprofit website to share and download documents. To the running of this website, we need your help to support us.

Thanks to everyone for your continued support.

No, Thanks
SAVE OUR EARTH

We need your sign to support Project to invent "SMART AND CONTROLLABLE REFLECTIVE BALLOONS" to cover the Sun and Save Our Earth.

More details...

Sign Now!

We are very appreciated for your Prompt Action!

x