1. Discussion of Monetary Policy and the Money Market Yield Curve Conference on the Analysis of the Money Market European Central Bank November 14, 2007 Eric T. Swanson Federal Reserve Bank of San Francisco Note: The views expressed in this presentation are the authors’ and do not necessarily reflect the views of the management of the Federal Reserve Bank of San Francisco or any other individuals within the Federal Reserve System.

2. Hiona Balfoussia, “An Affine Factor Model of the Greek Term Structure” Michael Fleming and Monika Piazzesi, “Monetary Policy Tick by Tick” Maria Athanasopoulou, Claus Brand, and Rasmus Pilegaard, “Does Real-Time Macroeconomic Information Affect the Yield Curve?” (ABP) Three Papers

3. Balfoussia, “Affine Model of Greek Term Struct” Standard affine no-arbitrage model of yield curve: Greek yield data, March 1999 – February 2007 Weekly frequency Three latent factors: “level”, “slope”, “curvature” Plots factors, relates them to Greek macro environment Special issues related to Greece: Acceptance of Greece into EMU, process of convergence 2004 Olympics Increasing maturity of Greek sovereign debt

4. Fleming-Piazzesi, “Monetary Policy Tick by Tick” Paper has two parts: 1.U.S. Treasury market microstructure In response to FOMC announcements: Volatility is higher Trading volume is higher Bid-Ask spreads are wider 2. Effects of FOMC announcements Measure yield curve response using tick data Find ex ante slope of yield curve has been correlated with market response to FOMC announcement Idea: yield curve that slopes up or down signals “stress”, long-term yields more sensitive to FOMC

5. ABP, “Does Real-Time Info Affect Yield Curve?” Measure response of euro area yield curve to major macroeconomic announcements using tick data: U.S. studies regress yield curve movements on “surprise” component of macro announcement But in euro area, survey expectations data are poor ABP use market-based approach to measure news, same idea Gurkaynak, Sack, Swanson (2005) used to measure FOMC statements Incorporate these surprises into dynamic latent-factor term structure model

6. Comparison of the Three Papers Countries Greece, US, Euro Area Frequency Tick-by-tick (ABP, FP), daily (ABP), weekly (B) Effects of Monetary Policy ABP: two dimensions, FP: one dimension, time-varying, B: not specifically modeled Incorporation of high-frequency data Monetary policy announcements (FP, ABP), macro data releases (ABP) Relationship between high-frequency data and yield curve ABP: high-frequency data affects yield curve dynamics FP: shape of yield curve affects high-frequency response

7. Advantages of Tick Data Markets incorporate information quickly: Source: Gurkaynak, Sack, and Swanson (2005 IJCB)

8. Intraday data increases precision, can eliminate bias: Advantages of Tick Data Source: Gurkaynak, Sack, and Swanson (2005 IJCB)

9. The Big Picture What do yield curve movements tell us about: Monetary policy expectations? Risk premia? Inflation expectations? Measure effects of monetary policy announcements on: Monetary policy expectations Risk premia Inflation expectations Asset prices The macroeconomy Information Control

10. Effects of Monetary Policy FOMC announcements consist of two parts: 1.Action: change in the target federal funds rate 2.Statement: rationalizing action, describing outlook ECB announcements potentially even more complex: 1.Action 2.Brief statement 3.One-hour press conference, with questions & answers

11. Effects of Monetary Policy Gurkaynak, Sack, and Swanson (2005): Test dimensionality of financial market responses to FOMC announcements, find # dimensions = 2 Extract two latent factors from financial market responses –rotate so that first factor is monetary policy action (Kuttner) –second factor (“path” factor) is then measure of statement Measure effects of monetary policy actions and statements on yield curve ABP: follow this approach for ECB FP: follow one factor approach (actions only)

12. Effects of Monetary Policy Source: Gurkaynak, Sack, and Swanson (2005 IJCB)

13. Effects of Monetary Policy Fleming-Piazzesi

14. Effects of Monetary Policy ABP consider “target” and “path” factor responses to major euro area macro announcements, not just monetary policy Idea: macro news monetary policy exp yield curve Potential problems with this generalization: Are there 2 dimensions to ECB monetary policy announcements? (or more?) When we consider all major macro announcements, even less clear that 2 dimensions is enough—maybe 5 or 6 dimensions are required? (GSS, Stock-Watson) Maybe macro news has effects on yield curve beyond monetary policy expectations (r*, risk premia, changes in higher moments)

15. Risk Premia Accounting for risk premia is potentially very important 3-month US$ Interest Rates in 2007

16. Balfoussia, ABP use affine no-arbitrage model to account for risk premia However: Estimates of risk premia can be very sensitive to model, especially assumptions about long-run behavior (π*, r*) Risk Premia

17. Estimates of risk premia can be sensitive: Source: Rudebusch, Sack, and Swanson (2007)

18. Balfoussia, ABP use affine no-arbitrage model to account for risk premia However: Estimates of risk premia can be very sensitive to model, especially assumptions about long-run behavior (π*, r*) For weekly or daily data, model dynamics are trivial: every factor is an independent random walk How much would authors’ risk premium estimates differ using standard monthly frequency data? When estimating risk premium on long-term bonds, which short-term interest rate is being used? (fed funds rate, overnight repo, 1-mo T-bill, 3-mo T-bill?) Risk Premia

19. High-Frequency Data and Models of Risk ABP include high-frequency “target” and “path” factor responses as shocks in affine no-arbitrage model Idea seems to be to rotate latent factors so that first two are “target” and “path” instead of “level” and “slope” Alternative: keep latent factors as “level” and “slope”, and regress on “target” and “path” surprises. FP also seem to want to relate high-frequency response of yields to risk premia, but this relationship is never specified

20. Final Thoughts High-frequency data are the state of the art for measuring effects of monetary policy Monetary policy announcements are not one-dimensional; two or more dimensions are required Ideally, we would like to control for risk premia when studying movements in the yield curve However, measures of risk premia are sensitive to model used, assumptions about long-run relationships (π*, r*), frequency, and which short-term rate is the benchmark Convincingly controlling for risk premia currently impossible Alternative: be aware that risk premia may be an issue, try to show that results are not systematic risk responses