1. BY: VIGNESH NATHAN ADVISOR: ANDREW PATTON Impact of Macroeconomic Announcements on Exchange Rates

2. Basic Ideology Initial Foreign Exchange Rate Macroeconomic Announcement Reactionary Foreign Exchange Rate 5 minutes It is a central assumption of this study that within this 10-minute interval, the announcement drives all foreign exchange trading and is responsible for all resulting fluctuations in prices. For this precise reason, we can treat the announcement as a “natural experiment.” This will allow me to use a relatively simple model.

3. Methodology The model is incredibly simple: R t = α + βS t + u t Where, R t is the ten minute return on the currency, given by: R t = ln(FX t+5min ) – ln(FX t-5min ) S t is the announcement surprise, given by: S t = Announcement t - Expectation t u t is the error term.

4. Asymmetric Response Very widely cited phenomenon in finance, states that foreign exchange markets respond in an unbalanced manner to different types of surprises, either given by its value (positive vs. negative, large vs. small) or its environment (good times vs. bad times) Asymmetric Response Tests Test #1Test#2 Positive SurprisesNegative Surprises Large Surprises (S t > median[abs{S t }]) Small Surprises (S t <= median[abs{S t }]) Pre-Recession (Pre-January 2008) Recession (Post-January 2008)

5. Testing for Asymmetric Response I R t = α + β 1 S t *1{Test #1==TRUE}+ β 2 S t *1{Test #2==TRUE}+ u t Β 1 measures the impact of surprise type 1. Β 2 measures the impact of surprise type 2. From there, we have the following hypothesis test: H o : β 1 -β 2 = 0 H 1 : β 1 -β 2 ≠ 0 Given by the following distributions: β 1 ~ (β 1, σ 1 2 ) β 2 ~ (β 2, σ 2 2 )

6. Testing for Asymmetric Response II Therefore, (β 1 -β 2 ) ~(β 1 -β 2, σ 1 2 + σ 2 2 - 2σ 1 σ 2 ρ 1,2 ) And, t-statistic = β 1 -β 2 / √[(σ 1 2 + σ 2 2 - 2σ 1 σ 2 ρ 1,2 )] In certain cases, such as pre-January 2008 vs. post- January 2008, we can assume that the events are independent, and thus uncorrelated. t-statistic = β 1 -β 2 / √[(σ 1 2 + σ 2 2 )]

7. Testing for Asymmetric Response III R t = α + β 3 S t *1{Test #1==TRUE & Test #2==TRUE)+ β 4 S t *1{Test #2==TRUE}+ u t In such a way that β 4 is the marginal impact of a Test #2 surprise. When we compare this to our old regression, given by: Original Regression: R t = α + β 1 S t *1{Test #1==TRUE}+ β 2 S t *1{Test #2==TRUE}+ u t We can derive the following: β 1 = β 3 β 2 = β 3 +β 4 β 2 -β 1 = β 3 +β 4 -β 3 = β 4 (β 2 -β 1 ) ~(β 2 -β 1, σ 1 2 + σ 2 2 - 2σ 1 σ 2 ρ 1,2 ) = (β 4, σ 4 2 )

8. Current Workflow Completed regression tables for Canada, Australia and United States (using USD-AUD).  Still need USD-EUR, USD-AUD not showing many significant results. Plan to expand time return to 60- and 120-minute interval surrounding announcement, to see if the effect ever diminishes.

9. Previous Concerns Robust standard errors were implemented. Chose to keep nominal value of surprise instead of relative value to avoid the confusion of independent variable representing a percent change of a percent change. Chose unemployment rate over number of employed persons in order to maintain consistency of the unit of the independent variable.