The world market risk premium and U.S. macroeconomic announcements

Journal of International Money and Finance 58 (2015) 75–97

Contents lists available at ScienceDirect

Journal of International Money and Finance j o u r n a l h o m e p a g e : w w w. e l s e v i e r. c o m / l o c a t e / j i m f

The world market risk premium and U.S. macroeconomic announcements Ding Du a, Ou Hu b,* a b

The W. A. Franke College of Business, Northern Arizona University, PO Box 15066, Flagstaff, AZ 86011, USA Department of Economics, Youngstown State University, Youngstown, OH 44555, USA

A R T I C L E

I N F O

Article history: Available online 18 August 2015 JEL Classification: F30 G15

Keywords: Macroeconomic announcements International capital asset-pricing model Cross-listed firms

A B S T R A C T

Conditional tests of the International CAPM in previous studies (e.g., Harvey, 1991) help identify predictability but not causality. In this paper, we take an event-study approach to examine if the world market risk premium is particularly higher on prescheduled US macroeconomic announcement days. Empirically, we apply the Savor and Wilson (2014) methodology to daily US stocks as well as foreign stocks cross-listed in the US. Our findings suggest that there is a causal relationship from the state of the global economy to the world market risk premium. Published by Elsevier Ltd.

1. Introduction Solnik (1974a) and Grauer et al. (1976) develop an international version of the capital assetpricing model (International CAPM) in which the world market return is used as the market proxy. The International CAPM has been a benchmark in international finance (Lewis, 2011). Early studies (e.g., Solnik, 1974b; Stehle, 1977) focus on unconditional tests of the International CAPM. Subsequent research uses conditional tests to understand how the world market risk premium varies with the state of the global economy.1 For instance, Harvey (1991) finds that the world market risk premium varies with global state variables proxied by US dividend yield, US term structure, and US default spread.

Part of this research was conducted while Ding Du was visiting the Finance Department at the University of Maryland’s Robert H. Smith School of Business. * Corresponding author. Tel.: +1 330 941 2061. E-mail address: [email protected] (O. Hu). 1 See also Stulz (1981), Adler and Dumas (1983), Ferson and Harvey (1993), Dumas and Solnik (1995), De Santis and Gerard (1998), Fama and French (1998), Koedijk et al. (2002), Harvey et al. (2002), Ng (2004), Hou et al. (2011), Fama and French (2012), Du and Hu (2012), and Balvers and Klein (2014). http://dx.doi.org/10.1016/j.jimonfin.2015.08.006 0261-5606/Published by Elsevier Ltd.

76

D. Du, O. Hu / Journal of International Money and Finance 58 (2015) 75–97

It is important to point out that the conditional tests in previous studies (e.g., Harvey, 1991) help identify predictability/association but not causality, because the predictability/association could be due to reverse causality and/or confounding factors. For instance, it may not be the case that the global state variables considered by Harvey (1991) (such as US default spread) drive the world market risk premium. Instead, these global state variables may merely reflect the expectations about future world market risk premium. Intuitively, if investors expect that the price of risk in an integrated global market will increase (i.e., more systematic risk in the global market), they may bid down prices of risky assets (such as junk bonds) and therefore drive up the default spread. How can we identify causal effects? A standard approach in the finance literature is the event study, which has been used extensively since Fama et al. (1969). We therefore take the event study approach in this paper and examine if the world market risk premium is particularly higher on prescheduled US macroeconomic announcement days. By focusing on days with prescheduled macroeconomic announcements, we can draw causal inferences about the effects of the state of global economy on the world market risk premium, because our results will less likely be driven by reverse causality and/or confounding factors. For instance, if the world market risk premium is on average higher on prescheduled unemployment announcement days, it is unlikely because the government agency (such as the US Bureau of Labor Statistics) determines the news releases schedule based on the expectations of the (higher) world market risk premium well ahead of time. Instead, such evidence is more consistent with the notion that investors demand higher returns on macroeconomic announcement days because investors are exposed to more systematic risk at such times. After all, important information about the state of the economy is revealed on macroeconomic announcement days. Motivated by recent macroeconomic-announcement literature (e.g., Savor and Wilson, 2013, 2014) and in line with Harvey (1991), we focus on prescheduled US unemployment, inflation and interest rate announcements. US macroeconomic announcements may reveal important information about the state of the global economy, because the US is an important trading partner for many economies (Rapach et al., 2013). Empirically, a growing literature suggests that US macroeconomic announcements have significant impact on global financial markets. For instance, Wongswan (2006) finds that US macroeconomic announcements affect the Korean and Thailand equity markets. Andersen et al. (2007), Faust et al. (2007), and Evans and Lyons (2008) document the reaction of exchange rates to US macroeconomic announcements. Ammer et al. (2010) show that “foreign firms on average are roughly as sensitive to U.S. monetary policy as U.S. firms” (p. 179). Hausman and Wongswan (2011) find that US monetary policy announcements have significant effects on foreign equity indexes, short- and long-term interest rates, and exchange rates in 49 countries. To test if the world market risk premium is higher on prescheduled US macroeconomic announcement days, we apply the Savor and Wilson (2014) methodology to US firms as well as foreign firms cross-listed in the US. The motivation of using cross-listed foreign firms is to circumvent the nonsynchronous trading problem in previous studies (Ammer et al., 2010). Our findings can be easily summarized. Although there is no significant relationship between International CAPM beta and mean excess returns for US stocks as well as for cross-listed foreign stocks on nonannouncement days, the positive risk–return tradeoff predicted by the International CAPM holds robustly on US macroeconomic announcement days. In addition, we find that there is also a significant risk–return relationship at the market level on US macroeconomic announcement days in the global equity market. Our results therefore suggest that the world market risk premium depends on the state of global economy. Our study adds to the International CAPM literature (e.g., Harvey, 1991) by providing new evidence that there is a causal relationship from the state of the global economy to the world market risk premium. Our study also adds to the macroeconomic-announcements literature. Specifically, in the language of factor models, while previous announcements studies (e.g., Ammer et al., 2010; Hausman and Wongswan, 2011) focus on the exposure of US and foreign securities to global market risk, we focus on the premium of global market risk. Our findings have important theoretical as well as practical implications. In terms of theoretical implications, our findings suggest that future international asset-pricing models should explore the mechanisms through which the state of the global economy affects the world market risk premium. In terms of practical implications, strengthening Rapach et al. (2013), our findings imply that in capital budgeting, portfolio evaluation, investment, and risk analysis

D. Du, O. Hu / Journal of International Money and Finance 58 (2015) 75–97

77

decisions, an international asset-pricing model that explicitly takes into account the role of the US may be more informative. The remainder of the paper is organized as follows: Section 2 discusses our data; Section 3 documents the US macroeconomic announcement effects in the global equity market; Section 4 reports asset-pricing test results; Section 5 provides further evidence; Section 6 concludes the paper with a brief summary.

2. Data Our sample period begins on January 2nd, 1974, because dollar exchange rates began floating in 1973 (Bartov et al., 1996), and we need one year of data to construct our test assets. Our sample period ends on December 30th, 2013, which is dictated by the availability of daily individual stock returns data from CRSP. For the US sample, we focus on common stocks with CRSP share codes 10 or 11. For the foreign sample, we use American Depository Receipts (CRSP share codes 30 or 31) and common stocks of firms incorporated outside the US (CRSP share code 12). We retrieve country code as well as accounting data from Compustat. We use the world market Index from Datastream, TOTMKWD (RI), as our proxy for the world market equity index. Following previous studies (e.g., Fama and French, 1998), the US risk-free rate is employed as the proxy for the world risk-free rate. Following Savor and Wilson (2013, 2014), we focus on US inflation, unemployment and interest rate announcements. Inflation and unemployment announcement dates are from the US Bureau of Labor Statistics. Producer Price Index (PPI) announcements instead of Consumer Price Index (CPI) announcements are used, since PPI is released earlier in our sample. The dates for the US Federal Open Market Committee (FOMC) scheduled interest rate announcements are from the US Federal Reserve. Unscheduled FOMC meetings are not included in the sample. To examine the risk–return relationship across different types of trading days, we use daily data as in Savor and Wilson (2013, 2014). A major challenge in international asset pricing with daily data is the nonsynchronous trading problem. This problem is particularly troublesome in our setting. For instance, while inflation and unemployment announcements are released at 8:30 am EST when European markets are still open, interest rate announcements are released at 2:15 pm EST after European markets are closed. To circumvent this problem, following Ammer et al. (2010), we use foreign stocks cross-listed in the US that trade contemporaneously with US stocks to capture the impact of US macroeconomic announcements on foreign stocks. The International CAPM assumes an integrated global market. Since developed markets are more integrated (e.g., Bekaert et al., 2011; Campbell and Hamao, 1992), we focus on 21 developedmarkets. They are Australia, Austria, Belgium, Canada, Denmark, Finland, France, Germany, Hong Kong, Israel, Italy, Japan, Netherlands, New Zealand, Norway, Portugal, Singapore, Spain, Sweden, United Kingdom, and United States. Panel A of Table 1 shows country representation of our sample. Column “Firm” and “Obs” show the number of firms and the number of (daily) observations in each country, respectively. As we can see, seven out of 20 non-US countries have less than 10 firms in our sample, suggesting that it is infeasible to study each non-US country in isolation. Therefore, in empirical tests, we study all non-US countries as a group to understand the impact of US macroeconomic announcements on foreign equity. If the world market risk is priced in the global equity market, mean excess returns of assets should vary systematically with the exposure to this factor. Therefore, in the same spirit of Savor and Wilson (2014), we construct 10 value-weighted beta-sorted portfolios for the US sample and 10 valueweighted beta-sorted portfolios for the foreign sample as our base test assets. More specifically, at the beginning of a month, the International CAPM is estimated to obtain the beta (βi) for a stock, with the prior one year of daily data.

ri,t = α i + βi rW ,t + ε i,t

(1)

where ri,t is the excess return on stock i on day t, rW,t is the daily excess return on the world market index, and εi,t is the disturbance. After obtaining individual stocks’ beta, we rank US stocks into 10

78

D. Du, O. Hu / Journal of International Money and Finance 58 (2015) 75–97

Table 1 Summary statistics. Panel A:Country representation Country

Firms

Obs

Country

Firms

Obs

Australia Austria Belgium Canada Denmark Finland France Germany Hong Kong Israel

38 1 6 426 6 7 38 35 14 148

101,753 1627 9641 1,079,288 17,009 18,401 88,235 68,849 30,696 396,845

Italy Japan Netherlands New Zealand Norway Portugal Singapore Spain Sweden United Kingdom United States

17 38 63 8 10 3 9 13 18 164 22,421

49,674 225,639 152,706 11,157 19,201 10,027 24,407 41,862 46,859 378,928 56,803,863

Panel B: Summary statistics for beta-sorted portfolios US beta-sorted portfolios

Low 2 3 4 5 6 7 8 9 High

Foreign beta-sorted portfolios

Average beta

Percentage positive

Percentage significant at the 10% level

Size ($1000)

Average beta

Percentage positive

Percentage significant at the 10% level

Size ($1000)

−1.16 0.14 0.32 0.48 0.62 0.76 0.93 1.13 1.40 2.45

0.28 0.78 0.97 1.00 1.00 1.00 1.00 1.00 1.00 1.00

0.08 0.21 0.43 0.58 0.70 0.79 0.85 0.89 0.92 0.93

171,872 904,764 1,458,331 1,744,349 1,743,336 1,770,598 1,670,941 1,718,184 1,734,877 1,257,343

−3.37 0.29 0.47 0.62 0.76 0.91 1.08 1.27 1.54 2.21

0.44 0.86 0.99 1.00 1.00 1.00 1.00 1.00 1.00 1.00

0.17 0.42 0.62 0.75 0.84 0.90 0.93 0.95 0.96 0.96

458,741 1,230,037 1,488,317 1,776,425 1,730,030 1,742,676 1,856,692 1,800,737 1,853,526 1,411,108

The table shows summary statistics for our base test assets.

beta-sorted portfolios and foreign stocks into 10 beta-sorted portfolios, respectively. These portfolios are held for one month and rebalanced monthly as in Savor and Wilson (2014). Panel B of Table 1 shows the relevant summary statistics of the beta-sorted portfolios. It seems that US and non-US portfolios are comparable. For instance, the average size of the US beta-sorted portfolios is $1,417,460,000, while that of the foreign beta-sorted portfolios is $1,534,829,000. 3. US macroeconomic announcements and global equity returns If US macroeconomic announcements reveal important information about the state of the global economy, global investors take more systematic risk on US macroeconomic announcement days and should demand higher returns. Therefore, we start by examining mean excess returns of our test assets across two types of trading days, namely US announcement and non-announcement days. To estimate the mean (excess) return of an asset, we regress time-series (excess) returns on a constant. To estimate mean (excess) returns separately for announcement and non-announcement days, we adopt the regression approach of Cooper et al. (2004). Essentially, we regress time-series (excess) returns on an announcement-day dummy and a non-announcement-day dummy, with no intercept. In all cases, the t-statistics are based on Newey–West HAC standard errors with the lag parameter set equal to 5 for our daily data. The results are reported in Panel A of Table 2. Section “US” shows the results for the US sample, while Section “Foreign” presents those for the foreign sample. Columns “All”, “A Day”, and “N Day” show mean excess returns on all days, those on announcement days, and those on non-announcement days, respectively. Eighteen out of 20 beta-sorted portfolios have higher mean excess returns on

D. Du, O. Hu / Journal of International Money and Finance 58 (2015) 75–97

79

Table 2 Mean excess returns and cumulative log returns. Panel A: Mean excess returns

Low 2 3 4 5 6 7 8 9 High

US beta-sorted portfolios

Foreign beta-sorted portfolios

All

A Day

N Day

All

A Day

N Day

4.08 (4.72) 3.65 (5.14) 3.82 (5.12) 3.38 (4.19) 4.01 (4.52) 3.54 (3.63) 3.72 (3.42) 3.67 (2.93) 3.33 (2.22) 3.36 (1.67)

7.03 (3.34) 4.59 (2.35) 6.32 (2.98) 4.28 (1.76) 5.27 (1.91) 5.58 (1.82) 7.52 (2.12) 10.32 (2.47) 12.97 (2.58) 16.14 (2.54)

3.68 (4.04) 3.52 (4.73) 3.48 (4.39) 3.26 (3.86) 3.84 (4.12) 3.27 (3.18) 3.21 (2.83) 2.77 (2.16) 2.03 (1.30) 1.62 (0.78)

3.45 (2.30) 2.78 (2.47) 4.09 (3.75) 4.33 (3.56) 3.22 (2.40) 3.25 (2.31) 4.64 (3.34) 2.78 (1.73) 1.14 (0.59) 4.46 (1.95)

5.43 (1.37) 2.18 (0.68) 6.82 (2.15) 9.37 (2.69) 2.64 (0.71) 7.93 (1.86) 10.91 (2.57) 3.93 (0.83) 12.13 (2.08) 24.38 (3.53)

3.17 (1.99) 2.86 (2.42) 3.72 (3.19) 3.65 (2.87) 3.30 (2.32) 2.61 (1.78) 3.79 (2.61) 2.62 (1.56) −0.35 (−0.18) 1.75 (0.73)

Panel B: Cumulative log return

Low 2 3 4 5 6 7 8 9 High

US beta-sorted portfolios

Foreign beta-sorted portfolios

All

A Day

N Day

All

A Day

N Day

3.73 3.38 3.51 3.03 3.56 3.01 3.06 2.82 2.12 1.37

0.80 0.51 0.71 0.46 0.57 0.59 0.80 1.09 1.35 1.62

2.94 2.86 2.80 2.56 3.00 2.41 2.26 1.72 0.77 −0.25

2.33 2.18 3.46 3.59 2.35 2.30 3.62 1.53 −0.63 1.89

0.53 0.19 0.73 1.02 0.22 0.81 1.16 0.31 1.19 2.53

1.80 1.99 2.73 2.57 2.13 1.49 2.46 1.22 −1.82 −0.64

The table shows mean excess returns and cumulative log excess returns for our base test assets. “All”, “A Day”, and “N Day” refer to all days, announcement days, and non-announcement days, respectively.

announcement days. For the US sample, the average mean excess return on announcement days is 7.77 basis points, while that on non-announcement days is 2.60 basis points. For the foreign sample, the average mean excess return on announcement days is 8.40 basis points, while that on nonannouncement days is 2.21 basis points. To put these numbers into perspective, the mean excess return of the US beta-sorted portfolios in Savor and Wilson (2014) is 8.28 basis points on announcement days and 1.16 on non-announcement days over a longer sample period from 1964 to 2011. Thus, our results suggest that US macroeconomic announcements affect not only US but also foreign stocks. Furthermore, the impact on foreign stocks is not smaller than that on US stocks, implying that US macroeconomic announcements do reveal important information about the state of the global economy. Following Savor and Wilson (2014), we calculate the cumulative log excess returns of our betasorted portfolios earned on US announcement and non-announcement days. The idea is to demonstrate the empirical importance of US macroeconomic announcement days. As we can see from Panel B of

80

D. Du, O. Hu / Journal of International Money and Finance 58 (2015) 75–97

Table 2, although US macroeconomic announcement days are only about 12% of trading days, our test assets earn a disproportionate fraction of their total excess returns at such times. For the US sample, the average cumulative log excess return earned on announcement days is 0.85, while that on nonannouncement days is 2.11. Therefore, the US beta-sorted portfolios earn nearly 30% of their total excess returns on 12% of trading days with macroeconomic announcements. For the foreign sample, the average cumulative log excess return earned on US announcement days is 0.87, while that on non-announcement days is 1.39. Thus, the foreign beta-sorted portfolios earn nearly 40% of their total excess returns on about 12% of trading days. Thus, the evidence suggests that US macroeconomic announcement play an important role in international asset pricing.

4. International CAPM and US macroeconomic announcements Section 3 shows that mean excess returns of US and foreign beta-sorted portfolios are higher on US macroeconomic announcement days, suggesting that US macroeconomic announcements may reveal important information about the state of the global economy. In this section, we formally test if the world market risk premium is higher on US macroeconomic announcement days.

4.1. Preliminary tests We first examine if the exposure to world market returns varies with types of trading days. Following Savor and Wilson (2014), we run the following time-series regressions asset by asset over the entire sample period.

ri,t = α i + βi,W rW ,t + βi, ArW ,t × DA_Day,t + ε i,t

(2)

where DA_Day,t is the announcement-day dummy. The t-statistics are based on Newey–West HAC standard errors with the lag parameter set equal to 5 for our daily data. The regression results are reported in Table 3. As we can see, only three out of 20 beta-sorted portfolios have significantly different exposure to world market returns on announcement days. Thus, the exposure to world market returns generally does not vary with types of trading days, suggesting that higher mean excess returns of betasorted portfolios on US macroeconomic announcement days should be mainly due to higher risk premium on the world market factor (i.e., a more positive trade-off between International CAPM beta and mean excess returns). Here, we provide preliminary graphic evidence. Since there are generally no significant differences in exposure to world market returns between two types of trading days, we use the simple International CAPM (i.e., Eq. 1) to estimate International CAPM beta. To visualize the relationship between International CAPM beta and mean excess returns, we plot scatter graphs in Fig. 1. In each graph, the horizontal axis measures the International CAPM beta, while the vertical axis measures the mean excess return. We first focus on the US sample in the first row. Panel A1 shows the relationship between International CAPM beta and mean excess returns on all days for 10 US beta-sorted portfolios. As we can see, the relationship is negative. Panel A2 shows the relationship separately for announcement days (square-shaped points and line) and non-announcement days (diamond-shaped points and line). The non-announcement-day points show a weakly negative relationship between International CAPM beta and mean excess returns, while the announcement-day points show a clear positive relationship. The evidence thus suggests that the risk premium on the world market factor is higher on US macroeconomic announcement days. We repeat our exercises for 10 foreign beta-sorted portfolios and for all 20 beta-sorted portfolios (US and foreign). In Panel B1, on all days, there is a negative relationship between International CAPM beta and mean excess returns for foreign beta-sorted portfolios. In Panel B2, the different relationships between International CAPM beta and mean excess returns on announcement and nonannouncement days again suggest that the risk premium on the world market factor is higher on US macroeconomic announcement days. Panels C1 and C2 report similar results for all 20 beta-sorted portfolios.

D. Du, O. Hu / Journal of International Money and Finance 58 (2015) 75–97

81

Table 3 Exposure of beta-sorted portfolios to the world market index. Panel A: Exposure of US portfolios

Low 2 3 4 5 6 7 8 9 High

Panel B: Exposure of Foreign portfolios 2

Constant

βι,W

βι,Α

R

3.16 (4.13) 2.67 (4.54) 2.59 (4.60) 1.88 (3.45) 2.23 (3.96) 1.50 (2.49) 1.34 (2.09) 0.93 (1.29) 0.06 (0.07) −0.73 (−0.57)

0.41 (11.35) 0.42 (17.35) 0.52 (26.43) 0.63 (33.04) 0.75 (39.16) 0.85 (46.67) 0.98 (54.01) 1.12 (43.97) 1.34 (39.70) 1.68 (34.15)

−0.08 (−1.92) −0.04 (−0.84) −0.00 (−0.06) 0.03 (0.62) 0.03 (0.51) 0.05 (0.89) 0.11 (1.83) 0.17 (2.37) 0.20 (2.31) 0.22 (2.05)

0.20 0.29 0.38 0.46 0.51 0.53 0.56 0.56 0.55 0.52

Constant

βι,W

βι,Α

R2

2.29 (1.62) 1.61 (1.59) 2.56 (2.83) 2.59 (2.59) 1.20 (1.14) 0.83 (0.80) 2.01 (2.08) −0.35 (−0.33) −2.55 (−1.87) 0.06 (0.04)

0.53 (8.46) 0.50 (21.34) 0.67 (23.17) 0.74 (31.08) 0.86 (29.33) 1.01 (46.95) 1.09 (46.12) 1.30 (52.06) 1.53 (48.58) 1.82 (42.90)

−0.14 (−1.82) −0.03 (−0.49) −0.06 (−0.96) −0.00 (−0.05) −0.01 (−0.22) 0.06 (0.87) 0.10 (1.64) 0.12 (1.72) 0.12 (1.33) 0.20 (1.89)

0.08 0.15 0.26 0.28 0.31 0.40 0.44 0.49 0.47 0.46

We run the following time-series regressions asset by asset over the entire sample period. ri ,t = α i + βi ,W rW ,t + βi , ArW ,t × DA_Day ,t + ε i ,t

where ri,t is the excess return on portfolio i on day t, rW,t is the daily excess return on the world market index, DA_Day,t is announcement-day dummy, and εI,t is the disturbance. The t-statistics are based on Newey– West HAC standard errors with the lag parameter set equal to 5 for our daily data.

4.2. Asset-pricing test results To formally examine if the premium on the world market factor is higher on US macroeconomic announcement days, we use the standard Fama and MacBeth (1973) two-pass regression. Since the exposure to world market returns is generally not significantly different between two types of trading days, our first-pass regression is based on Eq. (1). In the robustness check section, we estimate beta separately for announcement and non-announcement days, and find similar results. For the second stage, we follow Savor and Wilson (2014) and estimate the risk premium separately for announcement and non-announcement days. Specifically, for each day t, we estimate the following cross-sectional regressions:

ri A,t = γ 0A + γ WA βˆ i ,t + ei ,t

(3a)

N ˆ riN,t = γ 0N + γ W βi ,t + ei ,t

(3b)

and

where ri ,At and riN,t are excess returns of portfolio i on announcement and non-announcement days, and βˆ i ,t is portfolio i’s exposure to world market returns estimated with prior one year of daily data from the first-pass regression of Eq. (1). Again, following Savor and Wilson (2014), we estimate the risk premium as the average across time of the cross-sectional estimates, and the standard error equals the time-series standard deviation of the cross-sectional estimates divided by the square root of the respective sample lengths. We then test whether the risk premium is significantly different on announcement days by applying a simple t-test for a difference in means.

82

D. Du, O. Hu / Journal of International Money and Finance 58 (2015) 75–97

Panel A1: US beta-sorted portfolios on all days

Panel A2: US beta-sorted portfolios on announcement and non-announcement days

3.7 3.6 3.5 3.4 3.3 3.2 3.1 3.0 2.9 2.8 0.25

0.50

0.75

1.00

1.25

1.50

1.75

16 14 12 10 8 6 4 2 0 0.25

Panel B1: Foreign beta-sorted portfolios on all days

0.50

0.75

1.00

1.25

1.50

1.75

Panel B2: Foreign beta-sorted portfolios on announcement and non-announcement days

4.0

30

3.5

25

3.0

20

2.5

15

2.0

10

1.5

5

1.0

0

0.5

-5

0.0

0.5

1.0

1.5

0.0

2.0

Panel C1: US and foreign beta-sorted portfolios on all

1.0

1.5

2.0

Panel C2: US and foreign beta-sorted portfolios on

days

announcement and non-announcement days

4.0

30

3.5

25

3.0

20

2.5

15

2.0

10

1.5

5

1.0 0.5 0.25

0.5

0 0.50

0.75

1.00

1.25

1.50

1.75

2.00

-5 0.25

0.50

0.75

1.00

1.25

1.50

1.75

2.00

Fig. 1. Relationship between International CAPM beta and mean excess returns for base test assets. The figure shows the relationship between International CAPM beta and mean excess returns for our base test assets. The horizontal axis measures the International CAPM beta, while the vertical axis measures the mean excess return. In each panel, the scatter plot on the left hand side shows the relationship between International CAPM beta and mean excess returns on all days, while the graph on the right hand side shows that same relationship separately for announcement days (square-shaped points and line) and non announcement days (diamond-shaped points and line).

The two-pass regression results are reported in Panel A of Table 4. Columns “All”, “A Day”, “N Day”, and “Diff” contain premium estimates on all days, those on announcement days, those on nonannouncement days, and premium differences between announcement and non-announcement days, respectively. Section “US” shows the results based on 10 US beta-sorted portfolios. If we do not distinguish announcement from non-announcement days, the world market factor on all days carries a statistically insignificant risk premium of 0.78 basis points per day with a t-statistic of 0.49. Interestingly, on announcement days, the world market factor has a significantly positive risk premium of 11.44 basis points per day with a t-statistic of 2.34. In contrast, on non-announcement days, it has a statistically insignificant risk premium of −0.67 basis points per day with a t-statistic of −0.40. The premium difference is 12.11 basis points per day with a t-statistic of 2.35. Furthermore, the intercept is insignificantly different from zero on announcements days but significantly positive on non-announcement days. Section “Foreign” presents the results based on 10 foreign beta-sorted portfolios. On all days, the world market factor carries an insignificant risk premium of −0.35 basis points per day with a t-statistic of −0.23. However, as soon as we distinguish announcement from non-announcement days, the results change significantly. Specifically, on announcement days, the world market factor carries a significantly

D. Du, O. Hu / Journal of International Money and Finance 58 (2015) 75–97

83

Table 4 Two-pass regression results for beta-sorted portfolios. Panel A: Value weighted portfolios US (10 portfolios)

Constant rW R2

US (10 portfolios)

All

A-Day

N-Day

Diff

All

A-Day

N-Day

Diff

3.26 (4.37) 0.78 (0.49) 0.41

−0.40 (−0.19) 11.44 (2.34) 0.42

3.75 (4.70) −0.67 (−0.40) 0.41

−4.15 (−1.89) 12.11 (2.35)

9.47 (19.53) −3.16 (−2.19) 0.46

10.06 (7.50) 6.08 (1.40) 0.48

9.39 (18.06) −4.42 (−2.89) 0.46

0.67 (0.47) 10.50 (2.27)

Foreign (10 portfolios)

Constant rW R2

Foreign (10 portfolios)

All

A-Day

N-Day

Diff

All

A-Day

N-Day

Diff

3.83 (3.36) −0.35 (−0.23) 0.17

−0.54 (−0.17) 10.09 (2.24) 0.19

4.42 (3.63) −1.77 (−1.07) 0.17

−4.96 (−1.43) 11.86 (2.47)

9.86 (9.20) −4.29 (−2.75) 0.17

10.31 (3.45) 5.39 (1.18) 0.18

9.80 (8.54) −5.60 (−3.38) 0.16

0.51 (0.16) 11.00 (2.27)

US and foreign (20 portfolios)

Constant rW R2

Panel B: Equal weighted portfolios

US and foreign (20 portfolios)

All

A-Day

N-Day

Diff

All

A-Day

N-Day

Diff

3.68 (4.70) −0.07 (−0.05) 0.19

0.27 (0.12) 9.02 (2.30) 0.20

4.15 (4.97) −1.30 (−0.92) 0.19

−3.87 (−1.61) 10.32 (2.47)

10.05 (15.95) −4.19 (−3.37) 0.18

10.30 (5.76) 5.14 (1.39) 0.19

10.01 (14.88) −5.45 (−4.13) 0.18

0.28 (0.15) 10.60 (2.70)

We use the standard Fama and MacBeth (1973) two-pass regression. Since the exposure to the world index is generally not significantly different on announcement days, our first-pass regression is based on the following equation. ri ,t = α i + βirW ,t + ε i ,t

where ri,t is the excess return on asset i on day t, and rW,t is the daily excess return on the world market index. For the second stage, we estimate risk premium separately for announcement and nonannouncement days. Specifically, for each period t, we estimate the following cross-sectional regressions: ri A,t = γ 0A + γ WA βˆ i ,t + ei ,t

and N ˆ riN,t = γ 0N + γ W βi ,t + ei ,t

where ri A,t and riN,t are the excess returns of test asset i on announcement and non-announcement days, and βˆ i ,t is test asset i’s exposures to the world index estimated with prior one year of daily data from the first-pass regression. We estimate the risk premium as the average across time of the crosssectional estimates, and the standard error equals the time-series standard deviation of the crosssectional estimates divided by the square root of the respective sample lengths. We then test whether the risk premium is significantly different on announcement days by applying a simple t-test for a difference in means. Columns “All”, “A Day”, “N Day”, and “Diff” contain premium estimates on all days, those on announcement days, those on non-announcement days, and premium differences between announcement and non-announcement days, respectively. Panels A and B show the results for valueweighted and equal-weighted beta-sorted portfolios, respectively.

positive risk premium of 10.09 basis points per day with a t-statistic of 2.24. In contrast, on nonannouncement days, its premium is −1.77 basis points with a t-statistic of −1.07. Moreover, the intercept is insignificantly different from zero on announcements days but significantly positive on nonannouncement days. Panel C shows the results based on all 20 beta-sorted portfolios. For 20 beta-sorted portfolios, the world market factor has a significantly positive risk premium of 9.02 basis points per day with a t-statistic of 2.30 on announcement days. In contrast, its premium on non-announcement days is −1.30 basis points

84

D. Du, O. Hu / Journal of International Money and Finance 58 (2015) 75–97

per day with a t-statistic of −0.92. All the evidence is consistent with the graphic evidence in Fig. 1, suggesting that the world market risk premium is higher on US macroeconomic announcement days. We emphasize that the evidence in Panel A of Table 4 suggests that there is a causal relationship from the state of the global economy to the world market risk premium, because our event-study results are not likely due to reverse causality and/or confounding factors. For instance, it is unlikely that government agencies (such as FOMC) determine the news release schedules based on the expectations of the world market risk premium. Instead, our event-study evidence is more consistent with the idea that investors demand higher returns on prescheduled macroeconomic announcement days because important information about the state of the global economy is revealed at such times. 4.3. Robustness checks 4.3.1. Equal-weighted beta-sorted portfolios We first repeat our exercises with equal-weighted beta-sorted portfolios. Such portfolios are more dominated by the effects of small stocks and may suffer from various biases associated with small stocks (e.g., nonsynchronous trading). The results are reported in Panel B of Table 4. Although the results on announcement days are weaker, the general pattern is the same. For instance, while the world market factor carries a positive risk premium of 5.14 basis points per day with a t-statistic of 1.39 on announcement days, its premium is −5.45 basis points with a t-statistic of −4.13 on non-announcement days. The premium difference is 10.60 basis points per day with a t-statistic of 2.70. 4.3.2. Expanded sets of test assets We next expand our set of test assets. For the US sample, we obtain 10 size, 10 book-to-market, and 10 industry portfolios from Kenneth French’s website.2 For the foreign sample, along the same line, we construct 10 size, 10 book-to-market, and 10 industry portfolios.3 Because these portfolios are formed according to very different characteristics, as Savor and Wilson (2014) point out, this is a stringent robustness test. In Panel A of Table 5, we report the results based on beta and size portfolios. Section “US (20 portfolios)” shows the results for the US sample. While the world market factor has a significantly positive risk premium of 10.96 basis points per day with a t-statistic of 2.33 on announcement days, its premium is −0.27 basis points with a t-statistic of −0.17 on non-announcement days. The premium difference is 11.22 basis points per day with a t-statistic of 2.26. The results are similar for the foreign sample (in Section “Foreign (20 portfolios)”) as well as for the combined US and foreign sample (in Section “US and Foreign (40 portfolios)”). We present the results based on beta, size, and book-to-market portfolios in Panel B of Table 5, and those based on beta, size, book-to-market, and industry portfolios in Panel C, respectively. In all cases, our results are robust. For instance, in Panel C, for the combined US and foreign sample, there are four different types of test assets formed based on very different characteristics and 80 test assets in total. Again, while the world market factor has a significantly positive risk premium of 8.33 basis points per day with a t-statistic of 2.42 on US macroeconomic announcement days, its premium is 0.26 basis points with a t-statistic of 0.20 on non-announcement days. The premium difference is 8.07 basis points per day with a t-statistic of 2.20. Fig. 2 plots the relationship between International CAPM beta and mean excess returns based on the expanded sets of test assets. For the US and foreign combined sample in the third row, Panel C1 shows the relationship between International CAPM beta and mean excess returns on all days, and Panel C2 shows the same relationship separately for announcement days (square-shaped points and line) and non-announcement days (diamond-shaped points and

2

We thank Fama and French for making these data available at http://mba.tuck.dartmouth.edu/pages/faculty/ken.french/. At the end of a year, we rank cross listed foreign stocks into 10 size and 10 book-to-market portfolios, and we also group those stocks into 10 industry portfolios. We rebalance all those portfolios on an annual basis. For each foreign stock, size (market value) is calculated as closed price in December times shares outstanding, book value is calculated with fiscal year-end data as subtracting total liabilities and preferred stock from total assets and then adding deferred taxes and convertible debt (if preferred stock is missing, we replace it with redemption value of preferred stock). In forming the industry portfolios, we borrow the definitions of 10 industry portfolios from Kenneth French’s data library. 3

D. Du, O. Hu / Journal of International Money and Finance 58 (2015) 75–97

85

Table 5 Two-pass regression results for expanded sets of test assets. Panel A: Beta and size portfolios US (20 portfolios)

α rW R2

US (30 portfolios)

rW R2

rW R2

US (40 portfolios)

A-Day

N-Day

Diff

All

A-Day

N-Day

Diff

All

A-Day

N-Day

Diff

3.19 (4.42) 1.07 (0.70) 0.35

0.77 (0.40) 10.96 (2.33) 0.36

3.51 (4.53) −0.27 (−0.17) 0.35

−2.74 (−1.31) 11.22 (2.26)

3.46 (4.84) 0.75 (0.50) 0.29

0.73 (0.38) 10.59 (2.33) 0.31

3.83 (4.97) −0.59 (−0.37) 0.29

−3.11 (−1.51) 11.18 (2.32)

3.43 (4.71) 0.66 (0.45) 0.25

0.99 (0.50) 9.82 (2.23) 0.26

3.76 (4.80) −0.59 (−0.38) 0.25

−2.77 (−1.31) 10.41 (2.23)

Foreign (30 portfolios)

Foreign (40 portfolios)

All

A-Day

N-Day

Diff

All

A-Day

N-Day

Diff

All

A-Day

N-Day

Diff

6.70 (6.37) −1.95 (−1.35) 0.12

1.20 (0.41) 10.66 (2.55) 0.14

7.45 (6.61) −3.66 (−2.37) 0.12

−6.25 (−2.00) 14.33 (3.21)

6.03 (5.83) −1.42 (−1.04) 0.11

1.14 (0.39) 9.55 (2.39) 0.12

6.70 (6.06) −2.91 (−2.00) 0.11

−5.56 (−1.77) 12.46 (2.93)

5.34 (5.17) −0.71 (−0.52) 0.10

1.35 (0.46) 8.77 (2.22) 0.12

5.88 (5.33) −1.99 (−1.38) 0.10

−4.53 (−1.45) 10.76 (2.56)

US and foreign (40 portfolios)

α

Panel C: Beta, size, BM and industry portfolios

All

Foreign (20 portfolios)

α

Panel B: Beta, size and BM portfolios

US and foreign (60 portfolios)

US and foreign (80 portfolios)

All

A-Day

N-Day

Diff

All

A-Day

N-Day

Diff

All

A-Day

N-Day

Diff

5.00 (6.75) −0.46 (−0.37) 0.14

1.59 (0.76) 9.49 (2.63) 0.15

5.46 (6.90) −1.81 (−1.37) 0.14

−3.87 (−1.72) 11.30 (2.94)

4.72 (6.26) −0.37 (−0.31) 0.13

1.86 (0.85) 7.95 (2.29) 0.14

5.11 (6.36) −1.50 (−1.18) 0.13

−3.25 (−1.39) 9.45 (2.55)

4.33 (5.70) 0.03 (0.03) 0.12

1.99 (0.91) 7.49 (2.18) 0.13

4.64 (5.74) −0.98 (−0.77) 0.12

−2.65 (−1.14) 8.47 (2.31)

We use the standard Fama and MacBeth (1973) two-pass regression. Since the exposure to the world index is generally not significantly different on announcement days, our first-pass regression is based on the following equation. ri ,t = α i + βirW ,t + ε i ,t

where ri,t is the excess return on asset i on day t, and rW,t is the daily excess return on the world market index. For the second stage, we estimate risk premium separately for announcement and non-announcement days. Specifically, for each period t, we estimate the following cross-sectional regressions: ri A,t = γ 0A + γ WA βˆ i ,t + ei ,t

and N ˆ riN,t = γ 0N + γ W βi ,t + ei ,t

where ri A,t and riN,t are the excess returns of test asset i on announcement and non-announcement days, and βˆ i ,t is test asset i’s exposures to the world index estimated with prior one year of daily data from the first-pass regression. We estimate the risk premium as the average across time of the cross-sectional estimates, and the standard error equals the time-series standard deviation of the cross-sectional estimates divided by the square root of the respective sample lengths. We then test whether the risk premium is significantly different on announcement days by applying a simple t-test for a difference in means. Columns “All”, “A Day”, “N Day”, and “Diff” contain premium estimates on all days, those on announcement days, those on nonannouncement days, and premium differences between announcement and non-announcement days, respectively.

line). The non-announcement-day points show a weakly negative relationship between International CAPM beta and mean excess returns, while the announcement-day points show a clear positive relationship. Thus, all the evidence reinforces the notion that the world market risk premium is higher on US macroeconomic announcement days. 4.3.3. Conditional beta Section 4.1 shows that the exposure to the world market risk generally does not vary with types of trading days. Therefore, we use unconditional beta in the two-pass regressions. In this section, we allow the exposure to vary with types of trading days. Essentially, we divide our sample into two subsamples, the announcement-day subsample and the non-announcement-day subsample, and apply

86

D. Du, O. Hu / Journal of International Money and Finance 58 (2015) 75–97

Panel A1: US beta, size, book-to-market and industry

Panel A2: US beta, size, book-to-market and industry

portfolios on all days

portfolios on announcement and non-announcement days

5.5

17.5

5.0

15.0

4.5

12.5 10.0

4.0

7.5

3.5

5.0

3.0

2.5 0.0

2.5 0.4

0.6

0.8

1.0

1.2

1.4

1.6

0.4

1.8

0.6

0.8

1.0

1.2

1.4

1.6

1.8

Panel B1: Foreign beta, size, book-to-market and

Panel B2: Foreign beta, size, book-to-market and

industry portfolios on all days

industry portfolios on announcement and non-announcement days

20.0 17.5 15.0 12.5 10.0 7.5 5.0 2.5 0.0 0.50

25 20 15 10 5 0

0.75

1.00

1.25

1.50

1.75

2.00

-5 0.50

0.75

1.00

1.25

1.50

1.75

Panel C1: US and foreign beta, size, book-to-market

Panel C2: US and foreign beta, size, book-to-market

and industry portfolios on all days

and industry portfolios on announcement and

2.00

non-announcement days 20.0

25

17.5

20

15.0 12.5

15

10.0

10

7.5

5

5.0

0

2.5 0.0 0.25

0.50

0.75

1.00

1.25

1.50

1.75

-5 2.00 0.25

0.50

0.75

1.00

1.25

1.50

1.75

2.00

Fig. 2. Relationship between International CAPM beta and mean excess returns for the expanded set of test assets. The figure shows the relationship between International CAPM beta and mean excess returns for the expanded set of test assets. The horizontal axis measures the International CAPM beta, while the vertical axis measures the mean excess return. In each panel, the scatter plot on the left hand side shows the relationship between International CAPM beta and mean excess returns on all days, while the graph on the right hand side shows that same relationship separately for announcement days (square-shaped points and line) and non announcement days (diamond-shaped points and line).

the two-pass regression methodology independently in each sample. Thus, the exposure to the world market risk on announcement/non-announcement days is estimated with only announcement/nonannouncement sample. The test assets are the base beta-sorted portfolios. The results are reported in Panel A of Table 6, and are similar as those based on unconditional beta. For instance, for the combined US and foreign sample, while the world market factor carries a significantly positive risk premium of 12.70 basis points per day with a t-statistic of 2.85 on US macroeconomic announcement days, its premium is −1.26 basis points with a t-statistic of −0.87 on non-announcement days. 4.3.4. Robust Fama and MacBeth (1973) cross-sectional regressions The simple Fama and MacBeth (1973) two-pass regression methodology used in Savor and Wilson (2014) can easily help to test the premium differences between announcement and non-announcement

D. Du, O. Hu / Journal of International Money and Finance 58 (2015) 75–97

87

Table 6 Two-pass regression results based on conditional beta and robust regressions. Panel A: Conditional beta

Panel B: Robust regressions

US (10 portfolios)

Constant rW R2

US (10 portfolios)

All

A-Day

N-Day

All

A-Day

N-Day

3.26 (4.37) 0.78 (0.49) 0.41

−2.53 (−1.00) 14.85 (2.94) 0.41

3.65 (4.52) −0.50 (−0.29) 0.41

3.54 (4.70) −0.37 (−0.28) 0.28

1.76 (0.84) 6.39 (1.74) 0.79

3.88 (4.81) −1.47 (−1.04) 0.80

Foreign (10 portfolios)

Constant rW R2

Foreign (10 portfolios)

All

A-Day

N-Day

All

A-Day

N-Day

3.83 (3.36) −0.35 (−0.23) 0.17

−0.33 (−0.09) 12.24 (2.57) 0.21

4.45 (3.62) −2.00 (−1.18) 0.17

3.58 (2.88) −0.64 (−0.43) −0.03

−1.12 (−0.34) 9.27 (2.33) 0.51

4.56 (3.40) −2.37 (−1.47) 0.53

US and foreign (20 portfolios)

Constant rW R2

US and foreign (20 portfolios)

All

A-Day

N-Day

All

A-Day

N-Day

3.68 (4.70) −0.07 (−0.05) 0.19

−0.98 (−0.37) 12.70 (2.85) 0.22

4.05 (4.82) −1.26 (−0.87) 0.19

3.59 (4.29) −0.54 (−0.42) 0.06

0.35 (0.15) 7.85 (2.18) 0.58

4.24 (4.74) −1.97 (−1.42) 0.59

In Panel A, we allow the exposure to vary with the types of trading days. Essentially, we divide our sample into two subsamples, the announcement-day subsample and the non-announcement-day subsample, and apply the Savor and Wilson (2014) methodology independently in each sample. In Panel B, we report the results based on the Shanken and Zhou (2007) robust standard errors.

days. However, this methodology may bring in the well-known errors-in-variables (EIV) bias (since this methodology estimates factor loadings in the first pass and using those to obtain risk premiums in the second pass) and misspecification (MIS) biases (since the International CAPM may have model misspecifications). To test if our results are robust to these biases, we modify the Savor and Wilson (2014) methodology. First, we divide our sample into two subsamples, the announcement-day subsample and the non-announcement-day subsample. Second, within each subsample, following the empirical asset-pricing literature (e.g., Du, 2013), we apply the Shanken and Zhou (2007) methodology to obtain EIV- and MIS-robust standard errors. The test assets are the base beta-sorted portfolios. The results are reported in Panel B of Table 6, and are similar as those based on the Savor and Wilson (2014) methodology. For instance, for the combined US and foreign sample, while the world market factor carries a significantly positive risk premium of 7.85 basis points per day with a robust t-statistic of 2.18 on US macroeconomic announcement days, its premium is −1.97 basis points with a robust t-statistic of −1.42 on non-announcement days. 4.3.5. Subsample evidence The integration of the global economy has been particularly significant in recent years. For instance, Evans and Hnatkovska (2014) find “that gross capital flows between industrialized countries (the sum of absolute value of capital inflows and outflows) expanded 300% between 1991 and 2000” (p. 16). Thus, we expect that US macroeconomic announcements may be more informative about the state of the global economy in recent years. To test this conjecture, we divide our sample into two equal-length subsamples (i.e., 1974–1993 and 1994 to 2013) and repeat our exercises. The results for a variety of test assets are reported in Table 7 and are consistent with our conjecture. For instance, on US macroeconomic announcement days, for 80 US and foreign portfolios formed on beta, size, bookto-market and industry, while the world market factor carries a significantly positive risk premium

88

D. Du, O. Hu / Journal of International Money and Finance 58 (2015) 75–97

Table 7 Subsample results. Panel A: 1974–1993

Panel B: 1994–2013

20 beta portfolios

Constant rW R2

20 beta portfolios

All

A-Day

N-Day

Diff

All

A-Day

N-Day

Diff

4.64 (4.28) −0.76 (−0.42) 0.12

4.09 (1.24) 0.48 (0.09) 0.13

4.71 (4.10) −0.91 (−0.47) 0.12

−0.63 (−0.18) 1.39 (0.25)

2.77 (2.46) 0.59 (0.30) 0.25

−3.03 (−0.98) 16.42 (2.89) 0.26

3.60 (2.97) −1.68 (−0.81) 0.25

−6.63 (−1.99) 18.09 (2.99)

40 beta and size portfolios

Constant rW R2

40 beta and size portfolios

All

A-Day

N-Day

Diff

All

A-Day

N-Day

Diff

4.60 (4.39) 0.10 (0.06) 0.10

3.91 (1.29) 3.67 (0.79) 0.11

4.69 (4.20) −0.36 (−0.20) 0.10

−0.78 (−0.24) 4.03 (0.81)

5.38 (5.14) −1.00 (−0.54) 0.18

−0.41 (−0.14) 14.54 (2.69) 0.19

6.21 (5.53) −3.22 (−1.64) 0.18

−6.63 (−2.12) 17.75 (3.09)

60 beta, size and BM portfolios

Constant rW R2

Constant rW R2

60 beta, size and BM portfolios

All

A-Day

N-Day

Diff

All

A-Day

N-Day

Diff

4.81 (4.49) −0.15 (−0.09) 0.11

4.95 (1.53) 1.88 (0.41) 0.11

4.79 (4.21) −0.41 (−0.24) 0.11

0.16 (0.05) 2.29 (0.47)

4.64 (4.37) −0.58 (−0.33) 0.15

−0.81 (−0.27) 13.21 (2.57) 0.16

5.43 (4.77) −2.55 (−1.36) 0.15

−6.24 (−1.95) 15.76 (2.88)

80 beta, size, BM and industry portfolios

80 beta, size, BM and industry portfolios

All

A-Day

N-Day

Diff

All

A-Day

N-Day

Diff

4.79 (4.46) 0.09 (0.06) 0.10

5.42 (1.74) 1.12 (0.25) 0.10

4.71 (4.12) −0.04 (−0.02) 0.10

0.71 (0.21) 1.16 (0.24)

3.88 (3.62) −0.02 (−0.01) 0.14

−0.99 (−0.33) 13.02 (2.54) 0.15

4.58 (4.00) −1.89 (−1.01) 0.14

−5.57 (−1.71) 14.90 (2.73)

We use the standard Fama and MacBeth (1973) two-pass regression. Since the exposure to the world index is generally not significantly different on announcement days, our first-pass regression is based on the following equation. ri ,t = α i + βirW ,t + ε i ,t

where ri,t is the excess return on asset i on day t, and rW,t is the daily excess return on the world market index. For the second stage, we estimate risk premium separately for announcement and nonannouncement days. Specifically, for each period t, we estimate the following cross-sectional regressions: ri A,t = γ 0A + γ WA βˆ i ,t + ei ,t

and N ˆ riN,t = γ 0N + γ W βi ,t + ei ,t

where ri A,t and riN,t are the excess returns of test asset i on announcement and non-announcement days, and βˆ i ,t is test asset i’s exposures to the world index estimated with prior one year of daily data from the first-pass regression. We estimate the risk premium as the average across time of the crosssectional estimates, and the standard error equals the time-series standard deviation of the crosssectional estimates divided by the square root of the respective sample lengths. We then test whether the risk premium is significantly different on announcement days by applying a simple t-test for a difference in means. Columns “All”, “A Day”, “N Day”, and “Diff” contain premium estimates on all days, those on announcement days, those on non-announcement days, and premium differences between announcement and non-announcement days, respectively.

D. Du, O. Hu / Journal of International Money and Finance 58 (2015) 75–97

89

of 13.02 basis points per day with a t-statistic of 2.54 in the 1994–2013 sample, its premium is 1.12 basis points with a t-statistic of 0.25 in the 1974–1993 sample. 4.4. Discussion 4.4.1. Synchronous world market factor The world market factor we use in this paper is the world market index from Datastream, which is not synchronized to the U.S. trading day. In this section, we show that using a synchronous world market factor does not change our results materially. We construct the synchronous world market index in two steps. First, we construct a value-weighted country index in the US dollar for each country in Table 1 based on the firms in our sample. Second, we construct the value-weighted world market index in the US dollar by using the lagged market capitalization of each country in the world market from Datastream as our weights. The results are presented in Panel A of Table 8. Sections “US”, “Foreign”, and “US and Foreign” show the results based on 40 US (beta, industry, size and BM) portfolios, 40 foreign (beta, industry, size and BM) portfolios, and 80 US and foreign (beta, industry, size and BM) portfolios as test assets, respectively. As we can see, the results are generally consistent with those in Panel C of Table 5. For instance, for 80 US and foreign test assets, if we do not distinguish announcement from non-announcement days, the synchronous world market factor on all days carries a statistically insignificant risk premium of −0.22 basis points per day with a t-statistic of −0.17. However, on US macroeconomic announcement days, it has a significantly positive risk premium of 9.76 basis points per day with a t-statistic of 2.55. In contrast, on non-announcement days, it has a statistically insignificant risk premium of −1.58 basis points per day with a t-statistic of −1.15. The premium difference is 11.34 basis points per day with a t-statistic of 2.79. Furthermore, the intercept is insignificantly different from zero on announcements days but significantly positive on non-announcement days. Although the world market index we construct in this section is synchronized to the U.S. trading day, it might be rather noisy because Table 1 shows that many countries have only few firms crosslisted in the US. Therefore, in this paper, we use the Datastream world market index, which covers most firms in the global equity market and should be less noisy. 4.4.2. World equity risk versus currency risk The world market factor we use is in the US dollar. Thus, it may contain not only world equity risk but also currency risk. Furthermore, Andersen et al. (2007), Faust et al. (2007), and Evans and Lyons (2008) document the reaction of exchange-rate movements to US macroeconomic announcements in intraday data. Thus, a natural question is: Is what we identify as a world market equity risk effect actually a currency risk effect? To answer this question, we examine the changes in exchange rates across two types of trading days, namely US macroeconomic announcement and non-announcement days. Since we focus on firms in developed economies in this paper, we examine the (daily) Major Currencies Index (MCI) from the Board of Governors of the U.S. Federal Reserve System. The MCI is a weighted average of the foreign exchange values of the U.S. dollar against currencies of major industrial countries, and is defined such that an increase in the MCI represents an appreciation of the U.S. dollar. The MCI includes the Euro Area, Canada, Japan, United Kingdom, Switzerland, Australia, and Sweden. Panel B of Table 8 reports the mean exchange-rate changes for the MCI and its component currencies on all days (“All”), those on announcement days (“A-Day”), those on non-announcement days (“N-Day”), and the differences in exchange-rate changes between announcement and non-announcement days, respectively. In general, exchange rates do not tend to behave differently on US macroeconomic announcement days, with one exception of Japan. For instance, the average changes of the MCI are −0.44 basis points per day (t = −1.00) on all days, 0.66 basis points per day (t = 0.53) on US macroeconomic announcement days, and −0.59 basis points per day (t = 1.27) on non-announcement days, respectively. The difference in the MCI changes between announcement and non-announcement days is 1.26 basis points per day (t = 0.95), which is not significant at the conventional levels. Thus, it is world equity risk, not currency risk, that drives the premium of the world market factor on US macroeconomic announcement days.

90

Table 8 Synchronous world market index and exchange-rate risk.

US

Constant rW R2

Foreign

US and foreign

All

A-Day

N-Day

Diff

All

A-Day

N-Day

Diff

All

A-Day

N-Day

Diff

3.29 (4.55) 1.12 (0.69) 0.25

1.82 (0.93) 9.09 (1.80) 0.27

3.49 (4.48) 0.04 (0.02) 0.25

−1.67 (−0.80) 9.05 (1.70)

5.38 (5.53) −1.03 (−0.73) 0.10

0.95 (0.35) 10.85 (2.59) 0.12

5.98 (5.73) −2.64 (−1.76) 0.10

−5.03 (−1.75) 13.49 (3.03)

4.40 (6.10) −0.22 (−0.17) 0.12

1.65 (0.83) 9.76 (2.55) 0.13

4.78 (6.17) −1.58 (−1.15) 0.12

−3.13 (−1.47) 11.34 (2.79)

Panel B: Exchange rate changes

MCI Australia Canada Euro

All

A-Day

N-Day

Diff

−0.44 (−1.00) 0.10 (0.14) 0.04 (0.10) −0.95 (−0.89)

0.66 (0.53) −0.10 (−0.05) −0.93 (−0.72) −5.04 (−1.76)

−0.59 (−1.27) 0.12 (0.16) 0.18 (0.40) −0.33 (−0.29)

1.26 (0.95) −0.22 (−0.10) −1.11 (−0.80) −4.70 (−1.52)

Japan Sweden Switzerland UK

All

A-Day

N-Day

Diff

−0.99 (−1.45) 0.08 (0.12) −1.33 (−1.77) −0.18 (−0.28)

3.43 (1.81) 0.28 (0.15) −0.00 (−0.00) 0.58 (0.33)

−1.62 (−2.22) 0.05 (0.07) −1.51 (−1.91) −0.28 (−0.42)

5.05 (2.49) 0.23 (0.11) 1.51 (0.65) 0.86 (0.47)

Panel A reports the results based on the synchronous world market index we construct. Sections “US”, “Foreign”, and “US and Foreign” show the results based on 40 US (beta, industry size and BM) portfolios, 40 foreign (beta, industry size and BM) portfolios, and 80 US and foreign (beta, industry size and BM) portfolios, respectively. Columns “All”, “A-Day”, “N-Day”, and “Diff” contain premium estimates on all days, those on announcement days, those on non-announcement days, and premium differences between announcement and nonannouncement days, respectively. Panel B reports the mean exchange-rate changes for the MCI and its component currencies on all days (“All”), those on announcement days (“A-Day”), those on non-announcement days (“N-Day”), and the differences in exchange-rate changes between announcement and non-announcement days, respectively.

D. Du, O. Hu / Journal of International Money and Finance 58 (2015) 75–97

Panel A: Synchronous world market index (80 US and foreign portfolios)

D. Du, O. Hu / Journal of International Money and Finance 58 (2015) 75–97

91

4.4.3. Country portfolios as test assets Country portfolios we construct in Section 4.4.1 have less of a factor structure and therefore are more stringent as test assets. In this section, we examine if our results are robust to using the country portfolios as test assets. Intuitively, we expect that US macroeconomic announcements have slightly weaker effects on the markets that are economically and/or geographically further from the U.S. (e.g., Asia Pacific markets). We test this conjecture by examining the mean excess returns of the country portfolios across two types of trading days (i.e., US macroeconomic announcement and non-announcement days). The results are reported in Section “Mean excess returns” of Panel A of Table 9. Columns “All”, “A-Day”, “N-Day”, and “Diff” present the mean excess returns on all days, those on announcement days, those on nonannouncement days, and the return differences between announcement and non-announcement days, respectively. First, like beta, industry, size and BM portfolios, country portfolios tend to have higher mean excess returns on US macroeconomic announcement days. Specifically, all 21 country portfolios except two earn higher excess returns on US macroeconomic announcement days, of which 11 are statistically significant at the 5% level for a one-sided test. Second, Asia Pacific economies seem to have slightly weaker announcement effects. For instance, the return differences between announcement and non-announcement days (“Diff”) are not statistically significant for three out of five Asia Pacific markets (i.e., Australia, Japan and Singapore). We also examine if the exposure of the country portfolios to the world market index varies with types of trading days with Eq. (2). The results are presented in Section “Exposure to the world market index” of Panel A of Table 9. Consistent with the results based on beta, industry, size and BM portfolios, the exposure of the country portfolios to the world market index does not vary with types of trading days. For instance, only two out of 21 country portfolios have significantly different exposure to the world market index on US macroeconomic announcement days at the 10% level. Therefore, we repeat our exercises in Section 4.2, except with the country portfolios as test assets, to examine if the premium on the world market factor is higher on US macroeconomic announcement days. The results are reported in Panel B of Table 9 and are in line with the results based on beta, industry, size and BM portfolios. For instance, if we do not distinguish announcement from non-announcement days, the world market factor on all days carries a statistically insignificant risk premium of 1.84 basis points per day with a t-statistic of 1.04. However, on US macroeconomic announcement days, it has a significantly positive risk premium of 14.63 basis points per day with a t-statistic of 2.81. In contrast, on non-announcement days, it has a statistically insignificant risk premium of 0.10 basis points per day with a t-statistic of 0.05. The premium difference is 14.54 basis points per day with a t-statistic of 2.62. Furthermore, the intercept is insignificantly different from zero on announcement days but significantly positive on nonannouncement days at the 10% level. Thus, our results are robust to using country portfolios as test assets.

4.4.4. CAPM versus International CAPM Savor and Wilson (2014) find that the US market risk premium is higher on US macroeconomic announcement days. Because U.S. market returns are a significant part of world market returns, an important question is: Is higher world market risk premium on US macroeconomic announcement days mechanically driven by higher US market risk premium at such times? To answer this question, we construct a world ex US index (as in Section 4.4.1 except that we exclude US firms). Panel A of Table 10 reports the mean excess returns of the world market index (including the US), the US market index, and the world ex US index on all days (“All”), those on announcement days (“A-Day”), those on non-announcement days (“N-Day”), and the return differences between announcement and nonannouncement days, respectively. If higher world market risk premium on US macroeconomic announcement days is mechanically driven by higher US market risk premium at such times, we expect that the world ex US index will not have higher returns on US macroeconomic announcement days. This conjecture is marginally rejected. While the mean excess return of the world ex US index is 7.84 basis points per day (t = 2.17) on announcement days, it is 2.82 basis points per day (t = 2.26) on nonannouncement days. The return difference is 5.02 basis points per day, marginally significant at the 10% level for a one-sided test. It is important to point out that since we construct the world ex US

92

D. Du, O. Hu / Journal of International Money and Finance 58 (2015) 75–97

Table 9 Country portfolios as test assets. Panel A: Mean excess returns and exposure to the world market index Mean excess returns

Australia Austria Belgium Canada Denmark Finland France Germany Hong Kong Israel Italy Japan Netherlands New Zealand Norway Portugal Singapore Spain Sweden UK USA

Exposure to the world market index

All

A-Day

N-Day

Diff

Constant

βι,W

βι,Α

R2

3.12 (1.68) 9.36 (2.15) 4.67 (1.09) 1.95 (1.42) 5.24 (2.65) 5.69 (1.78) 4.66 (2.35) 2.11 (1.11) 5.88 (2.40) 3.52 (1.67) 4.73 (2.39) 3.76 (2.31) 4.70 (3.62) 1.16 (0.21) 4.78 (2.16) 3.02 (1.23) 7.23 (1.61) 5.96 (3.22) 4.97 (1.90) 3.64 (2.76) 2.81 (2.65)

10.42 (1.97) 3.19 (0.28) 20.24 (1.81) 12.86 (3.09) 18.26 (3.53) 26.19 (2.80) 14.06 (2.48) 5.80 (1.01) 22.19 (2.66) 15.55 (2.74) 11.49 (1.84) 3.17 (0.67) 11.64 (2.89) 26.14 (1.67) 12.29 (1.90) 12.01 (1.62) 13.75 (1.03) 12.61 (2.41) 20.58 (2.85) 10.08 (2.63) 8.77 (2.63)

2.14 (1.09) 10.27 (2.19) 2.44 (0.53) 0.48 (0.34) 3.42 (1.62) 2.78 (0.81) 3.32 (1.56) 1.61 (0.82) 3.56 (1.35) 1.90 (0.84) 3.77 (1.78) 3.84 (2.21) 3.77 (2.69) −2.29 (−0.39) 3.72 (1.57) 1.74 (0.67) 6.30 (1.35) 5.02 (2.54) 2.83 (1.02) 2.77 (1.96) 2.01 (1.82)

8.28 (1.48) −7.08 (−0.58) 17.80 (1.49) 12.38 (2.88) 14.83 (2.68) 23.41 (2.33) 10.74 (1.77) 4.19 (0.70) 18.63 (2.08) 13.65 (2.25) 7.73 (1.16) −0.68 (−0.13) 7.88 (1.81) 28.44 (1.67) 8.57 (1.24) 10.27 (1.30) 7.45 (0.53) 7.59 (1.36) 17.76 (2.31) 7.32 (1.78) 6.76 (1.94)

0.13 (0.09) 8.82 (2.07) 2.08 (0.54) −0.72 (−0.76) 3.14 (1.74) 1.88 (0.65) 0.98 (0.71) −1.34 (−0.90) 2.60 (1.29) 1.14 (0.60) 2.35 (1.51) 0.26 (0.25) 1.76 (1.95) −0.02 (−0.00) 1.72 (0.90) 0.52 (0.24) 2.40 (0.65) 2.81 (1.92) 1.02 (0.46) 1.10 (1.13) 0.07 (0.14)

0.98 (27.39) 0.21 (3.19) 0.86 (20.85) 0.85 (46.43) 0.67 (17.36) 1.14 (25.14) 1.06 (46.10) 1.12 (36.35) 1.15 (37.76) 0.75 (25.09) 0.86 (33.85) 1.15 (54.74) 0.97 (46.62) 0.40 (5.64) 0.94 (21.21) 0.80 (24.20) 1.56 (22.38) 0.97 (40.04) 1.24 (33.04) 0.85 (47.53) 0.88 (75.75)

−0.00 (−0.06) 0.13 (1.18) 0.07 (0.84) 0.06 (1.50) −0.04 (−0.66) −0.03 (−0.23) −0.06 (−1.52) 0.06 (1.00) 0.09 (1.05) 0.09 (1.15) −0.05 (−0.87) −0.03 (−0.61) −0.01 (−0.40) −0.05 (−0.39) −0.05 (−0.61) 0.02 (0.25) −0.02 (−0.17) −0.08 (−1.68) −0.01 (−0.10) −0.06 (−1.49) 0.05 (2.39)

0.33 0.02 0.15 0.51 0.17 0.21 0.56 0.45 0.39 0.15 0.39 0.58 0.54 0.01 0.30 0.26 0.31 0.46 0.31 0.44 0.79

Panel B: Two-pass regression results

Constant rW R2

All

A-Day

N-Day

Diff

2.51 (1.71) 1.84 (1.04) 0.07

−1.09 (−0.25) 14.63 (2.81) 0.07

3.00 (1.91) 0.10 (0.05) 0.07

−4.09 (−0.90) 14.54 (2.62)

The table reports the time-series and two-pass regression results based on 21 country portfolios as test assets.

D. Du, O. Hu / Journal of International Money and Finance 58 (2015) 75–97

93

Table 10 CAPM versus International CAPM. Panel A: Mean excess returns of the factors The world market index

Mean

The US market index

The world ex US index

All

A-Day

N-Day

Diff

All

A-Day

N-Day

Diff

All

A-Day

N-Day

Diff

3.06 (2.92)

8.74 (2.43)

2.29 (2.11)

6.46 (1.72)

2.81 (2.74)

8.77 (2.23)

2.01 (1.92)

6.76 (1.66)

3.41 (2.86)

7.84 (2.17)

2.82 (2.26)

5.02 (1.33)

Panel B: Two-pass regression results 80 test assets

Constant rW rUS R2

101 test assets

All

A-Day

N-Day

Diff

All

A-Day

N-Day

Diff

6.59 (9.28) −0.60 (−0.47) −0.40 (−0.75) 0.23

4.00 (2.11) 9.14 (2.36) 0.12 (0.08) 0.24

6.94 (9.08) −1.93 (−1.40) −0.47 (−0.83) 0.22

−2.94 (−1.43) 11.07 (2.70) 0.59 (0.36)

6.03 (7.92) 0.05 (0.04) −0.40 (−0.82) 0.18

2.95 (1.41) 10.50 (2.75) 0.45 (0.31) 0.19

6.45 (7.90) −1.37 (−1.00) −0.52 (−0.99) 0.18

−3.51 (−1.56) 11.87 (2.92) 0.96 (0.63)

Panel A reports the mean excess returns of the world market index (including the US), the US market index, and the world ex US index on all days (“All”), those on announcement days (“A-Day”), those on non-announcement days (“N-Day”), and the return differences between announcement and non-announcement days, respectively. In Panel B, we estimate the following two-factor model with the two-pass regression: ri ,t = α i + βW ,irW ,t + βUS ,irUS ,t + ε i ,t

where rW,,t is the daily excess return on the world market index, rUS,t is the daily excess return on the US index that is orthogonalized to the world market index. While Section “80 test assets” presents the results based on 80 US and foreign beta, industry, size and BM portfolios as test assets, Section “101 test assets” present those based on 101 US and foreign beta, industry, size, BM and country portfolios. Columns “All”, “A-Day”, “N-Day”, and “Diff” contain premium estimates on all days, those on announcement days, those on non-announcement days, and premium differences between announcement and non-announcement days, respectively.

index with the cross-listed firms and many countries only have few firms cross-listed in the US, the world ex US index we construct can be very noisy, leading to the low test power. The evidence in Panel A of Table 10 suggests that higher world market risk premium on US macroeconomic announcement days is not mechanically due to higher US market risk premium at such times. Therefore, our results are new, relative to those in Savor and Wilson (2014). Next we explore if the International CAPM also provides a better description of mean excess returns of stocks in the global equity market. Essentially, we consider a two-factor model that includes both the world market factor and the US market factor (that is orthogonalized to the world market index):

ri,t = α i + βW ,i rW ,t + βUS ,i rUS ,t + ε i,t

(4)

where rUS,t is the daily excess return of the orthogonalized US market index. If the idiosyncraticcomponent of US market returns captured by rUS,t also helps to explain the cross-section of mean excess returns in the global equity market, the International CAPM is not a better asset-pricing model. Panel B of Table 10 presents the Fama and MacBeth (1973) test results based on the two-factor model (i.e., Eq. (4)). While Section “80 test assets” presents the results based on 80 US and foreign (beta, industry, size and BM) portfolios as test assets, Section “101 test assets” reports those based on the expanded set of test assets (which also includes 21 country portfolios). Interestingly, the orthogonalized US market return is not priced in the global equity market. For instance, based on 101 test assets, the premia of the orthogonalized US market index are −0.40 basis points per day (t = −0.82) on all days, 0.45 basis points per day (t = 0.31) on US macroeconomic announcement days, and −0.52 basis points per day (t = −0.99) on non-announcement days, respectively. Our results therefore suggest that the International CAPM is a better model for the global equity market.

94

D. Du, O. Hu / Journal of International Money and Finance 58 (2015) 75–97

5. Risk–return tradeoff at the market level The evidence in Section 4 suggests that in the global equity market there is a robust positive relationship between risk (International CAPM beta) and expected returns across assets on US macroeconomic announcement days, particularly in the subsample period from 1994 to 2013. In this section, following Savor and Wilson (2014), we provide further evidence that there is also a robust risk–return tradeoff at the market level on US macroeconomic announcement days in the global market. Our proxy of world market risk is a conditional forecast of one-quarter-ahead variance of daily world market returns (EVt), because “(t)heories, such as the CAPM, that relate expected returns to variance do so for the ex ante measure, not the innovation” (Savor and Wilson, 2014, p. 188). Quarterly variance of daily world market returns is defined as the average squared daily excess world market return over a quarter, RVt. In the same spirit of Savor and Wilson (2014), we use quarter t’s realized world market variance, log announcement-day world market excess returns and non-announcement-day world market excess returns to predict quarter t + 1’s world market variance. Fig. 3 shows the predicted world market variance (EVt) against the realized world market variance (RVt+1). It seems that EVt tracks both low-frequency movements and high-frequency spikes in the realized world market variance particularly well in recent years. With this estimate of EVs, we follow Savor and Wilson (2014) to test the relationship between risk (expected market variance) and returns at the market level. The results for the whole sample are presented in Panel A of Table 11. In all cases, the t-statistics are based on Newey–West HAC standard errors with the lag parameter set equal to 4 for our quarterly data. We also include an equation estimating the dynamics of EVt as in Savor and Wilson (2014). First, we regress log world market excess returns in quarter t + 1 (rW,t+1) on the expected world market variance for quarter t + 1 (EVt). The lagged log world market excess return (rW,t) is also included. The results are reported in row (1) of Panel A. Consistent with the US evidence in Savor and Wilson (2014), in the global market, expected world market variance is generally not correlated with expected world market returns. The coefficient on EVt is 0.02 with a t statistic of 0.62. Next, we separately estimate the correlation between expected world market variance (EVt) and expected log excess world market returns on announcement and non-announcement days. The results are presented in rows (3) and (4) of Panel A. Although the correlation between expected world market variance (EVt) and expected log excess world market returns on announcement days (rAW,t+1) in row (3) is positive, it is not statistically significant. Row (4) shows that non-announcement-day world market returns are negatively correlated with expected world market variance, although not significant either. As we point out in Section 4.3.5, US macroeconomic announcements may be more informative about the state of the global economy in recent years due to rapid global integration. Thus, the correlation

12 10 8 6 4 2 0 1974 1976 1978 1980 1982 1984 1986 1988 1990 1992 1994 1996 1998 2000 2002 2004 2006 2008 2010 2012 RV

EV

Fig. 3. World market variance. The figure shows the predicted world market variance (EV) against the realized world market variance (RV).

D. Du, O. Hu / Journal of International Money and Finance 58 (2015) 75–97

95

Table 11 Risk–return trade-off at the market level. Panel A: 1974–2013

(1)

rW,t+1

(2)

EVt+1

(3)

rAW,t+1

(4)

rNW,t+1

(5)

EVt+1

Intercept

rW,t

−0.00 (−0.16) 0.49 (6.80) −0.00 (−0.45) −0.00 (−0.12) 0.51 (6.65)

0.16 (1.81) −0.11 (−0.27)

rAW,t

0.15 (1.13) 0.04 (0.09) −1.06 (−0.24)

rNW,t

EVt

R2 0.00

−0.02 (−0.50) 0.17 (2.17) −0.07 (−0.21)

0.02 (0.62) 0.30 (2.61) 0.01 (1.46) 0.01 (0.49) 0.27 (4.33)

0.09 0.00 0.00 0.08

Panel B: 1974–1993

(1)

rW,t+1

(2)

EVt+1

(3)

rAW,t+1

(4)

rNW,t+1

(5)

EVt+1

Intercept

rW,t

0.02 (1.31) 0.43 (7.20) 0.01 (0.37) 0.01 (1.30) 0.44 (7.99)

0.07 (0.73) −0.62 (−1.17)

Intercept

rW,t

0.02 (1.21) 0.46 (5.40) −0.01 (−1.93) 0.01 (1.32) 0.48 (7.19)

0.06 (0.63) −0.23 (−0.43)

rNW,t

EVt

R2 −0.00

0.01 (0.02) −0.11 (−0.27) −3.13 (−0.90)

−0.07 (−0.83) 0.11 (1.34) −0.26 (−0.56)

−0.01 (−0.54) 0.34 (3.45) −0.01 (−0.24) −0.01 (−0.71) 0.32 (4.37)

rAW,t

rNW,t

EVt

R2 −0.00

0.05 (0.91) 0.11 (1.31) −0.10 (−0.17)

−0.01 (−0.54) 0.35 (2.60) 0.02 (3.45) −0.01 (−0.78) 0.32 (3.45)

rAW,t

0.16 0.03 0.00 0.18

Panel C: 1994–2013

(1)

rW,t+1

(2)

EVt+1

(3)

rAW,t+1

(4)

rNW,t+1

(5)

EVt+1

0.47 (2.46) −0.17 (−0.45) −1.59 (−0.48)

0.12 0.02 0.00 0.12

rW,t is log world market excess returns in quarter t. rAW,t and rNW,t are log announcement-day and nonannouncement-day world market excess returns in quarter t, respectively. EVt is predicted world market variance in quarter t + 1.

between world market variance and world market returns on announcement days may be time varying. To test this conjecture, we repeat our exercises in the two equal-length subsamples and report the results in Panels B and C. Consistent with our conjecture, although the correlation between expected world market variance and expected world market returns is not significant in the 1974–1993 subsample, it is significantly positive in the 1994–2013 period. More specifically, for the 1974–1993 subsample, the coefficient on EVt in row (3) of Panel B is −0.01 with a t statistic of −0.24. However, for the 1994– 2013 subsample, the coefficient on EVt in row (3) of Panel C is 0.02 with a t statistic of 3.45. 6. Conclusions We apply the Savor and Wilson (2014) methodology to daily US stocks as well as foreign stocks cross-listed in the US to examine if the world market risk premium is higher on US macroeconomic announcement days. Our findings can be easily summarized. Although the world market risk premium

96

D. Du, O. Hu / Journal of International Money and Finance 58 (2015) 75–97

is insignificantly different from zero on non-announcement days, it is robustly positive on US macroeconomic announcement days. Furthermore, we show that higher world market risk premium on US macroeconomic announcement days is not mechanically due to high US market risk premium at such times, and that the International CAPM provides a better description of the cross-section of stock returns in the global equity market. We interpret our results as suggesting that there is a causal relationship from the state of the global economy to the world market risk premium. Our findings have important theoretical as well as practical implications. In terms of theoretical implications, our findings suggest that future theoretical research should explore the mechanisms through which the state of the global economy affects the world market risk premium. In terms of practical implications, strengthening Rapach et al. (2013), our findings imply that in capital budgeting, portfolio evaluation, investment, and risk analysis decisions, an international asset-pricing model that explicitly takes into account the role of the US may be more informative.

Acknowledgments The authors thank the editor Joshua Aizenman and one anonymous referee for their valuable and insightful comments. The responsibility of any remaining errors is ours.

References Adler, M., Dumas, B., 1983. International portfolio choice and corporation finance: a synthesis. J. Finance 38, 925–984. Ammer, J., Vega, C., Wongswan, J., 2010. International transmission of US monetary policy shocks: evidence from stock prices. J. Money Credit Bank. 42, 179–198. Andersen, T.G., Bollerslev, T., Diebold, F.X., Vega, C., 2007. Real-time price discovery in global stock, bond and foreign exchange markets. J. Int. Econ. 73, 251–277. Balvers, R., Klein, A.F., 2014. Currency risk premia and uncovered interest parity in the international CAPM. J. Int. Money Finance 41, 214–230. Bartov, E., Bodnar, G.M., Kaul, A., 1996. Exchange rate variability and the riskiness of U.S. multinational firms: evidence from the breakdown of the Bretton Woods system. J. Financ. Econ. 42, 105–132. Bekaert, G., Harvey, C.R., Lundblad, C.T., Siegel, S., 2011. What segments equity markets? Rev. Financ. Stud. 24, 3841–3890. Campbell, J.Y., Hamao, Y., 1992. Predictable stock returns in the United States and Japan: a study of long-term capital market integration. J. Finance 47, 43–69. Cooper, M.J., Gutierrez, R.C., Hameed, A., 2004. Market states and momentum. J. Finance 59, 1345–1365. De Santis, G., Gerard, B., 1998. How big is the premium for currency risk? J. Financ. Econ 49, 375–412. Du, D., 2013. Another look at the cross-section and time-series of stock returns: 1951 to 2011. J. Empir. Finance 20, 130–146. Du, D., Hu, O., 2012. Foreign exchange volatility and stock returns. J. Int. Financ. Mark. Inst. Money 22, 1202–1216. Dumas, B., Solnik, B., 1995. The world price of foreign exchange risk. J. Finance 50, 445–479. Evans, M.D.D., Hnatkovska, V.V., 2014. International capital flows, returns and world financial integration. J. Int. Econ. 92, 14–33. Evans, M.D.D., Lyons, R.K., 2008. How is macro news transmitted to exchange rates? J. Financ. Econ. 88, 26–50. Fama, E., Fisher, L., Jensen, M., Roll, R., 1969. The adjustment of stock prices to new information. Int. Econ. Rev. 10, 1–21. Fama, E.F., French, K.R., 1998. Value versus growth: the international evidence. J. Finance 53, 1975–1999. Fama, E.F., French, K.R., 2012. Size, value, and momentum in international stock returns. J. Financ. Econ. 105, 457–472. Fama, E.F., MacBeth, J.D., 1973. Risk, return and equilibrium: empirical tests. J. Polit. Econ. 81, 607–636. Faust, J., Rogers, J.H., Wang, S.B., Wright, J.H., 2007. The high-frequency response of exchange rates and interest rates to macroeconomic announcements. J. Monet. Econ. 54, 1051–1068. Ferson, W.E., Harvey, C.R., 1993. The risk and predictability of international equity returns. Rev. Financ. Stud. 6, 527–566. Grauer, F.L.A., Litzenberger, R.H., Stehle, R.E., 1976. Sharing rules and equilibrium in an international capital market under uncertainty. J. Financ. Econ. 3, 233–256. Harvey, C.R., 1991. The world price of covariance risk. J. Finance 46, 111–157. Harvey, C.R., Solnik, B., Zhou, G., 2002. What determined expected international asset returns? Annals of Economics and Finance 3, 249–298. Hausman, J., Wongswan, J., 2011. Global asset prices and FOMC announcements. J. Int. Money Finance 30, 547–571. Hou, K., Karolyi, A., Kho, B., 2011. What factors drive global stock returns? Rev. Financ. Stud. 24, 2527–2574. Koedijk, K.G., Kool, C.J.M., Schotman, P.C., van Dijk, M.A., 2002. The cost of capital in international financial markets: local or global? J. Int. Money Finance 21, 905–929. Lewis, K.K., 2011. Global asset pricing. Annu. Rev. Financ. Econ. 3, 435–466. Ng, D.T., 2004. The international CAPM when expected returns are time-varying. J. Int. Money Finance 23, 189–230. Rapach, D.E., Strauss, J.K., Zhou, G., 2013. International stock return predictability: what is the role of the United States? J. Finance 68, 1633–1662. Savor, P., Wilson, M., 2013. How much do investors care about macroeconomic risk? Evidence from scheduled economic announcements. J. Financ. Quant. Anal. 48, 343–375. Savor, P., Wilson, M., 2014. Asset pricing: a tale of two days. J. Financ. Econ. 113, 171–201.

D. Du, O. Hu / Journal of International Money and Finance 58 (2015) 75–97

97

Shanken, J., Zhou, G., 2007. Estimating and testing beta pricing models: alternative methods and their performance in simulations. J. Financ. Econ. 84, 40–86. Solnik, B.H., 1974a. An equilibrium model of the international capital market. J. Econ. Theory 8, 500–524. Solnik, B.H., 1974b. The international pricing of risk: an empirical investigation of the world, capital market structure. J. Finance 29, 365–378. Stehle, R.E., 1977. An empirical test of the alternative hypotheses of national and international pricing of risky assets. J. Finance 32, 493–502. Stulz, R.M., 1981. A model of international asset pricing. J. Financ. Econ. 9, 383–406. Wongswan, J., 2006. Transmission of information across international equity markets. Rev. Financ. Stud. 19, 1157–1189.