Market Impact of Macroeconomic Announcements: Do Surprises Matter?

Transcription

1 Market Impact of Macroeconomic Announcements: Do Surprises Matter? Alec Schmidt Kensho; Financial Engineering Stevens Institute of Technology Financial Risk and Engineering NYU School of Engineering; 1

2 History I Chen, Roll, and Ross (1986): Macroeconomic developments are undiversified risk within the APT framework and hence merit excess returns. However, economic theory does not determine which economic variables are responsible for excess returns. Influential works: Balduzzi et al (2001); Flannery & Protopapadakis (2002); Andersen et al (2003); Hautsch & Hess (2007); Gilbert et al (2010); Evans (2011). Excess returns for US equity indexes and bonds on the days of macroeconomic announcements prior to 2009: Savor & Wilson (2013) no surprises included in the model. 2

3 History II Nadler & Schmidt (2014) examined returns of several US equity ETFs on the days of 18 major macroeconomic announcements for the period of January 2009 July The ARMA+GARCH model with external regression terms that describe announcement events and their surprises is used. It was found that ISM Manufacturing Reports, Non- Farm Payrolls, International Trade Balance, Index of Leading Indicators, Housing Starts, Jobless Claims were statistically significant. Several trading strategies based on realizing daily returns on the announcement days outperform buy-and-hold strategy. 3

4 Statistics for SPY daily returns on the days of macroeconomic announcements Economic Indicators SPY (mean daily return=0.0006) Announcements # of data Mean corr(r, s) p-value Return/surprise, % Model daily return +/+ +/- p25 p5 Real Activity Non-farm payroll Retail sales Industrial Production Personal Income & Outlays Consumer Credit GDP Consumption New Home Sales Investment Durable Goods Orders Factory Orders Government Purchases Treasury budget Net Exports International Trade Balance Prices Producer Price Index Consumer Price Index Forward-Looking Consumer Confidence Index ISM manufacturing report Housing Starts Index of Leading Indicators Jobless claims

5 History III Details in Performance of announcement-based trading strategies: SPY B&H Effective Sharpe ratio: Sh = R/(σ T); R - compound return; σ - standard deviation of mean daily returns on T announcement days in the sample when the strategy is exposed to market risk. SPY Days with Announcements Model #18 Model p25 Model p5 T Mean daily return p-value na σ Compound return Sharpe ratio

6 Do surprises matter? Surprise: S k (t) = [A k (t) E k (t)]/σ Sk (1) A k (t) and E k (t) are actual and expected (consensus) values of macroeconomic indicator k at time t, Σ Sk is standard deviation of S k. Consensus: company guidance; Wall st. estimates; Estimize crowd estimates (estimize.com). Problems: E k is subjective and A k has a look-ahead bias. Actual change: AC k (t) = [A k (t) A k (t-1)]/σ Ak (2) Expected change: EC k (t) = [E k (t) A k (t-1)]/σ Ek (3) 6

7 Specifics of This Work Details in Time period: January 2004 April major macroeconomic indicators S&P 500 SPDR ETF (SPY) The autoregressive model includes both the announcement event factors (dummy variables) and the announcement surprise/actualchange/expected change factors. Volatility is accounted by the GARCH process. A k and E k from Econoday.com 7

8 The Model ARMA(p, q) + GARCH(m, n) RUGARCH software (Ghalanos, 2013) Parsimonious model: 1 st minimum of Akaike criterion => m=2 and n=p=q=1. r i (t) = μ i + a i,1 r i (t-1) +b i,1 ε i (t-1) + k=1 N d ik D k t + N k=1 s ik X k (t) + ε i (t) (4) ε i (t) = z(t)σ i (t), z(t) ~ N(0,1) (5) σ i2 (t) = ω i + α i,1 ε i2 (t-1) + α i,2 ε i2 (t-2) + β i,1 σ i2 (t-1) (6) X k = {S k, AC k, EC k } 8

9 Results I Correlations between Actual Change (AC), Expected Change (EC), and Surprises (S) Macroeconomic Correlations Indicators (EC, AC) (S, AC) Consumer Confidence Consumer Credit Consumer Price Index Durable Goods Orders Factory Orders GDP Housing Starts Industrial Production Trade balance ISM Mfg Index Jobless Claims Leading Indicators New Home Sales Non-farm payrolls Personal Income Producer Price Index Retail Sales Treasury Budget

10 Results II For entire period, Jan 2004 Apr 2014, Consumer Credit, Housing Starts, ISM Manufacturing Report, and Non-Farm Payrolls had p-value < 0.1 for all three models (S-based, AC-based, and EC-based). In (including bear market of ), Consumer Credit, Housing Starts, ISM Manufacturing Report, and Jobless Claims had p-value < 0.1 for all three models while Non-Farm Payrolls had p- value < 0.1 for both AC- and EC-based models. In (bull market), Housing Starts, ISM Manufacturing Report, Leading Indicators, and Non-Farm Payrolls had p-value < 0.1 for all three models, and International Trade Balance had p-value < 0.1 for S-, and AC-based models. 10

11 Macroeconomic announcement indexes I Are macroeconomic announcements the major driver of returns? Citigroup Economic Surprise Index log return R i (t) = k=1 N d ik D k t + N k=1 x ik X k (t) (7) p i (t) = Ap i (t-1)exp(r i (t)) (8) A = exp(-r a ), t > τ = 7-Jan-2005 (9) R a is R i (t) averaged over the period Feb 2004 Dec Works for AC- and EC-based indexes but not for S-based index. 11

12 SPY fits (7) (9) with 18 macroeconomic indexes SPY data AC-based fit S-based fit EC-based fit S P Y I n d e x e s /14/2004 5/28/ /10/2006 2/22/2008 7/6/ /18/2010 4/1/2012 8/14/ /27/

13 Macroeconomic announcement indexes II S-based index has the highest volatility with σ = 5.49 while the AC-based index has the lowest volatility with σ = The EC-based index has σ = 3.47, and the SPY return has σ = S index AC Index EC Index SPY return I n d 100 e x e s 90 S 130 P Y r 120 e t u 110r n /14/2004 5/28/ /10/2006 2/22/2008 7/6/ /18/2010 4/1/2012 8/14/ /27/2014

14 Macroeconomic announcement indexes III p i (t) = Ap i (t-1)exp(i i (t)) (10) I i (t) = αr i (t) + (1-α)I i (t-1) (11) A = exp(-i a ), t > τ = 7-Jan-2005 (12) I a is I i (t) averaged over the period Feb 2004 Dec 2004; α =2/(T+1); T = 7;

15 SPY fits (10) (12) using 18 macroeconomic indexes T = 28 15

16 SPY fits (10) (12) using 7 macroeconomic indexes with p < 0.1 T = 7 16

17 Macroeconomic announcement indexes III Correlations between the SPY price and its S/AC/EC fits Index # of Indicators 18 7 S AC EC Dropping statistically insignificant announcements improve results for S-based index. 17

18 Conclusions Surprises (S) in the macroeconomic indicators that are used in current research of market impact of macroeconomic announcements seem not to have advantages over actual changes of the indicators (AC) and expected changes of the indicators (EC). All three variables (S, AC, and EC) yield similar statistically significant macroeconomic indicators. The disadvantage of S versus AC is that the former is much noisier, which may be related to its intrinsically subjective nature. The advantage of EC over both S and AC is that the former does not have a look-ahead bias and hence can be used for short-term forecasting. With proper fitting, all three variables (S, AC, and EC) can be used for qualitative modeling the SPY price dynamics. However, the effects of macroeconomic announcements alone do not describe the strength of the bull market of

19 19

20 Q & A 20