Realized and Anticipated Macroeconomic Conditions Forecast Stock Returns

Transcription

1 Realized and Anticipated Macroeconomic Conditions Forecast Stock Returns Alessandro Beber Michael W. Brandt Maurizio Luisi Cass Business School Fuqua School of Business Quantitative City University Duke University Investment and CEPR and NBER Solutions London Quant Group May,

2 Big picture question How do we characterize the variation of the risk premium on stocks, both through time and across countries? 2

3 Big picture question How do we characterize the variation of the risk premium on stocks, both through time and across countries? Empiricist use financial market forecasting variables (e.g., val. ratios). Models are based on economic fundamentals (e.g., consumption). Disconnect, why? Fundamental data are infrequent, backward-looking, restated... Market predictors, however, reflect also other things (preferences, misvaluation) 2

4 Big picture question How do we characterize the variation of the risk premium on stocks, both through time and across countries? Empiricist use financial market forecasting variables (e.g., val. ratios). Models are based on economic fundamentals (e.g., consumption). Disconnect, why? Fundamental data are infrequent, backward-looking, restated... Market predictors, however, reflect also other things (preferences, misvaluation) We want to take a structured economic news flow approach 2

5 Big picture question How do we characterize the variation of the risk premium on stocks, both through time and across countries? Empiricist use financial market forecasting variables (e.g., val. ratios). Models are based on economic fundamentals (e.g., consumption). Disconnect, why? Fundamental data are infrequent, backward-looking, restated... Market predictors, however, reflect also other things (preferences, misvaluation) We want to take a structured economic news flow approach Multifactor model for fundamentals Focus on realized ex-post measures (eg, quarterly GDP releases) and anticipating ex-ante information (eg, survey of consumers/ firm managers) Incorporate the entire cross-section of publicly released data Continuous (at least daily) updating 2

6 The paper in a nutshell Construct real-time systematic macroeconomic factors 3

7 The paper in a nutshell Construct real-time systematic macroeconomic factors Statistical evidence of predictability Realized growth forecasts stock market returns 1-4 months ahead. Anticipated growth (orthogonal to realized) adds to this predictability. Effect far exceeds and not subsumed by predictability of usual suspects. 3

8 The paper in a nutshell Construct real-time systematic macroeconomic factors Statistical evidence of predictability Realized growth forecasts stock market returns 1-4 months ahead. Anticipated growth (orthogonal to realized) adds to this predictability. Effect far exceeds and not subsumed by predictability of usual suspects. Return predictability by fundamentals is state dependent 3

9 The paper in a nutshell Construct real-time systematic macroeconomic factors Statistical evidence of predictability Realized growth forecasts stock market returns 1-4 months ahead. Anticipated growth (orthogonal to realized) adds to this predictability. Effect far exceeds and not subsumed by predictability of usual suspects. Return predictability by fundamentals is state dependent Consistent evidence internationally. 3

10 The paper in a nutshell Construct real-time systematic macroeconomic factors Statistical evidence of predictability Realized growth forecasts stock market returns 1-4 months ahead. Anticipated growth (orthogonal to realized) adds to this predictability. Effect far exceeds and not subsumed by predictability of usual suspects. Return predictability by fundamentals is state dependent Consistent evidence internationally. Economic relevance of predictability Profitable long-short country selection strategy 3

11 Outline Data Methodology Preliminaries Empirical results Conclusion 4

12 Data As released data on macroeconomic news from Bloomberg Only exactly time-stamped and non-restated data 43 US, 43 UK, 183 European, and 45 Japanese releases Jan 1997 through Dec

13 Data As released data on macroeconomic news from Bloomberg Only exactly time-stamped and non-restated data 43 US, 43 UK, 183 European, and 45 Japanese releases Jan 1997 through Dec 2011 For each release we obtain the announcement date and time, the released statistic, its consensus expectation, and the complete cross-section of economist forecasts Detailed UK and European forecasts are available from June 1997 and detailed Japanese forecasts are available from May

14 Data As released data on macroeconomic news from Bloomberg Only exactly time-stamped and non-restated data 43 US, 43 UK, 183 European, and 45 Japanese releases Jan 1997 through Dec 2011 For each release we obtain the announcement date and time, the released statistic, its consensus expectation, and the complete cross-section of economist forecasts Detailed UK and European forecasts are available from June 1997 and detailed Japanese forecasts are available from May 2000 Daily S&P 500, FTSE 100, Euro STOXX 50, and Nikkei 225 index returns (net of local risk-free rates) 5

15 Data As released data on macroeconomic news from Bloomberg Only exactly time-stamped and non-restated data 43 US, 43 UK, 183 European, and 45 Japanese releases Jan 1997 through Dec 2011 For each release we obtain the announcement date and time, the released statistic, its consensus expectation, and the complete cross-section of economist forecasts Detailed UK and European forecasts are available from June 1997 and detailed Japanese forecasts are available from May 2000 Daily S&P 500, FTSE 100, Euro STOXX 50, and Nikkei 225 index returns (net of local risk-free rates) Usual suspect risk premium predictors used in the literature including VIX and the volatility risk premium constructed using daily realized volatility based on 5-min index returns 5

16 Categorizing economic news Inflation Growth { Output Realized Growth Employment Anticipated Growth 6

17 Economic news flow Conference Board Consumer Confidence Chicago Purchasing Managers Index Inflation University Michigan Consumer Survey Employment ADP National Employment Report Output ISM Manufacturing PMI Macro Sentiment Nonfarm Payrolls Total,Manufacturing + Unemployment Rate + Average Weekly Hours ISM Non-Manufacturing PMI Retail Sales + Retail Sales Less Auto Import Price Index PPI + PPI Core Industrial Production + Capacity Utilization Empire State Manufacturing Survey Manufacturing Trade Inventories CPI + CPI Core Durable Goods Orders Conference Board Leading Index GDP + GDP Price Index Personal Income + Pers. Consum. Exp. + PCE Price Index Manufacturers New Orders reference month M month M+1 month M+2 7

18 Transformations and temporal alignment We first-difference all non-stationary news series Dickey-Fuller test and economic intuition 8

19 Transformations and temporal alignment We first-difference all non-stationary news series Dickey-Fuller test and economic intuition We then populate the data into T N matrices by country j=1 j=2 j=5 j=6 j=n missing missing missing t-22 A t-22,1 not released... missing not released t-21 not released A t-21,2... missing A t-21, not released not released... missing not released t A t,1 not released... A t,5 not released t+1 not released A t+1,2... not released A t+1, not released not released... not released discontinued discontinued T discontinued j=1 j=2 j=5 j=6 j=n missing

20 Transformations and temporal alignment (cont) Two data issues Missing data in between release dates Different starting and ending dates 9

21 j=1 j=2 j=5 j=6 j=n missing missing Transformations... and... temporal... missing alignment (cont)... t-22 A t-22,1 not released... missing not released t-21 not released A t-21,2... missing A t-21, Two data issues Missing data in between release dates Different starting and ending dates not released not released... missing not released t A t,1 not released... A t,5 not released t+1 not released A t+1,2... not released A t+1, not released not released... not released discontinued discontinued T discontinued We solve the missing data problem by forward filling j=1 j=2 j=5 j=6 j=n missing missing missing t-22 A t-22,1 E[A t-22,2 ]=A t-43,2... missing E[A t-22,6 ]=A t-43, t-21 E[A t-21,1 ]=A t-22,1 A t-21,2... missing A t-21, E[A...,1 ]=A t-22,1 E[A...,2 ]=A t-21,2... missing E[A...,2 ]=A t-21, t A t,1 E[A t,2]=a t-21,2... A t,5 E[A t,2]=a t-21, t+1 E[A t+1,1 ]=A t,1 A t+1,2... E[A t+1,5 ]=A t,5 A t+1, E[A...,1 ]=A t,1 E[A...,2]=A t+1,2... E[A...,5 ]=A t,5 discontinued discontinued T discontinued

22 j=1 j=2 j=5 j=6 j=n missing missing Transformations... and... temporal... missing alignment (cont)... t-22 A t-22,1 not released... missing not released t-21 not released A t-21,2... missing A t-21, Two data issues Missing data in between release dates Different starting and ending dates not released not released... missing not released t A t,1 not released... A t,5 not released t+1 not released A t+1,2... not released A t+1, not released not released... not released discontinued discontinued T discontinued We solve the missing data problem by forward filling j=1 j=2 j=5 j=6 j=n missing missing missing t-22 A t-22,1 E[A t-22,2 ]=A t-43,2... missing E[A t-22,6 ]=A t-43, t-21 E[A t-21,1 ]=A t-22,1 A t-21,2... missing A t-21, E[A...,1 ]=A t-22,1 E[A...,2 ]=A t-21,2... missing E[A...,2 ]=A t-21, t A t,1 E[A t,2]=a t-21,2... A t,5 E[A t,2]=a t-21, t+1 E[A t+1,1 ]=A t,1 A t+1,2... E[A t+1,5 ]=A t,5 A t+1, E[A...,1 ]=A t,1 E[A...,2]=A t+1,2... E[A...,5 ]=A t,5 discontinued discontinued T discontinued We address the different data length explicitly in our methodology 9

23 Methodology For each country and ex-ante categorized subset of macro news series (for inflation, output, etc.), we extract the first principal component from the correlation matrix of the data 10

24 Methodology For each country and ex-ante categorized subset of macro news series (for inflation, output, etc.), we extract the first principal component from the correlation matrix of the data Besides the news categorization, the critical ingredient to our methodology is the correlation matrix 10

25 Methodology For each country and ex-ante categorized subset of macro news series (for inflation, output, etc.), we extract the first principal component from the correlation matrix of the data Besides the news categorization, the critical ingredient to our methodology is the correlation matrix Use Stambaugh (1997) to accommodate different data lengths 10

26 Methodology For each country and ex-ante categorized subset of macro news series (for inflation, output, etc.), we extract the first principal component from the correlation matrix of the data Besides the news categorization, the critical ingredient to our methodology is the correlation matrix Use Stambaugh (1997) to accommodate different data lengths Coarse sub-sampling scheme to overcome extreme local persistence Subsample forward-filled data monthly to estimate correlation matrix Repeat for all sampling starting dates over past the month Average all sub-sampled estimates over the past month Analogous to Ait-Sahalia et al. (2005) for microstructure data 10

27 Methodology For each country and ex-ante categorized subset of macro news series (for inflation, output, etc.), we extract the first principal component from the correlation matrix of the data Besides the news categorization, the critical ingredient to our methodology is the correlation matrix Use Stambaugh (1997) to accommodate different data lengths Coarse sub-sampling scheme to overcome extreme local persistence Subsample forward-filled data monthly to estimate correlation matrix Repeat for all sampling starting dates over past the month Average all sub-sampled estimates over the past month Analogous to Ait-Sahalia et al. (2005) for microstructure data Telescoping estimates starting with an initial five year sample (60 monthly observations) to obtain true real-time factor estimates 10

28 Existing alternative approaches Two general approaches to real-time macroeconomics 11

29 Existing alternative approaches Two general approaches to real-time macroeconomics 1. Balanced panel coincident indices Monthly or quarterly weighted averages of a large cross-section of data E.g., Stock and Watson (1989) = CFNAI 11

30 Existing alternative approaches Two general approaches to real-time macroeconomics 1. Balanced panel coincident indices Monthly or quarterly weighted averages of a large cross-section of data E.g., Stock and Watson (1989) = CFNAI 2. Nowcasting Kalman filter with AR factor dynamics applied to only a few indicators E.g., Evans (2005) or Arouba et al. (2009) (later ADS) 11

31 Existing alternative approaches Two general approaches to real-time macroeconomics 1. Balanced panel coincident indices Monthly or quarterly weighted averages of a large cross-section of data E.g., Stock and Watson (1989) = CFNAI 2. Nowcasting Kalman filter with AR factor dynamics applied to only a few indicators E.g., Evans (2005) or Arouba et al. (2009) (later ADS) Traditionally optimized to forecast GDP (and used for that) 11

32 Factor correlations Output (blue) and employment (red) are highly correlated 2 Output and Employment collapse to one series (= realized growth) 12

33 Factor correlations (cont) Realized (blue) and Anticipated (red) growth are highly correlated 2 Realized and Anticipated Growth orthogonalize anticipated to realized growth or collapse to one series (= growth) 13

34 Factor correlations (cont) Growth (blue) and inflation (red) are fairly independent Growth and Inflation

35 Growth factor 2 Growth Index

36 Growth factor versus CFNAI 2 1 CFNAI and Growth Index CFNAI Growth

37 Growth factor versus ADS ADS and Growth Index ADS Growth

38 Growth factor versus SPF and realized GDP 3 2 GDP, SPF, and real-time Growth Index GDPact SPF GROWTH

39 Growth factor versus S&P index GROWTH SP

40 Growth factor uncertainty 3.5 Dispersion

41 Growth factor uncertainty versus VIX DISP VIX

42 Growth factor uncertainty versus S&P index DISP SP

43 International Evidence Growth US -4 EU UK JP

44 Correlations Growth Indices Other Predictors R m R f All Realized Anticipated Dispersion VRP ln P E ln D P DEF TERM Summary Statistics Mean Std Dev Skew Kurtosis Correlation Matrix R m R f Growth Indices: All Realized Anticipated Dispersion Other Predictors: VRP ln(p/e) ln(d/p) DEF TERM

45 Correlations Growth Indices Other Predictors R m R f All Realized Anticipated Dispersion VRP ln P E ln D P DEF TERM Summary Statistics Mean Std Dev Skew Kurtosis Correlation Matrix R m R f Growth Indices: All Realized Anticipated Dispersion Other Predictors: VRP ln(p/e) ln(d/p) DEF TERM

46 Correlations Growth Indices Other Predictors R m R f All Realized Anticipated Dispersion VRP ln P E ln D P DEF TERM Summary Statistics Mean Std Dev Skew Kurtosis Correlation Matrix R m R f Growth Indices: All Realized Anticipated Dispersion Other Predictors: VRP ln(p/e) ln(d/p) DEF TERM

47 Main empirical results Predictability regressions R m;t,t+lead R f ;t,t+lead = α + βf t + γz t + ɛ t, with F t is either realized growth, orthogonal anticipated growth, or both, and Z t is a set of control variables 27

48 US return predictability Horizon Δ t 22,t Realized (1.70) (3.76) (3.25) (2.90) (2.45) (1.49) (1.24) Adj. R 2 (%)

49 US return predictability (cont) Horizon Δ t 22,t Realized (1.70) (3.76) (3.25) (2.90) (2.45) (1.49) (1.24) Adj. R 2 (%) Horizon Anticipated (1.77) (1.96) (3.24) (4.07) (4.46) (4.67) (4.59) Adj. R 2 (%)

50 US return predictability (cont) Horizon Δ t 22,t Realized (1.70) (3.76) (3.25) (2.90) (2.45) (1.49) (1.24) Adj. R 2 (%) Horizon Anticipated (1.77) (1.96) (3.24) (4.07) (4.46) (4.67) (4.59) Adj. R 2 (%) Horizon Δ t 22,t Realized (1.67) (3.66) (3.15) (2.80) (2.41) (1.39) (1.0) Anticipated (1.72) (1.90) (3.20) (4.03) (4.40) (4.64) (4.56) Adj. R 2 (%)

51 Realized/Anticipated Growth vs. Other Predictors Panel A: 20-day horizon (1) (2) (3) (4) (5) (6) (7) (8) (9) Δ t 22,t Realized (3.66) (4.25) (3.96) (3.79) Anticipated (1.90) (1.71) (2.37) (2.22) VRP (0.64) (1.76) ln P E (-1.75) (-2.71) ln D P (1.76) DEF (-0.87) (-0.14) TERM (-0.44) (-1.57) Adj. R 2 (%)

52 Realized/Anticipated Growth vs. Other Predictors Panel B: 60-day horizon (1) (2) (3) (4) (5) (6) (7) (8) (9) Δ t 22,t Realized (2.80) (3.02) (3.25) (2.86) Anticipated (4.03) (3.94) (4.91) (4.19) VRP (0.51) (0.99) ln P E (-2.05) (-3.94) ln D P (1.86) DEF (-0.68) (-0.19) TERM (-0.34) (-2.03) Adj. R 2 (%)

53 Conditional U.S. Stock Market Predictability Panel A: Growth dispersion above median Horizon Δ t 22,t Realized (2.93) (3.97) (3.05) (2.83) (1.58) (0.72) (-0.22) Anticipated (-0.02) (1.14) (3.18) (3.98) (4.41) (4.75) (4.76) Adj. R 2 (%) Panel B: Growth dispersion below median Horizon Δ t 22,t Realized (-0.87) (0.18) (-0.32) (-0.15) (0.79) (-0.01) (0.56) Anticipated (2.72) (1.84) (1.25) (1.18) (1.45) (1.59) (1.40) Adj. R 2 (%)

54 Conditional U.S. Stock Market Predictability (cont) Panel C: Recession (growth level < 0) Horizon Δ t 22,t Realized (1.83) (2.92) (2.08) (1.80) (1.27) (0.28) (-0.57) Anticipated (1.93) (2.36) (3.92) (4.70) (5.28) (5.83) (6.12) Adj. R 2 (%) Panel D: Expansion (growth level > 0) Horizon Δ t 22,t Realized (-0.88) (0.85) (1.18) (0.96) (1.00) (0.34) (0.72) Anticipated (-1.41) (-2.35) (-1.67) (-0.55) (-0.31) (-0.15) (-0.01) Adj. R 2 (%)

55 International evidence: Local and Global Factors Panel A: U.S. Local Global U.S. as Global Horizon Δt 22,t Realized (3.66) (3.15) (2.80) (2.41) (2.04) (1.32) (1.34) (0.65) (3.66) (3.15) (2.80) (2.41) Anticipated (1.90) (3.20) (4.03) (4.40) (2.03) (3.58) (4.15) (4.48) (1.90) (3.20) (4.03) (4.40) Adj. R 2 (%) Panel B: Europe Local Global U.S. as Global Horizon Δt 22,t Realized (-0.31) (-1.72) (-1.01) (-1.50) (1.88) (0.84) (1.13) (0.46) (3.67) (3.35) (2.91) (2.42) Anticipated (4.39) (5.72) (5.89) (5.54) (2.11) (3.46) (3.76) (4.04) (1.50) (2.53) (3.47) (3.78) Adj. R 2 (%)

56 International evidence: Local and Global Factors Panel C: U.K. Local Global U.S. as Global Horizon Δt 22,t Realized (1.26) (1.58) (1.62) (1.06) (2.10) (1.24) (1.11) (0.16) (3.33) (3.37) (2.81) (2.34) Anticipated (2.03) (1.91) (2.18) (2.01) (1.96) (3.47) (4.53) (4.92) (1.41) (2.48) (3.84) (4.28) Adj. R 2 (%) Panel D: Japan Local Global U.S. as Global Horizon Δt 22,t Realized (-0.32) (-1.71) (-1.09) (-0.78) (0.89) (-0.10) (-0.36) (-0.92) (3.57) (2.93) (1.78) (0.89) Anticipated (3.01) (3.43) (2.89) (2.63) (3.36) (4.76) (4.54) (3.93) (1.95) (2.53) (2.89) (2.89) Adj. R 2 (%)

57 International evidence: Local and Global Factors U.S. Europe U.K. Japan Pooled Δ t 22,t Global Realized (2.80) (-0.08) (2.24) (0.34) (2.83) Global Anticipated (4.03) (4.11) (2.91) (3.05) (7.87) Δ t 22,t Local Realized Δ t 2,t Global Realized (-1.05) (0.53) (-1.80) (-0.96) Local Anticipated Global Anticipated (4.27) (0.44) (0.71) (3.63) Adj. R 2 (%)

58 Economic significance We form a simple macro timing portfolio 38

59 Economic significance We form a simple macro timing portfolio 1. Long (short) four stock market indexes in equal risk proportions if current global economic activity is above (below) level 60-day before 38

60 Economic significance We form a simple macro timing portfolio 1. Long (short) four stock market indexes in equal risk proportions if current global economic activity is above (below) level 60-day before 2. Long (short) four stock market indexes if difference between global sentiment and economic activity is positive (negative) 38

61 Economic significance We form a simple macro timing portfolio 1. Long (short) four stock market indexes in equal risk proportions if current global economic activity is above (below) level 60-day before 2. Long (short) four stock market indexes if difference between global sentiment and economic activity is positive (negative) 3. Long (short) the two stock market indices with smaller (larger) divergence between local and global economic activity. Do the opposite with net sentiment 38

62 Macro timing portfolio SR= Macro Timing Portfolio

63 Forecasting future economic growth We estimate a simple regression: Growth t+lead = α+ρgrowth t +β 1 t 22,t Realized+β 2 Anticipated t +ɛ t Growth (331.16) (88.96) (40.77) (25.31) (17.84) (13.97) (11.65) Δ t 22,t Realized (2.44) (2.16) (2.09) (2.04) (1.64) (1.21) (0.86) Anticipated (6.71) (6.31) (5.22) (4.76) (4.79) (4.62) (4.53) Adj. R 2 (%) AR Residual Adj. R 2 (%)

64 Forecasting future economic growth Panel B: Conditional on growth dispersion above median Growth (292.76) (87.34) (38.81) (23.26) (16.80) (13.01) (10.74) Δ t 22,t Realized (3.95) (4.90) (3.93) (3.60) (3.39) (2.82) (2.32) Anticipated (6.1670) (4.29) (3.50) (3.02) (3.27) (3.29) (3.21) Adj. R 2 (%) AR Residual Adj. R 2 (%) Panel C: Conditional on growth dispersion below median Growth (108.19) (31.17) (17.82) (12.53) (8.31) (6.52) (5.50) Δ t 22,t Realized (0.09) (-1.28) (-0.86) (-0.14) (-0.48) (-0.74) (-0.86) Anticipated (2.43) (3.56) (3.45) (4.05) (4.26) (3.81) (3.45) Adj. R 2 (%) AR Residual Adj. R 2 (%)

65 Conclusion Constructed real-time systematic macro factors for realized and anticipated growth (and inflation, output, employment) Constructed corresponding real-time measures of uncertainty Compared our growth factor to other nowcasting techniques Provided statistical evidence of predictability Macro factors significantly forecast stock market returns Effect is not subsumed by traditional predictors Predictability by fundamentals is state-dependent Consistent evidence internationally Demonstrated economic relevance of predictability Profitable LS country selection strategy (and in new paper: profitable cross-sectional stock selection model) 42