Momentum Strategies in Futures Markets and Trend-following Funds

Transcription

1 Momentum Strategies in Futures Markets and Trend-following Funds Akindynos-Nikolaos Baltas and Robert Kosowski Imperial College London 2012 BK (Imperial College London) Momentum Strategies in Futures Markets / 27

2 Motivation Capacity constraints have limited these funds in the past. [...] It s a problem for trend-followers: the larger they get, the more di cult it is to maintain the diversity of their trading books. While equity or bond futures markets are deep and liquid, markets for most agricultural contracts -soy or wheat, for example- are less so. (The Financial Times, November 27, 2011, Winton s head is a proud speculator,) We carry out (i) in-depth study of momentum strategy (larger cross-section, more frequencies, longer sample period) and (ii) analysis of capacity constraints in these strategies, which matter for investors Momentum phenomenon widespread and CTA s a popular investment in recent years We combine research on momentum strategies in futures markets with research on hedge funds/ctas BK (Imperial College London) Momentum Strategies in Futures Markets / 27

3 Presentation Outline Motivation Related Literature Methodology Data Empirical Results Conclusions We thank INQUIRE Europe for nancial support BK (Imperial College London) Momentum Strategies in Futures Markets / 27

4 Main Findings In depth analysis of momentum strategies (larger cross-section, longer sample, more frequencies) BK (Imperial College London) Momentum Strategies in Futures Markets / 27

5 Main Findings In depth analysis of momentum strategies (larger cross-section, longer sample, more frequencies) Low correlation of monthly, weekly daily strategies BK (Imperial College London) Momentum Strategies in Futures Markets / 27

6 Main Findings In depth analysis of momentum strategies (larger cross-section, longer sample, more frequencies) Low correlation of monthly, weekly daily strategies Evidence of serial correlation and implications for EMH BK (Imperial College London) Momentum Strategies in Futures Markets / 27

7 Main Findings In depth analysis of momentum strategies (larger cross-section, longer sample, more frequencies) Low correlation of monthly, weekly daily strategies Evidence of serial correlation and implications for EMH Evidence of counter-cyclical performance of MWD; MWD not explained by standard factor models BK (Imperial College London) Momentum Strategies in Futures Markets / 27

8 Main Findings In depth analysis of momentum strategies (larger cross-section, longer sample, more frequencies) Low correlation of monthly, weekly daily strategies Evidence of serial correlation and implications for EMH Evidence of counter-cyclical performance of MWD; MWD not explained by standard factor models CTA indices statistically and economically signi cantly exposed to M, W and D BK (Imperial College London) Momentum Strategies in Futures Markets / 27

9 Main Findings In depth analysis of momentum strategies (larger cross-section, longer sample, more frequencies) Low correlation of monthly, weekly daily strategies Evidence of serial correlation and implications for EMH Evidence of counter-cyclical performance of MWD; MWD not explained by standard factor models CTA indices statistically and economically signi cantly exposed to M, W and D No evidence of capacity constraints: negative M,W,D performance - ow relationship, but time-varying and not signi cant; consistent with liquid futures markets BK (Imperial College London) Momentum Strategies in Futures Markets / 27

10 Main Findings In depth analysis of momentum strategies (larger cross-section, longer sample, more frequencies) Low correlation of monthly, weekly daily strategies Evidence of serial correlation and implications for EMH Evidence of counter-cyclical performance of MWD; MWD not explained by standard factor models CTA indices statistically and economically signi cantly exposed to M, W and D No evidence of capacity constraints: negative M,W,D performance - ow relationship, but time-varying and not signi cant; consistent with liquid futures markets Implications for CTA and HF investors. BK (Imperial College London) Momentum Strategies in Futures Markets / 27

11 Related Literature Cross-sectional Momentum: BK (Imperial College London) Momentum Strategies in Futures Markets / 27

12 Related Literature Cross-sectional Momentum: Equities (Jegadeesh and Titman, 1993, 2001), futures (Pirrong, 2005, Mi re and Rallis, 2007), currency markets (Menkho, Sarno, Schmeling and Schrimpf, 2012), everywhere (Asness, Moskowitz and Pedersen, 2009) BK (Imperial College London) Momentum Strategies in Futures Markets / 27

13 Related Literature Cross-sectional Momentum: Equities (Jegadeesh and Titman, 1993, 2001), futures (Pirrong, 2005, Mi re and Rallis, 2007), currency markets (Menkho, Sarno, Schmeling and Schrimpf, 2012), everywhere (Asness, Moskowitz and Pedersen, 2009) Time-series momentum in futures markets BK (Imperial College London) Momentum Strategies in Futures Markets / 27

14 Related Literature Cross-sectional Momentum: Equities (Jegadeesh and Titman, 1993, 2001), futures (Pirrong, 2005, Mi re and Rallis, 2007), currency markets (Menkho, Sarno, Schmeling and Schrimpf, 2012), everywhere (Asness, Moskowitz and Pedersen, 2009) Time-series momentum in futures markets Moskowitz, Ooi and Pedersen (2012), Burnside et al (2011) BK (Imperial College London) Momentum Strategies in Futures Markets / 27

15 Related Literature Cross-sectional Momentum: Equities (Jegadeesh and Titman, 1993, 2001), futures (Pirrong, 2005, Mi re and Rallis, 2007), currency markets (Menkho, Sarno, Schmeling and Schrimpf, 2012), everywhere (Asness, Moskowitz and Pedersen, 2009) Time-series momentum in futures markets Moskowitz, Ooi and Pedersen (2012), Burnside et al (2011) Rational and behavioural explanations of serial correlation in returns BK (Imperial College London) Momentum Strategies in Futures Markets / 27

16 Related Literature Cross-sectional Momentum: Equities (Jegadeesh and Titman, 1993, 2001), futures (Pirrong, 2005, Mi re and Rallis, 2007), currency markets (Menkho, Sarno, Schmeling and Schrimpf, 2012), everywhere (Asness, Moskowitz and Pedersen, 2009) Time-series momentum in futures markets Moskowitz, Ooi and Pedersen (2012), Burnside et al (2011) Rational and behavioural explanations of serial correlation in returns Rational (Berk et al. (1999), Johnson (2002), Ahn, Conrad and Dittmar (2003), Sagi and Seasholes (2007), Liu and Zhang (2008)) and the behavioural explanations (e.g. Barberis et al. 1998, Daniel et al., 1998, Hong and Stein, 1999) BK (Imperial College London) Momentum Strategies in Futures Markets / 27

17 Related Literature Cross-sectional Momentum: Equities (Jegadeesh and Titman, 1993, 2001), futures (Pirrong, 2005, Mi re and Rallis, 2007), currency markets (Menkho, Sarno, Schmeling and Schrimpf, 2012), everywhere (Asness, Moskowitz and Pedersen, 2009) Time-series momentum in futures markets Moskowitz, Ooi and Pedersen (2012), Burnside et al (2011) Rational and behavioural explanations of serial correlation in returns Rational (Berk et al. (1999), Johnson (2002), Ahn, Conrad and Dittmar (2003), Sagi and Seasholes (2007), Liu and Zhang (2008)) and the behavioural explanations (e.g. Barberis et al. 1998, Daniel et al., 1998, Hong and Stein, 1999) Volatility and return sign predictability (Christo ersen and Diebold, 2006, and Christo ersen, Diebold, Mariano, Tay and Tse, 2007) BK (Imperial College London) Momentum Strategies in Futures Markets / 27

18 Related Literature Cross-sectional Momentum: Equities (Jegadeesh and Titman, 1993, 2001), futures (Pirrong, 2005, Mi re and Rallis, 2007), currency markets (Menkho, Sarno, Schmeling and Schrimpf, 2012), everywhere (Asness, Moskowitz and Pedersen, 2009) Time-series momentum in futures markets Moskowitz, Ooi and Pedersen (2012), Burnside et al (2011) Rational and behavioural explanations of serial correlation in returns Rational (Berk et al. (1999), Johnson (2002), Ahn, Conrad and Dittmar (2003), Sagi and Seasholes (2007), Liu and Zhang (2008)) and the behavioural explanations (e.g. Barberis et al. 1998, Daniel et al., 1998, Hong and Stein, 1999) Volatility and return sign predictability (Christo ersen and Diebold, 2006, and Christo ersen, Diebold, Mariano, Tay and Tse, 2007) Capacity constraints in hedge fund strategies and ow-performance BK (Imperial College London) Momentum Strategies in Futures Markets / 27

19 Related Literature Cross-sectional Momentum: Equities (Jegadeesh and Titman, 1993, 2001), futures (Pirrong, 2005, Mi re and Rallis, 2007), currency markets (Menkho, Sarno, Schmeling and Schrimpf, 2012), everywhere (Asness, Moskowitz and Pedersen, 2009) Time-series momentum in futures markets Moskowitz, Ooi and Pedersen (2012), Burnside et al (2011) Rational and behavioural explanations of serial correlation in returns Rational (Berk et al. (1999), Johnson (2002), Ahn, Conrad and Dittmar (2003), Sagi and Seasholes (2007), Liu and Zhang (2008)) and the behavioural explanations (e.g. Barberis et al. 1998, Daniel et al., 1998, Hong and Stein, 1999) Volatility and return sign predictability (Christo ersen and Diebold, 2006, and Christo ersen, Diebold, Mariano, Tay and Tse, 2007) Capacity constraints in hedge fund strategies and ow-performance Jylha and Suominen (2011), Della Corte et al (2011), Koijen and Vrugt (2011), Naik, Ramadorai and Stromqvist (2007) BK (Imperial College London) Momentum Strategies in Futures Markets / 27

20 Data Futures Data Tick data; Dec Jan. 2012; futures prices for 71 assets(26 commodities, 23 equity indices, 7 currencies and 15 intermediate-term and long-term bonds) Construct returns following MOP (2012): roll so that we trade most liquid contract; Ratio-adjust backwards prices; CTA/Hedge fund data: BarclayHedge database: see Joenvaara, Kosowski and Tolonen (2012) for comparison of di erent data bases; 3834 unique CTA funds between June 1959 and January 2012; total AUM end of 2012 of $444 billion (down from $507 billion in August 2011); CTA index data: BarclayHedge CTA index as well as our AuM-weighted index of CTA universe BK (Imperial College London) Momentum Strategies in Futures Markets / 27

21 CTAs: Number, AuM and Flows (F5) BK (Imperial College London) Momentum Strategies in Futures Markets / 27

22 Methodology Construction of the return series of the (aggregate) time-series momentum strategy: R K J = 1 M t M t SIGN i (t i=1 J, t) 40% σ i (t; D) R i (t, t + K ), (1) where M t is the number of available assets at time t, σ i (t; D) denotes an estimate at time t of the realized volatility of the i th asset Volatility Estimates: Volatility estimates a ect turnover and transaction costs We use Yang and Zhang (2000) volatility estimator: σ 2 YZ (t; D) = σ2 OPEN (t; D) + kσ2 STDEV (t; D) + (1 k) σ2 RS (t; D) with 60-day estimates of volatility. YZ (2000) most e cient estimator in a pool of range estimators; unbiased volatility estimator that is independent of both the opening jump and the drift of the underlying price process (Shu and Zhang (2006) and Baltas (2011)) BK (Imperial College London) Momentum Strategies in Futures Markets / 27

23 Serial Correlation and Return Predictability Assess amount of return predictability that is inherent in lagged returns on the monthly, weekly and daily frequencies by running the following pooled time-series cross-sectional R (t 1, t) σ YZ (t 1; 60) = α + β R (t h 1, t h) λ σ YZ (t h 1; 60) + ɛ (t), So we regress the excess return for an instrument in month t on its return lagged h months, where both returns are scaled by their ex ante volatilities The quantity of interest in these regressions is the t-statistic of the coe cient β λ for each lag. Large and signi cant t-statistics essentially support the hypothesis of time-series return predictability. Results show signi cant t-statistics for monthly frequency for rst 12 months (Figure 1, Panel A); clustered around two distinct periods for weekly frequency (Panel B); also two subperiods for daily frequency (Panel C) BK (Imperial College London) Momentum Strategies in Futures Markets / 27

24 Preliminary Evidence of Return Predictability (Figure 1) BK (Imperial College London) Momentum Strategies in Futures Markets / 27

25 Momentum Pro tability Following evidence of return predictability proceed to construction of time-series momentum strategies for a grid of lookback (J) and investment periods (K) Instead of forming a new momentum portfolio every K periods, when the previous portfolio is unwound, we follow overlapping methodology MWD: Panel A presents the results for the monthly strategy and K, J = f1, 3, 6, 9, 12, 24, 36g months Panel B presents the results for the weekly strategy and K, J = f1, 2, 3, 4, 6, 8, 12g weeks Panel C presents the results for the daily strategy and K, J = f1, 3, 5, 10, 15, 30, 60g days. time-series momentum strategy generates a statistically and economically signi cant mean return and alpha for all three rebalancing frequencies. BK (Imperial College London) Momentum Strategies in Futures Markets / 27

26 Momentum Pro tability - Monthly Frequency (T3A) BK (Imperial College London) Momentum Strategies in Futures Markets / 27

27 Sharpe Ratios for MWD (T3 A, B and C) BK (Imperial College London) Momentum Strategies in Futures Markets / 27

28 Best Performing Strategies Pairs (12, 1), (9, 3) and (1, 12) as the three best monthly strategies Weekly: best strategy pair is (8, 1), followed by the pairs (12, 2) and (1, 8) We decide to work with the strategies M12 1, W 8 1 remaining of the paper (MWD in short) and D1 15 for the BK (Imperial College London) Momentum Strategies in Futures Markets / 27

29 Summary Statistics for Best Strategies - Table 4A BK (Imperial College London) Momentum Strategies in Futures Markets / 27

30 Summary Statistics for Best Strategies - Table 4B BK (Imperial College London) Momentum Strategies in Futures Markets / 27

31 Decomposition of the MWD Strategies T5 (part 1) BK (Imperial College London) Momentum Strategies in Futures Markets / 27

32 Decomposition of the MWD Strategies T5 (part 2) BK (Imperial College London) Momentum Strategies in Futures Markets / 27

33 Historical Performance of Momentum Strategies (F2) BK (Imperial College London) Momentum Strategies in Futures Markets / 27

34 Sharpe Ratios and Correlations of Univariate Time-Series Momentum Strategies (F3) BK (Imperial College London) Momentum Strategies in Futures Markets / 27

35 Comparison of MWD to CTA Indices over time (T6A) BK (Imperial College London) Momentum Strategies in Futures Markets / 27

36 Comparison of MWD to CTA Indices over time (T6B) BK (Imperial College London) Momentum Strategies in Futures Markets / 27

37 Return Decomposition of the AUM-Weighted CTA Index (T7) BK (Imperial College London) Momentum Strategies in Futures Markets / 27

38 60-Month Rolling adjusted R2 (F4) BK (Imperial College London) Momentum Strategies in Futures Markets / 27

39 Time-Series Momentum Pro tability and CTA Fund Flows (T8) BK (Imperial College London) Momentum Strategies in Futures Markets / 27

40 Rolling t-statistics of CTA Fund Flow Variables (F6) BK (Imperial College London) Momentum Strategies in Futures Markets / 27

41 Conclusions In depth analysis of momentum strategies (larger cross-section, longer sample, more frequencies) BK (Imperial College London) Momentum Strategies in Futures Markets / 27

42 Conclusions In depth analysis of momentum strategies (larger cross-section, longer sample, more frequencies) Evidence of serial correlation and implications for EMH BK (Imperial College London) Momentum Strategies in Futures Markets / 27

43 Conclusions In depth analysis of momentum strategies (larger cross-section, longer sample, more frequencies) Evidence of serial correlation and implications for EMH Evidence of counter-cyclical performance of MWD; low correlation between M,W and D BK (Imperial College London) Momentum Strategies in Futures Markets / 27

44 Conclusions In depth analysis of momentum strategies (larger cross-section, longer sample, more frequencies) Evidence of serial correlation and implications for EMH Evidence of counter-cyclical performance of MWD; low correlation between M,W and D MWD not explained by standard factor models BK (Imperial College London) Momentum Strategies in Futures Markets / 27

45 Conclusions In depth analysis of momentum strategies (larger cross-section, longer sample, more frequencies) Evidence of serial correlation and implications for EMH Evidence of counter-cyclical performance of MWD; low correlation between M,W and D MWD not explained by standard factor models CTA indices statistically and economically signi cantly exposed to M, W and D BK (Imperial College London) Momentum Strategies in Futures Markets / 27

46 Conclusions In depth analysis of momentum strategies (larger cross-section, longer sample, more frequencies) Evidence of serial correlation and implications for EMH Evidence of counter-cyclical performance of MWD; low correlation between M,W and D MWD not explained by standard factor models CTA indices statistically and economically signi cantly exposed to M, W and D No evidence of capacity constraints: negative M,W,D performance - ow relationship, but time-varying and not signi cant; consistent with liquid futures markets BK (Imperial College London) Momentum Strategies in Futures Markets / 27

47 Conclusions In depth analysis of momentum strategies (larger cross-section, longer sample, more frequencies) Evidence of serial correlation and implications for EMH Evidence of counter-cyclical performance of MWD; low correlation between M,W and D MWD not explained by standard factor models CTA indices statistically and economically signi cantly exposed to M, W and D No evidence of capacity constraints: negative M,W,D performance - ow relationship, but time-varying and not signi cant; consistent with liquid futures markets Implications for CTA and HF investors. BK (Imperial College London) Momentum Strategies in Futures Markets / 27