A Blessing or a Curse? The Impact of High Frequency Trading on Institutional Investors

Transcription

1 Second Annual Conference on Financial Market Regulation, May 1, 2015 A Blessing or a Curse? The Impact of High Frequency Trading on Institutional Investors Lin Tong Fordham University

2 Characteristics and Strategies of HFT Concept Release on Equity Market Structure, SEC (2010) Characteristics: High-speed Co-location Large intraday volume but finish the trading day as close to a neutral position as possible Popular Strategies: Passive market making (rebate capturing) Arbitrage Structural vulnerabilities: e.g. Flash orders, Michigan Consumer Confidence Index Directional trading: Order anticipation Momentum ignition (Spoofing): Flash Crash of 2010 May pose particular problems for long-term investors

3 Institutional Investors Concerns about HFT Concerns about increased trading costs Norwegian Sovereign Wealth Fund, NY Times, October 2013 Order anticipation Institutional investors questioned whether our market structure meets their need to trade efficiently and fairly, in large size speech by Mary Schapiro, former SEC Chairperson. Increased popularity of ATSs (Alternative Trading Systems) Dark pools: account for 40% of stock trading volume now - NY Times, April 6, 2013 Flash Boys boosts IEX trades 40%

4 Current Literature 1 Narrower bid ask spreads, reduced volatility, and faster price discovery Hendershott et al. (JF 2011), Hasbrouck and Saar (JFM 2013), Menkveld (JFM 2013), Brogaard, Hendershott, and Riordan (RFS 2014), Chaboud et al. (JF 2014), Boehmer et al.(2014) No additional costs to other investors: Brogaard et al. (FR 2014): London Stock Exchange HFT participate in about 20% of the trading volume in 2008 and rise to 40% in 2012 much lower than the 70% in the U.S. in 2009 Korajczyk and Murphy (2015): Canadian stock exchanges; Need to identify HFT and large trades in trade and quotes data Malinova, Park and Riordan (2013): Retail investors

5 Current Literature 2 Problems of HFT: Illusory liquidity: Kirilenko et al. (2011) and Easley et al. (2011a) Order anticipation: Hirschey (2013) Increased volatility of quotes: Hasbrouck (2014)

6 Why institutions are not better off? Institutions trade in large orders Millions of shares per order To reduce price impact, a large order is divided into small pieces Institutional trading costs Bid-ask spreads are only a small component of trading costs. Market impact can be 5-10 times of bid-ask spreads major component of institutions trading costs.

7 This Paper (1) Examines the impact of HFT on trading costs of traditional institutional investors in the U.S market Mutual funds, pension funds, endowments, foundations (low frequency traders) I find that HFT increases the trading costs of institutional investors. One STDEV. increase of HFT activity is associated with an increase in average trading costs by 1/3. Various analyses to rule out reverse causality (i.e., HFT attracted to stocks with high trading costs)

8 This Paper (2) When and how does HFT increase institutional investors costs? Opportunistic liquidity provision when there is large institutional buy-sell imbalance Short-lived liquidity: when institutional imbalance is higher, HFT more intense, but HF traders keep zero positions at market close Expensive: HFT impact on trading costs is strongest when institutions have large buying imbalance Directional trading: momentum ignition or order anticipation HFT impact most pronounced when HF trade directions are not random Less negative effects on institutions with good trading skills

9 HFT Data NASDAQ HFT data All trades on a sample of 120 randomly selected stocks, ; 40 largecap, 40 midcap, and 40 smallcap Identify trades by HF traders and non-hf traders Papers that use this dataset: e.g., Brogaard, Hendershott and Riordan (RFS 2014) Measure of HFT intensity HFT Intensity: total HFT volume for stock i on day t, divided by the stock s average trading volume in past 30 days Why not separating HFT liquidity supply/demand? An HFT strategy can be combine with market and limit orders, e.g., order anticipation

10 Institutional Transaction Data Ancerno institutional transaction dataset Formerly known as Abel/Noser (Anand et al., RFS 2012, JFE 2013) Execution-level data, including institution identity code, stock ticker, stock price at placement, execution price, number of shares executed, direction of trades, etc. 204 institutions traded on the 120 stocks during sample period

11 Measure of Trading Costs Execution Shortfall = [(P 1 P 0 )/P 0 ] D P 1 : value-weighted execution price of the order P 0 : price at time of order placement D = 1 for buy order and D = 1 for sell order Captures bid-ask spreads, market impact, and price slippage Volume-weighted average over all orders for a given stock on a given day process.pdf

12 T1: Summary Statistics All Large Cap Mid Cap Small Cap Average Market Capital (in billions) Average HFT Total Trading Volume (in millions) Average Execution Shortfall (in percentage) Amihud Illiquidity Measeure E Average Institutional Order Size 244, , ,823 63,943 Average Trades Per Order

13 Execution Shortfall (in %) HFT intensity Fig 1,2: HFT, Liquidity, and Execution Shortfall HFT is positively correlated with liquidity Trading costs are negatively correlated with liquidity. HFT intensity and liquidity Large Stocks Mid Stocks Small Stocks Illiquid stocks Morderate liquid stocks Liquid stocks Execution shortfall and liquidity Large Stocks Mid Stocks Small Stocks Illiiquid stocks Morderate liquid stocks Liquid stocks

14 Execution Shortfall (in %) Fig 3: HFT, Liquidity, and Execution Shortfall (Cont d) Trading costs are positively correlated with HFT intensity HFT intensity and execution shortfall Large Stocks Mid Stocks Small Stocks Low HFT intensity Mid HFT intensity High HFT intensity

15 Multivariate Panel Regression Execution Shortfall it = α i +γ t +a HFT Intensity it +b X it +ɛ it HFT Intensity it : HFT activity for stock i on day t Control variables X it : 1) firm size, 2) book-to-market ratio, 3) event dummy, 4) daily return volatility, 5) absolute daily institutional trading imbalance, 6) prior 1-day, 1-month, and 12-month stock returns, 7) Amihud illiquidity ratio, 8) daily dollar turnover, 9) average institutional order size, 10) absolute institutional imbalance, 11) average trades per order, 12) prior 1-month market volatility, 13) lagged market daily return Firm-fixed effects (α i ) and day-fixed effects (γ t ) Two-way clustered standard errors

16 T3: HFT s Impact on Execution Shortfall Dependent Variable Execution Shortfall Execution Shortfall Coefficient t-value Coefficient t-value Intercept (0.24) (-1.77) HFT Intensity (4.48) (3.37) Log Market Cap (-0.66) (1.08) Book-to-Market Ratio (-0.95) (1.23) Prior 1-day Return (-0.24) (-0.64) Prior 1-month Return (0.25) (-0.69) Prior 12-month Return (0.92) (-0.26) Amihud Illiquidity Ratio (3.14) (3.36) Daily Return Volatility (1.42) (0.30) Daily Dollar Turnover (-1.66) (-0.19) Average Institutional Order Size (1.37) (1.42) Absolute Institutional Imbalance (2.56) (2.67) Average Trades Per Order (0.16) (-0.44) Prior 1-month Market Volatility (3.24) Prior 1-day Market Return (-0.05) Day-fixed Effects No Yes Stock-fixed Effects No Yes Two-way Clustered Standard Deviations Yes Yes

17 Endogeneity Reverse causality: is HFT more prevalent among stocks with high trading costs? Unlikely since HFT likes liquidity HFT are most popular among large and liquid stocks (Fig. 1) Unlikely since HFT withdrew at the 2008 short selling ban when institutioanl trading costs are high. Unlikely since Granger causality test results Omitted variables: are there any unobserved factors that influence both HFT and institutional trading costs? Control for time- and firm-fixed effects (Table 3) Control for stock and institutional trading characteristics documented to affect liquidity/trading costs (Table 3) Control for events that may jointly affect HFT and trading costs (earnings announcements and M&A announcements)

18 T5: HFT s Impact on Execution Shortfall on Event Days Dependent Variable Execution Shortfall Coefficient t-value Intercept (1.74) HFT Intensity*Event Dummy (1.29) HFT Intensity*No Event Dummy (3.88) Event Dummy (1.39) Log Market Cap (1.03) Book-to-Market Ratio (1.23) Prior 1-day Return (0.65) Prior 1-month Return (0.70) Prior 12-month Return (0.31) Amihud Illiquidity Ratio (3.37) Daily Return Volatility (0.26) Daily Dollar Turnover (0.24) Average Institutional Order Size (1.40) Absolute Institutional Imbalance (2.69) Average Trades Per Order (0.49) Day-fixed Effects Yes Stock-fixed Effects Yes Two-way Clustered Standard Deviations Yes Adjusted R-squared (%) 3.49 Number of Observations 54963

19 Fig 4,5: Short selling Ban - Sep 19, stocks in my sample are subjected to the ban. Excution shortfall (in %) for banned and unbanned stocks /04/08 09/09/08 09/14/08 09/19/08 09/24/08 09/29/08 10/04/08 10/09/08 10/14/08 10/19/ Banned Exec. Shortfall Unanned Exec. Shortfall HFT Activity for Banned and Unbanned Stocks /09/08 09/09/08 14/09/08 19/09/08 24/09/08 29/09/08 04/10/08 09/10/08 14/10/08 19/10/08 Banned HFT Unbanned HFT

20 Granger Causality Test Determining causal relations between HFT and trading cost (ES) in statistical lead/lag sense ( ) ( ) ( ) ( ) ESi,t a1,i b11,i b = + 12,i ESi,t 1 HFT i,t a 2,i b 21,i b 22,i HFT i,t 1 ( ) ɛ1,i,t + ɛ 2,i,t b 12,i 0: HFT Granger causes ES; b 21,i 0: ES Granger causes HFT Making inference on causalities jointly on 120 stocks: Estimating above VAR(1) for all 120 stocks; obtain sample cross-sectional averages of b 12,i and b 21,i ; Compare the sample averages with bootstrapped averages of b 12,i and b 21,i Bootstraps are performed under the null of no causality, while keeping the correlations between residuals across stocks

21 T6: Granger Causality Test: Results Results: Variable: b12 Q1 Mean Median Q3 Sample Coefficients Sample t-statistic (-0.456) (0.311) (0.265) (0.977) Bootstraped p-value [0.043] [0.002] [0.010] [0.008] Variable: b21 Q1 Mean Median Q3 Sample Coefficients Sample t-statistic (-0.725) (0.039) (-0.031) (0.793) Bootstraped p-value [0.695] [0.341] [0.583] [0.141] HFT Granger cause trading costs, but not vice versa

22 Further Analysis How and when HFT affects institutional trading costs Liquidity provision when institutions have large trade imbalance Directional trading: momentum ignition or order anticipation Less negative effects on institutions with good trading skills

23 HFT and Institutional Buy-Sell Imbalance Does HFT provide liquidity when institutions have large trade imbalance? Sorted portfolio analysis: each day, sort all stocks by size and institutional trade imbalance. HFT intensity, HFT imbalance Panel regression on trading costs in subsamples with different levels of institutional trade imbalance

24 T9: HFT and Institutions Buy-Sell Imbalance (Sorted Portfolio) Panel A: Distribution of HFT and institution buy-sell imbalance Mean Q1 Median Q3 HFT Buy-Sell Imbalance Institution Buy-Sell Imbalance Panel B: Institutional buy-sell imbalance Institutions net selling Institutions balanced Institutions net buying Large Stocks Mid Stocks Small Stocks Panel C: HFT intensity Institutions net selling Institutions balanced Institutions net buying Large Stocks Mid Stocks Small Stocks Panel D: HFT buy-sell imbalance Institutions net selling Institutions balanced Institutions net buying Large Stocks Mid Stocks Small Stocks

25 T10: Impact of HFT and Institutional Buy-Sell Imbalance Dependent Variable Execution Shortfall Institutions net selling Institutions balanced Institutions net buying Coefficient t-value Coefficient t-value Coefficient t-value Intercept (2.82) (-1.37) (-2.86) HFT Intensity (-1.77) (2.24) (4.78) Log Market Cap (-2.79) (1.18) (2.36) Book-to-Market Ratio (2.72) (0.34) (2.41) Prior 1-day Return (1.50) (-0.99) (-1.01) Prior 1-month Return (1.40) (-0.45) (-1.22) Prior 12-month Return (1.76) (0.61) (-1.31) Amihud Illiquidity Ratio (1.38) (0.96) (-1.31) Daily Return Volatility (0.89) (-0.75) (0.20) Daily Dollar Turnover (2.36) (-1.05) (-2.14) Average Institutional Order Size (1.04) (0.30) (1.11) Absolute Institutional Imbalance (3.12) (2.02) (2.09) Average Trades Per Order (0.12) (-1.28) (-0.23) Day-fixed Effects Yes Yes Yes Stock-fixed Effects Yes Yes Yes Two-way Clustered Std. Yes Yes Yes

26 Detecting Directional Trading HFT strategies and randomness of HF trade directions (buy or sell): Market making: B-S-B-S-B-S-B-S-B-S Directional trading (momentum ignition or order anticipation): B-B-B-B-B-S-S-S-S-S Random walk: B-S-S-B-B-B-S-B-S-S Detecting HFT strategies: Runs test (classic non-parametric test of random walks) Market making: negative & significant Directional trading: positive & significant Random walk: not significant

27 T11: Directional HF Trades Dependent Variable Execution Shortfall Directional Market Making Random Walk Coefficient t-value Coefficient t-value Coefficient t-value Intercept (0.14) (-0.18) (-1.44) HFT Intensity (2.60) (1.94) (1.64) Log Market Cap (-0.30) (0.69) (1.63) Book-to-Market Ratio (2.49) (0.42) (-0.15) Prior 1-day Return (0.21) (-0.65) (-0.66) Prior 1-month Return (-0.20) (0.38) (-1.48) Prior 12-month Return (1.38) (0.03) (-0.85) Amihud Illiquidity Ratio (4.61) (2.43) (1.09) Daily Return Volatility (-1.41) (0.69) (0.62) Daily Dollar Turnover (-1.70) (0.43) (0.92) Average Institutional Order Size (0.95) (-1.45) (2.95) Absolute Institutional Imbalance (1.18) (3.83) (0.88) Average Trades Per Order (-0.53) (-0.20) (-0.39) Day-fixed Effects Yes Yes Yes Stock-fixed Effects Yes Yes Yes Two-way Clustered Std. Yes Yes Yes

28 T11: Heterogeneity of the impact of HFT Institutions with better trading skills are less affected by HFT Dependent Variable Execution Shortfall Coefficient t-value Intercept (-1.74) HFT Intensity (4.01) HFT Intensity High Trading Skill Dummy (-2.55) High Trading Skill Dummy (-0.19) Log Market Cap (0.88) Book-to-Market Ratio (0.99) Prior 1-day Return (-0.81) Prior 1-month Return (-0.98) Prior 12-month Return (-0.32) Amihud Illiquidity Ratio (3.72)) Daily Return Volatility (0.51) Daily Dollar Turnover (0.35) Average Institutional Order Size (0.96) Absolute Institutional Imbalance (2.74) Average Trades Per Order (-0.85) Day-fixed Effects Stock-fixed Effects Two-way Clustered Standard Deviations Yes Yes Yes

29 The negative effects of HFT is weaker on institution with good trading skills. Conclusions HFT significantly increases the trading costs of institutional investors Robust to the controls of stable stock liquidity characteristics and events that might jointly affect HFT and trading costs Analyses on the Short-selling ban and Granger causality tests further rule out reverse causality/simultaneity HFT serves as an opportunistic liquidity provision when institutional investors have large trade imbalances (especially for imbalance on buy side) The impact of HFT is most pronounced when their trading is directional (likelihood of momentum ignition or order anticipation)