order dynamics recent evidence from the nyse by ellul et al. (2007 jef)
TRANSCRIPT
Order Dynamics: Recent Evidence from the
NYSE
Andrew Ellul, Indiana UniversityCraig W. Holden, Indiana UniversityPankaj Jain, University of MemphisRobert Jennings, Indiana University
Goal and Motivation Empirically characterize traders’ order
choice under dynamic market conditions Order submission and cancellation
decisions determine liquidity-provision and liquidity-usage better understanding of the price formation
process determinants of execution quality
Analyze NYSE in 2001 All trades are in decimals Allows automatic execution (Direct+) Have all SuperDOT orders, but no orders
worked by floor brokers
Theoretical Predictions Tested
Negative Serial Correlation in Order Type -Parlour 98
Prior trader submits market buy consumes liquidity relatively more attractive to submit limit sell replenish liquidity
Cycle of consuming and replenishing liquidity negative serial correlation in order type (MB less likely to follow MB)
Puzzle: Biais, Hillion and Spatt (1995) found positive serial correlation in order type (MB more likely to follow MB) LOB will grow more imbalanced and eventually break down
Trader’s Patience – Handa and Schwartz (1996) Model trader’s timeframe for completing trade Impatient choose market orders; Patient choose limit order Extend logic highly impatient choose fast auto ex with little
regard for market conditions; More patient choose floor executions with meaningful regard for market conditions
More Literature Other Variables Suggested By Theory
Depth – Parlour (1998) Time-of-day – Harris (1998), Parlour (1998), Bloomfield,
O’Hara and Saar (2002) and others Quoted Spread – Cohen, Maier, Schwartz, Witcomb
(1981), Harris (1998), Foucault (1999) and others Volatility – Foucault (1999) Market and Own Return – Lo, Mamaysky and Wang
(2000) Related Empirical Studies
Harris (1998) Hollifield, Miller and Sandas (1999,2002) Ahn, Bae and Chan (2001) Bae, Jang and Park (2002) NYSE using 90-91 TORQ Beber and Caglio (2002) NYSE using 90-91 TORQ Hasbrouck and Saar (2002) Ranaldo (2002)
Data NYSE System Order Database (SOD)
over April – June 2001 148 stocks stratified by price and
volume Detailed info on:
order submission (time, type, size, etc.) order’s outcome: execution,
cancellation, or expiration
Hypotheses 1 Negative Serial Correlation in Order
Type for Individual Orders Hypothesis: Prior limit buy (sell) lower prob limit buy (sell)
Negative Serial Correlation in Order Type for Aggregated Order Flows Hypothesis: Higher Δ limit buys (sells) in prior 5 min
lower Δ limit buys (sells) in current 5 min
Short-term Forecasting Hypothesis: Large ask (bid) depth prob limit sell (buy)
and prob market sell (buy)
Hypotheses 2 Jump-The-Queue Hypothesis:
Large ask (bid) depth Prob inside limit sell (buy) and prob at or behind limit sell (buy)
Time of Day Hypothesis: Time Prob various order types
changes
Trader Impatience Hypothesis: Auto-ex sensitivity to market conditions
< floor sensitivity to market conditions
Methodology Analyze choices of representative trader
= weighted average over all trader types Multinomial logit with 7-way event
structure Market Buy, Market Sell, Limit Buy, Limit
Sell, Cancel Buy, Cancel Sell, No Activity (45% of limit orders are cancelled) No activity event = stock-specific amount of
clock time with no order submissions or cancellations
Median time between successive order events or 5 minutes, whichever is less
For 8 stocks = 1 second; for 50 stocks = 5 minutes
By Aggregated Order Flow Aggregate buys and sells Aggregate same 7 events by 5 min
int. Market buy, market sell, limit buy, limit
sell, cancel buy, cancel sell, no activity Estimate 7 OLS equations with the
same independent variables aggregated by 5 min int
13-way Multinomial Logit Dependent variable = 13-way event
structure
4 types of Limit Buys, 4 types of Limit Sells, Market Buy, Market Sell, Cancel Buy, Cancel Sell, No Activity Event
Limit Buy Limit Sell
Above $30.10 Marketable Worse-than-the-quote
Ask = $30.10 Marketable At-the-quote
$30.01 - $30.09
Inside-the-quote Inside-the-quote
Bid = $30.00 At-the-quote Marketable
Below $30.00 Worse-than-the-quote
Marketable
More aggressive
More aggressive
Explanatory Variables Last Event Market Buy (dummy) Last Event Market Sell (dummy) Last Event Limit Buy (dummy) Last Event Limit Sell (dummy) Last Event Cancel Buy (dummy) Last Event Cancel Sell (dummy) Relative National Best Bid size (/ shrs outstand) Relative National Best Ask size (/ shrs
outstand) Time (number of 5 min since midnight)
Control Variables Percent Spread Relative Volume (ln vol(t-5)/Shrs
outstand) Own Return (% chg in midpoint (5 min)) Own Return Squared Volatility Market Return (5 min % chg SP500 ETF) Time Squared (deviation from midday
sqr) U-shape patterns Day Return
Order by Order 13 way: positive serial correlation
Change in aggregated order flow: negative correlation
Conclusion Neg Serial Corr in Order Type by Ind Orders = Reject Neg Serial Corr in Order Type by Agg Flows=Support Resolves puzzle and supports Parlour model Short-term Forecasting = Support Jump-the-Que = Support Depth impacts in both ways Time-of-day = Support Supports Bloomfield, O’Hara, and Saar
experimental evidence Trader Impatience = Support Supports Handa & Schwartz logic applied to Auto-ex