Trading Costs and Intraday Patterns

1
Trading Costs and Intraday Patterns

• Rotman Distinguished Lecture Series
• March 2008
• Ingrid M. Werner
• Martin and Andrew Murrer Professor of Finance
• Fisher College of Business, The Ohio State
  University

2
Empirical Predictions Trading Costs

• Measuring trading costs using only trades
• Measuring trading costs by referencing quotes
• Trading costs should be increasing in trade size
• Spreads reflect several components
• Order processing costs
• Inventory risk
• Adverse selection

3
Roll (1984)Trade-Based Transactions Costs

• How should we measure transactions costs from
  trade-by-trade data (no quotes)?
• Roll (1984) came up with a clever technique that
  uses the idea that trade-prices should have a
  negative autocorrelation on average due to
  bid-ask bounce.
• In practice, the spread varies over the day and
  there are lots of continuations
• Tested in Harris (1990)

½
1
qt
-1
½
qt is independent of ut and iid
4
Transactions Costs using Quotes

• It is not so obvious how we should measure
  transactions costs!
• Petersen Fialkowski (1994) emphasized the
  difference between the quoted spread, effective
  spread, and realized spread
• Should we weigh measures?
• To determine whether a particular trade is a buy
  or a sell, we need to use a trade-classification
  rule.
• Lee Ready (1991) proposed to use the following
  simple rule
• A trade above the mid-quote is a buy
• A trade below the mid-quote is a sell
• A trade at the mid-quote is classified based on a
  tick-test (the direction of recent prices)
• Trade are matched to quotes lagged 5 seconds

1
3
2

• Is a buy
• Is a sell
• Is a buy (uptick)

5
Comparative Studies of Trading Costs

• For example, Huang and Stoll (1996) compare
  trading costs for Nasdaq and NYSE firms in 1991
  based on a matched sample
• Matching is on SIC code, LT debt, closing price,
  SHOUT, and BV equity
• NYSE quoted and effective spread are lower than
  Nasdaq quoted and effective spreads.
• Price impact of trades are larger on NYSE than on
  Nasdaq, so differences in adverse selection
  cannot explain result
• Rolls spread is also larger on Nasdaq than on
  the NYSE
• Cross-sectional regressions suggest that the
  observed differences cannot be accounted for by
  the matching techniqueNYSE trading costs are
  lower than Nasdaq trading costs???

6
Updated Trade Classification Algorithms

• More recently, several authors have proposed more
  reliable ways of classifying trades as buyer and
  seller initiated.
• The LR (1991) algorithm does poorly for trades
  that are within the spread, and for trades
  outside of the spread.
• Odders-White (2000) discusses the occurrence and
  consequences of trade misclassification of NYSE
  stocks.
• Werner (2003) uses NYSE audit trail data to
  analyze traing costs.
• Actual effective spreads for liquidity-demanding
  orders are about 50 less than those measured
  using the LR (1991) algorithm.
• Trade composition (SP, FB, MO, LO, etc) matters
  for costs
• Ellis, Michaely, and OHara (2001) propose an
  alternative trade-classification method for
  Nasdaq stocks.
• Classify a trade at the ask as a buy and a trade
  at the bid as a buy, and use a tick-test for all
  other trades.
• While the classification success is not
  significantly larger, it turns out to have a very
  large effect on measures of effective spreads.
• Moreover, a zero second delay is now recommended.

7
Comparative Studies of Trading Costs

• Bessembinder (JFM, 2003) uses 1998 data to show
  that correcting for the misclassification
  (relying on EMO (2001)) does not affect the
  conclusion that Nasdaq has higher trading costs
  than the NYSE.

Bessembinder (JFQA, 2003) Shows that spreads
and depth after decimalization in 2001 are lower
on both Nasdaq and the NYSE. The NYSE is still
cheaperthan Nasdaq
8
Trading Costs and Trade Size

• Theoretical models (e.g., Easley OHara (1987))
  suggest that large trades should pay a higher
  transactions cost
• Inventory
• Information
• Lots of empirical papers have verified that this
  is the case based on NYSE data.
• Reiss and Werner (1996) instead show that large
  trades in London actually get significant price
  improvement!
• Bernhardt, Dvoracek, Hughson and Werner (2003)
  show that London price improvements are based on
  reputation/relationships (measured as trading
  between the dealer and broker in the past).

9
Trading Costs and Trade Size
10
Transactions Costs References

• Bernhardt, D., E. Hughson, V. Dvoracek, and I.
  Werner, 2003, Why do larger orders receive
  discounts on the London Stock Exchange?, Review
  of Financial Studies, .
• Bessembinder, H., 2003, Issues in assessing trade
  execution costs, Journal of Financial Markets 6,
  .
• Huang, R., and H. Stoll, 1996, Dealer versus
  auction markets A paired comparison of execution
  costs on Nasdaq and the NYSE, Journal of
  Financial Economics 41, 313-357.
• Reiss, P., and I. Werner, 1996, Transaction costs
  in a dealer markets Evidence from the London
  Stock Exchange, in A. Lo Ed. The Industrial
  Organization and Regulation of the Securities
  Industry, University of Chicago Press, 125-175.
• Werner, I., 2003, NYSE Spreads, order flow, and
  information, Journal of Financial Markets 6,
  309-335.

11
Glosten and Harris (1988)Decomposing Spreads

• What part of the spread is attributable to order
  processing and asymmetric information
  respectively?
• Develop a model where both information shocks and
  order flow affect the fundamental value, and
  transactions costs are assumed to be increasing
  in volume.
• Use TS to estimate c0 and z1 for each stock.
• Study CS of c0 and z1 to figure out if z1 is a
  significant proportion of the spread..

Z1adverse selection z1(c0/P (), insider conc.
(0), SHH (-))
c0order processing c0(1/Trd.Frq (), s ())
12
Ho and Stoll (1997)Decomposing Spreads

• What part of the spread is attributable to order
  processing, asymmetric information, and inventory
  respectively?
• Exclude continuations
• Cannot separate out asymmetric from inventory
  costs ?(aß).
• Rest is order processing (1-?)
• Results
• S0.1222
• ?11.4
• (1-?) 88.6
• Fixed costs are much more important than
  information/inventories!

Adverse selection
Inventory
13
Ho and Stoll (1997)Decomposing Spreads

• What part of the spread is attributable to order
  processing, asymmetric information, and inventory
  respectively?
• Try to separate a from ß
• Model serial correlation in trade flows
• Results
• a -3 !!!!
• ß 19
• (1-?) 84
• Fixed costs are still much more important!
• Adverse selection costs are negative!

• Woops those results violate the model
• Perhaps the problem is that orders are broken up.
• Bunch trades at the same price
• Results
• a 9 !!!!
• ß 29
• (1-?) 62
• Cheating???
• Other researchers (e.g., Jones and Lipson (1995))
  have tried to fix it, but they keep having
  trouble getting the adverse selection component
  to behave

14
Spread Decomposition References

• Hasbrouck, J., 2007, Chapter 3.4 and Chapter 9.8.
• George, T., G. Kaul, and Nimalendran, 1991,
  Estimation of the bid-ask spread and its
  components A new approach, Review of Financial
  Studies 4, 623-656.
• Glosten, L., and L. Harris, 1988, Estimating the
  components of the bid-ask spread, Journal of
  Financial Economics 21, 123-142.
• Huang, R., and H. Stoll, 1997, The components of
  the bid-ask spread A General approach, Review of
  Financial Studies 10, 995-1034.
• Petersen, M., and D. Fialkowski, 1994, Posted
  versus effective spreads, Journal of Financial
  Economics 35, 269-292.
• Roll, R., 1984, A simple implicit measure of the
  effective bid-ask spead in an efficient market,
  Journal of Finance 39, 1127-1139.
• Stoll, H., 1989, Inferring the components of the
  bid-ask spread theory and empirical tests,
  Journal of Finance 44, 115-134.

15
Intraday Patterns

• Several authors have examined the intraday
  patterns of returns, volatility, volume, and
  spreads.
• NYSE-listed stocks
• Jain and Joh (1988) returns and volume
• Wood, McInish, and Ord (1984) volume and
  volatility
• Wood, McInish (1992) spreads
• Nasdaq-listed stocks
• Chan, Christie, and Schultz (1995) volume,
  volatility, and spreads
• London stock exchange
• Werner and Kleidon (1996) volume, volatility, and
  spreads
• Returns are declining
• Volatility is U-shaped over the day
• Volume is U-shaped over the day
• Spreads are U-shaped over the day
• Why?

16
Werner and Kleidon (1996)London Intraday Patterns
17
Intraday (Day-of-the Week) Patterns

• Theoretical models with asymmetric information
  make predictions on how actions of strategic
  traders translate into patterns in the data.
• For example
• Kyle (1985) predicts that prices should be linear
  in signed order flow and that informed traders
  should smooth out their trades over time
• Admati and Pfleiderer (1988) predicts that
  trading will be clustered to periods with low
  trading costs and high volatility
• Foster and Viswanathan (1993)
• Estimate based on 60 NYSE firms, deciles 1, 5,
  and 10 (20 each)
• Day of the week effect as well as intraday
  patterns
• Interday pattern (decile 10)
• Mondays are characterized by low volume and high
  adverse selection.
• Tuesdays have low volatility and adverse
  selection
• Intraday pattern
• Volume volatility is U-shaped for all deciles
• Volume and adverse selection are weakly
  positively correlated

18
Foster and Viswanathan (JF, 1993)
Volume
Volatility
19
Foster and Viswanathan (1993)
Spreads
C fixed component
Lambda adverse selection component
20
Madhavan, Richardson, and Roomans (1997)Intraday
Patterns

• Theoretical structural Bayesian model of
  specialist pricing that addresses both
  decomposition of spread and intraday patterns.
• Builds on Madhavan and Smidt (1991)
• Accounts for asymmetric information and inventory
  risk.
• Tested based on NYSE-listed stocks
• Results
• Asymmetric information declines over the day
• The inventory component of the spread increases
  over the day
• Price impact (trading costs) declines over the
  day
• The autocorrelation of order flow is U-shaped
  over the day

21
Madhavan, Richardson, and Roomans (1997)Intraday
Patterns
Asymmetric information
Transaction cost
22
Intraday Pattern References

• Chan, K.C., W. Christie, and P. Schultz, 1995,
  Market structure and the intraday patterns of
  bid-ask spreads for Nasdaq securities, Journal of
  Business 68, 35-60.
• Foster, D., and S. Viswanathan, 1993, Variations
  in trading volume, variance, and trading costs
  Evidence on recent price formation models,
  Journal of Finance 48, 187-211.
• Jain, P. and G. Joh, 1988, The dependence between
  hourly prices and trading volume, Journal of
  Financial and Quantitative Analysis 23, 269-284.
• Madhavan, A., Richardson, M., and M. Roomans,
  1997, Why do security prices change? A
  transaction-level analysis of NYSE-listed stocks,
  Review of Financial Studies 6, 345-374.
• Madhavan, A., and S. Smidt, 1991, A bayesian
  model of intraday specialist pricing, Journal of
  Financial Economics 31, 99-134.
• McInish, T., and R. Wood, 1992, An analysis of
  intraday patterns in bid/ask spreads for NYSE
  stocks, Journal of Finance 47, 753-764.
• Werner, I., and A. Kleidon, 1996, UK and US
  trading of British cross-listed stocks An
  intraday analysis of market integration, Review
  of Financial Studies 9, 619-664.

23
Price Process

• The structural models that we have discussed so
  far have only limited dynamics.
• One way to accommodate richer dynamics is to cast
  the price and order flow problem in a Vector Auto
  Regression (VAR) framework.
• The easiest example is to consider the Roll model
  as a two-equation system
• where A, B, and C are 2x2 matrices
• Well known techniques exist for specifying and
  estimating VARS.
• Dynamics are studied by impulse response
  functions
• Study how a shock to an exogenous variable, i.e.,
  ut or vt, affects current and future price
  changes and order flow

24
Hasbrouck (1991)Price Process

• Hasbrouck has pioneered modeling the joint
  dynamics of trades and quotes as a VAR.
• Idea (from Hasbrouck (1988)) is to separate the
  effect of inventory (temporary) from information
  (permanent) on price changes
• Key is that only unanticipated changes in order
  flow should have permanent price impact
• Hasbrouck (JF, 1991) measures the information
  effects as the permanent price impact of a trade
  (impulse response).
• Inventory, order processing, and other frictions
  should have transient impact on prices
• Price impact takes time
• Price impact is a positive and concave function
  of trade size
• Large trades cause a widening of the spread
• When spreads are wide, the price impact is larger
• There are more information asymmetries for small
  firms

25
Hasbrouck (1991)Price Process
26
Hasbrouck (1991)Price Process
27
Hasbrouck (1991)Price Process

• Hasbrouck (RFS, 1991) proposes a method to
  decompose the variance of changes in the
  efficient price to those that are trade related
  (information based), and those that are unrelated
  to trades.
• Finds that trades (information) account for about
  34 of total variation
• The fraction of variances attributable to
  information is higher for less liquid stocks than
  for more liquid stocks.
• Finds support for Foster and Viswanathan (1993),
  and contradicts Admati Pfleiderer (1988) when it
  comes to intraday patterns
• Price impact is largest at the open

28
Hasbrouck (1993)Measure of Market Quality

• Hasbrouck (1993) identifies a new measure of
  market quality
• The standard deviation of the difference between
  the transaction price and the implicit
  unobservable efficient price
• The efficient price is assumed to follow a random
  walk component
• Problem is that the efficient price is
  unobservable!!!
• The stationary component is labeled pricing
  errors
• Propose a clever VAR is used to estimate the
  variance of the efficient price
• Pricing error standard deviation is about 0.33
  of stock price, corresponding to a cost of 0.26
  of stock price.
• Dispersion of pricing errors are elevated around
  the open and the close

29
Hasbrouck (1995)Off-Exchange Competition

• Hasbrouck (1995) studies what part of the
  information comes from the NYSE versus the
  Regional exchanges
• A VAR is used for cross-market analysis of quotes
  for Dow 30 stocks
• Idea is that the efficient price is a common
  process across markets trading the same stock
• What fraction does the NYSE versus the regionals
  contribute to changes in the efficient price?
• NYSE is the main source of price discovery (92.7
  information share)
• Regional exchanges are cream-skimming

30
Hasbrouck (1995)Off-Exchange Competition
31
Hasbrouck (1995)Off-Exchange Competition
32
Engle and Russell (2003)The ACD Model

• Engle and co-authors have developed a time-series
  model called the ARCH model (AutoRegressive
  Conditional Heteroscedasticity) to model the
  variance of returns.
• Most market microstructure work samples trades,
  regardless of how close together or far apart
  those trades occur in time.
• The Autoregressive Conditional Duration model
  addresses the fact that data is often irregularly
  spaced.
• Technically, the time between trades is modeled
  as a point process with dependent arrival rates.
• The ACD model is essentially a ARCH model for the
  time between trade arrivals.
• Additional papers by Engle and co-authors extend
  these ideas to analyze the joint dynamics of
  quotes and trades.

33
Price Process References

• Hasbrouck, J., 2007, Chapter 9
• Chakravarty, S., 2001, Stealth trading Which
  traders trades move prices? Journal of Financial
  Economics 61, 289307.
• Engle, R., and J. Russell, 2003, Autoregressive
  conditional duration A new model for irregularly
  spaced data, Econometrica 66, 1127-1162.
• Hasbrouck, J., 1991a, Measuring the information
  content of stock trades, Journal of Finance 46,
  179-207.
• Hasbrouck, J., 1991b, The summary informativeness
  of stock trades An econometric analysis, Review
  of Financial Studies 4, 571-595.
• Hasbrouck, J., 1993, Assessing the quality of a
  security market A new approach to
  transaction-cost measurement, Review of Financial
  Studies 6, 191-212.