Equity Trading by Institutional Investors. To Cross or Not To Cross? Randi Næs and Bernt Arne Ødegaard
Research problem Institutional investor, have money, will buy US Equity Submission strategy for institutional investor Decision investigated: Venue for trading Traditional exchange ECN s Crossing Networks What do we look at? Cost of trading, focus on crossing networks. HOW?
Trading venues Innovation in current US equity markets. Traditional exchanges, NYSE, NASDAQ, Regionals. ECN s Electronic Limit Order Markets. Independent Price discovery Crossing Networks. Submit quantities desired. Prices derived off main market (NYSE/NASDAQ)
Multiple trading places, theory Usual intuition: Winner takes most (Chowdhry and Nanda (1991)) Coexistence exchange and crossing network (Hendershott and Mendelson (2000)) Crossing network lower cost of running No price discovery in crossing network.
Possible submission strategies Submit to crossing network only. Submit to crossing network first, then to exchange if not crossed. Submit to exchange only.
Observation In a crossing network there is no independent role for the price mechanism. Information based trading will only affect the probability of execution in a crossing network.
Prior Research, empirical Overview of cost of trading for institutional investors: Keim and Madhavan (1998). Cost of institutional trading in ECN s: Conrad, Johnson and Wahal [2001]
Our investigation Use data on US equity trades of one large insitutional investor, the Norwegian Government Petroleum fund. Fund: Invest in international bond and equity markets. Index tracking, FT/S&P Actuaries World Index. Built up an equity position of $7 billion from scratch in 6 months, start 1998. Submission strategy of the fund: Cross as much as possible. Buy the rest in the primary market.
Measuring the costs of transacting equity. Method of Keim and Madhavan (1998). Implementation shortfall approach: commission per share Measure of explicit cost = P d Measure of implicit cost = Pa P d 1 P a = the average price of all executed trades in the order P d = the closing price for the stock on the day before the decision to trade
Cost estimates Average trading costs for the Fund s transactions Costs n Total Implicit Explicit All trades mean 0.12 0.09 0.03 3909 stdev (1.95) (1.95) (0.14) vw avg 0.30 0.29 0.01 Crosses only mean 0.09 0.06 0.03 3252 stdev (2.01) (2.01) (0.15) vw avg 0.27 0.27 0.01 Market only mean 0.30 0.25 0.05 657 stdev (1.60) (1.60) (0.04) vw avg 0.46 0.43 0.03
Determinants of costs Regression analysis of total trading costs for all orders coeff (pvalue) Constant -0.0076 (0.04) ln(mcap i ) -0.0008 (0.04) ln(ord i ) 0.0014 (0.00) ln(mktvlm i,t 1 ) 0.0004 (0.20) R i,t 3,t 1 0.0001 (0.58) HL i,t 1 0.0001 (0.60) Di EC -0.0068 (0.00) Di M 0.0036 (0.00) Di P -0.0099 (0.00) n 3522 R 2 0.04
The additional costs of crossing networks In terms of measured costs crossing networks seem superior. However, given the submission strategy of the fund, what does this really mean? Conditionally on not being crossed, the stocks transacted in the market had a higher cost.
Adverse selection Use knowledge of the fund s submission strategy: First try to cross, then trade in the market. In the absense of informed traders, the availability of a stock in the crossing network is unrelated to the quality of the stock. In the presence of informed traders, the higher the quality of the stock the less likely it will be available in a crossing network: For a high quality stock, No informed traders will sell the stock Some informed will try to buy the stock Both these cases will lower the probability of the stock being available in the crossing network.
How to investigate this? Two implementations Event study Choice theoretic regression
Event study Under the null of no informed traders: No relation between the quality of a stock and the probability of execution in a crossing network. Event study: Estimate of quality differences across two groups: Stocks that were crossed. Stocks that were tried crossed the same date, but could not, and hence had to be bought in the market. Use ex post excess return as an estimate of quality (short lived information being revealed)
Event Study 0.02 dates in June 0.015 0.01 0.005 CAR 0-0.005-0.01-0.015-0.02 0 5 10 15 20 event date market orders crosses
Formally test whether there is a difference: Use a dummy variable Di cross as explanatory variable. Is this significant? Determinants of CAR coeff (pvalue) Constant -0.081 (0.00) ln(mcap i ) 0.006 (0.06) ln(ord i ) 0.007 (0.04) R i,t 6,t 1-0.000 (0.67) HL i,t 1 0.013 (0.00) σ(r i,t 60,t 1 ) -0.182 (0.62) RelVlm i,t 2,t 1-0.010 (0.10) Di Cross -0.027 (0.01) Date 2-0.032 (0.00) Date 3-0.031 (0.00) Date 4-0.050 (0.00) n 1100 R 2 0.05
Choice theoretic regression. Verify conclusions of event study by controlling for other properties of a trade. Order size Market cap Still use ex post excess return as an estimate of quality
Directly investigating determinants of crossing. Probit model estimating determinants of probability of a cross Variable coeff (pvalue) constant -2.147 (0.00) CAR i,t,t+20-1.372 (0.01) ln(mcap i ) 0.333 (0.00) ln(ord i ) -0.129 (0.01) R i,t 6,t 1-0.047 (0.00) HL i,t 1-0.004 (0.95) RelVlm i,t 2,t 1 0.216 (0.06) RelVlm i,t -0.333 (0.00) n 950 Pseudo R 2 0.059
B Chowdhry and V Nanda. Multi market trading and market liquidity. Review of Financial Studies, 4:483 511, 1991. Jennifer Conrad, Kevin M Johnson, and Sunil Wahal. Alternative trading systems. Working paper, University of North Carolina at Chapel Hill, forthcoming in Journal of Financial Economics, November 2001. Terrence Hendershott and Haim Mendelson. Crossing networks and dealer markets: Competition and performance. Journal of Finance, 55(5), October 2000. Donald B Keim and Ananth Madhavan. The cost of institutional equity trades. Financial Analysts Journal, pages 50 69, July/August 1998.