Journal of Banking & Finance 25 (2001) 1589±1603 www.elsevier.

com/locate/econbase

The use of undisclosed limit orders on the Australian Stock Exchange
Michael J. Aitken a, Henk Berkman
a b

b,*

, Derek Mak

a

University of Sydney, Sydney, NSW 2006, Australia Department of Accounting and Finance, University of Auckland, Private bag 92019, Auckland, New Zealand Received 15 January 1999; accepted 22 June 2000

Abstract This paper investigates the use of undisclosed limit orders on the Australian Stock Exchange (ASX). Our ®ndings suggest that undisclosed limit orders are used to reduce the option value of limit orders. We ®nd no evidence that undisclosed limit orders are more frequently used by informed traders than disclosed limit orders. The e€ects of recent changes in undisclosed order regulation are also examined. We ®nd that the enhancement in pre-trade transparency, through tightening the undisclosed order regulation in October 1994, resulted in a signi®cant decline in trading volume. The impact of the second regulation change in October 1996, which further tightened undisclosed order regulation, resulted in a less signi®cant trading volume reduction. Ó 2001 Elsevier Science B.V. All rights reserved. JEL classi®cation: G14 Keywords: Market transparency; Undisclosed limit orders; Quote-matching; Order option values; Market liquidity

*

Corresponding author. Tel.: +64-9-3737599 ext. 7181; fax: +64-9-3737406. E-mail address: h.berkman@auckland.ac.nz (H. Berkman).

0378-4266/01/$ - see front matter Ó 2001 Elsevier Science B.V. All rights reserved. PII: S 0 3 7 8 - 4 2 6 6 ( 0 0 ) 0 0 1 4 9 - 7

the greater the incentive to hide his identity and trading intentions. 1 Informed traders might therefore prefer undisclosed over disclosed limit orders (DLOs) as the former provides a lower degree of pre-trade transparency. The free option problem exists because monitoring costs and delays in cancellation and execution prevent investors from continuously updating their limit orders for changes in market conditions. 1 . the Paris Bourse and the Toronto Stock Exchange. Some markets. 1996). Aitken et al. Keim and Madhavan (1995). Chakravarty and Holden (1995). An alternative explanation for the use of ULOs comes from Harris (1996). the quote-matcher may be able to limit his loss by selling to the large order. Undisclosed orders do not remove exposure of orders to new For related theoretical papers see. This study investigates the factors that a€ect the use of these Ôundisclosed limit ordersÕ on the ASX. Introduction Exchanges that adopt electronic order book systems rely on limit orders as the major source of liquidity. give traders the option to hide the quantity of their limit orders and display price only. accounting for approximately 28% of volume.1590 M. For example. a quote-matcher might place a buy limit order at a slightly higher price. 2 Limit order submitters face the risk that other traders take advantage of the trading options implicit in exposed limit orders. Admati and P¯eiderer (1991) and Forster and George (1992). ULOs increase the trading risk borne by quote-matchers by creating uncertainty over the size of the Ôsafety netÕ. He argues that undisclosed limit orders are used as a defensive strategy against quote-matchers. Their reasoning is based on the idea that the more private information a trader possesses. / Journal of Banking & Finance 25 (2001) 1589±1603 1. The e€ect of recent changes in undisclosed order regulation is also examined. the quote-matcher will pro®t to the full extent of the rise. Aitken et al. A defensive strategy against quotematchers is to hide the quantity of limit orders. Limit orders are therefore exposed to the arrival of adverse new public information and might get unfavourable execution (see Berkman.J. for example. Handa and Schwartz (1996). was undisclosed. (1996) report that in 1993 about 6% of orders on the ASX. for example. Why do traders use undisclosed limit orders (ULOs)? According to some market participants ULOs are sometimes used by informed traders. Harris and Hasbrouck (1996) and Harris (1997). Another problem that could be less severe for ULOs is what has been referred to as the Ôsitting duckÕ or free option problem of limit orders. the Australian Stock Exchange (ASX). If his limit order is hit and the stock value subsequently rises. if a large order to buy a stock is entered at a certain price. If the stock price falls. 2 Other papers that consider tradersÕ order submission strategies are Kumar and Seppi (1994).

suggesting that undisclosed orders are not used more frequently by informed traders. which further tightened undisclosed order regulation. resulted in a less signi®cant reduction in trading volume. First. we test the information content of the submission of ULOs and compare this with DLOs and market orders. However. the ASX tried to enhance pre-trade transparency by forcing traders to display more of their orders. since picking o€ is more complicated than simply hitting the stale limit order for a known total number of disclosed shares. thereby enhancing liquidity. We ®nd that the use of ULOs increases in volatility and average order value and decreases in tick size and trading activity. for a cross-section of stocks. we analyse the impact of ULOs on market liquidity. the quantity ®eld on the trading screen displays a ÔUÕ for undisclosed.J. Our results are consistent with the use of ULOs as a strategy to reduce the option value of limit orders. reduce their willingness to provide liquidity. Note that on the ASX the quantity of ULOs is hidden. Similarly. We hypothesise that ULOs serve a useful purpose as a potential strategy to reduce the option value of limit orders. This regulation change reduced the use of ULOs and resulted in a signi®cant decline in trading volume. The impact of the second regulation change in October 1996. 4 These results are consistent with Berkman and Hayes (2000). we expect that the free option problem is less severe for ULOs than for DLOs. the execution skills of ¯oor brokers are utilised to reduce the option value of limit orders. Section 4 presents the cross-sectional analysis of the use of ULOs and Section 5 analyses the change in market liquidity around the regulation changes. In October 1994. we examine the e€ect on market liquidity of changes in undisclosed order regulation on the ASX. if a trader submits an ULO. They show that on ¯oor trading systems. Our ®ndings indicate that there is a strong permanent stock price reaction to the submission of market orders. Section 6 provides the summary and conclusion. Section 2 outlines the institutional framework of the ASX and the data set. / Journal of Banking & Finance 25 (2001) 1589±1603 1591 public information. We ®nd no di€erence in stock price reaction between disclosed and undisclosed limit orders. 4 Finally. This study provides an empirical analysis of the use of ULOs. the free option problem is expected to reduce the willingness of traders to provide liquidity. Section 3 presents evidence on the stock price reactions around order submission. Aitken et al. 3 Next. 3 . However. we relate the use of ULOs to factors that are expected to a€ect the quote-matching and free option problem. The remainder of this paper is organised as follows.M. Harris (1990) argues that quote-matchers are free-riding on orders submitted by primary liquidity providers and quote-matching might therefore. Other market participants can therefore ÔobserveÕ the ULO immediately after submission. To test this hypothesis.

This study examines only orders submitted between 10:00 am and 4:00 pm when the market was open for continuous trading. The average stock price across the sample stocks is $1.5 million dollars per stock and the average market value of equity across our sample stocks is 313 million dollars.709 and the median $160. .000 on 16 October 1996. The relative tick ranges from 0.000. / Journal of Banking & Finance 25 (2001) 1589±1603 2. The data are drawn from the ASX microstructure databases maintained by the Securities Industry Research Centre of Asia±Paci®c (SIRCA). measured from 1 January 1994 to 24 October 1994. this uncertainty could act as a deterrent for quote-matching and picking o€ of stale limit orders. we collected data for the 30 days before and after the two regulation changes on 24 October 1994 and 16 October 1996.000. The average value of the ULOs in this period is $356. 5 No such rules exist in SEATS.1% to 67% and the daily high low averages 2% of the stock price.4 million.3 million. For the event study. The quantity ®eld on SEATS for such orders displays a ÔUÕ for undisclosed. which contain detailed information for all orders submitted to SEATS. broker number and the quantity associated with each limit order in the book. Other descriptive statistics for our initial sample are given in Table 1. The initial sample consists of all ULOs placed on the ASX between 1 January 1994 and 24 October 1994. (1995) and Harris (1996) for more detailed discussions on the institutional framework of CAC (Paris) and CATS (Toronto). Description of the market and the data set The ASX uses an electronic screen-trading system called the Stock Exchange Automated Trading System (SEATS). This was increased to $25. Our initial sample has 114. the disclosure threshold was $10.000 on 24 October 1994 and further increased to $100. Initially. Average daily trading volume is 12.J. Aitken et al. The numbers above show that there is considerable variability in the value of ULOs. On the Toronto Stock Exchange and the Paris Bourse.657 observations. As argued before.1592 M. The standard deviation of ULO value is $1. The minimum value is $10. This U is visible for all market participants as soon as the order is entered in SEATS.7 and the bid±ask spread is 4% on an average. SEATS provides an order driven market which automatically executes matched bids and asks entered by brokers based on ®rst price and then time priority. SEATS screens display price. undisclosed orders have less precedence than disclosed orders at a given price.004 and the maximum value is $24. The table uses average values of the variables per stock. Brokers have the option to hide the quantity of their limit orders when the total value of the transaction is over the level of a disclosure threshold. 5 See Biais et al.

04 0. Admati and P¯eiderer (1991).D.220 0.03 0.J. However.5 312. 1593 3.M. traders have a greater incentive to hide their identity the more private information they possess. informed traders might avoid limit orders altogether and submit market orders to exploit their information (Glosten and Milgrom. (iii) within 30 days of each other. To reduce the in¯uence of ex-post expectational errors regarding order size.036 0. within two ticks of the spread of the ULO. 1985). we only include limit orders placed at the current best quote. informed traders prefer undisclosed over disclosed limit orders as the former provides a lower degree of pre-trade transparency. ($ m) 1.105 0. Direct matching of the order size for every ULO in the sample might be inappropriate as the order size of an ULO is not revealed.025 0. Aitken et al. Hasbrouck. for each stock only the middle one-third of all ULOs in terms of order size are used. Bid±ask spread. To test these hypotheses. 3. The foremost control is for order size.01 0. / Journal of Banking & Finance 25 (2001) 1589±1603 Table 1 Descriptive statisticsa Mean Price ($) Bid±ask spread Relative tick High±low range Volume ($ m) Market cap. for example.71 0.402 1090 33. 1987.3 1496 a This table reports descriptive statistics for 1041 stocks with ULOs in the period from 1 January 1994 to 24 October 1994.5 0.21 0.374 Maximum 47. The table uses average values of the variables per stock. as di€erent order sizes have di€erent information content (Easley and OÕHara.8 Minimum 0. (ii) in the same direction (buy/sell).001 0.001 0. we use a matched sample design to ensure that any di€erence in stock price movements around order submission can be attributed to order type.667 0. (iv) not within 30 min before or after the submission of each other. We assume the marketÕs initial guess for the quantity of a ULO for a particular stock equals the historical average size of all ULOs for that stock. That is. Stock price behaviour around order submission To test for the information content of the di€erent order types.02 12. For each ULO in this group. measured over the sample period. DLOs and ULOs. this section analyses permanent price e€ects around the submission of a matched sample of market orders. To select limit orders that are likely to have the highest information content. Volume is the on market trading volume and market capitalisation is the market value of equity. .001 2. a matching DLO and MO of the same stock are then selected such that they are (i) of order sizes within 10% of the ULO. only medium-sized ULOs are used. and (v) with a current spread in terms of ticks. relative tick and the average daily high±low range are all measured as a fraction of the stock price. 1991). or better.021 59. As shown by. According to this logic. We use a number of controls to construct matched MO±ULO±DLO subsamples.634 S.

This result does not support the hypothesis that ULOs are more attractive to informed traders than DLOs.63% after 1 minute (signi®cant at the 0. The price reactions following both disclosed and undisclosed limit orders are much smaller. That is. / Journal of Banking & Finance 25 (2001) 1589±1603 The sample of buy orders has 966 matched observations and the sample of sell orders has 1024 matched observations. All marketable limit orders are hence deleted from the sample. 5 and 30 minutes after order submission is analysed. and 1. For market orders.01% level). After market buy order submission there is a price increase of. The price reaction following market orders remains signi®cant. the best opposite-side quote prevailing at the time of submission is taken as the reference price. 6 3. To measure the information content of the di€erent order types. Cross-sectional evidence on the relative use of undisclosed limit orders In this section. For market sell orders there is an average price decrease of 0. which hit the opposite quote and then become the best same-side quote. both stock reaction measures are problematic. for buy limit orders we measure the ask-to-ask return following submission of the order and for sell limit orders we use bid-to-bid returns. Aitken et al. We use opposite-side quotes for limit orders to avoid the problem that after submission of a new limit order a quote improvement could simply re¯ect an increased supply of liquidity and does not necessarily re¯ect new information. and for sell market orders ask-to-ask returns) this problem is avoided.J. because the execution of a market order results in a change in the opposite-side of the order book. we relate the use of ULOs to variables proxying for the quote-matching and free option problem for a cross-section of stocks. For limit orders. the stock price reaction at the end of the day. Results Table 2 presents the results regarding the information content for the matched MO±ULO±DLO subsamples. a signi®cant stock price reaction following limit order submission using the end-of-day quotes.31% 1 minute after submission. on an average. Most important for our purposes is that the di€erence between the mean price reaction for disclosed and undisclosed limit orders is never signi®cantly di€erent from zero (t-statistic in bottom row of Table 2). 4. we use same-side quotes. Using same-side quotes (for buy market orders bid-to-bid returns. As For marketable limit orders. 6 .1. 0. 2. implying a permanent price impact. There is. This change could simply re¯ect the consumption of liquidity and does not necessarily re¯ect new information.1594 M. however.

61) 0.68 a The ULO subsample consists of the middle one-third of all ULOs in terms of order volume for each stock in the period 1 January 1994 to 24 October 1994.141 (2. / Journal of Banking & Finance 25 (2001) 1589±1603 Table 2 Stock price reaction after order submission for the matched ULO±DLO±MO subsamplea 1 min Panel A: Buy-orders Market orders DLOs ULOs t-statistic (DLO ˆ ULO) Panel B: Sell-orders Market orders DLOs ULOs t-statistic (DLO ˆ ULO) 0.027 (2.83) 0.315 (10.77) )0.034 (0.42) 0.60) 0.377 (10. The ®nal sample consists of 966 buy-orders and 1024 sell-orders.006 (0.17) 0.040 (1.13) )0.M.627 (14.29) 0. Only limit orders placed at or better than the best bid or ask are included in this sample.007 (0.44) )0.03) )0.66 30 min 0. In markets that enforce time precedence. We therefore.61) 0.42) 0. All returns are expressed as percentage (%).021 (0.312 (10.005 (0.89) 0.11) 0.18) 0.342 (10.090 (2.55) 0.65 5 min 0.11) 0. Quote-matchers must pay the cost of a minimum tick when they jump ahead of a limit order to gain price priority. Values in parentheses are t-values for the null hypothesis that the mean return is zero.06) 0.36) )0. we expect the use of ULOs to be higher for these stocks.000 dollars (all ÔundisclosableÕ limit orders).99) 0.687 (14.32) 0.632 (14. RelTicki is de®ned as the average .764 (14.001 (0.88) 0. Aitken et al.51 2 min 0. The t-statistic in the bottom row tests the null hypothesis that the di€erence between the mean returns for disclosed and undisclosed limit orders is zero. Each ULO is matched by a DLO and a market order according to the ®ve criteria listed in Section 3. Since quote-matching is expected to be more prevalent for stocks with a low relative tick. because they are less exposed.39 )0.001 (0. expect the use of ULOs to be relatively high for stocks that are characterised by a relatively high option value of limit orders.12) 0. If the tick is small.53) 0.29) 0. 1996). ULOs are likely to mitigate quote-matching and have lower option value than DLOs. For each stock i.878 (12.45) 0. All stocks without ULOs in the sample period are excluded from the analysis.67) 0.J.166 (3.012 (0.016 (0.53 )0.001 (0. mentioned before.87) 1.052 (1. quote-matchers can obtain price priority by just slightly improving price.85 1595 Day-end 0.36) 0.294 (4.352 (6.118 (2.009 (1. the cost of quote-matching is higher.62 )0.354 (5. is measured as the log of the value of all ULOs submitted in the period from 1 January 1994 to 24 October 1994 divided by the log of the value of all limit orders with a value of more than $10. USEi . the larger the minimum tick (Harris.14 )0.64) 0.53 )0. The use of ULOs for each stock i.

which is more likely to be exogenous: USEi ˆ a0 ‡ a1 RelTicki ‡ a2 Riski ‡ a3 LnFsizei ‡ ei : …2† Table 3 presents the results for models (1) and (2). the relative use of ULOs in order to reduce this option value. 7 4. . Aitken et al. Riski . as the expected time during which an order is outstanding decreases in trading activity. LnFsizei .000. Our model uses the log of the weighted average value of undisclosable limit orders for stock i as measure of order value. The weights for undisclosed and disclosed limit orders are determined by the respective total dollar value in the sample period of the ULOs and DLOs with a value of more than $10. replacing trading activity and average order value by the log of the market value of the ®rm. These problems might also a€ect model (1). The option value of limit orders. ceteris paribus. is expected to increase in volatility (Copeland and Galai. where the weights are the number of undisclosed orders used to compute the dependent variables for each stock: USEi ˆ a0 ‡ a1 RelTicki ‡ a2 Riski ‡ a3 LnTraActi ‡ a4 LnOrdVali ‡ ei : …1† As discussed in Harris (1996). microstructure models typically su€er from endogeneity problems. LnOrdVali .1596 M. As argued above. is measured as the average daily high±low range as a fraction of price.J.000. LnTraActi .1. Empirical model and results We estimate the following weighted least squares model. or average DLO order value does not materially change any of the results. In order to address this issue. ceteris paribus. To proxy for trading activity our model includes. Model (1) shows that the use of undisclosed limit orders is negatively related to relative tick size and trading 7 The use of average ULO order value. In our regression. we estimate model (2). 1983). / Journal of Banking & Finance 25 (2001) 1589±1603 minimum tick size of stock i during the sample period divided by the average stock price. DLOs are more exposed than ULOs and are expected to have a higher option value. which is measured as the log of the number of market orders in the sample period with an order value of more than $10. Limit order traders give trading options to other market participants. volatility for stock i. and therefore. We expect the option value of orders to be smaller for more actively traded stocks. Larger orders have higher option value.

J. LnTraActi is measured as the log of the number of market orders with an order value of more than $10.26) 1041 0. It is interesting to note that the ®rm size has a positive coecient.47) 2. RISKi is the average daily high-low spread as fraction of price for stock i. After 30 minutes. but the signs of the coecients are consistent with the results for model (1). This result suggests that despite the relatively higher trading activity for large stocks. Initial analysis on our data indicates that ULOs takes longer time to execute.672 (3.73) 1. we ®nd that 16% of the ULOs and 26% of the DLOs are executed within 5 minutes. 38% of the ULOs and 48% of the DLOs are executed.021 ()6. Aitken et al. USEi is the log of the value of all ULOs submitted in the period from 1 January 1994 to 24 October 1994 for stock i divided by the log of the value of all limit orders with a value of more than 10. We also ®nd that the relative use of ULOs is positively related to volatility and order value. For example. These results are consistent with the hypothesis that undisclosed limit orders are used to mitigate the quote-matching and free option problem.000 dollars for stock i (all ULOs).461 ()32.085 ()11.52 )1. 8 . LnFsizei is the log of the market value of ®rm i.418 (26.54) 1597 0. (1) and (2).312 ()3. LnOrdVali is the log of the weighted average value of undisclosable limit orders for stock i. the larger average order size for these stocks results in an increase in the use of ULOs as ®rm size increases.26) )2.000 in the sample period.191 a See Eqs. The explanatory power of model (2) is substantially lower. activity. for a sample of ULOs and DLOs that are fully executed within 3 hours. The weights for undisclosed and disclosed limit orders are determined by the respective dollar value of the ULOs and DLOs over $10.000 for stock i. RelTicki is the tick size for stock i as a proportion of the average price in the sample period. / Journal of Banking & Finance 25 (2001) 1589±1603 Table 3 Weighted least squares models for factors a€ecting the use of ULOsa Use of ULOs Intercept RelTick Risk LnTraAct LnOrdval LnFsize Sample size Adjusted R2 )5.04) )6.214 (5.93) 1041 0.M. tStatistics are in parentheses.51) )0.98) 0.034 (5. The weights used in the estimation of model (1) and (2) are the number of ULOs used to compute the dependent variable for each stock.704 ()13. 8 In a more complete analysis of tradersÕ order strategies one could also include the time it takes to execute di€erent order types.

Pj the average of closing bid and ask prices for stock j. Everything else being the same. Since the dummy for the observations with ULOs before the regulation change is omitted. we also analyse changes in volume and spreads for stocks with no ULOs in the period before and after the regulation change. Vj the dollar volume of total trading on ASX for stock j. the bid±ask spread and volume. the increased exposure of limit orders might drive away primary liquidity providers. the other dummy variables indicate the di€erence in volume and spread compared to this sample. one observation before the regulation change and one observation after the regulation change.J. Fj the dollar …3a† . Aitken et al. The impact of disclosure regulation changes on market liquidity In October 1994 and 1996. To control for omitted variables that might a€ect our results. …3b† where Ln indicates the natural log. the resultant increase in quote-matching activity causes the market bid±ask spread to narrow as quote-matchers more often jump ahead of the existing limit order queue. Assuming that the regulation change did not a€ect the liquidity of these stocks. and after the regulation change. we analyse the impact of the disclosure regulation changes on both trading volume and bid±ask spreads as measures of liquidity. Ln Vj ˆ b0 ‡ b1 Ln Sj ‡ b2 Ln Rj ‡ b3 Ln Fj ‡ b4 ULOAFTERj ‡ b5 NO ULOBEFOREj ‡ b6 NO ULOAFTERj ‡ vj .1598 M. / Journal of Banking & Finance 25 (2001) 1589±1603 5. All variables are averaged over a 30-day period before. The dummy variables are of most interest as they indicate the change in the two liquidity measures around the regulation changes for both the sample with ULOs (ULOAFTER) and the sample with no ULOs (NO_ULOBEFORE and NO_ULOAFTER). In this section. the ASX tried to enhance pre-trade transparency by increasing the disclosure threshold. These increases made it relatively more dicult for liquidity providers to submit ULOs and thus rendered a potential strategy to reduce the option value of limit orders less e€ective. the average stock price and the market value of equity) and three dummy variables. a comparison with the control sample should improve the robustness of our analysis. The dependent variables are measures of liquidity. Rj the average daily high±low price range as a fraction of price of stock j. The model we estimate is Ln Sj ˆ a0 ‡ a1 Ln Vj ‡ a2 Ln Rj ‡ a3 Ln Pj ‡ a4 ULOAFTERj ‡ a5 NO ULOBEFOREj ‡ a6 NO ULOAFTERj ‡ uj . The simultaneous equations model has two observations for each ®rm. The independent variables are control variables (the average daily high±low price range. However. Since volume and bid±ask spreads are interdependent we use a simultaneous equations model similar to Hedge and Miller (1989). Sj is the time-weighted intraday bid±ask spread as a fraction of price for stock j. which could adversely a€ect the bid±ask spread and trading volume.

We also ®nd no signi®cant changes in average order size and the daily high±low return. For the sample of ®rms with no ULOs. Around event 2. These numbers drop to 2. indicates that compared to the pre-event period there is a signi®cant decrease in trading volume and no signi®cant change in the bid±ask spread (the pre-event dummy for ®rms with ULOs is not included in the model). but ®nd no signi®cant change in the proportion of o€-market trades. the mean percentage of ULOs drops from 1.7% and the dollar value of these orders drops from 13.8% of total trading volume. Aitken et al. also analyse o€-market volume. and zero otherwise. Also note that the on-market volume drops and the bid±ask spread increases signi®cantly around event 1.and post-period. there is a signi®cant reduction in bid±ask spread and a signi®cant increase in trading volume. For the sample of stocks with no ULOs.6% of orders accounting for 19% of trading volume is undisclosed in the 30 days prior to event 1.1.M. Table 4 shows that an average of 3. 5. Results The top half of Table 4 gives descriptive statistics for the sample ®rms with ULOs in the 30 days surrounding the regulation changes. Note that the dummy variables for the sample of stocks with no ULOs indicate that for these stocks spread and volume increased around event 1. NO_ULOBEFORE ˆ 1 if the observation is pre-regulation change and the stock had no ULOs in the pre.1% to 8. ULOAFTER. / Journal of Banking & Finance 25 (2001) 1589±1603 1599 value of shares outstanding for stock j. The results for ®rms with no ULOs submitted in the 30 days before and after each regulation change are reported in the bottom half of Table 4. and zero otherwise and NO_ULOAFTER ˆ 1 if the observation is post-regulation change and the stock had no ULOs in the pre.06) whereas the increase in on-market trading volume is insigni®cant. We therefore. respectively after the regulation change. .J. Panel A documents the impact of event 1. ULOAFTER ˆ 1 if the observation is post-regulation change and the stock had ULOs in the pre. Panel A presents data for the event around 24 October 1994 (event 1 hereafter) and Panel B presents the data for the October 1996 event (event 2 hereafter). the post-regulation dummy variable for the group of ®rms with ULOs. 9 Table 5 presents the results for the simultaneous equations models. the changes in bid±ask spread and volume around event 1 are insigni®cant.6% to 0. The bid±ask spread decreases (P-value is 0. 9 The increases in the level of the disclosure threshold have presumably improved the level of pretrade market transparency on the ASX as more order quantity is revealed in the limit order book.and post-period. if investors shift o€-market to avoid the new undisclosed order requirement then the change in transparency might be negligible.6% and 17%.or post-period. All control variables are signi®cant and have the expected signs. However. and zero otherwise.

337 0 0 0.J. Only ®rms with no ULO submitted in the 30 days before and after the regulation changes are included in the no-ULO sample (129 ®rms around event 1 and 270 ®rms around event 2).766 P-values Panel B Event 2: $25.0466 12.223 30-days after Mean 0.006 0 0 0.601 0 0 0.001 0.0260 0.000±$100.000 ULO sample Fraction of ULOs (#) Fraction of ULOs ($) Fraction bid±ask spread Log volume ($) No-ULO sample Fraction of ULOs (#) Fraction of ULOs ($) Fraction bid±ask spread Log volume ($) 0.000 ULO sample Fraction of ULOs (#) Fraction of ULOs ($) Fraction bid±ask spread Log volume ($) No-ULO sample Fraction of ULOs (#) Fraction of ULOs ($) Fraction bid±ask spread Log volume ($) 0.001 0.235 0.067 0 0 0.001 0.0882 0.1309 0.0336 14. a The purpose of including ®rms with no ULOs in our sample is to control for the impact of omitted variables on our liquidity measures.007 0.0249 15.0422 13.1894 0. P-values are generated from the Wilcoxon rank test.260 0.0521 12.0362 0.231 0. / Journal of Banking & Finance 25 (2001) 1589±1603 Table 4 Descriptive statistics before and after the two regulation changesa 30-days before Mean Panel A: Event 1: $10.1600 M.064 0. which tests the null hypothesis that the di€erence in medians between the pre.1724 0.0157 0.803 0.0382 14.025 0. .231 0 0 0.000±$25.165 0 0 0. Aitken et al.0554 12.0278 15.001 The table presents mean values in the 30 days pre-regulation and the 30 days post-regulation. Only ®rms with at least one ULO submitted in the 30 days before or after the regulation changes are included in the ULO sample (521 ®rms around event 1 and 474 ®rms around event 2).001 0. F-tests can then be used to test whether the change in spread and volume around the events di€ers for the sample of stocks with ULOs from the sample of stocks without ULOs.0072 0.and post-event period is zero.

03) 1488 0.95) )0.78) )0.372 (6.93) 1488 0.J.037 ()1.82) )0.72) )0.229 ()4.058 ()7.48) )0.000 Constant Spread Volume Volatility Price Firm size ULOAFTER NO_ULOBEFORE NO_ULOAFTER Sample size Adjusted R2 6.133 ()2.75 a See Eqs.72) ± ± )0. Fj is the dollar value of shares outstanding.245 (9.289 (13.676 ()7.002 (0.96)b )1. Sj is the time-weighted percentage intraday bid±ask spread over the sample period for stock j.07) )0.10) 0.267 ()3.64) )0.311 ()2. ULOAFTER ˆ 1 if the observation is post-regulation change and the stock had ULOs in the pre.136 ()3.81 21.78) ± ± )0. Pj the average of closing best bid and ask prices. (3a) and (3b).545 ()6.and post period. and zero otherwise and NO_ULOAFTER ˆ 1 if the observation is post-regulation change and the stock had no ULOs in the pre.541 ()5.97) ± ± 0.76) 1300 0.164 ()4.56) ± ± )0.000±$25.51) )0.605 (14.95) )0.350 ()16.M.000 to $100. Rj the average high±low range as a fraction of price over period.82) ± ± 0.02) ± ± 0.000 and from $25. .46) 0. NO_ULOBEFORE ˆ 1 if the observation is pre-regulation change and the stock had no ULOs in the pre.000a Dependent variables Spread Panel A: Event 1: $10. Vj is the dollar volume of trading on the ASX.08 (7.12 ()0. t-values are in parentheses.33) 1300 0.98) )0.000±$100.64) )2.741 (26. and zero otherwise.132 (4.652 (15. Aitken et al.148 ()2.or post period.453 ()27.108 ()1.000 Constant Spread Volume Volatility Price Firm size ULOAFTER NO_ULOBEFORE NO_ULOAFTER Sample size Adjusted R2 Panel B: Event 2: $25.149 ()6.69 8.09 ()0.82) ± ± )0.74 Volume 11.000 to $25.591 (24. and 0 otherwise. / Journal of Banking & Finance 25 (2001) 1589±1603 1601 Table 5 Stage least squares estimations: Change of ULO-threshold from $10.33) )0.and post period.42) )0.34) ± ± 0.

which might be due to the reduced e€ectiveness of ULOs around this event. However. For trading volume.1602 M. The use of ULOs increases in volatility and order value and decreases in tick size and trading activity. due to the tightening of undisclosed order regulation in 1994. Our ®ndings suggest that ULOs are used to mitigate the quote-matching and free option problem of limit orders. we ®nd that trading volume for the sample with ULOs decreases and trading volume for the sample with no ULOs increases. Summary and conclusions This study investigates factors a€ecting the use of ULOs on the ASX. We also examine the impact of two recent changes in undisclosed order regulation on market quality. Consistent with event 1. A stock price reaction analysis shows there is no di€erence in price reaction after submission of disclosed and undisclosed limit orders. the di€erence between these changes is not signi®cant (Pvalue is 0. / Journal of Banking & Finance 25 (2001) 1589±1603 We ®nd no evidence of a signi®cant di€erence between the two samples in the change in the bid±ask spread (P-value is 0. These results con®rm the hypothesis that tightening the undisclosed order regulation had the e€ect of discouraging primary liquidity suppliers. 6. The impact of a similar regulation change in 1996 resulted in a less signi®cant decrease in trading volume. we ®nd that the decrease for the ULO sample is signi®cantly smaller that the increase in trading volume for the no-ULO sample (P-value is 0. . F-tests indicate that the change in the bid± ask spread for the sample with no ULOs is not signi®cantly di€erent from the sample with ULOs. The lower signi®cance of the results around the second event might be due to the strongly reduced e€ectiveness of ULOs around this event. We thank two anonymous referees for their constructive comments and Mike Young and Darryl Young for their programming assistance.11).01).15). This result suggests that ULOs are not more frequently used by informed traders than DLOs.J. The regulation dummy variables. Acknowledgements We are grateful to SIRCA and the ASX for making the data available to us. ULOAFTER. Aitken et al. for the change in spread and volume around the regulation event are not signi®cant. resulting in a reduction in trading volume for stocks with ULOs. led to a signi®cant decline in trading volume. We ®nd that an improvement in pre-trade market transparency. This ®nding is consistent with primary liquidity suppliers being discouraged when it is more dicult for them to limit their order exposure. Panel B of Table 5 presents the results for event 2. Again most of the control variables have the expected signs and are signi®cant.

D. . P. D. forthcoming. Biais.M. Chakravarty. 179±208.. 1835±1861. Hasbrouck. Milgrom. 1989. Journal of Financial and Quantitative Analysis 24. Journal of Finance 38 (5). Working Paper..B.P.. Market-making in initial public o€erings of common stocks: An empirical analysis.. Anatomy of the trading process: Empirical evidence on the behavior of institutional traders.R. Harris.. C.. Brown. Sunshine trading and ®nancial market equilibrium. Journal of Financial Economics 37. L.. D. 1992.. L.R. Review of Financial Studies 9 (3. Aitken et al.. Schwartz. Holden.J. 1996. M. Hegde.. 1996. T. R. L. Spatt. market makers. Working Paper.. Journal of Financial Economics 14. Berkman. B. C. Galai. L. Seppi.J. Keim. 1983. Harris. M. Journal of Financial Economics 19. Liquidity. Anonymity in securities markets. University of Southern California. Easley. Aitken. Monograph 1990±1994.. 1994.. George. R. Berkman. 69±90. 1995.. 371±398. Limit and market orders with optimizing traders. Journal of Financial and Quantitative Analysis 31. Hasbrouck. 1985. New York University Salomon Center Monograph Series in Finance in Economics. Handa.. Optimal dynamic order submission strategies in some stylized trading problems. Journal of Futures Markets.. Miller. H. P. Hayes. Limit order trading. 1996. 71±100. 1987..A. J. M. 1996. and information in securities markets. Working Paper. Information e€ects of the bid±ask spread.. S. 1990. Journal of Finance 50. 1995. 75±91. 2000.. 1655±1689. P. 977±1002. J. L.. A. H. 1995. University of Southern California... 1997.. 1991. Fall). Hillion. Carnegie Mellon University. Kumar. 213±241. Journal of Finance 46 (1). Measuring the information content of stock trades.. Harris. Large option trades. ask and transaction prices in a specialist market with heterogeneously informed traders. trade size..E. P¯eiderer.. Review of Financial Studies 4. T. 443±481. Infrequent trading and ®rm size as explanations for the intra-day patterns in returns on SEATS.W. Price. L.. Forster. Journal of Financial Intermediation 4. limit orders: The SuperDOT evidence on order submission strategy. / Journal of Banking & Finance 25 (2001) 1589±1603 1603 References Admati. Harris. A.E. An integrated model of market and limit orders.. 1991. and limit orders... Does a large minimum price variation encourage order exposure? Working Paper. 1996. Madhavan... Securities Industry Research Centre of Asia±Paci®c. An empirical analysis of the limit order book and the order ¯ow in the Paris Bourse. OÕHara. Glosten. The role of ¯oor brokers in the supply of liquidity: An empirical analysis.R. D. 213±231. 1457±1469. trading rules. Walter. Market vs.. P. P. and electronic trading systems. 168±206.. S. P.. Copeland. Journal of Financial Intermediation 2. Bid. Journal of Finance 51. T.