Hidden Liquidity Inside the Spread

Transcription

1 Hidden Liquidity Inside the Spread James Holcomb Associate Professor of Economics University of Texas at El Paso James Upson* Assistant Professor of Finance University of Texas at El Paso Abstract: Using a sample of 300 NYSE listed securities and excluding trade reporting facility trades, we examine the impact of order executions inside of the posted spread on market centers, an inside the spread trade (IST). The posted spread on a market center may differ from the National Best Bid and Offer Quote (NBBO). IST volume represents a substantial portion of daily volume executed at market centers, averaging 14%. However, only 51% of these trades improve on NBBO prices. Although the volume is substantial, cost savings from IST s is small, averaging $132 per stock day. We find evidence that hidden liquidity inside the spread tends to act as a loss leader, drawing additional trades after the successful execution of liquidity inside the spread. When an exchange s quotes are competitive with the NBBO, the probability of observing an IST is lower but when intermarket competition is high the probability of observing an IST increases. We find that IST s are generally uninformed. Our results are robust to firm size and trading intensity. *Contact Author 1

2 Abstract: Using a sample of 300 NYSE listed securities and excluding trade reporting facility trades, we examine the impact of order executions inside of the posted spread on market centers, an inside the spread trade (IST). The posted spread on a market center may differ from the National Best Bid and Offer Quote (NBBO). IST volume represents a substantial portion of daily volume executed at market centers, averaging 14%. However, only 51% of these trades improve on NBBO prices. Although the volume is substantial, cost savings from IST s is small, averaging $132 per stock day. We find evidence that hidden liquidity inside the spread tends to act as a loss leader, drawing additional trades after the successful execution of liquidity inside the spread. When an exchange s quotes are competitive with the NBBO, the probability of observing an IST is lower but when intermarket competition is high, the probability of observing an IST increases. We find that IST s are generally uninformed. Our results are robust to firm size and trading intensity. 2

3 1.0 Introduction The use of hidden liquidity, concealing all or part of an order from exposure in the limit order book (LOB) is a common feature of electronic order driven markets around the world. Hidden liquidity generally comes in two forms. Iceberg orders refer to a market state where there is depth displayed at a given price and there is depth available at a non-displayed price. When the display portion of an iceberg order is executed, a portion of the hidden liquidity becomes visible. Just as if you chop the top off an iceberg, the hidden portion under the water floats to the surface. For example, consider an order for 1,000 shares with a display portion of 200 shares and a hidden portion of 800 shares. When the 200 share display part executes, 200 shares from the 800 hidden are moved to display, with 600 remaining hidden. 1 The key aspect of iceberg order is that the hidden depth is advertised in the sense that there is a displayed interest in trading at a given price. Bessembinder, Panayides, and Venkataraman (2009) find that the locations and size of the hidden part of iceberg orders can be predicted, though imperfectly. Fully hidden liquidity is an order where there is no displayed trading interest. Fully hidden orders can be posted inside of the market centers displayed quotes and require an explicit search effort on the part of the liquidity demander to locate. Hasbrock and Saar (2009) investigate the search for hidden liquidity inside the spread and identify the use of fleeting orders, orders that appear and disappear from the limit order book very quickly, as evidence of this search effort. The liquidity demander is rewarded for the search effort by receiving a price that is better than the display price of the market center quote. Our research focuses on this second type of order, fully hidden liquidity inside of the quoted spread on a market center. We exclude from our analysis trades reported through a trade reporting 1 The recharge order size need not be constant but can be set as a percentage of the hidden base or as a random number. See as an example of iceberg orders. 3

4 facility, which are executed on an off exchange venue, such as a dark pool. Since by construction fully hidden liquidity cannot be observed, we infer the hidden liquidity by observing a trade. An Inside the Spread Trade (IST) is defined as a trade that occurs at a price that is inside of the in force quote from the market center at the time of the trade and is inside the next quote update from the market center. The second requirement controls for potential lags in reporting of trades and quotes to the Daily Trade and Quote (DTAQ) database used in our analysis. 2 Our sample consists of 300 NYSE listed stocks, randomly selected from low, medium, and high trading intensity firms. The sample period is the first quarter of IST volume represents 14% of the volume executed on exchanges over our sample period. We find that IST volume is higher than average at the start of the trading day and lower than average at the end of the trading day. Comparing IST trade prices against NBBO quotes, we find that only 51% of the IST trades are inside of the NBBO quote, while 46% of IST trades are at NBBO quote prices, with the remaining IST s occurring at adverse prices to the NBBO quote. While the IST may represent a price improvement over the executing market centers displayed quote, relative to the NBBO of the market, price improvement of IST s is basically a coin flip. We estimate the savings, relative to NBBO prices, for IST s and find that on average only $ is saved per stock day. Even for high trade intensity stocks in our sample, the average daily savings is only $ There are a number of stock days in our sample where the savings from IST trades are negative with the worst day displaying a savings of -$9,770.99, while the best day of savings is $11, The costs to liquidity demanders associated with searching for fully hidden liquidity are unknowable; however, even with relatively small search costs, the benefits of IST s to liquidity demanders appear to be small. 2 Ding, Hanna, and Hendershott (2014) find that the consolidated tape National Best Bid and Offer quote differ from the prices in faster proprietary data feeds several times a second, but the differences resolve in one to two milliseconds indicating that the latency is relatively small. 4

5 In iceberg orders there is a tradeoff between the exposed portion of an order and the hidden portion of the order. 3 Hidden liquidity generally loses time priority of execution and cannot execute until all displayed liquidity executes. However, fully hidden liquidity inside the spread maintains price priority over displayed liquidity at adverse prices and would execute before the displayed liquidity. Thus, the tradeoff for fully hidden liquidity inside the spread must balance the increase in execution probabilities based on price priority with the decrease in execution probability due to the lack of an advertised price and displayed trading interest. The execution probabilities will depend on specific liquidity characteristics of the market center where the hidden liquidity will be posted. We evaluate the percentage of IST volume relative to total volume executed on a market center and find a wide range of values. On the National Stock Exchange (exchange C in DTAQ), roughly 64% of volume is IST, while on the NYSE only 4.63% of volume is IST. Even on duel platform exchanges such as DirectEdge and BATS there is substantial differences in IST volume. On DirectEdge 38% of volume is IST but on DirectEdge A 52% of volume is IST. On BATS IST volume is 22% while on BATS Y 14% of volume is IST. The wide disparity between IST volumes at the exchange level indicates that there is a significant strategic component to the placement of fully hidden liquidity. Potentially, one reason for posting fully hidden liquidity is for the hidden order to act as a loss leader. The hypothesis is once the market observes an IST, additional orders are routed to the market center reporting the IST. We define two trade types, an IST, and a Display Trade (DT) which is a trade at the displayed quote price on the market center. Given a DT, we calculate the number of seconds until the next trade execution, IST or DT, occurs on the market center. In addition we calculate given an IST, how many seconds elapse before the next trade execution, IST 3 See Bessembinder et el (2009) or Frey and Sandas (2009) for a more complete discussion on the tradeoff between display and hidden portions of Iceberg orders. We refrain from a more complete discussion here because iceberg orders are not the focus of our research. 5

6 or DST occurs. If IST s are acting as a loss leader, then the elapsed time between the next trade and an IST should be shorter than the elapsed time between a DT and the next trade. Our results are mixed. On 6 of the 11 active exchanges in our sample, we find that the elapsed time between trades after an IST trade is significantly shorter. The findings are roughly equivalent for each trade intensity group. As a second test of the loss leader hypothesis, we estimate a logistic regression on the probability of an IST trade occurring. The regression is run for each stock day and then the coefficients are aggregated. We find that the probability of observing an IST decreases if the exchange is posting prices that are equal to the market NBBO. This is true if the market is only competitive on one side of the market. However, we show that as intra-market competition for order flow increases, the probability of observing an IST increases. Pardo and Pascual (2012) investigate the information content of iceberg orders on the Spanish Stock Exchange and find that the market does not attribute relevant information content to the hidden portion of the order. Gozluklu (2014) finds the hidden liquidity does not affect informational efficiency of markets. Boulatov and George (2013) find that the ability to hide liquidity increases the tendency for informed traders to provide liquidity. Bloomfield, O Hara, and Saar (2014), using a laboratory market, claim informed traders tend to hide liquidity inside of the spread. We use the information shares method of Hasbrouck (1995) to investigate the contribution to price discovery of process of IST s. For large stocks the price discovery of IST s is significantly smaller that would be expected by the volume share of these trades. For medium stocks the price contribution of IST s is roughly proportional to the volume share. However, for small stocks, price discovery from IST s is much larger than would be expected given the volume share. The market impacts of fully hidden liquidity inside the spread are studied by Hautsch and Huang (2012) using Nasdaq ITCH data. Similar to Bessembinder et el (2009), they find that fully 6

7 hidden liquidity prices are predictable based on the observable market state on Nasdaq. Our analysis complements theirs in several ways. First, we find that roughly 46% of IST s execute at the market NBBO price. This adds another explanatory variable to the selection on the price grid for the placement of fully hidden liquidity. Second, we show that when market centers display quotes are competitive with the NBBO, execution of IST s is less likely, which implies, but does not prove, that hidden orders are less prevalent when display prices are more competitive. Finally, we show that when intermarket competition is high, the probability of IST s also increases. This indicates that the competition for the hidden liquidity provision increases as the competition for the display liquidity provision increase on a market wide basis. The balance of the paper is formatted as follows. In section 2.0 we conduct a literature review and in section 3.0 develop our hypotheses. Section 4.0 reviews the data and sample characteristics. Section 5.0 presents the results and section 6.0 concludes the paper. 2.0 Literature Review There is significant theoretical and empirical research on the use, strategies, and market impacts of hidden liquidity and pre-trade transparency. Madhavan, Porter, and Weaver (2005) investigate the change in pre-trade transparency on the Toronto Stock Exchange where the exchange increasing the reporting of the limit order book resulted in a decrease in market liquidity. Boulatov and George (2013) and Baruch (2005) model markets with varying degrees of pre-trade transparency and find that opaque markets have a higher liquidity provision. Buti and Rindi (2013) model the choice of uninformed traders choice to expose or hide all or a portion of an order. Moinas (2010) models the choice of informed traders to expose or hide liquidity. Her model indicates the informed traders will post larger orders and improve market liquidity when hidden orders are allowed. 7

8 Bessembinder, Panayides, and Venkataraman (2009), investigate hidden liquidity on the Euronext-Paris market and find that both the presence and magnitude of hidden liquidity can be predicted. Frey and Sandas (2009) investigate the impact of iceberg orders and find they attract liquidity when discovered. Aitken, Berkman, and Mak (2001) investigate hidden liquidity on the Australian Stock Exchange and find hidden liquidity is used to reduce the option value of limit orders. Pardo and Pascual (2012) investigate the information quality of hidden liquidity and find that market participants do not associate hidden liquidity with relevant information content. Hautsch and Huang (2012) look at hidden liquidity inside of the spread. The find that liquidity hidden inside of the spread is predictable based on observable market conditions. Empirical papers are only able to observe data and are unable to identify the motivations of traders positing hidden liquidity. Two papers conduct laboratory experiments to help identify the motivation of traders posting hidden liquidity. Gozluklu (2014) finds that when there is competition between informed traders book depth increases when liquidity can be hidden. Bloomfield, O Hara, and Saar (2014) find that the ability to hide liquidity impacts the trading strategies of market participants in terms of display quantities and aggressiveness of trading. However, they find that market liquidity and informational efficiency are not significantly affected with hidden liquidity. 3.0 Hypothesis Development Why hide liquidity? The literature indicates that a major reason to hide liquidity is to hide the trading intention of the liquidity supplier. Buti and Rindi (2011) indicate that the display of large orders can increase completion for the liquidity provision, leading to price undercutting. Harris (1997) indicates that posting of limit orders can trigger front-running strategies from other traders and Harris (1996) indicates that pick-off risk increases with the posting of limit orders. We propose 8

9 a different economic incentive for posting fully hidden liquidity. Our theory is that fully hidden liquidity acts as a loss leader. Once the presence of hidden liquidity is identified with an IST trade, liquidity demanders route additional trades to the market center which executed the IST. These trades may execute against additional hidden liquidity or execute against display liquidity. We define a Display Trade as a trade at a price equal to the displayed quoted price. This leads to the following hypotheses: H1: The time between an IST and the next trade will be shorter than the time between a Display Trade and the next trade. H2: As intermarket competition increases, the probability of observing an IST increases. The second hypothesis is also based on the loss leader concept. If, for example, all active exchanges are posting liquidity at the NBBO price, there is no economic incentive for a liquidity demander to route a trade to any particular market center. However, if liquidity suppliers post hidden liquidity inside of the NBBO spread and an IST trade executes against the liquidity, it signals the market of potential price improvement on a particular market center, possibly improving the chances of having additional trades routed to the market center. Boulatov and George (2013) develop a model of hidden liquidity supply and show that informed traders provide more liquidity when liquidity is hidden. Although uninformed traders may still supply hidden liquidity, if the hidden liquidity provision is dominated by informed traders, price reactions to IST trades should contributed to price discovery in the market. We investigate the information quality of fully hidden liquidity using the information shares method of Hasbrouck (1995). Specifically we form two price channels, one for IST s and the second for Display trades. We estimate the information share for each stock day in the sample. The information shares method generates a high and low limit to the contribution of a price channel to price discovery. We average the high and low limit to generate a point estimate of the information share. Following 9

10 Chakravarty, Jain, Upson, and Wood (2012) we compare the information share of a price channel to the volume share of the price channel. A price channel is more informed if the information share is greater than the volume share of the price channel. Our hypothesis is: H3: The information share of IST s is strictly greater than the volume share of IST s. Operationally, we take the last trade in each second as the point estimate of the price for a price channel to limit the computational challenges of the information shares method. All trade volume is used to calculate the volume share of a price channel. 4.0 Data and Sample 4.1 Data Our data is the Daily Trade and Quote (DTAQ) database for the first quarter of The DTAQ database includes the exchange calculated NBBO, additional condition codes, and is time stamped to the millisecond. Holden and Jacobsen (2013) recommend the use of the exchange calculated NBBO for calculation of transaction costs as the first best option. We consider trades and quotes from market open at 9:30 am EST to market close at 4:00 pm EST. There are a number of filters we place on trades in the DTAQ database. First all trades reported through a Trade Reporting Facility (TRF) are dropped from the analysis. While a significant number of TRF trades execute inside of the spread, they are not related to hidden liquidity at the market center reporting the trade. Second, we drop the opening trade on the listing exchange. These trades are identified with condition code O. The opening trade is often at the prevailing quote midpoint but does not represent hidden liquidity. Trades with the following conditions are also dropped from the analysis: Average Price Trade (code B), Cash Trade (C), Price Variation Trade (H), CAP Election Trade (I), Rule 127 (K), Sold Last (L), Next Day Trade (N), 10

11 Prior Reference Price (P), Seller (R), Stock Option Trade (V), Cross Trade (X), Sold (Z), Derivatively Priced (4), Market Center Re-Open (S), and Market Center Closing Trade (6). Special attention is paid to quotes to track the withdrawal of a market center from active participation in the market. A market center can issue a quote with a positive ask and bid price but with zero ask and bid depths. In such cases, the exchange is considered as having an infinite spread. If the depth is positive on one side of the market but zero on the other side, the side with positive depth is considered. A market can indicate a sided withdrawal by issuing a quote with a zero ask price, indicating that no interest is available on the ask side of the market, or a zero bid price, indicating that there is no interest on the bid side of the market Identifying Hidden Liquidity Trades Inside the Spread Trades (IST s) are identified based on the following algorithm. Two tests are applied to each trade. First, the trade price is tested against the in force quote from the exchange executing the trade. If the trade price is between the ask and bid prices, the trade passes the first test for an IST. If the prevailing quote is a market withdrawal quote, as detailed in the previous section, the trade passes the first test. If the market has withdrawn from only one side, for example there is no active display interest on the ask side, then the trade passes the test if it is greater than the prevailing bid. The second test is based on the first quote update from the market center executing the trade, after the trade has occurred. For the trade to pass this test it must again be between the ask and bid price of the updating quote. If the updating quote is a full market withdrawal quote, the trade passes the test. If the updating quote is a partial market withdraw, for example on the ask side, the trade passes the test if the execution price is above the bid. This second test controls for potential latency issues in reporting trades and quotes. 11

12 4.2 Sample The sample selection process starts with the CRSP data base. CRSP is used to identify only common stocks to be included in the sample, security codes 10 and 11. Only NYSE listed securities are considered. Next we use the DTAQ data base to calculate the average number of trades per day for each equity in the data base. We merge the CRSP and average trade samples, keeping only common stocks. We next apply two filters. First, the average price of the security must be between $10 and $500 dollars. In addition, we require stocks to have a minimum of 500 trades per day on average and a maximum of 30,000 trades per day. We wish to eliminate hyper-liquid stocks such as Bank of America where a typical trading day has over 100,000 trades. Such stocks often have bid ask spreads that are bound at the minimum tick size of one cent, quoted depths that skew liquidity calculations, and multiple NBBO updates in the same millisecond. The remaining sample is then ranked by trade frequency into three groups, High, Medium, and Low. 100 stocks are then randomly selected from each group for a final sample of 300 stocks. Table 1 shows descriptive statistics of the sample and sub-samples. We present the mean and standard deviation (std) of market capitalization, average daily trade frequency, and average daily returns. Market capitalization is based on the first day of our sample, January 3, Although we group by trade frequency, the market capitalization results indicate the High, Medium, and Low trade frequency stocks are also Large, Medium, and Small relative to market capitalization. For the balance of the paper we refer to our groups as Large Stocks, Medium Stocks, and Small Stocks since these designations correspond to both trade intensity and market value. The small, medium, and large subsamples have an average number of daily trades of 3,082, 5,868, and 12,642. Although there is an increase in the trading intensity over the sub-groups, our small stocks have significant trading interest over the day. We believe that our sample of stocks 12

13 represents an active group of securities that are traded by institutions and retail traders, yet avoid analytical problems associated with hyper liquid stocks and very low liquidity stocks Results 5.1 Daily and Intraday IST volume Our analysis starts by examining the times series plots of the percentage of volume executed with IST s. Figure 1 shows the plot of the percent of IST volume for each day in our sample segmented by firm size/trading intensity. The plot indicates that volume executed with IST s is relatively constant over the sample period with the exception of March 16, We investigate the market events of this day and find that a large amount of positive economic news is released on this day. 5 Hautch and Huang (2012) find that the use of hidden liquidity decreases when the time horizon of trading becomes shorter or when time priority becomes more important. On a day with large amounts of positive economic news, traders may move away from hidden liquidity in order to fill desired positions quickly, before the economic information is fully absorbed into the market price. The drop in IST volume on this date is consistent with the findings of Hautch and Huang. The plot does not indicate that there is an end-of-month effect on IST volume. Market centers offer rebates to liquidity suppliers that post resting orders which are then executed. The level of rebate is contingent on the total volume executed on the market center by the liquidity supplier, with higher rebates going to liquidity suppliers executing more volume. An end-of-month effect could arise on IST volume if liquidity suppliers use hidden liquidity to draw additional transaction volume to the market in order to reach a break point in the volume rebate schedule. The 4 An example of a problem with very low liquidity stocks is the tendency of market withdrawal. Specifically, a market center may issue a quote with positive ask and bid prices but with zero depth. Such quotes are not NBBO eligible. For very low liquidity stocks, there are portions of the trading day where no NBBO exists. 5 The information on the economic announcements on March 16, 2012 can be found at 13

14 lack of a clear end-of-month signal argues against the possibility as a potential motivation for posting hidden liquidity. In Figure 2 we plot the intra-day percentage of IST volume. The plot shows the results for the full sample. In unreported results, the plot is conditioned on firm size/trade intensity with identical results. Hautch and Huang (2012) also test for intra-day patterns in the execution of hidden liquidity but find little evidence to support the hypothesis. From our market wide perspective, we find that IST volume is higher than average in the first 30 minutes of trading and decreases sharply during the final 10 minutes of trading. IST volume is relatively constant during the middle of the trading day. 5.2 IST NBBO Price Grid Evaluation While, by definition, all IST trades occur inside of the spread of the executing exchange, this section of the analysis locates the placement of IST s relative to the NBBO price grid. The results are presented in Table 2. For the full sample, only 51% of IST s execute inside of the NBBO spread. In addition, roughly 23.25% of IST s execute at prices equal to the NBBO ask or NBBO bid price. These results indicate that the price improvement implied by a trade inside the spread is less that would be expected. 6 We also find that roughly 1.25 % of IST s execute below (above) the NBBO bid (ask). Table 2 also presents results for small, medium, and large stocks in our sample. The percentage of trades executing inside of the NBBO spread is decreasing in firm size/trading intensity. While 54.84% of trades execute inside the NBBO for small stocks, only 45.77% of trades execute inside of the NBBO spread for large stocks. As the percent of IST s inside of the NBBO decrease, more IST s execute at prices equal to the NBBO quote. However the percent of trades 6 We assume that IST s at the ask are buyer initiated and that IST s at the bid are seller initiated. It is possible that hidden liquidity is being used to stealth lock the market meaning the ask trade could be seller initiated. Liquidity suppliers would lose the benefit of the spread but receive the liquidity supplier rebate. 14

15 outside of the NBBO quote remains roughly constant over the range of firms. Given that, as demonstrated by Hasbrouck and Saar (2009), search costs are incurred by the liquidity demander to locate hidden liquidity, the transaction cost saving for IST s may be relatively modest. We estimate these savings in the next section. 5.3 Transaction Cost Savings Our transaction cost analysis uses the following calculations. IST dollar volume is calculated at the IST transaction price*trade size. The transaction cost saving for an IST is calculated as follows. If the trade is below the NBBO midpoint, the transaction cost savings are equal to (Price IST - NBBO Bid )*Trade Size. If the trade is at the NBBO midpoint, or above the NBBO midpoint, the transaction cost savings are equal to (NBBO Ask -Price IST )*Trade Size. Total spread costs is calculated as the absolute value of (Price-NBBO Midpoint )*Trade Size. All trades, IST and Display trades (trades at displayed prices) are included in this calculation. Each of these variables is summed for each stock day in the sample. % Savings is the ratio of Spread Savings to Total Spread Costs. Finally, % Return is the ratio of Spread Savings to Dollar Volume (IST). In Table 3 we present the mean, standard deviation, minimum and maximum of each variable for the full sample and conditioned on firm size. Panel A presents the results for the full sample. Perhaps the most striking finding is that the average savings per stock day from IST trades is only $132.58, with a standard deviation of $ It should also be noted, that while $132 on average is saved by the market, these savings must be divided between the various liquidity demanders that execute IST s, likely resulting in much lower savings per liquidity demander. For example, if only 10 liquidity demanders, on average, are splitting the savings equally from IST s, the savings per demander is only $ Given even minimal search costs to find hidden liquidity, the 15

16 actual savings, on average, could be smaller or even negative. In addition, when we examine the minimum value of spread savings for the full sample it is negative, with a value of -$9,770. Relative to total transaction cost, the spread savings represent roughly 2.2%. The percent of spread savings is higher for smaller stocks and decreases as firm size increases. One can view the spread savings as an instantaneous return on the dollar volume invested in the security. For the full sample the return percentage on spread savings is a quite modest 0.005%. While positive, our analysis does not take into consideration the potential search costs associated with hidden liquidity. Overall our cost analysis for IST s indicates a very modest savings in explicit transaction costs. 5.4 Market Center Analysis In the integrated stock markets of the U.S., liquidity suppliers have several choices to make when submitting a limit order: Price, Quantity, and which market center the order will be placed on. The same options are available to suppliers of fully hidden liquidity. In this section we evaluate the use of fully hidden liquidity on a market center by market center basis. Our results are shown in Table 4. The results are presented for the full sample, Panel A, and conditioned on firm size/trading intensity, Panels B-C. %Hid Total is the percentage of fully hidden liquidity executed on a market center with respect to the total volume executed on exchanges. %Hid Ex is the percentage of hidden liquidity executed on a market center with respect to the total volume executed on the market center. Joint is the χ 2 statistic of joint equality of % Hid Ex. The test rejects joint equality below the 1% level. Our results show there is substantial variation in the proportion of fully hidden liquidity executed on individual market centers. Even for market centers that are under the same umbrella exchange, such as Direct Edge and Bats, there is variation. 7 Direct Edge A (DirA) market center has 7 In 2014, BATS purchased Direct Edge to become the second largest exchange in the U.S. 16

17 1.57% of the total hidden volume on exchanges and 51.98% of volume on Direct Edge A is from IST s. However, on Direct Edge 3.8% of the hidden volume executes but 38.18% of market center volume is hidden. The National Stock Exchange (Nat, the old Cincinnati Stock Exchange) executes only 0.29% of hidden volume from a market perspective but 63.55% of market center volume is IST. All of our sample stocks are listed on the NYSE which executes 1.63% of hidden volume, which is 4.62% of volume executed on the NYSE. The conditional results based on firm size/trading intensity remain very consistent. For each market center both the percentage of hidden volume relative to market volume, %Hid Total, and the percentage of hidden volume relative to exchange volume, %Hid Ex, are economically indistinct for many of the market centers. Direct Edge executes 3.8% of hidden volume in the market for small and large stocks and 3.79% for medium stocks. The percentage of hidden volume on Direct Edge goes from a low for small stocks at 37.24% to a high of 39.19% for large stocks. The stable results for each conditioning group, on each market center, imply that liquidity providers on these market centers adopt a consistent strategy for the use of hidden liquidity. 5.5 Time Between Trades Hypothesis 1 is based on the concept that hidden liquidity is being offered as a reward for future trades routed to the market center executing the IST trade. Liquidity demanders may be looking for additional hidden liquidity and the small but potential price improvement from executing against hidden liquidity inside of the NBBO quote. We test hypothesis 1 as a conditional expectation on a stock by stock basis. First, we estimate the average time between trades, given that the last trade was a Display trade. We call this metric D Time. Next, we estimate the average time between trades, given that the last trade was an IST. We call this metric H Time. We then conduct a paired t-test between D Time and H Time. The results are presented in Table 5. 17

18 Panel A of Table 5 reports the results for the full sample. Six of the eleven active market centers in our study show that H Time is shorter than D Time, supporting the contention of hypothesis 1. Three of the market centers that have H Time being longer than D Time, Nat, DirA, and DirX, also have the three highest percentages of IST volume executed on the market centers, 63.55%, 51.98%, and 38.18% respectively. Given the high percentage of IST volume on these exchanges, liquidity suppliers may have adopted a different trading strategy for hidden liquidity. Nasd, Arca, CBOE, Phil, BatsY, and Bats all have lower times between trade executions given and IST trade, than the time between trades given a Display trade execution. 8 The NYSE, the listing exchange for the sample stocks, has a shorter trade time after a Display trade than compared to an IST trade. The NYSE also has the shortest time between trades for both IST and Display trades. While the NYSE is a special case, since it is the listing exchange, the result still fails to support the hypothesis. Panel B reports the results for the small stocks in our sample. The results are stronger for small stocks than for medium stocks. Six of the eleven market centers have shorter times between trades after and IST. For medium stocks, Panel C, only five of the market centers have shorter H Times. For large stocks, Panel D, we again find six of the market centers with shorter H Times. By random chance, given eleven active exchanges, we could find a significant difference in 1 of the exchanges at the 10% level, so we believe that our results are more than just chance. One question that arises is the economic significance of, for example, a 2.2 second improvement on the time between trades on the Nasdaq (Nasd) market center. However, in a market that now moves on a nanosecond time step, 2.2 seconds is a long period of time. Liquidity 8 The DTAQ database we are using does not report odd lot trades. O Hara, Yao, and Ye (2013) show that a large portion of trades are odd lot. This may induce an upward bias in our results of the time between trades since we cannot see odd lot trades executing. Davis, Roseman, Van Ness, and Van Ness (2014) find that one share trades are used to search for hidden volume. The inclusion of odd lot trades in the analysis may improve the results. 18

19 suppliers gaining a 2.2 second advantage on other liquidity suppliers on the same market center can manage inventory positions and structure future limit orders on the market center, hidden or display, based on the information generated from the trade execution. 5.6 Market Competition In this section of the analysis we evaluate the impact of market conditions on observing and IST. We use a logistic regression of the following form: IST AtAsk AtAskBid AtBid QFrag (1.1) ikt,, i 1 ikt,, 2 ikt,, 3 ikt,, 4 it, it, Where IST is 1 if the trade is against hidden liquidity for stock i, on market center k, at time t, and zero otherwise. AtAsk is 1 if the market center executing the trade has an ask price equal to the NBBO ask price, while AtBid is 1 if the market center executing the trade has a bid price equal to the NBBO bid price. If the market center has ask and bid prices equal to the NBBO ask and bid price, AtAskBid is set to 1 and AtAsk and AtBid are set to zero. If the market centers quote is not competitive AtAsk, AtBid, and AtAskBid are all set to zero. We capture intermarket competition with the variable Qfrag. Following Madhavan (2011), Qfrag is calculated 1-(HerfAsk+HerfBid)/2. HerfAsk is the Herfindahl Index of all ask depth with a price equal to the NBBO ask and HerfBid is the Herfindahl Index of all bid depth with a price equal to the NBBO. Ask and example the HerfAsk is calculated as K ( AskDpth / ) 1 k Ik NBBO AskDpth where I k is an indicator variable that is 1 if the market center ask k price is equal to the NBBO ask and zero otherwise. Higher levels of Qfrag indicate increased intermarket competition. We estimate the regression for each stock day in the sample. We then aggregate the results based on the following method. The coefficient, β is the weighted average of all β s where the weighting is based on the precision of the estimate. Specifically : 19

20 ˆ it, 2 it, 1 2 it, (1.2) Where β i is the estimated coefficient for stock i on day t and σ 2 is the standard error for the coefficient. The t statistic for the aggregated results is then: ˆ t stat (1.3) 1 2 it, The regression results are reported in Table 6. The results are presented for the full sample and conditioned on firm size/trading intensity. The results are very consistent in sign, magnitude, and significance of the confidents so we discuss the results for the full sample to conserve space. Our results indicate that the probability of observing and IST decreases significantly when a market center is posting prices that are competitive with the NBBO. The reduction in probability is identical whether the market center is quoting only on the ask side of the market or only on the bid side of the market. In addition, when the market center is positing a quote that is equal to the NBBO on both the ask and bid side of the market, the probability of observing an IST decrease even more. The evidence indicates when liquidity suppliers on a market center are posting competitive quotes there is a significant decrease in the likelihood of observing an IST. The coefficient of Qfrag is positive and significant. Qfrag measure the amount of inter-market competition for the liquidity provision. As inter-market completion increases liquidity suppliers may post hidden liquidity inside the spread so that an IST execution may draw additional order routings to the market center executing the IST. Our regression results support hypothesis 2. 20

21 5.7 Information Share The model of Boulatov and George (2013) indicates that informed traders will provide more liquidity when resting orders can be hidden. If the market believes that hidden liquidity inside the spread is provided by informed traders, the price contribution of IST may add disproportionality to the price discovery in the market. We test this hypothesis, H3, by estimating the information share of IST s based on the method of Hasbrouck (1995). The information shares method produces and upper and lower bound of contribution to price discovery for each price channel used in the analysis. We average the upper and lower bound to generate a point estimate of the information share. We then compare the point estimate of the information share to the volume share of the price channel. If the information share is higher than the volume share it indicates that the order flow contribute more to the price discovery process than would be indicated by the volume share of the order flow. We also test the information ratio, defined as the information share point estimate divided by the volume share. An information ration of 1.0 indicates that the order flow carry information directly proportional to the volume share. The results are shown in Table 7. For the full sample the mean information share if IST s is while the mean volume share is The difference is statistically significant at the 1% level. The mean information ratio is indicating that IST volume contributes 22.3% more to price discovery than would be implied by the volume share. The information ratio is statistically different from 1.0 at the 1% level. Turning to the information share results conditioned on firm size/trade intensity, IST volume on large stocks contributes less to price discovery than would be implied by the volume share. The information share for IST volume is while the volume share is In addition the information ratio is only which is statistically different form 1.0 at the 1% level. For medium sized stocks the information share of IST volume is roughly equal to the volume share. The difference of is statistically significant but not truly economically significant. The 21

22 information ratio is Again, the value is statistically different from 1.0 but economically insignificant. The results for the full sample are being driven by the small/low trade intensity stocks in the sample. The information ratio of IST volume is while the volume share is The difference is and is significant at the 1% level. The information ratio is and is significantly different from 1.0 at the 1% level. Our results indicate that IST volume for small stocks contributes 70% more to the price discovery process than would be implied by the volume share. The high amount of price discovery coming from IST volume indicates that informed traders use hidden liquidity extensively for stocks with lower trade intensity. For high trade intensity stocks it might be easier for informed traders to hide in the sheer volume of trading or informed traders may adopt a passive liquidity supply approach to trading but post order as Icebergs or as display orders, rather than fully hide the orders. Small, low trade intensity stocks might be more difficult for informed traders to take positions in without having a large price impact. The use of fully hidden liquidity may mitigate the price impact of informed trader s trades. However, the information share method does not allow us to discern which aspect of IST s drives the price discovery process; active demand orders or passive supply orders of fully hidden liquidity. We can say that fully hidden liquidity is an important aspect to price discovery for small, lower trade intensity firms. 6.0 Conclusion Hidden liquidity is an important source of liquidity supply in modern electronic markets. There are two basic types of hidden liquidity: iceberg orders and fully hidden orders. Our study focuses on fully hidden resting orders that execute inside of the quoted spread of a market center. Using a sample of 300 NYSE listed firms over the first quarter of 2012 we investigate the market 22

23 impact of fully hidden liquidity. Our analysis does not include off exchange trades that are reported through a Trade Reporting Facility. Inside the spread trades (IST s) account for roughly 14% of the volume executed in the market. IST volume is roughly equal for small, medium, and large firms as a percentage of total volume. Surprisingly, transaction cost savings for IST s is only $132 per stock day. Given that search costs are required to find fully hidden liquidity, the transaction costs savings are quite low in our view. We find some evidence that fully hidden liquidity acts as a loss leader for liquidity suppliers. The hypothesis is the liquidity suppliers post fully hidden liquidity so that when an IST is reported additional volume is routed to the exchange reporting the IST. We find that the time, given an IST execution, between trades is smaller than the time between a trade at the display price and the next trade for six of the 11 active exchanges in our sample. In addition, the probability of observing an IST is higher when inter-market competition is also high. However, the probability of observing an IST when a market center is posting competitive quotes is lower. We use the information share method of Hasbrouck (1995) to examine the price discovery contribution of IST volume. For the large stocks in our sample we find that IST volume has lower contribution to price discovery than would be implied by the volume share of IST volume. For medium sized stocks the information share of IST volume is economically equal to what would be implied by the volume share if IST s. However, IST volume for small stocks carries 70% more contribution to price discovery than would be implied by the volume share if IST s. Our analysis contributes to the growing literature on the use of hidden liquidity supply in modern markets. 23

24 References Aitken, Michael, Henk Berkman, and Derek Mak, 2001, The use of undisclosed limit orders on the Australian stock exchange, Journal of Banking and Finance 25, Chakravarty, Sugato, Pankaj Jain, James Upson, and Robert Wood, 2012, Clean Sweep: Informed trading through intermarket sweep orders, Journal of Financial and Quantitative Analysis 47, Bessembinder, Hendrik, Marios Panayides, and Kumar Venkataraman, 2009, Hidden liquidity: An analysis of order exposure strategies in electronic stock markets, Journal of Financial Economics 94, Bloomfield, Robert, Maureen O Hara, and Gideon Saar, 2014, Hidden liquidity: Some new light on dark trading, Journal of Finance, forthcoming. Boulatov, Alex, and Thomas George, 2013, Hidden and displayed liquidity in securities markets with informed liquidity providers, The Review of Financial Studies 26, Buti, Sabrina, and Barbara Rindi, 2013, Undisclosed orders and optimal submission strategies in a limit order market, journal of Financial Economics 109, Ding, Shengwei, John Hanna, and Terrence Hendershott, 2014, How slow is the NBBO? A comparison with direct exchange feeds, Financial Review 49, Frey, Stefan, and Patrik Sandas, 2009, The impact of iceberg orders in limit order books, AFA 2009 San Francisco Meetings Paper, available at SSRN: Gozluklu, Arie, 2014, Pre-trade transparency and informed trading: Experimental evidence on undisclosed orders, working paper University of Warwick, available at SSRN: Harris, Larry, 1996, Does a large minimum price variation encourage order exposure, working paper University of Southern California. Harris, Larry, 1997, Order exposure and parasitic traders, working paper University of Southern California. Hasbrouck, Joel, 1995, One security, many markets: Determining the contributions to price discovery, Journal of Finance 50, Hasbrouck, Joel, and Gideon Saar, 2009, Technology and liquidity provision: The blurring of traditional definitions, Journal of Financial Markets 12, Hautsch, Nikolaus, and Ruihong Huang, 2012, On the dark side of the market: Hidden order placements, working paper, available at SSRN: 24

25 Holden, Craig, and Stacey Jacobsen, 2014, Liquidity measurement problems in fast, competitive markets: Expensive and cheap solutions, Journal of Finance 69, Madhavan, Ananth, 2011, Exchange-traded funds, market structure, and the flash crash, working paper BlackRock, available at SSRN: Madhavan, Ananth, David Porter, and Daniel Weaver, 2005, Should securities markets be transparent? Journal of Financial Markets 8, Moinas, Sophie, 2010, Hidden limit orders and liquidity in limit orders market, working paper University of Toulouse, O Hara, Maureen, Chen Yao, and Mao Ye, 2014, What s not there: Odd lots and market data, Journal of Finance 69, Pardo, Angel, and Roberto Pascual, 2012, On the hidden side of liquidity, The European Journal of Finance 18,

26 17.0% Percent of Volume Traded Inside the Spread 16.0% 15.0% Percent Total Volume 14.0% 13.0% 12.0% 11.0% Small Stocks Medium Stocks Large Stocks 10.0% 9.0% Figure 1: Time series plot of the percent of total volume traded inside the exchange posted spread for small, medium, and large stocks from January 3, 2012 through March 30, Date 26

27 25.0% Intraday Percent of Volume Traded Inside the Spread 20.0% Percent of Volume 15.0% 10.0% Percent Volume 5.0% 0.0% 9:31 9:40 9:49 9:58 10:07 10:16 10:25 10:34 10:43 10:52 11:01 11:10 11:19 11:28 11:37 11:46 11:55 12:04 12:13 12:22 12:31 12:40 12:49 12:58 13:07 13:16 13:25 13:34 13:43 13:52 14:01 14:10 14:19 14:28 14:37 14:46 14:55 15:04 15:13 15:22 15:31 15:40 15:49 15:58 Time of Day Figure 2: Intraday plot of the percent of volume executed inside the spread for the full sample. 27