Canceled Orders and Executed Hidden Orders Abstract:

Transcription

1 Canceled Orders and Executed Hidden Orders Abstract: In this paper, we examine the determinants of canceled orders and the determinants of hidden orders, the effects of canceled orders and hidden orders on market quality and their day-of-the-week patterns. We find that effective spread, market capitalization, stock volatility, order volume, odd-lots and hidden orders are the determinants of canceled orders and that effective spread, odd-lots and stock volatility are the determinants of hidden orders. However, both the determinants of canceled orders and the determinants of hidden orders are not the same across exchanges. More canceled orders and hidden orders in the market will both result in higher effective spread and quoted spread. Their effects on depth depend on how we measure depth. The number of canceled orders is highest on Wednesday and lowest on Monday and Friday. However, we detect no pattern for cancel-to-trade. The number of hidden orders, hidden volume, hidden order rate and hidden volume rate are all lowest on Monday, which contradicts our hypothesis. 1

2 1. Introduction Canceled orders and hidden orders constitute a large proportion of activity in the securities markets and are frequently examined by researchers. Van Ness, Van Ness, and Watson (2015) examine all securities exchanges in the U.S. and find that the average cancellation rate is 72.9% in their sample period from 2006 to Bessembinder, Panayides, and Venkataraman (2009) find that 44% of the Euronext-Paris Stock Exchange sample order volume is hidden orders. Moreover, the proportion of canceled orders and hidden orders in different exchanges vary significantly. The proportion of canceled orders in Island ECN and London Stock Exchange are 40% and 95%, respectively (Hasbrouck and Saar, 2002; Machain and Dufour, 2013). The proportion of hidden orders in Australian Stock Exchange and Frankfurt Stock Exchange are 28% and 16%, respectively (Aitken, Brown, and Walter, 1996; Frey and Sandas, 2009). Why are the proportions of canceled orders and hidden orders in different exchanges so various? What are the determinates of canceled orders and hidden orders? Are the determinates the same in different exchanges? To answer those questions, in this study, we focus on 12 exchanges and investigate the factors that influence canceled orders and executed hidden orders and the effects of canceled orders and executed hidden orders on market quality. 1 We also investigate if the factors that affect cancellation activity and executed hidden orders are the same across the 12 U.S. stock exchanges. This paper has two focuses. The first focus is canceled orders. Canceled orders not only constitute a large percentage of orders, but also catch regulators attention, which make studying canceled orders on the market quality even more important. It is widely believed that excessive canceled orders in the market are detrimental to the market quality; therefore, regulations assess a daily cancellation fee and still try to make amends on the fee structure. On July 1, 2013, the Securities Exchange Act Release (No ) assessed a daily cancellation fee per Account Symbol and security if 1 In the dataset we have, we can only observe executed hidden orders. 2

3 the order cancellation ratio exceeded a designated threshold. On January 27, 2014, CHX proposed to amend this cancellation fee. CHX proposed to provide an exemption from Order Cancellation Fees for a given month if an Account Symbol met an Average Daily Volume requirement for that month. 2 Machain and Dufour (2013), and Van Ness, Van Ness, and Watson (2015) find that canceled orders (or canceling potential liquidity) can hurt market quality. Knowing what drives the canceled orders in each exchange can help regulators make better rules for the market. One paper studying canceled orders is closely related to our paper. Van Ness, Van Ness, and Waston (2015) examine limit order cancellation activity using Dash-5 monthly data from 2006 to Our research differs in three ways. First, we use daily data from the SEC MIDAS database. Daily data can better identify the determinants of cancellation activity than monthly data because monthly data aggregates market activities. We reexamine the determinants of canceled orders and the effects of canceled orders on the market quality using daily stock data. Second, we add to Van Ness, Van Ness, and Waston by determining, for example, if odd-lots and hidden orders impact cancellations. Third, we test if the determinants of order cancellations are the same across different exchanges. The second focus of our paper is executed hidden orders. Whether transparency is good or not for the security market has been a concern of by researchers and regulators. 3 One way to study the transparency issue is through examining hidden orders. Microstructure literature has not reached an agreement regarding whether hidden orders improve market quality. Some research finds that allowing hidden orders in the market improves market quality (Frey and Sandas, 2009; Bessembinder, Panayides and Venkataraman, 2009; Boulatov and George, 2013), but Hendershott and Jones (2005) find that Zhao and Chung (2007) study SEC adopted Rule 605, which requires market centers to make monthly public disclosure of execution quality. They find that the SEC s goal to improve execution quality through more transparent markets has been achieved. However, Comerton-Forde and Tang (2009) examine the effects of the removal of broker identifiers from the central limit order book of the Australian Stock Exchange and find that anonymous trading improves price competition and liquidity especially for large stocks. 3

4 when Island stopped displaying its limit order book, market quality decreased, and when Island redisplayed its orders, market quality improved. Therefore, if hidden orders improve market quality is uncertain. Our study of the determinants of executed hidden orders can help us better understand market transparency issue. Our paper is closely related to two other papers. Bessembinder, Panayides, and Venkataraman (2009) study the determinants of the decision to hide order size using intraday data by focusing on Euronext-Paris stocks. They examine the determinants of order exposure from the perspective of the initiating trader and market participants. They find that order attributes and market conditions are useful in explaining traders exposure decisions, which is also the findings in De Winne and D Hondt (2007). Our study is different from Bessembinder, Panayides, and Venkataraman and De Winne and D Hondt in three ways. First, those two papers both focus on only one stock market. Bessembinder, Panayides, and Venkataraman study 100 Paris stocks from the Euronext Stock Exchange, and De Winne and D Hondt study 40 Canadian stocks from the Euronext Stock Exchange. However, our sample includes all U.S. traded stocks in all U.S. stock exchanges that report hidden order rates (eight out of 12 exchanges). Second, we test if the determinants of hidden orders are the same across different exchanges. Third, with our dataset, we are able to answer some other questions. For example, do people use odd-lots and hidden orders at the same time to hide their information? Do hidden orders hurt market quality? In this paper, we first study the determinants of canceled orders and the determinants of executed hidden orders. We define canceled orders as the orders either fully or partially canceled, and executed hidden orders as the trades which execute against hidden orders. We find that effective spread, market capitalization, stock volatility, order volume, odd-lots and executed hidden orders are among the determinants of canceled orders and that effective spread, odd-lots and stock volatility are determinants of executed hidden orders. However, determinants of canceled orders and determinants 4

5 of executed hidden orders are not the same across exchanges. Next, we examine the effects of canceled orders and executed hidden orders on market quality and find that whether canceled orders and executed hidden orders hurt the market quality depends on the metric of market quality. More specifically, on the canceled order side, more cancellations in the market lead to higher spreads and lower depth when the depth is measured as half penny, or one penny, or one and half penny, or two penny above and below the quote midpoint. However, more cancellations in the market lead to more depth when the depth is measured as at quote midpoint. On the executed hidden order side, a higher proportion of executed hidden orders in the market results in higher spreads and more depth when the depth is measured as at the quote midpoint, or half penny and one, or half penny above and below the quote midpoint. However, a higher proportion of executed hidden orders in the market results in lower depth when the depth is measured as one penny, or two penny above and below the quote midpoint. Finally, we investigate day-of-the-week patterns for cancellations and executed hidden orders. For canceled orders, we find that the number of canceled orders is highest on Wednesday and lowest on Monday and Friday, and that there is no pattern for cancel-to-trade ratio which is defined as the daily number of cancellations divided by the number of trades. For executed hidden orders, we find that the number of executed hidden orders, executed hidden volume, hidden order rate and hidden volume rate are all lowest on Monday. Our findings contribute to the cancellation activity and market transparency literature. We find that two conflict hypotheses can both be correct if we use different exchange samples to test the hypotheses. For example, we find a negative relation between spreads and cancellation activities in the Boston Stock Exchange and the Philadelphia Stock Exchange, which confirms the finding in Liu (2009). We also find a positive relation between spreads and cancellation activities in Archipelago Exchange, BATS Y-Exchange, BATS Z-Exchange, EDGA Exchange, EDGX Exchange and National Association of Securities Dealers Automated Quotations, which is consistent with the finding in Van Ness, Van Ness, 5

6 and Waston (2015). Moreover, we confirm that it is too early to get to the conclusion that cancellation activities and hidden orders are detrimental to the market quality. We find that whether canceled orders and hidden orders are detrimental to market quality or not depends on the metric of market quality. Specifically, we find that more canceled orders and hidden orders in the market will both result in higher effective spread and quoted spread and that the effects of canceled orders and hidden orders on depth depend on how we measure depth. 2. Data and Descriptive Statistics We obtain our data from the SEC website s Market Information Data Analytics System (MIDAS) database, which records all U.S. listed stocks daily activities in twelve exchanges from 2012 to Specifically, the MIDAS dataset includes market capitalization rank, volatility rank, stock price rank, canceled trades, total trades, odd-lots trades, executed hidden orders, order volume, trade volume, executed hidden order volume, effective spread, quoted spread and depth. Since we are concerned with spreads and the MIDAS database has spreads and depth for the calendar year of 2013 only, we limit our sample period to include January to December, There are twelve exchanges in the sample, which are Archipelago Exchange (Arca), American Stock Exchange (Amex), BATS Y-Exchange (Bats-Y), BATS Z-Exchange (Bats-Z), Boston Stock Exchange (Boston), Chicago Stock Exchange (CHX), EDGA Exchange (Edga-A), EDGX Exchange (Edge-X), National Association of Securities Dealers Automated Quotations (Nasdaq), National Stock Exchange (NSX), The New York Stock Exchange (NYSE) and Philadelphia Stock Exchange (PHLX). All exchanges report canceled orders, but not all report hidden orders. We focus on only the exchanges that report hidden orders. We exclude Amex, CHX, NSX and NYSE because they do not report 4 MIDAS database webpage: 6

7 hidden orders. Therefore, we have a total of eight exchanges in the sample. Table 1 lists the exchanges and number of sample observations. [Insert Table 1 here] 2.1 Descriptive Statistics Table 2 lists descriptive statistics for canceled order measures, hidden order measures, market quality measures and other variables used in our analysis. Table 2 shows the daily descriptive statistics for the sample, aggregated from the eight exchanges. Panel A reports the summary statistics for canceled orders. Cancel-to-trade ratio is defined as the daily number of cancellations divided by the number of trades. The mean (median of) cancel-to-trade ratio is (25.15). It is quite striking that the number of canceled orders is times of the number of total trades on average. There is a large difference of cancel-to-trade ratio across stocks. We use the cancel-to-trade ratios in our examination of the determinants of canceled orders in the next section. [Insert Table 2 here] Panel B of table 2 reports summary statistics for hidden orders. Hidden orders are the daily number of trades which execute against hidden orders. Therefore, we do not have the total number of hidden orders in the market. We have only the number of hidden orders which execute in the market. Hidden order rate is defined as hidden orders as a percentage of total trades. Hidden volume rate is hidden volume divided by total volume in the market. On average, 19.09% of trades in the market are executed against hidden orders. Hidden volume rate is lower than hidden order rate, suggesting that hidden orders have lower volume than visible orders. We use both hidden order rate and hidden volume rate in our study of the determinants of hidden orders. 7

8 Panel C of table 2 reports the summary statistics for market quality measurements. We use share volume-weighted effective spread, share volume-weighted relative effective spread, timeweighted quoted spread and time-weighted relative quoted spread and depth to measure market quality. We use both dollars and shares to measure depth. Specifically, we use Depth0.0, Depth0.5, Depth1.0, Depth1.5, and Depth2.0. Depth0.0 is the average quoted dollar or share size at the quote midpoint. DepthX.X is the average quoted depth (dollar or share size) at X.X cents above and below the quote midpoint. For example, depth2.0 is the average quoted dollar size or share size two pennies above and below the quote midpoint. The number of observations drops significantly when depth is at the quote midpoint. The means and medians of depth0.5 and depth1.5 appear higher than that of depth0.0, depth1.0 and depth2.0. Panel D of table2 presents the summary statistics for other variables in the analysis. 3. Determinants of Canceled Orders In this section, we first explain what factors affect traders cancellation activity. Then, we introduce the empirical model that we use in the analysis. Finally, we show the results Hypothesis Development Liu (2009) argues that when a trader places an order, he faces two types of risk, free trading option risk and non-execution risk. Limit order traders can monitor the market to lower these two types of risk. Thus, traders need to balance the cost of monitoring and free trading option risk and nonexecution risk. When spreads are wide, the cost of monitoring is low and fewer orders are canceled. Liu finds weak evidence of negative relation between spreads and canceled orders. Liu also finds that large firms have more canceled orders because greater levels of media and analyst coverage encourages more traders to monitor information flows and be ready to cancel orders. However, Van Ness, Van Ness, 8

9 and Waston (2015) find the opposite. They find a positive relation between effective spreads and canceled orders and a negative relation between firm size and canceled orders. The reason could be that they use monthly data, which aggregates market activities. We use daily data to reexamine the relation between effective spreads and canceled orders and the relation between firm size and canceled orders. Fong and Liu (2010) argue that high volatility leads to high free trading option risk. High free trading option risk leads to a high cost of monitoring, which implies a positive relation between volatility and order cancellation activities. They also argue that when the relative stock volume is low, opportunity costs of monitoring the limit order book is high, which implies a positive relation between order volume and order cancellation activities. O Hara, Yao, and Ye (2014) and Davis, Roseman, Van Ness, and Van Ness (2015) find that the second most frequent odd-lot trade size is trades of one share. However, one share trades have no economic meaning since the cost of one share trades is highest. Davis, Roseman, Van Ness, and Van Ness also find evidence that traders use one-share orders to detect the market liquidity and 95% of all one-share orders are canceled. If odd-lots are used to detect market liquidity, we should expect that these odd-lot orders would be canceled later on. Therefore, we expect the positive relation between odd-lots activities and order cancellation activities. Hasbrouck and Saar (2002) find that 27.7% of all visible orders are fleeting orders. They define fleeting orders as limit orders that are canceled within two seconds. One possible explanation provided by Hasbrouck and Saar is that the trader wants to fish for hidden orders that better the opposing quote. In order to find more liquidity, the trader submits a limit order, and if it is not executed, the trader will immediately cancel it. If the limit order is executed against a hidden order, then it becomes a hidden trade. Therefore, if there are more hidden trades (in our dataset, hidden orders are trades executed against hidden limit orders, so hidden orders are hidden trades), we should observe less order 9

10 cancellation activities. However, if the hidden orders are partially executed, the other part of hidden order will be canceled (citation). D Hondt, De Winne, and Francois-Heude (2001) find that, compared to fully displayed orders, hidden orders are less likely to be totally executed and, therefore, lead to more cancellations. Bessembinder, Panayides, and Venkataraman (2009) also find that when there are more hidden orders in the market, the probability of full execution of all ordrs will be decreased. Therefore, we expect to see that more hidden orders (hidden limits that execute) lead to more canceled orders in the market. We do not have an expectation for the relation between hidden liquidity and canceling liquidity. Prior literature finds that the proportion of canceled orders differs between exchange (Machain and Dufour, 2013; Van Ness, Van Ness, and Waston, 2015). We also find the same phenomenon in our sample. Table 3 and figure 1 show that large differences of cancel-to-trade ratio across exchanges. PHLX has the highest cancel-to-trade ratio, which indicates PHLX has times canceled orders to total trades. Edge-X has the least cancel-to-trade ratio, which indicates Edge-X has times canceled orders to total trades. Therefore, whether the determinants of canceled orders in different exchanges are the same is questionable. We first examine the determinants of canceled orders in the full sample, which means that we aggregate each stock s daily activities from eight exchanges. Then, we examine the determinants of canceled orders in the eight exchange subsamples and investigate whether the results hold in different exchanges. [Insert Table 3 and Figure 1 here] 3.1. Empirical Method In multivariate analysis, we use two-stage least square (2SLS) model. Our model of the determinants of canceled orders is defined in equation (2): 10

11 cancel_to_trade i,t = b 1 + b 2 effective_spread i,t + b 3 market_capitalization i,t + b 4 volatility i,t + b 5 volume i,t + b 6 odd_lots i,t + b 7 hidden_order_rate i,t + instrument i,t + u (1) Effective spread is a determinant of canceled orders and also one of the market quality measurements. We study the effect of canceled orders on the market quality in section 5. Therefore, effective spread may be endogenous. We use 2SLS to deal with the endogeneity problem. We use twice-lagged relative effective spread for stock i at time t as the instrument variable. Successful instrumental variable candidates must satisfy two criteria. The first criterion is that instruments correlate with effective spread. The second criterion is that instruments do not correlate with error term in equation (1). Hansen s J-test of overidentifying restrictions is not significant at all, indicating that the instruments are uncorrelated with the error term in equation (1), and the instrument is valid. Dependent variable in the model is cancel-to-trade ratio, defined as the number of all canceled orders, either full or partial, over total trades for i at time t. Effective spread is share volume-weighted relative effective spread for stock i at time t. Market capitalization is decile ranked market capitalization for stock i at time t. Volatility is decile ranked stock volatility for stock i at time t. Volume is the total order volume for stock i at time t. Odd lot is odd-lots trades over total trades for stock i at time t. Hidden order rate is defined as the number of trades against hidden orders over the number of trades for stock i at time t. We also include month indicator variables (not shown) to control for a time trend Results [Insert Table 4 here] We first report canceled order rates by market capitalization quintiles and volatility quintiles in Table 4. In panel A, we present canceled order rates by market capitalization quintile. By looking at the median, we see that orders of small stock (in quintile 1) are canceled more frequently compared to that 11

12 of large stocks (in quintile 5), a result consistent with findings in Van Ness, Van Ness and Waston (2015) but inconsistent with the theoretical predictions of Liu (2009). In panel B, we present canceled order rates by stock volatility quintile. By looking at the median, we see that stocks in the smallest quintile (quintile 1) are canceled more frequently than stocks in the largest quintile (quintile 5), a result inconsistent with the findings in Fong and Liu (2010). [Insert Table 5 here] Next, we conduct regression to investigate the determinants of canceled orders. The 2SLS estimated coefficients from equation (2) are presented in Table 5. In the aggregated sample, we aggregate the data from eight exchanges for each stock by day. We use boldface for the numbers that are not significant above 5% level. We find that effective spread is positively significant related to cancel-to-trade ratio in the aggregated sample, Arca, Bats-Y, Bats-Z, Edge-A, Edge-X and Nasdaq, which is consistent with the findings in Van Ness, Van Ness and Waston (2015) but inconsistent with the theoretical predictions of Liu (2009). However, we find the positive relation between effective spread and cancel-to-trade ratio in two exchanges, Boston and PHLX. The results suggest that the relation between spread and cancel-to-trade ratio is not the same across exchanges. Market capitalization is negatively significant related to cancel-to-trade ratio in the aggregated sample and all exchanges, except Edge-X, which confirms the findings in Van Ness, Van Ness and Waston (2015) but is inconsistent with the theoretical predictions of Liu (2009). We find a negative relation between volatility and cancel-to-trade ratio in the aggregated sample, Arca, Bats-Y, Boston, Edge-A, Nasdaq and PHLX exchanges. The result is inconsistent with the findings in Fong and Liu (2010). Fong and Liu (2010) only study the Australian Securities Exchange. The differences between our findings could be because we use different exchanges. Different exchanges have different market structures, which could lead to different relations between volatility and cancel- 12

13 to-trade ratio. We also find no significant relation between volatility and cancel-to-trade ratio in Bats-Z and Edge-X. This is an interesting finding since Bats-Z and Edge-X all have rebates to the traders who provide the liquidity to the market. When traders have maker rebates, traders will become less sensitive to the costs of monitoring. Therefore, we see there is no relation between volatility and cancel-to-trade ratio. Fong and Liu (2010) find a positive relation between volumes and cancel-to-trade ratio. We find a positive relation between volumes and cancel-to-trade ratio in Arca, Bats-Y, Bats-Z, Boston, Nasdaq and PHLX exchanges confirming the findings in Fong and Liu (2010). However, in the aggregated sample, Edge-A and Edge-X exchanges, we find no relation between volumes and cancel-to-trade ratio. We find a significantly positive relation between odd-lots rates and cancel-to-trade ratio in the aggregated sample and all exchanges except the Boston exchange, supporting our hypothesis that oddlots trades are used to detect the market liquidity, therefore we see the more odd-lots trades in the market the more order canceled. We also find that hidden order is significantly positive related to cancel-to-trade ratio in Arca, Bats-Y, Bats-Z, Boston, Edge-A, Nasdaq and PHLX exchanges. D Hondt, De Winne and Francois-Heude (2001) also find a positive relation between hidden orders and canceled orders. The positive relation between hidden orders and cancel-to-trade ratio could be because hidden orders are more likely to be partially executed. The non-executed hidden orders will be canceled soon afterwards. Overall, we document mixed results of the determinants of canceled orders. In some exchanges, our findings are consistent with prior literature s theoretical predictions and empirical findings. In other exchanges, we find the opposite of the prior literature predictions. However, we do confirm that the determinants of canceled orders are not the same across exchanges. 13

14 4. Determinants of Hidden Orders Hidden orders are the orders not displayed when traders submit them, which means that other traders cannot see those orders. In our dataset, we only have hidden orders against trades, which means that only when hidden orders execute and become trades are they recorded in the MISDAS database as hidden orders. In this section, we first explain what factor affects hidden orders. Then, we introduce the empirical model we use in the analysis. Finally, we show the results. 4.1 Hypothesis Development As suggested by Harris (1996), a trader exposing order faces the risk that other traders will use the information in the exposed order to his disadvantage. The other traders can either front-run his order or they will refuse to supply liquidity to his side. However, the benefit of exposing an order is to attract other traders to obtain more liquidity. The decision of submitting a hidden order or a display order depends on the risks and the benefits of exposing an order. When an order contains more information that will easily be detected by other traders, the trader will choose to hide the order to protect his information. Therefore, when more information is in the market, there will be more hidden order executions. Chakrabarty and Shaw (2008) find that the number of hidden orders executed increases when more information is in the market (around earnings announcements). Bloomfield, O Hara, and Saar (2011) use a laboratory market and find that informed traders submit more hidden orders than liquidity traders when the market allows them to submit hidden orders. O Hara, Yao and Ye (2014) find that 35% to 39% of price discovery is coming from odd-lots, consistent with informed traders attempting to hide trade from the market. Johnson, Van Ness and Van Ness (2015) also find that odd-lots contribute to price discovery. Therefore, traders use odd-lots to hide their information. If traders use hidden orders and odd-lots to hide their information at the same time, we would see a positive relation between hidden orders and odd-lots. If traders access hidden orders 14

15 and odd-lots as a substitute way to hide information, we would see a negative relation between hidden orders and odd-lots, which means that traders using hidden orders to hide their information would decrease the usage of odd-lots to hide information. Even if an order does not contain any information, the order still has option value. According to Harris (1996), when volatility is high order exposure is smaller because option values increase with volatility. Therefore, when volatility is high in the market, we would observe more hidden orders. Aitken, Berkman and Mak (2001) and Fleming and Mizrach (2009) also find a positive relation between hidden orders and volatility. We expect a positive relation between volatility and hidden orders. Bessembinder, Panayides and Venkataraman (2009) argue a positive relation between the bidask spread and hidden orders. The reason is that a large spread makes front-running strategies more profitable. Therefore, traders will hide more volume when the spread is large since the probability of being front-run increases. However, De Winne and D Hondt (2007) find a negative relation between the bid-ask spread and hidden orders. The opposite findings could be because two papers examine the relation in two different exchanges. We reexamine the relation between bid-ask spread and hidden orders in all U.S. exchanges. Prior literature finds that the proportion of hidden orders is different exchange (Aitken, Brown and Walter, 1996; Frey and Sandas, 2009; Bessembinder, Panayides and Venkataraman, 2009). We also find the same phenomenon in our sample. Table 3 and Figure 1 show that hidden order rate and hidden order volume across different exchanges are different. Boston stock exchange has the most hidden order rate and hidden volume rate, and PHLX has the least. The difference between Boston and PHLX in the average hidden order rate is 21.37%. Therefore, whether the determinants of hidden orders in different exchanges are the same is questionable. We first examine the determinants of hidden orders in the full sample, which means that we aggregate the stock activities from eight exchanges. Then, we 15

16 examine the determinants of hidden order in eight different exchanges and investigate whether the results hold in different exchanges Empirical Method In multivariate analysis, we use two-stage least square (2SLS) model. Our model of the determinants of hidden orders is defined in equation (1): hidden_rate it = b 1 + b 2 oddlots it + b 3 volatility it + b 4 effective_spread it + instrument it + u (2) Effective spread is a determinant of hidden order and also a market quality measurement. We study the effect of hidden orders to the market quality in section 5. Therefore, effective spread may be endogenous. We use 2SLS to deal with the endogeneity problem. We use twice-lagged effective spread for stock i at time t as the instrument variable. Hansen s J-test of overidentifying restrictions is not significant at all, indicating that the instruments are uncorrelated with the error term in equation (2), and the instrument is valid. We use both hidden order rate and hidden volume rate as our dependent variable. Hidden order rate is defined as the number of trades against hidden orders over total trades. Hidden volume rate is defined as the sum of trade volume for trades against hidden orders over total trades volume. Odd-lots is odd-lots trades over total trades for stock i at time t. Volatility is decile ranked stock volatility for stock i at time t. Effective spread is share volume-weighted relative effective spread for stock i at time t. We also include month indicator variables (not shown) to control for a time trend Results [Insert Table 6 here] 16

17 The 2SLS estimated coefficients from equation (1) are presented in Table 6. In the full sample, we aggregate the data from eight exchanges for each stock by day. We use boldface for the numbers that are not significant above 5% level. We find a positive relation between effective spread and hidden rate, which is consistent with the findings in Bessembinder, Panayides and Venkataraman (2009). The positive relation between effective spread and hidden rate is consistent with the argument that traders will hide more volume when the spread is large since the probability of being front-run increases. However, when we run the regression model for each exchange, we find a positive relation between effective spread and hidden rate only in Edge-X and NASDAQ. We find a negative relation between effective spread and hidden rate in Arca, Bats-Y, Bats-Z, Edge-A, Edge-X and PHLX, which is consistent with the findings in De Winne and D Hondt (2007). In the Boston sample, we find that effective spread is not significantly related to hidden order rate and is negatively significant related to hidden volume rate. Therefore, we find a mixed result for the relation between effective spread and hidden rate, but we do confirmed that the determinants of hidden orders are not the same across exchanges. In Table 6, we find a positive relation between odd-lots rate and hidden rate in the full sample and all subsamples, which is consistent with the hypothesis that informed traders use odd-lots and hidden orders to hide their information at the same time. We also find a positive relation between volatility and hidden rate across all exchanges, which is consistent with Harris s (1996) argument that when volatility is high, order exposure is smaller because option values increase with volatility. The result is also consistent with the findings in Aitken, Berkman and Mak (2001) and Fleming and Mizrach (2009). 5. Market Quality 5.1. Canceled Order Effect on Market Quality 17

18 Regulators, traders and exchange executives all care about market quality. Academic research also frequently focuses on market quality. Cancellation activity detrimental to market quality is already found by Van Ness, Van Ness and Watson (2015). They use monthly data from the SEC s Dash-5 to examine the effect of cancellation on market quality. However, monthly data aggregate market activities. Here, we use daily data to reexamine the effect of cancellation to market quality. We conduct a two-stage least square model to examine the effect of canceled orders to market quality. It is necessary to consider other factors that may influence spreads such as price, volatility and activity suggested by McInish and Wood (1992). Our model is defined in equation (3): Market_quality i,t = b 1 + b 2 cancel_to_trade i,t + b 3 volatility i,t + b 4 inverseprice i,t + b 5 number_of_transcations i,t + instrument i,t + u (3) Since market quality is a determinant of canceled orders, cancel-to-trade may be endogenous. We use 2SLS to deal with the endogeneity problem. We use twice-lagged cancel-to-trade for stock i at time t as the instrument variable. Hansen s J-test of overidentifying restrictions is not significant at all, indicating that the instruments are uncorrelated with the error term in equation (3), and the instrument is valid. [Insert Table 7 here] We follow prior literature to use spread and depth to measure market quality (Zhao and Chung, 2007; Boehmer, Jones and Zhang, 2013). For spread, we use share volume-weighted effective spread, share volume-weighted relative effective spread, time-weighted quoted spread and time-weighted relative quoted spread. For depth, we use depth at bid-ask quote midpoint denoted as depth0.0, and depth of each half-penny increment from bid-ask quote midpoint. Independent variables are cancel-to trade, stock volatility, inversed stock price and the number of transactions. Cancel-to-trade is the 18

19 number of canceled orders to the number of trades. We also include month indicator variables (not shown) to control for a time trend. The results are presented in Table 7. We use boldface for the numbers that are not significant above 5% level. Panel A shows the effect of cancel-to-trade on spreads. We find that more cancel-totrade leads to higher effective spread, relative effective spread and relative quoted spread, indicating cancellation activity is detrimental to market quality. We find the same result as in Van Ness, Van Ness and Watson (2015). Panel B presents the effect of cancel-to-trade on depth in dollars, and Panel C presents the effect of cancel-to-trade on depth in shares. We do find that when cancellation activity is high, depth, measured by dollar and share, is less at half penny, one penny, one and half penny and two penny above and below the quote midpoint. The finding that the more the cancellation activity the less the depth is consistent with the findings in Van Ness, Van Ness and Watson (2015). However, we find that when cancellation activity is high, depth, measured by dollar and depth, is higher at the quote midpoint. This could be because traders cancel unattractive quotes and submit more attractive quotes. Therefore, we observe more cancellations lead to more depth at the quote midpoint. In summary, the effect of cancellation activity to market quality is mixed. On one hand, we find that more cancellations lead to higher spreads and lower depth of half penny, one penny, one and half penny and two penny above and below the quote midpoint. On the other hand, we find that more cancellations lead to more depth at quote midpoint Hidden Order Effect on Market Quality Prior literature finds that allowing hidden orders in the market improves market quality (Frey and Sandas, 2009; Bessembinder, Panayides and Venkataraman, 2009; Boulatov and George, 2013). However, Hendershott and Jones (2005) find that when Island stopped displaying its limit order book, 19

20 market quality decreased, and when Island redisplayed its orders, market quality improved. Therefore, whether hidden orders improve market quality is uncertain. In this section, we examine the effect of hidden orders on the market quality. We conduct a two-stage least square model to examine the effect of hidden orders on market quality. It is necessary to consider other factors that may influence spreads such as price, volatility and activities suggested by McInish and Wood (1992). Our model is defined in equation (4): Market_quality i,t = b 1 + b 2 hidden_order_rate i,t + b 3 volatility i,t + b 4 inverseprice i,t + b 5 number_of_transcation i,t + instrument i,t + u (4) Since market quality is a determinant of hidden orders, cancel-to-trade may be endogenous. We use 2SLS to deal with the endogeneity problem. We use twice-lagged hidden order rate for stock i at time t as the instrument variable. Hansen s J-test of overidentifying restrictions is not significant at all, indicating that the instruments are uncorrelated with the error term in equation (4), and the instrument is valid. [Insert Table 8 here] We follow prior literature to use spread and depth to measure market quality (Zhao and Chung, 2007; Boehmer, Jones and Zhang, 2013). For spread, we use share volume-weighted effective spread, share volume-weighted relative effective spread, time-weighted quoted spread and time-weighted relative quoted spread. For depth, we use depth at bid-ask quote midpoint denoted as depth0.0, and depth of each half-penny increment from bid-ask quote midpoint. Independent variables are cancel-totrade rate, stock volatility, inversed stock price and the number of transactions. Hidden order rate is defined as the number of hidden orders to the number of trades. We also include month indicator variables (not shown) to control for a time trend. 20

21 The results are presented in Table 8. We use boldface for the numbers that are not significant above 5% level. Panel A presents the effect of hidden orders on spreads. We find that when hidden order rate is high, effective spread, relative effective spread and relative quoted spread are higher, suggesting hidden orders hurt market quality. Panel B shows the effect of hidden orders on depth in dollars, and Panel C presents the effect of hidden orders on depth in shares. We find that the higher the percentage of hidden orders results in more depth, measured by dollar and share, at the quote midpoint (depth0.0), half penny and one and half penny above and below the quote midpoint (depth0.5 and depth1.5), suggesting that hidden order is beneficial to the market. However, we also find that the higher the percentage of hidden orders results in less depth of one penny and two penny above and below the quote midpoint when depth measured by both dollar and share (depth1.0 and depth2.0). In summary, higher proportion of hidden orders in the market result in higher spread and more depth at the quote midpoint, half penny and one and half penny above and below the quote midpoint. However, higher proportion of hidden orders leads to less depth of one penny and two penny above and below the quote midpoint. 6. Day-of-the-week Patterns 6.1. Day-of-the-week Patterns of Canceled Orders Day-of-the-week patterns of many stock characteristics are documented in prior literature (Lakonishok and Levi, 1982; Kiymaz and Berument, 2001; Kiymaz and Berument, 2003). Kiymaz and Berument (2003) investigate day-of-the-week patterns on volume, volatility and return of major stock market indexes. Johnson (2014) examines day-of-the-week patterns of odd-lots proportions and finds that odd-lots proportions are highest on Wednesdays and Friday. A volatility day-of-the-week effect is documented on many exchanges in the world (Balaban, Bayar and Kan, 2001; Alagidede, 2008; 21

22 Kenourgios and Samitas, 2008). Kiymaz and Berument (2003) document a volatility day-of-the-week pattern in the U.S. stock market and find that volatility is highest on Wednesdays and Fridays. In the previous section, we show that volatility is a determinant to cancel-to-trade rate. In the aggregated sample, we find a negative relation between volatility and cancel-to-trade rate. Therefore, we hypothesize that the cancel-to-trade rate will follow a day-of-the-week pattern, with cancellations being the lowest on Wednesdays and Fridays. [Insert Table 9 here] We conduct an ordinary least square regression to show day-of-the-week pattern of cancellations. The dependent variable is the number of canceled orders or cancel-to-trade rate. The independent variables are Monday, Tuesday, Thursday and Friday. Therefore, the regression coefficients all compare to the cancellations on Wednesday. Table 9 presents the results. We use boldface for the numbers that are not significant above 5% level. We first examine the day-of-the-week pattern of the number of canceled orders. We find a humped shape of the number of canceled orders. The number of canceled orders is highest on Wednesday and lowest on Monday and Friday. However, Kiymaz and Berument (2003) find that on the U.S. stock market, volume is lowest on Monday and Friday. Day-ofthe-week pattern of the number of canceled orders we find could be fully explained by the pattern of volume. Therefore, we further examine the day-of-the-week pattern of cancel-to-trade, which is the total canceled orders over total trades. No independent variable is significant. We do not find a pattern for cancel-to-trade Day-of-the-week Patterns of Hidden Orders Prior literature documents that traders use hidden orders to hide their information (Harris, 1996; Chakrabarty and Shaw, 2008; Bloomfield, O Hara, and Saar, 2011). Monday usually contains more information since people usually cannot trade on the weekend. Recently, Jiang, Likitapiwat and Mclnish 22

23 (2012) find that firms prefer to announce earnings after trading hours. Thus, information on the weekend could be incorporated into trading on Monday. Therefore, if Monday contains more information, we should expect more hidden orders in the market. [Insert Table 10 here] We use an ordinary least square regression to show day-of-the-week pattern of hidden orders. The dependent variable is the number of hidden orders, or hidden volume, or hidden order rate, or hidden volume rate. The independent variables are Tuesday, Wednesday, Thursday and Friday. Therefore, the regression coefficients all compare to the hidden orders on Monday. Table 10 presents the results. We use boldface for the numbers that are not significant above 5% level. We find that the number of hidden orders, hidden volume, hidden order rate and hidden volume rate are all lowest on Monday, which contradicts our hypothesis. The explanation could be that hidden orders are used primarily by uninformed traders, which is found by Frey and Sands (2009) and Bessembinder, Panayides and Venkataraman (2009). 7. Conclusion In this paper, we focus on canceled orders and hidden orders, which are both important to the securities markets since they are a non-trivial proportion of total orders and potentially effect market qualities. First, we examine the determinants of canceled orders and determinants of hidden orders. We find that effective spread, market capitalization, stock volatility, volume, odd-lots rate and hidden orders are the determinants of canceled orders, and that effective spread, odd-lots rate and stock volatility are the determinants of hidden orders. However, when we use the stock market aggregated sample and exchange sub samples to examine the determinants of canceled orders and the determinants of hidden orders, we get different results, suggesting that different market characteristics 23

24 or market structures can lead to the determinants of canceled orders and the determinants of hidden orders being different. Second, we investigate the effects of canceled orders and hidden orders on market quality. More cancellations on the market lead to higher spreads and lower depth of half penny, one penny, one and half penny and two penny above and below the quote midpoint but more depth at quote midpoint. Higher proportion of hidden orders in the market results in higher spread and more depth at the quote midpoint, half penny and one and half penny above and below the quote midpoint but lower depth of one penny and two penny above and below the quote midpoint. Therefore, whether canceled orders and hidden orders hurt market quality depends on how you want to measure the market quality. Finally, we examine day-of-the-week patterns of canceled orders and hidden orders. The number of canceled orders is highest on Wednesday and lowest on Monday and Friday. However, this pattern could be fully explained by the pattern of volume. Therefore, we further examine the day-ofthe-week pattern of cancel-to-trade, but we did not find a pattern for cancel-to-trade. For hidden orders, we find that the number of hidden orders, hidden volume, hidden order rate and hidden volume rate are all lowest on Monday, which contradicts our hypothesis. 24

25 References Alagidede, P., Day of the week seasonality in African stock markets, Applied Financial Economic Letters 4, Aitken, M.J., Brown, P., Walter, T., 1996, Infrequent trading and firm size as explanations for the intraday patterns in returns on SEATS, Working Paper. Balaban, E., A. Bayar, and O. Kan, Stock returns, seasonality and asymmetric conditional volatility in world equity markets, Applied Economics Letters 8, Bessembinder, H., Panayides, M., Venkataraman K., 2009, Hidden liquidity: an analysis of order exposure strategies in electronic stock markets, Journal of Financial Economics, 94, Berument, H., Kiymaz, H., 2001, The day of the week effect on stock market volatility, Journal of Economics and Finance, 25, Brunsden, J., 2012, Traders May Face Nordic-Style EU Fees for Canceled Orders, retrieved October 2, 2013, from Boehmer, E., Jones, M., C., Zhang X., 2013, Shackling Short Sellers: The 2008 Shorting Ban, The Review of Financial Studies, Boulatov, A., George. T., 2013, Hidden and displayed liquidity in securities markets with informed liquidity providers, Review Financial Studies. Boni, L., Brown, D., Leach, J., 2013, Dark pool exclusivity matters, Working Paper. Chakrabarty, B., Shaw, K., 2008, Hidden liquidity: order exposure strategies around earnings announcements, Journal of Business Finance & Accounting, 35(9) & (10), De Winne, R., D Hondt, C., Hide-and-seek in the market: placing and detecting hidden orders, Review of Finance, 11, D Hondt, C., De Winne, R., Francois-Heude, A., 2001, Hidden orders: an empirical study on the French segment of Euro,NM, working paper. Fleming, M.j., Mizrach, B., 2009, The Microstructure of a U.S. Treasury ECN: The Brokertec Platform, Working Paper. Fong, K., Liu, W., 2010, Limit order revisions, Journal of Banking & Finance, 34, Frey, S., Sands, P., 2009, The impact of iceberg orders in limit order books, Working Paper. Harris, L., 1985, A transaction data study of weekly and intradily patterns in stock returns, Journal of Financial Economics, Harris, L., 1996, Does a minimum price variation encourage order exposure? Working Paper. 25

26 Hasbrouck, J., Saar, G., 2002, Limit orders and volatility in a hybrid market: the island ECN, Working Paper. Hasbrouck, J., Saar, G., 2009, Technology and liquidity provision: The blurring of traditional definitions, Journal of financial markets, 12, Hendershott, T., Jones, C., 2005, Island goes dark: Transparency, fragmentation, and regulation, Review of Financial Studies, 18, Johnson, H., 2014, Odd Lot Trades: The Behavior, Characteristics, and Information Content, Over Time, The Financial Review, 49, Johnson, Van Ness and Van Ness (2015) Are All Odd-Lots the Same? Odd-Lot Transactions by Order Submission and Trader Type, Working Paper. Jones, C., 2013, What do we know about high-frequency trading?, Working Paper. Kiymaz, H. and H. Berument, The day-of-the-week effect on stock market volatility and volume: International evidence, Review of Financial Economics 12, Kenourgios, D., Samitas, A., 2008, The day of the week effect patterns on stock market return and volatility: Evidence for the Athens stock exchange, International Research Journal of Finance and Economics, 15, Lakonishok, J., Levi, M., 1982, Weekend effects on stock returns: A note, Journal of Finance, 37, Liu, W., 2009, Monitoring and Limit Order Submission Risks, Journal of Financial Markets, 12, O'Hara, M., Yao, C., Ye, M., 2014, What's not there: odd lots and market data. The Journal of Finance, 69, Machain, M., Dufour, A., 2013, Liquidity supply and the intensity of limit order cancellations, Working Paper. McInish, H., T., Wood, A., R., 1992, An analysis of intraday patterns in bid/ask spreads for NYSE stocks, The Journal of Finance, 47, Tuttle, L., 2003, Hidden Orders, Trading Costs and Information, working paper Van Ness, B., Van Ness R., Watson, E., 2015, Canceling liquidity, Journal of Financial Research, Zhao., X., Chung, H., K., 2007, Information disclosure and market quality: the effect of SEC rule 605 on trading cost, Journal of Financial and Quantitative Analysis, 42,