SEARCH

SEARCH BY CITATION

Abstract

  1. Top of page
  2. Abstract
  3. 1. INTRODUCTION
  4. 2. DATA AND METHOD
  5. 3. RESULTS
  6. 4. FURTHER ANALYSIS AND ROBUSTNESS TESTS
  7. 5. CONCLUSION AND SUGGESTIONS FOR FUTURE RESEARCH
  8. APPENDIX A
  9. APPENDIX B
  10. REFERENCES

This study examines the impact of allowing traders to co-locate their servers near exchange servers on the liquidity of futures contracts traded on the Australian Securities Exchange. It provides evidence of an increase in proxies for high-frequency trading activity following the introduction of co-location. There is strong evidence of a decrease in bid–ask spreads and an increase in market depth after the introduction of co-location. We conclude that the introduction of co-location enhances liquidity. We conjecture that co-location improves the efficiency with which liquidity providers (including market maker high-frequency traders) are able to make markets. © 2013 Wiley Periodicals, Inc. Jrl Fut Mark 34:20–33, 2014

1. INTRODUCTION

  1. Top of page
  2. Abstract
  3. 1. INTRODUCTION
  4. 2. DATA AND METHOD
  5. 3. RESULTS
  6. 4. FURTHER ANALYSIS AND ROBUSTNESS TESTS
  7. 5. CONCLUSION AND SUGGESTIONS FOR FUTURE RESEARCH
  8. APPENDIX A
  9. APPENDIX B
  10. REFERENCES

A number of studies document that algorithmic trading, in general, and high-frequency trading (HFT), in particular, are positively related to liquidity (e.g., Brogaard, 2010; Hasbrouck & Saar, 2012, December; Hendershott, Jones, & Menkveld, 2011; Hendershott & Riordan, 2009, August). Not surprisingly, exchanges have modified their market structure to attract more algorithmic trading. As Hendershott et al. (2011, p. 31) point out, one way of doing so is by permitting “algorithmic traders to co-locate their servers in the market's data center.” Co-location reduces latency (the time it takes to make and execute trading decisions) and therefore enables co-located HFT market makers to more rapidly adjust their quotes as market conditions change. It also levels the playing field among competing HFT market makers who are also co-located by ensuring that none has a latency advantage over the other from an exchange perspective. Consequently, the introduction of co-location facilities is expected to increase liquidity by encouraging HFT and increase competition among HFT market makers.

Other things equal, the introduction of co-location by an exchange should lead to greater trading volume by algorithmic traders. However, it is not immediately apparent how, if at all, the increased trading volume from algorithmic traders will affect liquidity and other dimensions of market quality. As Hendershott et al. (2011, p. 3) identify:

But it is not at all obvious a priori that AT [algorithmic trading] and liquidity should be positively related. If algorithms are cheaper and/or better at supplying liquidity, then AT may result in more competition in liquidity provision, thereby lowering the cost of immediacy. However, the effects could go the other way if algorithms are used mainly to demand liquidity … In fact, AT could actually lead to an unproductive arms race, where liquidity suppliers and liquidity demanders both invest in better algorithms to try to take advantage of the other side, with measured liquidity the unintended victim.

The decision by the Australian Securities Exchange (ASX) to allow futures traders to co-locate their servers starting February 20, 2012 provides a natural experiment by which to assess how much liquidity is affected by the introduction of co-location, as well as the impact of algorithmic trading on liquidity.1 In this study, we test the proposition that HFT increased following the introduction of co-location on the ASX by examining various proxies of the amount of HFT activity. We also directly test whether the introduction of co-location is associated with an improvement in liquidity.

This study also extends previous research by examining futures markets where the mix of traders are different to equities markets studied in previous research examining HFT. Futures markets include a substantial portion of arbitrageurs attempting to exploit price discrepancies between near and deferred contracts as well as futures and spot markets. We examine data for the four most actively traded futures contracts on the ASX before and after the introduction of co-location, to determine the effects on liquidity of the decision to allow co-location near the exchange server. Specifically, we examine the: Share Price Index (SPI), 90-day Bank Accepted Bills (BABs), 3-year Government Bond, and 10-year Government Bond futures contracts.

We document that trading volume, the order-to-trade ratio, message traffic, and depth increased for the three interest rate futures contracts following co-location. In contrast, trading volume, the order-to-trade ratio and message traffic fell while depth increased for the SPI contract following co-location. We argue that the puzzling behavior of the SPI is largely explained by the introduction of a cost recovery charge (i.e., a tax) on cash market equity message traffic that raised the cost and lowered the benefits of stock index arbitrage. Consistent with Chaboud, Hjalmarsson, Vega, and Chiquoine (2011), we do not observe any change in volatility after the introduction of co-location. This is contrary to Boehmer, Fong, and Wu (2012), who report evidence of increased short run volatility after the introduction of co-location and Hasbrouck and Saar (2012, December) and Hagströmer and Nordén (2013) who report a decrease in volatility when algorithmic trading increases. We also quantify the economic significance of the increase in market liquidity after co-location.

2. DATA AND METHOD

  1. Top of page
  2. Abstract
  3. 1. INTRODUCTION
  4. 2. DATA AND METHOD
  5. 3. RESULTS
  6. 4. FURTHER ANALYSIS AND ROBUSTNESS TESTS
  7. 5. CONCLUSION AND SUGGESTIONS FOR FUTURE RESEARCH
  8. APPENDIX A
  9. APPENDIX B
  10. REFERENCES

Intraday trade and quote data used in this study is sourced from Thomson Reuters Tick History (TRTH) and is managed and distributed by the Securities Industry Research Centre of Asia Pacific. Time stamped to the nearest microsecond, the data include price and volume details for each trade, and order book information for up to 10 price levels. The data includes qualifiers that identify open bid and ask quotes, and daily high and low prices. Our analysis considers the four most actively traded futures contracts traded on the ASX derivative market, including: 3- and 10-year Government Bonds, 90-day BABs, and the SPI. The sample period extends August 20, 2011–August 20, 2012, coinciding with a 12-month event window centered on the introduction of co-location facilities. The sample is restricted to trades in the nearest to maturity contract, during daytime trading hours.2

To gauge the change in the extent of algorithmic trading in futures markets we adopt several proxies consistent with Hendershott et al. (2011) as our data does not permit explicit identification of algorithmic traders. We consider three related measures: (1) Message Traffic, (2) Order-to-Trade Ratio, and (3) Algo Trade. Boehmer et al. (2012) and Hendershott et al. (2011) highlight message traffic can include new order submissions, modifications, or order cancellations. Order book or market depth data sourced from TRTH summarizes such information. Consequently, message traffic calculated for each day is the accumulation of changes in the order book for each microsecond of the trading day.

Order-to-Trade Ratio for contract i on day t is calculated as:

  • display math

Hendershott et al. (2011) suggest normalizing message traffic by trading volume, consistent with their approach we examine, Algo Trade calculated as follows for contract i on day t:

  • display math

To test the impact of the introduction of co-location on ASX Trade24,3 we estimate for each futures contract:

  • display math

where Coloit is a dummy variable set to zero prior to the introduction of co-location facilities on February 20, 2012 and set to one following its introduction. Yit includes our proxies of the degree of algorithmic trading or a series of market quality and liquidity metrics, such as: bid–ask spreads, depth, open interest, trading activity, and volatility.

Bid–ask spreads for each trade executed on ASX Trade24 are calculated from prevailing best bid and ask quotes. For each trading day we calculate the average bid–ask spread in ticks and percentage bid–ask spread, which standardizes the dollar tick spread by the quote midpoint. We introduce a third measure related to bid–ask spreads, Percentage Trades at Minimum Tick, providing a frequency count of trading throughout the day at the minimum tick. The minimum tick for the SPI (one tick), BABs (0.01%), and 10-year Government Bonds (0.005%) did not change during the sample period considered. In relation to 3-year Government Bonds, the ASX mandates a minimum tick of 0.01% (equivalent to approximately $24), except for trading days between the 8th of the expiry month and expiry date. During this period the minimum tick for 3-year Government Bonds is reduced to 0.005%. The change in the minimum price movement for 3-year Government Bonds applies to all contract months, irrespective of expiry. To avoid any confounding effects of this regime change we exclude trading days in the 3-year Government Bond futures from the 8th of the expiry month to settlement.

Our variable Depth, aggregates over the trading day the average available total depth at the best bid and ask quotes. Trade Size, measures the average daily trade parcel, while Transactions counts the number of trades. Volume measures the total size of trade for trading day t, Open Interest measures the open interest at the start of the trading day, while Volatility indicates contract i's price volatility on day t, proxied by the daily price movement (see Parkinson, 1980)

  • display math

The first objective of our examination is to identify the relation between algorithmic trading and co-location. Ex ante, it is expected the introduction of co-location should increase the quantity of algorithmic trading. However, the implementation of co-location coincides with the introduction of a cost recovery charge by the Australian Securities and Investments Commission (ASIC) to charge participants a message traffic and trading fee to recover expanded market supervision costs.4 The cost recovery charge intended for equities and derivative markets applied solely to equity products in 2012, however its introduction may indirectly impact equity based futures contracts. Arbitrage in the SPI futures contract involves trading in the underlying equity market. If arbitrage opportunities diminish as a result of the charge, this may lead to a reduction in message traffic for equity based futures contracts vis-à-vis interest rate futures contacts. To test this proposition we estimate the following regression for each futures contract i:

  • display math

where ATit, is a proxy for algorithmic trading: Message Traffic, Order-to-Trade Ratio, or Algo Trade. Coloit is a dummy variable set to zero prior to the introduction of co-location facilities on February 20, 2012 and set to one following its introduction. CRCit is a dummy variable set to zero prior to the introduction of the cost recovery charge on January 1, 2012 and set to one following its introduction.

In this study, our principal objective is to test the impact of co-location on market liquidity. To evaluate and measure the effect of co-location we estimate the following regression for our four measures of market liquidity:

  • display math

where Liquidityit is proxied via: dollar tick spread, percentage tick spread, percentage of trades at the minimum tick, and total depth at the best prevailing quotes. Consistent with the literature that seeks to explain the determinants of liquidity, we control for price volatility and the level of trading activity. The theory underpinning these variables is well-developed in Demsetz (1968), Stoll (1978), and Copeland and Galai (1983). Price volatility measures the risk to liquidity providers per unit of time. Hence, the higher the price volatility, the greater the compensation sought by liquidity providers. To control for the level of trading activity, in the spirit of Wang, Yau, and Baptiste (1997) we introduce open interest as a control for the level of liquidity. Wang et al. (1997) argue changes in the information collected by derivative traders are reflected in the expected changes in physical positions. Open interest or the total number of unsettled contracts in the futures markets captures such changes. A high degree of open interest would indicate greater trading activity and liquidity in the future. The level of open interest should not be associated with the degree of algorithmic trading as most HFT firms typically do not hold significant open positions.

3. RESULTS

  1. Top of page
  2. Abstract
  3. 1. INTRODUCTION
  4. 2. DATA AND METHOD
  5. 3. RESULTS
  6. 4. FURTHER ANALYSIS AND ROBUSTNESS TESTS
  7. 5. CONCLUSION AND SUGGESTIONS FOR FUTURE RESEARCH
  8. APPENDIX A
  9. APPENDIX B
  10. REFERENCES

We begin by examining the change in the proxies for algorithmic trading for interest rate futures following the introduction of co-location. Table I reports summary statistics for a number of algorithmic trading proxies. Table I reports an increase in raw message traffic for all interest rate futures contracts, which is statistically significant at all conventional levels of significance. For example, for BABs futures raw message traffic increases from 7.65 messages per minute to 10.98 messages per minute. Table I also reports changes in message traffic, which have been scaled by trading interest, including the order-to-trade ratio and Hendershott's proxy. On the basis of these variables, Table I provides some evidence that HFT increased following the introduction of co-location. The algorithmic trading proxy increases significantly for BABs futures following the introduction of co-location, while the order-to-trade ratio increases significantly for 3- and 10-year Government Bond futures.

Table I. Liquidity and Market Quality Measures Centered Around the Introduction of Co-Location
 Tick spread% Tick spread% Trades at minimum tickDepthMessages per minuteOrder-to-trade ratioAlgo tradeTrade sizeTransactionsVolumeOpen interestVolatility
  1. This table reports summary statistics of market quality pre and post the introduction of co-location facilities on the ASX. To test the impact of co-location on market quality measures we estimate the following specification for each futures contract i:

  2. inline image

  3. where Coloit is a dummy variable set to zero prior to the introduction of co-location facilities on February 20, 2012 and one thereafter. Yit is the relevant dependent variable: bid–ask spreads, depth, messages per minute. The analysis is based on daily observations and the event window extends 6 months around the introduction of co-location. Panel A reports statistics for 10-year Government Bonds, Panel B reports 3-year Government Bonds, Panel C reports BABs, and Panel D reports SPI. t-Statistics are reported in parenthesis.

A. 10-Year Government Bonds
Pre0.0050520.00526898.9926136.977.0396−0.0110836.832332159973562710.000662
Post0.0050280.00520699.5135447.796.4395−0.0111616.843081213383675600.000653
Difference−0.000024−0.0000630.519210.90−0.5862−0.0000620.02747534111288−0.000006
 (−2.88)(−7.04)(3.34)(6.92)(7.06)(−1.21)(−0.14)(0.14)(6.79)(5.93)(1.73)(−0.19)
B. 3-Year Government Bonds
Pre0.0100240.01038399.80212024.215.0181−0.05391626.261968526355016500.000794
Post0.0100130.01030599.90325932.685.5974−0.05534230.392365725714550000.000798
Difference−0.000010−0.0000780.0911338.580.5883−0.0013404.1640020083−466500.000008
 (−1.94)(−9.94)(2.48)(9.24)(6.97)(5.91)(−0.76)(5.84)(4.10)(5.62)(−4.94)(0.16)
C. 90-Day Bank Accepted Bills
Pre0.0100870.01053799.2311597.657.5381−0.04294729.72425122582067860.000569
Post0.0100720.01045799.36161910.989.5969−0.03505731.82492147731598410.000577
Difference−0.000013−0.0000800.124623.332.07860.0078702.18662525−469460.000008
 (−0.70)(−4.00)(0.81)(5.60)(6.08)(6.98)(3.40)(1.63)(2.07)(2.53)(−9.35)(0.15)
D. Share Price Index
Pre1.0589030.02542194.3033339.1110.8755−0.0020622.0013668272972044930.012968
Post1.0353780.02464496.5446225.158.8269−0.0024952.0111253226312123990.009196
Difference−0.023530−0.0007652.2513−114.39−2.0424−0.0004350.01−2427−46877905.63995−0.003750
 (−4.43)(−4.84)(4.60)(16.37)(−9.99)(−2.56)(−6.86)(0.90)(−4.98)(−4.62)(2.39)(−6.45)

Despite the relatively strong evidence for interest rate futures, which supports the proposition that HFT activity increased following the introduction of co-location, the evidence for SPI futures appears to contradict this proposition. Specifically, message traffic, the order-to-trade ratio, and the Hendershott measure for HFT all decline from the pre- to the post-co-location period, and this decline is statistically significant at all conventional levels of significance. This is contrary to the notion that the introduction of co-location increases HFT activity. However, this is likely to be attributable to the introduction of the message traffic charge for equity markets during the post-co-location period, which therefore confounds our analysis of the introduction of co-location.5 We return to this issue below.

Table I also enables us to provide preliminary evidence on the impact of co-location on liquidity. Table I shows that the bid–ask spread in ticks and as a percentage of the price, declines across all futures contracts (including the SPI) from the pre- to the post-co-location period. This decline is statistically significant for 3- and 10-year Government Bond futures as well as SPI futures contracts. The table also reports that the portion of trades executed at the minimum tick increased significantly for all contracts except the BABs. We now turn to an assessment of market depth: another dimension of liquidity. Table I shows that depth available at the best bid and ask increased significantly following the introduction of co-location, across all futures contracts. In aggregate, the descriptive statistics reported in Table I therefore provide prima facie evidence that the liquidity of futures contracts improved following the introduction of co-location. However, these results can only be regarded as preliminary, as Table I shows various determinants of liquidity that are unlikely to be related to co-location were also changing significantly from the pre- to the post-co-location period, including price volatility and open interest. Hence, this confirms the importance of controlling for these variables in order to properly assess the impact of co-location on liquidity.

Table II re-examines the impact on the proxies for HFT activity, after controlling for market variables including price volatility and trading interest. The table enables us to draw similar conclusions to those drawn using raw proxies for HFT, and supports the proposition that HFT trading increased following the introduction of co-location.

Table II. Message Traffic Taxes and Algorithmic Trading
Variable10-Year Government Bonds3-Year Government BondsBank Accepted BillsShare Price Index
  1. This table presents regression analysis of the impact of the introduction of co-location facilities on algorithmic trading behavior. The following specification is estimated:

  2. inline image

  3. where ATit is the relevant algorithmic trading proxy. Panel A reports results for our first proxy, message traffic per minute. Results for Order-to-Trade Ratio measured as the number of messages divided by number of transactions are reported in Panel B. Panel C reports results for Algo Trade (volume) measured as (−1 × volume/100)/message traffic. Our regression specification controls for daily open interest, volatility, and a dummy variable CRC, which is set to zero prior to the introduction of the cost recovery charge imposed by the regulator ASIC on January 1, 2012, and set to one following its introduction. The CRC dummy is only included for the SPI contact as the recovery charge was imposed only for equities markets. Dummy variable, Colo is set to zero prior to the introduction of co-location facilities on February 20, 2012, and set to one following introduction of co-location. The event window extends 6 months around the introduction of co-location. t-Statistics are reported in parenthesis.

A. Messages
Intercept23.97990.71482.6672204.0512
 (5.06)(0.23)(2.41)(4.95)
Colo11.20019.49883.4110−29.7039
 (8.68)(10.23)(7.72)(−2.00)
CRC   −70.6745
    (−4.49)
Open interest−0.0000120.0000220.0000030.000173
 (−0.98)(3.48)(0.61)(0.94)
Volatility26217.5015508.637695.489152.81
 (10.64)(12.99)(16.46)(8.67)
B. Order-to-trade ratio
Intercept8.98326.02858.483812.4288
 (5.04)(17.59)(9.99)(3.54)
Colo−0.56900.53082.0514−0.5941
 (−1.17)(5.24)(6.06)(−0.47)
CRC   −3.1206
    (−2.33)
Open interest−0.000002−0.000001−0.0000010.000009
 (−0.52)(−1.88)(−0.22)(0.60)
Volatility−1625.73−445.54−1367.11−203.54
 (−1.75)(−3.42)(−3.81)(−2.26)
C. Algo Trade (volume)
Intercept−0.0076−0.0431−0.0694−0.0017
 (−4.83)(−7.15)(−10.75)(−6.20)
Colo0.0000−0.00140.0142−0.0003
 (0.00)(−0.81)(5.53)(−2.76)
CRC   −0.0003
    (−3.22)
Open interest0.0000000.0000000.0000000.000000
 (−1.59)(−0.32)(4.79)(0.09)
Volatility−1.66−11.16−2.64−0.02
 (−2.03)(−4.87)(−0.97)(−3.19)

Table III reports results of the regression analysis, which enables us to assess the impact of co-location on liquidity after controlling for changes in the known determinants of liquidity from the pre- to the post-co-location period such as trading activity and price volatility. Consistent with previous research, Table III reports a positive relation between price volatility and the various measures of bid–ask spreads, and a negative relation between our proxy for trading interest (open interest) and bid–ask spreads. Most importantly, an examination of the dummy variable for co-location implies that there was a decrease in bid–ask spreads across all four futures contracts examined. Panel A of Table III shows a significant decreases in bid–ask spreads (in ticks) for all futures contract, except BABs that do not change significantly. The coefficient on the co-location dummy for the regressions based on the bid–ask spread, as a portion of price, is negative and statistically significant for all futures contracts. Finally, Panel C of Table III reports the regression results based on the percentage of trades at the minimum tick suggests that co-location resulted in a significant decrease in bid–ask spreads for trades in 3- and 10-year Government Bond futures and SPI futures contracts. Hence, on the balance, the results reported in Table III confirm that the introduction of co-location was associated with an improvement in liquidity as measured by bid–ask spreads.

Table III. The Impact of Co-Location on Market Liquidity
Variable10-Year Government Bonds3-Year Government BondsBank Accepted BillsShare Price Index
  1. This table presents regression analysis of the impact of the introduction of co-location facilities on market liquidity. The following specification is estimated:

  2. inline image

  3. where Liquidity is proxied via; Dollar Tick Spread, Percentage Tick Spread, Percentage of Trades at the Minimum Tick, and Depth. Included in the estimated specification is a dummy variable, Colo, which is set to zero prior to the introduction of co-location facilities on February 20, 2012, and set to one following co-location. Panel A reports regressions for bid–ask spread in ticks, Panel B for percentage bid–ask spreads, Panel C for trades at the minimum tick, and Panel D for depth. The event window extends 6 months around the introduction of co-location. t-Statistics are reported in parenthesis.

A. Bid–ask spread (ticks)
Intercept0.0050450.0100210.0098821.051120
 (162.13)(568.16)(217.06)(48.18)
Colo−0.000024−0.0000140.000004−0.023611
 (−2.81)(−2.66)(0.24)(−4.03)
Open interest−0.000000−0.000000−0.000000−0.000000
 (−0.39)(−1.82)(−2.05)(−0.34)
Volatility0.028250.045030.210710.05195
 (1.74)(6.72)(10.96)(0.09)
B. Bid–ask spread (percent)
Intercept0.0052840.0103850.0103640.024457
 (160.30)(379.83)(204.69)(38.07)
Colo−0.000062−0.000082−0.000068−0.000643
 (−6.88)(−10.17)(−3.35)(−3.73)
Open interest−0.000000−0.000000−0.000000−0.000000
 (−1.07)(−1.29)(−1.34)(−0.77)
Volatility0.026610.042850.196560.03753
 (1.55)(4.12)(9.19)(2.17)
C. Trades at minimum tick (percent)
Intercept98.8799.66100.6194.92
 (173.15)(777.37)(257.26)(47.31)
Colo0.5042620.1275010.0117292.242069
 (3.24)(3.36)(0.08)(4.16)
Open interest0.0000010.0000010.0000030.000003
 (0.53)(2.62)(1.54)(0.28)
Volatility−252.02−248.63−1469.67−7.28
 (−0.85)(−5.10)−(8.90)(−0.14)
D. Depth
Intercept234.303242.352328.4442.97
 (4.96)(7.70)(10.60)(13.57)
Colo88. 301078.61278.8912.31
 (6.86)(8.66)(3.18)(14.47)
Open interest0.0002710.0012770.0040000.000031
 (2.17)(1.49)(4.16)(2.08)
Volatility−105043.791−607274.799−600907.774−243.522
 (−4.27)(−3.79)(−6.48)(−2.86)

Table III also reports the results of regression analysis based on depth. In relation to depth, Table III confirms that the introduction of co-location was associated with an increase in market depth across all futures contracts. These increases are significant at all conventional levels of significance. We conclude that the introduction of co-location facilities was associated with an increase in liquidity as measured by market depth.

One of the interesting conclusions that can be drawn from the results above is that an increase in HFT was not essential to an improvement in liquidity following the introduction of co-location. For example, while the changes in the proxies for HFT activity for 10-year Government Bond futures are mixed (raw message traffic increases significantly, while the order-to-trade ratio and the Hendershott proxy for algorithmic trade do not change significantly following the introduction of co-location), the change in the proxies for liquidity for 10-year Government Bond futures all imply that liquidity improved significantly following the introduction of co-location. Perhaps more stark is the evidence that message traffic decreased significantly while liquidity increased for SPI futures. Hence, co-location may have improved the efficiency with which HFT firms and other market participants are able to make markets, and therefore, resulted in an improvement in liquidity, without the introduction of any additional HFT activity.

The results above also enable us to estimate the overall benefit in terms of transaction costs associated with the introduction of co-location. Specifically, based on estimates of the change in bid–ask spreads (in ticks) reported in Table III, the size and value (in dollars) of the minimum tick and the annual contract turnover for each of the futures contracts for financial year 2012, we estimate the benefit through reduced bid–ask spreads associated with the introduction of co-location facilities. In aggregate, we estimate that the introduction of co-location was associated with a $12 million improvement in the cost of trading based on the bid–ask spread alone across the four futures contracts examined in this paper.6

4. FURTHER ANALYSIS AND ROBUSTNESS TESTS

  1. Top of page
  2. Abstract
  3. 1. INTRODUCTION
  4. 2. DATA AND METHOD
  5. 3. RESULTS
  6. 4. FURTHER ANALYSIS AND ROBUSTNESS TESTS
  7. 5. CONCLUSION AND SUGGESTIONS FOR FUTURE RESEARCH
  8. APPENDIX A
  9. APPENDIX B
  10. REFERENCES

In this section we undertake several analyses to probe our conclusions and also to assess their robustness. It was earlier suggested that the apparently conflicting evidence on the impact of the introduction of co-location on HFT was likely to be related to the introduction of a message traffic charge at the beginning of January 2012 for equities markets trades. This is likely to have increased the cost for high frequency traders that arbitrage equities and futures markets. Figure 1 depicts the ratio of message traffic to trades in the securities, which underpin the SPI futures. The diagram clearly demonstrates that the order-to-trade ratio declined following the introduction of the cost recovery charge in January 2012, from approximately 10 messages per trade to approximately eight messages per trade (or indeed 20%). This is similar to the decline in message traffic documented for SPI futures contracts in Table I that declined from the pre- to the post-co-location period from approximately 10.9 to 8.8 messages per trade (or by approximately 20%). This provides evidence to support our conjecture that the decline in message traffic for the SPI is related to the introduction of a message traffic and trade charge by the market regulator.

image

Figure 1. Algorithmic Trading in ASX Equity Securities. This figure presents daily observations of the order-to-trade ratio for the underlying ASX equity securities that comprise the SPI futures contract. The data include 2 months prior to the introduction of the cost recovery charge on January 1, 2012 (only for equities markets) by the market regulator ASIC. [Color figure can be viewed in the online issue, which is available at wileyonlinelibrary.com.]

Download figure to PowerPoint

While our results reported earlier examines the impact of co-location on liquidity by considering a 6-month observation interval before and after the introduction of co-location, such long periods increase the probability that extraneous variables enter our analysis and confound analysis. In order to test the robustness of our results, we replicate our analysis sampling only 3 months of data before and the introduction of the co-location facility. The results of this analysis are reported in Table IV. The results documented in Table IV confirm that our analysis is robust to the sample period used, and support our conclusion that the introduction of co-location is associated with an increase in liquidity.

Table IV. Replication with a 3-Month Event Window
Variable10-Year Government Bonds3-Year Government BondsBank Accepted BillsShare Price Index
  1. This table presents results of robustness tests of the impact of co-location on market liquidity. We shorten the event window considered in the estimation of the following specification:

  2. inline image

  3. where Liquidity is proxied via, Dollar Tick Spread, Percentage Tick Spread, Percentage of Trades at the Minimum Tick, and Depth. Panel A reports regressions for bid–ask spread in ticks, Panel B for percentage bid–ask spreads, Panel C for trades at the minimum tick, and Panel D for depth. The event window extends 3 months around the introduction of co-location.

A. Bid–ask spread (ticks)
Colo−0.0000−0.00000.0000−0.0165
 (−2.21)(−2.96)(1.25)(−2.16)
B. Bid–ask spread (percent)
Colo−0.0000−0.00000.0000−0.0010
 (−2.38)(−1.67)(0.66)(−5.05)
C. Trades at minimum tick (percent)
Colo0.51270.1743−0.22051.5112
 (2.44)(2.93)(−1.42)(2.13)
D. Depth
Colo102.76561262.3748387.963210.8637
 (5.22)(6.35)(2.78)(8.49)

The securities examined in this study are all futures contracts, which rely on underlying assets such as stocks and bonds. Changes in the liquidity of these underlying assets may have an impact on the liquidity of the futures market (see Chang, Jain, & Locke, 1995; Fabre & Frino, 2004; Jegadeesh & Subrahmanyam, 1993). For example, the cost of hedging market maker trades in futures is likely to increase if liquidity in the underlying asset decreases, and this is therefore likely to increase the cost of making markets in futures leading to a decrease in the liquidity of futures. In order to test the robustness of our conclusions to changes in the liquidity of underlying assets, we introduce trading volume in the spot market for three of the futures contracts examined in this study as a control variable in examining the association between co-location and liquidity. Specifically, we examine daily trading volume in the ASX/S&P 200 index, and the basket of bonds which make up the 3- and 10-year Government Bonds.7 Table V (which replicates and extends the equation underlying Table III) implies that the conclusions we draw from our analysis are robust to the inclusion of underlying volume in our analysis.

Table V. Inclusion of Underlying Trade Volume as Robustness Test
VariableShare Price Index3-Year Government Bonds10-Year Government Bonds
  1. This table presents results of robustness tests of the impact of co-location on market liquidity. Trading volume in the underlying instrument is included as a control variable in the following specification:

  2. inline image

  3. where Liquidity is proxied via, Dollar Tick Spread, Percentage Tick Spread, Percentage of Trades at the Minimum Tick, and Depth. Panel A reports regressions for bid–ask spread in ticks, Panel B for percentage bid–ask spreads, Panel C for trades at the minimum tick, and Panel D for depth. The event window extends 6 months around the introduction of co-location. t-Statistics are reported in parenthesis.

A. Bid–ask spread (ticks)
inline image−0.023004−0.000012−0.000024
 (−3.71)(−2.30)(−2.83)
Underlying volume−0.000000−0.000000−0.000000
 (−0.30)(−1.15)(−0.41)
B. Bid–ask spread (percent)
inline image−0.000632−0.000083−0.000062
 (−3.46)(−9.88)(−6.79)
Underlying volume−0.000000−0.000000−0.000000
 (−0.19)(0.17)(0.15)
C. Trades at minimum tick (percent)
inline image2.181750.113470.51738
 (3.83)(2.91)(3.29)
Underlying Volume0.0000000.0000000.000000
 (0.33)(1.49)(0.65)
D. Depth
inline image11.85661027.429090.2927
 (13.24)(8.04)(6.96)
Underlying Volume0.0000000.0000000.000000
 (1.57)(1.65)(1.20)

5. CONCLUSION AND SUGGESTIONS FOR FUTURE RESEARCH

  1. Top of page
  2. Abstract
  3. 1. INTRODUCTION
  4. 2. DATA AND METHOD
  5. 3. RESULTS
  6. 4. FURTHER ANALYSIS AND ROBUSTNESS TESTS
  7. 5. CONCLUSION AND SUGGESTIONS FOR FUTURE RESEARCH
  8. APPENDIX A
  9. APPENDIX B
  10. REFERENCES

The ASX introduced co-location for ASX futures markets on February 20, 2012. We provide evidence of an increase in HFT activity following the introduction of co-location. We also provide strong evidence of a decrease in bid–ask spreads and an increase in market depth following the introduction of co-location, across all futures contracts examined. We conclude that the introduction of co-location resulted in an improvement in liquidity of futures contracts.

A number of avenues for future research are possible. First, while we examine traditional measures of liquidity (bid–ask spreads and depth) in assessing the impact of co-location on liquidity, it may be possible to examine other measures including the market impact of institutional trades. Second, while we have examined the impact of co-location on the general level of liquidity, it may be useful to examine whether these results hold around important information announcements. Third, we suggest further exploring the differentiating impact of co-location and the cost recovery charge to disentangle the opposing impacts, particularly if the regulator applies these charges to derivative securities and not solely equity securities. Finally, HFT is related to the latency of the trading system and some exchanges have introduced new trading platforms that per se change latency (e.g., Singapore Exchange in 2011). It would be useful to examine whether the results of this study also apply to these other microstructure changes that alter the attractiveness of futures markets to high frequency trading.

APPENDIX A

  1. Top of page
  2. Abstract
  3. 1. INTRODUCTION
  4. 2. DATA AND METHOD
  5. 3. RESULTS
  6. 4. FURTHER ANALYSIS AND ROBUSTNESS TESTS
  7. 5. CONCLUSION AND SUGGESTIONS FOR FUTURE RESEARCH
  8. APPENDIX A
  9. APPENDIX B
  10. REFERENCES
AUSTRALIAN LIQUIDITY CENTRE

On February 20, 2012, the ASX introduced a co-location facility for trading of all futures contracts. The facility allows futures brokers to co-locate their computer servers in the same room as the computer server, which operates the futures trading system of the exchange, known as ASX Trade24.8 The co-location facility, named the Australian Liquidity Centre, is located in a dedicated building in a Sydney suburb approximately 5 km from the Central Business District of Sydney (where the head office of the ASX is located). According to ASX sources, on the first day of operation five futures brokers had co-located their servers in the co-location facility and approximately 20% of bid and ask orders in futures contracts were submitted through co-located servers.

The ASX co-location facility, which provides cabinet space for futures brokers or their clients includes a standard power supply (of 2 kW per month) and a fiber optic cable that connects each cabinet to ASX Trade24. These cabinets are installed in the same room as the computer server that runs ASX Trade24. The length of the fiber optic cable is equal (65 m) for each cabinet regardless of where it is physically located in the building. An order created by the server of a futures broker or their clients must travel from the cabinet of the broker down the fiber optic line to the computer server of the broker that manages the order known as the Automated Order Entry Interface (AOEI) or Gateway and then onto the matching engine (ASX Trade24).

The introduction of the co-location service significantly reduced the amount of time it took for an order to travel from the computer server of a broker (previously located in the broker's office) down a dedicated line to the trading system of the exchange and back. This time lapse is known as latency. ASX estimates of this latency are approximately 2 milliseconds. While the latency created by the 65 m fiber optic line is <2 microseconds, the bulk of the latency originates from the Gateway and the matching engine of ASX Trade24.9 Since the cable connecting each cabinet to the server of the Exchange is equal in length regardless of where the cabinet is located in the co-location facility, all co-located servers have equal latency.

Before the introduction of the ASX co-location facility, the ASX operated its ASX Trade24 trading system on a server housed at its headquarters in the central business district of Sydney. Each broker had a gateway in their office, and this was connected to the server, which operated ASX Trade24 via a dedicated line rented from a commercial telecommunications infrastructure supplier such as Telstra or Optus. Large brokers were located around the city of Sydney of up to approximately 5 km from the ASX, and hence, a message sent from their server to the exchange would have had to travel through a number of kilometers of line and through a number of networks or switches. One broker we spoke to suggested that the latency that this created could have exceeded 4–5 milliseconds, implying that the latency benefit from co-location was 2–3 milliseconds. Furthermore, the amount of latency was affected by the proximity of the brokers office to the ASX (which was costly to reduce as it required rental of office space close to the Exchange), the quality of the infrastructure in the building in which a broker was located and the quality of the infrastructure provided by the telecommunications supplier of a broker.

The cost of co-locating a server is approximately $5,000 per month. The cost of hiring a cabinet and the power supply is approximately $2,500 per month. The AOEI or gateway costs approximately $2,500 per month each and allows the entry of 12 messages per second.10 Most brokers would have two or more AOEI's. The marginal cost to the broker from allowing one of their clients (e.g., a HFT firm) to set up a cabinet is also $2,500 per month.11

The ASX charges an exchange fee of $0.90 per side for trades in 3-year Government Bonds, 10-year Government Bonds, BABs, and SPI futures. Arbitrage transactions involving SPI futures also involve transactions in stocks. Since June 30, 2010, transactions in stocks have incurred an exchange fee of 0.15 basis points per side.

APPENDIX B

  1. Top of page
  2. Abstract
  3. 1. INTRODUCTION
  4. 2. DATA AND METHOD
  5. 3. RESULTS
  6. 4. FURTHER ANALYSIS AND ROBUSTNESS TESTS
  7. 5. CONCLUSION AND SUGGESTIONS FOR FUTURE RESEARCH
  8. APPENDIX A
  9. APPENDIX B
  10. REFERENCES
MESSAGE TRAFFIC AND TRADE FEES

On August 1, 2010, the Australian government transferred the role of market supervision from the ASX to the regulatory body, ASIC. This decision was part of broader shift of market supervision from market operators to a streamlined single supervision and enforcement agency and the introduction of competing exchanges in Australia. A consequence of this regime shift was the introduction by ASIC, to charge a fee or collect revenue from regulated entities to fund market supervision services. For the period January 1, 2012–June 30, 2013, ASIC introduced an additional fee for messages and transaction in stocks executed on the ASX.12 The objective of the additional fees were to recover $2.487 million per quarter from brokers proportionate to their share in transactions and $0.215 million per quarter proportionate to their share in message traffic which was incurred by ASIC in supervising the market. Clearly, ASIC's fee per trade in stocks and fee per message in stocks could change each quarter with the volume of transactions and message traffic. The magnitude of these fees has not been disclosed publically, but is estimated it to be in the order of $0.03 per trade side for the fee component levied on the basis of the number of transaction.13

REFERENCES

  1. Top of page
  2. Abstract
  3. 1. INTRODUCTION
  4. 2. DATA AND METHOD
  5. 3. RESULTS
  6. 4. FURTHER ANALYSIS AND ROBUSTNESS TESTS
  7. 5. CONCLUSION AND SUGGESTIONS FOR FUTURE RESEARCH
  8. APPENDIX A
  9. APPENDIX B
  10. REFERENCES
Notes
  1. 1

    See Appendix A for further details regarding the co-location facility constructed by the ASX, including information on latency and access fees.

  2. 2

    Our results are robust to the inclusion or exclusion of dates close to expiration. Frino and McKenzie (2002) find trades occurring within 10 days of expiration of the near contract are likely to form part of rollover strategies associated with increased activity, declining spreads, and lower market impact costs in the period before contract maturity.

  3. 3

    ASX Trade24 is the trading platform used by the ASX to trade derivative contracts. ASX Match is the system used to trade equity products.

  4. 4

    See Appendix B for further details on the scheme.

  5. 5

    Regression analysis of trading volume in the futures contract 30 days pre and post the announcement of the cost recovery measure made on November 25, 2011, report no significant changes in anticipation of the impending tax on January 1, 2012. The authors thank the referee for this suggestion.

  6. 6

    We calculate the value of the introduction of co-location facility using the following equation: inline imageValue of minimum tick × Contract turnover in 2012 × [Coefficient on Colo Dummy (in ticks)/minimum tick]. For example for 10-year Government Bond futures this is equal to 38 × 17,220,000 × 0.000024/0.005 or $ 3.1 million.

  7. 7

    The basket of Government Bonds is published by the ASX as each interest rate futures contract is listed. We are unable to source underlying volume for the BABs.

  8. 8

    The Australian Securities Exchange originated out of the merger of the Australian Stock Exchange and the Sydney Futures Exchange. The trading system of the Exchange was previously called SYCOM and is described in Frino and Oetomo (2005) and Frino, Kruk, and Lepone (2007).

  9. 9

    A millisecond is one thousandth of a second while a microsecond is one millionth of a second.

  10. 10

    A message includes: a new bid or ask order, cancellation or amendment of an existing order.

  11. 11

    See ASX (2012a; 2012b; 2012c).

  12. 12

    See Market Supervision Cost Recovery Impact Statement, ASIC (2011).

  13. 13

    The Annual Report of the ASX for the year ended June 30, 2012 reported that there were 165 million trades in stocks for the year, or approximately 4.125 million per quarter. Given that the total fee to be recovered by ASIC each quarter was $2.487 million, then the approximate size of the fee was $0.03 cents per trade side.