A Tale of Two Consequences: Intended and Unintended Outcomes of the Japan TOPIX Tick Size Changes
AA Tale of Two Consequences:Intended and Unintended Outcomes of the Japan TOPIX Tick SizeChanges
Ravi KashyapSolBridge International School of Business / City University of Hong Kong
May 9, 2019Japan; Venue; Analysis; Tick; Size; Change; Exchange; Execution; Uncertainty; Costs; TradingJEL Codes: G15 International Financial Markets; D53 Financial Markets; G12 Trading Volume
Edited Version: Kashyap, R. (2015). A Tale of Two Consequences. The Journal of Trading,10(4), 51-95.
Contents a r X i v : . [ q -f i n . T R ] M a y Possibilities for a Deeper Dive 226 Does Tick Size Matter? Tick Size Does Matter! 227 Acknowledgements and End-notes 238 References 239 Appendix - I (Bird’s Eye View Comparisons) 2610 Appendix - II (Deep Dive Comparisons) 32
We look at the effect of the tick size changes on the TOPIX 100 index names made by the Tokyo StockExchange on Jan-14-2014 and Jul-22-2014. The intended consequence of the change is price improvementand shorter time to execution. We look at security level metrics that include the spread, trading volume,number of trades and the size of trades to establish whether this goal is accomplished. An unintended effectmight be the reduction in execution sizes, which would then mean that institutions with large orders wouldhave greater difficulty in sourcing liquidity. We look at a sample of real orders to see if the execution costshave gone up across the orders since the implementation of this change.We study the mechanisms that affect how securities are traded on an exchange, before delving into thespecifics of the TSE tick size events. Some of the topics we explore are: The Venue Menu and How to IncreaseRevenue; To Automate or Not to Automate; Microstructure under the Microscope; The Price of Connectionsto High (and Faraway) Places; Speed Thrills but Kills; Pick a Size for the Perfect Tick; TSE Tick SizeExperiments, Then and Now; Sergey Bubka and the Regulators; Bird’s Eye View; Deep Dive; Possibilitiesfor a Deeper Dive; Does Tick Size Matter? Tick Size Does Matter!
The more that shoppers shop, the more shops there will be and the more the shops will try to woo theshoppers. Similar has been the effect of increasing financialization across the globe. Greater levels of trading,by both retail and institutional investors, have resulted in more exchanges springing up and offering moreproducts that can be traded. The longer the menu of venues, more the competition among them, and thisnaturally requires attempts at trying to attract and retain customers, by a venue to increase its revenue. Ifpeople are willing to pay (or bid) more than what is asked (or offered), then perhaps, we would not havespecialized venues to trade, a small price to pay for, let us just say, peace on Earth. Forgetting about2topia - but keeping in mind that conceivably, the Bid-Offer spread can be a barometer to a civilization’sprogress, till it becomes irrelevant, indicating that a society has transcended beyond mere material mattersof accumulating and allocating wealth - a brief and worthy digression would be to look at how exchangeshave evolved and what factors drive the future development of trading venues.An exchange, as the word implies, is the process during which people give and take things of similarvalue. At a place where this transfer happens, also an exchange, shares or holdings can be liquidated andhence the primary mission of an exchange is to provide liquidity. For the rest of the discussion, we ignorethe exchange of OTC (Over The Counter) securities, which are traded wherever, whenever and however onecan trade them; but we leave the reader with the analogy that if Exchange Trading is similar to collectingtolls on a road; OTC Trading is like highway robbery.As with most historical matters, there is no agreement on when and where the first stock exchange wasstarted. There seems to be some consensus that the first exchanges were started to finance East Indiacompanies that provided investment for merchants that sailed on the high seas to conduct business withvarious countries in Asia. In addition to raising capital, these trading arenas, offered a means for the transferand sharing of risk. The effect of this simple movement of securities between owners, ripples across, getsmagnified and affects the entire economy. See, (Michie 2001), for an excellent exposition on the questionsexchanges faced centuries ago and how they are similar to the issues that are cropping up today, with a focuson the London Stock Exchange (LSE). Conceptually, the trading venues of today, still perform the sameduties, since all of finance, through time, has involved three simple outcomes – “Buy, Sell or Hold”. Thecomplications are mainly to get to these results.
Alongside the progression of exclusive trading locations, a parallel development has been the increased useof automation and technology in the buying and selling of securities. This has removed the traditional conceptof a brick and mortar building where specialists or jobbers (the market makers on the NYSE - New YorkStock Exchange - and LSE) acted as the counter-parties for brokers, who were undertaking orders on behalf ofthe end investors. Since the advent of the National Association of Securities Dealers Automated Quotations(NASDAQ), the first electronic stock market, many more electronic trading networks have proliferated andvirtual trading has been adopted even by the remaining physical exchanges to a great degree.There is no dearth of evidence that the rules of trading affect the profitability of various trading strate-gies. Venkataraman (2001) compares securities on the NYSE (floor-based trading structure with humanintermediaries, specialists and floor brokers) and the Paris Bourse (automated limit order trading structure).He finds that execution costs might be higher on automated venues even after controlling for differencesin adverse selection, relative tick size and economic attributes. A trade occurs when an aggressive tradersubmits a market order and demands liquidity, hence the rules on a venue are designed to attract demanders3f liquidity, and nudge liquidity providers to display their orders. Displaying limit orders involves risks. First,the counter-parties could be better informed and liquidity providers could get picked off. Hence, they wouldlike the trading system to allow them to trade selectively with counter-parties of their choice. Second, theyrisk being front-run by other traders with an increase in the market impact of their orders. Hence, largetraders want to hide their orders and expose them only to traders who are most likely to trade with them.This means fully automated exchanges, which anecdotally seems to be the way ahead, need to take specialcare to formulate rules to help liquidity providers better control the risks of order exposure.
Market microstructure is the investigation of the process and protocols that govern the exchange ofassets with the objective of reducing frictions that can impede the transfer. In financial markets, wherethere is an abundance of recorded information, this translates to the study of the dynamic relationshipsbetween observed variables, such as price, volume and spread, and hidden constituents, such as transactioncosts and volatility, that hold sway over the efficient functioning of the system. Madhavan (2000) provides acomprehensive survey of the theoretical and experimental literature relating to: price formation, the dynamicprocess by which prices come to impound information; market structure and design, the relation betweenprice formation and trading rules; and transparency, the ability of market participants to observe informationabout the trading process.Differences in microstructure, such as tick size, spread, trade depth and clustering of prices (clustering isthe tendency for prices to fall on a subset of available prices), could be due to differences in market structuresuch as whether a given market is a dealer market or an auction market. Auction markets are order driven,where buy orders seek the lowest available prices and sell orders seek the highest available prices. Thisprocess is called the price discovery process because it reveals the prices that best match buyers to sellers.Dealer markets are quote-driven because prices are set only by dealer quotes in the market. Huang and Stoll(2001) study securities on the LSE and American Depository Receipts (ADRs) on the NYSE. The key featuredifferentiating the two market structures is the treatment of public limit orders. In an auction market, limitorders are displayed and may trade against incoming market orders. In a dealer market, limit orders areheld by each dealer, are not displayed, and can only be traded against the dealer’s quote. Spreads could behigher in dealer markets since they are set by dealers. A minimum tick is necessary in an auction market toencourage liquidity provision by limit orders and by dealers. Without a minimum tick (or a minimum tradesize), a limit order can cheaply step ahead of another limit order or a dealer quote. If there is no minimumtick, it is easy to avoid time priority. Dealer markets do not require time priority across dealers and they haveless need for a minimum tick. However, each dealer quotes in depth even in the absence of a tick rule becausehe wishes to maintain a reputation for liquidity or because dealer markets set standards as to depth. Quoteclustering is highly correlated with spreads and with the stock characteristics that determine spreads. If a4arket has higher spreads it has greater clustering. Trade clustering is lesser relative to quote clustering indealer markets indicating that negotiations with dealers can be successful even though negotiation for betterprices by customers takes place off the screen whereas negotiation in an auction market takes place on thescreen via limit order placement. In auction markets, limit orders break up quote clustering as they seek togain priority and trades cluster to a similar degree as quotes. Higher spreads are accompanied by greaterdepth. Trade sizes are larger consistent with the large depth, but the difference in trade size is not as greatas the difference in depths.While a number of variables can be observed on an exchange, the primary lever available for adjustmentis the spread. (Roll 1984 and Stoll 1989) connect stock price changes to the bid-offer spread. The spread isdetermined due to order processing costs, adverse information or inventory holdings costs. A key distinctionto be aware of is between the quoted spread and the realized spread. The quoted spread is the differencebetween the ask price quoted by a dealer and the bid price quoted by a dealer at a point in time. The realizedbid-ask spread is the average difference between the price at which a dealer sells at one point in time andthe price at which a dealer buys at an earlier point in time. The quoted spread is related to characteristicsof securities such as the volume of trading, the stock price, the number of market makers, the volatility, andother factors. If the spread reflects only order processing costs, the bid and the offer always straddle thetrue price. The dealer covers costs by buying at the bid and selling at the offer (on average). Sequencesof purchases at the bid price are ultimately offset by sequences of sales at the ask price. In this case, therealized spread and quoted spread are the same. An implication of both the inventory cost model and theadverse information cost model is that the realized spread earned by a dealer is less than the spread quotedby the dealer (empirical studies show this to be the case; references mentioned earlier in the paragraph havemore details). Under the inventory cost model, this is because the dealer lowers both bid and ask prices aftera dealer purchase in order to induce dealer sales and inhibit additional dealer purchases and raises both bidand ask prices after a sale in order to induce dealer purchases and inhibit dealer sales. The net effect of bidand ask price changes are such that future transactions that will equilibrate inventory. New prices are setsuch that the dealer is indifferent between a transaction at the bid price and a transaction at the ask price.Under the adverse information cost model, bid and ask prices are changed in a similar way to reflect theinformation conveyed by transactions. After a sale to the dealer, bid and ask prices are lowered because atransaction conveys information that the expected equilibrium price of the security is lower. Transactionsconvey information under the assumption that some traders are better informed than others.
Technology coupled with globalization is causing financial markets to be linked. While the global financialconnection is mostly implicit at this stage, explicit links across exchanges are being added (See Kashyap 2015bon the Hong Kong Shanghai Connect). Caldarelli, Marsili, and Zhang (1997) show that even a simple model5f an exchange, operating as a completely closed system with no external influences, where the participantstrade with the sole purpose of increasing their capital after observing the price history, can produce richand complex fluctuations in prices. In our numerical explorations, we return to this premise and restrict ourobservations to measurements that can be directly gathered on an exchange. Structuring the study in thisway abstracts away from the subjective decision regarding what external variables can influence a system.This benchmark scenario, which is a simplified platform to glean illuminative lessons, differs from a realisticsetting with regards to the changes in demand from an exchange participant, which can be influenced byexternal factors as observed by the participant himself or from the order flow he receives from investorsthat are not exchange members, who could be acting due to other extraneous forces. Karolyi and Stulz(1996) explore factors that affect cross-country stock return correlations using dollar-denominated returnsof U.S. and Japanese shares trading in the U.S. They find that U.S. macroeconomic announcements, shocksto the Yen/Dollar foreign exchange rate and Treasury bill returns, and industry effects have no measurableinfluence on U.S. and Japanese return correlations or the co-movements between U.S. and Japanese sharereturns. However, large shocks to broad-based market indices (Nikkei Stock Average and Standard and Poor’s500 Stock Index) positively impact both the magnitude and persistence of the return correlations. Possibly,movements in prices are the biggest contributors to further price movements.Adding technology enabled buying and selling, which can support more trades, to the ability to buy andsell in distant lands, accelerates the transfer of securities, causing prices to be displaced back and forth fromany equilibrium (rather pseudo equilibrium) which causes more transactions to happen affecting the entirecycle of investment management (Kashyap 2014a). Advocates of setting daily limits on how much the priceof a security can change over the course of a trading day believe that such measures can decrease stock pricevolatility, counter overreaction, and do not interfere with trading activity. Daily price limit critics claimthat price limits cause higher volatility levels on subsequent days (volatility spillover hypothesis), preventprices from efficiently reaching their equilibrium level (delayed price discovery hypothesis), and interfere withtrading due to limitations imposed by price limits (trading interference hypothesis). Kim and Rhee (1997)study the effect of daily price limits on the Tokyo Stock Exchange (TSE) and despite small sample size issuesfind evidence supporting all three hypotheses, suggesting that price limits may be ineffective.This constant jumping of prices and the associated transfer of capital gives rise to a highly specializedwork force that seeks to plug every source of inefficiency and profit from it (Kashyap 2015a). As funds flowfrom the benefactor to the beneficiary, chunks of it are taken by players along the financial food chain. Thetouch point during this transfer of capital, the exchange, has to seen as reducing any potential losses. Butintermediaries and their actions can possibly aid the overall process of liquidation, making it quicker andefficient; it also brings more trading volumes, a main source of income for the exchanges. There is also anindirect channel as more participants and more volume begets lower spreads, which lowers execution costs,6hich induces more volume, which then generates more profits. Easley and O’Hara (2010) demonstrate thepotential benefits to exchanges, investors and firms from reducing ambiguity over how markets work or assetprices are formed. Uncertainty can cause some traders to be overly influenced by “worst case” outcomes,even when these outcomes have little objective possibility of occurring. This, in turn, can cause such naiveinvestors to opt not to participate in markets, a result detrimental to both markets and the economy alike.Microstructure features (which in their study refers to listing thresholds; monitoring to ensure fair andnon-manipulative trading; and operational oversight of clearing and settlement to insure that a trader whobuys stock actually receives it and that one who sells stock actually delivers it) can be used to reduce thisambiguity, and thereby induce greater participation in markets. A side effect of excess liquidity might beinvestor passivity and fragmentation of stockholdings (Bhide 1993), since investors can exit stocks they don’tlike easily and their holdings are not big enough to be a voice for better corporate governance.
The technical arms race will give rise to a situation where participants focus their efforts on tradingfaster once certain kinds of new information is received since being first would mean the difference betweenprofits and losses (or perhaps, just the difference between profits and lesser profits, which can sometimes seemequally worse in an atmosphere where milliseconds matter). Ye, Yao & Gai (2012) confirm the old adage,speed thrills but kills. They find evidence that increasing the speed of trading from the microsecond levelto the nanosecond level, lead to dramatic increases in message flow. The increases in message flow are duelargely to increases in order cancellations without any real increases to actual trading volume. Spread doesnot decrease following increase in speed; market efficiency, in terms of price formation, does not improve;market depth decreases and short-term volatility increases, probably as a consequence of more cancellations.A fight for speed increases high-frequency order cancellation but not real high-frequency order execution.Increased cancellation generates more noise to the message flow. Low-frequency traders then subsidize high-frequency traders because only executed trades are charged a fee. The exchanges continually make costlysystem enhancements to accommodate higher message flow, but these enhancements facilitate further ordercancellations, not increases in trading volume. Investment in high frequency trading with sub-millisecondaccuracy may provide a private benefit to traders without consummate social benefit; therefore, there maybe an over-investment in speed.The point that warrants further consideration is whether High Frequency Trading (HFT) is leading tobenefits by either directly providing additional liquidity or indirectly via the spawning of numerous techno-logical innovations in computer networking hardware, software or other items used to facilitate HFT, whichcan then be beneficial to other sectors. An example that concerns high speeds is from the car racing industry,which comes up with innovations that produce faster and safer cars. Many of these innovations slip into themainstream automobile industry over time. Completely restricting any endeavor is not ideal since it hard to7now where the next life changing idea might spring up; but regulating the dangerous or unfavorable ones isprudent. Learning further from this example, we do not see race-cars cruising down our town streets (though,not something to rule out entirely in the not too distant future); they hustle around in an exclusive arena,indicating that perhaps we need a similar mandate for the HFT industry.
From a theoretical point of view (Harris 1994), the tick size is the lower bound of the bid-ask spread.We can then expect that a reduction in the tick size would decrease the quoted spread. Nevertheless, thereduction in the spread could also decrease order exposure because liquidity provision is less profitable andmore risky. As a consequence, the quoted depth could also decline. Goldstein and Kavajecz (2000) andLipson and Jones (2001) explore the effects of the tick size reduction on the NYSE and find that whiledecrease in tick sizes might improve the liquidity for small size orders, institutional traders were worse offbecause they had to bear an increase in trading costs following the decline in depth throughout the entireorder book. Bessembinder (2003) reports evidence regarding trade execution costs and market quality afterthe 2001 decimilization on NYSE and NASDAQ, which includes narrower average quoted, effective, andrealized bid-ask spreads on both markets, lower volatility on both markets, and the absence of systematicreversals of quote changes on either market, indicating that market quality has indeed been improved, whileadmitting that a complete assessment of the impact of decimalization on market quality will also requireaccess to proprietary data on institutional trading programs, in order to assess whether trading costs forlarge institutions have also declined.Bourghelle and Declerck (2004) look at the consequences of a change in the tick size at the Paris Bourse,where there was a both a decrease and increase in tick size on different groups of securities and hence offeredan opportunity to simultaneously explore the issues involved. Decreased tick sizes induced a decrease in depthat the quotes. However, in contrast with results obtained for US markets, (reflecting differences in marketdesign between European and US exchanges), this change neither generated a change in the bid-ask spreadnor a reduction in liquidity provision for large trades. Limit order submission inside the best quotes (on thebest quotes) increases significantly, and investors use more hidden quantity orders to reduce exposure. Stocksthat experienced an increase in the tick size did not have altered spreads, but it increased the depth at thebest quotes, showing evidence of a larger display of liquidity. Limit order submission inside the best quotessignificantly decreases, and investors use fewer hidden quantity orders. By increasing the per-share rent, alarger relative tick makes liquidity supply more profitable and probably attracts new limit order traders inthe market. To attract liquidity demanders, designers of trading systems have to stimulate investors to fullydisplay their orders. A relatively coarse pricing grid does not always result in excessively large spreads, butenhances quoted depth, encourages liquidity providers to expose their trading interest and stimulate investorsto quote the competitive spread. 8hn, Cao and Choe (1998) examine the impact of decimalization in Canada and find a significant reductionin the spread and quotation depth on the Toronto Stock Exchange (ToSE) and a significant reduction in thespread on NASDAQ for ToSE stocks indicating that NASDAQ dealers might not operate as efficiently asperfect competition warrants and could quote narrower spreads without any rule change on the NASDAQ.However, the decimalization does not affect the spread on the NYSE and American Stock Exchange (AMEX)for ToSE cross-listed stocks. The most important finding is that despite an economically significant reductionin the spread on the ToSE, orders for the cross-listed stocks do not migrate from U.S. markets to the ToSE.This result contrasts with the ToSE’s objective to attract order flows from the U.S. markets and to increasethe market share of the ToSE in cross-listed stocks. The savings in transaction costs on the ToSE are notsufficient to offset the benefits of trading (which include the ease of trading and superior execution of blocks)on the NYSE and AMEX. The practice of payments for order flow has existed between Canadian brokersand U.S. dealers for years. Given the restriction that a Canadian broker cannot accept payments for thepurchased order from a Canadian dealer but can accept payments from a U.S. dealer, there is little incentiveto direct the order to the TSE for execution, even though the TSE offers lower trading costs. Chung and VanNess (2001) look at the intraday effect on spreads due to the Order Handling Rules (OHR, which includedquote depth and tick size reductions) implemented on the NASDAQ in 1997. They find that the tick-sizereduction led to a significant decline in spreads with the magnitude of the decline being largest (smallest)during the last (first) hour of trading and to a significant decrease in quoted depths with the magnitude ofthe decline being smallest during the first hour of trading. Bessembinder (1999) finds that executions costson the NASDAQ remain higher compared to the NYSE even after the OHR was put in place, though thecross-market differential has decreased steadily over time. Some explanations for the difference could be that:one, NASDAQ securities have different economics characteristics (such as greater return volatility or smallerinvestor base); two, NASDAQ’s quote driven dealer structure could be less efficient than the order drive NYSEstructure; three, or the NASDAQ facilitates a form of collusion that can keep spreads higher. A point worthnoting is the preferencing arragements whereby orders are routed by brokers to dealer based on preexistingagreements rather than to the market maker displaying the best quote might have been responsible for alack of competition on the NASDAQ. Bessembinder (2000) examines changes in trade execution costs andmarket quality for a set of NASDAQ listed firms whose tick size changed as their share prices passed through$10. Though there was apparently no written rule, the convention on Nasdaq during 1995 was to use ticksizes of 1/8 dollar for bid quotations at or above $10 per share and 1/32 dollar for bid quotations below $10per share. The empirical results indicate that spreads decreased and there was no evidence of a reduction inliquidity.Bollen and Busse (2006) measure changes in mutual fund trading costs following two reductions in thetick size of U.S. equity markets: the switch from eighths to sixteenths and the subsequent switch to decimals.9hey estimate trading costs by comparing a mutual fund’s daily returns to the daily returns of a syntheticbenchmark portfolio that matches the fund’s holdings but has zero trading costs by construction. Smaller ticksizes lower depth, thereby penalizing institutional investors. Large institutional orders are sensitive to marketdepth for at least two reasons. First, filling a large order may take several days and multiple transactions;hence a large order likely suffers price concessions as market depth is consumed. Second, information leakagemay move prices adversely as the institutional investor attempts to fill the order. Investors who trade smallquantities of individual equities benefit from the tighter spreads following the switch to decimal pricing, andare largely unaffected by any decline in depth.These results underscore the view that market structure has a significant effect on trading costs and hencethe design of optimal trade and quote dissemination protocols coupled with proper regulatory oversight of theinvestment process are essential for investor welfare and market quality. Of all the weapons in the regulatoryarsenal, it seems, changes that can influence the price formation process without directly setting price levels,hold the greatest power. One takeaway from these studies is that regulators may be well advised to avoidreducing tick size if they want to attract liquidity providers, and if order exposure is profitable for a market.
The TSE has tried its hand earlier at tick size changes when it introduced a change in its minimum ticksizes on April 13, 1998. The TSE is one of the largest limit order markets using a tick size that is a stepfunction of share price. The reduction in tick size therefore depends on price ranges. Ahn, Cai, Chan andHamao (2007) investigate the liquidity and market quality of the stocks affected by this change. They find thatthe quoted spread and the effective spread declined significantly. Reductions in spread are greater for firmswith greater tick size reductions, greater trading activity, and higher monopoly rent proportion in the bid-askspread component. There is an increase in the quote revision (relative to the number of trades), suggestingthere is more price competition among limit order traders in providing liquidity. Although investors aremore aggressive in posting quotes, there is no definite evidence of an increase in trading volume reflecting adecrease in depth provided to the market.The current change has a three phase implementation over a period of close to two years.• Phase 1 was a pilot phase that went live from 14th January 2014. It covered stocks from TOPIX100 index and reduced tick sizes only for stocks with quote price above ï¿œ3000. Due to the highconcentration of stocks with lower market price in TOPIX, this pilot phase had an impact on only 39stocks. For simplicity, we consider the affected securities based on the prices as of the ex-date.• Phase 2 commenced on 22nd July 2014 over the same universe of stocks, i.e. TOPIX 100. It introduceddecimal yen tick sizes for stocks trading below ï¿œ5000. This phase had an impact on 80 stocks. For10implicity, we consider the affected securities based on the prices as of the ex-date.• Phase 3, expected in September 2015, will be based on the tick structure of Phase 2. However, TSE willannounce the final tick sizes and list of target stocks after evaluating the impact of previous 2 phases.Figure 1: Tick Size Change Schedule
Any attempt at regulatory change is best exemplified by the story of Sergey Bubka (End-note 2), theRussian pole vault jumper, who broke the world record 35 times. Attempts at regulatory change can becompared to taking the bar higher. In this case, the intended effect of the change is price improvement andshorter time to execution. We look at security level metrics that include the spread, trading volume, numberof trades and the size of trades to establish whether this goal is accomplished. Despite all the uncertainty(Kashyap 2014b), we can be certain of one thing, that certain market participants will find some way overthe intended consequences, prompting another round of rule revisions, or raising the bar, if you will. In thiscircumstance, an unintended effect caused by the reactions of participants to the new rules, might be thereduction in execution sizes, which would then mean that institutions with bigger orders would have greaterdifficulty in sourcing liquidity. We look at a sample of real orders to see if the executions costs have gone11p across larger orders since the implementation of this change. So far, we have talked about the unknowns(or unintended consequences) that we know about (or can anticipate). What about the unknowns that wedon’t know about (or cannot even imagine). The only thing, we know about these unknown unknowns arethat, there must be a lot of them, hence the need for us to be eternally vigilant, compelling all attempts atrisk management to make sure that the unexpected, even if it does happen, is contained in the harm it cancause, while being cognizant that this is easier said than done; a topic best saved for another time.
Before doing an in-depth study over a large sample of data, we perform a cursory check around the dayssurrounding the two ex-dates to see if we can spot any hints of change. We did three groups of high levelcomparisons to ensure that the observations are not restricted to the altered dynamics of trading on anysingle day. The key metrics we used in this analysis are the average spread, measured in yen per share andthe average execution size measured in number of shares. We adjust the price and spread according to thesplit ratio for securities that had stock splits during the time period of our data sample (Figure 4).1. We compared the metrics on the day of the change to the metrics on the immediately previous tradingday; i.e. between Jan 14th and Jan 10th for the first phase of the change; and between Jul 22nd and Jul18th for the second phase of the change (Jan 13th and Jul 21st were public holidays in Japan).2. We compared the metrics one and two days after the change for the first phase and second phaserespectively with the metrics exactly a week before this day; i.e. between Jan 15th and Jan 8th; and Jul 24thand Jul 17th.3. We compared the metrics between the earliest date in this limited sample and the last day we haveobserved after the change; i.e. between Jan 16th and Jan 6th for the first phase of the change; and betweenJul 28th and Jul 14th for the second phase of the change.
As expected, after the change to decrease the tick size on the TOPIX 100 index names on Jul 22 2104and Jan 14 2104, the average spreads have decreased consistently, across the names that had prices in therange which would be affected by the new rules. On the days immediately following the change, the spreadsare down for almost all the securities affected by the change. This also fits in with the fact that the minimumtick size after the change is 0.1 yen across some of the names. Another observation due to this change isprobably the unexpected change in the average execution size on these names. While this change is not assignificant and consistent as the spread changes, the initial results do indicate a trend towards decreasedexecution sizes.For the first phase, the average spreads have decreased across 100%, 100% and 100% of the affected namesover the three sets of comparisons that we did (It is across 63%, 72% and 84% over the full set of names and12he change is across 96% of the names when we consider only the names where the spread was on averagein excess of 1 yen earlier) and for the second phase the spreads have decreased across 100%, 100% and 96%of the affected names over the three sets of comparisons (It is across 100%, 90% and 86% of the full set ofnames and the magnitude of the change is considerably smaller as compared to the change that was done inJan 2014).For the first phase, the average execution size has decreased across 100%, 95% and 95% of the affectednames over the three sets of comparisons that we did (It is across 89%, 70% and 65% over the full set of namesand there seems to be no threshold over which the decreased size is entirely consistent, we need additionalobservations for this metric) and for the second phase the average execution size has decreased across 93%,88% and 79% of the affected names over the three sets of comparisons (It is across 81%, 77% and 72% ofthe full set of names). The summary of the results (Figures 2 and 3) and the details (Figures 5 to 8) acrossindividual securities are given in section 9, Appendix - I.
A fly through of the data across the two ex-dates reveals that a detailed analysis would indeed be aworthy endeavor. As part of this deep dive, first, we perform stationarity checks on prices, spreads, tradesizes and volumes and also consider whether these variables have become more volatile since the changes. Wesupplement these with statistical tests and evaluate properties that can establish trends regarding whether thevariables are either increasing or decreasing, after the two event dates. While this is interesting information, aquestion of paramount importance is, “what is the effect of these changes on the trading costs across differentsize orders”? For this we consider trading costs on a sample of close to 250,000 real orders starting six monthsbefore the first event and ending six months after the second event. Before we tackle the crucial conundrumof trading costs, we review some basics regarding the measurement of transaction costs.
The unique aspect of our approach to trading costs is a method of splitting the overall move of thesecurity price during the duration of an order into two components (Collins and Fabozzi 1991; Treynor 1994;Yegerman and Gillula 2014). One component gives the costs of trading that arise from the decision processthat went into executing that particular order, as captured by the price moves caused by the executionsthat comprise that order. The other component gives the costs of trading that arise due to the decisionprocess of all the other market participants during the time this particular order was being filled. Thissecond component is inferred, since it is not possible to calculate it directly (at least with the present stateof technology and publicly accessible data) and it is the difference between the overall trading costs and thefirst component, which is the trading cost of that order alone. The first and the second component arisedue to competing forces, one from the actions of a particular participant and the other from the actions of13veryone else that would be looking to fulfill similar objectives. Naturally, it follows that each particularparticipant can only influence to a greater degree the cost that arises from his actions as compared to theactions of others, over which he has lesser influence, but an understanding of the second component, can helphim plan and alter his actions, to counter any adversity that might arise from the latter. Any good traderwould do this intuitively as an optimization process, that would minimize costs over two variables directimpact and timing, the output of which recommends either slowing down or speeding up his executions.With this measure, traders now actually have a quantitative indicator to fine tune their decision process.When we decompose the costs, it would be helpful to try and understand how the two sub costs could varyas a proportion of the total. The volatility in these two components, which would arise from different sources(market conditions), would require different responses and hence would affect the optimization problemmentioned above invoking different sorts of handling and based on the situation, traders would know whichcost would be the more unpredictable one and hence focus their efforts on minimizing the costs arising fromthat component. Another popular way to decompose trading costs is into temporary and permanent impact[See Almgren and Chriss (2001); Almgren (2003); and Almgren, Thum, Hauptmann and Li (2005)]. Whilethe theory behind this approach is extremely elegant and considers both linear and nonlinear functions ofthe variables for estimating the impact, a practical way to compute it requires measuring the price a certaininterval after the order. This interval is ambiguous and could lead to lower accuracy while using this measure.We now introduce some terminology used throughout the discussion.1. Total Slippage - The overall price move on the security during the order duration. This is also a proxyfor the implementation shortfall (Perold 1988; and Treynor 1981). It is worth mentioning that there aremany similar metrics used by various practitioners and this concept gets used in situations for which it isnot the best suited (Yegerman and Gillula 2014). While the usefulness of the Implementation Shortfall,or slippage, as a measure to understand the price shortfalls that can arise between constructing aportfolio and while implementing it, is not to be debated, slippage need to be supplemented with moregranular metrics when used in situations where the effectiveness of algorithms or the availability ofliquidity need to be gauged.2. Market Impact (MI) - The price moves caused by the executions that comprise the order under consid-eration. In short, the MI is a proxy for the impact on the price from the liquidity demands of an order.This metric is generally negative or zero since in most cases, the best impact we can have is usually noimpact.3. Market Timing - The price moves that happen due to the combined effect of all the other marketparticipants during the order duration. 14. Market Impact Estimate (MIE) - This is the estimate of the Market Impact, explained in point twoabove, based on recent market conditions. The MIE calculation is the result of a simulation whichconsiders the number of executions required to fill an order and the price moves encountered whilefilling this order, depending on the market microstructure as captured by the trading volume and theprice probability distribution, over the past few days. See Kashyap 2015b for a dynamic programmingapproach to minimize the Market Impact under various formulations of the law of motion of prices.This simulation can be controlled with certain parameters that dictate the liquidity demanded onthe order, the style of trading, order duration, market conditions as reflected by start of trading andend of trading times. In short, the MIE is an estimated proxy for the impact on the price from theliquidity demands of an order. Such an approach holds the philosophical viewpoint that making smallerpredictions and considering their combined effect would result in lesser variance as opposed to makinga large prediction; estimations done over a day as compared to estimations over a month, say. Ageometrical intuition would be that fitting more lines (or curves) over a set of points would reduce theoverall error as compared to fitting lesser number of lines (or curves) over the same set of points. Whencombining the results of predictions, of course, we have to be mindful of the errors of errors, which canget compounded and lead the results astray, and hence, empirical tests need to be done to verify thesuitability of such a technique for the particular situation.5. All these variables are measured in basis points to facilitate ease of comparison and aggregation acrossdifferent groups. It is possible to measure these in cents per share and also in dollar value or othercurrency terms.6. The following equations, expressed in pseudo mathematical terms to facilitate easier understanding,govern the relationships between the variables mentioned above.
Total Slippage = Market Impact + Market Timing{Total Price Slippage = Your Price Impact + Price Impact From Everyone Else (Price Drift)}Market Impact Estimate = Market Impact Prediction = f (Execution Size, Liquidity Demand)Execution Size = g(Execution Parameters, Market Conditions)Liquidity Demand = h(Execution Parameters, Market Conditions)Execution Parameters <->vector comprising (Order Size, Security, Side, Trading Style, Timing Decisions)Market Conditions <-> vector comprising (Price Movement, Volume Changes, Information Set)Here, f, g, h are functions. We could impose concavity conditions on these functions, but arguably, similar15esults are obtained by assuming no such restrictions and fitting linear or non-linear regression coefficients,which could be non-concave or even discontinuous allowing for jumps in prices and volumes. The specificfunctional forms used could vary across different groups of securities or even across individual securities oreven across different time periods for the same security. The crucial aspect of any such estimation is thecomparison with the costs on real orders, as outlined earlier. Simpler models are generally more helpful ininterpreting the results and for updating the model parameters. Hamilton [1994] and Gujarati [1995] areclassic texts on econometrics methods and time series analysis that accentuate the need for parsimoniousmodels.The Auxiliary Information Set could be anything under our Sun or even from under other heavenly objects.A useful variable to include would be the blood pressure and heart rate time series of a representative groupof security traders.
As a brush up, the total slippage or implementation shortfall is derived below with the understandingthat we need to use the Expectation operator when we are working with estimates or future prices. (Kissell2006) provides more details including the formula where the portfolio may be partly executed. The list ofsymbols we use are,• ¯ S , the total number of shares that need to be traded.• T , the total duration of trading.• N , the number of trading intervals.• τ = T /N , the length of each trading interval. We assume the time intervals are of the same duration,but this can be relaxed quite easily. In continuous time, this becomes, N → ∞ , τ → .• The time then becomes divided into discrete intervals, t k = kτ, k = 0 , ..., N .• For simplicity, let time be measured in unit intervals giving, t = 1 , , ..., T .• S t , the number of shares acquired in period t at price P t .• P can be any reference price or benchmark used to measure the slippage. It is generally taken to bethe arrival price or the price at which the portfolio manager would like to complete the purchase of theportfolio.• Any trading trajectory, would look to formulate an optimal list of total pending shares, W , ..., W T +1 .Here, W t is the number of units that we still need to trade at time t . This would mean, W = ¯ S and16 T +1 = 0 implies that ¯ S must be executed by period T . Clearly, ¯ S = T (cid:80) j =1 S j . This can equivalently berepresented by the list of executions completed, S , ..., S T . Here, W t = W t − − S t − or S t − = W t − − W t is the number of units traded between times t − and t . W t and S t are related as below. W t = ¯ S − t − (cid:88) j =1 S j = T (cid:88) j = t S j , t = 1 , ..., T. Using the above notation, Paper Return = ¯ SP T − ¯ SP Real Portfolio Return = ¯ SP T − (cid:32) T (cid:88) t =1 S t P t (cid:33) Implementation Shortfall = Paper Return − Real Portfolio Return = (cid:32) T (cid:88) t =1 S t P t (cid:33) − ¯ SP The innovation we introduce would incorporate our earlier discussion about breaking the total impact orslippage, Implementation Shortfall, into the part from the participants own decision process, Market Impact,and the part from the decision process of all other participants, Market Timing. This Market Impact, wouldcapture the actions of the participant, since at each stage the penalty a participant incurs should only bethe price jump caused by their own trade and that is what any participant can hope to minimize. A subtlepoint is that the Market Impact portion need only be added up when new price levels are established. If theprice moves down and moves back up (after having gone up once earlier and having been already countedin the Impact), we need not consider the later moves in the Market Impact (and hence implicitly left outfrom the Market Timing as well). This alternate measure would only account for the net move in the pricesbut would not show the full extent of aggressiveness and the push and pull between market participants andhence is not considered here, though it can be useful to know and can be easily incorporated while runningsimulations. Our measure of the Market Impact, for a buy order, then becomes,Market Impact = T (cid:88) t =1 { max [( P t − P t − ) , S t } The Market Timing is then given by,Market Timing = Implementation Shortfall − Market Impact = (cid:32) T (cid:88) t =1 S t P t (cid:33) − ¯ SP − T (cid:88) t =1 { max [( P t − P t − ) , S t } .4 Deep Dive Results We perform many levels of comparisons and tests to gauge the impact of the changes. Our informationset for the deeper dive consists of two datasets. One is the daily close price, average spread, total volumeand total number of trades across each of the 100 securities starting from July-01-2013 to Dec-10-2014. Theother dataset comprises orders on these securities for the same time period. This information set containsall the standard order level information like number of shares, value, number of executions taken to fill theorder and also includes Market Impact, Timing and the Total Slippage. All the variables can be measuredbased on observations done on an exchange, since our study is structured as a self contained closed system,except for the inclusion of the FX rate, which is required to construct notional buckets in USD and helpsrelate to a broader audience and to facilitate inferences to be drawn easily. All the results are depicted usingsummary tables and graphical elements in section 10, Appendix - II.We first calculate the Equal Weighted, Volume Weighted and Trade Weighted Spread, Prices, the ratio ofthe Spread and Price and the Trade Size. We use the average spread, the close price, the total daily volumeand the total number of daily trades at the security level. The fall in the spread, both the average spread andthe ratio of the spread by the price, and the trade size around the two event days is easily seen in the timeseries graph (Figures 9, 10 and 11). The trend in the trade size is better inferred when we smooth it usinga ten day moving average filter, being conscious of the fact there will be a lag before we observe the valuesgoing down. We supplement all the individual variables with the 90 day moving volatility of each of thetime series. The volatilities of the spread moves upward around the two event days, but we cannot concludethat a new higher volatility level is established. When we consider the price volatility (Figure 10), it is notclearly evident that volatility has trended upwards. This is also not clearly established from the volatilitytime series at the security level, hence we do not report the security level volatilities. Both the raw valuesand filtered values for the volumes and the number of trades, do not show any discernible trend (Figure 12).A point to bear in mind is that these events would have a greater effect on intraday volatility and this issomething to be checked for in later studies. Greater volatility results in more efforts at managing a moreuncertain environment. In addition to affecting the Market Impact numbers, this would be reflected in theMarket Timing as well and hence in the overall Slippage numbers.We perform standard stationary tests on price, volume, spread and other variables at the individualsecurity level. We employ the Augmented Dickey-Fuller (ADF) Test, the KPSS test and Phillips-Perron(PP) test. The null hypothesis for the ADF and PP test is that there is a unit root against the alternate thatthe series is explosive or stationary. The KPSS null hypothesis is that the series is level or trend stationaryagainst the alternate that there is a unit root. It is easily apparent that total daily volume, daily averagespread, average volume, ratio of spread by price and the number trades are stationary. Prices, Inverse of thePrice and USD/JPY FX rates are not. We repeat these tests across the below six samples that we create18rom the overall dataset. We see that the first difference of the non stationary variables results in a stationarytime series. We report the count of securities with a p-value less than 0.05 in Figure 20. In all our regressions,we include the first difference of the variables which are non stationary.1. Sample Full , SF: The entire dataset, from Jul-01-2013 to Dec-10-2014.2. Sample One, S1: The start of the dataset to the first event, from Jul-01-2013 to Jan-10-2014.3. Sample Two, S2: The first event to the second event, from Jan-14-2014 to Jul-18-2014.4. Sample Three, S3: The second event to the end of the dataset, from Jul-22-2014 to Dec-10-2014.5. Sample Four, S4: From the start of the dataset to the second event, from Jul-01-2013 to Jul-18-2014.6. Sample Five, S5: From the first event to the end of dataset, from Jan-15-2014 to Dec-10-2014.Next, we fit a trend line with a non-zero intercept across each of the variables at the security level and countthe number of securities that show an increasing trend (Figure 21 summarizes the results for each of thevariables, z t ). This is also equivalent to checking a deterministic time trend in the variables as shown in (Eq:1). z t = β + β t + ε t (1)It is clearly seen from this that spreads have come down and the average volume has decreased consistentlyacross most of the names around the two events. To clarify an apparent divergence, the counts for thethird sample show a high number for increased spread, but this sample includes only days after the secondevent and the fall in spreads have occurred before the start of this sample. There is a jump in the spreadaround Oct-31-2014 (also the volume, number of trades and notational) which causes the spread trend linesto have increased slope in the last sample period, overcoming the effect of the earlier decrease on Jul-22-2014.Barring this outlier, which was caused by a sudden surge in prices, possibly attributable to the Bank ofJapan unexpectedly adding stimulus by targeting a $726 billion USD annual expansion in the central bank’smonetary base and the $1.2 trillion USD Government Pension Investment Fund announcing plans to morethan double its target allocation to Japanese stocks to 25 percent of assets, (See End-notes 3 and 4) theresults are consistent and as expected. To supplement the above six samples, whenever the sample includesthe last date Dec-10-2014, we run an extra set of regressions until Oct-30-2014, allowing us to judge theresults after removing the effects of this abnormal jump.The volume and number of trades trend is inconclusive just by looking at the slope of the time trend.Hence to assess this further, we run some regressions across each of the six samples. We run three sets of19egressions. In the first (Eq: 2; Figure 23 shows all the variables and summarizes the results), the volume isthe dependent variable, y t . Spread and Number of Trades are the key independent variables. In the second,we exclude the number of trades. The third regression is similar to the second except that we set the numberof trades as the dependent variable. y t = β + β t + β ln ( AvgBidAskSpread t ) + β ln (cid:18) Spread t P rice t (cid:19) + β ln ( T otalT rades t ) (2) + β ln ( U SDJP Y F irstDif f erence t ) + β ln ( CloseP riceF irstDif f t ) + β ln (cid:18) CloseP riceF irstDif f t (cid:19) + ε t (3)Here, the ratio of the spread by price and the inverse of the price act as control variables. We also includethe USD/JPY as an additional control variable. All the variables except the FX rate are significant. Thecorrelation matrix is in Figure 22. It is clearly seen from the regression coefficients (Figure 23) that thevolume and number of trades increase when spreads fall. We get similar results when we take the lag of theindependent variables by one day and by one week.Saving the best for last, we look at trading costs. We run separate regressions with all three of our costmetrics described earlier, Market Impact, Market Timing and Total Slippage as the independent variables.We find that, the results are similar across all three proxies of the trading cost, but the adjusted R-Squaredis higher with the Market Impact, M I t (Eq: 4; Figure 25 shows all the variables and summarizes the results).The real cost associated with a trader’s effort in seeking liquidity is given by the Market Impact, hence wereport and discuss only those results. We need to interpret the results keeping in mind that trading costsare notoriously difficult to predict, and models relating costs to other variables come with a high level ofvariance. We include a whole smattering of variables as the independent variables, including usual suspectssuch as spread, spread by price, inverse price, FX Rates, order notional, number of executions, moving 90 dayvolatilities of price, spread, volume, number of trades and the FX rate. The correlation matrix is in Figure 24.As additional control variables, we include the liquidity demanded by the order as a percentage of the totaldaily volume bucketed into five categories and the USD notional bucketed into four categories. We repeat theregressions for two different sets of categorizations of the Notional buckets (Figure 13). With such a setup,the orders in the smallest notional bucket, 0-1MM (million) USD, become the benchmark against which wemeasure the trading costs in the other buckets. The costs are on a decreasing trend from the beginning of oursample (Figures 14, 15 and 16). Hence, later studies should try to include explanatory variables to accountfor this phenomenon. We do not run these regression with a time lag since we are primarily interested in thecontemporaneous relationship between changes in the variables, but lag effects are not to be ruled out and20an be pursued later. M I t = β + β t + β ln ( AvgBidAskSpread t ) + β ln (cid:18) Spread t P rice t (cid:19) + β ln ( T otalV olume t ) (4) + β ln (90 DayM ovingCloseP riceV olatility t ) + β ln ( T otalV olumeV olatility t ) + ... + ε t (5)As primary evidence of increased trading costs, we see from the regression results (Figure 25) that thecoefficients are more negative in the 10MM+ notional buckets in the sample periods after either of the twochanges have happened, indicating that the costs in this bucket are higher relative to the other buckets. Forexample, in the S2, S3 and S5 samples, the 10MM+ bucket coefficients are higher by 5%, 2% and 0% ascompared to the 5-10MM bucket coefficients which are lower by 13%, 30% and 19% respectively. We canobserve this effect in the graph of the Market Impact by notional buckets over time. In the other set ofregressions with 25MM+ notional size categorization, for the S2, S3 and S5 samples, the 10-25MM bucketcoefficients are higher by 9%, 5% and 11% as compared to the 1-10MM buckets coefficients which are lowerby 29%, 7% and 14% respectively. The sample size is much smaller for the the 25MM+ bucket than the10MM+ buckets and hence the results are not as reliable, but we include it for completeness. Includinginteraction effects between the liquidity demand and notional buckets or excluding the notional buckets doesnot improve or change the results significantly.As secondary evidence of increased costs, we present daily 90-day moving volatilities on all three of tradingcosts metrics in Figures 17, 18 and 19. The higher volatility in finding liquidity is seen in the Market Impactvolatilities which has risen consistently since the changes. We need to keep in mind that the metric weare using is a moving 90 day volatility hence the actual effects of the change start to show up after a fewdays time. Also, the values near the start of the sample are not yet fully incorporating many days of dataand hence need to be overlooked. The Market Timing and Total Slippage can vary from positive to negativenumbers, hence to calculate the corresponding volatilities we cannot use continuous compounding and insteadwe use a 5 day moving average and the percentage difference between successive values. Because these valuescan fluctuate more widely than the Market Impact, the consistent increasing volatility pattern is not easilyinferred for these two variables. Further explorations using intraday data are required to establish whetherhigher volatility levels have been reached and could be one possible explanation for higher trading costs.In short, we have decreased spreads, decreased trading size, increased number of trades, increased volumeand increased trading costs for larger orders. The implications of this and to whom the immediate benefitswill accrue should be fairly obvious. Buying and selling smaller sizes more frequently can be less expensive,but buying (or selling) and holding (after holding) larger chunks of securities for longer periods of time mighthave become more costly. The aftereffects of the change are not exactly a win-win situation for everyone,there seem to be some winners and some losers. 21 Possibilities for a Deeper Dive
A key metric that would be useful to understand the effects of the change would be the intraday volatilitiesof the prices and execution sizes. Volatility is proportional to trading costs, hence, measuring intradayvolatility before and after the change could provide some answers to why costs have increased. When pricevolatility and execution size volatility increase, a trader faces a more uncertain environment. He has to factorin his decisions the possibility of the price and liquidity slipping away from him which results in higher overallcosts or uncertainty about costs, which is also costly. This happens through greater swings in the MarketImpact; in addition, greater movements in the Market Timing will cause higher overall Slippage numbers.Intraday data will allow the depth posted at the quotes to be analyzed and this could explain the reductionin the execution sizes and the increased difficulty in sourcing large orders. It would be interesting to trace thenumber of cancelled orders, quotes, the additional messages being relayed, changes to instructions and otherforms of noise, and the technology infrastructure being deployed to accommodate any additional processingburdens, both by the exchange and (if possible to estimate) across other participants. Another intendedeffect was to reduce the time to execution which can be measured using tick by tick data. Again, it is worthpondering the reasons why filling an order in 10 milliseconds or 50 milliseconds would make such a differenceto the loftier goal of providing a secondary market for the transfer of firm ownership and risk. For the deepdive, we have not considered the results by the securities affected during each event, since the results seemto hold strongly across the entire set. Checking this extra box might show other potentially interesting orunexpected outcomes. The sudden spread increase and surrounding market events on Oct-31-2014 are worthyof a closer inspection. This study has been performed as a completely closed system. To further this avenue ofapproach, having external control variables could help account for some of the cyclical or structural variationsin the variables and establish the trends strongly. This is particularly important for the trading costs whichare on a decreasing trend from the beginning of our sample and including explanatory variables could explainthis phenomenon. We have not included moving trading cost volatilities in any of our regressions, but thesecould be useful control variables since these change at a slower pace compared to the actual variable and pickup long term trends.
We conclude that one set of changes have happened as anticipated, with the reduction in the averagespread size. The unanticipated change is the reduction in the average execution size. We also see that thetotal volume and the number of trades have increased. The increased volume is not necessarily provingbeneficial in adding liquidity to all exchange participants. On one hand, investors who are trading smallerorder sizes are likely to experience a decreased cost of trading; on the other hand, large institutional investorstrading bigger orders, might require additional efforts to source in the liquidity to fill their trades; with the22et effect being that this additional effort might even lead to a slightly increased effective cost of trading.Once the dust from the last set of changes, which are yet to be implemented, settles down, supplementingthis study with more intraday indicators will go a long way towards determining conclusively which group ofinvestors will be the ultimate beneficiary. In our attempt to answer the question, “Does Tick Size Matter?”, weunequivocally find that, “Tick Size Does Matter”. The significant competition between trading mechanismsand venues, highlights the need for future research related to the consequences of tick size on trading costsand the dynamics of liquidity supply.
1. The following individuals have been a constant source of inputs and encouragement, more continuousthan the flow of orders in an extremely liquid venue: Brad Hunt, Henry Yegerman, Samuel Zou, AlexGillula and Ronald Ang at Markit; Dr. Isabel Yan, Dr. Yong Wang, Dr. Vikas Kakkar, Dr. Fred Kwan,Dr. Costel Daniel Andonie and Dr. Humphrey Tung at the City University of Hong Kong. The viewsand opinions expressed in this article, along with any mistakes, are mine alone and do not necessarilyreflect the official policy or position of either of my affiliations or any other agency.2. Sergey Nazarovich Bubka (born 4 December 1963) is a Ukrainian former pole vaulter. He representedthe Soviet Union until its dissolution in 1991. Sergey has also beaten his own record 14 times. Hewas the first pole vaulter to clear 6.0 metres and 6.10 metres. Bubka was twice named Athlete of theYear by Track & Field News and in 2012 was one of 24 athletes inducted as inaugural members of theInternational Association of Athletics Federations Hall of Fame. Sergey Bubka, Wikipedia Link3. Japanese stocks soared, with the Nikkei 225 Stock Average closing at a seven-year high, as the Bankof Japan unexpectedly boosted easing and the nation’s pension fund prepared to unveil new assetallocations. Japan’s Nikkei 225 Soars to Seven-Year High on BOJ, GPIF4. Japanese shares are swinging by the most on record after a double boost by the nation’s central bankand pension fund sent the Topix index to a six-year high two weeks after it entered a correction. Sell,Buy, Sell Again in 27 Days Amid Record Topix Swings
1. Ahn, H. J., Cao, C. Q., & Choe, H. (1998). Decimalization and competition among stock markets:Evidence from the Toronto Stock Exchange cross-listed securities. Journal of Financial Markets, 1(1),231-87.2. Ahn, H. J., Cai, J., Chan, K., & Hamao, Y. (2007). Tick size change and liquidity provision on theTokyo Stock Exchange. Journal of the Japanese and International Economies, 21(2), 173-194.3. Almgren, R., & Chriss, N. (2001). Optimal execution of portfolio transactions. Journal of Risk, 3, 5-40.4. Almgren, R. F. (2003). Optimal execution with nonlinear impact functions and trading-enhanced risk.Applied mathematical finance, 10(1), 1-18.5. Almgren, R., Thum, C., Hauptmann, E., & Li, H. (2005). Direct estimation of equity market impact.Risk, 18, 5752.6. Bessembinder, H. (1999). Trade execution costs on Nasdaq and the NYSE: A post-reform comparison.Journal of Financial and Quantitative Analysis, 34(3).7. Bessembinder, H. (2000). Tick size, spreads, and liquidity: An analysis of Nasdaq securities tradingnear ten dollars. Journal of Financial Intermediation, 9(3), 213-239.8. Bessembinder, H. (2003). Trade execution costs and market quality after decimalization. Journal ofFinancial and Quantitative Analysis, 38(04), 747-777.9. Bhide, A. (1993). The hidden costs of stock market liquidity. Journal of financial economics, 34(1),31-51.10. Bollen, N. P., & Busse, J. A. (2006). Tick size and institutional trading costs: Evidence from mutualfunds. Journal of Financial and Quantitative Analysis, 41(04), 915-937.11. Bourghelle, D., & Declerck, F. (2004). Why markets should not necessarily reduce the tick size. Journalof banking & finance, 28(2), 373-398.12. Caldarelli, G., Marsili, M., & Zhang, Y. C. (1997). A prototype model of stock exchange. EPL(Europhysics Letters), 40(5), 479.13. Chung, K. H., & Van Ness, R. A. (2001). Order handling rules, tick size, and the intraday pattern ofbid–ask spreads for Nasdaq stocks. Journal of Financial Markets, 4(2), 143-161.14. Collins, B. M., & Fabozzi, F. J. (1991). A methodology for measuring transaction costs. FinancialAnalysts Journal, 47(2), 27-36.15. Easley, D., & O’Hara, M. (2010). Microstructure and Ambiguity. Journal of Finance, 65(5), 1817-1846.16. Goldstein, M. A., & Kavajecz, K. A. (2000). Eighths, sixteenths, and market depth: changes in ticksize and liquidity provision on the NYSE. Journal of Financial Economics, 56(1), 125-149.247. Gujarati, D. N. (1995). Basic econometrics, 3rd. International Edition.18. Hamilton, J. D. (1994). Time series analysis (Vol. 2). Princeton university press.19. Harris, L. E. (1994). Minimum price variations, discrete bid-ask spreads, and quotation sizes. Reviewof Financial Studies, 7(1), 149-178.20. Huang, R. D., & Stoll, H. R. (2001). Tick size, bid-ask spreads, and market structure. Journal ofFinancial and Quantitative Analysis, 36(04), 503-522.21. Jones, C. M., & Lipson, M. L. (2001). Sixteenths: direct evidence on institutional execution costs.Journal of Financial Economics, 59(2), 253-278.22. Michie, R. (2001). The London stock exchange: A history. OUP Catalogue.23. Karolyi, G. A., & Stulz, R. M. (1996). Why do markets move together? An investigation of US-Japanstock return comovements. The Journal of Finance, 51(3), 951-986.24. Kashyap, R. (2014a). Dynamic Multi-Factor Bid-Offer Adjustment Model. Institutional Investor Jour-nals, Journal of Trading, 9(3), 42-55.25. Kashyap, R. (2014b). The Circle of Investment. International Journal of Economics and Finance, 6(5),244-263.26. Kashyap, R. (2015a). Financial Services, Economic Growth and Well-Being: A Four Pronged Study.Indian Journal of Finance, 9(1), 9-22.27. Kashyap, R. (2015b). Hong Kong - Shanghai Connect / Hong Kong - Beijing Disconnect (?). SocialScience Research Network (SSRN), Working Paper.28. Kim, K. A., & Rhee, S. (1997). Price limit performance: evidence from the Tokyo Stock Exchange.The Journal of Finance, 52(2), 885-901.29. Kissell, R. (2006). The expanded implementation shortfall: Understanding transaction cost compo-nents. The Journal of Trading, 1(3), 6-16.30. Madhavan, A. (2000). Market microstructure: A survey. Journal of financial markets, 3(3), 205-258.31. Perold, A. F. (1988). The implementation shortfall: Paper versus reality. The Journal of PortfolioManagement, 14(3), 4-9.32. Roll, R. (1984). A simple implicit measure of the effective bid-ask spread in an efficient market. TheJournal of Finance, 39(4), 1127-1139. 253. Stoll, H. R. (1989). Inferring the components of the bid-ask spread: theory and empirical tests. TheJournal of Finance, 44(1), 115-134.34. Treynor, J. L. (1981). What does it take to win the trading game?. Financial Analysts Journal, 37(1),55-60.35. Treynor, J. L. (1994). The invisible costs of trading. The Journal of Portfolio Management, 21(1),71-78.36. Venkataraman, K. (2001). Automated versus floor trading: An analysis of execution costs on the Parisand New York exchanges. The Journal of Finance, 56(4), 1445-1485.37. Ye, M., Yao, C., & Gai, J. (2012). The externalities of high frequency trading. SSRN Working Paper:http://papers.ssrn.com/abstract_id=206683938. Yegerman, H. & Gillula, A. (2014). The Use and Abuse of Implementation Shortfall. Markit WorkingPaper.
Figure 2: Bird’s Eye View Comparison Summary for Jan 14, 2014Figure 3: Bird’s Eye View Comparison Summary for Jul 22, 201426igure 4: Stock Split Ratios During Study Time Period27igure 5: Bird’s Eye View Spread Comparison Detail for Jan 14, 201428igure 6: Bird’s Eye View Execution Size Comparison Detail for Jan 14, 201429igure 7: Bird’s Eye View Spread Comparison Detail for Jul 22, 201430igure 8: Bird’s Eye View Execution Size Comparison Detail for Jul 22, 2014310 Appendix - II (Deep Dive Comparisons)