econstor Make Your Publications Visible.

Transcription

1 econstor Make Your Publcatons Vsble. A Servce of Wrtschaft Centre zbwlebnz-informatonszentrum Economcs Bond, Steve; Hawkns, Mke; Klemm, Alexander Workng Paper Stamp duty on shares and ts effect on share prces IFS Workng Papers, Insttute for Fscal Studes (IFS), No. 04/11 Provded n Cooperaton wth: Insttute for Fscal Studes (IFS), London Suggested Ctaton: Bond, Steve; Hawkns, Mke; Klemm, Alexander (004) : Stamp duty on shares and ts effect on share prces, IFS Workng Papers, Insttute for Fscal Studes (IFS), No. 04/11, Insttute for Fscal Studes (IFS), London, Ths Verson s avalable at: Standard-Nutzungsbedngungen: e okumente auf EconStor dürfen zu egenen wssenschaftlchen Zwecken und zum Prvatgebrauch gespechert und kopert werden. Se dürfen de okumente ncht für öffentlche oder kommerzelle Zwecke vervelfältgen, öffentlch ausstellen, öffentlch zugänglch machen, vertreben oder anderwetg nutzen. Sofern de Verfasser de okumente unter Open-Content-Lzenzen (nsbesondere CC-Lzenzen) zur Verfügung gestellt haben sollten, gelten abwechend von desen Nutzungsbedngungen de n der dort genannten Lzenz gewährten Nutzungsrechte. Terms of use: ocuments n EconStor may be saved and coped for your personal and scholarly purposes. You are not to copy documents for publc or commercal purposes, to exhbt the documents publcly, to make them publcly avalable on the nternet, or to dstrbute or otherwse use the documents n publc. If the documents have been made avalable under an Open Content Lcence (especally Creatve Commons Lcences), you may exercse further usage rghts as specfed n the ndcated lcence.

2 STAMP UTY ON SHARES AN ITS EFFECT ON SHARE PRICES Steve Bond Mke Hawkns Alexander Klemm THE INSTITUTE FOR FISCAL STUIES WP04/11

3 STAMP UTY ON SHARES AN ITS EFFECT ON SHARE PRICES Steve Bond (IFS and Unversty of Oxford) Mke Hawkns (IFS) Alexander Klemm (IFS and UCL)* June 004 Abstract: Ths paper provdes a dscusson of stamp duty and ts effects. Ths s followed by an emprcal study usng changes n the rate of stamp duty n the UK as natural experments. Because shares wll be affected dfferently dependng on how frequently they are traded, we can employ a dfference-n-dfferences methodology. We fnd that the announcements of cuts n stamp duty had a sgnfcant and postve effect on the prce of more frequently traded shares compared to other shares. As expected under the effcent markets hypothess, the mplementaton of cuts (when at a dfferent date from the announcement) dd not affect returns dfferentally. Key words: Stamp duty, transacton tax, Tobn-tax, natural experment, tax reform. JEL classfcaton: G14, H9, E6. Acknowledgments: Ths research s part of the Large Busness Taxaton Programme at the IFS, funded by the 100 Group and the Inland Revenue. We would lke to thank Paula Landow, Julan McCrae and semnar partcpants at an Inland Revenue semnar, especally Stephen Matthews, for comments and advce. Any remanng errors are solely the responsblty of the authors. * Correspondng author: e-mal: a.klemm@fs.org.uk, postal address: Insttute for Fscal Studes, 7 Rdgmount Street, London WC1E 7AE, UK.

4 Executve Summary Stamp duty s a tax on share transactons n UK ncorporated companes, currently leved at ½% of the purchase prce of shares. Ths rate has been changed over the years, most recently n 1986 when t was cut from 1% to ts current rate. In the 1990 Budget the Conservatve government of the tme announced ts abolton to concde wth the ntroducton of the London Stock Exchange s new settlements system, Taurus. However, ths was never mplemented and stamp duty remans. ebate contnues about whether stamp duty on shares should be abolshed, and t s therefore useful to examne some emprcal evdence on ts effects. If stamp duty s captalsed nto share prces, t s expected to depress prces more for shares that are more frequently traded, and where future stamp duty payments are expected to be hgher at any gven rate. Past announcements of cuts n the rate of stamp duty allow us to test ths predcton, by studyng the effect of these announcements on share prces for groups of frms wth dfferent levels of tradng. ata on tradng volumes n the 1990s shows that frm sze s a good predctor of share turnover rates. As ths data on tradng volumes s not avalable pror to 1986, our man results use frm sze as a proxy for turnover. We also use data on actual turnover rates to check our results for the 1990 announcement of abolton. We fnd evdence that stamp duty ndeed has a detrmental effect on share prces. Specfcally, the prce of shares that are more frequently traded ncreases relatvely to that of shares that are less frequently traded on the announcement dates of cuts n stamp duty n 1984, 1986 and The analyss of the events n 1986 s of partcular nterest, because n that year the actual cut took place around sx months after the announcement. We fnd a sgnfcant effect only on the announcement date, not on the date of the mplementaton. Ths s the expected result n an effcent market, and provdes some support for our emprcal approach. Stamp duty s thus shown to depress share prces, partcularly for frms whose shares are frequently traded. Ths may ncrease the cost of captal faced by frms, whch n turn could have negatve repercussons on nvestment. Stamp duty also dstorts the sgnals that share prces send about the proftablty of frms, as share prces are also affected by expectatons of future turnover volumes and stamp duty rates. Our results show that these effects are real and measurable. Ths fndng s mportant both n dscusson about the benefts of abolshng stamp duty, as well as for the wder debate on the merts of transacton taxes, ncludng the controversal Tobn tax. It remans open, however, whether the negatve effects of alternatve sources of tax revenue would be smaller than those of stamp duty.

5 1 Introducton Ths paper nvestgates the mpact of UK stamp duty on share prces. Beng a transacton tax, stamp duty on shares s controversal. Research on the effects of transacton taxes s therefore polcy relevant. Not only because of the potentally large effect on stock markets, but also because of the wder dscusson about the merts of ntroducng a transacton tax on nternatonal currency transactons (the Tobn tax ). 1 For all the theoretcal arguments on the effects of stamp duty, there s surprsngly lttle emprcal evdence on the effects of stamp duty on share prces and turnover. The lttle that exsts s summarsed n secton 3.. Our approach s to use changes n the rate of stamp duty as natural experments. In partcular, we use a dfference-n-dfferences methodology, comparng the dfferental mpact on the prce of low turnover and hgh turnover shares of the announcements and mplementatons of stamp duty reductons n 1984, 1986 and Ths paper s structured as follows. The next secton descrbes the background of the current stamp duty regme and the changes made to t. Secton 3 descrbes the theory of a transacton tax and brefly revews the emprcal evdence. Secton 4 explans our emprcal strategy to dentfy effects on share prces usng tax reforms. Secton 5 presents the results and secton 6 concludes. An appendx dscusses the effect of market movements on our results, and whether usng abnormal returns would be approprate n ths context. Background Stamp duty s a tax on share transactons n UK ncorporated companes, currently leved at ½% of the purchase prce of shares. It s chargeable whether the transacton takes place n the UK or 1 The Tobn tax was frst suggested n Tobn (1974). Strctly, stamp duty s chargeable on the purchase prce of a share where there s a legal nstrument of transfer. Ths accounted for around 10% of total revenue from share transactons n The remanng revenue was collected through stamp duty reserve tax (SRT), whch s the equvalent tax on an agreement to transfer the share where there s no wrtten nstrument of transfer. Snce the ntroducton of an electronc settlements system, CREST, n 1996, SRT has taken over as the man tax on share transactons. 3

6 overseas, and whether ether party s resdent n the UK or not. It s not chargeable on securtes ssued by companes ncorporated overseas. The rate of stamp duty has changed over the years. Before 1974 t stood at 1%. Then t was ncreased to %. In the 1984 Budget t was reduced to 1% agan. In 1986 t was further reduced to ts current rate of ½%. In the 1990 Budget the Conservatve government of the tme announced that stamp duty on shares would be abolshed wth the ntroducton of the London Stock Exchange s new settlements system, Taurus. However, n the end Taurus was abandoned n March 1993 and the abolton was never mplemented. One of the outstandng features of stamp duty s that t s the cheapest of all UK taxes to collect, wth a collecton cost of just 0.11 pence per pound rased. For comparson, the correspondng fgure for ncome tax, the most mportant revenue raser, s 1.59 pence (Inland Revenue, 00). As the fgure reported for stamp duty ncludes the cost of collectng stamp duty from land and property purchases, t s lkely that the correspondng fgure for stamp duty on share transactons s even lower, gven that most transactons on the stock exchange are now electronc and stamp duty can thus be deducted automatcally. The amount of revenue rased vares from year to year around an average of about 3bn, 3 whch corresponds roughly to one tenth of that rased by corporaton tax. Whle stamp duty s n law leved on the purchaser of a share n a UK regstered company, the true economc ncdence of the tax, and therefore ts effect on the economy, s ambguous. In the long run, ndvduals wll bear the tax ether n lower returns on ther savngs, hgher product prces due to lower nvestment, or a combnaton of the two. At the moment of ntroducton or unantcpated ncrease, the tax may fall predomnantly on exstng shareholders f share prces adjust to delver the same expected post-tax returns as are avalable on substtute assets that are not subject to stamp duty. Therefore the benefcares of a cut n stamp duty are also most lkely to be the current owners of shares that are lable to stamp duty. 3 Accordng to Inland Revenue Statstcs, stamp duty on shares rased the followng amounts durng the last fve years:.5bn (1998/99), 3.7bn (1999/000), 4.5bn (000/01),.9bn (001/0) and.5bn (00/03). 4

7 In recent years there has been ncreasng pressure from the London Stock Exchange and others for the abolton of stamp duty. 4 One of the prncpal arguments used to support ts abolton s that stamp duty rases the cost of captal for UK frms and therefore reduces nvestment. In ts calculatons for the London Stock Exchange, OXERA (001) estmated that stamp duty rases the cost of captal by 0.7 to 0.87 percentage ponts. Ths s calculated by dervng the mpled mpact of stamp duty on share prces and therefore company dscount rates. In practce, the effect on nvestment may depend on the margnal source of fnance for dfferent types of frms. For frms whose margnal source of fnance s retaned earnngs or new equty, we would ndeed expect stamp duty to rase the cost of captal. Ths s because t s leved on the share prce, whch ncludes the captalsed value of retaned profts that have not yet been pad out to shareholders. Hawkns and McCrae (00) calculate that stamp duty at ts current rate could add between 0.15 and 0.65 percentage ponts to the cost of retaned earnngs fnance, dependng on the frequency wth whch a company s shares are bought and sold. For other frms,.e. those whose margnal source of fnance s debt, the effect on ther cost of captal may be less mportant. 3 Economc effects of securtes transactons taxes The potental effects of stamp duty are wde-rangng. It may affect share prces, as nvestors wll prce n future payments of ths transacton tax. It may also affect tradng volumes, as the dfference between the valuaton of a prospectve buyer and seller needs to be larger to cover not only general transacton costs, but also the amount of stamp duty. 5 Furthermore there may be effects on volatlty, whch could be reduced f the latter effect prevented tradng when valuatons change only margnally, although t may also ncrease, because adjustments wll take 4 E.g. London Stock Exchange Urgent acton on Stamp uty needed to boost the economy Press Release 4 March 003. See also Stamp Out Stamp uty Campagn at 5 Emprcal evdence has consstently found such a negatve relatonshp, see Jackson and O onnell (1985), Lndgren and Westlund (1990), Ercsson and Lndgren (199), Atken and Swan (000) and Swan and Westerholm (001) 5

8 place more suddenly. 6 Whle all of these effects may be mportant, the focus of ths paper s on the effect on the share prce. 3.1 Theoretcal mpact of a transactons tax on share prces Accordng to the standard dvdend valuaton model, 7 the equlbrum prce of a share should be equal to the expected present dscounted value of future dstrbutons after all taxes. In the specal case where the dvdend per share grows at the constant rate g and the dscount rate r s also constant, the prce P s gven n the absence of transacton taxes by: ( 1+ g) ( 1+ r) ( 1+ g) + = 3 ( 1+ r) r g P = r where s the dvdend per share pad at the end of the current perod, and r - g s the dvdend yeld. Stamp duty can be ncorporated nto ths smple model by assumng a fxed number of transactons t per perod, so that the prce n the presence of stamp duty becomes: = r g + st ( stp)( 1+ g) ( 1+ r) ( stp)( 1+ g) 3 ( 1+ r) stp P = = 1+ r stp r g where s s the rate of stamp duty. To calculate the effect of an unantcpated permanent change n the rate of stamp duty, we make the addtonal assumptons that the change n the stamp duty rate does not affect turnover, the dscount rate or the growth rate of dvdends. The proportonal change n prce mpled by a P P ( s s ) t change n the rate of stamp duty from s to s s then gven by: =. P r g + s t 6 Emprcal evdence on the effect on volatlty has been rather mxed. Umlauf (1993) and Saporta and Kan (1997) found no effect, Atken and Swan (000) and Swan and Westerholm (001) found that transactons taxes tend to ncrease volatlty and Hau and Chevaller (000) found that they slghtly reduce t. 7 See, e.g. Brealey and Myers (000). 6

9 t In the specal case of the abolton of stamp duty (s = 0), ths smplfes to s and the r g assumpton that stamp duty has no effect on share turnover becomes unnecessary. Table 1 shows the ncrease n share prces predcted by ths smple model f the current stamp duty rate of 0.5% were to be abolshed, for a range of shares wth dfferent turnover rates and dvdend yelds. The hgher the frequency of transactons, the more stamp duty would be saved n the future by shareholders as a result of ts abolton, and the hgher s the predcted ncrease n the current share prce. Ths basc predcton holds for cuts n the rate of stamp duty more generally, provded the turnover rate ncreases less than proportonately as the rate of stamp duty s reduced. 8 Ths s the key predcton that we test n our emprcal analyss. Table 1: Predcted share prce mpact of abolshng stamp duty vdend Turnover yeld 10% 30% 50% %.5% 7.5% 1.5% 3% 1.7% 5.0% 8.3% 4% 1.3% 3.8% 6.3% In practce, there are a number of reasons why the observed prce mpact of abolshng stamp duty could dffer from that predcted by ths smple model. For nstance, nvestors may not expect the abolton to be permanent, whch would reduce the predcted mpact on prce. In cases where the rate s cut rather than abolshed, an ncrease n turnover would partally offset the effect of the lower tax rate on prce, although ths n turn may produce an offsettng effect through market lqudty. That s, nvestors may requre a lower rsk premum from shares that are traded n a more lqud market, whch would tend to reduce the dscount rate, r. We would also expect many factors other than the rate of stamp duty to affect a company s share prce. In prncple, any news that affected nvestors expectatons of future profts, or ther expected post-tax return on other assets (real dscount rate), would potentally change the share prce. Examples of such news mght nclude announcements of recent profts or updated analysts forecasts of future profts, changes n macroeconomc polcy that may affect future real 8 That s, provded a 50% reducton n the rate of stamp duty, from say 0.5% to 0.5%, would not produce a 50% ncrease n the number of transactons. 7

10 nterest rates or the volatlty of nflaton, or a new nventon that changed the relatve cost of a company or opened up a new market for ts products. The emprcal challenge s to dentfy the effect of a stamp duty rate change gven that there are so many other nfluences on share prces. 3. Revew of emprcal evdence The followng provdes a bref revew of both UK and nternatonal emprcal work on the effects of stamp duty or equvalent foregn transactons taxes Evdence from the UK espte the changes n stamp duty tax rates and the debate surroundng them, there s not very much UK emprcal work. Jackson and O onnell (1985) estmate the mpact of transacton costs on average real share prces over the perod usng aggregate annual data for the UK stock market. They fnd an elastcty of share prce wth respect to transacton costs of 0.3. If true, ths mples that the 1984 reducton n the stamp duty rate from % to 1% would have led to an 8% rse n share prces. 9 Saporta and Kan (1997) look at the share prce mpact of changes n the rates of stamp duty by observng share prce movements on the days that the last three rate changes were announced n the UK n 1974, 1984 and They fnd that share prces moved n the expected drecton and by a statstcally sgnfcant amount. However, as noted by Saporta and Kan, the change cannot be drectly attrbuted to changes n stamp duty, because other nformaton may have been httng the market on the same day, partcularly as the announcements were made on budget days Jackson and O onnell estmate that other transacton costs were 0.75% of transacton values. Ths mples that total transacton costs were.75% pror to the stamp duty cut n The effect of a 1 percentage pont cut n stamp duty on prces can then be calculated as the share prce elastcty tmes the percentage change n transacton costs,.e. 0.3 (1%/.75%)= 8.4%. 10 They also try to learn somethng from comparng the mpact on shares traded n the UK and stamp duty free Amercan epostory Recepts (AR). Ths s made dffcult by arbtrage, whch wll lead to smlar prces between the two assets. Nevertheless ther results are as expected (.e. returns on ARs are hgher after announcements of tax cuts) but ths dfference s not sgnfcant. Because of very small sample szes (between 4 and 11 frms) they should anyway be treated wth cauton. 8

11 3.. Overseas evdence Related evdence has also been presented for other countres. Umlauf (1993) employs a smlar methodology to that n Saporta and Kan to consder the effect of the mposton of a 1% tax on Swedsh share transactons n He fnds that the share prce ndex declned by 5.3% durng the month of the announcement, compared wth a predcted 6.75% declne. However, hs paper s subject to the same crtcsm as Saporta and Kan s namely, that t fals to control for other possble nfluences on share prces such as nterest rate changes and other tax changes. Swan and Westerholm (001) attempt to dentfy the share prce mpact of the abolton of securtes transacton taxes n Sweden and Fnland n 1991 and 199 respectvely. In order to allow ncreases n turnover due to lower transacton costs to have an addtonal postve mpact on share prces through mprovements n lqudty, they frst model the effects on share turnover of changes n transacton costs due to tax changes. They use pooled daly data for roughly a twoyear perod ether sde of the tax change for 30 Fnnsh stocks and 80 Swedsh stocks. They attempt to correct for other nfluences on share prces by ncludng control varables such as nterest rates and exchange rates. They also estmate equvalent aggregate equatons for the whole market and separate equatons for dfferent sub-samples of stocks. Ths allows them to test whether transacton taxes mpact dfferentally on turnover of stocks n small and large companes, whch may dffer n terms of lqudty and other transacton costs. They estmate that the elastcty of share prces wth respect to transacton costs s 0.0 for Sweden and 0.1 for Fnland. These are smlar magntudes to those found by Jackson and O onnell for the UK. If the true elastcty for the UK s around 0. and transacton costs are between 1% and 3% of transacton values, ths mples that a 0.5% cut n stamp duty would ncrease share prces by between 3.5% and 14% a smlar range to that presented n Table 1. 4 Methodology We follow a dfferent approach to the exstng lterature by usng a dfference n dfferences approach, n whch we use changes n stamp duty rates over tme, and announcements thereof, as natural experments. Specfcally, from the smple model of share prces outlned above, we would expect the prce mpact of a gven change n the rate of stamp duty to be greater for stocks 9

12 that have hgh turnover than for those wth low turnover, other thngs beng equal. By splttng stocks nto hgh and low turnover categores, n prncple ths enables us to test whether stamp duty does ndeed affect share prces n the way predcted by the model. 4.1 Econometrc specfcaton The approach that we start wth s a basc dfference-n-dfferences specfcaton, as follows: r = α + β 1 + β + u t 1 t where r t s the rate of return on a share at tme t, random error. 1 and are dummy varables and u t s a There may be more than two dummes, but they wll always be defned n the same manner: whenever 1 equals one, wll also equal one,.e. determnes the addtonal effect of beng n group for frms that are already n group 1. E.g. when frms are grouped accordng to sze, 1 wll be one for all frms above a certan sze and even hgher sze threshold. wll be one for all frms above an The coeffcents are then nterpreted as follows. α wll be the average return on all shares not n a group defned by one of the dummes. β 1 gves the addtonal return on shares, whch are n that group defned by 1, compared to the control group of shares not defned by ether of the dummes. β wll gve the addtonal return on shares that are n the group defned by only, relatve to all the shares n the group defned by 1. The sum of β 1 and β would thus be the addtonal return of shares n the group defned by relatve to shares not n We measure r t as the cumulatve total return over the perod (event wndow). Ths ncludes both dvdends and captal gans, although n most cases there are no dvdend payments durng the relatvely short event wndows

13 4. Events consdered The four events that we use as natural experments are the three announcements of stamp duty cuts n the Budgets of 1984, 1986 and 1990, and the mplementaton of the 1986 cut sx months after ts announcement. If the stock market were reasonably effcent, n the sense that new nformaton about or affectng a frm s valuaton s reflected n ts share prce soon after t s revealed to the market, we would expect the man effect of a stamp duty cut on share prces to take place mmedately after the announcement date. There should be no effect on the date of the mplementaton. In practce, the dates of announcement and mplementaton were the same n In 1986, the rate cut was announced before t was mplemented although there was some uncertanty surroundng the precse date of mplementaton, as t was to concde wth the Bg Bang. 11 The fnal event consdered was John Major s announcement n Budget 1990 that stamp duty would be abolshed wth the ntroducton of Taurus. 1 Ths proposal was never actually mplemented, as the Taurus project was eventually abandoned. Hence, f the ntal announcement had any credblty we would expect share prces to adjust n March 1990, after whch pont the effect would presumably unwnd as t became clear to market partcpants that Taurus would never become operatonal. The detals of the three announcements and the subsequent mplementatons of the rate cuts are summarsed n Table below. Table : Key dates and detals of stamp duty rate changes Announcement date Implementaton date Actual mplementaton date Old rate New rate announced 13 March March March 1984 % 1% 18 March 1986 Autumn October % ½% 0 March 1990 Late n 1991/9 (1) Not mplemented ½% 0% (1) The Budget costng assumed abolton from 1 January199. Around each announcement date we consder event wndows wth dfferent start and end dates and varyng duratons. Budget speeches were delvered to Parlament n the afternoon on the dates n queston. We focus on an event wndow commencng at close of tradng the day before 11 The date from whch share transactons on the London Stock Exchange would be settled usng CREST. 1 Taurus was the ntended replacement for the Talsman settlements system. It was abandoned n March

14 the announcement. 13 It s not clear whether market partcpants would have had suffcent tme to react fully to the announcement on the day n queston. Allowng a longer tme perod for the market to react would ncrease the lkelhood of pckng up the full prce mpact of the stamp duty announcement n our data, but would at the same tme ncrease the rsk that we would also be pckng up the effects of extraneous tems of news completely unrelated to stamp duty. We consder event wndows stretchng from one to up to fve tradng days after the announcement. 4.3 Identfcaton of low- and hgh-turnover stocks atastream provdes daly data on the volume of shares traded after Bg Bang (October 1986) for stocks where, on average, 000 or more shares are traded daly. Snce three of the four events that we are consderng took place before or on the same date as Bg Bang, we are unable to use ths turnover data to group stocks by actual turnover n those cases. In our man results we therefore use market captalsaton as a predctor of turnover. To justfy ths methodology we examne the relatonshp between tradng volumes and market value durng the perod n whch we do have data. We defne a turnover rate as the fracton of outstandng shares traded n a year. Frst, we rank stocks where we have both market captalsaton and turnover data by market captalsaton n each year from 1991 onwards (the perod where the sample sze exceeds 500). 14 Fgure 1 plots the average annual turnover rates for four groups of stocks grouped by market captalsaton. 15 It shows that the average turnover rates of stocks n groups wth hgher market captalsaton s consstently hgher than that of groups wth lower market captalsaton. 13 It s possble that rumours may have leaked out pror to the announcement, but ther credblty would have been suspect. 14 For producng these charts we have dropped any frms that had less than 50 observatons per year or less than 4 years of data. Frms wth tradng ratos less than 1% per year are lkely to have unrelable prce date, because of an llqud market, and were also dropped. Fnally we dropped frms wth tradng ratos n excess of 00%, because of doubts over the relablty of ther data. 15 The fgure shows the unweghted average. A very smlar chart s obtaned f averages are weghted by market value, except that n the frst three years, the weghted average turnover rate of the top 100 frms s slghtly below that for the followng group (101 st to 50 th frm). A fracton of shares traded equal to 1 (or a tradng rato of 100%), ndcates that the annual number of transactons s equal to the number of shares outstandng. 1

15 Fracton of shares traded n year year Largest 100 frms 50th to 500th frm 100th to 50th frm Rest Mean turnover rates Fgure 1: Average turnover of stocks n dfferent sze classes n a gven year In order to avod problems of endogenety, we also show that rankng by market captalsaton n perod t-1 s a good predctor of a stock s rankng by share turnover n perod t. Table 3 shows that market captalsaton s a good predctor of turnover n the 1990 s, both for the current perod and for one perod ahead. Table 3: Rankng by sze n year t-1 as a predctor of rankng by turnover n year t-1 and t Market captalsaton at t-1 Average annual turnover at t-1 Average annual turnover at t top (.011).509 (.010) (.007).45 (.007) (.005).35 (.005) Rest.67 (.004).75 (.004) Notes: Robust standard errors gven n parentheses. An alternatve way to dentfy hgh-turnover stocks would be to use turnover data from the 1990s to allocate frms to groups and then to assume that turnover rates reman constant over tme. 13

16 Table 4 shows the correlaton between turnover rates (turnover / market value) at dfferent tme ntervals. We see that there s a hgh correlaton between adjacent perods, but ths correlaton becomes lower as we consder perods that are farther apart. Ths suggests that usng turnover data from later years may work for the 1990 announcement of the abolton of stamp duty, but t cannot be justfed for any of the earler changes. Table 4: Correlaton between stock turnovers at varous dates 16 t t-1 t- t-3 t-4 t-5 t-6 t-7 t 1.00 t t t t t t t Our preferred ndcator of turnover therefore s the average market captalsaton n the perod precedng the event, whch appears to be a good predctor of turnover n the current perod durng the 1990 s. Unfortunately we are unable to test whether the same s true for the 1980 s because the data are ether unavalable pror to 1991 or scarce. 17 Nevertheless, we proceed on the bass that the relatonshp between market captalsaton and turnover durng the 1980 s was smlar to that observed n the 1990 s. If ths were not the case, and sze were a less good predctor of turnover n the 1980 s, then f anythng our results would tend to be based aganst fndng a dfferental mpact of stamp duty cuts on the market value of hgh turnover stocks. Ths s because each sze groupng would contan a mxture of low- and hgh-turnover stocks, basng the estmated coeffcent on stocks classfed as hgh turnover towards zero. 16 Each correlaton was calculated from the same sample of 1531 observatons, therefore the coeffcents are comparable. 17 Some data are avalable from 1986 onwards, but for less than 500 companes, whch s not suffcent for our approach of groupng frms nto four sze bands. 14

17 5 Emprcal results Ths secton summarses the man results for each event referred to earler. We frst present the results usng sze as a proxy for turnover rates, as ths approach s possble for all the dates we consder. We then show, for the 1990 announcement, addtonal results usng the actual turnover data. 5.1 Results usng sze as a proxy Table 5 shows the results from regressng the cumulatve return from the day before to the day after the event on sze dummes. These dummes are defned as follows: The dummy labelled largest 500 s one for each frm that belongs to the largest 500 frms of the sample. A postve co-effcent thus ndcates that the largest 500 frms experenced hgher returns over the relevant perod than the smaller ones. The next dummy, largest 50, equals one, for frms that are among the 50 largest ones. The coeffcent thus tells us how much larger (or smaller) the returns of the 50 largest frms are compared to the group of the 500 largest. The remanng dummy largest 100 s defned smlarly. Table 6 shows the same regressons but for a cumulatve return over a longer tme nterval, from the day before the reform to fve days after. 15

18 Table 5: Results of regressons of two-day return (t-1 to t+1) on sze bands. ate of stamp duty change announcement / mplementaton ep var: Return 13 March 1984 (announcement and 18 March 1986 (announcement) 7 October 1986 (mplementaton 0 March 1990 (announcement) mplementaton) date) (1) () (3) (4) Largest (0.004) (0.004) (0.003) (0.00)** Largest (0.003) (0.003)** (0.00) (0.00)*** Largest (0.00)*** (0.00)*** (0.00)* (0.001) Constant (0.001)*** (0.001)*** (0.001)*** (0.001)*** Observatons R-squared F-statstc Robust standard errors n parentheses * sgnfcant at 10%; ** sgnfcant at 5%; *** sgnfcant at 1% Table 6: Results of regressons of one-week return (t-1 to t+5) on sze bands. ate of stamp duty change announcement / mplementaton ep var: Return 13 March 1984 (announcement and 18 March 1986 (announcement) 7 October 1986 (mplementaton 0 March 1990 (announcement) mplementaton) date) Largest (0.007)** (0.007) (0.008) (0.005)* Largest (0.005) (0.006) (0.007) (0.004)** Largest (0.003)*** (0.004)** (0.005) (0.00) Constant (0.00)*** (0.00)*** (0.00)*** (0.001)*** Observatons R-squared F statstc Robust standard errors n parentheses * sgnfcant at 10%; ** sgnfcant at 5%; *** sgnfcant at 1% The results are broadly as predcted by theory. We fnd that the shares of larger frms, whch are on average more frequently traded, earn hgher returns than those of smaller frms, at the tme that reductons are announced, though not on the mplementaton date of the 1986 rate cut. The precse pattern across sze bands s not always the same. ependng on the event and on the event wndow, dfferent sze dummes may have sgnfcantly postve coeffcents. Generally the 16

19 dfference between the largest 500 frms and the remanng frms tends to be more sgnfcant than dfferences wthn the group of the largest 500 frms, except n 1990 when the greatest effect was seen for the largest 50 frms. The general absence of dfferental prce mpacts around mplementaton date of the 1986 rate cut s reassurng, because, under the assumpton of effcent markets, the beneft of a lower stamp duty rate should have been prced n at the tme of announcement. The results for the 1990 announcement are also nterestng. We fnd a strong prce mpact n the expected drecton, suggestng that the market consdered the announcement of the abolton to be credble. 5. Results usng share turnover volume data As mentoned above, turnover volume data are generally avalable from 1991 onwards. For some frms, however, wth average turnover n excess of 000 shares per day, they become avalable from late 1986 onwards. It s therefore possble to use these data drectly to allocate frms nto turnover bands, at least for the 1990 event. The data are scarce though: we have suffcent volume nformaton for 166 frms only. These are all relatvely large frms. 96 of the 100 largest frms are ncluded n ths group and more than three quarters of these 166 frms are among the largest 150 frms. Gven the lmted number of observatons on volume data, we splt these frms nto only two groups at the medan turnover rate (defned as turnover dvded by market value) wthn ths sample. As well as comparng the returns of the most heavly traded frms to the remanng frms wth tradng volume data, we can also compare the return of all frms wth volume data to those wthout, because we know that only frms wth frequently traded shares wll have volume data for ths perod. The results of regressons of cumulatve returns over two days and one week are shown n columns (1) and () of Table 7. 17

20 Table 7: Results of regressons of cumulatve return on turnover band around March 1990 announcement. (1) () (3) (4) turnover rate > medan Cumulatve return Cumulatve return Cumulatve Cumulatve (t-1 to t+1) (t-1 to t+5) abnormal ret. abnormal ret. (t-1 to t+1) (t-1 to t+5) (0.003) (0.006) (0.003) (0.006) has turnover data (0.00)** (0.005)** (0.00)** (0.004)** Constant (0.001)*** (0.001)*** (0.001)*** (0.001)*** Observatons R-squared F-statstc Robust standard errors n parentheses * sgnfcant at 10%; ** sgnfcant at 5%; *** sgnfcant at 1% These results confrm that frms whose shares have hgher turnover rates benefted from relatvely hgher rates of return around the March 1990 announcement of the abolton of stamp duty. The dfference wthn the group of hghly traded frms wth volume data s small compared to the dfference between all hghly traded frms and the rest of the market. Another experment that we can do wth the volume data s to repeat the regresson usng abnormal returns rather than total returns as the dependent varable. Whle ths was not approprate when usng sze as the groupng varable (see Appendx), there s less reason why ths should be unnformatve when we use actual volume data. Indeed columns (3) and (4) reveal that the results wth abnormal returns are very smlar to the results usng total returns. 6 Summary and concluson In ths paper we have analysed the effect of changes n stamp duty rates on share prces. We fnd that stamp duty clearly depresses share prces, partcularly for frms wth more frequently traded shares. Ths may ncrease the cost of captal faced by frms, whch n turn could have negatve repercussons on nvestment. Stamp duty also dstorts the sgnals that share prces send about the proftablty of frms, as share prces are also affected by expectatons of future turnover volumes and stamp duty rates. Our results show that these effects are real and measurable. Ths fndng s 18

21 mportant both n dscusson about the benefts of abolshng stamp duty, as well as for the wder debate on the merts of transacton taxes, ncludng the controversal Tobn-tax. It remans open, however, whether the negatve effects of alternatve sources of tax revenue would be smaller than those of stamp duty. 19

22 Appendx: Abnormal returns It s common n event studes to use ether a statstcal or economc model to control for changes n share prces that were caused by factors unrelated to the events studed (see MacKnlay, 1997). The most wdespread method s to focus on abnormal returns by calculatng for each share ts normal correlaton wth market movements ( beta ), and then focussng on any return n excess of that explaned by market movements. 18 The event studed n ths paper dffers fundamentally from the events analysed n most event studes, as t affects all shares (although to dfferent degrees) rather than just a few. Unlke n other event studes, the event tself wll thus have a consderable mpact on the market return. To see ths more clearly, consder the followng equaton, specfyng the relatonshp between the return on a share and the market return: r = R β r m where r s the abnormal return, R s the total return, r m s the market return and hstorcal correlaton between the market and ndvdual returns. β s the In an event study the am s usually to dentfy the effect caused by the event, rather than by other nformaton httng the market on the same day. In most studes ths s straghtforward, as the event wll have (vrtually) no effect on the market return. A merger announcement, for example, s unlkely to affect the market return, as long as the mergng frms are small relatve to the market. In these cases the mpact of the event on the return s smply the abnormal return. If however the event beng studed has a substantal mpact on the market return, and partcularly f the β vary systematcally wth the characterstcs of the dfferent groups of frms beng consdered, ths approach can gve a msleadng mpresson of the mpact of the event on the share prces of dfferent groups of frms. Ths concern n present n our context, as we llustrate below. 18 Although ths method s wdespread, t has not been used n emprcal studes of stamp dutes. These studes have ether focussed on total returns, or n the case of OXERA (001) on total returns n excess of the market return, but wthout allowng the mpact of the market return to dffer by frms,.e. assumng beta=1 for all frms. 0

23 The dfference-n-dfferences approach, whch restrcts ts attenton to relatve dfferences n return, makes the calculaton of abnormal return unnecessary as long as market betas are not systematcally dfferent across groups. So the dscusson above mght appear unnecessary. However, when we calculated betas and compared them to the sze dstrbuton, we found that they are clearly ncreasng n sze, as s demonstrated n Table 8. Table 8: Betas by frm sze 1990 Rank of frm Avg. beta S.e Notes: Betas calculated by regressng daly returns of ndvdual shares on the returns of the FT All Share ndex over the year pror to the 1990 event. The fndngs from Table 8 suggest that there s a fundamental dffculty: If betas are ncreasng n sze, and share turnover s too, then t s dffcult to dstngush between a postve aggregate shock to the stock market and a change to stamp duty, because both would cause dfferentally hgher returns for the shares of larger frms. Unfortunately usng abnormal returns does not provde a soluton to ths. Suppose there s an announcement of the abolton of stamp duty (as n 1990). Assume for smplcty that on ths day nothng else happens on stock markets and that the market return wll only be postve because of ths event. Table 9 shows the abnormal returns one would expect to see n that case. Table 9: Theoretcal abnormal returns (1990) Sze band Turnover rate Expected return MV share Market return beta Abnormal return Top % 74.1% 6.3% % % 15.3% 6.3% % % 6.4% 6.3% % rest % 4.% 6.3% % Turnover rates are ncreasng n sze, as documented n secton 4.3. The expected returns for each sze group are calculated as n Table 1, assumng a dvdend yeld of 4%. Usng the share of the market value made up by each sze group, the market return s obtaned. Clearly ths s manly drven by the effect on large frms. The betas are calculated for each sze group and are 1

24 ncreasng n sze. Applyng then the formula for the abnormal return, we fnd that the largest frms had the lowest abnormal return. Concludng from ths that they had not been affected by the stamp duty cut would be a mstake, as by assumpton the stamp duty cut was the only event occurrng. The problem s that the largest group prmarly drves the market return. At the same tme, wth a beta close to one, almost all of the effect s subsequently deducted. For practcal purposes ths means that f one s studyng an event wth mportant effects on market returns, focussng on dfferences across frms n abnormal returns may be msleadng. If betas dffer across groups, then ths wll even affect dfference-n-dfferences regressons. In the partcular case of dfferences by frm sze, snce betas are ncreasng n frm sze, the entre effect can be wped out f usng abnormal returns. In the opposte case of decreasng betas, the effect would be overstated. It may therefore be preferable to use total returns, partcularly f the market return over the perod s to a large extent drven by the event beng studed. Ths s the reason why ths paper has used total returns, as have all other papers n ths lterature, 19 though wthout provdng an explanaton. In prncple ths argument also holds for regressons usng actual tradng volume data to group frms. However, because the correlaton between sze and tradng volume s not perfect, t s possble that abnormal returns wll not be fully elmnated by the mechansm descrbed above. Indeed, n our regressons based on volume data, we fnd that ths s case (see secton 5.). References Atken, M. and Swan, P. (000), The mpact of a transacton tax on securty market traders: the case of Australa s tax reducton, Workng Paper, Unversty of Sydney Brealey, R. A. and S. C. Myers (000) Prncples of corporate fnance Sxth Edton, Irwn/McGraw-Hll. Ercsson, J. and Lndgren, R. (199), Transacton taxes and tradng volume on stock exchanges: an nternatonal comparson, Stockholm School of Economcs, epartment of Fnance, Workng Paper no. 39. Hau, H. and Chevaller, A. (000), Estmatng the volatlty effect of a securty transacton tax, ESSEC, Graduate Busness School, mmeo. 19 Except for OXERA (001) as mentoned above. Ths paper however assumes a fxed beta of 1, whch would only affect the constant n a dfference-n-dfferences regresson.

25 Hawkns, M. and J. McCrae (00) Stamp duty on share transactons: s there a case for change? Commentary 89, London: Insttute for Fscal Studes. Inland Revenue (00) Annual Report for the Year Endng 31 March 00, The Statonary Offce. Jackson, P. and O onnell, A. (1985), The effects of stamp duty on equty transactons and prces n the UK Stock Exchange, Bank of England, scusson Paper no. 5. Lndgren, R. and Westlund, A. (1990), How dd the transacton costs on the Stockholm Stock Exchange nfluence tradng volume and prce volatlty?, Skandnavska Ensklda Banken Quarterly Revew, /1990, pp MacKnlay, A. Crag (1997) Event Studes n Economcs and Fnance Journal of Economc Lterature, Vol pp OXERA (001) Impact of Stamp uty on the Cost of Captal of UK Lsted Companes paper prepared for the London Stock Exchange / The Hundred Group of Fnance rectors Saporta, V. and K. Kan (1997), The effects of stamp duty on the level and volatlty of UK equty prces, Bank of England, Workng Paper no. 71. Swan, P. L. and J. Westerholm, 001. The Impact of Transacton Costs on Turnover and Asset Prces; The Cases Of Sweden's and Fnland's Securty Transacton Tax Reductons, epartmental Workng Papers 144, Tor Vergata Unversty, CEIS Tobn, J. (1974) The new economcs one decade older Prnceton, NJ: Prnceton Unversty Press. Umlauf, S. (1993), Transacton taxes and the behavour of the Swedsh stock market, Journal of Fnancal Economcs, 33, pp