Market microstructure matters when imposing a Tobin tax-Evidence from the lab.

TRADING IN FX MARKETS IS DOMINATED BY TWO MICROSTRUCTURES: exchanges with market makers and OTC-markets without market makers. Using laboratory experiments we test whether the impact of a Tobin tax is different in these two market microstructures. We find that (i) in markets without market makers an unilaterally imposed Tobin tax (i.e. a tax haven exists) increases volatility. (ii) In contrast, in markets with market makers we observe a decrease in volatility in unilaterally taxed markets. (iii) An encompassing Tobin tax has no impact on volatility in either setting. Efficiency does not vary significantly across tax regimes.

Introduction

The idea of implementing a transaction tax on foreign exchange (FX) markets was first circulated by James Tobin in the early 1970s as a reaction to the high volatility in FX markets after the fall of the Bretton-Woods system of fixed exchange rates (Tobin, 1978). He argued that the tremendous increase in trading volume since then had mainly been due to speculative behavior.1 Tobin assumes two archetypes of traders on financial markets: stabilizing long-term investors who base their trades on fundamentals, and short-term speculators who try to detect time trends in prices. Hence, a small transaction tax would mainly harm the frequently trading speculators who would either leave the market or at least markedly reduce their trading volume. Consequently, this would lead to a decrease in volatility and to an increase in market efficiency, while potential tax revenues are a “side effect” for Tobin.2 The attitude on the last point has changed in the past few years, at least among politicians, as during the financial crisis 2007–2009 the idea of a Tobin tax has become very popular among them. The tax seems to target “evil speculators” and promises substantial tax revenues which do not have to be paid (directly) by the “normal” tax payer/voter.

Scientific research on the impact of a Tobin tax has started mainly in the 1990s with studies on the more general issue of how transaction taxes affect financial markets.3 There is broad consensus in the literature on some “trivial” issues such as negative effects of a Tobin tax on trading volume and market shares of taxed markets (compared to untaxed markets, i.e. tax havens).4

While the direction of the volume effects seems to be clear, other issues, especially the impact of a Tobin tax on volatility and on market efficiency, are still hotly debated, with strong academic backers for both sides. Parts of the controversy regarding volatility are likely due to different methodological approaches and different model assumptions: the main body of literature supporting the hypothesis of Tobin relies on agent-based models,5 while studies opposing the hypothesis of Tobin are mainly empirical, but suffer from the problem that they can only infer the impact of transaction taxes indirectly, since a Tobin tax has not yet been implemented.6

One common feature of all the papers mentioned so far is that the market microstructure has been ignored. This paper is an attempt to close this research gap. Currently, global trading in foreign exchange is dominated by two market microstructures: part of global volume is handled by exchanges where market makers ensure permanent liquidity provision.7 An even larger share of global volume is traded OTC (over-the-counter) between individual parties without market makers.8 The only paper to directly address the important market microstructure issues with respect to a Tobin tax is by Pellizzari and Westerhoff (2009). They use an agent-based model with the chartist/fundamentalist approach and explore the effect of a Tobin tax in different market microstructures, namely dealership and double-auction markets. They report that liquidity decreases in reaction to the imposition of a Tobin tax in a double-auction market and thus a given market order has a stronger price impact. As a consequence the imposition of a Tobin tax does not decrease price volatility, since the stabilizing effect of a reduction in speculative orders and the destabilizing effect of an increased price impact of orders due to lowered liquidity offset each other. By implementing a dealership-market with artificial market makers providing constant liquidity provision, they find that the introduction of a Tobin tax reduces volatility in dealership markets as speculation is reduced. Thus, a Tobin tax has different effects, depending on the market microstructure.9

Our paper can be understood as a cross-test of agent-based models and laboratory experiments, as we base our research question on and compare our results to Pellizzari and Westerhoff (2009). We implement two important alterations to their setup by (i) conducting laboratory markets with real humans interacting, and by (ii) implementing trade on two markets (for the same currency pair) simultaneously, thereby allowing for tax havens. For the experiment we build on and extend the setup in Hanke et al. (2010). In particular, we compare the impact of a Tobin tax under different market microstructures in laboratory markets: in Treatment OTC no market makers are present and thus each human trader can post limit and market orders. Hence, liquidity evolves endogenously through the actions of the human traders. In Treatment MM computerized market makers constantly post limit orders irrespective of the tax regime and thus keep liquidity provision constant.

We observe very strong and significant differences in the effects of a Tobin tax under different market microstructures: (i) in markets without market makers (Treatment OTC) an unilaterally imposed Tobin tax (i.e. a tax haven exists) increases volatility. (ii) In contrast, in markets with market makers (Treatment MM) an unilaterally imposed Tobin tax decreases volatility, while (iii) an encompassing Tobin tax has no impact on volatility in either setting. We do not find any significant differences in market efficiency across tax regimes, as all markets are fairly efficient.

The paper is organized as follows: In Section 2 we present the market model and experimental design. Sections 3 and 4 report the econometric model and the results, respectively. Finally, Section 5 concludes the paper by relating our results to previous studies and by discussing the practical implications of our results.

Experimental setup and procedure

The setup of the experiment follows the one presented in Hanke et al. (2010), who explore the effects of a Tobin tax in one or both of two continuous double-auction markets. As a useful preliminary, we provide the following definitions. A market is a sequence of 10 trading periods for the currencies A and B. A session consists of 2 markets (LEFT and RIGHT), in which traders can act on both markets simultaneously.10 A tax rate scenario defines when and on which markets within a session a two-way Tobin tax is imposed (possibilities LEFT, LEFT and RIGHT, no tax on both markets). A treatment consists of 6 tax rate scenarios (with 2 sessions in Treatment OTC and 4 sessions in Treatment MM for each tax rate scenario). With all other things equal, Treatment OTC uses a double-auction market architecture, whereas Treatment MM is set up as a dealership market with computerized market makers posting limit orders.

Market setup

In each session a different cohort of 16 (Treatment OTC) or 8 (Treatment MM) human subjects trade currency A for currency B on two markets (denoted LEFT and RIGHT and placed accordingly on the screen; see screenshot in Appendices A and B). Both markets are displayed on the trading screen at the same time and traders can be active on both markets simultaneously. Buying a currency on one market and selling it on the other is possible, as is buying on both markets or selling on both markets.

The fundamental value of A (expressed in units of B) is modelled as a geometric Brownian motion without drift:FV denotes the fundamental value in period k and γ is a normally distributed random variable with a mean of zero and a standard deviation of 5%. FV0 is set to 60. We draw one fundamental value path randomly (path I) and its counterpart mirrored at the unconditional expected value of FV is used as path II. For each tax rate scenario in Treatment OTC (MM) two (four) sessions are run, one (two) with path I, the other with path II.

For the sake of simplicity we introduce a symmetric information structure where at the beginning of each period each subject receives a private signal (SIGNAL) on the fundamental value of currency A. This signal is calculated as the FV plus a noise term with a mean of zero and a standard deviation of 2.5%. Estimation errors cancel out across subjects in each period to ensure that “the market” has an unbiased estimate of the FV.11

In each session half of the subjects are initially endowed with 75 A and 1500 B. The other half starts with 25 A and 4500 B. Given an initial fundamental value FV0 of 60 B per A each trader's initial wealth is 6000 in currency B. The holdings in A and B are carried over from one period to the next and going short up to 100 (6000) units of currency A (B) is possible.12 Any order size and the partial execution of limit orders is possible as long as the endowments in A and B are above −100 and −6000, respectively. Order books are emptied, i.e. all orders deleted, before the beginning of a new period. To keep things simple there are no interest payments on holdings in either currency.

Tax rate scenarios

The tax rate scenarios shown in Table 1 differ with respect to when and on which markets a (two-way) Tobin tax of 0.1% of the transaction value (price in B multiplied by units of A traded) is levied.

Table 1

Tax rate scenarios in each treatment.

Tax rate scenario	Periods 1–5		Periods 6–10
	LEFT	RIGHT	LEFT	RIGHT
0L	–	–	0.1%	–
02	–	–	0.1%	0.1%
L0	0.1%	–	–	–
L2	0.1%	–	0.1%	0.1%
20	0.1%	0.1%	–	–
2L	0.1%	0.1%	0.1%	–

Entries show the two-way tax rate (0.1% for each side) for taxed markets. Dashes indicate the absence of taxes. In Treatment OTC (MM) two (four) sessions are run for each tax rate scenario.

In particular, each session consists of two phases p of 5 periods each. Hence, the treatment abbreviations in Table 1 are to be read as follows: the numbers “0” and “2” specify whether no market (“0”) or both markets (“2”) are taxed in a given phase. If only one market is taxed, we chose to tax only the left market (“L”) to reduce the number of possible scenarios. In this case the right market is always the tax haven.13

Before the beginning of the first and the sixth period subjects are informed about the imposition of a tax with an announcement screen. This screen outlines in detail which markets are taxed and provides a calculation example for taxation. Subjects do not get any information about the potential implementation of transaction taxes before the main experiment starts and they are not informed whether and when the tax regime is changed again. Furthermore, the tax rate is also placed on the trading screen once a tax has been introduced.14

Experimental treatments

We ran 12 (24) sessions with 16 human (8 human subjects and 2 computerized agents) traders each in Treatment OTC (MM).15 The main experiment lasted 10 periods of 4 min each.16 To avoid strategic behavior towards the end of the experiment, subjects were told that the experiment will end between periods 8 and 14 with equal probability. At the end of the experiment all units of A were bought back at the fundamental value of the last period. Therefore, the final wealth comprised the value of the holdings in A (units of A multiplied by the fundamental value of the last period) plus the holdings in B and is converted into EUR at an exchange rate of 1 EUR = 400 Taler. All these were public knowledge. All 384 subjects were business students at the University of X, recruited with ORSEE (Greiner, 2004).17 Sessions were computerized (using zTree 3.2.8 by Fischbacher, 2007) and lasted about 90 min.

Treatment 1: over-the-counter – OTC

Subjects trade in a continuous double auction market and are able to post limit and market orders. Limit orders are executed according to price and then time priority.18 Market orders have priority over limit orders and are always executed instantaneously.19 The order books are open which means that all limit orders are immediately visible to all subjects.

Treatment 2: market maker – MM

Treatment MM deviates from Treatment OTC in one crucial aspect: subjects are not able to post limit orders, as those are provided by market makers. To achieve a constant liquidity inflow we implement one computerized market maker in each market (similar to Pellizzari and Westerhoff, 2009).20 Every several seconds (see process and parameters below) a market maker places both a bid and an ask at the same time t:Here P denotes the last transaction price at time t in period k and |ɛ| is the absolute value of a standard normally distributed random variable. Hence, the bids and asks of a market maker mostly dependent on the last transaction price, since he places a bid and an ask with a spread of 2|ɛ| with the same ɛ on each side of the current market price. Thus, if prices go up (through excess demand of the experimental subjects), market makers quickly incorporate this in their bids and asks. Note that they process no fundamental information. Market efficiency will thus be determined solely by the actions of human subjects. Market makers have no constraints on how many units of A and B they can hold, but in real markets they usually try to keep their long- and short-positions balanced, i.e. have a net exposure of zero. We therefore add a parameter δ to ensure that the market maker has a tendency to keep his net holdings in A close to his initial endowment A0 of zero. δ is calculated as (A − A0)/(|A − A0|) · (− |A − A0|)/100 which reduces to −A/100 with A (A0) denoting the holdings in currency A at time t of period k (at the beginning of the experiment). If, for example, his holdings in A are 20 units below the initial holdings A0, he adds 0.2 to the bid and the ask to make his bids more attractive for subjects to accept and to sell. Consequently, the further away the current holdings in A are from the initial holdings the more aggressively a market maker tries to bring his holdings in A back to a net position of zero.

In order to mimic the order flow generated in market phases with both markets being untaxed in Treatment OTC, 90% of the generated limit orders are posted between 1 and 23 s after the previous order was posted (with a mean of 7.2 s). The distribution of this stochastic “waiting time” is drawn from a Weibull-distribution.21 For the quantity posted with each limit order a Poisson-distribution with a mean of 4.5 units of A fits the distribution in Treatment OTC best. Parameter λ is half the average limit order size in all periods with tax regime “0” in Treatment OTC.22

Thus, the eight human traders in each session of Treatment MM are provided with a limit order flow very similar to untaxed markets in Treatment OTC. While the order flow (liquidity) by humans in Treatment OTC is likely to change after the introduction of a tax, this is not the case in Treatment MM, where the order flow is independent of the tax regime applied. This offers the advantage that we can measure the impact of a transaction tax on trader behavior when liquidity provision is held constant. However, it also means that some results need to be interpreted with caution, as in real markets market makers may change their order flow as a reaction to the tax.23

Definition of variables and econometric model

We use the following panel regression model:Here, y is a generic placeholder for the dependent variables explained below, m indicates cross-section (market) and p phase (i.e. five consecutive periods in which a certain tax regime was applied). TT_encompassing equals 1 when both markets are taxed, zero otherwise. TT_unilateral is a binary dummy to define unilaterally taxed markets with the other market being taxed and Tax_haven is a binary dummy for the tax haven. Consequently, intercept α represents the tax regime in which both markets are untaxed. Importantly, we apply clustered standard errors on a session level to allow for correlation within sessions and independence of observations between sessions. In particular, we implement the “vce(cluster varname)” method in STATA in all panel regressions in this paper.

Table 2 provides formulae for the dependent variables used in this paper: normalized trading volume, normalized returns, acceptance ratio (measure for trading behavior), relative absolute deviation (measure for market efficiency) and normalized tax revenues. We normalize trading volume (VOL) by the mean and the standard deviation of trading volume in each session s to avoid idiosyncratic impacts of individual sessions, since trading volumes differ by a factor of more than three between sessions.24 As one can see from Table 2 the respective means and standard deviations are calculated from period data. To arrive at normalized volume of phase p of market m the average of the respective five period values is calculated. A similar approach is applied for the volatility measure, standard deviation of normalized returns (SDRET). Log-returns, RET, with i denoting transaction, are normalized by the mean and the standard deviation in each session (see the discussion in Plerou et al., 1999 on the importance of normalizing returns from different observations). The standard deviation of these normalized returns in each market phase serves as dependent variable. Hence, independent observations with differences in the absolute level of volatility are easily comparable. ACCRATIO is calculated as the number of market orders divided by the number of limit orders. This variable is a proxy for the cautiousness of traders to accept limit orders, since low values hint at a very careful execution of market orders and a trend towards limit orders to gain the bid-ask spread. This behavior is especially expected in taxed markets, since the tax adds to the bid-ask spread as additional transaction costs. Note that ACCRATIO can be higher than 1 in cases where many small market orders are placed and thus partial execution of limit orders happens quite frequently. Beside tax revenues (TAXREV) where we normalize the tax revenues (in currency B) by the mean and the standard deviation of tax revenues in each session, relative absolute deviation (RAD) completes the set of variables. It serves as measure for mispricing and is the absolute difference between mean prices per period and the respective FVs, benchmarked at the average FV in the market (see Stöckl et al., 2010). Hence, the higher RAD, the stronger is mispricing and the lower is market efficiency. For the variables VOL, ACCRATIO, TAXREV, and RAD period values are calculated first and the mean per phase p and market m is used in the regression.

Table 2

Formulae for the calculation of variables.

Measure	Calculation
Normalized trading volume	VOLs,m,kNORM=(VOL_s,m,k−VOL_s¯)/σsVOL
Normalized returns (tick data)	RETs,m,iNORM=(RET_s,m,i−RET_s¯)/σsRET
Acceptance ratio	ACCRATIOs,m,k = MOs,m,k/LOs,m,k
Relative absolute deviationa	RAD_s,m,k=P_s,m,k¯−FV_s,m,k/FV_s¯
Normalized tax revenues	TAXREVs,m,kNORM=(TAXREV_s,m,k−TAXREV_s¯)/σsTAXREV

s, session; m, market; k, period; i, trades.

VOL = units of currency A traded in period k; ; ; RET = ln(P) − ln(P); P = trading price of trade i; ; ; MO = number of market orders; LO = number of limit orders. ; FV = fundamental value; ; TAXREV = tax revenues in currency B in period k; ; ;.

Stöckl et al. (2010).

Results

We observe very active trading in our markets, with an average of 764 transactions per session in Treatment OTC and 812 in Treatment MM. This is on average roughly one transaction every 3 s. Average trading volume per session is 1712 units of A in Treatment OTC, and 1314 in Treatment MM, which means that each unit of A is turned over 2.1 (3.3) times in OTC (MM). Fig. 1 presents the respective averages of normalized trading volume, volatility, acceptance ratio and market efficiency. The left four bars of each panel show data for Treatment OTC, the right four bars Treatment MM.

Fig. 1

Descriptive statistics. Averages per phase of the dependent variables conditional on treatment and tax regime. VOL (normalized trading volume), SDRETNORM (standard deviation of normalized returns), ACCRATIO (acceptance ratio – market orders divided by limit orders) and RAD (relative absolute deviation of prices compared to fundamentals). no_Tax: both markets untaxed, Tax_hav: this market untaxed, but other market taxed, TT_uni: this market taxed, but other market untaxed, TT_enc: both markets taxed.

The first panel on the top left presents data on normalized trading volume VOL. Here one observes that trading volume increases markedly in both treatments in the Tax havens (second bar) and falls even more in the unilaterally taxed markets (third bar). When an encompassing Tobin tax is implemented, trading volume decreases, but not dramatically (fourth bar). The effects seem to be stronger in Treatment OTC than in MM.

The level of volatility, measured by the standard deviation of normalized returns, is presented in the top right panel. While volatility is unchanged in both treatments when both markets are untaxed (first bar) compared to both being taxed (fourth bar), the unilateral introduction of a Tobin tax strongly increases volatility in Treatment OTC, but markedly decreases it in Treatment MM (third bar). The opposite holds for the Tax haven (second bar), where volatility decreases slightly in OTC, but increases in MM.

The bottom left panel presents data on the acceptance ratio. Here we see similar patterns (highest ratios when there is no tax and in the Tax haven, markedly lower ratios in single- or double-taxed markets), but at different levels: the acceptance ratio in Treatment MM is only roughly half the number in Treatment OTC. This is due to the fact that the computerized market makers in MM always post a bid and an ask simultaneously (with on average half the volume of the human traders in OTC), thus there are roughly two times as many orders in this treatment than in OTC.

Market efficiency, measured by RAD and shown in the bottom right panel, does not vary much across treatments and tax regimes. With values between 4.0 and 6.5 percent average deviation from the respective fundamental values, efficiency was quite high in all tax regimes.

Short selling, while allowed up to 100 percent of the initial total endowment (A and B combined), is not excessively used. Overall only 10.4 percent of subjects have short-positions in Treatment MM at the end of a period, while the number is higher at 19.2 percent in OTC. In both treatments short positions in currency A are more common than in currency B. For those subjects which hold short positions at the end of a period, the position is on average −2150 in B (up to −6000 possible) and −29 in A (−100 possible). Thus, the possibility to go short is used in only one-tenth to one-fifth of cases, and when it is used on average one third of the possible maximum is used. On average only 3–6 percent of the initial holdings were shorted.

After these descriptive statistics we now turn to the detailed econometric analysis provided in Tables 3 and 4. In the former table results of the panel regression according to Eq. (4) are shown, while the latter provides results of a pairwise Mann–Whitney U-test which serves as a non-parametric robustness check.

Trading volume and liquidity

Focussing on normalized trading volume (VOL) in Table 3, we find that trading volume drops, though not significantly, when a tax is introduced in both markets (TT_encompassing) compared to double-untaxed markets (intercept α) in both treatments. This is in line with Hanke et al. (2010), where the tax rate is five times as high in most of their treatments. Also comparable to Hanke et al. (2010), when the tax is introduced in only one market (TT_unilateral), the drop is highly significant in both treatments, while trading volume in the untaxed market (Tax_haven) increases significantly in both treatments (with the effect being stronger in OTC).

Table 3

Panel regression for both treatments.

	VOLNORM		SDRETNORM		ACCRATIO		RAD		TAXREVNORM
	OTC	MM	OTC	MM	OTC	MM	OTC	MM	OTC	MM
TT_encompassing	−0.257	−0.252	−0.078	−0.064	−0.129	−0.109**	−0.017	0.005	−0.329	−0.180
	(−0.999)	(−1.326)	(−0.546)	(−0.771)	(−1.458)	(−2.036)	(−1.198)	(0.975)	(−1.120)	(−0.640)
TT_unilateral	−1.227***	−0.815***	0.561**	−0.357***	−0.223*	−0.192***	0.008	0.013	−1.247***	−1.106***
	(−6.109)	(−4.915)	(1.969)	(−4.240)	(−1.814)	(−3.717)	(0.571)	(1.494)	(−6.060)	(−5.040)
Tax_haven	1.027***	0.334*	−0.111	0.247**	0.115*	0.049	−0.004	0.013
	(4.746)	(1.754)	(−1.084)	(2.184)	(1.669)	(0.866)	(−0.400)	(1.375)
α	0.119	0.164	0.987***	0.917***	1.031***	0.598***	0.058***	0.048***	0.155	0.132
	(0.800)	(1.525)	(15.982)	(14.616)	(10.176)	(10.921)	(3.219)	(10.522)	(0.950)	(0.860)
N	48	96	48	96	48	96	48	96	48	96

Dependent variables: VOL (normalized trading volume), SDRET (standard deviation of normalized returns), ACCRATIO (acceptance ratio – market orders divided by limit orders), RAD (relative absolute deviation of prices compared to fundamentals), and TAXREV (normalized tax revenues). z-Values are given in parentheses. TT_encompassing: both markets taxed; TT_unilateral: this market taxed, but other market untaxed; Tax_haven: this market untaxed, but other market taxed.

10% significance level of a double-sided test.

5% significance level of a double-sided test.

1% significance level of a double-sided test.

Turning to differences between the two treatments one can see that most effects of an unilateral imposition of a Tobin tax are weaker in MM than in OTC. Thus, the market microstructure clearly has an influence on the effects a Tobin tax has on markets. The main reason is that the constant order flow in MM ensures enough liquidity in the form of limit orders to facilitate trade. Due to the design, there are no significant differences in the number of limit orders across tax regimes in Treatment MM. In Treatment OTC, however, liquidity varies significantly across tax regimes, with an average of 190 and 195 limit orders per phase and market when both markets are untaxed and both markets are taxed (TT_encompassing), respectively. Only 92 limit orders are posted in unilaterally taxed markets (TT_unilateral) in each phase on average, as liquidity moves to the tax haven, where we observe an average of 267 limit orders per phase.25

Volatility

One of the most important, but also most evasive and controversial issues surrounding the Tobin tax is the development of volatility in taxed markets. The first – rather surprising – result is that volatility is not significantly affected by an encompassing Tobin tax in both treatments. We attribute this to the relatively small changes the tax triggers in trading volume and order flow in both treatments.

However, when the tax is introduced in only one market we observe that the standard deviation of normalized returns (SDRET) develops differently in the two microstructures: volatility increases significantly in unilaterally taxed markets (TT_unilateral) in Treatment OTC, but decreases significantly in the respective markets in Treatment MM compared to double-untaxed markets. The former runs counter, the latter is in line with the hypothesis of Tobin. This result merits a deeper analysis. As we find it is mainly a consequence of order flow, i.e. limit orders posted in the market. In MM the order flow is unaffected by the tax, i.e. the computerized market makers post as many limit orders as before. However, as human subjects in tax regime TT_unilateral submit fewer market orders (see development of trading volume), orderbooks become very liquid (large number of limit orders in the order book), and as a consequence the price impact of market orders and thus volatility decrease. By contrast, in Treatment OTC human subjects submit fewer limit orders in the unilaterally taxed market (92 limit orders per phase compared to 190 when both markets are untaxed) and a large share of liquidity shifts to the untaxed market (where on average 267 limit orders are posted), with the result of unilaterally taxed markets becoming relatively illiquid. Trading volume and the number of market orders posted decrease as well and thus the few transactions that are carried out have a stronger price impact and lead to comparatively high volatility in unilaterally taxed markets. One last finding we want to mention: volatility increases in the tax haven in Treatment MM due to the strong shift of trading activities to this market in combination with constant liquidity.

Trading behavior

ACCRATIO is the number of market orders divided by the number of limit orders which serves as a proxy for the cautiousness of subjects to accept posted limit orders. In Treatment MM this ratio is driven solely by the number of market orders placed by subjects, as liquidity provision is kept constant on both markets. As subjects mainly trade on the tax haven (Tax_haven), ACCRATIO in the unilaterally taxed market decreases significantly compared to when both markets are untaxed.

By contrast, in Treatment OTC also the limit order flow is affected by a tax. In taxed markets subjects are more cautious and so accept fewer limit orders. Hence, ACCRATIO decreases in markets with an unilateral tax (TT_unilateral), while it increases in the Tax haven (Tax_haven).

Market efficiency

Turning to market efficiency, which we measure by relative absolute deviation (RAD), we find that it is not significantly affected by a Tobin tax in any of the treatments. This result from the regression of Table 4 is supported by pairwise Mann–Whitney U-tests (see Table 4) which all deliver p-values above 0.20. This result differs from Hanke et al. (2010), who find lower efficiency in an unilaterally taxed market. We attribute this to the high level of efficiency we observe throughout the experiment. With average RADs of 5.8 and 4.8 percent in OTC and MM, respectively, the deviations from the fundamental values are never very large.

Table 4

Robustness checks for the dependent variables. Pairwise Mann–Whitney U-tests (z-values and p-values in parenthesis are provided).

OTC	TT_unilateral	TT_encompassing	Tax_haven
VOLNORM
no_tax	3.797***	1.319	−3.246***
	(0.000)	(0.187)	(0.001)
TT_unilateral		−3.858***	−3.361***
		(0.000)	(0.001)
TT_encompassing			−3.552***
			(0.000)

VOL (normalized trading volume), SDRET (standard deviation of normalized returns), ACCRATIO (acceptance ratio – market orders divided by limit orders), RAD (relative absolute deviation of prices compared to fundamentals), and TAXREV (normalized tax revenues). no_tax: both markets untaxed; TT_encompassing: both markets taxed; TT_unilateral: this market taxed, but other market untaxed; Tax_haven: this market untaxed, but other market taxed.

10% significance level of a double-sided test.

5% significance level of a double-sided test.

1% significance level of a double-sided test.

Tax revenues

The development of tax revenues (TAXREV) mainly depends on the development of trading volume. As trading volume reduction is stronger in unilaterally taxed markets of Treatment OTC, the negative effect on tax revenues there is nearly two times stronger than in Treatment MM (compared to hypothetical tax revenues in periods where both markets are untaxed). Looking at non-normalized data we find that tax revenues are more than 40% higher in unilaterally taxed markets in Treatment MM than in Treatment OTC. Only when an encompassing Tobin tax is imposed (TT_encompassing) tax revenues are substantial.

Conclusion and discussion

We examined the effect the introduction of a Tobin tax had in laboratory markets which were set up either as OTC-markets or as dealership markets where computerized market makers (MM) provided limit orders and thus liquidity provision irrespective of the tax regime applied. The main findings of the paper are on the controversial issue of volatility: (i) in markets without market makers an unilaterally imposed Tobin tax increased volatility. (ii) In contrast, in markets with market makers an unilaterally imposed Tobin tax decreased volatility, while (iii) an encompassing Tobin tax had no impact on volatility in either setting. In particular, the mechanisms of the results were mainly due to different flows of liquidity in both treatments: in markets of Treatment OTC an unilaterally introduced Tobin tax decreased trading volume and the number of limit orders significantly, leading to lower orderbook liquidity, an increased price impact of market orders and thus higher volatility. By contrast, in unilaterally taxed markets in Treatment MM subjects traded less as well. In combination with the constant order flow provided by the computerized market makers, this lead to highly liquid orderbooks, a decrease in the price impact of market orders and thus lower volatility. At the same time volatility increased in the tax haven due to increased trading activity in combination with a constant level of liquidity provision. Trading volume decreased much less than in markets with an unilateral Tobin tax and hence tax revenues were substantial.

Thus, from the perspective of volatility, Treatment OTC stands in contrast to the hypothesis of Tobin and supports most empirical studies on transaction taxes (e.g. Aliber et al., 2003; Habermeier and Kirilenko, 2003; Hau, 2006). Instead, Treatment MM is in line with Tobin's conjectures and supports many agent-based studies on transaction taxes (e.g. Frankel, 1996; Westerhoff, 2003; Ehrenstein et al., 2005; Westerhoff and Dieci, 2006). We also report similar, mostly even more pronounced effects compared to Pellizzari and Westerhoff (2009). The most likely reason for the less pronounced results in their double-auction setting is that they do not allow for tax avoidance by implementing only one market. As agents cannot shift volume to another market, their “liquidity effect” is probably weaker and hence volatility remains almost unaffected. Thus, following the results of Pellizzari and Westerhoff (2009) and our findings, future research on the Tobin tax should take into account the strong impact of market microstructure, especially for volatility.

When trying to relate our results to real FX markets, we have to acknowledge the limitations of our laboratory markets, as especially Treatment MM was set up with very basic computerized market makers. Also, as with any experimental study, the size of the data-set is limited and we have to keep in mind that we operate in stylized markets. However, if even such simple markets where few subjects with comparatively low incentives trade deliver strong and robust results (e.g. strong tax avoidance, increased volatility in humans-only-markets), we are confident that these results translate well to larger markets where the stakes are much higher and subjects thus more eager to find “optimal” responses to tax regime changes. Thus, we think the results are clear and robust enough to state that (i) markets with market makers (MM) could lead to the desired outcome of Tobin (i.e. lower volatility as high-frequency traders trade less than before) when it can be assured that liquidity provision is hardly affected by the tax. This finding is remarkable as it would be beneficial for governments to impose a Tobin tax on such markets even without international coordination, since an unilateral imposition lowers volatility without affecting efficiency. Furthermore, (ii) no matter which microstructure is applied, an encompassing Tobin tax would not increase volatility or affect efficiency and would raise substantial tax revenues if introduced on all major markets.