Securities transaction taxes and stock price informativeness: evidence for France and Italy

This empirical study addresses the impact of securities transaction taxes (STTs) on price informativeness. Two specific events are assessed: the entry into force of STTs in France and Italy in 2012 and 2013, respectively. Specifically, it is gauged whether those events produced effects on the future earnings response coefficient (FERC)—a proxy for the capacity of investors to anticipate future earnings—of the stocks eligible for the STT. Our findings indicate that the new taxes on securities transactions that entered into force in France and Italy were neutral with respect to the effects on price informativeness. That is, no material changes in the FERC of affected stocks are detected in the aftermath of the STTs’ entry into force. This result applies when considering the events in France and Italy separately, and when analyzing different subsets of the sample based on the size, trading activity, and liquidity of the stocks.


Introduction
Do securities transaction taxes (STTs) influence price informativeness, namely the capacity of stock prices to track a firm's fundamentals? Indeed, the taxation of capital market transactions is a controversial issue that has generated keen interest among financial researchers. While there is almost consensus that trading activity 1 3 falls in the wake of the introduction of STTs, research on the sign and magnitude of impacts over other financial outcome variables, such as liquidity, volatility, and price informativeness, has remained inconclusive. Our empirical analysis sheds additional light on this question while evaluating the impact on price informativeness of the introduction of STTs by France andItaly in 2012 and2013, respectively. Two opposing views emerged in the debate about the potential benefits and costs of the introduction of STTs for market quality. On the bright side, some economists argue that too much trading heightens (excess) market volatility. According to Keynes (1936), trading activity is driven by short-term speculation and not by fundamentals. By discouraging speculative trading, STTs curtail wasting of resources and excess volatility. Likewise, Stiglitz (1989) claims that too much noise trading exacerbates "excess volatility," whereas Summers and Summers (1989) advocate that STTs discourage participation of frequent traders that do not base their decisions on fundamental information (e.g., "positive feedback traders" and "trend followers"). When decisions to trade are not justified by information, stock prices tend to deviate more from their correct level.
Accordingly, noise trading and short-term speculation constitute negative externalities, whereby the reduction of such "futile" trading via STT fosters market quality. 1 The more sensitive these short-term investors and noise traders are to transaction costs, the greater the share of trading that will be diverted to economic agents possessing fundamental information. This is because, in equilibrium, assets with higher transaction costs are held by investors with longer average holding periods ("composition effect"). If long-term shareholders' influence on price formation increases, it is reasonable to assume that prices will become more aligned with fundamentals.
On the negative side, STTs discourage the participation of liquidity providers. It is clear that STTs will drive away both rational agents and noise traders for tax saving reasons (Palley 1999). Kupiec (1996) and Schwert and Seguin (1993) highlight the role of speculation in boosting liquidity and stabilizing financial markets. Ross (1989) emphasizes the negative impact of short-term traders' withdrawal on liquidity and price efficiency. Wider bid-ask spreads and higher volatility might emerge with taxation. By increasing the costs of inventory positions and arbitrage, STTs could widen the deviation of prices from correct levels. 2 Jointly, the decay in participation, trading activity, and arbitrageurs' activity reinforce the initial impact of the tax on transaction costs and price efficiency. 1 If noise traders (either uninformed or not perfectly rational) prevent stock prices from converging to fundamental value, then mounting trading activity is destabilizing. The deterrence of noise traders' participation vis-à-vis informed trading ("composition/clientele effect") could improve price informativeness. This assertion is in line with Subrahmanyam (1998) who states that STTs encourage the acquisition of long-term information (vis-à-vis short-term information), despite the potential detrimental effect on liquidity and trading activity. Demary (2011) suggests that fundamental traders (chartists) become more (less) aggressive with the introduction of STTs. 2 In this regard, Habermeier and Kirilenko (2001) echo that STTs produce effects on market quality similar to those found by previous literature for capital controls.

3
Securities transaction taxes and stock price informativeness:… On the theoretical side, the equilibrium models of Constantinides (1986) and Vayanos (1998) show that larger transaction costs are accompanied by the change of investors' trading patterns, namely lower portfolio turnover ("turnover adjustment") and trading activity. Matheson (2012) and Mannaro et al. (2008) predict lower trading volume with the introduction of STTs. These predictions are underpinned by empirical literature showing that STTs impair trading activity (Baltagi et al. 2006;Chou and Wang 2006;Hanke et al. 2010;Pomeranets and Weaver 2018;Umlauf 1993). 3 Also, in an analysis of the French and Italian STT events, Capelle-Blancard and Havrylchyk (2016), Meyer et al. (2015), Colliard and Hoffmann (2017), and the EU Commission report (2013) 4 find a detrimental effect on trading activity.
As for liquidity, Bloomfield et al. (2009) use a laboratory experiment to show that taxation should not affect spreads. Pomeranets and Weaver (2018) document higher relative spreads and price impact after a rise in STTs in the state of New York. Capelle-Blancard and Havrylchyk (2016), Meyer et al. (2015), and Becchetti et al. (2014) do not find impacts on liquidity of the French STT, but Colliard and Hoffmann (2017) document detrimental effects. Cappelletti et al. (2017), Galvani and Ackman (2021), and Hvozdyk and Rustanov (2016) pinpoint lower liquidity with the introduction of STTs in Italy. 5 Capelle-Blancard (2017) shows that the Italian FTT exerted no effects on market liquidity or volatility using a regression discontinuity design.
Our investigation departs from previous studies by speaking directly to the impact of STTs on price informativeness. While prior research essentially explored effects on liquidity, volatility, and volumes, we address the capacity of stock prices to track a firm's fundamentals. In effect, it is not clear whether price informativeness is buoyed (or hampered) by STTs. The tax induces turnover adjustment, and discourages portfolio reallocation and short-term speculation. Investors with a longer investment horizon (more prone to acquiring fundamental information) could increase their holdings ("holdings adjustment effect") in affected securities relative to noise 3 Hu (1998) and Bloomfield, O'Hara, and Saar (2009) do not find impacts on trading volume. 4 Available at https:// ec. europa. eu/ taxat ion_ custo ms/ sites/ defau lt/ files/ docs/ body/ effect_ french_ ftt. pdf. 5 Other studies stress the relevance of external factors on the conclusions. Information asymmetry (Dupont and Lee 2007), firm size (Galvani and Ackman 2021), and market type-e.g., order driven versus market driven (Pellizzari and Westerhoff 2009)-could influence inferences with respect to liquidity, whereas fundamental risk and supply risk (Song and Zhang 2005) or market cycle (Phylaktis and Aristidou 2007) could weigh on inferences with regard to volatility.

3
traders. However, at the same time, the tax also impairs market liquidity, depth, and participation, with potentially negative consequences for price informativeness.
The key question we attempt to answer is whether the ability of stock prices to mirror firms' fundamental value and, more compellingly, future earnings, is affected by STTs. An extension of the future earnings response coefficient (henceforth, FERC) model is employed in the empirical investigation. In the baseline FERC model, current returns are regressed against a constant, 1 year lagged past earnings, current earnings, the sum of earnings for fiscal year t + 1 through t + 3, and stock returns for fiscal year t + 1 through t + 3. 6 The FERC-i.e., the estimated loading on future earnings-measures the capacity of stock prices "to bring the future forward" with respect to profitability (Collins et al. 1994;Gelb and Zarowin 2002;Lundholm and Myers 2002) and is employed in the literature as a price informativeness indicator.
The setting of Lundholm and Myers (2002) and Tucker and Zarowin (2006) is modified to extract difference-in-differences estimates pertaining to the inception of STTs. The identification strategy underlying that difference-in-differences analysis is the comparison of the FERC of stocks affected (treatment group) and unaffected (control group) by the application of STTs, before and after the tax event. The "treatment" group contains stocks from France and Italy that are eligible for STTs, whereas the "control" group includes stocks not affected by the tax event (although sharing the same characteristics as those from the treatment bin). The inclusion of the control group in the assessment is justified by the necessity to generate counterfactual evidence, namely, to control for time-varying factors that drive price informativeness but are unrelated with the application of STTs. As with event study analyses, two different time frames are compared: a pre-event window, encompassing the period preceding STTs in France and Italy, and a post-event window covering the subsequent phase.
The initial analysis considers the tax events in France and Italy in tandem. Thus, the treatment group comprises the French stocks with a market capitalization above 1 billion euros in one of the relevant tax event dates; the pre-event window for these stocks covers the period 2008-2011, whereas the post-event window comprises the span 2012-2016. It also includes Italian stocks with a market capitalization above 0.5 billion euros in one of the relevant dates; the pre-event window for Italian stocks ranges from 2008 to 2012, whereas the post-event window spans the period 2013 to 2016. Stocks from Austria, Belgium, Germany, Finland, Spain, and the Netherlands are used as controls.
To implement the difference-in-differences analysis, three binary variables are created and added to the FERC model setup: one indicating inclusion in the treatment group, another specifying the timing of the tax event, and a third denoting the stock eligibility for the STT each year. The three binary variables are interacted with covariates from the baseline FERC model and are added to the FERC specification. In our simpler approach, the baseline FERC model is extended with the introduction of the binary variable specifying the eligibility of the stock to the tax event in 1 3 Securities transaction taxes and stock price informativeness:… a given year and interactions of that variable with other covariates from the original setting. The coefficient on the interaction of future earnings with that binary variable captures the impact of the STT on the FERC, and hence, on price informativeness.
Remarkably, the point estimate for that variable is not statistically meaningful. These findings conform to the hypothesis that the impact of STTs on price informativeness about future earnings was neutral. We reach similar findings when the French and Italian events are assessed separately, as well as when running alternative model specifications. These supplementary tests include the estimation of less parsimonious models, the introduction of additional control variables to account for the stock's information environment, allowing time-varying FERC in the model, and the implementation of a nearest-neighbor matching methodology. Overall, our inferences remain intact when running these alternative approaches. Finally, a crosssectional analysis (with respect to firm size, trading activity, and liquidity) reveals similar results in different sample partitions, thereby confirming that conclusions are not driven by specific subsets of the analyzed data.
Our assessment is noteworthy for several reasons. Firstly, theoretical and empirical research is ambiguous about the net benefits of transaction taxes. Second, there are no studies looking directly into the impact of STTs on price informativeness on future profitability, which makes this analysis a novelty in the literature. Third, further evidence on STTs is highly desirable at this point, as several countries, including those of the european Union, are considering following France and Italy in the introduction of such taxes. Overall, these findings inform that neither the French nor the Italian STT weighed on the price informativeness of the stocks affected by the new taxation. Implementation of STTs in these countries hardly affected the capacity of stock returns to anticipate future earnings. This outcome is, in fact, aligned with others, suggesting that STTs produce negligible effects on market quality variables, such as liquidity and volatility. 7 Still, we must caution that in both cases, the STT is only applicable to large stocks, and market makers are exempt from the tax, which could have smoothed its impact on overall market quality.
The remainder of the paper is set out as follows. Section 2 presents the methodology and the research design, while Sect. 3 depicts the dataset. The analysis and discussion of the results are shown in Sect. 4. Finally, Sect. 5 presents the final remarks.

Research design and methodology
This study takes advantage of the unique setting offered by the implementation of STTs by France and Italy to investigate the impacts of such event types on stock price informativeness. As discussed in the introductory section, such effects are 7 Colliard and Hoffmann (2017) document a decline (increase) of holdings of French stocks by investors with high (low) portfolio turnover, consistent with STTs curbing noise trading and short-term speculation. However, the benefits of this "clientele effect" (composition of the trading population) were offset by a modest increase in transaction, hedging, and adverse selection costs. Thus, not only irrational investors may have been driven away but also rational ones. theoretically unclear and have been overlooked by the empirical literature, which has focused instead on other financial outcome variables such as trading activity, volatility, and liquidity. Our analysis departs from prior literature because it looks directly into the effects on stock price informativeness about fundamentals, particularly the accuracy of investors' expectations about future earnings.
To evaluate the impact of the STT on stock price informativeness, a differencein-differences approach is carried out, where two events are analyzed in tandem: the introduction of STTs by France and Italy in 2012 and 2013, respectively. With respect to France, an STT of 20 bps was introduced on stock purchases, effective from 1 August 2012. The STT covers stock trades of listed companies incorporated in France with market capitalization exceeding one billion euros irrespective of the trading platform used [e.g., stock exchanges, multilateral trading facilities, and over the counter (OTC)] and of the country of residence of the investor. In the first year, the relevant date for the computation of the market capitalization was 1 January 2012, whereas in the following years, the reference date became 1 December of the previous year. The Italian STT came into effect on 1 March 2013, being applied to stock trades and equity-based derivatives. Only trades on stocks issued by domestic (resident) firms with a market capitalization above 0.5 billion euros in November of the prior year are eligible.The base rate is 10 and 20 bps for trades conducted on stock exchanges and OTC, respectively.
Price informativeness is retrieved from the FERC model of Collins et al. (1994), Gelb and Zarowin (2002), and Lundholm and Myers (2002). The financial outcome variable (FERC) has roots in the accounting literature and reflects the association between returns and future earnings realizations. Ball and Brown (1968), Beaver et al. (1980), and Strong and Walker (1993), inter alia, explore the earnings-returns relation (ERC model) and find a weak association between current earnings and stock returns. Collins et al. (1994) suggest that the explanatory power of the ERC model is substantially improved with the inclusion of future earnings in the regression. Effectively, they find the point estimate on future earnings to be statistically and economically significant; the model's explanatory power also increases substantially, meaning that investors anticipate and price in future earnings. 8 Accordingly, current stock returns mirror both the unexpected current earnings and the contemporaneous variation in expectations about future earnings. Empirically, the FERC has been utilized to assess the capacity of stock returns to track long-term fundamentals. The underlying rationale is that returns become more informative when investors price in a firm's future profitability: a positive (and statistically significant) sensitivity of current returns to future earnings realizations is interpreted as stock prices embedding accurate expectations about future earnings. Our baseline setting is an extension of the model used by Tucker and Zarowin (2006) and Lundholm and Myers (2002), and is expressed as follows: where R i,t stands for stock returns over a 12 month time frame that begins 3 months after the beginning of fiscal year t; R3 i,t is the log buy-and-hold return for the 3 year period that begins 3 months after the end of fiscal year t; X i,t is the net income before extraordinary items available to common shareholders (NIBE) in fiscal year t scaled by market capitalization at 3 months after the beginning of fiscal year t; X3 i,t is calculated as the sum of the NIBE for the years t + 1, t + 2, and t + 3 deflated by market capitalization at 3 months after the beginning of fiscal year t. All these variables are winsorized at the 1st and 99th tails of their distribution. 9 The FERC estimate (the point estimate on X3 i,t , i.e., â 3 ) is expected to be positive. A higher â 3 signals greater sensitivity of current returns to future profitability, or in other words, greater capacity of stock prices to "bring the future forward" about firm profitability. The estimated coefficient(s) for X i,t ( X i,t−1 and R3 i,t ) is (are) expected to be positive (negative). Following earlier research (Haw et al. 2012), country, industry (two-digit SIC) and year fixed effects are added to the setting. 10 In the standard specification, CONTROLS are excluded to reduce concerns of collinearity. In an extended specification, they are also employed in the model and include market capitalization at the beginning of fiscal year t ( SIZE i,t ); asset growth ( TAG i,t ) computed as the percentage change in total assets from the beginning of year t − 1 to the end of year t + 1; analyst coverage ( ACov i,t ) defined as the number of analysts issuing earnings per share (EPS) forecasts at the beginning of year t; earnings variability calculated as the standard deviation of NIBE for fiscal year t + 1 through t + 3, divided by the market capitalization of the firm at 3 months after the beginning of year t ( EVol i,t ); market-to-book ratio ( M∕B i,t ) at the beginning of year t; and a binary variable indicating whether X3 i,t is negative, and set to 0 otherwise ( LOSS i,t ). Aside from LOSS i,t , these variables are converted into decile ranks within year bins. CONTROLS include the interactions of The difference-in-differences approach is designed to assess whether the introduction of STT's was accompanied by changes in the FERC. To that end, the FERCs 9 The intention of including future returns in the regression is to control for changes in expectations about future earnings after the current period, thereby allaying concerns about measurement error (Lundholm and Myers 2002). Three leads on future earnings and returns (k = 3) are utilized because previous research (e.g., Collins et al. 1994) showed that adding more leads of those variables hardly affected the explanatory power of the regression. 10 The original setup is expressed as follows: with X t−1 , X t and R t as defined above; X t+k corresponds to the firm's future NIBE in the k year period following the current year (scaled by the market capitalization of the firm at t-1); and R t,t+k stands for the log buy-and-hold annual return for the k year period following the current year. To make the model more parsimonious, Tucker and Zarowin (2006) and Lundholm and Myers (2002), inter alia, advocate the aggregation of future earnings and future returns.
of the stocks affected by the implementation of STTs (i.e., treatment group) before and after the tax event are compared. To generate counterfactual evidence, a control group of stocks (i.e., not affected by the STTs) is included in the analysis to control for time-varying factors that influence the price informativeness of the overall market but are not related to the tax event. In short, we compare the change of the FERC of treated stocks with that registered by control stocks in the wake of the onset of STTs. Note that the major assumption behind difference-in-difference analysis is that treated units and control units follow parallel trends with respect to outcome variables.
Previous research pertaining to the impact on the liquidity and volatility of the French STT considered a similar research design, but control groups vary: French mid and small caps (Becchetti et al. 2014 (2017) uses German firms and Italian small caps as control groups. Our control group includes tax-exempt Italian and French firms and firms headquartered in Austria, Belgium, Germany, Finland, Spain, and the Netherlands. Therefore, the control group comprises firms located geographically close and economically linked to those of the treatment group. Moreover, major stock index returns of these "control" countries are highly correlated with the ones of France and Italy. To ensure homogeneity of the sample, we exclude firms with less than 350 million euros of market capitalization, unless stated otherwise.
To address our research question, we proceed in the following way. First, three binary variables are created. TreatG i is a binary variable specifying whether the stock belongs to the treatment group (i.e., it is eligible for the STT at least once during the analyzed time frame; notice that this variable is time invariant). YearSTT t is a binary variable set to one after the entry-into-force of the French STT, and zero otherwise. STT i,t is a binary variable set to one if the stock is eligible for the STT in year t, and otherwise zero. Then, Eq. (1) is extended to obtain difference-in-differences estimates. Specifically, interactions of covariates from the baseline model (constant, X i,t−1 , X i,t , X3 i,t and R3 i,t ) with TreatG i , YearSTT t and STT i,t are included in the following regression models: (2) Securities transaction taxes and stock price informativeness:… with all variables as previously defined. The impact of the STT is measured by the coefficient on X3 i,t * STT i,t . If it is positive (negative) and significant, the tax produced positive (detrimental) effects on price informativeness. Interactions with TreatG i capture differences between treatment and control group units in the period that precedes the analyzed event, whereas interactions with YearSTT t control for common shocks (thus, unrelated to the STT) that affected both groups alike. Importantly, if X3 i,t * TreatG i lacks explanatory power, it may be inferred that the FERC is virtually identical in the two groups before the adoption of STTs. If X3 i,t * YearSTT t is not statistically significant, then FERCs would stay the same in the absence of the event.
The next section presents the data sources and summary statistics.

Data sources and descriptive statistics
Our data are assembled from three different databases. Data on trading activity (prices and volumes) are gathered from Datastream. Firms' financial and accounting data are sourced from Worldscope. Analyst coverage is obtained in I/B/E/S. Following earlier research (Lee 2018;Tucker and Zarowin 2006), firms in the financial industries (SIC code 6000-6999) are dropped from the investigation. The sample covers eight countries (Austria, Belgium, Finland, France, Germany, Italy, the Netherlands, and Spain) and the period 2007-2019. The tax event only affected stocks with market capitalization above 0.5 billion euros in Italy and above 1 billion euros in France. Other stocks are tax-exempt. To form homogeneous groups with respect to size, we exclude stocks with market capitalization below 350 million euros from the assessment. Moreover, we also exclude firms with negative book equity or net losses above the previous year's book equity. Following Lundholm and Myers (2002), we employ additional filters to the main sample to allay concerns that conclusions are driven by outliers. Thus, firm-year observations with outliers regarding earnings and returns are excluded. More precisely, if X i,t is greater (lower) than 1 (−1), the observation is dropped; similarly, if X3 i,t is above (below) 3 (−3), the observation is removed. Finally, firm-year observations with | | R3 i,t | | above 1000% are also dropped. Table 1 presents sample statistics. The average (median) R i,t stands at 1.6% (7.1%), whereas the average (median) X i,t hovers around 4.5% (5.7%). With respect to R i,t , the dispersion is high with the interquartile range standing above 40%. The average X3 i,t is approximately 14.6%, and about 13% of the firms recorded three-year-ahead The average (median) market value of equity amounts to 10 400 million USD (303 million USD). These large numbers are explained by the sample filters that removed firms with an equity market value below 350 million euros. The average (median) number of analysts issuing EPS forecasts is about 13 (15).
The median earnings volatility ( EVOL i,t ) is 2.9%, and the median percentage sales growth is 4.5%. The average ROA-return on assets (earnings before interest and taxes deflated by lagged total assets) is 8.4%, whereas the average cash flows from operations scaled by lagged market capitalization (CF) is 13%. The average market-to-book ratio is 2.4, whereas the average leverage (total liabilities deflated by total assets) is 19.7%. Panel B allows a side-by-side comparison between "treated" units and "control" units in the span prior to 2012, i.e., before the inception of STTs. In general, "treated" units had a slightly better market performance, but reached lower profitability. In either case, the subsamples comprise large companies, thanks to the sample restrictions imposed. In the following section, we move on the presentation of the main results and conclusions of the analysis.

Main inferences
Our assessment begins with the estimation of the baseline FERC model, i.e., Eq. 1. The observations covering the span 2008-2016 from treated and control groups are included in the regression. The sample is restricted to firms with market capitalization above 350 million euros, irrespective of the location. 11 The treatment group comprises firms headquartered in France with a market capitalization above 1 billion euros after 2012 and firms located in Italy with a market capitalization above 0.5 billion euros after 2013, whereas the control group encompasses firms headquartered in Austria, Belgium, Finland, Germany, the Netherlands and Spain, and other firms headquartered in France and Italy that are not part of the treatment group.
Information in column [1] of Table 2 shows that the point estimate on X3 i,t (i.e., the FERC estimate) is positive (0.58) and statistically significant (p-value < 0.01). The point estimate on X i,t is also positive and statistically significant. These outcomes suggest co-movement between current returns and current and future earnings. Finally, the point estimates for X i,t−1 and R3 i,t are negative and statistically meaningful. All these results conform to those of Tucker and Zarowin (2006) and Lundholm and Myers (2002).
Next, we compare point estimates for treated and control stocks before and after the tax entered into force. The sample is split using that date breakpoint. A variant of Eq. 1 where X i,t−1 , X i,t , X3 i,t and R3 i,t are interacted with TreatG i is estimated in the two subsets. Column [2] reports results for the period that precedes the tax event. The estimated loading on X3 i,t is positive and statistically meaningful, whereas TreatG i * X3 i,t lacks statistical significance. We redo the previous equation for the post-event period (after 2012 for French stocks and 2013 for Italian stocks). Interestingly, TreatG i * X3 i,t continues to lack statistical significance (see column [3]). Taken together, these results imply a similar FERC for treated and control stocks before and after the introduction of the tax in France and Italy.
This preliminary evidence casts doubt regarding the existence of a meaningful impact of STTs on price informativeness. Going forward, Eqs. 2 and 3 are estimated 11 The conclusions of the analysis are qualitatively the same using other thresholds or even including all the firms, regardless of size. with the aim of extracting difference-in-differences estimates for FERC. Note that in Eq. 3, we account for differences in treated and control groups before the event and control for the common path of FERC in the absence of the event (see column [5]). Critically, STT i,t * X3 i,t lacks statistical significance in both regressions, consistent with the reasoning that price informativeness suffered no impacts from the introduction of the tax. TreatG i * X3 i,t also lacks explanatory power, substantiating previous inferences that both groups had similar FERCs before the tax event. As to YearSTT t * X3 i,t , the point estimate is negative and statistically significant at the 10% level, suggesting a slight decline of the FERCs, but for all stocks, during the analyzed span (see column [5]). Column [6] of Table 2 reports results for Eq. 3, removing the Italian firms from the sample. In this case, the sample is restricted to firms with market capitalization above 750 million euros, irrespective of the location. This way, we turn our attention exclusively to the impact of the French STT, while ensuring relatively homogenous groups with respect to size. Remarkably, STT i,t * X3 i,t remains statistically meaningless, as found when using the full sample, complying with the hypothesis that the French STT produced residual effects on price informativeness about future earnings of the affected stocks. As to TreatG i * X3 i,t and YearSTT t * X3 i,t , they also lack predictive power.
The results for the analysis of the Italian STT are reported in column [7]. Now, the treatment group comprises stocks eligible to the novel Italian tax in the period 2013-2016 (i.e., stocks issued by Italian domestic firms with a market capitalization above 0.5 billion euros). The control group encompasses firms headquartered in Austria, Belgium, Finland, Germany, the Netherlands and Spain, and other firms headquartered in Italy with a market capitalization below 0.5 billion euros. To ensure homogeneity of treated and control units, the sample is again restricted to firm-year observations with market capitalization above 350 million euros. We note that the threshold for inclusion in the sample is below the one used for the French STT's analysis (0.75 billion euros), but conclusions prevail with alternative thresholds. Remarkably, STT i,t * X3 i,t lacks explanatory power when estimating Eq. 3, suggesting no impact of the tax event on the price informativeness of Italian stocks. TreatG i * X3 i,t and YearSTT t * X3 i,t are also statistically meaningless.
Thus far, our findings indicate a negligible impact of STTs on the FERC. That outcome prevails when considering the events on France and Italy separately. Nevertheless, we did yet not consider control variables in Eqs. 2 and 3. The most prominent reason is that adding further interaction variables to the model could elicit collinearity issues and increase standard errors of point estimates. Still, it is compelling to rule out the hypothesis that conclusions are being driven by other determinants of price informativeness. Column [1] of Table 3 presents results when Eq. 3 is extended with CONTROLS ; column [2] displays the results for a more parsimonious version of Eq. 3 with CONTROLS , where interaction variables other than those involving X3 i,t are removed. To conserve space, only point estimates and corresponding t-stats for STT i,t * X3 t are tabulated. Remarkably, STT i,t * X3 i,t continues to lack explanatory power in both these settings. Column [3] presents estimation results from a variant of Eq. 3, where the coefficient pertaining to X3 i,t is time varying. To put it another way, we relax the assumption that FERC is invariant over time, by interacting X3 i,t with year binary variables. As can be seen, the conclusions are not affected by the consideration of a fully time-varying FERC. Although not reported, we emphasize that the results are similar when restricting the treatment group to French firms and Italian firms separately.
On balance, according to these outcomes, both the French and Italian STTs had no effect on the price informativeness about future profitability of the affected stocks.

Supplementary tests
Several tests were conducted to ascertain the robustness of the results. First, we checked whether the parallel trend's assumption holds, which is crucial to attest the validity of the difference-in-differences estimates. The rationale behind differencesin-differences techniques is that the control group should deliver a suitable counterfactual of the variation of the outcome variable in the treatment group in the absence of exposure to the STT. This is to say that the variation of the FERC in the control group mimics the hypothetical variation of the FERC of treated stocks if STTs had not been introduced. Earlier, we saw that prior to the event, both groups had, on average, equal FERC (e.g., see column (2) of Table 2). Following the insights of Roberts and Whited (2013), a more thorough examination was undertaken, where the "average" FERC during the "pre-shock era" of both groups was compared. Specifically, the pre-event window was subdivided into yearly subperiods to verify whether the FERCs of treatment and control groups followed a similar path.
where D k,t is set to one for observations recorded in year k, and zero otherwise. All other variables are as previously defined. Only pre-event observations (i.e., before 2012) are utilized in the regression. Our interest lies in ̂ 3,k . Figure 1 (left side) allows identification of departures in the FERCs of treated and control units over time. Notably, the FERCs from the two bins moved in tandem during the pretreatment window because neither of the estimated loadings for ̂ 3,k is statistically meaningful. Accordingly, the parallel trend assumption is not rejected in the 3 year time frame prior to the entry into force of the STT.
Next, we tried to break down the effects of entry into force of the STTs during the post-event window, with a view to seeing how the impact on FERC evolved over time in the aftermath of the event. To this end, an approach similar to the one followed by Chen et al. (2017) was carried out: where Year(k) t is a binary variable for which the value of one is assigned in year k. This setting differs from Eq. 3 in two aspects: (i) FERC is allowed to change every year ( ∑ 2016 h=2008 3,h * Year h,t * X3 i,t ) and (ii) the effects of the STT's entry into force on FERC are also permitted to change over time ( ∑ 2016 k=2012 3,k * STT i,t * Year k,t * X3 i,t ). The right side of Fig. 1 presents point estimates for STT i,t * Year k,t * X3 i,t (and corresponding 95% confidence intervals). Interestingly, ̂ 3,k (k = 2012-2016) are not statistically significant, reinforcing previous inferences that the tax event had no effect on price informativeness.
In a different vein, nearest-neighbor matching is also employed. Stocks from the treatment group are paired with stocks from the control group using the Mahalanobis distance. For each treated stock, we selected the top three and five firm-year observations from the control group with nearest distances with respect to two matching covariates (market capitalization and the log of the number of analysts issuing EPS forecasts at the end of the previous calendar year). Non-exempt stocks from France and Italy were paired (with replacement) with their counterparts located in Austria, Belgium, Finland, the Netherlands, Germany, and Spain within two-digit SIC groups. Afterwards, Eqs. 2 and 3 and a more parsimonious variant of those settings were re-estimated in the matched samples. Table 4 exhibits the regression results considering the three and five best matches. Reassuringly, the main inferences are retained with respect to X3 i,t * STT i,t , which remains non-significant, irrespective of the setting and subsample.
Finally, we also conducted cross-sectional tests with respect to size, liquidity, and trading activity. Small-and mid-cap stocks have a smaller investor base. As they are more prone to a decay in investor participation, effects of the tax are likely to be stronger among them. However, we note that the STT is applied only to large stocks. The Amihud measure and traded value are proxies for information asymmetry, heterogeneity of expectations/disagreement, and liquidity. Correspondingly, it is not clear how they may condition price informativeness. If the share of informed investors increases vis-à-vis noise traders, trading costs will eventually surge, but price informativeness may also increase.
The sample is split into two bins of observations based on these stock-specific variables measured prior to the event (pre-event window averages). We formed several subsamples based on whether the stock was positioned above or below the median for each breakpoint variable. Afterwards, Eq. 2 was re-run, conditional on each specific subsample. Table 5 presents side-by-side comparisons of the point estimate on X3 i,t * STT i,t in different subsets. Considering firm size, point estimates on X3 i,t * STT i,t stay statistically meaningless for above and below-median stocks with respect to market capitalization. The interaction variable lacks explanatory power in both bins. We obtained qualitatively similar inferences for the Amihud measure and turnover value. Here, the differences between FERC estimates become wider, though they remain nonsignificant.

3
Securities transaction taxes and stock price informativeness:…

Conclusions
The implications of STTs for market quality constitute an important and controversial topic in financial economics. After the 2008 financial crisis, this issue gained renewed attention, with EU countries discussing the basis for the implementation of an STT across the economic region. In fact, in 2011, the european Commission presented a plan aiming to introduce an STT at the EU level, but no consensus among member states was achieved. Nevertheless, two countries (France and Italy) paved the way for future STTs in the EU and introduced their own versions of the tax. Measuring the impact of such a tax on overall social welfare is a difficult task. Prior empirical studies addressed specific outcome variables and conducted a partial analysis of these events. One specific feature that has not been addressed concerns impacts of the STT on stock price informativeness. By speaking directly to the capacity of stock prices to converge to a firm's fundamental value (manifested in the information content of stock prices about future profitability), we provide an incremental contribution to this literature.
Our results suggest that French and Italian STTs produced neutral effects on price informativeness. A difference-in-difference analysis on the aforementioned events reveals that the future earnings response coefficient did not change with the introduction of the tax. These results survive multiple robustness exercises, including implementation of nearest-neighbor matching approaches, inter alia. Furthermore, inferences are insensitive to subsample decomposition, namely by size, liquidity, and trading activity.
Prior studies on these events detected lower trading activity in the aftermath of the introduction of the tax, with impacts mainly concentrated on illiquid stocks. However, investigation regarding the impact of STTs on liquidity and volatility is inconclusive, although several studies suggest neutral to modest negative effects on those variables (Capelle-Blancard and Havrylchyk 2016; Colliard and Hoffmann 2017; Meyer et al. Table 5 Cross-sectional analysis The table presents the results of the estimation of Eq. 2. Different partitions of the sample are employed, namely with respect to size (market value of equity), Amihud measure, and traded value. The main sample is split into bins (considering whether the firm is above or below the median prior to the event for each variable). To save space, the table only reports point estimates (and corresponding t-statistics in parenthesis) on X3 t * STT t . T-statistics are calculated using heteroskedastic robust standard errors clustered by firm. (***), (**), and (*) indicate statistical significance at the 1%, 5%, and 10% level, respectively

3
Securities transaction taxes and stock price informativeness:… 2015). Our conclusions are, thus, consistent with these prior studies showing residual impacts of the French and Italian taxes. While the topic under analysis has interest on its own, we stress the possibility of other second-order repercussions on the capital market's functioning and the efficiency of resource allocation, which constitutes an interesting avenue for future research.

Appendix: STTs in France and Italy
The French parliament approved the taxation of financial transactions on 14 March 2012, with the vote on the French First Amended Finance Bill, which became effective on 1 August 2012 (Section XX: Taxe sur les transactions financiers-Article 235 ter ZD, French Government (2013)). In the initial announcement, the rate was set to 0.1%. On 4 July, via the 2nd Amended Finance Law of 2012, the tax was increased to 0.2%. The exemptions are the following: • Pure intraday transactions (the tax only applies to daily net position changes, i.e., ownership transfers) • Newly issued shares (primary market operations) • Transactions by clearing houses and market makers • Employee stock ownership plans.
At the same time, the French government introduced a levy on HFT activity. The tax amounted to 1 bp on the notional amount of modified or cancelled messages by HFTs exceeding an order-to-trade ratio of 5:1. The HFT tax applies to transactions conducted by French HFT traders on (all) French stocks. Most HFT traders became exempt, whereby this additional levy produced limited effects and revenues.
The Italian FTT was introduced by the (Italian) Stability Bill (Law 228) on 24 December 2012. The bill was then published on 29 December 2012 in the Italian Official Gazette. The Italian tax is applicable to stocks traded on exchange and OTC issued by companies that reside in Italy with 500 million euros or above market capitalization. Shares traded on exchange and OTC are taxable at a rate of 0.12% and 0.22%, respectively, for 2013 and 0.1% and 0.2% for 2014 and afterwards. Both rates were exceptionally set at 0.12% and 0.22%, respectively, for 2013 only. Securities representing these shares (ADRs, GDRs, etc.) are also under the scope of the STT. The tax is based on net transfer of ownership position at the end of the trading day, like the French STT. material. If material is not included in the article's Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http:// creat iveco mmons. org/ licen ses/ by/4. 0/.