Market impacts of the 2020 short selling bans

At the height of the COVID-19 related market stress in March 2020, six European countries jointly implemented market-wide short selling bans. Utilizing regulatory data on share trading volumes and short positions, and complementing existing literature, we conduct a comprehensive analysis of these bans. Based on a difference-in-difference approach, our estimation finds that the 2020 short selling bans are associated with a deterioration in liquidity and trading volumes, and a decrease in volatility. We do not find evidence that the bans supported nor harmed the prices of banned shares over the application period. Remarkably, the negative impact on liquidity persisted even after the bans were lifted. The deterioration of liquidity appears stronger for shares deemed as liquid, i.e., more pronounced for large-cap stocks, highly fragmented stocks, and stocks with listed derivatives. Sectoral effects are observed for the stocks most affected by the market stress, namely the healthcare and the consumer cyclical sectors. Shares already shorted before the bans saw stronger declines in liquidity and volatility, implying a more challenging price discovery process. Finally, no evidence of a displacement effect between banning and non-banning jurisdictions was observed.


Introduction
As a result of the COVID-19 pandemic, financial markets have been hit by an external shock of unprecedented size in 2020.During the initial stage of the crisis in 1Q20, markets experienced one of the fastest declines in recent history, including surges in volatility and liquidity contractions (Chart 2 and 3 in Appendix 1), testing the resilience of market infrastructures and financial institutions.The STOXX Europe 600 index recorded a peak-totrough fall of -35.5% in February and March 2020.As investor sentiment and equity market performance turned negative, short selling activity -a widespread phenomenon during market downturns -increased from late February 2020, reflecting investors' pessimism (Chart 1 in Appendix 1).
While in regular market conditions short selling can play a role in price discovery, in extreme market conditions it can contribute to accelerate price volatility and to exacerbate market losses.To mitigate possible adverse developments and provide protections to issuers and investors, six European National Competent Authorities (NCAs) (Austria, Belgium, France, Greece, Italy, Spain) coordinated to impose for the first-time a long-term 1 exchange-wide short selling bans on their markets, which started on 18 March 2020 and were lifted on 18 May 2020 as market conditions improved. 2The remaining European regulators opted against implementing a ban on short selling.In parallel, to improve the monitoring of such positions and in accordance with the European short selling regulation 3 , the European Securities and Markets Authority temporarily lowered the reporting threshold for net short positions (NSPs) from 0.2% to 0.1%. 4  This paper provides an impact analysis on the effects of the short selling bans adopted during the first wave of the COVID-19 pandemic in 2020, with the aim to assess their effects on market quality, looking at liquidity, trading volumes, price returns, and volatility indicators, as well as the possibility of a displacement effect between countries which had introduced short selling bans and countries which had not done so.Consistent with Siciliano and Ventoruzzo (2020), we believe that multi-country evidence, rather than individual country data, should be less affected by idiosyncratic effects arising from other country-specific policy interventions that occurred during the crisis period.Thus, the analysis is conducted at the European level, taking into account all European Economic Area (EEA) countries as well as the UK 5 , relying on a difference-in-difference design combined with matching techniques, in order to isolate the cross-sectional effects of the bans6 , similarly to Beber et al. (2018).It follows previous analyses of the possible market impacts of the EU Short Selling Regulation undertaken by ESMA -such as Mazzacurati (2018), who focuses on the concentration level of short selling activities and on how the public disclosure threshold influences the market outcome of net short positions.
The paper complements the existing literature in three ways: first, the analysis is enhanced by the use of regulatory data on European share trading volumes, allowing to present for the first time an assessment of the link between share market fragmentation and short-selling bans, with a significant discrepancy between the bans' impact on shares with low and high trading fragmentation.Secondly, through the use of net short positions data, we examine the bans' impacts on already shorted shares.Lastly, given that the COVID-19 related market stress affected a large number of economic sectors, the proposed sectoral analysis is extended to the sectors that were most affected by the market stress -a new and broader perimeter than the traditional focus of short selling impact analyses (i.e., the financial sector).
Consistent with prior theoretical and empirical work, the 2020 short selling bans are associated with a liquidity deterioration, measured by significantly higher bid-ask spreads (+8% for stocks in banned jurisdictions during the restriction, compared to the control group) and Amihud illiquidity values (+5.8%).The bans also had the effect of decreasing the volumes traded (-14.9%), as well as the volatility of the shares traded under the bans (-8.4%).Effects on abnormal returns do not appear significant.
To check whether the effects of the bans were long-lasting, in an alternative regression framework we find that liquidity remained impacted by the bans after they were lifted (significant higher bid-ask spreads and illiquidity indicator).Similarly, volatility and trading 5 The dataset covers the 30 members of the European Economic Area (EEA) and the UK.Since the data used in the analysis are encompassing the years 2019 and 2020, i.e. before the end of the Brexit transition period, and the amount of trading activities in the UK allow for increasing accuracy of the analysis during the matching process, the UK shares are included in the dataset in the control group of the matching process and in the regressions.volumes continued to be lower in the months following the lift of the bans compared to the exante period, and abnormal returns were slightly lower.
Furthermore, separating our dataset by stock characteristics to estimate whether the bans had differentiated effects according to their ex-ante liquidity, we observe that the negative impact of the bans on liquidity is more pronounced for stocks with a large market capitalisation than for smaller capitalisation.Similarly, taking into account the level of fragmentation of those shares, a significant discrepancy is observed between shares with low trading fragmentation, for which liquidity slightly improved and trading volumes were unaffected; and highly fragmented shares, for which a major deterioration of liquidity and a decrease in volumes is observed under the bans.Additionally, since these bans were explicitly extending the short selling ban not only to changes in position via shares, but also via a derivative instrument, the regression is extended to assess if the impact is different for shares that have derivatives or not, assuming no difference between the sub-samples.The results show that the bans had a stronger impact on the liquidity of the stocks with listed derivatives, with an additional decrease of volumes traded and increasing volatility.
To assess whether sectoral dynamics influenced the effects of the bans, we perform the same analysis focusing on sectors that were especially hit during the COVID-19 crisis.In contrast to other analyses done on previous crises of a financial nature (e.g., Beber et al. 2018), financial sector stocks did not seem to be particularly affected by the bans, which may be explained by the fact that the COVID-19 market stress impacted both the financial sector and the real economy from the outset.The stocks that were at the heart of the crisis, namely from the healthcare sector, saw an important increase in their volatility, trading volumes as well as a surge in their liquidity, suggesting that information revelation was high for them, while shares under the bans from the consumer cyclical sector saw a deterioration of their liquidity and a decrease in their volatility.
Finally, using the net short positions' reporting regime put in place during the market stress, a final regression assesses the effect of the bans on shorted shares.Shares that were shorted before the bans observed an even stronger increase in their illiquidity indicator and larger decrease in their traded volumes compared to the other shares.Moreover, they saw their volatility increase during the ban, signalling more uncertainty regarding their value.
Additionally, to gauge the possibility of a shift of short selling activity from banning jurisdictions to non-banning ones (i.e., a 'displacement' effect), a study of short-sellers activity patterns is presented in Appendix 2. 7 We employ both regulatory short selling disclosures at share level, 7 Given the fact that it is impossible to disentangle the effect of the bans from the effect of deteriorating economic conditions as a consequence of the COVID-19 pandemic, a causal impact of the short selling bans on displacement is not estimated.Nevertheless, descriptive evolution of the position of short sellers provide useful indications as to whether such an effect actually materialised during the bans.and publicly disclosed short positions at position holder level, to analyse the evolution of short sellers' positions during the period of the bans.Short sellers were classified according to their historical behaviour between the start of 2020 and the enactment of the bans, and the evolution of their positions.Both the overall minor increase of NSPs in non-banning jurisdictions, and the absence of observable modification of active short sellers' behaviour before and after the bans, suggest that no visible displacement of short selling activity was observed during the bans.
This empirical analysis seeks to contribute to supervisory convergence in the context of the latest review of the EU short selling regulation8 , with the aim of taking stock of the experience gathered in 2020.It helps promote financial stability and understand the consequences of such policies, and improve their overall efficiency.The Final Report on the review of the short selling regulation9 proposed targeted amendments to improve its operation, such as clarifying the procedures for the issuance of short and long-term bans, the prohibition of naked short selling and the calculation of net short positions and their publication.
The remainder of this paper is organized as follows.Section 2 introduces the paper in the literature, and section 3 presents the data, the matching process, and the regression set-up.Section 4 presents the results, first in the overall set-up and then with specific analysis for the post-bans period, characteristics of the shares, sectoral effects, and for shorted shares.Finally, the analysis on displacement effect is presented in Appendix 2, while Appendix 1 contains additional tables and results.

Literature review
Most of the academic literature on short selling is framed with reference to Diamond and Verrecchia (1987), stating that, under efficient market conditions, short sellers are informed traders and constraining short sales reduces the informational efficiency of prices.Thus, constraining short sales is likely to have a negative impact on market quality.For instance, Saffi and Sigurdsson (2011) document that stocks with higher short selling constraints, measured by low stock lending supply and high borrowing fee, have lower price efficiency, and relaxing the short selling constraints does not lead to instability in the form of a higher probability of large negative stock returns.However, Hong and Stein (2003), modelling heterogenous market agents' behaviour, find that short selling constraints aggravate price declines.This concern -i.e., that predatory short sellers can exacerbate downwards price movements -appears critical during stress conditions or for vulnerable companies: Brunnermeier and Oehmke (2014) find that predatory short selling can be responsible for a higher probability of default, especially for fragile financial stocks and during crisis periods, by contributing to a decline in stock prices.Furthermore, institutional differences and regulatory frameworks appear key in the informational level of short sellers (Boehmer et al., 2022).
These concerns led many regulators to prohibit or constrain short sales during the 2008 financial crisis and the Euro area debt crisis in 2012, providing natural experiment configuration assessing the impact of short selling activities in abnormal market conditions.However, this empirical literature assesses bans that affected financial stocks, introduced over financial stability concerns for the banking system, thus not entirely comparable to market-wide bans. 10 The most encompassing analysis, from Beber and Pagano (2013), covers the short selling bans imposed during the 2008 crisis in 30 countries and indicates that the bans were associated with a decrease in liquidity, with an increase in bid-ask spreads of 1.28 to 1.98 percentage points (compared to a sample average of 4%), and with a significant increase of the Amihud illiquidity indicator.In contrast, the introduction of disclosure requirements in some countries are associated with a significant improvement in market liquidity, i.e. associated with a reduction of 0.65 percentage points in the bid-ask spread.Similarly, Boehmer et al (2013), using intraday data on the 2008 short sale ban in the US, document a deterioration of liquidity and market quality in response to the ban.Beber et al. (2018) also shows that short selling bans imposed both during the subprime crisis and the eurozone crisis had differentiated effects on non-financial corporations and financial institutions, whose banned shares experienced higher probability of default, greater volatility and price declines, particularly for banks.Some NCAs carried out impact evaluations of the 2020 short selling bans.Consistent with prior empirical work in other crisis settings, Lopez and Pastor (2020) and Benhami et al. (2022) compare differences in national market quality between stocks subject to short selling restrictions and stocks not affected by the same restrictions.Comparing the French and Dutch equity markets, Benhami et al. (2022) observe that the ban is associated with a deterioration of the liquidity variables (+7 bps for the quoted spreads of banned shares), a decrease in volatility, with more pronounced effects on large caps, and no significant effect on price returns.Comparing the Spanish and German equity markets, Lopez and Pastor (2020) identify similarly a negative impact of the ban on bid-ask spreads, but an increase in depth based on the Amihud illiquidity indicator. 11Furthermore, the authors do not find evidence that the securities subject to the ban experienced a decrease in their trading volume or volatility, and no significant impact of the ban on prices, CDS spreads nor any specific impact on financial stocks was observed.10 Furthermore, one of the challenging aspects of assessing the impact of short selling bans from an empirical perspective is that the bans themselves (as well as the selection of banned stocks in most of the bans put in place before the 2020 European bans) may be endogenous, and thus the empirical identification of the effect of the intervention on market characteristics is challenging.11 However, since this result is surprising and not in line with the rest of the empirical literature, the authors point out that further research on this issue is necessary, especially since Amihud levels for Spanish securities were higher than their German control group at the time of the ban, a difference that is attributed by the authors as country risk -with higher premia, such as those seen in Spain, resulting in shallower markets.
Extending the analysis by looking at 14 selected EU countries and the UK 12 during the European bans, Siciliano and Ventoruzzo (2020) estimate a significant increase of bid-ask spreads of 14% for banned shares, and a decrease of the inverse of the Amihud illiquidity indicator by 0.1%, with more pronounced effects when separating between financial and nonfinancial stocks (the bid-ask spread of financial firms increased by 25.6% versus 14.4% for non-financial firms during the banned period).
Finally, making use of the publicly disclosed short selling positions from NCA websites in 2019 and 2020, Greppmair et al. (2020) study how short sellers evaluate the importance of fiscal space for individual companies.Their results indicate that short sellers adapted quickly, since short selling activity shifted upon the onset of the pandemic towards companies with low financial flexibility and in countries with limited fiscal space.Consistent with the notion that short sellers are informed investors, they entered their short positions prior to the enactment of the bans and many of these positions were closed before the end of the market crash.This confirms that short sellers were quick to incorporate complex information during market stress period, either from their ability to process available information (Engelberg et al., 2012, Reed, 2013), or with the help of potential private information (Boehmer et al., 2018, Boehmer et al., 2020).Indirectly, this confirms the findings of our descriptive analysis on the displacement effect, namely that no material changes of NSPs took place from banning to non-banning jurisdictions during the bans period. 13This strategy was profitable, as shorted firms with low liquidity buffers in low-rated countries experienced an abnormal return of -10% during the period of COVID-19 related market stress.
Furthermore, the European bans imposed from March to May 2020 stated clearly that not only did the prohibition apply to shares, but it also prohibited any transaction which might constitute or increase a NSP on shares, involving any type of financial instruments including saving/preferred shares, index-related instruments, depositary receipts, and derivatives. 14 During the global financial crisis in 2008, most bans did not explicitly rule out the possibility of carrying out synthetic short strategies -for example through options. 15However, this substitute is costly and typically available only to sophisticated investors.This aspect is relevant since, Analysing the US stocks during the 2008 financial crisis, Kolasinski et al. (2013) documents 12 The rationale behind the selection is not presented in the paper per se; however the 15 countries appear to be the biggest EU countries for share trading at the time (6 banning countries: Austria, Belgium, France, Greece, Italy, Spain; and Denmark, Finland, Germany, Ireland, Luxembourg, Netherlands, Portugal, Sweden, United Kingdom).13 See Appendix 2.
14 Since the NSP is calculated as the sum of all short positions minus all long positions held by an investor in relation to restricted shares and related instruments, all relevant positions were prohibited; stemming from the purchases or sales of shares, options, swaps, futures, depositary receipts, index-related instruments, covered warrants, certificates and any other structured product whose effect is to create a NSP in a share subject to the ban.
Since the global financial crisis bans and in accordance with Regulation (EU) No 236/2012 on short selling, most European bans put in place during the sovereign debt crisis were prohibiting all transactions in any financial instrument that create, or increase, a net short position, meaning including derivatives.
that the ban decreased market liquidity and increased the informativeness of short sales, and that both these changes were particularly strong for stocks with listed options.Similarly, Beber and Pagano (2013) observed more stringent effects of the bans on liquidity for small-cap stocks and for stocks that do not have listed options, suggesting that the availability of an option market allows investors to express short views on the banned stocks.However, since the 2020 European bans did not allow for the possibility to use the derivative markets, their effects on share liquidity should not be affected by the availability of listed derivatives.Thus, we test for this hypothesis by looking at the differentiated effect of short selling bans on shares with or without listed derivatives, and expect no difference between the two sub-samples.
Finally, an important feature of the SSR regime is the market-making exemption that implies that a number of market participants, mainly large banks, are still allowed to take short positions.Indeed, while bans may effectively curtail speculative short selling behaviours, the market making exemption means that the bans do not constitute a full constraint on short sales, an assumption on which the empirical literature analysing the short selling bans in 2008 is based.

Dataset construction
The following analysis focuses on the impact of the two months exchange-wide bans from 18 March to 18 May 2020, employing data from 13 January (2 months before the ban) to 31 July 2020 (2 months after the ban).The days of the short-term bans, on 13 and 17 March 2020, are excluded. 16The regressions are estimated on daily data, first for all EEA and UK shares 17 , and then for selected samples of the dataset, to investigate whether the liquidity impacts vary with share characteristics.
The starting list of shares is created using the European Financial Instrument Reference Data System (FIRDS), from which the relevant instruments traded in Europe and UK are extracted for the full year 2019 and 2020 (common/ordinary shares, depositary receipts on equities, and preferred/preference shares based on the Classification of Financial Instruments), taking into account possible ISIN changes.Shares terminated before 2020, or for which the last received data is before 2020, as well as shares which have their principal trading venue located in a 16 The one-or two-day bans imposed in March are too short-lived for a dedicated econometric analysis, and including those days in the data will impact the results.17 Since the data used in the analysis are encompassing the years 2019 and 2020, i.e. before the end of the Brexit transition period, and the amount of trading activities in the UK allows for increasing accuracy of the analysis during the matching process, the UK shares are included in the dataset as a control group.third country outside of the EEA are excluded. 18For all these instruments, the relevant variables are extracted using ESMA Financial Instruments Transparency System (FITRS) and FIRDS databases, as well as market data.Table 7 in Appendix 1 provides descriptive information of the variables used and their definition.
Penny stocks, i.e., shares with an average price below EUR 1 in 4Q19 or in 1H20, are excluded from the sample, as well as new stocks without data in 2019.Furthermore, shares with daily price returns equal to zero during more than 30% of our timeframe (signalling stale prices) are also excluded.Finally, to deal with possible data quality issues, we exclude shares with negative, missing, or extreme average bid-ask spreads (higher than 10% during 2020), and the data is winsorized by eliminating the observations corresponding to the top and bottom 1% of bid-ask spreads.This process results in a removal of 4,935 stocks out of 8,290, corresponding to 60% of our starting sample and to 23% of its market capitalization.Each step of this cleaning process is described in Table 8 in Appendix 1.The final sample is composed of 3,355 stocks, 911 belonging to the treatment group (stocks from countries with short selling bans) and 2,444 to the control group.Table 9 in Appendix 1 provides descriptive information of the final sample by country.

Matching process
To increase the robustness of the analysis, sample matching is added as a pre-analysis step: the goal is to balance the treatment and the control groups (i.e., to achieve similar share characteristics in the two groups).Matching is a statistical technique used to construct a comparison group in order to enable a comparison of outcomes among treatment and control groups while controlling for possible confounding factors (Gertler et al. (2011)).
Even though several matching techniques can be employed, we focused on two: Nearest Neighbour Matching (NNM) and Coarsened Exact Matching (CEM). 19NMM selects the closest counterpart of each observation based on a weighted function of a defined set of covariates (i.e. a 'distance' measure), while CEM performs exact matching on a set of covariates, 'coarsening' the continuous ones into strata and discarding the strata that do not contain at least one treated and one control observation.In the final estimation, only the retained observations are used, but they are weighted by the size of the corresponding "stratum".

18
In the same manner, shares that are traded on a Regulated Market and that have a foreign issuer are excluded, in order to remove non-EEA depository receipts or shares that are from a foreign issuer and thus have very high market cap while not being really traded in the EEA.

19
Propensity Score Matching (PSM) was assessed; however, since our setting did not satisfy the required assumptions, it was deemed not suitable for the analysis.The Synthetic Control Methods (SCM) approach was also considered but given our large sample of analysis it would have been impractical to construct a 'synthetic' control unit for each treatment share.
Hence, this method allows for multiple control observations to be matched to a single treated observation, with weights correcting any potential imbalance of observations.Both methods have been evaluated based on (i) the balance of results, measured by comparing share characteristics prior to and after the matching, (ii) the impact of each method on the sample size.In terms of balance, CEM is more powerful than the NMM method, but it is also more restrictive and has a higher chance of reducing the sample size. 20Chart 4 in Appendix 1 shows the standardised mean differences between the treatment and control groups after carrying out both CEM and NNM matching in our sample.In this setting, CEM achieves a more balanced outcome, as shown by the fact that absolute standardised mean differences between treatment and control group for all matching variables are below the standard threshold value of 0.1.NNM method results in greater imbalance for specific information, despite performing well overall.This comes at the expense of a contained sample reduction: from a starting point of 911 treatment shares, CEM matches 858 (94%) while NNM matches all of them.
Based on these reasons, CEM has been chosen for the matching procedure, resulting in a final sample composed of 1,716 stocks, half belonging to the treatment group and half to the control group.Among a range of potential variables considered, the following set of information has been employed for the matching procedure 21 : • average market capitalisation in 4Q19, in order to avoid any confounding impacts due to the pandemic and to the short selling bans; • the share sectoral information, i.e., the classification of economic sectors from Refinitiv Eikon; • liquidity status, using the liquidity assessment from ESMA transparency calculations, based on the free float, average daily number of transactions and average daily turnover at the share level for 2019 and 2020. 22 Table 10 in Appendix 1 shows the differences between the treatment and control groups, before and after the matching, and Table 11 presents the matched sample by country.Furthermore, to fulfil the parallel trends assumption needed for the difference-in-difference analysis, Charts 5 to 8 in Appendix 1 present the graphical evolution for both groups after the 20 Due to the coarsening of continuous variables, using CEM the observations for which a suitable match is not found will be dropped; on the other hand, the 'distance' approach is less powerful but also less restrictive in terms of observation filtering.

21
The list is not exhaustive, and many potential alternatives could be considered.However, these choices are in line with related works in the literature and, in our opinion, guarantee a balanced outcome without losing a significant number of observations due to missing data.

22
According to MiFID II/MiFIR, a share is liquid if all of the following conditions are satisfied: (a) the free float of the share is not less than EUR 100 million for shares admitted to trading on a regulated market; (ii) the market capitalisation is not less than EUR 200 mn for shares that are only traded on MTFs; (b) the average daily number of transactions in the share is not less than EUR 250 mn; (c) the average daily turnover for the share is not less than EUR 1 million.
matching process and show that the distance between the treatment and control group remains constant over time prior to the bans' enactment.Finally, Table 1 below presents a description of the main variables of the dataset, after the matching process.

Regression set-up
The difference-in-difference analysis of the impact of short selling bans on our set of dependent variables employs the following baseline regression model: where: •   is one of the dependent variables for which the regression is estimated, i.e. bid-ask spreads, Amihud illiquidity indicator, abnormal return, intraday volatility or volumes traded. is a constant,  represents the share,  the country and  the time index.
•   is a dummy variable equal to one over all trading days for shares in banned countries.
•   is a dummy variable equal to one for all shares during the validity of the short selling ban.
•   *   is the interaction variable which isolates the effect of the treatment on the affected stocks.Its estimated coefficient is, hence, the most important coefficient of this regression.
•   is the set of control variables, either at stock level (market capitalisation, fragmentation indicator, volumes traded, intraday volatility) or at country level (stringency index).Additionally, VSTOXX is added as a proxy for the evolution of volatility on European markets.
•  are the fixed effects included in the regressions.Stock fixed effects control for unobserved variables linked to each stock, and time fixed effects take into account the commonality in liquidity or price returns.We present the regressions with both fixed effects combined (two-way fixed effects), keeping in mind the issues caused by multicollinearity. 23  The difference-in-difference model estimates the impact of the bans on five main variables of interest: two variables to assess the liquidity of the equity market (bid-ask spreads and the Amihud illiquidity indicator), abnormal returns (to represent the evolution of prices), volumes traded, and a volatility measure.Standard errors are clustered at the stock level, following Beber and Pagano (2013).
Following Degryse et al. (2015) and ESMA (2017), we compute the bid-ask spreads using the closing ask and bid prices for each share (the difference between the highest price a buyer is willing to pay for a share and the lowest price a seller is willing to accept), and we normalise daily bid-ask spreads using the following formula to correct for nominal differences and make reliable comparisons across companies and countries: The Amihud illiquidity indicator is the second measure of liquidity considered.It calculates the daily ratio of the absolute value of a stock return to its dollar volume, and thus serves as a rough measure of price impact: Increasing values of Amihud indicate that the price return is less affected by trading volumes, and thus higher values indicate less-liquid stocks.
The volumes traded variable is based on the quantity of share traded daily, in order to observe the evolution of trading volumes, but separating this variable from price evolution.We normalise the bid-ask spreads, trading volumes and Amihud using a log transformation.
We calculate abnormal returns, which allow for an assessment of prices evolution and stock over/under-performing with respect to its reference market benchmark, by comparing for each 23 When adding one or multiple fixed effects, share-invariant or time-invariant variables will be dropped from the regressions, as well as the variables 'Event' or 'Treatment' in the regression above.
share its daily price returns with the price returns of the benchmark national index multiplied by the share market beta 24 , using: Finally, in order to increase the robustness of the analysis, we use two different measures of volatility: first, we compute historical volatility as the standard deviation of a stock's log returns over two days.Second, we calculate the so-called high-low range volatility, or intraday volatility, based on daily high and low trading prices, using Parkinson's (1980) approach: where ℎ   is the stock s highest trading price on day t, and    is the stock s lowest trading price.
Furthermore, the following control variables were included: •    , to control for firm size.As larger firms generally benefit from larger coverage by financial analysts, they tend to have larger trading volumes and possibly higher market liquidity.
•    , calculated as the inverse of the Herfindahl-Hirschman Index, which is a widely used measure to determine the concentration of a market.Fragmentation can have a significant impact on market liquidity, since higher fragmentation can improve liquidity aggregated over all trading venues, but may lower liquidity in the reference market.The indicator is calculated daily at share level for onexchange trading and is defined as: , with M being the total number of venues that displayed trading in share s.
•   , to control for daily volatility, •    , a daily measure of the strength of containment policies linked to the spread of the pandemic, which controls for the impact of the pandemic at the country level.This variable is measured as the daily average for each country of nine indicators pertaining to containment and lockdown policies (Oxford COVID-19 Government Response Tracker 2021).

24
The market beta is the monthly historical beta calculated by Refinitiv Datastream for each share, and the variable employed as benchmark is the return of each national index of reference, using daily price from Refinitiv Datastream.
• Following Beber and Pagano (2013), we also control for volatility among the explanatory variables, since it can affect market liquidity by changing the inventory risk of market makers.
• Other variables, such as turnover, Return on Equity, Return on Assets, Economic sentiment at the country level and country risk were tested but did not affect results or add value, and hence they are not presented.
The full dataset covers 1,716 European stocks between 13 January 2020 (i.e., two months before the ban) and 31 July 2020.Regarding the chosen timeframe, the same regressions were estimated using different timeframes (both shorter and longer timeframes were tested) with similar results, and these alternative estimations are not presented here.The paper presents the results of estimating the model for the proposed timeframe and, in addition, an alternative specification obtained by adding a dummy variable for the period after the ban, to assess whether the effects of the bans persisted after its termination (see subsections 4.2).
In line with previous empirical studies, the hypothesis tested in the analysis is that the imposition of a short selling ban, by preventing potentially informed investors to take new short selling positions, will slow down the price discovery process, and that such delayed resolution of uncertainty about fundamentals will decrease the liquidity of shares (measured here with bid-ask spreads and the Amihud illiquidity indicator). .Existing studies document mixed evidence on the relation between short selling and volatility during short sale constraints.Bans targeting specific stocks and/or sectors can amplify concerns about specific shares, amplifying volatility, heightening uncertainty, and increasing information asymmetry about the fundamentals (Liu (2015)).Boehmer et al. (2013) find an increase in volatility in US stocks under the short sale ban in place in 2008, suggesting that this volatility might stem from deteriorating market quality, or mirror greater instability in the underlying fundamentals during that period.However, a ban on short selling can also mechanically restrict price movement and thus volatility (Barardehi et al. (2023)).Since the usual objective of regulators and policymakers that introduce short selling ban during periods of financial distress is the decrease of volatility, past bans were also usually share-specific, meaning targeting specific shares of sectors, but seldom market-wide.This is the first-time 6 important European equity markets close all possibilities of short selling, including by explicitly prohibiting other type of short selling position through derivatives' use, for instance.Thus, the effects of those sharespecific bans on volatility may be different.The hypothesis that we test here, since the European bans of 2020 are market-wide and concern all shares, is a decrease of short-term volatility, as well as trading volumes.
The expected effects on prices are more ambiguous: if the ability to short sell stocks increases the informational efficiency of market prices (see Saffi and Sigurdsson (2011)), a constraint on short selling, by slowing down price discovery, will be expected to sustain prices.However, short-sales constraints, by increasing the risk perceived by uninformed investors, can lead them to require higher expected returns (inducing lower prices), or lead to negative information not being incorporated into share prices under the ban, which would aggravate the price decline after the end of the bans (Hong and Stein (2003)).
4 Effects of the bans on market quality

Main results
The effects of the short selling bans on market liquidity for the concerned shares appear to be negative, as indicated by the sign and the statistical significance of the main variable of interest in the regression, the interaction between Treatment and Event (Table 2).
The short selling bans are correlated with a deterioration of the bid-ask spread of the concerned shares: the regression coefficient (0.077) is statistically significant and implies an average increase of 1.08 (= 0.077 ), meaning bid-ask spreads increased by 8% for stocks in banned jurisdictions during the restriction, compared to the control group.Similarly, the coefficient for Amihud is significant and the bans are associated with an increase of 5.8% of the Amihud illiquidity indicator, confirming that the bans reduced liquidity of the concerned shares.To gauge the economic significance of the impact, we compare those numbers with the standard deviation of daily variable changes before and during the market stress periods, as a point of reference.In countries that did (not) implement a ban, daily bid-ask spreads standard deviation in the last quarter of 2019 was 1.5% (2.3%), but 12.3% (8.8%) during the first quarter of 2020.Similarly, for the Amihud illiquidity measure, the standard deviation of daily Amihud indicator was 1.5% (2.3%) in countries that did (not) impose a ban in the last quarter of 2019, and 12.3% (8.8%) in the first quarter of 2020.The bans also had the effect of decreasing the volumes traded, with the significant coefficient showing an important decrease of -14.9% for shares traded under the ban.
Moreover, the impact on abnormal returns appears non-statistically significant and, furthermore, the adjusted R-squared for this dependent variable is small, showing that the percentage of the variance of abnormal returns explained by our model is minor.This suggests that the bans did not harm nor sustain market prices over the enactment period.In the literature, the effectiveness of short selling bans in supporting stock prices is ambiguous.Looking at excess returns during the 2008 financial crisis, Beber and Pagano (2013) show that the bans have not been associated with better stock price performance globally, with the US being the only exception.Similarly, Siciliano and Ventoruzzo (2020) estimates that shares' excess returns in the banned period in 2020 were, on average, 0.1% lower compared with firms in European countries that did not impose short selling bans.
Finally, the volatility analysis highlights that shares in banned countries exhibited a lower degree of volatility during the ban period.The coefficients displayed in Table 2 imply a statistically significant reduction in volatility: compared to the sample average and the sample median in non-banning days (equal to 2.90 and 3.07, respectively), the coefficient implies a decrease of -8.4% (i.e., -0.245/2.9)and -8.0%(i.e., -0.245/3.07),respectively.Since the results for the historical volatility measure are similar, only the intraday volatility model is presented.These results confirm that the bans attained the objective expected by the policy-makers: a stabilization of price movements.

Post-bans effect
Since market conditions did not normalise right after the lift of the bans, and to investigate any long-lasting effect of the bans, we propose an alternative specification of our baseline difference-in-differences analysis which considers separately the post-bans period, according to the following specification: Where  is a dummy variable taking value 1 from 19 May to end of July (the post-bans period).Interacting this variable with the treatment allows to differentiate between the effects of the bans period itself and any prolonged effects afterwards (coefficient );the latter are to be compared to the pre-bans period.
The estimates in Table 3 show that even though the period of the bans had the stronger impact on liquidity variables, the period immediately afterwards is also associated with significantly higher bid-ask spreads (+3.8%) and illiquidity (+9.7% for the Amihud illiquidity indicator) in countries where a ban was implemented.Volatility and trading volumes continued to be lower for the shares constrained by the bans after it was lifted (respectively -3.9% and -4.6%). 25The effect on abnormal return is significant, albeit small (-0.2%). 26

Differentiated effects of the bans by share characteristics
In order to assess whether short selling bans had differentiated effects on the liquidity of stocks with specific characteristics, we repeat the analysis taking into account different groups of stocks (Table 4).
Firstly, we separate between the smallest (small-cap shares) and the largest (large-cap) market capitalisation of our sample 27 , since liquidity is lower for small-cap stocks even in the absence of short selling constraints.Boehmer et al. (2013) argue that lower impacts of short selling restrictions on small-cap stocks should be expected, given that small stocks experience 25 The impact on volatility is calculated here by comparing the regression coefficient with the sample median observed during the pre-ban period (-3.9%=-0.117/2.98).

26
Similar results are observed in the same regression without controls and presented in Table 13 in Appendix 1.

27
The regression is estimated separately for each quartile of the companies by capitalization, and the first (small-cap) and last (large-cap) quartiles of the dataset in terms of market capitalisation are presented in the table.minor changes in the amount of shorting during the ban, and that the liquidity supply provided by market makers is less pervasive for small-cap stocks than for large-cap.While this explanation might not be entirely comparable to the European situation during the COVID-19related market stress, it can be expected that in countries where large-cap stocks are overrepresented, the bans had stronger impact and was associated with larger increases in bid-ask spreads and in the Amihud illiquidity indicator, as well as stronger changes in volumes traded and volatility.
The results in Table 4 confirm that the bans greatly affected the liquidity of large-cap shares, with a strong increase of their bid-ask spreads (+15.9%),whereas they did not have significant impacts on small-cap shares.The Amihud illiquidity indicator corroborates that the adverse liquidity effect of bans was more pronounced for large-cap shares, with a statistically significant increase of the Amihud illiquidity indicator for large-cap of +32.7%, and for small-cap of -17.1%, i.e., an increase in liquidity for small-cap shares (Table 4).Volumes traded increased significantly for small-cap shares traded under the bans (+10.8%)but decreased markedly for large-cap (-31.3%).Volatility decreased significantly for small-cap (-7.9% compared to the small-cap median in non-banning days) and did not vary significantly for large-cap (Table 14 in Appendix 1).Furthermore, to assess whether the bans affected shares differently depending on their fragmentation level, i.e., how their trading volumes are distributed across trading venues in Europe, we introduce a market fragmentation indicator calculated as the inverse of the Herfindahl-Hirschman Index for volumes traded by venue, and separate shares with and high and low market fragmentation. 28  The impact of the bans on bid-ask spreads does not appear statistically significant for shares with low trading fragmentation, meaning shares for which trading is heavily concentrated on one or few venues.However, bid-ask spreads widened significantly for shares with high trading fragmentation (+14.7%).The same discrepancy in liquidity deterioration is observed looking at the Amihud illiquidity indicator, with low fragmentation shares seeing an improvement of their liquidity (-9.6% of the illiquidity), while highly fragmented shares saw a major deterioration (+34.8%).Similarly, highly fragmented shares observed an important decrease of their traded volumes under the bans (-30.8%,Table 14 in Appendix 1), while the bans had no significant effect on trading volumes for low fragmented shares.Neither abnormal returns nor volatility were significantly affected by the bans regardless of the degree of fragmentation.
Finally, since the 2020 short selling bans are explicitly extending the constraint not only to shares, but also their derivatives 29 , we modify our baseline regression to assess the impact for shares that have derivatives.Without the extension of the bans to derivatives, the effects of short selling restrictions on bid-ask spreads should be stronger for stocks without derivatives, as Beber and Pagano (2013) concluded.Therefore, we anticipate no distinction between the two groups in the 2020 bans setup.In order to investigate whether the availability of derivatives plays a role in the liquidity effects of the bans, we create a dummy variable for shares for which there are listed derivatives (options, futures, and warrants) in FIRDS.To investigate differentiated effects for subgroups of shares (i.e., shares with listed derivatives and, as described in the following sections, shares with net short positions prior to the bans and shares belonging to specific economic sectors), we perform the following triple difference regression: Where variable   is the dummy identifying whether a share s belongs to the subgroup of interest.In this setting, the coefficient of interest is , i.e., the one related to the triple difference (the interaction between treated group, treatment period and relevant subgroup).Moreover, since we observe that stocks with listed derivatives are usually large-cap, we estimate the regression only on the shares with the largest market capitalisation (i.e. the third and fourth 28 Similarly, low (highly) fragmented stocks are defined as the first (last) quartiles in terms of market fragmentation of the dataset.

29
During past short-selling bans, the use of derivatives was not always explicitly constrained, leaving room to effectively take short positions by trading in the option markets.Beber and Pagano (2013) However, the 2020 bans explicitly ban the creation or increase of short net positions, that applies to shares but also to all related instruments included in the calculation of net short position, like saving/preferred shares, index-related instruments, listed and OTC derivatives, depositary receipts, covered warrants, etc. quartiles of market capitalisation in the sample), allowing us to single out the effect of having listed derivatives from the market size effect.
Results show that the bans had a stronger impact on the liquidity of the stocks with listed derivatives, with an additional deterioration of the illiquidity indicator (+18.3%), an additional decrease of volumes traded (-14.4%) and increasing volatility (+23.2% compared to the median value in non-banning days), but no statistically significant effect on the bid-ask spreads.

Differentiated effects of the bans across sectors
To assess whether sectoral dynamics influenced the effects of the bans, we perform the same analysis focusing on sectors that were especially hit during the COVID-19 crisis.Most of the empirical literature on short selling bans analysed their effects during the past financial crisis, when bans were prompted by concerns about the stability of financial institutions, and in some countries targeted only financial institutions' shares.Thus, this literature focused on the differentiated effects of the bans on financial institutions, observing stronger negative impact on financials (Marsh and Payne (2012)).
However, the COVID-19 market stress was different, and so was the distribution of its adverse effects across firms and sectors of the economy.Thus, to test whether the sectoral effects were present during the COVID-19 related market stress, sectoral dummies were added to the baseline regression drawing on the triple difference setting introduced in the previous section.In particular, we look at the sectors that were mostly affected by the crisis, using the others as a benchmark.During the pandemic the stock market has priced in various dimensions of resilience, including each industry's immunity to social distancing requirements (Pagano et al. (2020)).To take this into account, the proposed benchmark group is constituted by the industrials, materials, consumer non-cyclical, energy and utilities sectors; while the sectors highlighted in the regression, and thus deemed as mostly affected by the market stress, are the financials, consumer cyclical, healthcare, real estate, technology and telecom sectors. 30 This latter group constitutes 38% of the shares in the matched dataset and 40% of its market capitalisation, assuring the representativeness of the results.
The first finding is that contrary to other analyses covering past crises of a financial nature (such as Beber et al. 2018), the financials subset did not seem to behave differently than other stocks during the COVID-19 market stress, with no significant effect for the shares under the bans except for an increase in bid-ask spreads (+7.5%).This might be explained by the fact that this crisis impacted both the financial sector and the real economy from the outset, which 30 The proposed list of groups is based on the observed returns for each sector in February and March 2020, as well as on the literature.However, the sectors deemed as "resilient" the COVID-19 related market stress can be subject to discussion and adjustments, and are constrained by the level of granularity accessible in the analysis.
were also both supported by accommodative monetary policies and fiscal support at the European and national levels.Furthermore, strong effects are observed for some of the sectors that were mostly hit by the pandemic-related market stress.The healthcare sector is the most impacted by the bans, with an improvement of liquidity (decrease in bid-ask spreads by -8.3% and in the illiquidity indicator by -15.6%), an increase in volatility (24.6%) and in volumes traded (+15.7%) for those shares during the bans period.On the contrary, shares under the bans from the consumer cyclical sector saw a deterioration of their liquidity (+12.0% of the Amihud indicator), a decrease in their volatility (-14.9%),but an increase in their traded volumes (+12.0%)during the period.Shares from the real estate sector saw a decrease in their volatility (-10.4%).The technology and telecom sectors were not significantly affected by the bans, suggesting that other characteristics have more importance in explaining the liquidity evolution of those shares under the ban.The impact on abnormal returns appears as non-statistically significant for all sectors.
These results confirm that, contrary to past crisis of a financial nature, where short selling bans affected in particular financial stocks, this crisis was of unprecedented nature, with many sectors affected by the market stress and by the bans.In this configuration the stocks that *** p<0.01, ** p<0.05, * p<0.1.NA: Not Applicable -these variables were not included in the regression.The presence of stock and day fixed effects may result in dummy variables (here, VSTOXX) to be removed from the estimation due to multicollinearity.Sources: ESMA.
were at the heart of the crisis, namely from the healthcare sector, saw an important increase in their volatility, trading volumes as well as a surge in their liquidity, suggesting that information revelation was high for them.

Differentiated effects for shorted shares
Finally, to assess whether shorted shares were more heavily impacted by the bans, we identify those shares with short selling activity prior to the bans and modify our baseline regression according to the triple difference framework introduced in Section 4.3.Shares with short selling activity in this setting are defined as those for which a positive net short position has been notified in the days before the implementation of the coordinated European bans, that is for which at least one net short position holder reported a position equal to or higher than 0.1% of the issued share capital of the company.Information on shorted shares comes from regulatory data on aggregated net short position per issuer, notified to NCAs following the requirements introduced by the Short Selling Regulation. 31The comparison between these two groups is feasible as it is based on pre-bans information, and aims at exploring whether the long term bans had any effects on those shares that were subject to short selling activity in the earlier phase of the COVID-19 market stress without considering how NSPs evolved in non-banning jurisdictions.Furthermore, considering the findings of the descriptive analysis on displacement effect (see Appendix 2), we believe the absence of a shift in short selling activity between banning and non-banning jurisdictions validates our approach.
The results show that shares that were shorted before the bans saw an important increase in the illiquidity indicator (+27.3%),confirming that the bans reduced their liquidity, but no significant effect on their bid-ask spreads.Contrary to other subgroups, shorted shares saw their volatility increase (+14.7% with respect to the sample median in non-banning days), while their traded volumes decreased even further compared to shares that were not shorted prior to the bans (-22.7%).This effect can be explained by the suppression of negative information in the price discovery process, which may have created more uncertainty regarding their value in the absence of short selling possibilities.31 Shorted shares are defined as those for which at least one positive net short position was notified between the 11 and 18 March 2020, no matter its value.According to Article 5(2) of SSR, any natural or legal person who has a net short position in relation to the issued share capital of a company shall notify the relevant competent authority, where its position reaches or falls below a relevant notification threshold.During the COVID-19 market stress, ESMA took the decision to lower the notification threshold from 0.2% to 0.1%, 2020, 0,2 % of the issued share capital of the company concerned and each 0,1 % above that.The European Commission adopted the decision to permanently lower the threshold from 0.2% to 0.1% in September 2021.

Conclusion
Using regulatory data on share trading volumes and short positions, we analyse the equity market impacts of the European long-term short selling bans of 2020 on liquidity, trading volumes, stock prices and volatility.In addition, we assess the link between the bans and market fragmentation, and extend the analysis to the sectors most affected by the market stress.
Our results suggest that the short selling bans entailed a deterioration of market liquidity and a decrease in the volatility of the shares concerned.Curbing short selling activity with the purpose of avoiding disorderly downward price spirals appears to have had, as a collateral effect, a significant reduction in trading volumes and liquidity for banned shares.In line with our earlier findings and the literature on the subject, constraining short sellers from opening short positions contributed to higher bid-ask spreads and higher Amihud illiquidity values.At the same time, considering the uncertainty linked to the COVID-19 market stress, volatility decreased under the bans, which is consistent with the stated objectives of policy-makers.Furthermore, the econometric analysis undertaken did not identify statistically significant correlations with abnormal price returns, suggesting that the policy did not harm nor sustain market prices over the enactment period.
To further refine the impact assessment of short selling bans on liquidity, we take into account the fact that bid-ask spreads may be affected by stock-specific characteristics.The negative The presence of stock and day fixed effects may result in dummy variables (here, VSTOXX) to be removed from the estimation due to multicollinearity.Sources: ESMA.impact of the bans on liquidity is more pronounced for stocks with a large market capitalisation, highly fragmented shares, shorted shares, and for stocks with listed derivatives.The stocks that were at the heart of the crisis, namely from the healthcare sector, saw an important increase in their volatility, trading volumes as well as a surge in their liquidity, suggesting that information revelation was high for them during the bans.
To conclude, we employ both short selling disclosures at share level and publicly disclosed short positions at position holder to carry out a descriptive analysis of the possibility of a shift in short selling activity from banning jurisdictions to non-banning ones.Both the overall minor increase of short selling positions in non-banning jurisdictions, and the absence of observable modification of active short sellers' behaviour before and after the bans, suggest that no visible displacement of short selling activity was observed during the bans.This empirical analysis contributes to supervisory convergence in the context of the latest review of the EU short selling regulation 32 , with the aim of taking stock of the first coordinated market-wide short selling bans, and improving the overall efficiency of such policies.NSPs from March to May 2020 is only 15 basis points, when the observed increase from February to March was 27 basis points, suggesting there is no clear displacement effect of short selling bans or reversal of NSPs towards non-banned shares.
As a second step, the activity of short sellers is examined through publicly disclosed NSP data, with the purpose of understanding whether the short selling bans impacted the behaviour of short sellers with high NSPs (i.e., short sellers reporting NSPs above the 0.5% threshold).
During the short selling bans, public NSPs decreased by 23% in banning countries and by 5% in non-banning countries (Chart 9).The number of overall active position holders declined moderately -from 207 to 200 (-3%) (Chart 10).In addition, the number of active position holders dropped from 99 (174) to 86 (170) in banning (non-banning) countries, a percentage decline of -13% (-2%).Overall, these numbers do not convey a displacement effect from banning countries to non-banning countries.

Number of publicly disclosed NSPs Number of position holders with public NSPs
To check for further short selling pattern, position holders are grouped according to their historical behaviour between January 2020 and the enactment of the short selling ban, in order to obtain three classes: 1. 'Ban preference': holders that detain 50% (or more) of their positions in banned countries, on average.The number of investors classified in each of these three groups, and its evolution, is graphically described in Chart 11.Based on this classification, the evolution of NSPs and market exposure of these classes of position holders is analysed and is summarized in Chart 12 and 13 below. 35  Chart 12 Chart 13 'Ban preference': number of publicly disclosed NSPs 'No ban preference': number of publicly disclosed NSPs Despite the drop in outstanding NSPs in banning jurisdictions observed during the bans (Chart 11), investors with a preference for banning countries did not relevantly modify their shorting activity, and on 18 May 2020 still held 58% of their positions in banning jurisdictions, compared to 63% on the day of the bans enactment.Moreover, no significant impact is observed for the 'no ban preference' group, which displays a slight decrease in NSPs for both banning and nonbanning jurisdictions (Chart 12).Thus, these figures suggest that many investors who had already positioned themselves prior to the bans did not massively shift their NSPs from banning to non-banning jurisdictions as a consequence of the ban.
On the other hand, for the investors who were not active before the start of the short selling bans it is not possible to determine an ex-ante their propensity towards either banning or nonbanning countries (hence the 'no preference' group).During the bans, these short sellers had no choice but to take short positions in non-banning jurisdictions (Chart 14).Thus, for these investors the bans acted as a constraint on their short selling preferences: as soon as the bans end, their exposure to banning jurisdictions starts to increase.Overall, the number of position holders belonging to the 'no preference' group was lower than the other two groups during the 35 The same classification exercise has been carried out by employing the market value of outstanding NSPs rather than the number of NSPs, leading to comparable results.ban period (Chart 13), and the number and market exposure of their NSPs was also less important in size.
In conclusion, the analysis of publicly disclosed short selling data does not point towards a relevant impact of the short selling bans on position holders' preferences.The charts and analyses in this report are, fully or in part, based on data not proprietary to ESMA, including from commercial data providers and public authorities.ESMA uses these data in good faith and does not take responsibility for their accuracy or completeness.ESMA is committed to constantly improving its data sources and reserves the right to alter data sources at any time.The third-party data used in this publication may be subject to provider-specific disclaimers, especially regarding their ownership, their reuse by non-customers and, in particular, their accuracy, completeness or timeliness, and the provider's liability related thereto.Please consult the websites of the individual data providers, whose names are given throughout this report, for more details on these disclaimers.Where third-party data are used to create a chart or table or to undertake an analysis, the third party is identified and credited as the source.

2 .'
No ban preference': holders that detain 50% (or more) of their positions in non-banned countries, on average.3.'No preference': holders that became active only after 18 March 2020 and, thus, cannot be classified in either of the two previous groups.of NSP in EEA countries, split by countries with and without short selling bans during the course of 2020.Data is relative to publicly disclosed NSP (those above 0.5% of the outstanding amount issued)of active short sellers in EEA countries, split by countries with and without short selling bans during the course of 2020.Data is relative to publicly disclosed NSP (those above 0.5% of the outstanding amount issued).Sources: NCAs, ESMA.
of NSP in EEA countries by position holders with 50 % or more of their positions between 1 January and 18 March 2020 in banned countries.of NSP in EEA countries by position holders with 50 % or more of their positions between 1 January and 18 March 2020 in non-banned countries.Sources: NCAs, ESMA.
of active short sellers in EEA countries, split by preference in allocation of NSPs between 1 January and 18 March 2020of NSP in EEA countries by position holders who did not have active positions in the period between 1 January and 18 March 2020.Sources: NCAs, ESMA.ESMA Working Paper, No. 1, 2024 Authors: Alessandro Spolaore, Caroline Le Moign Authorisation: This Working Paper has been approved for publication by the Selection Committee and reviewed by the Scientific Committee of ESMA.

Table 2
Main regression resultsEstimates of the regression, robust standard errors in parentheses.Standard errors are clustered at the stock level.*** p<0.01, ** p<0.05, * p<0.1.NA: Not Applicable -these variables were not included in the regression.

Table 3
Regression results, post-bans period variable Estimates of the regression, robust standard errors in parentheses.Standard errors are clustered at the stock level.*** p<0.01, ** p<0.05, * p<0.1.NA: Not Applicable -these variables were not included in the regression.The presence of stock and day fixed effects may result in dummy variables (here, VSTOXX) to be removed from the estimation due to multicollinearity.

Table 4
Regression results, share characteristics, liquidity variables Not Applicable -these variables were not included in the regression.The presence of stock and day fixed effects may result in dummy variables to be removed from the estimation due to multicollinearity.Sources: ESMA.

Table 5
Regression results, sectoral effects Estimates of the regression, robust standard errors in parentheses.Standard errors are clustered at the stock level.

Table 6
Regression results, shorted shares Estimates of the regression, robust standard errors in parentheses.Standard errors are clustered at the stock level.*** p<0.01, ** p<0.05, * p<0.1.NA: Not Applicable -these variables were not included in the regression.
Amihud and volatility are weighted by market cap.Trading volumes in thousands, market capitalisation in EUR bn, bid-ask spread in %, and Amihud is multiplied by a factor of 100.Sources: Refinitiv, ESMA.

Table 10
Differences in the main variables for the treatment and control groups, before and after the matching Sources: Refinitiv Datastream, FIRDS, FITRS, ESMA.

Table 12 Main
Estimates of the regression, robust standard errors in parentheses.Standard errors are clustered at the stock level.*** p<0.01, ** p<0.05, * p<0.1.NA: Not Applicable -these variables were not included in the regression.

Table 13
Regression results, post-ban variable, without controls Estimates of the regression, robust standard errors in parentheses.Standard errors are clustered at the stock level.*** p<0.01, ** p<0.05, * p<0.1.NA: Not Applicable -these variables were not included in the regression.Sources: ESMA.

Table 14
Regression results, share characteristics, trading volumes and volatility variables Estimates of the regression, robust standard errors in parentheses, *** p<0.01, ** p<0.05, * p<0.1.NA: Not Applicable -these variables were not included in the regression.The presence of stock and day fixed effects may result in dummy variables to be removed from the estimation due to multicollinearity.Abnormal returns are never significant for each stock characteristic presented in the tables, and thus not shown in the table.Sources: ESMA.
© European Securities and Markets Authority, Paris, 2024.All rights reserved.Brief excerpts may be reproduced or translated provided the source is cited adequately.Legal reference for this Report: Regulation (EU) No. 1095/2010 of the European Parliament and of the Council of 24 November 2010 establishing a European Supervisory Authority (European Securities and Markets Authority), amending Decision No 716/2009/EC and repealing Commission Decision 2009/77/EC, Article 32 'Assessment of market developments, including stress tests', '1.The Authority shall monitor and assess market developments in the area of its competence and, where necessary, inform the European Supervisory Authority (European Banking Authority), and the European Supervisory Authority (European Insurance and Occupational Pensions Authority), the European Systemic Risk Board, and the European Parliament, the Council and the Commission about the relevant micro-prudential trends, potential risks and vulnerabilities.The Authority shall include in its assessments an analysis of the markets in which financial market participants operate and an assessment of the impact of potential market developments on such financial market participants.'The information contained in this publication, including text, charts and data, exclusively serves analytical purposes.It does not provide forecasts or investment advice, nor does it prejudice, preclude or influence in any way past, existing or future regulatory or supervisory obligations by market participants.
In each case, ESMA is cited by default as a source, reflecting any data management or cleaning, processing, matching, analytical, editorial or other adjustments to raw data undertaken.