The effect of the lifted short selling ban on market volatility and liquidity

(1)

1

The effect of the lifted short selling

ban on market volatility and liquidity

Student: Auke de Haan

Supervisor: Timotej Homar

This version: February, 2014

Student number: 10091114

Specialization: Finance & Organization

(2)

2

Abstract

The goal of this paper is to investigate the effect of the lifted short selling ban in the Dutch equity market. More precisely, it wishes to investigate the effects it had on volatility and liquidity. This paper does so by observing the lifted covered and uncovered short selling ban on the EURONEXT stock exchange. On October 10th_{, 2008 the Dutch Authority Financial} Markets (AFM) issued a notification about restricted short selling. The short selling regulation expired on June 1st_{, 2009. This study analyzes two different time events to gain a more} comprehensive picture about the effect of the lifted short selling ban on the Dutch market. By using an intraday measure for volatility and a bid-ask spread and Amihud’s measure for liquidity, the effects of the lifted short selling ban will become clear. This paper finds that there is a significant relationship between volatility and the lifted short selling ban. To be more specific, the lifted short selling ban causes a decrease in market volatility. In addition, no significant relationship is found in the Amihud’s liquidity measure. Referring to the bid-ask spread, only with a larger time window there is a positive significant value. Meaning, that the lifted short selling restriction causes a more illiquid market. A reason could be that before and during the removal of the ban the Dutch equity market was less stressed. This study could not make a clear conclusion whether the lifted short selling ban caused an upgrade in market quality.

(4)

4

1. Introduction

By definition, “Short selling”, refers to the sale of a financial instrument that the seller does not actually own at the time of the sale. It is usually performed to profit from an expected decline in the price of a financial instrument (Helmes, Henker, and Henker 2010). Short selling may be used to accomplish several objectives, for example, speculators may use it to seek profits from price declines. Other motivations for short selling include hedging, arbitrage and tax considerations (Woolridge and Dickinson, 1994). According to Woolridge and Dickinson (1994), the growth in volume of short selling had been attributed to the hedging and arbitrage associated with the program trading activities of market professionals.

To prevent prices from reflecting only the views of the most optimistic investors in the market, many financial economists believe that some short selling is necessary. Another benefit of short selling is that it permits liquidity to the market (Angel and McCabe 2008). Although short selling is necessary to reduce overvaluation of shares, for long short selling has been unpopular with security owners because they believe it can depress stock prices. Owners of a security cannot do very much to prevent permissible short selling (Culp and Heaton 2008).

There is a major difference between covered and uncovered (naked) short selling. Culp and Heaton (2008) state that naked short selling has been the focus of an increasing number of lawsuits. The definition of naked short selling is that the seller agrees to sell shares but does not intend to deliver the shares on the regular settlement date, if ever (Angel and McCabe 2008). According to Angel and McCabe (2008) the usual criticisms of short selling is that short selling creates an incentive for unethical activity, which means that short selling spreads false or misleading information about its targets. Moreover, Angel and McCabe clarify that short sellers not only profit from the misery of others but they also cause it. The most important argument of Angel and McCabe (2008) is that short selling increases the volatility of markets. In this paper the influence of short selling on the volatility in stock markets will also be investigated.

The crisis on the financial markets starting late 2007 brought back some long-standing questions. The reason why most regulators restrict short selling is because short sales restrictions can reduce the relative severity of a market panic (Bris, Goetzmann, and Zhu 2007).

(5)

5 Many researchers investigated what the impact is of the short selling constraints on financial markets. With the main question: Do they make markets more or less efficient? Most regulators around the world reacted to the 2007-09 crisis by imposing bans on short selling (Beber and Pagano 2013).

The emergency order temporarily banning short selling of financial stocks will restore equilibrium to markets (Christopher Cox, SEC Chairman, 19 September 2008, SEC News Release 2008–211). The main question was whether banning short selling will lead to a better market quality (Beber & Pagano, 2013).

The argumentation of the SEC about short selling restrictions on U.S. financial stocks during the crisis (2008-2011): “Unbridled short selling is contributing to the recent sudden price declines in the securities of financial institutions unrelated to true price valuation (Khor, 2008).” During the credit crisis various countries in Europe put a restriction on short selling.

There are several studies dealing with the consequences of that restriction and whether it is efficient. Boehmer, Johen and Zhang (2008) examine the ban’s effect on market quality, shorting activity, the aggressiveness of short sellers, and stock prices. The conclusion of their research is that stock prices appear unaffected by the ban and all but the smallest quartile of firms subject to the ban suffer a severe degradation in market quality. Baber and Pagano (2013) identified their effects on liquidity, price discovery, and stock prices. According to their study, bans were detrimental for liquidity, especially for stocks with small capitalization and no listed options, slowed price discovery, especially in bear markets and failed to support prices, except possibly for U.S. financial stocks. Even the federal reserve’s bank of New York states that the 2008 ban on short sales failed to slow the decline in the price of financial stocks; in fact, prices fell tremendously during the two weeks in which the ban was in effect and stabilized once it was lifted (Battalio, Mehran, and Schultz 2012).

“Knowing what we know now, I believe on balance the commission would not do it again. The costs (of the short selling ban on financials) appear to outweigh the benefits.” (Christopher Cox, telephone interview to Reuters, 31 December 2008) (Beber and Pagano, 2013).

It is clear that most researchers investigating the ban on short selling agreed that it will lead to degradation in market quality. Because of these poor reviews on short selling restrictions some

(6)

6 countries end their short sale ban (Scannel & Karmin, 2008). The French and Belgian financial regulators have lifted their bans on the short selling of certain financial stocks and also South Korea’s financial regulator lifted a five-year-old ban on short selling of financial shares. Reinstating the ability to sell the market short will probably lead to a short term correction in the market (Reuters, 2013). The next section will describe extensively what literature tells us about short selling restrictions.

This study deals with the consequences of that lifted ban restriction and whether it is effective. I analyze the market trade volume before and after the lifted ban. A market with low volume is very inefficient and does not attract investors (Reuters, 2013). Therefore liquidity could be a sufficient proxy for market quality. Another parameter that will be investigated is volatility. According to the paper of Daly (2011) volatility is an important issue on itself because when asset prices fluctuate sharply over time, this may lead to an erosion of confidence in capital markets and a reduced flow of capital into equity markets. Economic and financial theory suggests that consumers are risk averse. Increased risk associated with a given economic activity causes a reduced level of participation in that activity (Daly 2011).

In this paper both volatility and liquidity are used as a proxy for market quality. Short selling bans in most countries were introduced in September 2008 (Beber and Pagano 2013). Then the ban was lifted at different dates and this gives rise to the following question:

 Did the end of the restriction on short selling have a positive impact on the quality of the Dutch equity market?

Because of the fact that many studies conclude that a ban on short selling will reduce market quality, the hypothesis of this study is:

 The end of short selling restrictions increases market liquidity, reduces volatility and will lead to an upgrade of the Dutch equity market quality.

(7)

7

2. Literature

This section summarizes the literature that is related to short selling restrictions and the effect on volatility and liquidity in the equity market.

Short selling (restrictions) are widely documented and the ongoing debate about short selling started in the dawn of modern stock trading in Amsterdam in the early 1600s. Traders in the shares of the Dutch East India Company (Vereenigde Oost-indische Compagnie) discovered that they could make money when stock prices were falling. In 1610 the Dutch East India Company banned short selling before the Amsterdam Stock Exchange opened its first building in 1611 (Angel and McCabe 2008).

Although this paper investigates the lifted short selling ban on the market quality, it is important to review the literature of the introduction of the short selling ban. To evaluate the short selling introduction I might get a better understanding and comprehensive picture of the cause and effect of a short selling restriction. There are numerous studies that analyze the short selling ban during the credit crisis. It is assumed that the short selling ban decreases market quality.

One important, useful, research study about the short selling bans is from Beber and Pagano (2013). They exploit the regulatory interventions around the world in 2008 and 2009 and they analyze the effects of the ban on market liquidity, on the speed of price discovery, and on stock prices. Beber & Pagano (2013) observe the bid ask prices, volumes, and number of outstanding shares and their data set contains 5,992,679 stock-day observations. They apply the different-in-different estimation. Their results are divided into three main points. First of all they find that bans were detrimental for liquidity, especially for stocks with small capitalization and no listed options. Secondly, a short selling constrain causes slowed price discovery, especially in bear markets. Thirdly, short selling bans failed to support prices, except possibly for U.S. financial stocks. Figure (1) shows the period in which bans were enacted in the sample countries of Beber and Pagano (2013).

The results of Beber and Pagano (2013) are also confirmed by Helmes, Henker & Henker (2010), who examine the effects of the short selling ban, imposed by Australian regulators. The ban on short selling of the Australian financial stocks lasted eight months. Helmes, Henker &

(8)

8 Henker (2010) conclude that, as predicted by previous theoretical work, stocks subject to the short selling ban suffered a severe degradation in market quality. To be more specific about market quality, they state that constraints on short selling reduced trading activity, increased bid and ask spreads and increased intraday volatility.

Figure (1) Structure of the data set in the paper of Beber and Pagano (2013)

The study conducted by Boehmer, Jones and Zhang (2009) draws the same conclusionas the paper of Beber and Pagano (2013). They conclude that the ban may not have provided much of an artificial price boost. In comparison with Beber and Pagano (2008) their study is on a smaller scale (1.000 financial stocks). The prices of stocks subject to the ban increase sharply, something that should be prevented by the ban.

Furthermore, a study done by Battalio, Mehran and Schultz (2012) again shows that the short selling ban failed to slow the decline in the price of financial stocks in U.S. This study has the same conclusions as the other papers previously discussed.

(9)

9 According to Diamond and Verrecchnia (1987), stock markets are not efficient anymore because short sale constraints lead to a decrease in trading and prices take longer to adjust to private information. Because short selling constrains cause a removal of some relatively informative sell or short orders. As a result, price adjustment is particularly slow during a short selling restriction because of the presence of negative private information (Diamond and Verrecchia 1987).

Miller’s (1977) paper explores some of the implications of a market with restricted short selling. For example Miller (1977) states that restricted short selling reduces the risk premiums on the riskier securities, and may even make them negative. About liquidity, Miller explains that because of short sales the supply of stock on the market increases by the amount of the outstanding short position. Thereby, short sale constraints prevent pessimistic traders from short selling without restricting optimistic traders from buying them. This results in an upward bias to stock prices.

More recent papers about short selling around the world are from Bris, Goetzmann and Zhu (2007). They analyze the cross-sectional and time-series information from 46 equity markets around the world. They observe the market efficiency and the returns to individual stocks and market indices. Like every other research study about short selling restrictions, they find evidence that prices incorporate negative information faster in countries where short sales are allowed. The market returns display significantly less negative results which is the opposite most regulators think.

Most of the research contributing to the investigation of short selling restriction concludes that a ban will lead to market degradation. This means that the volatility will rise and liquidity will decrease. Moreover, price adjustment is particularly slow during a short selling restriction. Apparently, there has not been much research on the lifted short selling constrain. The reason for that could be that it will not contribute to science because most researchers agree that short selling will improve market quality. A study about the lifted short selling restriction could be superfluous. But, in science it is critical not to draw fast conclusions and to observe and analyze new data carefully. As this paper includes a new time window this study will definitely contribute to science.

(10)

10

3. Data

This section describes what data is used for stock information in order to do the required estimations of volatility and liquidity, and how this data is obtained.

The research conducted by Beber and Pagano (2013) performs a differences-in-differences estimation. They analyze only countries that imposed bans on financial stocks with a time window of 50 days before and 50 days after the ban inception date. As regulation of short sales during the crisis differs across countries along many dimensions, it is extremely hard to find comparable countries. Beber and Pagano (2013) point out those different dimensions in short sale restrictions:

i. Different ban inception dates; ii. Different lifting dates;

iii. The presence of countries that imposed no bans;

iv. Differences in the scope of bans, which applied only to financials in some countries and to all stocks in others; and

v. Differences in the stringency of bans, which were naked in some cases and covered in others.

This study only investigates the lifted short selling ban. Beber and Pagano (2013) and Bris, et all. (2007) analyze cross-sectional and time-series information from many equity markets around the world. This paper will only investigate the lifted short selling ban in two European countries. As Denmark and the Netherlands restricted the ban on naked and covered financial stocks, it is sufficient to compare these countries. Moreover, Denmark and the Netherlands have, roughly speaking, the same financial sector and both introduced the short selling restriction in the same period. This provides an opportunity to investigate the effects of the lifted short selling ban in the Netherlands and compare the results with Denmark.

According to the executive order issued by the Danish FSA (Financial Supervisory Authority), obtaining a short position in financial institutions is prohibited. As stated in section 1(3) of the order, borrowing of shares does not constitute ownership. This executive order therefore banned short selling in Danish financial shares and came into force on 13 October 2008

(11)

11 (Larsen, 2008). Again, the Danish ban on short selling in financial institutions was a total ban on short selling in such shares and not limited to uncovered short selling in such shares. The restriction was as indicated regarding shares in Danish listed banks on Nasdaq Omx Copenhagen (Larsen, 2008).

Up until 1st November 2012 the executive order on short selling was effective in Denmark. The EU regulation replaced the Danish short selling ban with an introduction on a naked short selling ban. So, the restriction in the EU on short selling and certain aspects of credit default swaps came into force on 1st November 2012 (ESMA, 2012).

On October 10th, 2008 the Dutch Authority Financial Markets (AFM) issued a notification about restricted uncovered short selling. The inception date of the uncovered short selling ban is September 21st, 2008 and covered several Dutch financial institutions listed on the EURONEXT. This notification was issued together with a request to report certain short positions. Because of the continuing turmoil in the financial markets the notification expanded to a general ban on short selling (uncovered and covered) on October 5th, 2008 (Government Gazette, 2008). The short selling regulation expires on June 1st, 2009.

Danish stocks will be used as a control group for the differences-in-differences estimation that will be used in order to test the effect of the lifted short selling ban. To perform a sharper differences-in-differences estimator, this paper will restrict the estimation to countries that imposed bans only on financial stocks and a ban on both uncovered and covered short selling. This estimation will be discussed in the methodology section.

For all the stocks June 1st_{, 2009 is the date of the event. I apply two different time windows:} data from 50 days and six months prior to this event (ex ante) will be compared to data from 50 days and six months after this event (ex post). First of all, I perform a test on the effect of lifted short selling bans over a window of 50 days around the inception date; which is 1st_June 2009. To be more specific, a 50 days window in this experiment is 36 trading days before and after 1st June 2009 (72 trading days in total). Secondly, I observe a window from January 1st, 2009 to December 12th, 2009. That is 192 trading days before the lifted short selling ban and 154 trading days after the lifted inception date. The choice of the time window has the advantage of concentrating on a time interval in which the effects of the ban should be less easily clouded by confounding factors (Beber and Pagano 2013).

(12)

12 The reason for analyzing two different time windows is mainly to obtain a better and comprehensive unbiased conclusion. The hypothesis is that, if this study analyzes a longer time period the statistic results should be more significant. I compare the two time periods to understand closely what the impact is on market volatility and liquidity on a short and longer period of time.

Figure (2), shows the period in which bans were enacted in Denmark and the Netherlands via color-coded lines. Dark blue lines correspond to naked bans of financial stocks and red lines indicate both covered and uncovered bans for financial stocks.

Figure (2) Blue line: Naked short selling ban. Red line: Covered and Naked short selling ban.

For all financial stocks data was obtained moderately simply using DataStream for the right time interval. These stocks are:

- Treatment group: Financial institutions of the Netherlands.

Aegon N.V., ING Group N.V., Fortis N.V., BinckBank N.V., Kas Bank N.V., SNS Reaal N.V., Van der Moolen Holding N.V. and Van Lanschot N.V.

- Control group: All financial institutions of Denmark (Nasdaq Omx Copenhagen). Danske Bank, Jyske Bank, Nordea Bank, Sydbank, Ringkjøbing Landbobank, Spar Nord Bank, Bank Nordik, Danske Andelskassers Bank, Djurslands Bank, Fynske Bank, Grønlandsbanken, Hvidbjerg Bank, Jutlander Bank, Kreditbanken, Lån og Spar Bank, Lollands Bank, Møns Bank, Nordjyske Bank, Nordfyns Bank, Nørresundby Bank, Østjydsk Bank, Salling Bank, Skjern Bank, Totalbanken and Vestjysj Bank.

The dataset which forms the basis for our empirical tests consist of a sample of 33 financial institutions (8 Dutch and 25 Danish).

(13)

13 Referring to the 50 days’ time window, daily high and low prices were obtained from DataStream for all trading days between April 10th, 2009 and July 20th, 2009. All weekends and public holidays were excluded from the sample, so the maximum number of trading days’ data for any single company was 65. For the six months window I obtained all high and low prices for all trading days between January 1st_{, 2009 and December 31}st_{, 2009. I deleted all} missing data, weekends and public holidays from the sample to gain the maximum number of trading days’ of 249 for any single company.

3.1. Descriptive statistics

This section summarizes the given data set, which is represented by multiple tables and figures in section 8. The output will include a boxplot of each dependent variable. This paper will discuss some descriptive statistics like the mean, median, mode, standard deviation and the minimum and maximum variables of each dependent variable. I analyze the effects of the lifted short selling ban on two variables: market liquidity and volatility. For each dependent variable I calculate a time-series average over the largest time interval researched in this paper, namely 1 year. In order to give a representation of the time-series, this paper uses histogram graphs. I will observe the time plots of both countries separately. There are four histogram graphs in this paper with a one year time period. I use the mean volatility and liquidity (both y-axis) for the histogram graphs. Descriptive statistics provide a useful summary to acquire a better understanding of the data.

First, I will describe the data set of the volatility dependent variable. This paper uses the intra-daily range to calculate volatility. The first time window of 36 trading days, provides 1342 observations. This includes the 25 Danish and 8 Dutch financial institutions. The sample mean of the volatility measures is 0.0236, with the given sample standard deviation of 1.70 percent. The boxplot, figure (1), gives a clear summary of the data set and the particular outliers. The second time window, which is 6 months around the inception date, has 5194 observations and has a mean of 0.0243 and a standard deviation of 1.97 percent. Both the mean and the standard deviation of the larger time window are higher than these values of the smaller time window. When analyzing the boxplot of the six month time event (figure (2)), I can conclude that there are more and higher outliers in comparison to the smaller time window. Referring to graph (1) and graph (2), the histogram graphs show the volatility in the beginning of 2009 of both

(14)

14 countries. The values are relative high. During the year, the volatility of both countries decreases, which follows logically because of the decreasing economic stress. I can conclude that the volatility of the Netherlands overall is lower than the volatility in Denmark and decreases towards the end of the year more than the observed Danish equity market. This may be in line with the hypothesis of this paper, that the lifted short selling ban in the Netherlands causes a decrease of volatility in the Dutch equity market.

The smaller time window of the data set of the Amihud’s liquidity measure has 1286 observations and the larger time window has 4959 observation. The mean of the data set is 4.142 and the standard deviation is 11.39 percent. Again, the mean and standard deviation refer to both Danish and Dutch financial institutions. The boxplot, figure (3), displays some extreme outliers. Interesting is the mean and standard deviation of the six month time event, which are 4.34 percent and 11.90 percent, respectively. This is not much different from the smaller time event. Moreover, the boxplot of the six month window is more or less the same as the 36 trading day’s window. When analyzing the time plots, graphs (3) and (4) of the Amihud’s liquidity measure of both countries, I can conclude that the measure in the Netherlands is systematically lower during the year 2009. The reason is that a lower Amihud’s measure denotes a liquid market. Therefore, the Dutch equity market is more liquid than the Danish equity market. It seems that the liquidity of the Dutch and Danish equity markets increases towards the end of the year.

Table (12) and (13) provides some descriptive statistics for the transaction cost measure. The smaller time event includes 1967 observations and the larger time window has 7403 observations. The standard deviations of both time windows are 3.48 percent and 3.73 percent. These values are relatively low compared to the standard deviations of Amihud’s liquidity measure for the same time windows.

3.2. Volatility

In this paper volatility is used as a proxy for market quality. According to Daly (2011) volatility is a fundamental important concept to the discipline of finance. Because, the fluctuation of asset prices may lead to an erosion of confidence in capital markets and a reduced flow of capital in equity markets. Also, volatility could be an important factor in determining the

(15)

15 probability of bankruptcy for individual firms. Moreover, the higher the volatility for a given capital structure, the higher the probability of default (Daly 2011). This section provides a survey of different measures of volatility. Further I will discuss the advantages and disadvantages of each measure. Moreover, I will describe exactly what determines volatility changes. There are multiple ways to calculate historical volatility apart from the standard deviation of changes in stock.

Volatility is used extensively for portfolio selection and a common measure of stock market volatility is the standard deviation of returns (Daly 2011). According to Daly (2011), a useful measure for characterizing the evolution of volatility is the sample standard deviation from daily returns. The standard deviation 𝜎 can be represented as follows:

𝜎 = √∑(𝑅𝑡− 𝑅̅

𝑇

1

)2_{/(𝑇 − 1) (1)}

In this formula the standard deviation and the returns are defined respectively as 𝜎 and 𝑅_𝑡 with a sample of T observations. The average return of the sample is, 𝑅̿ = ∑ 𝑅_𝑡/𝑇. The sample standard deviation formula is shown to develop a better understanding concerning different volatility measures but this paper chooses not to use this formula.

A measurement for forecasting volatility of a financial asset is the Black-Scholes option pricing formula. This forecasting method involves the market price of stock options. Referring to the paper of Daly (2011), the Black-Scholes formula applies a theoretical pricing function to relate the price of a contingent claim to the volatility of the underlying asset and solve it for the implied volatility given by the market price of the asset. One problem with this approach is that the true volatility must be constant. According to Daily (2011) this is more likely if the term is short enough so that stock prices can be considered to a normal distribution with constant variance. The interpretation of volatility calculated by the implied standard deviation is unclear if this is not the case. Moreover, the Black-Scholes formula can only be used for European Style Options, options that may only be exercised at the expiration date of the option. In this study historic volatility will be analyzed therefore this measure is redundant. The reason is that

(16)

16 the Black-Scholes formula is a forecasting model, but it may be of importance in further studies.

The squared daily returns, the realized volatility measure and the intra-daily range are the three leading measures of intraday volatility. The squared daily returns is a quite simple measure and uses squared daily returns but this is not an ample measure as it does not include price shifts during the day. This measure is not a comprehensive measure of volatility because usually the closing price is the extreme price of that day. The realized volatility measure sampled data at regular intra-daily intervals. Under some assumptions this volatility measure is a consistent estimator of the quadratic variation of the underlying diffusion process (Engle and Gallo 2003).

This paper uses the intra-daily range. This formula includes a spread between the highest recorded daily price and the lowest recorded daily price. The intra-daily range does take price fluctuations during the day into account. According to Engle and Gallo (2003) this spread has long been recognized as a function of the volatility during the day. Again, the intra-daily range suggested Parkinson (1980) is adopted in this paper for calculating the volatility of the Dutch and Danish equity market. The following formula states the intra-daily range:

𝜎 = 1

2√𝑙𝑛2∗ 𝑙𝑛

𝐻𝑖𝑔ℎ

𝐿𝑜𝑤 (2)

The dataset comprises daily high and low market price (data type PH and PL) series for two stock exchanges: EURONEXT and Nasdaq Omx Copenhagen. The dataset consists of 5,195 observations for each series.

3.3. Liquidity

For the liquidity multiple measures will be used to give a complete analysis of the effects on liquidity. The reason for this is that obtaining data to calculate liquidity is less straightforward than the volatility measure. There are many studies which have investigated how to measure liquidity and there is a continuous discussion about the definition. This paper summarizes an overview of these studies and the indicators that can be used to illustrate and analyze liquidity measurements.

(17)

17 Liquidity is a complex concept and is still an insufficiently researched topic. Liquidity is not observed directly but rather has a number of aspects that cannot be captured in a single measure (Amihud 2002). One definition of liquidity is, for example, the ability to convert stocks into cash without affecting the price (Bogdan, Bareša, and Ivanović 2012).

In general it is perceived that liquid markets are preferred over illiquid markets. To present a cohesive conclusion about the liquidity market quality before and after the inception date this paper includes two different liquidity measurements. The reason for this is that no single correct and universally accepted measure determines a market degree of liquidity (Sarr and Lybek 2002). And, according to a paper written by Sarr and Lybek (2002), some variables of liquid markets may change over time.

The paper of Sarr and Lybek (2002) conducted an important study about liquidity and explains that liquid markets offer three benefits. First of all liquid markets allow central banks to use indirect monetary instruments and contribute to a more stable monetary transmission mechanism. Secondly, liquid markets allow financial institutions to accept larger asset-liability mismatches. Thereby reducing the risk of the central bank having to act as lender of last resort for solvent but illiquid credit institutions. As a third reason they clarifies that a liquid market attracts more investors, who can transact in them more easily.

Sarr and Lybek (2002) explains the five characteristics of liquid markets in detail: tightness, immediacy, depth, breadth and resiliency. Tightness refers to the low transaction costs and closely to the bid-ask spread. The speed on which orders can be executed and the efficiency of the trading represent immediacy. Depth refers to the amount of orders or the amount of potential buyers and sellers. Breadth, which refers to how large order are. The variable resiliency defines the speed at which new orders compensate imbalances.

Furthermore, Bogdan et al. (2012) investigates the reason of stock illiquidity and their main findings are the height of both the transaction costs and the spread between bid-ask price of the stock. In this paper, I use this bid-ask spread and the transaction cost measure. Hence, two measures will be used in reference to the calculation of liquidity which will be carefully described below.

(18)

18 3.3.1. Amihud’s liquidity measure

The first measure in reference to liquidity is known as Amihud’s liquidity measure. For their calculation most of the liquidity measurements require unavailable microstructure data on transactions and quotes. The Amihud’s liquidity measure used in this study is calculated from daily data on returns and volume (Amihud 2002). Although the daily data (return and volume) is coarser and less accurate than those microstructure data, it is readily available for the study of the time series effects of liquidity. Amihud offers an estimation of illiquidity by investigating what the effect is on the return of a certain trading volume:

𝛼2 =|𝑟|

𝑉 (3) The daily return in this formula stands for r and V defines volume of trades. The market is liquid when high volumes of trade have a lower impact on price returns which results in a lower value of liquidity measure. In a liquid market the Amihud’s measure of liquidity decreases. So, a lower value of the Amihud’s liquidity measure defines a liquid market. When Amihud’s measure of liquidity increases, the market becomes less liquid.

Again, I collected the data from DataStream database and stock returns are expressed in percent. According to Amihud (2002), the illiquidity measure is the average across stocks of the daily ratio of absolute stock return to dollar volume. Trading volume data were obtained from DataStream. I extracted turnover by volume from DataStream data type VA, which shows the value of shares traded per day.

According to Amihud (2002), there are more measures of liquidity that use data on volume. One of the related measures is turnover (formula 4), the ratio of trading volume to the number of shares outstanding. Sarr and Lybek (2002) adds that the trading volume can be given more meaning by relating it to the outstanding volume of the asset being considered. One negative aspect of the turnover rate is that the trading volume may shift significantly during day, week or month. To get a more cohesive picture about liquidity using the turnover rate, volatility must be taken into consideration.

(19)

19

𝑇𝑛 = 𝑉

(𝑆 ∗ 𝑃) (4) Referring to equation (4); S stands for the outstanding stock of an asset and P for the average price. V, defines volume of sales and can be obtained directly from DataStream.

One measure that is as effective as Amihud’s liquidity measure, is Martin’s liquidity index (1975). This measure is used for calculating the liquidity of an index. Another formula that originally applied to the equity market is the Hui-Heubel Liquidity Ratio (formula 5). This measure relates the volume of trades to its impact on prices. The less liquid indices are, the higher the value of the Hui-Heubel liquidity ratio. A high ratio means that prices changes according to high volume, which is a sign of illiquidity (Sarr and Lybek 2002).

𝐿ℎℎ =

[(𝑃_𝑚𝑎𝑥 − 𝑃_𝑚𝑖𝑛)/𝑃_𝑚𝑖𝑛]

[𝑉/(𝑆 ∗ 𝑃̅)] (5) The equation of 𝐿_ℎℎ can be calculated as an average of the 5-day period or another defined interval in a sample to smooth volatility. Pmax and Pmin stands for, respectively, maximum and minimum price in that same time interval. The other variables are already described in the last two sections. Although this study investigates the Dutch and Danish indices, I only observe the financial institutions of equity markets. There is no need to use a Hui-Heubel liquidity ratio to calculate the liquidity of equity markets because this paper compares individual financial stocks with each other.

3.3.2. Transaction cost measure

One important measure that is described by the Sarr and Lybek (2002) paper is the bid-ask spread. According to Sarr and Lybek (2002), this measure captures nearly all of the transaction costs. The bid-ask spread reflects four different transaction costs:

i. Order-processing costs;

ii. Asymmetric information costs; iii. Inventory-carrying costs; and, iv. Oligopolistic market structure costs.

(20)

20 Important is that high transaction costs cause a reduction in the demand for trades. As a reaction the number of potential active participants in a market declines (Sarr and Lybek 2002). This is in contrast when transaction costs are small. In the latter, market-makers would prefer to use dealers in auction mechanisms to trade rather than incur direct search costs. So, the higher the transaction costs, and therefore a higher bid-ask spread, the lower the amount of trades will be. Illiquid markets are associated with a high spread and vice versa. The high trading costs will reduce the number of market participants and will also affects the breadth and resiliency of an equity market. According to Sarr and Lybek (2002) since large transaction costs (and therefore a large bid-ask spread) may deter trades, they reduce resiliency by preventing orders from flowing in promptly to correct order imbalances that tend to move price away from their fundamental level. In the end, liquid markets will result in small bid-ask spread and will often lead to fragmentation among market and a higher trade volume. A small bid-ask spread results in a more efficient market. The formula for calculating the transaction cost measure is as follows:

𝛼1 = (𝑃𝑎 − 𝑃𝑏)

(𝑃_𝑎+ 𝑃_𝑏)/2 (6) Daily closing bid and ask prices were obtained from DataStream for all trading days between April 10th_{, 2009 till July 20}th_{, 2009 and January 1}st_{, 2009 till December 31}st_{, 2009. These prices} are the best bid and ask prices quoted by market makers at the close of the market each day. Again, all weekends and public holidays were excluded from the sample. The Danske Andelskassers Bank is excluded from the dataset because DataStream was not able to find the data of this bank. It is worth discussing whether the daily closing prices obtained from Datastream are valid (Lee and Rui 2002). According to Lee et al (2002) the closing prices may not indicate the average market-maker behavior during the day. The closing prices could not be representative of intraday prices.

To summarize, this study uses multiple measures of liquidity to give a more comprehensive estimation of what the consequence is of decreased latency on liquidity. Several methods and formulas for calculating volatility and liquidity have been reviewed. Only the Amihud’s and transaction costs measure will be used in this paper to calculate liquidity. For volatility, this paper uses an intraday volatility measure. I observe two different time events and therefore I perform six regressions all together. This will be discussed carefully in section 5.

(21)

21

4. Methodology

This section provides explanation of the model that is used and of the several calculations that are made. In order to get a cohesive and detailed insight in the effects of the lifted ban in short selling, multiple measures of liquidity and one measure of volatility are used. I already discussed the advantages and disadvantages of each measure and argued carefully which measure I chose for liquidity and volatility in section 3.The method and the outcomes of the measures will be used to investigate what the effects are of the lifted ban in the Netherlands and if the effect is significant.

4.1. Model and differences-in-differences method

Studies conducted by Stock and Watson (2012) report that data from a randomized controlled experiment can be analyzed by comparing differences in means or by a regression that includes the treatment indicator. The so-called differences-in-differences estimation, which estimates causal relationships, has become increasingly popular. According to Bertrand, Duflo and Mullainathan (2003) the differences-in-differences estimation has some limitations. There might be possible signs of endogeneity of the interventions in a differences-in-differences model. It is appropriate when the interventions are as good as random, conditional on time and group fixed effects (Bertrand, Duflo, and Mullainathan 2003). Stock and Watson (2012) describe that if the treatment group is randomly assigned, the differences-in-differences estimator is unbiased and consistent.

According to Stock and Watson (2012) the differences estimator is the difference in the sample averages for the treatment and control groups, which can be computed by regressing the outcome of dependent variable Y on a binary treatment indicator X:

𝑌𝑖 = 𝛽1+ 𝛽1𝑋𝑖+ 𝑢𝑖, 𝑖 = 1, … , 𝑛. (7) To add some control variables in the regression, the difference estimator can be improved. Doing so leads to the differences estimator with additional regressors (Stock and Watson, 2012). But, even controlling for observed variables, some differences might remain between the treatment and control group (Stock and Watson, 2012). This paper analyzes the change in

(22)

22 the outcomes pre- and post-treatment and not a comparison between the outcomes Y of the two groups. This estimator is the difference across groups in the change, and difference over time.

The regression that is done in this paper is described in formula:

𝑌𝑖 = 𝛽0+ 𝛽1𝑇𝐼𝑀𝐸 + 𝛽2𝑇𝑅𝐸𝐴𝑇𝑀𝐸𝑁𝑇 + 𝛽3(𝑇𝐼𝑀𝐸 ∗ 𝑇𝑅𝐸𝐴𝑇𝑀𝐸𝑁𝑇) + 𝑢𝑖 (8) The particular estimation that is done in this paper is the dependent variable of the model (𝑌_𝑖). This study analyzes the impact on liquidity and volatility. This paper uses one measure of volatility and two measures of liquidity. The dependent variable is intraday volatility in reference to volatility, which was first suggested by Parkinson (1980). The two separate measures of liquidity are: the Amihud’s measure of liquidity and the transaction cost measure (so called bid-ask spread). So there are three different dependent variables. Further detail about the measurement of volatility and liquidity in section 3.1 and 3.2.

For each of the dependent variables, one control group will be used to give a clear picture of the effect of a lifted short selling restriction on volatility and liquidity. Both variables TIME and TREATMENT are binary variables. When “i” is before the event date (June 1, 2009) the binary variable Time equals 0 and “i” is 1 when “i” is after the event date. The binary variables equals 0 for the control group and 1 for the treatment group. TIME*TREATMENT is an interaction term between the two binary variables.

In this differences-in-differences model the interaction term is the one of interest, as it measures the effect of a lifted short selling ban on the treatment group. Using Ordinary Least Squares (OLS) I can estimate the differences-in-differences estimator in treatment and control groups before and after the short selling restriction. Because in this paper the interaction term (𝛽₃) is investigated, a simple OLS regression will be made to calculate the 𝛽3 from the differences-in-differences estimator. The 𝛽₃ measures the effect of a lifted short selling ban on the treatment group. Therefore the hypothesis I propose is as follows:

𝐻0: 𝛽₃ = 0, 𝑎𝑔𝑎𝑖𝑛𝑠𝑡 𝑡ℎ𝑒 𝑎𝑙𝑡𝑒𝑟𝑛𝑎𝑡𝑖𝑣𝑒 𝑡ℎ𝑎𝑡

(23)

23 Stock & Watson (2012) explain that the differences-in-differences estimator is the average change in Y for those in the treatment group, minus the average change in Y for those in the control group:

Where the average change in Y in the treatment group is ∆𝑌𝑡𝑟𝑒𝑎𝑡𝑚𝑒𝑛𝑡 and the average change in Y in the control group is ∆𝑌𝑐𝑜𝑛𝑡𝑟𝑜𝑙.

Besides the relation between the lifted short selling ban and the volatility and liquidity of the equity market is the significance of that particular coefficient, of importance. By using a t-statistic and the corresponding p-value I will determine the significance of a particular coefficient. From these p-values I can determine whether the coefficients are significantly different from zero. In other words, the significance tells me at what confidence level I can say that there is either a positive or negative relationship between the lifted short selling ban and the measures of volatility and liquidity.

5. Empirical results

In this paper volatility and liquidity in the Dutch and Danish equity market was investigated. This section discusses the results of the calculations that have been done with regard to the measure of volatility and the two measures of liquidity. I compare the countries Denmark and the Netherlands with each other and hopefully this paper obtains a clear significant conclusion about the market quality before and after the lifted ban. First the results from the volatility measure will be discussed and then the measure for liquidity. For each subsection, the differences-in-differences model will be used as described in formula (6), and the hypothesis as described in the methodology section will be tested. In the end, as stated in the introduction, this study tests the following hypothesis:

 The end of short selling restrictions increases market liquidity, reduces volatility and will lead to an upgrade of the Dutch equity market quality.

(24)

24 This hypothesis reflects the literature about short selling restrictions, which is shortly: short selling is an important tool for price discovery, volatility and liquidity in the financial marketplace. This paper only investigates the volatility and liquidity, in further study price discovery in equity markets could also be observed to gather a more cohesive picture of the lifted ban on the quality on the market. Because price discovery could be another proxy for market quality. I observe two time windows (50 days and six months), both are around the inception date which is June 1, 2009. The use of three dependent values and two different time windows, results in six regressions in total.

5.1. Volatility

This paper uses one measure of volatility, an intraday volatility measure. This measure uses the daily high and low prices as seen in formula (2). The formula describes the relationship between these prices and the volatility. The first part of the intra-daily formula could be seen as a scaling factor to the use of the daily high and low prices. There are several volatility measures but this one, in particular, can be calculated for each day in the sample period. The advantage is that I can use all the daily observations to do a regression analysis. As described in the data section all weekends and holidays are deleted from the data set. I evaluate two different time windows. The reason for that is to gather a more consistent representation about the quality of the market before and after the inception date. As stated earlier, a longer time window might present a better significant value.

For using one measure to calculate volatility for 25 Danish financial institutions listed on Nasdaq Omx Copenhagen and 8 Dutch banks listed on EURONEXT with two different time windows, results in two regressions in total. The results of the estimation of all the coefficient 𝛽₃, which describes the effect of a lifted short selling ban in the treatment group (the Netherlands), of all regressions can be found in table (1) in the output summary. The differences-in-differences method is used to calculate the coefficient of the interaction term. The interaction term in this study is declared as TIME*TREATMENT. I tested, with a simple t-test, this interaction term with the hypothesis that 𝛽3= 0, as described in the methodology section. I can draw conclusions about the significance of the coefficient 𝛽3, when I observe the t-statistics and the corresponding p-values. Table (1) presents a summary of the results of all the data I observed.

(25)

25 First let’s analyze column (1) of table (1) which is a summary of the regression using the intra-daily range formula with a 50 days window (36 trading days). The interpretation of the coefficients of the time dummy (𝛽1) tells how much volatility has increased or decreased in the control and treatment group from time before and after the inception date (June 1st, 2009). The value of the coefficient 𝛽₁ = −0.00805 and has a p-value of smaller then 1 percent. I am able to reject the null hypothesis with a 1 percent significant level. This means that after the lifted short selling ban in the Netherlands the volatility in the Dutch and Danish observed financial stocks decreased. I cannot conclude much about this statement because I am interested in the interaction term.

The second dummy is TREATMENT which defines the control group. According to the value of the coefficient, 𝛽₂ = 0.00785 (significant at the p<0.01 level), the 8 Dutch financial institutions are more volatile than the Danish 25 financial institutions before and after the lifted short selling restriction.

Finally, the third dummy and most important coefficient in this study, has a negative significant value of 𝛽₃ = −0.00331. I can reject the null hypothesis (with 10% critical alpha) which tells us that the volatility has decreased in the treatment group more than it has decreased in the control group after the lifted short selling ban (June 1st_{, 2009). Because, as explained above,} according to the first dummy the trend of volatility of both treatment and control groups was declining.Interesting is the fourth regression in table (1). Which represents the volatility results as well but with a longer time period. The 𝛽₃ value of the fourth regression is lower and more significant than the interaction term of the first regression. From this dataset I can conclude that the significant value will increase when I observe a longer time window. In the six month time period around the inception date the volatility will decrease more than a 36 trading day’s window. As the time window increases, there is enough evidence to conclude that the volatility decreases in the Dutch equity market after the short selling ban. This statement reflects the literature described in section 2.

Furthermore, I can conclude that the coefficient of the fourth regression (volatility with a six month time event) does differ from zero with a 95 percent confidence interval. The reason becomes clear when I investigate the Stata output of the regressions further in table 2 and 3. I am confident to say that the coefficients of the longer time interval differ from zero, as zero is

(26)

26 not included in the confidence interval. The fourth regression only includes negative value and a lower value of the intraday volatility formula means that there are low fluctuations in prices, resulting in the high and low prices that are close to each other. Therefore a negative coefficient means that volatility decreases when short selling constrain ends. Table (2) summarize the output of intraday volatility with a 50 days’ time window. It becomes clear from table (2) that for this particular regression the confidence interval includes positive and negative values. Thus, I can conclude with 95 percent confidence that the coefficient does not differ from zero, because the confidence interval includes both positive and negative values.

I use the R-squared measure for observing the fit of the regression. The dependent variable, volatility, is explained by the independent dummy variables. In other words, how well do the two binary variables and the interaction term explain the intraday volatility? The R-squared for the first regression (table 2) is 9.76 percent. The R-squared for the fourth regression (table 2) is much higher, 14.33 percent.

The adjusted R-squared measure is a modified version of the R-squared because it does not necessarily increase as the number of independent variables increase. The adjusted R-square for the 50 day time window is 9.56 percent and for the six month time window, 14.28. I can conclude that the R-squared measure as well as the adjusted R-squared measure, are quite low. These (binary) variables are not sufficient in explaining the variation in the dependent variable, which is intraday volatility.

In short, it has become apparent that I have sufficient evidence to prove a relation between decreasing volatility and the lifted short selling ban in the Dutch equity market. From the calculated t-statistic and p-values I can conclude that the null hypothesis 𝛽3 = 0, meaning that there is no relationship between volatility and the lifted short selling ban on the treatment group, can be rejected. There is a negative significant relationship between the two.

5.1. Liquidity

I examine the effects of lifted short selling bans on liquidity with two measures to give a more comprehensive estimation of the lifted ban on liquidity. I use the Amihud’s liquidity measure, formula (3), and the bid-ask spread, formula (6), as described in the methodology. The results

(27)

27 of the calculations will be discusses in the next section. Moreover, this study provide evidence based on regression analysis.

5.1.1. Amihud’s liquidity measure

Formula (3) offers the Amihud’s liquidity measure and defined as the ratio of the absolute value of daily return to trading volume. As explained in the methodology section, a market is said to be illiquid when the ratio is high. To be more specific: trading volume has a large effect on the daily price fluctuations. Again, I observe two different time windows for all dependent variables. The regressions are done referring to the Amihud’s measure which I will described extensively in this section. In the literature section of this paper I concluded that a short selling ban will lead to an illiquid market. But, this paper investigates a lifted ban after the crisis (June 1st, 2009).

I start first with describing the 50 days’ time window around the inception date of June 1st, 2009. The coefficient of the interaction tem is the one I am interested in. As described earlier, I am testing the following hypotheses 𝛽3 = 0. Observing table (4), which summarizes the output from the regressions, I can see that the estimated coefficient is 𝛽₃ = 0.03996. This value is absolutely not significant and I cannot reject the null hypothesis. There is no evidence that shows any correlation between the lifted ban in the Netherlands and the liquidity in the equity market. In fact, all the coefficients in table (4) are not significant. Moreover, by looking at the 95 confidence interval in table (4), the interval contains positive and negative values. This means that the coefficient does not differs significantly from zero.

In section 5.1, I noticed that a longer time window cause higher significant values on the coefficients. In this case the dependent variable is the liquidity measure but in the 50 days’ time window there were no significant values. The outcome of the Amihud’s six months window measure can be found in table (5) of the output summary. The coefficient of the interaction term has a value of 𝛽₃ = −0.0640591 so there seems to be a negative impact on Amihud’s measure. This indicates an increase in liquidity when the short selling ban (covered and uncovered ban) ended in the Netherlands. A negative value of the interaction term means that high volumes of trade have a lower impact on price returns. Unfortunately, there is no sign of any coefficient which has a significant value. Moreover, the R-squared and adjusted R-squared

(28)

28 seen in table (4) and (5) are both quite low. These binary variables are not sufficient in explaining the variation in Amihud’s liquidity measure.

In summary, the analysis of Amihud’s liquidity measure, for the eight Dutch financial institutions, results in coefficient of the interaction term 𝛽3 that is relatively low. The term is insignificant and does not differ from zero. So, there is not enough evidence to conclude that the lifted short selling ban in the Dutch equity markets results in a more liquid market (using the Amihud’s liquidity measure).

5.1.2. Transaction costs measure

Because liquidity is a complex variable I use two different measures and formula (2), explained in the methodology section, offers the bid-ask spread. When the bid-ask spread is large, the lower the amount of trades will be. Illiquid markets are associated with a high spread because of the higher transaction costs. As in all the regressions, I present two different time windows. Table (1) present a summary of the Stata output and in this section columns (3) and (6) will be discussed. Table (6) and (7) shows a more detailed overview about the regressions with the bid-ask spread as dependent variable. This section will compare the output of the two dependent liquidity variables with each other.

For calculating the bid-ask spread formula (2) is used, which is a measure of liquidity. Again, the model is described in the methodology section by formula (6). The dependent variable is the bid-ask spread and the independent variables are the two binary variables and the interaction between the two. For each dataset the ask spread is calculated. This returned the daily bid-ask spread for the eight Dutch financial institutions (treatment group) with the control group (the selected 25 Danish banks). The determination of the coefficient 𝛽₃ coincides with the coefficient that is computed in the regression. These are calculated as described in previous sections using the differences-in-differences method.

First I will analyze column (3) of table (1) which is the smaller time window. The coefficient of 𝛽1 has a slight negative value and the p-value turns out to be less than the 1% significance level. There is enough evidence (alpha 1 percent) to conclude that after the inception date (June 1st, 2009) the volatility decreased in the Dutch (treatment) and the Danish (control) equity

(29)

29 market. The reason for this will not be investigated and discussed in this paper. Again, this paper is only interested in the 𝛽₃ coefficient.

According to the value of 𝛽₂ = −0.0297 the treatment group has a lower spread and therefore more liquid than the control group with a significant value of 1 percent. This could be but the underlying reason for this will not be discussed or investigated in this paper.

When analyzing the 𝛽₃ in table (3), I noticed a low positive number but the t-statistics is not significant and I cannot reject the null hypothesis. According to table (6) the p-value is 0.257 and therefore I cannot draw any conclusions. Interesting is column (6) of table (1) which provide a summary of the bid-ask spread with a six months’ time window. The 𝛽3 has a significant (alpha 1 percent) positive value. The confidence interval contains only positive values. So, I can conclude that this coefficient differs significantly from zero. Such a result indicates that the observed result would be highly unlikely under the null hypothesis. This interaction of time and treatment tells how much the liquidity increased or decreased in the treatment group than it has in the control group. In this case the, 𝛽3, is positive meaning that the liquidity in the treatment group has decreased after the lifted short selling ban. This is an interesting result. This contradicts the literature which explains that the ban causes more liquidity in the equity market. Consequently, it is logical to expect a more liquid equity market when the short selling restriction was lifted in the Netherlands.

6. Conclusion

This paper has investigated the lifted short selling restriction on eight Dutch financial institutions. Only those eight financial institutions suffer from the short selling restriction. The ban was introduced on October 10th, 2008 and only constrained uncovered short selling. Not much later the Dutch government expanded the restriction on covered short selling (September 21st, 2008). The inception date of the lifted short selling ban was June, 1st 2009 and around this date I investigated the volatility and liquidity in the Dutch equity market. In this paper I explained that it is logical to use volatility and liquidity as a proxy for market quality. Hereby I can conclude whether the market quality decreased or increased after the lifted short selling ban in the Netherlands.

(30)

30 When I interpret the output of the regression one in reference to the intraday volatility measure, it has become apparent that I can prove a relation between the treatment group after the lifted short selling date (eight Dutch financial institutions) and volatility. I can conclude that for both observed time windows, the lifted short selling ban in the treatment group had a significant effect on intraday volatility. The calculated t-statistic and p-values prove that the null hypotheses 𝛽3 = 0, meaning that the lifted short selling ban had no effect on intraday volatility for the treatment group, can rejected. Conclusively I can say that there is a negative correlation between the dependent and independent variables. The calculated coefficients, which are quite small, are all proven to be significant. The significance and the coefficient value has increased as I used a larger time window. I can conclude that the lifted short selling ban in the Netherlands causes a decrease in volatility in the market.

Besides the volatility measure, this paper uses multiple measures for calculating liquidity. Amihud’s liquidity measure and the bid-ask spread are used to detect a correlation between liquidity and the lifted short selling restriction in the treatment group. When I study the output summary of the regressions, there are some interesting conclusions to be drawn.

When I investigate the output summary of the Amihud’s ratio, for which two regressions have been done, the lifted ban has no significant effect on the Amihud’s liquidity measure. The first time event, which is 50 days before and after the lifted ban, shows no significant value and I cannot reject the null hypothesis, meaning that the lifted short selling ban had no effect on the Amihud’s liquidity measure. I must draw the same conclusion from the regression in the second time event (six months around the inception date).

The second liquidity measure is the bid-ask spread. I cannot find any significant values in the output summary referring to the small time window. For this time event I cannot reject the null hypothesis, meaning that there is not enough evidence to conclude that the lifted short selling ban has a negative or positive impact on the liquidity in the Dutch equity market. When analyzing the longer time window, the interaction term is significant positive. Even though the coefficients are relatively small, there is a significant relationship. This means that the Dutch equity market has become more illiquid after the lifted short selling ban which is in contradiction with the literature.

(31)

31 It is important to be critical about this conclusion. There could be a possible sign of endogeneity in the last discussed regression. If policy makers tend to impose such short selling bans at times when stocks tend to become illiquid for some other reason, the correlation between short selling bans and market illiquidity could not be interpreted as a causal relationship. This problem could be solved by adding more explanatory variables in the model or by using instrumental variables (IV) regression. Where the first stage is a linear probability model determining the likelihood of a lifted ban and the second stage models its effects on liquidity. Another explanation would be that during the lifted short selling ban the equity market was less stressful than it was at the introduction of the short selling restriction.

Further study could investigate multiple time events of the lifted short selling ban. Also, more measures could be used, some of which are already mentioned in the Methodology section. This would give a more comprehensive picture of the effects of a lifted short selling ban on volatility and liquidity. I could also expand this study by observing other countries who also lifted their short selling ban. Or perhaps use a different control group with no ban at all or that imposed bans on all stocks, so that the control group is formed by stocks in countries that imposed no bans and exempt stocks in countries that imposed partial bans.

(32)

32

7. References

Amihud, Yakov. 2002. “Illiquidity and Stock Returns: Cross-Section and Time-Series Effects.” Journal of Financial Markets 5(1): 31–56.

Angel, James J., and Douglas M. McCabe. 2008. “The Business Ethics of Short Selling and Naked Short Selling.” Journal of Business Ethics 85(S1): 239–49.

Battalio, Robert, Hamid Mehran, and Paul Schultz. 2012. “Market Declines: What Is

Accomplished by Banning Short-Selling?” Current Issues in Economics and … (August 2011): 1–7.

Beber, Alessandro, and Marco Pagano. 2013. “Short Selling Bans Around the World: Evidence from the 2007–09 Crisis.” The Journal of Finance LXVIII(1).

Bertrand, Marianne, Esther Duflo, and Sendhil Mullainathan. 2003. “How Much Should We Trust Differences-in-Differences Estimates?”

Bogdan, S, S Bareša, and S Ivanović. 2012. “Measuring Liquidity on Stock Market: Impact on Liquidity Ratio.” Tourism and Hospitality … 18(2): 183–93.

Bris, Arturo, William N Goetzmann, and Ning Zhu. 2007. “Efficiency and the Bear : Short Sales and Markets Around the World.” LXII(3).

Culp, CL, and J.B. Heaton. 2008. “Economics of Naked Short Selling, The.” Regulation. Daly, Kevin James. 2011. “An Overview of the Determinants of Financial Volatility: An

Explanation of Measuring Techniques.” Modern Applied Science 5(5): 46–63.

Diamond, Douglas W., and Robert E. Verrecchia. 1987. “Constraints on Short-Selling and Asset Price Adjustment to Private Information.” Journal of Financial Economics 18(2): 277–311.

Engle, Robert F, and Giampiero M Gallo. 2003. “A Multiple Indicators Model For Volatility Using Intra-Daily Data.” (May): 1–27.

Helmes, Uwe, J Henker, and Thomas Henker. 2010. “The Effect of the Ban on Short Selling on Market Efficiency and Volatility.” Available at SSRN 1688135: 1–55.

Lee, Bong-Soo, and Oliver M. Rui. 2002. “The Dynamic Relationship between Stock Returns and Trading Volume: Domestic and Cross-Country Evidence.” Journal of Banking &

Finance 26(1): 51–78.

Sarr, A, and Tonny Lybek. 2002. “Measuring Liquidity in Financial Markets.” IMF working

(33)

33

8. Output summary

Table (1) Differences-in-differences estimation of the interaction term for volatility (intra-day measure) and liquidity (Amihud’s measure and

bid-ask spread). VARIABLES (1) Volatility 50 days window (2) Liquidity Amihud 50 days window (3) Liquidity Bid-Ask spread 50 days window (4) Volatility six months window (5) Liquidity Amihud six months window (6) Liquidity Bid-Ask spread six

months TIME -0.00805***1 -0.0369 -0.00444*** -0.0112*** 0.0587 -0.0154*** (0.00106)2 (0.117) (0.00168) (0.000625) (0.082927) (0.000932) TREATMENT 0.00785*** -0.287* -0.0297*** 0.0111*** -0.32077 -0.0362*** (0.00138) (0.155) (0.00265) (0.000892) (0.126074) (0.00159) TIME*TREATMENT -0.00331* 0.0400 0.00416 -0.00855*** -0.064059 0.0131*** (0.00192) (0.216) (0.00366) (0.00114) (0.163813) (0.00205) Constant 0.0258*** 0.299*** 0.0380*** 0.0295*** 0.35162*** 0.0442*** (0.000744) (0.0828) (0.00121) (0.000495) (0.0633447) (0.000727) Observations 1,342 1,343 1,967 5,194 5,194 7,403 R-squared 0.098 0.005 0.106 0.143 0.143 0.126

1_{When * is added in the coefficient is means that the particular coefficient is significant at the 90 percent confidence level. ** means it is} significant at the 95 percent confidence level and *** means it significant at the 99 percent confidence level.

(34)

34

Table (2) Output summary of intraday volatility with a 50 days’ time window.

(35)

35

Table (4) Output summary of Amihud’s liquidity measure with a 50 days’ time window.

(36)

36

Table (6) Bid-ask spread with a 50 days’ time window.

(37)

37

Table (8) Volatility, 50-days’ time window

Table (9) Volatility, six months’ time window

Table (10) Liquidity Amihud’s measure, 50-days’ time window

Table (11) Liquidity Amihud’s measure, six months’ time window

Table (12) Liquidity Transaction cost measure, 50-days’ time window

(38)

38

Figure (1) Volatility, 50-days’ time window

(39)

39

Figure (3) Liquidity Amihud’s measure, 50-days’ time window

(40)

40

Figure (5) Liquidity Transaction cost measure, 50-days’ time window

(41)

41

Graph (1) Time plot, volatility Denmark

The effect of the lifted short selling ban on market volatility and liquidity