The U.K. short saleban 2008/2009 and stock prices

The U.K. Short Sale Ban 2008/2009 and

Stock Prices

Abstract

The UK imposed a ban on short selling at the start of the financial crisis in September 2008. The goal was to raise stock prices in the financial sector and stabilise the market. However, short selling provides liquidity and stimulates price discovery. Two regressions are performed on stock prices of both affected firms and a control group. The results are significantly declining prices for all stock during the ban and there is a significant negative interaction effect found for financial stock and the short selling ban for both stock prices and trade volume.

Keywords: Short selling, stock price, regulation, FSA JEL Classification: G01, G18

Bachelor’s Thesis

Name: Maia ten Kortenaar

University: University of Amsterdam Student number: 10459979

Bachelor: Economics and Business Specialization: Economics and Finance Supervisor: J.J.G. Lemmen

i

Statement of Originality

This document is written by student Maia ten Kortenaar who declares to take full responsibility for the contents of this document.

I declare that the text and the work presented in this document is original and that no sources other than those mentioned in the text and its references have been used in creating it.

The Faculty of Economics and Business is responsible solely for the supervision of completion of the work, not for the contents.

ii

Table of Content

1. Introduction ……….. 1

1.1 Motivation ………... 1

1.2 Research Question and Hypotheses .………...…….. 1

1.3 Brief Results ………... 2

1.4 Structure ……..……….. 2

2. Literature Review ………...… 3

2.1 Short Selling ……….. 3

2.1.1 Types of Short Selling ……….... 3

2.1.2 Positive Effects of Short Selling ……….... 4

2.1.3 Negative Effects of Short Selling ……….. 5

2.1.4 Informed Investors ………. 6

2.1.5 Stock Price Manipulation ………7

2.2 Short Sale Bans ………. 8

2.2.1 The Short Sale Ban in the UK ……… 8

2.2.2 Similar Short Sale Bans in Other Countries ………... 9

2.3 Hypotheses ………... 11

3. Data and Methodology .……….. 12

3.1 Data ……….. 12

3.2 Model ………... 12

4. Results ………. 15

4.1 Price and Pooled OLS ……….. 15

4.2 Price and Fixed Effects ………...…. 16

iii

5. Conclusion ..………..….. 24 References ………...…… 26 Appendix ………. 29

1

1. Introduction

1.1 Motivation

There is much debate about whether short selling is ethically just or not. This is because investors that short sell profit from the misery of others or even create it. In some countries it is wholly prohibited and in others partly or not at all regulated, there is no general consensus among countries. Although it is a controversial subject with some downsides, short selling also provides the market with benefits. For example, prices that better reflect the true value of stock and their firms. There is still a lot to be debated and researched, as opinions are very diverse on the subject.

Not only the ethical side of short selling is frequently debated, there is another aspect of short selling under attention. In the recent financial crisis bans on short selling were imposed in some countries, generally the intention was to stabilize the capital market. Governments and financial regulators see a ban on short selling as a tool usable for a quick interference in the capital market. However, interfering is not without consequences. Banning short selling has a large impact on market efficiency and price discovery. While there is substantial research available today, mostly pointing in the direction that constraining short selling leads to market deterioration, still many countries choose to impose some kind of ban on short selling in times of market turmoil.

One of the countries imposing a ban on short selling as a reaction to the start of the financial crisis in 2008 was the United Kingdom. The Financial Services Authority imposed a ban on the 18th of September in 2008 and lifted the ban on the 16th of January in 2009. Only financial stocks were targeted, because these were particularly volatile and vulnerable. This period of banning short selling in the UK makes for a rich dataset and thus a good opportunity to see what the effects of such a ban are. In my opinion it is worth researching what the effects were on the stock market prices and if the UK achieved her goal, because existing literature points in the opposite direction.

1.2 Research Question and Hypotheses

The introduction above gives notion to the research question in this thesis. What was the effect of the short sale ban 2008-2009 on stock prices in the U.K.? The ban on short selling was a means to increase the stock prices of financial firms and to prevent the financial market from crashing.

2

The hypotheses below will help answering this question. Further support for the hypotheses will be provided in chapter two.

Hypothesis 1: Stock prices in general will rise during a ban on short selling and reverse when

the ban is repealed.

Hypothesis 2: Stock prices for financial firms will drop during the ban, financial stock reacts

differently to the short selling ban.

Hypothesis 3: Trading volumes for financial stock will drop to very low amounts relative to

non-financial stock. The volume will rise when the ban is lifted.

1.3 Brief Results

Two types of panel regressions are performed on stock prices, including dummy variables for the ban, financial stock and an interaction estimate between these two. In the pooled OLS regression the financial dummy is significant at 1% with a coefficient value of -0.4445, indicating lower prices for financial stock. The time dummies indicate that the lowest prices were during the short selling ban. The interaction variable is significant and has a negative effect of 20.46%. The fixed effect model used on stock prices generates roughly the same outcomes except for an insignificant interaction variable. A fixed effects regression on trade volume indicates a significant general increase of 30% in trade volume and an interaction effect of -47.3% for financial stock during the ban period.

1.4 Structure

In chapter two a literature review is provided, necessary background information is presented and relevant research is discussed. In the third chapter the data and manipulations will be explained as well as the methodology, a panel regression with fixed effects and a pooled OLS are used to analyse the impact of different variables on stock prices and trade volume. In chapter four results of the three regressions will be presented. Finally, in chapter five conclusions will be drawn and limitations will be discussed. Future recommendations for research will be suggested.

3

2. Literature review

This chapter presents the origin of short selling and explains the mechanisms of both covered and naked short selling. It also discusses the benefits and downsides of short selling. Finally the relevant support for the hypotheses will be presented and relevant literature and researches will be discussed.

2.1 Short Selling

Short selling originated in The Netherlands in the 1600s, around the same time and place where the first stock market came into existence. Traders of shares of the Dutch East Asia trading Company expected a drop in value of the institution because of a merger with a French rival. They discovered that there was money to be made from these falling stock prices. They formed a secret trading group and made great profits as the shares of the Dutch East Asia trading Company dropped around twelve percent in value. The government and other traders were not at all content with this new form of trading. Almost immediately after this discovery, short selling was found to be too controversial and was banned in 1610 (Bris et al., 2007).

2.1.1 Types of Short Selling

There are two types of short selling currently used in the markets, covered short selling and naked short selling. Naked short selling is more controversial and is more commonly prohibited. The short selling process for a covered short works as follows. An investor borrows shares from a shareholder. In most cases this process goes through a broker. The investor now has borrowed some stocks but he or she does not own them. The shares are sold and delivered to a third party, the buyer. At a predetermined settlement day the investor is required to have bought the shares back to return them to the original lender (Culp and Heaton, 2008). If the stock has decreased in price in the meanwhile, the investor was able to buy them back at a lower price than the selling price and thus made a profit.

When one doesn’t borrow the stock from a broker or some other market participant, the process is called naked short selling. This is known to be more controversial, as the investor might have no intention of delivering the shares or it is not believed that it is reasonably possible to locate and buy the stock by the settlement day (Culp and Heaton, 2008). When the number of written shares is extremely high, it might be impossible to buy the shares before settlement day. In this case there are no tangible or existing shares to begin with, so it is possible to give out and sell

4

an infinite number of shares, called phantom shares. In this way one can drive the stock price down to zero, which is one of the negative effects of naked short selling (Angel and McCabe, 2009). In conclusion, naked short selling can lead to more severe problems than covered short selling. This is because the number of shares that can be sold is unlimited, there is no collateral.

2.1.2 Positive Effects of Short Selling

There are certain benefits attached to short selling stock, though policy makers might indicate otherwise. Short selling provides market efficiency, this efficiency comes from the facts that short selling allows for better reflection of true values of stock, also called price efficiency or discovery, and it provides higher liquidity in the market.

Miller (1977) argues in his paper that in the case of short sale constraints, stock prices are upwardly biased. This is best intuitively understood by imagining a pessimistic investor. He or she expects prices to drop, but is not allowed to participate in the market because of the constraint on short selling. Because of this, part of the negative information is not included in the stock prices. This leads to optimistic investors bidding prices above what should be the average perceived fair price for the stock, thus the upward bias. This discovery has implications for the efficient market hypothesis, as it states that all available information should be included in prices. Miller’s findings contradict the efficient market hypothesis, as not all available information is incorporated. This might point towards an inefficient market.

In line with Miller’s paper is the research of Diamond and Verrecchia (1987). They model the short sale constraint in a rational expectations framework. The conclusion found is that restraining short sales slows down the speed at which the market reflects new private information in the stock prices, this effect is even stronger for negative private information. Thus, allowing short selling in the market may lead to prices better reflecting true values and faster information incorporation in prices. Boehmer and Juan (2013) found additional evidence for this statement, more active short sellers lead to more accurate stock prices. When short sellers become more active, the incorporation of information accelerates and stock prices will close in on their fundamental values. Short sellers even seem to change their behaviour in response to occurring extreme events, their behaviour stimulates price discovery and reduces stock price divergence from fundamental values.

By not imposing any constraints on short selling, the number of participants in the market increases. Trade volumes will increase because there are more participants, leading to lower

5

transaction costs. In the end this leads to more liquidity in the market (Financial Services Authority, 2009). Considering the number of participants in the financial market, not many of them are individual investors. Most of the short selling is done by hedge funds and pension funds, when short selling is constrained this becomes a severe problem for these parties. Leading to a great drop in market participants as mentioned above (Financial Services Authority, 2009). Short selling allows hedge funds and other financial institutions to operate the way they do, creating a liquid stock market and providing their product to the consumer. According to Duffie et al. (2002) institutional investors are preferable when it comes to lending stock. These institutions hold stock over longer periods of time and are not likely to recall the stock. Restricting this large group of institutional investors might lead to a severe drop in liquidity as mentioned above.

2.1.3 Negative Effects of Short Selling

The FSA (2009) provides a list of potential problems attached to short selling in a discussion paper on short selling. The first downside is abusing short selling to create misleading signals about the true value of a stock and the real supply. The stock price can be manipulatively pushed down, this is a larger problem for firms in the financial sector as they depend more on the confidence of their customers and counterparties. In times of market instability, the problem is even greater. The potential of abuse is more evident for naked short selling relative to covered short selling. As naked short selling is not limited by the need to borrow the stock, for naked short selling one only needs to find buyers. Though, naked short selling is not actively pursued in Europe.

The second problem addressed by the FSA (2009) is the self-fulfilling prophecy that can arise when investors overreact on downward trends in prices. When they react correctly, short selling can lead to better reflected values in stock prices. However, when they overreact and the confidence gets too low, prices might drop below their fundamental value. If this is the case, firms might get in trouble by not being able to raise funds. If the lack in confidence spreads throughout the market and contaminates other stock this might become a self-fulfilling prophecy. Bank runs can occur and firms will go bankrupt. This in combination with an already stressed market can in the end lead to disorderly markets.

6

Another possible downside found by Bris et al. (2007), is that short selling might have an effect on the size of negative returns. The frequency of negative returns is not affected by short selling only the size of the returns. Shkilko et al. (2009) have found results in line with this argument, short selling may cause price pressure and can substantially increase the price declines. Short selling is not flagged and disclosed in the UK, this means a less transparent market as uninformed market participants do not know the amount of stock being shorted. This can be seen as a potential problem (Financial Services Authority, 2009). If the market is informed on the aggregate short position, it can make a just estimation to which extent short selling is driving stock prices as well as the amount of overhang that needs to be covered at some point in time. Short selling has been blamed for market crashes, although there is not yet conclusive evidence for this or the opposite (Chang and Yu, 2004). Bris et al. (2007) agree with this statement, they mention short selling is as old as stock markets but there is still no crushing evidence indicating whether short selling prevents or facilitates market crashes.

2.1.4 Informed Short Sellers

According to Boehmer et al. (2008) short sellers provide more than 20% of the total trading volume. Short sellers are assumed to have relevant information on stock values and contribute substantially to price discovery as found by Diamond and Verrecchia (1987). However, informed investors do not always have incentives to disclose their knowledge. They might have motives to minimize information leakage (Boehmer et al., 2013). Engelberg et al. (2012) find that a great part of the advantage of short sellers comes from their ability to analyse publicly available information. Market makers seem to underperform relative to clients, the latter has better informed shorts.

There is enough evidence supporting above stated arguments, but there is not much proving that short sellers are exactly aware of what they are doing. Most shorting is done by institutions, these shorts are assumed to be more informed relative to shorting done by individuals (Boehmer et al., 2008). Institutions are informed by doing expensive research, though individuals might know important information because of their job position. It is forbidden to trade one’s corporate stock, but it is still possible to short stock in closely related substitute corporations. Boehmer et al. (2008) are the first to examine the information content and incidence of short sales in the U.S., they found short sellers to be extremely well informed. Especially institutional short sellers seem to identify and act on important value-relevant information, which has not

7

yet been incorporated in stock prices. These price effects are permanent, so stock price manipulation can be ruled out.

Short sellers can be seen as well informed investors. Institutional short sellers invest money in thorough research and find relevant information, whereas individuals have relevant information by being close to the source.

2.1.5 Stock Price Manipulation

Short selling can also be used to manipulate stock prices, this was one of the main fears of the FSA in 2008 at the beginning of the financial crisis. One manipulation strategy is called a bear raid, this happened in the early years of the first stock exchange in the Netherlands. An investor would massively start selling a particular stock, other investors got frightened and sold their shares as well. This large increase in supply would cause the stock price to drop. Then the first investor was able to buy back the initial shares at a lower price with a profit. Buying back the shares would lead to a rise in stock price, back to the original level (Allen & Gale, 1992). Trading pools work the other way around, a group of investors buy stocks of a particular firm and then spread favourable rumours about the firm. This causes a rise in stock price, the investors are able to sell their shares at a profit. The same mechanism works for unfavourable rumours and shorting stock, causing stock prices to drop (Allen & Gale, 1992).

Above mentioned methods are called action-based manipulation and information-based manipulation, around the world these activities were declared illegal early in the 20th century and they still are to this day, so these problems seem to be in the past (Allen & Gale, 1992). Regulation has been fairly effective with a few exceptions. Though, the effectiveness of any restriction depends on the costs of finding a loophole or avoiding the regulation (Grundy et al., 2012).

There is a third category of manipulation, called trade-based manipulation. A trader buys and sells stock large amounts of stock, not using any publicly observable tactics. Allen and Gale (2012) found that this strategy is profitable and might be socially desirable.

So there are ways to manipulate stock prices, financial authorities have regulated these channels but there are still exceptions found and large schemes rolled up.

8

2.2 Short Sale Bans

Governments can interfere in the short selling market in various ways. The most common manners to interfere will be discussed in this subchapter.

One way to constraint short selling is called a blanket ban by the Financial Services Authority, which covers both covered short selling and naked short selling in a prescribed market (2009). An alternative regulation is a ban only cast on naked short selling, this is more commonly done. Naked short selling is more controversial than covered short selling because it allows one to give out an infinite number of shares and the price to drop to zero. Some jurisdictions prohibit naked short selling in general or have temporarily banned this process.

Thirdly, a government can choose to ban a particular segment of the short selling market, usually the financial market because this is most vulnerable in times of financial crisis. The UK chose to do this in September 2008 (Financial Services Authority, 2009) with financial stock. There are some other ways to constrain short selling such as circuit-breakers and tick rules. A circuit-breaker is a pause on trading a particular stock when there is an abnormal increase or decrease in its price. Tick rules can be up-tick or zero-tick. An up-tick rule states that the last sale must be at a higher price than the sale preceding it, if this is not the case the stock cannot be sold short. If the zero-tick rule is active, the last price must be unchanged but higher than the preceding sale (Financial Services Authority, 2009).

Short selling bans have asymmetric effects for investors holding favourable and unfavourable information (Figlewski, 1981), in other words normal traders and short sellers. This can make significant differences in the integration of information in stock prices as found by Figlewski and many others after. If this is wished for, then it might be a wise choice to constrain short selling. Although at this moment, evidence is lacking in the case of banning short selling and reaching set goals.

2.2.1 The Short Selling Ban in the UK

The Financial Services Authority imposed a ban on short selling on the 19th of September in 2008. The ban imposed in the UK was a blanket ban, directed at a specified list of financial firms. Specifically, 29 firms were targeted at the time of announcement. Later this was altered to 32 firms. The statement provided by the Financial Services Authority states that active creation or increase of net short positions in publicly quoted financial companies is prohibited. They also required daily disclosure of all net short positions in excess of 0.25 percent of the

9

ordinary share capital of the relevant companies as well as disclosure of net short positions held at close on 19th _{September. The constraint was declared effective until 16}th _{January 2009, but}

is to be reviewed after 30 days. The blanket ban did not apply to market makers and existing short positions were left untouched. The reason to interfere was the extreme market turbulence at the time. Stock price volatility was very high and persistent. It was feared that financial stock was seriously undervalued, by banning short selling the FSA hoped to increase the stock prices. In the case of financial stock there was fear of downward pressure on prices (2009). By banning short selling manipulative short sellers could not drive the prices further down.

There is no definition in the law for short selling in the UK, so there is no way in determining whether a sale is long or short. Because of this, there is no tick rule effective in the UK. Lack of a tick rule can result in a quick build-up of short interest and related returns might be more unconstrained (Hodgkinson et al., 2012).

The intentions of the FSA were probably nothing but good, but recent research points out that banning short selling did not lead to the intended results. Marsh and Payne (2012) found that market conditions for the financial market were not significantly different from others prior to the ban. Prices were dropping in all sectors and trading costs were increasing. They argue that the ban was the cause of the UK equity markets deterioration, liquidity drained from the markets and market efficiency was greatly reduced. Trading costs increased massively. They state that the ban even exacerbated the existing problems.

2.2.2 Similar Short Selling Bans in Other Countries

In 2008, at the start of the financial crisis, other financial authorities came up with regulations for the stock market similar to the FSA’s, for example the SEC, Securities and Exchange Commission, in the United States. It started with a temporary ban on naked short selling on all U.S. stocks, the SEC chose to make the ban permanent on July 27, 2009. In line with the FSA, the SEC also imposed a ban on the financial sector on the 19th September in 2008. The restriction applies to a total of 799 financial firms. The SEC and FSA cooperate on an ongoing basis, so motives for banning shorting are similar. As stated in the publication of the SEC, the financial market is too dependent on market confidence and panic selling. In normal and stable circumstances short selling provides liquidity and contributes to price efficiency. Prices were declining and Lehman brothers just fell, these were no stable market conditions. The ban was

10

repealed on 8th _{October 2008, short selling was permitted again but the ban on naked shorting}

is still in effect to this day (SEC, 2008).

During the ban shorting activity for large capital stock declines by around 77% (Boehmer et al., 2009). Algorithmic traders cannot act as informal market makers, this leaves formal market makers in a market with less competition. The formal market makers are able to create greater returns at the expense of those demanding liquidity, thus market quality deteriorates.

Fotak et al. (2009) argue that the ban harmed market efficiency, in line with Boehmer et al. (2009). They found that price declines for firms hit hardest by the financial meltdown, were not caused by naked shorting. In fact, naked short sellers seem to respond to public news and price declines rather than trigger them. No evidence was found that covered or naked short selling triggered large stock price declines or credit rating downgrades.

Beber and Pagano (2013) report results that compliment Boehmer et al. (2009) as well, the price effect of the short selling ban was neutral at best, whereas the whole objective of the FSA and SEC was to increase stock prices. Furthermore, they found the effects of a short selling ban were worse for other countries than for the United States, the implied liquidity reduction was larger.

Chang and Yu (2004) researched the stock market and short selling constraints in Hong Kong, their conclusion was that constraining short selling causes stock overvaluation. This overvaluation gets worse with a greater dispersion of investor opinion (2004). They also report higher volatility of stock returns when there are no constraints in place.

Three days after the FSA and SEC initiated the short selling regulations, the Netherlands imposed a ban on naked short selling. The AFM (Authority Financial Markets) prohibited naked shorting for a number of financial institutions. The approach differs from the U.S. and the U.K. as it left covered short selling unregulated. The fees for naked shorting in Europe are substantially high such that not many investors dare to use this strategy. Leaving covered short selling unregulated leaves room for liquidity, but takes out the potentially harmful trading acts.

11

2.3 Hypotheses

This chapter discusses the hypotheses used along with the relevant support found in literature research.

Hypothesis 1: Stock prices in general will rise during a ban on short selling and reverse when

the ban is repealed.

This hypothesis is the key to the research question, the objective of the FSA was to raise stock prices to stable the market. It is supported by Miller (1977), by banning shorting a part of the available information is not incorporated in the prices. According to Miller this will lead to a change in prices, in this case the pessimistic view on prices is left out. So, the average price will rise on banning short selling because most information incorporated is positive. The same conclusion is found by Nagel (2005). He concludes that short sale constraints can prevent pessimistic views of the market and stock being included in stock prices.

Hypothesis 2: Stock prices for financial firms will drop during the ban, financial stock reacts

differently to the short selling ban.

This statement is supported by research done by Beber and Pagano (2013), they found the stock prices effect to be neutral at best. This might be because there is an interaction effect for financial stock and the ban period. Marsh and Payne (2012) conclude prices continued to be dropping in all sectors during the ban.

Hypothesis 3: Trading volumes for financial stock will drop to very low amounts relative to

non-financial stock. The volume will rise when the ban is lifted.

Support comes from Marsh and Payne (2012), they found trading volumes for financial stock to decrease substantially more than non-financial stock.

12

3. Data and Methodology

In this chapter the sample selection and performed manipulations will be explained. Next, the model used is provided and multiple regression methods are discussed.

3.1 Data

The FSA targeted a select group of financial firms at the time of the crisis, 32 firms in total. The FTSE 100 is used to be able to compare these firms with a control group. The historical constituent list of the FTSE 100 provides the firms in the FTSE during September 2008. This list is found in DataStream and includes financial stocks as well as stocks from all other sectors, DataStream provides a list of 102 firms. Firms that left the FTSE 100 or went bankrupt during the period of the ban are left in the sample to prevent survivorship bias. The non-financial stocks remaining after the subtraction of the affected firms will provide the control group. By using the entire data set, a selection bias is prevented. One firm is left out of the data set as trading volumes were missing for the first few months. This leaves a total number of 101, of which 17 are in the financial sector and were targeted by the FSA.

Using DataStream stock prices, trade volumes, market values and return on investment for all stocks are found, these are all daily observations. The stock price volatility is calculated for each stock by taking the standard deviation over the entire period. The period used is from the 1st _{of January 2008 to 31th of August 2009, this embraces the ban by 6 months pre and post,}

not too broad of a period but enough to give a good view of the emerging crisis and a regenerating market. Total number of trading days is 435, the ban was active for 86 trading days. In Stata all the data sets are merged and set to panel data.

3.2 Model

In this thesis a regression model with a difference-in-difference term is used to examine stock prices during the short selling ban. Marsh and Payne (2012) used a difference-in-difference regression in their research, a similar model is used here but without some of the interaction variables1. This design is made to measure the effect of a treatment on the target data by

1 _{The original model by Marsh and Payne (2012) is: 𝑦}

𝑡= ∝ +∝1𝐷𝑖,𝑡𝐹 +∝2𝐷𝑖,𝑡𝑃𝑟𝑒+∝3𝐷𝑖,𝑡𝐵𝑎𝑛+∝4𝐷𝑖,𝑡𝑃𝑜𝑠𝑡+∝5𝐷𝑖,𝑡𝐹 ×

𝐷_𝑖,𝑡𝑃𝑟𝑒_+∝

6𝐷𝑖,𝑡𝐹 × 𝐷𝑖,𝑡𝐵𝑎𝑛+∝7𝐷𝑖,𝑡𝐹 × 𝐷𝑖,𝑡𝑃𝑜𝑠𝑡+∝8𝑉𝑖,𝑡+∝9𝑇𝑖,𝑡+ 𝜀𝑖,𝑡. As a relatively small sample of data is used here, too

many explanatory variables could cause over fitting. That is why 𝐷𝑖,𝑡𝐹 × 𝐷𝑖,𝑡𝑃𝑟𝑒 and 𝐷𝑖,𝑡𝐹 × 𝐷𝑖,𝑡𝑃𝑜𝑠𝑡 are removed from the model. Since Marsh and Payne accounted for market capitalization by matching the financial firms to a group of similar sized non-financial firms, they don’t have a control variable for market capitalization. In this thesis the firms are not matched, so a control variable for market capitalization 𝑀𝑖,𝑡 is added.

13

comparing the treated group and the control group, often used in physics and other beta-research, this model is used extensively in studies on change in treatments. A clear distinction can be made between banned stock and a non-banned stock control group by using dummy variables.

To measure the effect on stock prices of the different variables, the model below is used. 𝑌_𝑡 = ∝ +∝₁ 𝐷_𝑖,𝑡𝐹 +∝₂ 𝐷_𝑖,𝑡𝑃𝑟𝑒+∝₃ 𝐷_𝑖,𝑡𝐵𝑎𝑛+∝₄ 𝐷_𝑖,𝑡𝑃𝑜𝑠𝑡+∝₅ 𝐷_𝑖,𝑡𝐹 × 𝐷_𝑖,𝑡𝐵𝑎𝑛+∝₆ 𝑀_𝑖,𝑡+∝₇ 𝑉_𝑖,𝑡

+∝8 𝑇𝑖,𝑡+∝9 𝑅𝑜𝐼𝑖,𝑡 + 𝜀𝑖,𝑡

Where 𝑌_𝑡 is the variable of interest, in this case the stock price. The right-hand side of the equation begins with a constant, α. The next parameter is 𝐷𝐹_{, this is a dummy variable for the}

type of stock. The value is one for financial stock and zero for non-financial stock. 𝐷𝑃𝑟𝑒 _{is one}

for the period 1st January to 18th September and zero otherwise. 𝐷𝐵𝑎𝑛 selects stock during the ban period, one for ban period and zero for no ban imposed. 𝐷𝑃𝑜𝑠𝑡 is a dummy for post ban, its value is one for the period after 19th January and zero in all other cases. Then there is an interaction variable 𝐷𝐹𝑥𝐷𝐵𝑎𝑛 which estimates whether the financial stock behaved differently during the ban relative to non-financial stock. The next variable is market capitalization or market value 𝑀𝑡, this is the share price multiplied by the number of ordinary shares outstanding.

𝑉_𝑡 is the volatility of the stock price over the entire period calculated separately for each stock and 𝑇_𝑡 stands for daily trading volume. Return on investment, 𝑅𝑜𝐼_𝑖,𝑡, is the daily calculated return index as calculated by DataStream. Trade volume, market capitalization and return on investment are added because they have been found to be significant in earlier papers by Boehmer et al. (2009) and Bris et al. (2007). The last term is the error term.

For hypothesis 1 ∝₁ and ∝₄ are of concern, when these coefficients are significant one can conclude that the ban had an effect on prices. ∝1 should be positive and ∝4 negative, this

indicates a rise in price for imposing the ban and a drop in prices after repealing the ban. Looking at ∝₃, one can see what happens to the dependent variable during the ban period but cannot conclude that this effect was caused by the ban. It is important to look at the time dummy variables relative to one another and keep in mind that not everything is cause and effect. Furthermore, if ∝₅ is significant this means that the ban had a different effect on financials stock relative to non-financial stock. Hypothesis 2 can be tested with ∝5, the

difference-in-difference term looks at the difference-in-difference in behavior between financial and non-financial stock prices. The same coefficient is convenient for hypothesis 3, only the independent variable for

14

the regression in this case must be trade volume. To prevent collinearity the control variable trade volume must be omitted for this regression.

With this panel data set one could do a fixed effects or random effects panel regression, random effects would be more efficient if the data is compatible. A Hausman test is performed for identifying if there is a significant result between the fixed and random effects. The test yields a Chi² value of 733.95 and the difference is significant at 1%, this indicates that effects differ and that the fixed effects model is the right model to use. The performed Hausman test can be found in the appendix as table 6. Though, fixed effects does not provide coefficients on 𝐷𝐹 _and

𝑉_𝑖 as these values do not change over time. Fixed effects controls for time invariant differences, the values of the coefficients would be incorporated in the total control. Since the dummy ban is of concern, a pooled OLS regression is performed alongside the fixed effects model. A pooled OLS estimator is more likely to be biased due to omitted variables, though better in this situation relative to a random effects regression as this would be inconsistent and still most likely be biased.

Before running these regressions, the natural logarithm is taken for price, market value, trade volume, return on investment and volatility. Histograms provided by Stata indicate that all these variables are positively skewed, taking the logarithm leads to a more normal distributed variable and thus a better fit for the model.

15

4. Results

In this chapter the results of pooled OLS regression are presented and discussed. In this first part descriptive statistics on the data are discussed.

The analysis of the data set starts with a graph on average stock prices for financial stock and non-financial stock over the entire sample period. Figure 1 can also be found in the appendix. At first glance stock prices seem to be falling for both types of stock. When the ban was imposed the fall in prices is even steeper but they seem to pick up halfway through the ban. After the ban is lifted one can identify an increasing trend in the price for both stocks. Though, the prices do not yet reach their old values from the pre ban period.

Figure 1. Average stock price

3 0 0 4 0 0 5 0 0 6 0 0 7 0 0 8 0 0 9 0 0 1 0 0 0 Ave ra g e st o ck p ri ce 01jan2008 01apr2008 01jul2008 01oct2008 01jan2009 01apr2009 01jul2009 01oct2009

16

Below four tables on descriptive statistics during the ban are found. From these tables it is already possible to see that the average price dropped for both financial and non-financial stock during the ban, this points in the opposite direction of what Miller (1997) argued in his paper. Both trade volume and market value decreased for financial and non-financial stock during the ban, this is in line with hypotheses 2 and 3. Return on investment seems to have decreased for both groups, but not in large amount. Tables 1a, b, c, d and e on the descriptive statistics can be found in the appendix.

Table 1b

Descriptive statistics Financials

Variable Mean Std. Dev. Min Max

id 50.23529 28.11204 1 86 Price 551.75 492.3132 19.99 3717.97 Marketvalue 14264.35 21382.61 321.47 116526.6 Trade volume 5672.566 5745.625 2 55713 RoI 17159.81 25212.91 12.99 120827.6 Price volatility 177.9359 258.0089 29.83212 1153.352 Table 1c

Descriptive statistics Financials during ban

Variable Mean Std. Dev. Min Max

id 50.23529 28.11965 1 86 Price 451.8357 329.4188 39.17 2205 Marketvalue 11953.87 20380.91 321.47 111840.4 Trade volume 4755.173 5511.302 2 55713 RoI 13918.75 20914.88 12.99 104555.4 Price volatility 177.9359 258.0787 29.83212 1153.352 Table 1d

Descriptive statistics Non-Financials

Variable Mean Std. Dev. Min Max

id 51.15476 29.35963 3 101 Price 788.9795 670.0028 45.79 5847.17 Marketvalue 11867.63 17560.1 676.61 122488 Trade volume 4559.222 3621.205 2 80375 RoI 29688.81 76026.87 26.66 603248.6 Price volatility 163.7335 209.6817 5.50185 1365.327

17 0 50 00 10 00 0 15 00 0 20 00 0 Av era ge tra de v ol um e 01jan2008 01apr2008 01jul2008 01oct2008 01jan2009 01apr2009 01jul2009 01oct2009 Date

Non-financial stock Financial stock

Figure 2 in the appendix is a graph of average trading volume per type of stock. Right before the ban trading volume for financial stock increased massively. During the ban financial trading volume is lower than non-financial, pre and post ban period this is reversed. One can derive from this that financial trade volume fell more relative to non-financial trade volume.

Figure 2. Average trade volume

4.1 Price and Pooled OLS

Before running the regression, a White test for heteroskedasticity is performed on the data set. The Chi² value is significant at 1% and thereby the null hypothesis indicating homoskedasticity can be rejected. So, when running the OLS regression robust standard errors are required to correct for heteroskedasticity. Next a Woolridge test is conducted to detect potential autocorrelation in the panel data. This test too, is significant at 1%. For these reasons clustered

Table 1e

Descriptive statistics Non-Financials during ban

Variable Mean Std. Dev. Min Max

id 51.15476 29.36123 3 101 Price 675.218 551.8218 45.79 3502.68 Marketvalue 10502.12 16253.53 821.84 104915 Trade volume 5486.003 4742.814 4 80375 RoI 25717.51 64946.31 26.66 453535.2 Price volatility 163.7335 209.6932 5.50185 1365.327

18

standard errors are reported, these are robust to heteroskedasticity and serial correlation. Both tests can be found in the appendix as table 6 and 7.

Table 2 in the appendix provides the pooled OLS regression, each column means an additional control variable added to the model. Looking at the final regression (5), all variables are included. Here 𝐷𝐹 _{is significant with a coefficient value of -0.4445. This indicates a 44.45%}

lower price if the stock is from a financial firm relative to a non-financial firm. So there is a pricing difference for both types of stock.

𝐷𝑃𝑟𝑒 _{has a coefficient of 0.3189 and is highly significant, before the regulation of short selling}

prices are 31.89% higher for both financial and non-financial stock. The coefficient of the ban dummy 𝐷𝐵𝑎𝑛 is also highly significant and would imply that stock during the ban had a value 15.55% higher compared to a period with no regulations. Because these are time dummies, there is always one active. This means the 15.55% increase for the banned period is actually a decrease of 31.89-15.55 = 16.34% for stock prices during the short selling ban. This is not in line with Miller’s (1997) and Nagel’s (2005) expectations on overvaluation for stock prices when short sellers are not allowed to trade. This finding disarms hypothesis one.

𝐷𝑃𝑜𝑠𝑡 _{is significant so there is a price change when the ban is lifted. When 𝐷}𝑃𝑜𝑠𝑡 _{is active (value}

of one) stock prices rise by 8%, but when this is compared to the contributions to stock price of pre period and ban period this means a drop in stock price relative to 31.89% pre ban and 15.55% during the ban respectively. So looking at hypothesis 1, there is a reduced stock price post ban. Stock prices keep on falling during the sample period.

Then the final variable is the interaction between the ban and financial stock 𝐷𝐹_{× 𝐷}𝐵𝑎𝑛_{, the}

coefficient is significant at 5% and has a value of -20.46%. This difference-in-difference estimate implies that financial stock does behave differently compared to non-financial stock during a short selling regulation. Prices for financials were 20.46% lower relative to non-financials. Hypothesis two is thereby confirmed as financial stock prices were predicted to be lower for financial stock.

Concluding one can say that the ban had a negative effect on prices for all stock, financial or non-financial. There is a significant interaction effect for financial stock and the short sale ban, meaning even lower prices for financial stock as compared to non-financial stock during the ban. 𝐷𝐹 is the main reason to run this pooled OLS as it is omitted in fixed effects. The variable is significant at 1% and has a large value of -44.45%, meaning a much lower stock price for financial stock.

19 Table 2 Pooled OLS regression - Price

lnprice (1) (2) (3) (4) (5) Df -0.4195* -0.4472* -0.4127* -0.0416* -0.4445*** (-1.83) (-1.81) (-1.73) (-1.84) (-2.89) Dban 0.0886*** 0.0711*** 0.102*** 0.0994*** 0.1555*** (3.24) (3.13) (3.20) (3.25) (5.69) Dpre 0.4217*** 0.3221*** 0.3027*** 0.2904*** 0.3189*** (8.09) (2.94) (6.54) (6.67) (8.03) Dpost 0.0846*** 0.0546*** 0.0617*** 0.0565*** 0.08*** (3.79) (2.94) (3.19) (3.00) (4.07) Df*Dban -0.0196 -0.0129 -0.1173 -0.1067 -0.2046** (-0.35) (-0.30) (-1.43) (-1.48) (-2.58) lnmktvalue 0.2551*** 0.3269*** 0.2932*** 0.2998*** (3.25) (4.07) (3.54) (7.28) lntradevol -0.1435* -0.1405* -0.3304*** (-1.75) (-1.75) (-7.21) lnroi 0.0723** 0.0673*** (2.18) (3.41) lnpricevol 0.61*** (13.14) constant 6.108*** 3.948*** 4.498*** 4.199*** 2.913*** (65.38) (5.97) (5.89) (5.67) (7.15) Number of obs. 42193 42193 42193 42193 42193 R² adj. 0.0677 0.1701 0.176 0.213 0.6714

20

4.2 Price and Fixed Effects

Since the Hausman test indicated fixed effects to be the best fit for the data set, a panel regression accounting for fixed effects is performed after the OLS. In the previous paragraph heteroskedasticity and autocorrelation are confirmed for the data. So in this fixed regression the option clustered is selected in Stata, this option corrects for heteroskedasticity and serial correlation in the data set. The fixed effects results can be found below and in the appendix in table 3. Again, in the table (1), (2), (3) and (4) denote an extra control variable added to the model. There is no 𝐷𝐹 _{because this is a time-invariant variable, Stata automatically omits this}

variable from the model. The same applies to 𝑉_𝑖, average stock price volatility does not change over the time period.

Focusing on model (4), R² within is 0.9979 which is very high. When using a fixed effects model, one should focus on the R² within provided. This measures the goodness of fit for the individual mean de-trended data which disregards all between information. 𝐷𝑃𝑟𝑒 has a value of 0.02193 and is significant at 1%. So when this dummy is active prices increase by 2.19%. Comparing this again with 𝐷𝐵𝑎𝑛, this coefficient is significant at 1% and indicates 0.36% higher stock prices during the ban period. 𝐷𝑃𝑜𝑠𝑡 is significant with a value of -0.01586. So post ban stock prices were lower by 1.586%. Again these time dummies need to be interpreted relatively to one another because there is always one active. This means the highest prices occurring pre ban and falling prices for the ban period and post ban. So hypothesis 1 cannot be confirmed, prices did not rise after imposing the short selling ban, these results are in line with results found by the OLS regression.

The difference-in-difference estimate (interaction variable) has a coefficient value of -0.0014 but is insignificant. This is in contradiction with the results found by the OLS method. Financial stock has stock prices 0.14% lower relative to non-financial stock. These findings are applicable to hypothesis two, stock prices for financial stock did not behave differently according to the fixed effects model and this hypothesis can thereby not be confirmed.

Comparing findings from both regression methods there are quite a few similarities and one substantial difference. Both indicate lower stock prices for all stocks during the short selling ban and even lower prices after the ban. There is a negative interaction effect for financials and the ban. The interaction effect is not significant in the fixed effects regression. This is an indicator that the ban had the same effect on all stock and not just affected the banned stock. Further research might by interesting on this matter because of the contradicting findings. The

21

dummy for financial stock in the pooled regression has a negative effect on prices by 44.45%, in the fixed effect model this dummy is accounted for in general fixed effects. Finally, post ban there seems to be significantly lower prices according to the fixed effects model and OLS model.

Table 3 Fixed effects panel regression - Price

lnprice (1) (2) (3) (4) Dban 0.0886*** 0.0214** 0.0135** 0.0036*** (3.24) (2.49) (2.04) (4.90) Dpre 0.4217*** 0.039** 0.0312** 0.0219*** (8.09) (2.59) (2.14) (14.94) Dpost 0.0846*** -0.0306*** -0.0318*** -0.0156*** (3.79) (-3.23) (-3.16) (-13.54) Df*Dban -0.0196 0.0061 -0.0097 -0.0014 (-0.35) (0.26) (-0.31) (-0.66) lnmktvalue 0.9798*** 0.9709*** 0.0245*** (26.99) (29.41) (3.12) lntradevol 0.0114 0.004*** (9.31) (7.16) lnroi 0.9776*** (129.43) constant 6.037*** -2.276*** -2.278*** -1.755*** (183.89) (-7.24) (-6.57) (-64.17) Number of obs. 43935 43935 42193 42193 R² within 0.3184 0.8417 0.8583 0.9976 R² overall 0.0363 0.1146 0.1185 0.0724

22

4.3 Trade Volume

To examine trade volumes for both types of stock, the same model as explained in chapter three can be used, this time with trade volume as independent variable. Of course the control variable trade volume is omitted when running the fixed effects regression. The decision for a fixed effects regression is again determined by a Hausman test. The White- and Wooldridge tests are conducted, both indicating the presence of heteroskedasticity and serial correlation. For this reason the regression is ran using robust and clustered standard errors. Results of the regression can be found on the next page and in the appendix along with the other test results in table 4, 5, 6 and 7.

𝐷𝐹 and price volatility are omitted from the regression as these variables are time invariant. The ban, 𝐷𝐵𝑎𝑛, seems to have a positive effect on trade volume by 30%, this coefficient is significant at 1%. This contradicts hypothesis 3 and is not in line with figure 2.

𝐷𝑃𝑟𝑒 and 𝐷𝑃𝑜𝑠𝑡 are both significant, with values of 0.151 and 0.125 respectively. These values need to be interpreted relative to 𝐷𝐵𝑎𝑛 _{as well, as they are time dummy variables. During the}

ban trade volumes is at the highest value, with approximately 50% lower volume pre and post ban. This means that there is no significant increase after the ban was repealed, so part of hypothesis 3 can be refuted.

The interaction variable 𝐷𝐹_{× 𝐷}𝐵𝑎𝑛 _{in this regression is very interesting. It is significant at 1%}

and has a value of -0.473. This means that for financial stocks during the ban trade volume decreased by 47.3% which is almost half the original value. So the short selling ban did have a different effect on the financial stock in the sample. This makes sense because short selling for these stocks was no longer allowed and around 20% of all trading is short selling. Hypothesis 3 can be confirmed partly for financial stock. There is no indication that post ban trading volume increased substantially but there is a strong difference for financial and non-financial stock and the effect on trade volumes during the ban.

23 Table 4 Fixed effects – Trade Volume

lntradevol (1) (2) (3) Dban 0.3022*** 0.2994*** 0.2979*** (6.80) (6.53) (6.38) Dpre 0.1682*** 0.1522** 0.151** (3.10) (2.47) (2.42) Dpost 0.1281*** 0.1234** 0.1252** (2.77) (2.54) (2.63) Df*Dban -0.4758*** -0.4746*** -0.4732*** (-6.96) (-7.01) (-7.05) lnmktvalue 0.0411 -0.072 (0.97) (-0.54) lnroi 0.1167 (0.81) constant 8.009*** 7.661*** 7.716*** (168.55) (22.31) (24.00) Number of obs. 42193 42193 42193 R² within 0.0324 0.0328 0.0337 R² overall 0.0126 0.1736 0.0007

24

5. Conclusion

In this thesis the effect of the short selling ban on stock prices in the United Kingdom is examined. The FSA found the financial market to be in disorder in September 2008 and chose to impose a short selling ban mainly to stabilise the market and raise prices. However, short selling is one of the most controversial subjects in the finance world and the FSA’s decision caused quite some turbulence. Interfering by banning particular trades is not without consequences. Short selling has a large impact on market efficiency, liquidity and price discovery, these are the main things usually provided by short sellers. So cutting them out of the market does not seem like the best thing to do.

A pooled OLS regression is performed on stock prices as well as a fixed effects regression, both with relevant control variables and dummy variables for periods and grouping. A Hausman test indicates fixed effects is preferred to random effects, though in that case no value is obtained for the financial firm dummy variable. For this reason a pooled OLS regression is added to the research, to be able to compare the regressions in hindsight. The main conclusions of the regressions are significant results for lower prices during the short selling ban. In both models the interaction effect between the financial dummy and ban dummy is negative, though not significant in the case of fixed effects. The OLS regression yields a significant value of -20.46% for the interaction estimator. Since these findings differ between both models it might be interesting to conduct some further research. The pooled OLS indicates that for financial stock the prices dropped even further. The dummy for financial stock in the OLS regression is highly significant with a value of -0.4445, so financial stock has substantially lower prices compared to non-financial stock. During the ban period trade volume was significantly rising for all stock, except financial stock. There is a highly significant decrease of 47.3% in trading volume for the latter during the ban.

Taking these findings into account along with recent research, for example by Marsh and Payne (2012) or Beber and Pagano (2013), the short selling ban did not have the effects the FSA had intended. Intentions were to raise stock prices, not drive them further down the spiral. By banning short sellers from the financial stock market the prices of these stock decreased further and their trading volume decreased massively. Short sellers have a bad name in the financial system, but research indicates they are not all that bad. The United Kingdom would probably have been better off with not interfering in the capital markets, though one cannot say with conviction what would have happened if the FSA did not interfere. The short selling ban was imposed during a very turbulent time, the start of the financial crisis.

25

As the chairman of the SEC Christopher Cox already stated, knowing what we know now we would not do it again. The costs seem to outweigh the benefits.

Further research can be directed at liquidity and short selling, banning short selling takes away a substantial amount of traders and their money. The effect of this can be measured by bid-ask spreads. The case examined in this thesis contains a ban on all types of short sales, there is no distinction between covered or naked short selling. It might be interesting to investigate whether these two have different effects on the market. For instance, as mentioned in the literature review, the Netherlands can be investigated on her naked short selling ban. Further research can also include cross-sectional data between different countries or look at intra-day data. Using intra-day data should give a good view of the continuous changes happening in the market, though the data is harder to come by.

26

References

AFM. (2008, June 2). Naked short selling verboden. Retrieved from:

http://www.afm.nl/nl/en/nieuws/2008/sep/naked-short-selling-verboden.aspx

Angel, J.J. & McCabe, D.M. (2009). The Business Ethics of Short Selling and Naked Short Selling, Journal of Business Ethics, 85(1), 239-249.

Allen, F. & Gale, D. (1992). Stock price manipulation. The Review of Financial Studies, 5(3), 503 – 529.

Autore, D.M., Billingsley, R.S. and Kovacs, T. (2011). The 2008 short sale ban: Liquidity, dispersion of opinion, and the cross-section of returns of US financial stocks, Journal of

Banking & Finance, 35(9), 2252-2266.

Beber, A. & Pagano, M. (2013). Short selling bans around the world: Evidence from the 2007-09 crisis, The Journal of Finance, 68(1), 343-381.

Boehmer, E., Jones, C. M., & Zhang, X. (2008). Which Shorts Are Informed? The Journal of

Finance, 63(2), 491-527.

Boehmer, E., C.M. Jones & X. Zhang (2009). Shackling short sellers: The 2008 shorting ban. Retrieved from http://papers.ssrn.com/sol3/papers.cfm?abstract_id=1412844

Boehmer, E., & Wu, J. J. (2013). Short selling and the price discovery process. Review of

Financial Studies, 26(2), 287–322.

Bris, A., Goetzmann, W.N., and Zhu, N. (2007). Efficiency and the Bear: Short Sales and Markets around the World. The Journal of Finance, 62(3), 1029-1079.

Chang, E. C., & Yu, Y. (2004). Short-Sales Constraints and Price Discovery : Evidence from the Hong Kong Market. The Journal of Finance, 62(5), 1-44.

27

Culp, C. L. & Heaton, J.B. (2008). The Economics of Naked Short Selling, Regulation, 31(1), 46-52.

Diamond, D. W. & Verrecchia, R. E. (1987). Constraints on short selling and asset price adjustment to private information, Journal of Financial Economics, 18(2), 277–312.

Duffie, D., Gârleanu, N., & Pedersen, L. H. (2002). Securities lending, shorting, and pricing.

Journal of Financial Economics, 66(2-3), 307–339.

Engelberg, J.E., Reed, A.V. and Ringgenberg, M.C. (2012) How are shorts informed? Short sellers, news, and information processing, Journal of Financial Economics, 105(2), 260-278.

Figlewski, S. (1981). The Informational Effects of Restrictions on Short Sales: Some Empirical Evidence. The Journal of Financial and Quantitative Analysis, 16(4), 463-476.

Figlewski, S., & Webb, G. P. (1993). Short Sales, and Market Completeness. The Journal of

Finance, 48(2), 761-777.

Financial Services Authority. (2009). Short selling. Financial Services Authority discussion paper 09/1.

Fotak, V., V. Raman & P.K. Yadav (2009). Naked short selling: The emperor’s new clothes? Retrieved from: https://traditions.ou.edu/content/dam/price/Finance/files/Naked_Short_ Selling.pdf

Grundy, B.D., Lim, B. and Verwijmeren, P. (2012). Do option markets undo restrictions on short selling? Evidence from the 2008 short-sale ban, Journal of Financial Economics, 106(2), 331-348.

Hodgkinson, L., Jafaar, A., Mohamad, A. and Wells, J. (2013). Short selling and stock returns: Evidence from the UK, The British Accounting Review, 45(2), 125-137.

28

Marsh, I.W. & Payne, R. (2012). Banning short sales and market quality: The UK’s experience,

Journal of Banking & Finance, 36(7), 1975-1986.

Miller, E. M. (1977). Uncertainty, and Divergence of Opinion, The Journal of Finance, 32(4), 1151-1168.

Nagel, S. (2005). Short sales, institutional investors and the cross-section of stock returns,

Journal of Financial Economics, 78(2), 277-309.

Security Exchange Commission. (2008, May 10). SEC halts short selling of financial stocks

to protect investors and markets, Retrieved from: https://www.sec.gov/news/press/2008/

2008-211.htm

Shkilko, A.V., Van Ness, B.F. and Van Ness, R.A. (2009). Aggressive short selling and price reversals. Working paper, available at SSRN: <http://ssrn.com/abstract=971210>.

29

Appendix

The first part of the appendix consists of descriptive statistics for different periods in the sample and distinguishing between the types of stock. The tables provide the mean, standard deviation, minimum and maximum values of all variables excluding the dummy variables. Id is the variable used to set the data as panel data.

Table 1a

Descriptive statistics

Variable Mean Std. Dev. Min Max

id 51 29.15509 1 101 Price 749.0498 649.6253 19.99 5847.17 Marketvalue 12271.04 18281.33 321.47 122488 Trade volume 4743.878 4072.384 2 80375 RoI 27579.97 70257.65 12.99 603248.6 Price volatility 166.124 218.6265 5.50185 1365.327 Table 1c

Descriptive statistics Financials during ban

Variable Mean Std. Dev. Min Max

id 50.23529 28.11965 1 86 Price 451.8357 329.4188 39.17 2205 Marketvalue 11953.87 20380.91 321.47 111840.4 Trade volume 4755.173 5511.302 2 55713 RoI 13918.75 20914.88 12.99 104555.4 Price volatility 177.9359 258.0787 29.83212 1153.352 Table 1b

Descriptive statistics Financials

Variable Mean Std. Dev. Min Max

id 50.23529 28.11204 1 86 Price 551.75 492.3132 19.99 3717.97 Marketvalue 14264.35 21382.61 321.47 116526.6 Trade volume 5672.566 5745.625 2 55713 RoI 17159.81 25212.91 12.99 120827.6 Price volatility 177.9359 258.0089 29.83212 1153.352

30 Table 1d

Descriptive statistics Non-Financials

Variable Mean Std. Dev. Min Max

id 51.15476 29.35963 3 101 Price 788.9795 670.0028 45.79 5847.17 Marketvalue 11867.63 17560.1 676.61 122488 Trade volume 4559.222 3621.205 2 80375 RoI 29688.81 76026.87 26.66 603248.6 Price volatility 163.7335 209.6817 5.50185 1365.327 Table 1e

Descriptive statistics Non-Financials during ban

Variable Mean Std. Dev. Min Max

id 51.15476 29.36123 3 101 Price 675.218 551.8218 45.79 3502.68 Marketvalue 10502.12 16253.53 821.84 104915 Trade volume 5486.003 4742.814 4 80375 RoI 25717.51 64946.31 26.66 453535.2 Price volatility 163.7335 209.6932 5.50185 1365.327

31

Figure 1. Graph of average stock prices during the sample period, the red lines indicate the period of the short selling ban.

3

0

0

4

0

0

5

0

0

6

0

0

7

0

0

8

0

0

9

0

0

1

0

0

0

Ave

ra

g

e

st

o

ck

p

ri

ce

32

0

5

0

0

0

1

0

0

0

0

1

5

0

0

0

2

0

0

0

0

01jan2008

01apr2008

01jul2008

01oct2008

01jan2009

01apr2009

01jul2009

01oct2009

Date

Non-financial stock

Financial stock

33

Table 2 Pooled OLS regression (clustered, robust standard errors) – Price

The independent variable is the natural logarithm of the daily stock prices. Df indicates the type of stock, a value of 1 means the stock is financial. Dban is a time dummy with value 1 from 18th September 2008 to 16th January 2009. Dpre is the period before the 18th of September and Dpost indicates the period after the 16th _{of January. Df*Dban is the difference-in-difference estimate, it multiplies the 1’s and 0’s from the separate}

variables, creating a new variable. Lnmktvalue, lntradevol and lnroi are the natural logarithms of the daily values of market value, trade volume and return on investment respectively. Lnpricevol is the natural logarithm of the stock price volatility over the entire sample period.

lnprice (1) (2) (3) (4) (5) Df -0.4195* -0.4472* -0.4127* -0.0416* -0.4445*** (-1.83) (-1.81) (-1.73) (-1.84) (-2.89) Dban 0.0886*** 0.0711*** 0.102*** 0.0994*** 0.1555*** (3.24) (3.13) (3.20) (3.25) (5.69) Dpre 0.4217*** 0.3221*** 0.3027*** 0.2904*** 0.3189*** (8.09) (2.94) (6.54) (6.67) (8.03) Dpost 0.0846*** 0.0546*** 0.0617*** 0.0565*** 0.08*** (3.79) (2.94) (3.19) (3.00) (4.07) Df*Dban -0.0196 -0.0129 -0.1173 -0.1067 -0.2046** (-0.35) (-0.30) (-1.43) (-1.48) (-2.58) lnmktvalue 0.2551*** 0.3269*** 0.2932*** 0.2998*** (3.25) (4.07) (3.54) (7.28) lntradevol -0.1435* -0.1405* -0.3304*** (-1.75) (-1.75) (-7.21) lnroi 0.0723** 0.0673*** (2.18) (3.41) lnpricevol 0.61*** (13.14) constant 6.108*** 3.948*** 4.498*** 4.199*** 2.913*** (65.38) (5.97) (5.89) (5.67) (7.15) Number of obs. 42193 42193 42193 42193 42193 R² adj. 0.0677 0.1701 0.176 0.213 0.6714

34

Table 3 Fixed effects panel regression (clustered, robust standard errors) - Price The independent variable is the natural logarithm of the daily stock prices. Dban is a time dummy with value 1 from 18th _{September 2008 to 16}th _{January 2009. Dpre is the period}

before the 18th of September and Dpost indicates the period after the 16th of January. Df*Dban is the difference-in-difference estimate, it multiplies the 1’s and 0’s from the separate variables, creating a new variable. Lnmktvalue, lntradevol and lnroi are the natural logarithms of the daily values of market value, trade volume and return on investment respectively. Time invariant variables are omitted from the regression.

lnprice (1) (2) (3) (4) Dban 0.0886*** 0.0214** 0.0135** 0.0036*** (3.24) (2.49) (2.04) (4.90) Dpre 0.4217*** 0.039** 0.0312** 0.0219*** (8.09) (2.59) (2.14) (14.94) Dpost 0.0846*** -0.0306*** -0.0318*** -0.0156*** (3.79) (-3.23) (-3.16) (-13.54) Df*Dban -0.0196 0.0061 -0.0097 -0.0014 (-0.35) (0.26) (-0.31) (-0.66) lnmktvalue 0.9798*** 0.9709*** 0.0245*** (26.99) (29.41) (3.12) lntradevol 0.0114 0.004*** (9.31) (7.16) lnroi 0.9776*** (129.43) constant 6.037*** -2.276*** -2.278*** -1.755*** (183.89) (-7.24) (-6.57) (-64.17) Number of obs. 43935 43935 42193 42193 R² within 0.3184 0.8417 0.8583 0.9976 R² overall 0.0363 0.1146 0.1185 0.0724

35

Table 4 Fixed effects panel regression (clustered, robust standard errors) – Trade Volume The independent variable is the natural logarithm of the daily trade volumes. Dban is a time dummy with value 1 from 18th September 2008 to 16th January 2009. Dpre is the period before the 18th of September and Dpost indicates the period after the 16th of January. Df*Dban is the difference-in-difference estimate, it multiplies the 1’s and 0’s from the separate variables, creating a new variable. Lnmktvalue and lnroi are the natural logarithms of the daily values of market value, trade volume and return on investment respectively.

lntradevol (1) (2) (3) Dban 0.3022*** 0.2994*** 0.2979*** (6.80) (6.53) (6.38) Dpre 0.1682*** 0.1522** 0.151** (3.10) (2.47) (2.42) Dpost 0.1281*** 0.1234** 0.1252** (2.77) (2.54) (2.63) Df*Dban -0.4758*** -0.4746*** -0.4732*** (-6.96) (-7.01) (-7.05) lnmktvalue 0.0411 -0.072 (0.97) (-0.54) lnroi 0.1167 (0.81) constant 8.009*** 7.661*** 7.716*** (168.55) (22.31) (24.00) Number of obs. 42193 42193 42193 R² within 0.0324 0.0328 0.0337 R² overall 0.0126 0.1736 0.0007

36

Table 5 Hausman test for fixed and random effects

H0: Difference in coefficients is not systematic (random effects) Dependent variable

Lnprice Lntradevolume Chi² 733.95 (6) 235.81 (6) Prob. > Chi² 0.0000*** 0.0000***

Degrees of freedom are presented in parentheses, *,**, *** denotes significance at 10%, 5% and 1% respectively

Table 6 White test for heteroskedasticity H0: Homoskedasticity

Dependent variable

Lnprice Lntradevolume Chi² 10074.65 (41) 3238.18 (31) Prob. > Chi² 0.0000*** 0.0000***

Degrees of freedom are presented in parentheses, *,**, *** denotes significance at 10%, 5% and 1% respectively

Table 7 Wooldridge test for autocorrelation in panel data H0: No first order autocorrelation

Dependent variable

Lnprice Lntradevolume

F(1,100) 220514.04 95.08

Prob. > F 0.0000*** 0.0000***