• No results found

The effect of hihg frequency trading on stock price volatility

N/A
N/A
Protected

Academic year: 2021

Share "The effect of hihg frequency trading on stock price volatility"

Copied!
24
0
0

Bezig met laden.... (Bekijk nu de volledige tekst)

Hele tekst

(1)

The effect of High Frequency

Trading on stock price volatility

Junhao Fu 10466584

Programme : Economics and Business

Track: Finance and Organization

Supervisor: Drs. P.V. (Pepijn) Trietsch

(2)

Abstract.

This paper investigates the relationship between High Frequency tradings(HFTs) and stock price volatility. As the huge increasing awareness of HFT activity in recent years, it has posed the crucial question of whether HFTs are beneficial for financial markets to both investors and firms. There are some existing researches have studied the impact of HFT on different measures of market efficiency, such as liquidity, price discovery and latency, but most of the results are not conclusive. Therefore, built on pervious researches, this study focuses on the impact of High frequency tradings on stock price volatility over the period 2014-2015 for a sample of 30 selected companies in U.S stock market. The amount of High frequency tradings (HFTs) are calculated by a formula built up based on Frank Zhang’s formula in 2010. Data is collected from CRSP and Thomason one until September 2015. The data of positions of Institutions investors and individual investors will be used to calculate HFT, and together with other common factors that affect volatility such as the past 12 months return, turnover rate, market value on equity, stock price level, interest rate and inflation rate, a model will be built up to analyze the contribution of HFT on stock price volatility. The result is significant, even though there is no sufficient evident to show that the HFT has impact on the stock price for those securities served in financial secto r, but in general the study shows there are negative impact on the HFTs in the price volatility among these 30 companies. However, the impacts are smaller compare to other factors that included into the model. Therefore, this paper concluded that HFT has negative impact on stock price volatility, but the impacts are relatively small.

1. Introduction

1.1 problem and central question

With an advanced and well developed computer science today, algorithmic trading system has become the most popular trading platform around the world. It saved a lot of time during the transaction in the market. It allows investor establish specific strategy beforehand both for buying and selling a certain security. Once the strategy has been programmed, it will be automatically executed by the computer.

In general, there are some advantages for High frequency trading, for example, it helps investor to minimize emotional investment. Due to the trading was set in program and it will execute tradings automatically, it’s so a rational investment behavior can be ensured. Secondly the trading strategy can be backtested, which means HFT can be applied into historical market data to determine the viability of the strategy. And there is no room for interpretation

(3)

after the program started. Therefore it helps the investor to test the expectancy of the market from computers and so as to adjust their strategy into real world. Moreover, HFT helps to improve the trading efficiency. Which investors are benefited from the immediate responds of computers so they would not have any delays in trading. As soon as the position is set, orders are automatically traded to prevent over losses. Also the trading system allows investor to diversify their interest in stocks. Therefore they can open severe accounts and trade with different strategy and securities at the same time. It also helps investors to hedge risks and prevent any unexpected losses. However, some drawbacks also come along with HFTs. Which it is highly rely on computers therefore if something goes wrong with the systems or any bugs in the software it will lead to a costly loss. Also investors who highly rely on HFT are always over-optimization. When they employ backtesting techniques the systems always provide the optimal outcomes and didn’t take into account with unexpected losses. Therefore when HFTs are implemented into real trading, the target price are always not set to be a most profitable level and sometimes even made a loss because of over expectation on the market.

There are some motivations for starting this research. Due to its characters of short holding period and high amount of trading, HFT are always seemed to be a major factor that interfere the market risk. One of examples is the black flash . which started at 2.30 pm on 6th may, the US stock market indexes, such as the

S&P 500, Dow Jones Industrial index and Nasdaq, collapsed and rebounded very rapidly within 36 minutes. The Dow Jones Industrial index had 9% dropped just within a very short time period. It was commonly thought that High-Frequency Tradings is one of the main reasons that caused the sudden crash on stock market. Data also shows that there are huge amount of HFT activities during the trading day and most investor start to blame the sudden price drop to HFT. Another example is lots of traditional traders complained that HFT has always push up or lower down the price level just before their deal done, which they use a high speed computers to finish the deal faster than traditional traders, so that they have to accept a price level of a deal that they did not expect for. And thus it increases the price level volatility.

However, according to some post analysis of the incident, supporters of HFT defended that HFT did not cause the crash. CME Group(2010) stated that, according to their investigation by using statistical data, there is no significant evidence to support that the huge number of HFT activities was related to the crash, in fact the report shows that HFT improved the stability of market price. The statement leads to another round of debate regarding to the effect of HFT to market price level.

Hence, I was motivated to analyze to see if HFT has impact on stock price volatility, the central question is: Does High-frequency contribute impact on stock price volatility? An empirical methodology will be applied in this research and it will be explained in the later paragraphs.

(4)

1.2 Historical background of High Frequency Trading

As one of largest subsets in algorithmic trading, High-frequency tradings (HFTs) is also highly rely on the strategy, but the price will not be set up either too high nor too low, therefore the program was expected to executed a deal within a very short time(normally not overnight). HFT is introduced in 1999 after the new authorization on E- stock exchange by the U.S. government, with characterized of short holding period, speedy trade and high turnover ratio, its proportion of market share was soon to begin rapid growth and start to dominate U.S stock market in early 21st century. According to data from the NYSE(2009), the HFT volume has been grew for 149% during the year of 2005 to 2008. There is approximately $141bn of hedge fund assets were implemented HFT strategies in the first quarter of year 2009, which accounted for more than 79% market shares in the hedge fund market.

The case of Renaissance Technologies is the first case that HFT has successfully implemented their investment strategies, which HFT investors are the market makers and they increased Renaissance Technologies’ liquidity but also lower the volatility and helped narrow bid-offer spreads, the risk for the stock itself was minimized and that makes trading and investing cheaper for other market participants.

The trading volume for HFTs is also a huge number, some market leaders in HFT industry are Chicago Trading, Timber Hill and Virtu Financial. Nowadays HFT firms account for 3/4 of stock trading volume with only 2% number of total operating intermediaries (Frank, 2010) .The Bank of England (2013) also estimates that in Europe HFT accounts for approximately 40% of equity orders volume and for Asia the number is about 5-10%, with potential for rapid growth.

After HFTs strategy is now being implemented widely, it gets more difficult to deploy their profitability. According to an estimation from Frederi Viens of Purdue University(2014), the operating income for HFTs has been declined from about $5 billion in 2010 , to about $1.25 billion in 2012. It means that even though the HFTs still account for a majority of trading in stock market nowadays, the increasing uncertainty and the intense competition within the market makes them more difficult to make profit.

(5)

2.1 Autotrading

As mentioned above , HFT is a subset from autotrading , which is also called Algorithmic trading . There have been some previous studies shows that the increasing number of auto trading did not bring negative effect on stock market, which it did not enhance the price volatility.

Hendershott and Riordan( 2009) , compared automated trading and other trades in German equity market. The paper found out that the automated trading accounted for more than half of the total trading volume in top 30 stocks, and in fact the study showed that automate trading contributes more on price efficiency and no evidence shows that the volatility has been increased. A similar study is done by Chaboud, Hjalmarsson, Vega and Chiquoine(2008). They used a database that identify auto trading and classical trading separately, and the result shows that in the EBS forex market, an increase in auto trades tend to produce more liquidity in the market but also associated decreasing volatility.

2.2 Price volatility

Stock price volatility is a main factor in a sufficient equity market, it was used to measure different kind of characteristics such as asset allocation, market efficiency and risk management within an equity market. High stock volatility is neither favorable for investors, nor the firms (Bushee and Noe 2000). Risk-averse investors are always seeking for a higher premium to hold high-volatility stocks, and they don’t have a quick reaction to when there are new changes on the HFT strategies (Zhang 2006). Froot, Perold and Stein (1992) suggested that companies does not want to have a high volatility on their stock price, since higher stock price volatility always lead to higher risk of negative return and thus investors would prefer not to invest on it. Also a higher risk leads to a higher cost of capital which is also a negative impact on the firm’s operation..

2.3 The Flash Crash

After the Flash Crash incident on 2010, the U.S. government published a research and concluded that high frequency traders have increased the stock price volatility, which is a main direct cause of the crash. Kirilenko et. al. (2014) studied the 2010 Flash Crash and found the similar conclusion. They find that HFTs initially bought contracts from fundamental sellers in the E-Mini during the price level is in the historical bottom, but then proceeded to sell contracts and

(6)

compete for liquidity with fundamental sellers. Hence, at the beginning crash HFTs didn’t change their trading behavior and that confuse other investors, they believe the market was more liquid than reality. Therefore when the crash comes, it creates a huge panic among the market. Based on this analysis, Kirilenko et al. (2011) concluded that HFTs did not directly cause the crash, but they should responsible for the huge selling volume on the trading day which causes investors’ irrational decision. Menkveld and Yueshen (2015) confirmed the U.S. government's and Kirilenko's narratives about the Flash Crash.

2.4 HFT and price efficiency

Previous research has shown that the advantage of speed trading of HTF improves the liquidity of stock market. (Brogaard and Garriott, 2014) and overall it facilitates price efficiency by leading permanent price changes in a correct direction and opposite direction to transitory pricing errors. (Brogaard, Hendershott, & Riordan, 2013). In their research, they compared HFT against other trades on NASDAQ during the period of 2008- 2010, they pointed out that HFT was positively correlated with permanent price changes and negatively correlated with transitory price changes, suggesting that HFT improves price discovery. Similar results are shown in Nicholas’s research(2011), by using the same method as Brogaard Hendershott & Riordan but with different time period, he concluded that HFT was positively correlated with non-HFT, corroborating with Brogaard, Hendershott & Riordan ‘s finding.

2.5 Does HFT affects stock price volatility?

There is not yet has a final answer for this question, the outcome of researches are always vary based on different stock markets, timing and methods.

Avramovic(2010, 2012) analyzed the US equity market from 2004-2012 and concluded in his two reports that with more HFT activities involved, Bid-ask spreads are substantially getting narrow, liquidity of stock market has increased but the short term volatility has declined. They also looked at the historical long term volatility and found out that the long-term volatility did not change even though the short time volatility declined. Therefore they concluded that market are not worse with the existence of HFT.

(7)

is London Stock exchange (LSE), and they labeled the investment types as HFT, investment companies, retail etc. And used a similar regression as Brogaard, Hendershott, & Riordan, to try to isolate the impact of HFT on the stock market. They found out that with an increasing number of HFT activities, the liquidity of the market increases when the spreads were wide, and when the spreads were narrow, the market tend to be more “smoothly and liquidly ” and not likely to increase stock price volatility

Finally in the research presented by Bershova and Rakhlin (2013), the data on Tokyo and London stock market have been studied and in their finding, long-term investors were fans of traditional investment strategy. They further find that with an increasing HFTs, there is a significant compression of bid-ask spreads, but also lead to an increasing in short-term volatility. Therefore they are partly disagreed with Avramovic. After the research, they conclude that the due to high amount of transaction costs for long-term investors, the increase in volatility were more than offset by the reduction in bid-ask spreads. However, the authors emphasize that their findings should not be generalized to all long-term investor orders. For example, they expect that the reduction in spreads would not offset the effect of higher short-term volatility for large orders. Under that situation, they find that HFT is associated with a significant compression of bid-ask spreads, with an increasing short-term volatility.

One last remark on HFT and market efficiency form Benos and Sagade (2013), they agrees that this method clarifies the relationship between price efficiency and volatility. Based on the previous research, they found that if the price volatility increased because HFT cannot associate with new information , then which means HFT increase the price efficiency . If, on the other hand, HFTs cause prices to move away from fundamentals, then it creates unexpected risk on the market and therefore deteriorates the market quality.

3 Methodology

The hypothesis of the model:

H0: There is not impact of HFT on stock price volatility H1: There is impact of HFT on stock price volatility

3.1 Measurement of HFT

(8)

for stock market Investors — (1)Institutional investors, which are investment from organizations or companies with large quantity of stocks. (2) Individual investors, which has relatively small amount of stocks and invested by an individual. (3) And lastly high-frequency traders. Hence the formula for total stock turnover can be rewritten as :

Total turnover= 𝑆ℎ𝑎𝑟𝑒 𝑜𝑢𝑡 𝑠𝑡𝑎𝑛𝑑𝑖𝑛𝑔𝑉𝑜𝑙 𝑡𝑜𝑡𝑎𝑙 = 𝑉𝑜𝑙 𝑖𝑛𝑠𝑡+𝑉𝑜𝑙 𝑖𝑛𝑑𝑖𝑣+𝑉𝑜𝑙 ℎ𝑡𝑓𝑠ℎ𝑎𝑟𝑒 𝑜𝑢𝑡𝑠𝑡𝑎𝑛𝑑𝑖𝑛𝑔 = 𝑉𝑜𝑙 𝑖𝑛𝑠𝑡 𝐼𝑛𝑠𝑡 ℎ𝑜𝑙𝑑 ∗ 𝐼𝑛𝑠𝑡 ℎ𝑜𝑙𝑑 𝑠ℎ𝑟 𝑜𝑢𝑡 + 𝑉𝑜𝑙 𝑖𝑛𝑑𝑖𝑣 𝐼𝑛𝑑𝑖𝑣 ℎ𝑜𝑙𝑑∗ 𝐼𝑛𝑑𝑖𝑣 ℎ𝑜𝑙𝑑 𝑠ℎ𝑟 𝑜𝑢𝑡 + 𝑉𝑜𝑙 ℎ𝑡𝑓 𝑠ℎ𝑟 𝑜𝑢𝑡 = 𝑉𝑜𝑙 𝑖𝑛𝑠𝑡 𝐼𝑛𝑠𝑡 ℎ𝑜𝑙𝑑 ∗ 𝐼𝑛𝑠𝑡 ℎ𝑜𝑙𝑑 𝑠ℎ𝑟 𝑜𝑢𝑡 + 𝑉𝑜𝑙 𝑖𝑛𝑑𝑖𝑣 𝐼𝑛𝑑𝑖𝑣 ℎ𝑜𝑙𝑑∗ 𝐼𝑛𝑑𝑖𝑣 ℎ𝑜𝑙𝑑 𝑠ℎ𝑟 𝑜𝑢𝑡 +HFT

Frank (2010) made three assumptions about HFT, which these assumptions will be also implied in this thesis.

(1) No high-frequency trading existed before 1999

(2) There is no position held by HFT investors by the end of the month because HFT normally will not hold any positions for longer than 24 hours.

(3) Individual investors are assumed to have similar trading behavior as institutional investors and this behavior lasts over a long time period. Therefore , based on the assumptions, he derived two more formulas:

(1) Inst + Indiv = 1 (base on assumption 2, HFT do no hold any position overnight)

(2) Inst*Instto + Indiv*Indivto = To ( before 1999, so we can assume that indivto/instto remains overtime )

HFT=TO-( 𝑉𝑜𝑙 𝑖𝑛𝑠𝑡 𝐼𝑛𝑠𝑡 ℎ𝑜𝑙𝑑 ∗ 𝐼𝑛𝑠𝑡 ℎ𝑜𝑙𝑑 𝑠ℎ𝑟 𝑜𝑢𝑡 + 𝑉𝑜𝑙 𝑖𝑛𝑑𝑖𝑣 𝐼𝑛𝑑𝑖𝑣 ℎ𝑜𝑙𝑑 ∗ 𝐼𝑛𝑑𝑖𝑣 ℎ𝑜𝑙𝑑 𝑠ℎ𝑟 𝑜𝑢𝑡 )

3.2 Data collection HFT

a) Data

The primary data come from Thomson One, which is a web platform that allows me to gather financial information such as institutional holding and individual holding on selected peers or industry.

Theoretically HFTs are only involved in some popular stocks, hence 30 companies in US stock that have the most trading volume by September 2015 were selected. Which I believe there is a higher involvement of HFTs in these

(9)

stocks than others. Among these 30 stocks, there are four stocks--- two “ISHARES TRUST” stocks and two “SECTOR SPDR TRUST” stocks are from the same company with another stock. “POWERSHARES QQQ TRUST” (31th place) and “COCA COLA CO”(32place) are used to replace each one of them to avoid multicollinearity. YAHOO(which is the 33rd) is also used to replace AETERNA ZENTARIS INC due to AETERNA has an extremely low price level (below than$ 0.12 throughout the month)

Moreover , the volume defined as the volume of shares that bought by institution investor and the volume of shares sold by institution investor. From the Thomson One database, the data of these 30 companies that the percentage of their shares are hold/ traded by institutional investors and individual investors, for the institutional holding and trading data were collected, the system selected those shares that have been filled in 13F form (which is normally has more than $100million value and institution investor owns a position –either traded or hold in it.) I also collected the data from individual investor with the same method. Finally the results(which are in terms of percentage ) multiply the total number of share outstanding equal to the volume that has.

b) Observations

From the Thomson One database, the institutional shareholder Momentum report shows all the institution momentum of the selected company with certain order, their “percentage of stock in most recent positions”, “percent of changes over a period of time” and “filling type” are the main factors that analyzed in this paper. Thomson One Ownership momentum is a useful tool it can compare up to 30 companies to see the proportion of each securities were hold and traded by institutional/individual investors. However, there is also a drawback in the database, due to the fact that there are maximum 30 stocks that are allowed to input in Thomson one to analyze in the same time. Therefore the observation sample for this essay is the top 30 stocks that have most trading volume in U.S market. The list of 30 companies is in the appendix.

Moreover, there is another drawback in the Thomson one which the system will only update quarterly therefore only the data in the last date of each quarterly will be uploaded. The latest upload date was 30th September 2015.

Some stocks such as PETROBRAS and Netflix has a short history after their first initial public offering(IPO), data for these two companies are missing before the year of 2013. Hence the earliest date of data was recorded on 1st march. In order

to match the time panel as same as other securities, this paper will started collecting data from 1st of March, and in the form of quarterly data for these 30

companies.

The sample period is from 1st of march 2014 until 31st of September 2015.

(10)

are 31st of march 2014, 30th June 2014,30th September 2014, 31st December 2014.

31st of march 2015, 30th June 2015,30th September 2015. There are 7 sample

days with 210 observations in total.

Even though it is better to collect observations as more as possible, it’s so the outcome from the model will be more accurate, the 210 is the largest number of observation that can be collected from Thomson One and CRSP under these restrictions. The good thing is the data are diverse, securities are from different sectors including financial sector, media sector, vehicle sector ect… the location is also diveresed which including companies from Canada, US and Brazil. Therefore the problematic multicollinearity can be minimized in the location perspective.

3.3 Model

In order to examine the effect of HTF on volatility, the stock volatility is controlled as dependent variable y, and HTF as dependent variable 1. I have also defined some control variables to investigate how they contribute to stock volatility. %(Inst), the amount of institutional holding/total outstanding , which is the percentage of inst. First of all, study from El-Gazzar(1998) shows that the market value on equity of stocks is also a factor that affects stock volatility, hence,

the Ln(MVE) will be added to model, which defined as logarithm of market value

of equity at each day, the MVE can be calculated by price-close monthly(from Compustata)* share outstanding. All items are from Compustat daily. 1/P, the inverse stock price .Data can be obtained from CRSP. Also, Frank(2010) found out that the stock price volatility is positively correlated with past 12 months return , which RET is defined as the weight-average return for the past 12months without dividends. Data will be collected from CRSP. Moreover, historically the inflation rate also push the stock prices higher hence it will be defined as a variable named INF . Last but not least , interest rate is also a factor that influences volatility, named INT, the To which is the turnover of the security, defined as traded volume/ total share outstanding, will be added into model as variable as well. The trading volume of institutions investors will be also considered into the model as well, which is the percentage of trading volume/total share outstanding, defined as %(Volinst)(the amount of individual trading is too small, which will not be considered in the model). However, according to the study from Hireschey and Nicholas(2011), HFT is positively correlate with other non-HFT, therefore the %(Volnst) will act as the last variable which it was believed to affect the stock price volatility.

Hence, the model will be designed as:

VOLATILITY = 𝛽0 + 𝛽1 * HFTs + 𝛽2*%(Inst)+ 𝛽3 * Ln(MVE) + 𝛽4 * (1/P) + 𝛽5 * RET + 𝛽6 * INF + 𝛽7 * INT + 𝛽8* To + 𝛽9*%(Volnst) + 𝜀𝜀

(11)

3.4 Dummy variable

There will be a dummy variable added into the model, which is defined as trust, among the 30 selected securities, there are both trust securities and normal firm’s securities , which I would like to test if there are differences on HFT trading for trust firms and non-trust company.

A trust security is a trustee who manages financial assets on behalf of another. In other words it can be said that in a trust fund company all the assets are held typically in the form of a trust, therefore normally a trust is serving in the financial industry.

A company, on the other hand, might serve variety industry, represents a combination of assets and individuals with a common goal of earning profits to increase the wealth of shareholders. It is a separate legal entity, and is in the form of corporate registered under the companies act. A company business doesn’t include a partnership business or other incorporated group of persons.

Theoretically HFT is prefer to involve more activities in trust securities rather than other securities, since they these securities sever in financial sector there HFTs have more knowledge about them and asymmetric information can be reduced. Hence, an additional dummy variable is to be used to test the difference of HFT contributes to trust securities against securities from operational companies.

Dummy variable (1 if the security is a trust fund firm and 0 if it does not)

Therefore the HFT would become: 𝛽10*dummy(1 or 0) Comparison of the models:

Model 1 : VOLATILITY = 𝛽0 + 𝛽1 * HFTs + 𝛽2*%(Inst)+ 𝛽3 * Ln(MVE) + 𝛽4 * (1/P) + 𝛽5 * RET + 𝛽6 * INF + 𝛽7 * INT + 𝛽8* To + 𝛽9*%(Volnst) + 𝜀𝜀

Model 2: VOLATILITY = 𝛽0 + 𝛽1 * HFTs + 𝛽2*%(Inst)+ 𝛽3 * Ln(MVE) + 𝛽4 * (1/P) + 𝛽5 * RET + 𝛽6 * INF + 𝛽7 * INT + 𝛽8* To + 𝛽9*%(Volnst) + 𝛽10*dummy(1 or 0)+ 𝜀𝜀

(12)

4. Results/ analysis

4.1 descriptive data

After I run the summarized descriptive data, the result is below:

Variable | Obs Mean Std. Dev. Min Max ---+---

To | 210 17.99636 64.40632 .5510215 812.8557 Volatility | 210 .0200323 .0179872 .0031835 .1360818 HFT | 210 17.45659 64.4428 -.0173901 812.4101

The number for HFTs is relatively high , it is because HFT holds a position and trade it frequently. Some of them the amount of shares that have been traded in a day are even larger than the total amount of share outstanding. Therefore the descriptive data can prove that HFT does exist in these stocks.

4.2 Does HFT has significant effects on stock volatility?

First of all, I run a simple regression between stock volatility and HFT, which is VOLATILITY = 𝛽0 + 𝛽1 * HTF the results can be find in Appendix . However, the coefficient of HFT is equal to0.0001911 which is almost equal to 0. It shows that HFT does not have direct relationship volatility if the model is simple regression. The P-value is equal to 0 which also shows that the null hypothesis should be rejected. The F-value is too small, it is only 0.4689 in stata. It shows there is only 46% of volatility can be explained by HFT. Hence it suggests that a simple regression is not the best model to answer the question.

Secondly, a competed regression is run in stata, which including 9 factors that mentioned before. The regression results for the HFTs are:

Coefficient of HFT= -0.0380479, t –value = -3.67, the p-value =0, R-square =0.73 , with 95% coefficient confident, HFTs will lie between -0.0585 and -0.01758. Therefore , the result shows that, during the period from 2014-2015, the p-value of HFT is equal to 0, which suggests that the null hypothesis should be rejected. Therefore, there is sufficient evidence shows that HFT has impact on stock price

(13)

volatility ,the coefficient is -0.0380, which it brings a negative impact on the stock price volatility, in other words it decreases the stock price volatility. Which in line with most of conclusions in prior literature. However, the impact is too small to make a huge change on the volatility, with 95% confident interval the largest coefficient is equal to -0.01758, which is close to 0. Also the R-square is equal to 0.73 which means the model is relatively fit the regression. Hence , it can be concluded that HFTs in these 30 securities has a negative impact on the price volatility but the effect is small.

With the same model , other important factors’ interpretations are:

There are four factors also contributed to the stock price volatility, including: Ln(MVE), coefficient = -0.0057, P-value =0.000

1/price, coefficient = 0.0032, P-value=0.000 Turnover, coefficient = 0.03819, P-value=0.000

Vol(inst), coefficient = -0.0553532, P-value=0.073 , For the rest of factor , null hypothesis should not be reject due to a large p-value.

Therefore, because the the p-value are too small , with 95% confident interval , the amount of Market value on Equity and inverse price level also the turnover rate for the securities has positive impact on the price volatility whereas the trading volume of institutional investor has a negative impact on the price volatility. The rest of four factors of interest rate and inflation rate level and the past 12 months return and the amount of institutional holding do not contribute their impact on stock price changes. When comparing the absolute value of coefficient with HFT, the amount of institutional trading and turnover has larger value than HFT which means they have strong impact on price volatility in the model whereas the market value of equity and the inverse price level have less impact than HFT.

4.3

Model analysis with dummy variable

After adding into a dummy variable , (1 if the security is a trust fund firm and 0 if it does not), the model will be restricted into HFT of trust securities against HFT for those non trust securities. The results for the model with dummy= 1 is:

Coefficient of HFT= -0.0411293, t –value = -3.77, the p-value =0, R-square =0.7239 , with 95% coefficient confident, HFTs will lie between -0.06266 and -0.01959

In the model with only non-trust securities, the result is similar to the completed model which HFT does not have sufficient impact on the stock price

(14)

volatility. And this result also suggests that the general variables that affect the price volatility in the market, has similar impact on the non-financial sector securities. Which is they have negative impact on the stock price volatility ,but the impacts are small.

Nevertheless, regarding to the model of trust securities, there is a complete different result compare to the original model and non-trust model, which it suggests that the interest rate, inflation rate, past 12 months return, HFT , inverse price level and the amount of institutional holding have negative impact on the stock price volatility, whereas the rest of factors contribute positively. In details , HFT has a large P value with 0.735 which the null hypothesis will not be rejected in a 95% confident level and the coefficient is -0.015 which also has a larger coefficient compare to other models. Even though the observation size for the model are quite small, the R-square = 0.8741 suggests that the independent variable (volatility) can be largely explainable by dependent variable, it also has the highest R-square value which means that it is a fitter model compare to other restriction models.

I then run up a descriptive data for this model, it summarizes that:

Variable | Obs Mean Std. Dev. Min Max ---+---

Volatility | 20 .0223973 .0195401 .0060242 .0648498 HFT | 20 28.45862 54.60217 -.01739 224.6022 To | 20 28.93759 54.54748 .5602172 225.0373

Compare to the descriptive data in section 1 , it suggests that there are more HFT activities involved into the financial sectors in average , the mean value is higher, but the maximum value and stat deviation are lower than the original model which also suggests that HFTs’ activities did not volatile a lot in each securities and in fact they might have similar strategies to deal with securities from financial sector.

Therefore, the study shows that HFT trading has involved more investment activities into financial sectors’ securities, but there is no sufficient evidence shows that they decreases or increases the stock prices changes.

(15)

5. Robustness checks

From the suggestions of Hireschey and Nicholas(2011), the HFTs are tend to have positive relationship with institutional trading volume(Volinst), which is the HFT has similar investing behavior as institutional investors. Therefore, with more investment from institutional investors, the amounts of HFTs are also higher. With more institutional holdings, the amount of HFT might be less. Hence, there might be problems of multicollinearity exists. Therefore, a robustness check will be will be implemented between HFT and institutional holding as well as the volume of institutional trading.

First of all, a regression between HFT and institutional holding/institutional trading is analyzed in stata, results shows that p-value for institutional holding is 0.108, and for the volume of institutional trading, p-value= 0.183. Where it shows that there are no significant evidences to show that HFT and institutional investors’ behavior has relationship in the sample. (detail table in appendix)

Moreover, a robust check will be run in the stata as well. With the robust check in the stata the results are mostly similar to the models that mentioned above. The HFTs has coefficient with -0.0380479 which is similar to the competed model mentioned above. And the P-value is = 0.041 therefore with 95% confident interval the null hypothesis should be reject, there is a negative impact of HFT activities on stock price volatility. For the rest of factors are mostly have different p-value but the conclusion are indifferent. One exception is the data of institutional trading. In the original model it suggest that p-value is equal to 0.073 whereas in the robustness model the results is 0.27, therefore in the robustness test the null hypothesis should be rejected and the institutional trading does not have effect on stock price volatility. It suggests that there is a robustness error in the “Volinst” (volume of institutional trading) factor. (detail table in appendix)

6. Conclusion

With the developing techniques on trading strategy and the development on financial engineering, the High frequency trading activities are involved increasingly in the stock market. With certain strategy they usually have very short time period of holding and stocks and high turnover rate with large amount quantity of stocks in the markets. Some investors blame that they push up the price level and volatile the market. One of example is the flash crash on 2010. Market index are increased and decrease rapidly and far over than the market expected, evidences shows that HFTs have involved a lot of activities during that time , hence study on whether HFTs affect the price volatility are subsequently put forth. Scholars such as Avramovic(2010, 2012), Bershova and Rakhlin(2013) ect agrees that HFT declined stock price volatility whereas some other

(16)

researches like a study from Jarnecic and Snaoe (2010) insisted that HFT decreases volatility in long-term but increased it in short-term

In this paper, based on the study results from pervious researches, HFT was empirically estimated in selected 30 stocks by using data from institutional investors and individual investors, which the data with mean average HFT volume involved into trading is about 17.45 times of share outstand of the security , with maximum value of 812.41. In the further analysis, I found out that in the completed model there is sufficient evidence to show that HFT has negative impact on the price volatility , but the coefficient is too small(nearly to 0), the R-square value is 72.96 which it indicates that most part of volatility can be explained by the factors in dependent part.

A similar outcome was found to reject null hypothesis in an additional model with dummy variable of non-trust securities.

More importantly, in the latter model which was restricted to model that only include trust securities, study shows that the HFT no correlation with volatility ,which the P- value is large enough to do reject null hypothesis. Therefore in this part the result suggests that HFT does not have effects on stocks.

However, there are still some limitations on this study:

First of all, the time series data are not continuous, due to the fact that Thomson One only provides quarterly data, which it turns that only monthly data are covered in the model. Also ,due to the size of stata’s limit. There are only 30 stocks are selected , the result will be more accurate if sample size is larger Secondly, the time period is too short. There are only data from 2014 and 2015 are calculated and analyzed, however HFT has been introduced more than 15 years therefore the result only can conclude year of 2014 and 2015 but not the years before.

Thirdly , due to the fact that HFT cannot be observed, some assumptions are applied in order to calculate the amount of HFT. To the extent that the calculation of HFT suffers from measurement error, some estimated regression coefficients could be biased.

Regarding to the conclusion for last model, the HFT did impact on stock price volatility regarding to different sectors, sample size for financial industry is too small, hence it may not represent for the whole situation of the industry, thus, I leave the question for further research.

(17)

Appendix 1, list of 30 selected shares MICROSOFT CORP

GENERAL ELECTRIC CO

DIREXION SHARES E T F TRUST PROSHARES TRUST II

FACEBOOK INC SPRINT CORP NEW APPLE INC

PFIZER INC

FRONTIER COMMUNICATIONS CORP

Over FORD MOTOR CO DEL RITE AID CORP

MICRON TECHNOLOGY INC INTEL CORP

BANK OF AMERICA CORP A T & T INC

CISCO SYSTEMS INC SIRIUS X M HOLDINGS INC FREEPORT MCMORAN INC YAHOO INC SPDR S & P 500 E T F TRUST SECTOR SPDR TRUST PETROLEO BRASILEIRO SA PETROBRAS VALE S A NETFLIX INC ISHARES TRUST

VANGUARD INTL EQUITY INDEX FUNDS

UNITED STATES OIL FUND L P MARKET VECTORS E T F TRUST POWERSHARES QQQ TRUST COCA COLA CO

(18)
(19)
(20)
(21)
(22)
(23)

Reference

Bollen, N. P., & Whaley, R. E. (2015). Futures Market Volatility: What Has Changed?. Journal of Futures Markets, 35(5), 426-454.

Brogaard, J., Hendershott, T., & Riordan, R. (2013). High frequency trading and price discovery.

Brogaard J, Garriott C, Pomeranets A. High-frequency trading competition[J]. 2014. Chaboud, A. P., Chiquoine, B., Hjalmarsson, E., & Vega, C. (2014). Rise of the

machines: Algorithmic trading in the foreign exchange market. The Journal of Finance, 69(5), 2045-2084.

Cheung Y. and L. Ng. 1992. Stock price dynamics and firm size: An empirical Investigation. The Journal of Finance 47 (5), 1985-1997.

Cumming, D., Zhan, F., & Aitken, M. (2012). High frequency trading and end-of-day manipulation. Available at SSRN 2145565.

Dr.John Bates.“Algorithmic Trading and High Frequency Trading, Experiences from the Market and Thought on Regulatory Requirements”

Domowitz, I., Glen, J., & Madhavan, A. (2001). Liquidity, volatility and equity trading c osts across countries and over time. International Finance, 4(2), 221-255.

Edwards, F. R. (1988). Does futures trading increase stock market volatility?. Financial Analysts Journal, 44(1), 63-69.

Hendershott, T., C. Jones, and A. Menkveld. 2010. Does algorithmic trading improve liquidity? Journal of Finance, forthcoming.

Hendershott, T., & Riordan, R. (2009). Algorithmic trading and information.Manuscript, University of California, Berkeley.

Hasbrouck, J., & Saar, G. (2013). Low-latency trading. Journal of Financial Markets, 16(4), 646-679.

Jones, C. M. (2013). What do we know about high-frequency trading?.Columbia Business School Research Paper, (13-11).

Jarnecic, E., & Snape, M. (2010, June). An analysis of trades by high frequency participants on the London Stock Exchange. In 17th Annual Conference of the

(24)

Multinational Finance Society MFS, Barcelona, Spain(Vol. 30).

Kirilenko, A. A., Kyle, A. S., Samadi, M., & Tuzun, T. (2014). The flash crash: The impact of high frequency trading on an electronic market.Available at SSRN 1686004. Zhang, F. (2010). The effect of high-frequency trading on stock volatility and price

discovery. SSRN eLibrary.

Menkveld, A. J. (2013). High frequency trading and the new market makers.Journal of Financial Markets, 16(4), 712-740.

Dische, A. (2002). Dispersion in analyst forecasts and the profitability of earnings momentum strategies. European Financial Management, 8(2), 211-228.

Referenties

GERELATEERDE DOCUMENTEN

Only the peak viral load-related parameters have been displayed for clarity, but the model also included the significantly associated parameters from the univariable

Auch eine Anwendung zu erzwingen, die Graphenstruktur des Werkzeugs zu verwenden, hilft nicht viel, da die Graphenstruktur des Werkzeugs nicht ausdrucksfähig genug für die

wie alle civiele functies waren overgedragen werden door de inlichtingendiensten gewezen op het feit dat de komst van de Nederlanders gevaar zou

When the American doctrine is applied to the case of InnovaThor v Generix, InnovaThor’s Swiss-claim could be considered a patented medical treatment method. This method can be

We implemented an algorithm based on a machine-learning approach to learn the level of engagement of the audience in such a way that it can be possible to measure

However, we argue that the work of clinical psychologists can improve the economic sustainability of chronic pain management in all the stages of the care, from the assessment phase

Utilization of opportunistic BLE network for animal mobility pattern RSSI/PathLoss Model Estimated Euclidean Distance Matrix Estimated Adjacency Matrix Critical Distance

Subject to section 86(9) and (10), a credit provider who receives notice of court pro- ceedings contemplated in section 83 or 85, or notice in terms of section 86(4)(b)(i), may