How do quarterly earnings announcements affect the prices of stocks of Silicon Valley companies? : a comparison of Silicon Valley growth and value companies

(1)

How do quarterly earnings announcements affect the

prices of stocks of Silicon Valley companies?

A comparison of Silicon Valley growth and value companies.

Author: B De Lima De Mello Student Number: 11148691

Thesis Supervisor: Dr. Jan Lemmen Finish Date: 31/01/2018

(2)

ABSTRACT

This study analyses the presence and consistency of Abnormal(AR) and Cumulative Abnormal Returns(CAR) in the Silicon Valley on the days surrounding quarterly earnings announcements, taking into consideration the biases present in investors reaction to growth and value stocks. I first analyze the significance of the aforementioned variables using a cross sectional t-test to find that both variables are consistently present in the periods analyzed. Afterwards, I make a comparison between AR and CAR of growth and value stocks, which yields a result that corroborates with the literature so far: abnormal returns are more present in value stocks than in growth stocks. However, a different reasoning is given for the causes of this issue: not only the bias already known is a cause for this difference, but also informative and expectational shocks.

Keywords:_{capital markets; earnings announcements; information content}

This document is written by Student Brunno De Lima De Mello who declares to take full responsibility for the contents of this document. I declare that the text and the work presented in this document are 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.

(3)

1. Introduction

Since its creation, the Efficient Market Hypothesis (EMH) has passed under great scrutiny. For many economists it seems absurd or not real, to assume that all available news and information is readily reflected into the pricing of stocks. Even in a greatly connected world like ours, where new information is as far as a click in a touch screen, it doesn’t seem that markets always operate efficiently like the EMH poses.

In an efficient market as Fama (1969) constructs, inefficiencies can be seen through the presence of abnormal returns in stock prices: if efficient markets don’t reflect all available information, investors who own privileged information may benefit leading to abnormal returns. The asymmetric information problem is not the only one pointed as the cause of market inefficiencies. Studies, such as the one from Lakonishok, Shleifer and Vishny (LSV) (1994), have shown that an agency problem might exist as well, since investors don’t always make rational decisions when investing.

Lakonishok et al. (1994) pose that investors’ suboptimal response can be seen through the different response investors have to earnings announcements disappointments towards growth and value stocks. Both defined in relation to the Book to Market (BTM) ratio: the first refers to low BTM stocks (growth stocks), and the latter to high ones (value stocks). Their findings show that investors constantly overstate future growth expectations of growth stocks by using past information to project future price movements. A later study by Donnelly (2014) showed that some part of optimism is embedded in the price valuation of growth stocks, which could explain its overvaluation when compared to value stocks.

This paper will examine past studies on earnings announcements effects on growth and value stocks, trying to add a different perspective by analyzing the effect of earnings announcements on the stock prices of Silicon Valley companies. The area, which is defined as a Global Cluster of Innovation, has unique characteristics which can aggravate the problems causing market inefficiencies, more specifically the biases and irrational decision making processes found in other studies (Skinner and Sloan, 2002; LSV, 1994), which will be discussed further on this paper.

(5)

The Silicon Valley (SV) has had the attention of investors for a long time now. Besides being home for big companies such as Apple, Alphabet and HP, it’s also the focus of emergence of many new cutting edge tech companies. The finance of startups in the region skyrocketed 57% in 2017 compared to 2016: more than 12.1 billion dollars were invested in the area only through venture capital (PwC, 2018). Engel (2015) defined some key aspects and behaviours that make the area so unique: heightened mobility of resources, increased velocity of business development and incentives and goals that lead to an affinity for collaboration. Each of these traits are somehow responsible for the numerous presence of growth stocks in the Silicon Valley.

Thus, by analyzing the effect of earnings announcements on stock prices of Silicon Valley companies, I expect to answer the three following questions: (1) Is there a presence of abnormal returns on SV stock prices after earnings announcements? (2) If so, are they consistent throughout the years? 3) And lastly, is there a difference between value and growth firms in relation to the market value correction after quarterly earnings announcements?

Section 3 of this paper will make a review on the existing literature on the Efficient Market Hypothesis and on the response of investors to earnings announcements. More specifically on the difference between their response to growth and value stocks. Additionally, evidence will be shown as to why Silicon Valley companies have characteristics which make them unique for this study.

In Section 4 I will conduct an event study comparing the prices of stocks from SV companies before and after earnings announcements. If the presence of abnormal returns is confirmed, I will examine the consistency of such returns throughout 3 different years. Ultimately, in Section 5 I will discuss my test results, give my conclusions and suggest further research on the topic.

(6)

2. Literature Review

The formal definition of the Efficient Market Hypothesis defines an efficient capital market as one where all available news and information are reflected in the prices. Fama _{(1965) first} theorized this by testing the predictability of price movements using two models: the chartist and the random walk theory. The chartist theory, or naive expectations model, holds that future price movements can be predicted using information on the past behaviour of a security. The random walk theory, on the other hand, constructs that price movements are as predictable as choosing a number randomly. That is, differently than on the first model, successive price changes are independent, identically distributed random variables.

The independence of prices changes is one of the two foundations of the random walk theory: it also assumes that price changes follow a certain probability distribution. This paper will focus on the independence assumption, as it directly clashes with the possibility of abnormal returns around earnings announcement dates. Fama (1965) argues that even if noise generating processes and successive new information are dependent, market mechanisms will produce independency in price changes for an individual stock. Sophisticated investors, those who consistently outperform the market, would exhaust any dependencies by identifying the investment opportunity they create, thus eliminating any possibility of consistent abnormal returns.

Although the market mechanisms assumption seems to work in theory, Fama (1965) points out that the foundation where the assumption is laid are rather unreal. He points out two criticisms: (1) there is no reason to think that individual’s intrinsic valuation of a stock is independent and (2) there is no reason to believe that successive information is independent. Another criticism towards his model is the underestimation of the psychological behaviour of investors. Recent studies (LSV, 1994; Ball and Brown, 1968) have shown that investors behave suboptimally, and if that is true, market mechanisms shouldn’t eliminate independency.

On a later revision of his work and other papers on the subject of Efficient Market Hypothesis, Fama (1991) poses that the proposition of a market where security prices reflect all available information is too strong. Therefore, using the existing literature, the defines three tests of efficiency:

(7)

● Weak form: How well can past returns predict future returns?

● Semi-strong form tests: How quickly do security prices reflect public information announcements?

● Strong form tests: Do any investors have private information that is not fully reflected in market prices?

This paper will focus on the semi-strong tests, or as Fama (1991) classified, even studies.

He points out that this is the least encumbered test affected by the joint hypothesis problem, which arises from the testing of market efficiencies along with the asset pricing model used to estimate expected returns. Still in this revision, Fama analyzes previous tests conducted by other scholars. Although he argues in favour of event studies as a proof of efficiency, some of the findings of these studies are rather contradictory to his theory (Bernard and Thomas, 1990; Ball and Brown, 1968). More specifically, the findings on post-announcement drifts and abnormal returns on the dates surrounding earnings announcements can be seen as proof of market inefficiencies.

The effect of earnings announcement on stock prices has been the focus of many studies (Donelly, 2014; Bernard and Thomas, 1990; Ball and Brown, 1968; Fama, 1965), because of the importance of the variable as a fundamental pivot of valuation and the presence of abnormal returns on the dates surrounding public announcements (Fama, 1965). Ball and Brown (1968) analyzed the effects of earnings announcements on stock prices. They conducted 3 OLS regressions which, in their opinion, represent the difference between the actual change in income and its conditional expectation. The latter can be defined as the part of a company’s income which is somehow correlated to the past income of all firms, as half of the variability of the average Earnings per Share is due to economy-wide effects (Ball and Brown, 1968), added with the present income of all firms. All regressions have the same format, and the error term is seen as the impact of the new information conveyed in the present income for a specific firm. The forecast error is constructed by subtracting change in actual income by change in expected income, both of which defined by OLS regressions on their paper. The results are striking, as they find that when the expected income differs from the actual one, the market tends to move the same way, i.e. the sign of the income forecast errors can be used to predict the sign of the rate of return prior to the annual earnings announcements. Even though they do signal that the drift towards the price changes that occur after the earnings announcements start even 12 months prior to the actual announcement, their result can still be seen as a proof of market inefficiency, after all, past

(8)

information shouldn’t be able to predict future price movements if they are completely random and independent. Fama (1965) argues that this is actually evidence of the price adjusting to new information as it is conveyed to the market, but if constant returns are made possible by exploiting the flow of new information in the market constantly, then the market mechanism mentioned by him wouldn’t generate independency, as intrinsic value investors don’t seem to be exhausting the causes of dependency.

Bernard and Thomas (1990) also make a great case for failures in the flow of information to prices. First, they assume that stock prices are partially affected by naive expectations, i.e. prices don’t reflect the extent to which time-series behaviour of earnings is different from naive expectations. As previously discussed, naive expectations for earnings announcements comprises of using information on earnings from year t - 1 to predict earnings in year t. With this, they open the possibility to the correlation of forecast errors, something that a market where all past information is impounded in prices couldn’t have. By using the naive model to predict the reaction to future earnings announcements, they were able to predict the three-day reaction to future earnings announcements up to 4 quarters ahead. The logic behind their study supports that the market passes on information to prices based on their reaction to information. Therefore, market inefficiency is founded in a failure of information to flow completely into prices.

The informational value of earnings announcements has long been the subject of study, and it is represented as ways for management to convey information to shareholders (Ball and Brown, 1968). By the definition of information (Beaver, 1968) as a change in the expectation regarding the outcome of an event, we can classify the annual income report as informational in content if it leads to a revision on future returns, resulting in a change in current price market equilibrium. Beaver (1968) findings, along with Mlonzi, Kruger & Nthoesane (2011) show that there is a significant volume and price reaction to earnings announcements. Thus, the informational value of such variable is irrefutable. However, the question that remains is if investors react rationally, i.e. as the EMH would expect, upon such informational event.

Findings on suboptimal investor behaviour have shown biases towards different types of stocks (Skinner and Sloan, 2002; LSV, 1994), namely, investors’ different responses to earnings announcement of value and growth stocks. The variables are classified under the dimension of their Book-to-Market (BTM) ratio: the first having low and the later having high

(9)

BTM. As a rule of thumb, high BTM stocks are the ones which have the ratio bigger than one, and, conversely, low BTM stocks are the ones which show a ratio smaller than one. By separating common stocks in these two categories, both studies were able to find that investors have overly optimistic expectations about the future earnings of growth stocks, which results in lower returns when these expectations aren’t met.

The overly optimistic expectations can be explained by the use of the previously discussed naive investment strategies: investors project the future prospects of a company based on its past performance and overreact to good or bad news. To test for differences in investors’ reaction, LSV (1994) analyzed the performance of value strategies in relation to glamour strategies, i.e. companies with high past prospects, and found that the low BTM stocks have an average annual return of 9.3 percent and the high BTM stocks have an average annual return of 19.8 percent (LSV, 1994, p. 1547). This happens, they infer, because the value strategy exploit errors in expectations regarding future growth of glamour stock which are embedded into the price.

Once the assumption that investors react differently to stocks classified under the dimension of the BTM ratio, a question arises as to whether companies with consistently low BTM present stronger biases. The reasons to why companies have low BTM are briefly discussed by LSV (1994): companies could have a low BTM because they possess a lot of intangible assets and growth opportunities or the company is actually an overvalued glamour stock, defined previously as a stock which performed well in the past. The reasons as to why companies have a low BTM ratio can all be found under the defining characteristics of Silicon Valley companies: (1) the presence of intangible assets in the area is irrefutable, especially in regards to intellectual property. More than 14618 patents were registered in the area on 2015 alone (_USPTO, 2015). (2) the presence of new businesses and startups is characteristic of the area (Engel, 2015). (3) many companies in the region are considered overvalued glamour stocks, such as Twitter.

Making the case for the presence of low BTM ratios in the SV area seems logical, however, the implications of such connection might have interesting empirical results. Along with the aforementioned traits, Engel (2015) defines the SV as a Global Cluster of Innovation: areas with a high density of interconnected companies and institutions which work in the same industry or field. Its distinctive characteristics regarding nature and behaviour are composed by: rapid emergence of new firms commercializing new technologies, creating new markets

(10)

and addressing global markets. Besides other definitions which I won’t analyze further on this paper, Engel (2015) defines key behaviours and components, specific to the area of the SV. Each contributing to the hypothesis that Silicon Valley companies are a strong case of low BTM ratio firms:

● Presence of Venture Capital Investors.

● High number of Research and Development centers. ● Heightened mobility of resources.

● Increased velocity of business development.

● Incentives and goals that lead to an affinity for collaboration.

The presence of Venture Capital Investors in the Valley is now characteristic, as 51% of startups say their most likely next source of funds is venture capital (PwC, 2018). I construct that the venture capital presence in the Silicon Valley creates an agency problem: Engel (2015) states that deal structures between investors and managers from startups encourage big wins and quick exits for the first, not the latter. If the main goal of the investor is to get a short term win, when selling the company its goal will be to overvalue it as much as possible, considering that after the deal the investor would follow the “Silicon Valley” model (Engel, 2015), and look to invest in a different company. Managers also seem to be under the possibility of agency problem since the job structure of the Silicon Valley is really job specific.

They also have incentives to get short term big wins and quick exits. Engel (2015) poses that an effective management career in the SV entails changing startups every 3-5 years to build a portfolio of capital return opportunities, which will be gained by selling the company. Therefore, executives in charge of these companies have as much incentives to overvalue their price as investors do. Once the contract between venture capital investor and company ends, it normally is sold to a different company or put into an initial public offering. Thus, if this overvaluation bias is true, it should be noticeable through stock prices movements after earnings announcements.

A proof of the investment irrationality present in the area comes in a study by Randall and Enzo (2017): they analyzed investors’ expectations from the market of two different regions with similar characteristics, the Silicon Valley and Copenhagen. Investors were asked to give grades for each region in five different aspects: economic activity; depth of capital markets; entrepreneurial culture and deals; human and social environment; and investor protection

(11)

and taxation. By analyzing their results through a mean analysis they found only the latter to be significantly different in the regions. Investors regarded the Silicon Valley better for protection than Copenhagen, and Stockholm better for taxation than the Silicon Valley. When taken into consideration this finding, the Silicon Valley was found to be 8.39% more attractive than Copenhagen, which seems rather irrational, as the area received 7.3 times more investment volume and closed 6.2 more deals per capita than Stockholm. Randall et al. (2017) argue that a heard effect, similar to the one found in the 1990s which lead to the bubble burst in 2001, is present in the area at the moment.

Combining the reviewed literature, it’s possible to infer that a strong overvaluation bias should be present in the Silicon Valley area. Because of multiple biases caused by the specific characteristics of the region, stock prices from these companies should present consistent abnormal returns surrounding earnings announcement dates. Thus, we arrive to the following hypothesis for the first research question:

H₀ : Abnormal Returns = 0 H₁: Abnormal Returns ≠0

H₀ : Cumulative Abnormal Returns = 0 H₁: Cumulative Abnormal Returns ≠0

For the second research question, the consistency of the results of the tests mentioned above will be taken into consideration when comparing them throughout 3 different years. If abnormal returns and cumulative abnormal returns are consistent, I expected to see tests results which reject the null hypothesis in at least 2 of the 3 years in question. The same rule applies for the periods surrounding the event date: only the ones most frequently present in the quarter analysis will be considered consistent.

Lastly, I will make a comparison between the AR and CAR of 30 different companies which showed the most extreme, i.e. lowest and highest BTM values. It is my assumption, considering the literature reviewed, that companies with high BTM ratio will show stronger biases than companies with low BTM ratios. This should be evident by making a t-test of paired mean comparison, where the null and alternative hypothesis are, respectively:

H₀ : _μ_d= 0

(12)

3. Methodology

The analysis on this paper has the goal of examining whether investors react irrationally during earnings announcements periods, and if so, how consistent is this reaction throughout different years.

The null and alternative hypothesis are formulated in regards to the Abnormal Returns(AR) and Cumulative Abnormal Returns(CAR), which are defined as:

R R E(R )

A _i,t = _i,t − _i,t

AR R

C _i,t = ∑ t=0

A _i,t

The first is defined by actual returns, minus the expected returns of firm _iin period _t, whereas the latter is constructed by adding all AR of firm _i for a given period.

The data used as input in these formulas was gathered from the Eventus tool. 50 Silicon Valley companies were chosen at random from the SV150 list, which ranks 150 companies in the region based on their sales. The companies are also publicly traded in the NASDAQ. After gathering data on the quarterly earnings announcements dates for each company in 2015, 2016 and 2017, the U.S. Daily Event Study Tool was used to test my hypothesis.

The tool is present in the Wharton software, and it calculates specific test statistics for event studies which comprise more than one company (cross sectional tests). Since the goal of this paper is to examine the presence of abnormal returns in the whole area of the SV, a cross sectional test will provide results which are more fit to answer this question. The assumptions made by this test also fit my data: by the Central Limit Theorem(CLT) the data set can be considered approximately normally distributed, as the number of companies exceeds 30; the companies were chosen at random, thus random sampling is respected.

Finally, the abnormal and cumulative abnormal returns have a continuous distribution. Since all assumptions are respected, the cross section_t-test was conducted, with different degrees of freedom for each year: 2015(_{df = 37)} , 2016_{(df=47) and 2017(df=46)}. The number of observations changed through the years analyzed when the data was plugged into the Eventus tool, since the program excludes companies which have missing data for any of the

(13)

days surrounding the event. However, the number of observations still respects the Central Limit Theorem (CLT) rule, so it can be considered approximately normally distributed.

The event date for each company (_t=0) was defined as the earnings announcement date for the quarter being analyzed. Abnormal returns were then calculated from 10 days before (_{t=-10) up to 10 days after (t=10) the announcement. Additionally, the estimation period,} used to calculate the expected return of a security, was set to 100 days prior to the beginning of the event (_t=-110).

In order to estimate the expected return of the designated security, the Capital Asset Pricing Model (CAPM) was used:

(R ) Rf β Rm Rf ) E _i,t = _t + _i_{* (} _t − _t

Where _Rf is the risk free rate of the market in question; β is a risk specific variable for firm _i which tells how sensitive it’s to market shocks; and _Rm is the market return for a given period _t.

Once the expected return for each period was calculated, the actual return for the same period was also calculated by using the following formula:

1 R_i,t = Pi, t

P_{i, t+1} −

Using the information on expected and actual returns, it was then possible to calculate the AR and CAR for each company in each quarter analyzed. After doing so, two cross section test statistics were applied to both variables separately. One using the standardized and the other using the regular or simple cross sectional test. Both tests were used in an attempt to control for robustness, as Boehmer, Musumeci and Poulsen (1991) construct that standardized tests are robust to event-induced variance increases of stock returns. The formulas are, respectively:

1. Standardized tests:

a. Abnormal Returns: z_t = ASARt , N(0, 1). S

√N * _ASARt

i. ASAR_t is the sum of the standardized abnormal returns (abnormal return divided by its standard deviation) of each firm in the given period; _N is the number of firms in the portfolio, which is 51 in this case; and _S_ASARis the standard deviation of the _ASAR_t_.

(14)

i. SCAR is the average standardized cumulative abnormal return for all firms in period _t; _Nis still 51 in this case and it represents the number of firms analyzed; and _S_SCARis the standard deviation of _SCAR. 2. Regular tests:

a. Abnormal Returns: t_AAR , N(0, S_AAR). t = √N *

AARt S_AARt

i. AAR_t is the average abnormal return, calculated by adding the AR for all firms in period _{t and dividing it by the number of firms; S} _AAR is the standard deviation of _AAR_t.

b. Cumulative Abnormal Returns: t_AAR , N(0, S_AAR). t = √N *

CAAR SCAAR

i. CAAR_t is the cumulative average abnormal return in period _t for all firms analyzed; _S_CAAR is the standard deviation of the _CAAR_t.

The results of the test statistics were then analyzed by comparing the quarters of the years being analyzed, i.e., 1st quarter of 2015 was compared to the 1st quarter of the other years and so on. The cross sectional test statistic for each quarter was calculated by combining the abnormal returns for all companies analyzed in the same period relative to the event. For instance, the cross sectional test statistics for abnormal returns in period -10 in the first quarter of 2015 will represent whether there is a presence of significant cross sectional AR in that period. For the cumulative abnormal returns, the cross sectional test statistics is calculated at the end of the event for all companies analyzed in that quarter, i.e., in the example above, the test would show if there are significant CAR in the first quarter of 2015. By separating the data this way I could better examine the patterns of consistent AR and CAR throughout the years in consideration.

In addition, to answer my last research question, I conducted a paired mean comparison t-test with the STATA software. The CAR and AR in the days surrounding the event were calculated for both the 15 highest and 15 lowest BTM firms. Then the periods were paired according to their quarter and year, and the t-test was conducted, using the following formula: t = _SE(μd)μd , with _SE(_μ_d) = Sd_√n ~ N(0,1) with _{n - 1}degrees of freedom.

(15)

4. Data Analysis and Results

The 21 days ( _{t = -10 until t = 10} ) window where all abnormal and cumulative returns for each company were calculated is going to be the focus of this analysis. The table below shows the descriptive statistics of the AR for all the quarters of years 2015, 2016 and 2017, respectively:

Table 1.0 - Descriptive Statistics Quarterly Mean Abnormal Returns.

Mean Abnormal Return

Q1 Q2 Q3 Q4 Event Time 2015 2016 2017 2015 2016 2017 2015 2016 2015 2016 -10 0.0022 0.0022 0.0042 0.0020 -0.2424 0.0035 -0.0010 -0.0002 0.0038 0.0003 -9 0,0029* 0.0029 0.0025 0.0005 0.3386 -0.0002 0.0005 -0.0028 0.0034 -0.0005 -8 0.0034 0.0034 -0.0004 0.0013 0.4797 -0.0033 0.0020 0.0051 0.0075 -0.0012 -7 0,0016* 0.0017 -0.0024 -0.0042 -0.2877 0.0057** -0.0029 0.0014 0.0013 -0.0016 -6 0.0000 0.0000 -0.0006 0.0039 -0.3102 0.0022 -0.0038* 0.0002 -0.0010 -0.0063* -5 -0.0055 -0,0055* 0.0030 0.0011 -0.0048 0.0034 0.0004 0.0023 -0.0068* -0.0003 -4 -0,0001** -0.0001 0.0026 -0.0063* 0.3464 -0.0029 -0.0024 0.0034 0.0008 0.0046 -3 -0.0006 -0.0006 -0,0045* 0.0008 -0.7320 -0.0008 -0.0064* 0.0013 -0.0021 -0.0053 -2 0.0083 0,0083** 0,0062** -0.0090** -0.0944 0.0007 -0.0011 0.0044 -0.0032 0.0004 -1 -0.0039 -0.0039 -0.0010 -0.0035 -0.0546 -0.0016 -0.0065 -0.0011 -0.0019** -0.0009 0 0.0081 0.0081 0,0057* 0.0046 -0.3975 -0.0041* 0.0013 -0.0019 0.0013 0.0047 1 -0.0062 -0,0061* -0.0037 -0.0059* -0.3880 -0.0064* -0.0023 -0.0034 -0.0122 0.0001 2 -0.0009 -0.0009 -0.0029 -0.0006 -0.4775 -0.0025 0.0028 -0.0006 -0.0095 0.0008 3 0,0010* 0.0010 -0.0004 0.0038 -0.3404 0.0025 0.0040 -0.0007 0.0069 -0.0022 4 0.0017 0.0017 0,0060* -0.0072 0.1523 0.0053** 0.0056 -0.0007 -0.0165** 0.0030 5 -0.0006 -0.0006 0.0050 0.0034 0.0001 -0.0011 0.0060 0.0022 -0.0123** 0.0061* 6 -0,0010* -0.0010 0.0019 -0.0026 0.0371 0.0077** -0.0055 0.0027 -0.0005 0.0022 7 -0,0004* -0.0004 0.0075 -0.0026 0.0809 0.0022 0.0027 -0.0015 -0.0021 -0.0032 8 -0.0083 -0,0082** 0.0021 -0.0057 -0.3627 -0.0039 -0.0044 -0.0051 -0.0027 0.0015 9 0.0104 0,0104* 0.0017 -0.0030 -0.0048 -0.0033 -0.0028 0.0082 0.0059 -0.0074** 10 0.0042 0.0042 0.0021 -0.0019 0.0361 0.0027 0.0036 -0.0036 -0.0010 -0.0040

Note: Results with * are rejected at 5%, and ** at 1% significance level.

(16)

Significant AR were found in almost every quarter analyzed. Only quarters 2 and 3 of 2016 showed no significant Abnormal Returns, thus there is strong statistical evidence to support that AR are present in companies situated in the SV. Other interesting facts can be drawn from the table: most AR are present in the 8 days window surrounding the event, supporting evidence found by Fama(1991) that the days closest to the event would show AR in a semi-strong efficient market, as a sign of adjustment of the market to the new information conveyed in the event. Another evidence of semi strong efficiency that corroborates previous findings (Bernard and Thomas, 1990) can be seen on the second day prior to the event, as it showed most significant AR when compared to other days. Anticipating information that will be conveyed in future events is a sign of a perfectly efficient market, as Fama(1964) argues that information is input in real time on market prices.

The graph below shows the distribution of mean AR for the periods analyzed:

Graph 1.0

The distribution of the Mean AR follows an approximate normal distribution, as the histogram shows a bell shaped curve, with skinny tails and very dense concentration surrounding 0. This ratifies the CLT, as the number of observations is higher than 30, but it also shows evidence to support Fama’s arguments that AR are close to, to expected to be 0.

(17)

In order to control for possible variances caused by the event itself, a standardized cross sectional t-test was conducted. The test robust results give important insight on the origins of the AR, and three separate cases can be highlighted: (1) When the ARs were significant only under the regular t-testing, for example in period _{t = -6} of the third quarter of 2015; (2) When ARs were significant only under standardized test, for instance in period _t= ₅ of the same quarter and year as the first case; (3) And lastly, when ARs were significant under both tests.The results of this testing can be seen in the Appendix section, on Table 2.0

Respectively, in the first case mentioned above, the difference in test results could be proof that the AR was actually significant because of expectational shocks caused by the event itself. For instance, if profits were lower or higher than expected, investors would react more intensely, i.e. overreaction bias, thus opening the opportunity for abnormal gains. That could be why under the standardized cross sectional test the hypothesis is not rejected. For the second case, it could be that the same happened as in the previous case, but now the variance change caused by the event itself could have actually balanced out the Abnormal Returns, making them significant only when considering that factor into the t-testing. Lastly, the third case should be a pure situation a market opportunity that outgains the market. This would be the most pure form of market inefficiency, as it would be present neither because of expectational shocks nor biases. On the other hand, following the time adjustment of prices argument given by Fama(1991) and mentioned above, this last case could also be evidence of new information being plugged into market prices, showing actually that the market is leaning towards efficiency, and to that, some periods of inefficiency are needed.

To answer the second research question, Cumulative Abnormal Returns were also analyzed for the periods in question, and a summary of the statistics of the variable can also be found on the Appendix section, as Table 3.0. Differently than on the case of AR, all test results were consistent throughout the analysis, i.e., both standardized and regular t-test yielded the same results. Moreover, only the first and last quarters of the years of 2015 and 2017 presented CARs, which could be evidence of the importance of such months, given that the last quarterly earnings announcements show the overall performance of a company in a given year, and the first show the possible profits prospects for the rest of the year in question. This would go inline with the argument given above that when both tests show significant abnormal returns, there is a market opportunity caused by the adjustment of prices to new information.Therefore, it’s possible to conclude that CARs are consistently

(18)

present in such quarters, corroborating with the assumption that the Silicon Valley area does have consistent AR and CAR.

Lastly, the results for the paired mean comparison analysis can be found on the table below:

Table 4.0 - Descriptive Statistics Mean Difference

Mean Difference between High and Low BTM

Abnormal Returns Cumulative Abnormal Returns

2015 2016 2017 2015 2016 2017

Q1 -0.0028* 0.0006 0.0004 0.0271** -0.0008 -0.0011

Q2 0.0006 0.0010 0.0012 -0.0094** 0.0164** -0.0125**

Q3 0.0004 -0.0024 - -0.0023 -0.0259** -

Q4 -0.0017 -0.0015 - -0.0282** -0.0099** -

Note: Results with * are rejected at 5%, and ** at 1% significance level.

Biases towards value stocks were significant at 95% confidence level only on the first quarter of 2015 when considering Abnormal Returns. That result is a contrast when compared to the CARs, which showed a significant bias in almost all quarters at a 99% confidence level. This corroborates the arguments built in the previous section that the bias present in the Silicon Valley includes all companies in the region, and is even stronger for value stocks, as CARs represent the sum of all AR. This could also be seen as evidence of a bias which is strongly present because of the specific characteristics of the region. Therefore, it’s possible to infer that market inefficiencies are indeed present in the area, and most likely they exist because of biases towards the region as a whole, not only because of the different perception investors have in relation to value and growth stocks.

(19)

5. Conclusions and Discussion

The objective of this paper was to analyze whether abnormal returns and cumulative abnormal returns were present in the days surrounding quarterly earnings announcements of Silicon Valley companies, if so, how consistent they were and what would be the difference between growth and value stocks in the dimension of these variables. At first sight, the results show that abnormal returns are indeed present in the 21 day window surrounding the event for most years analyzed. However, the presence of AR could be the result of a market which is adjusting to new information conveyed in the quarterly earnings announcements, and evidence for this argument is shown in the different results yielded from regular and cross sectional t-testing.

Cumulative Abnormal Returns were also shown to be significant and consistent throughout the periods analyzed, even though there seems to be little evidence that their presence is due to biases in the region. The pattern shown by the test results provide an insight similar to the one for the ARs when considering that the quarters which showed most significant CARs are the ones more susceptible to informational shocks, i.e. the first and the last quarter of the year.

Even though the explanations to the first two research questions seem to go along the way of informational shocks and price adjusting periods, the mean comparison analysis between growth and value stocks could show a different perspective of the subject in question. The strong presence of CAR could be evidence of a bias present in the region as a whole. Whether or not the bias towards growth and value stocks exists, it’s a fact that when polling together the ARs of growth and value stocks, a strong difference emerges between the possible abnormal gains of the two categories of stocks.

Therefore it’s possible to infer that not only AR and CAR are consistently present in the Silicon Valley area, but also growth and value stocks show significant diverging behaviour when considering the latter. That could happen due to either informational shocks, or biases which are present in the region and corroborate an already established bias towards value stocks.

(20)

Further research should focus on separating the effects of such shocks to better understand the causality of such difference. Moreover, a more broad spectrum of companies could be considered to have a better idea of the biases present in the region by comparing the AR and CAR of the SV with the ones of other regions/industries. It would also be interesting, considering the volatile and dynamic nature of the companies in the region, to apply the same event study analysis to intra hour changes in market prices to better split the effects of informational and expectational shocks on price changes.

(21)

REFERENCES

Ball, R., & Brown, P. (1968). An empirical evaluation of accounting income numbers. _Journal of Accounting Research,6(2), 159-178.

Beaver, W. H. (1968). The information content of annual earnings announcements. _Journal indent of Accounting Research, Vol.6, 67-92.

Bernard, V. L., & Thomas, J. K. (1990). Evidence that stock prices do not fully reflect the implications of current earnings for future earnings. _{Journal of Accounting and}

Economics, ₁₃(4), 305-340.

Boehmer, E., Masumeci, J., & Poulsen, A. B. (1991). Event-study methodology under conditions of event-induced variance. _{Journal of Financial Economics}, ₃₀(2), 253-272. Cottin, R. & Garry, E. (2017). The Venture Capital behavioral bias and the ecosystem

investment flows. Retrieved January 15, 2018, from

http://www.diva-portal.org/smash/get/diva2:1115351/FULLTEXT01.pdf.

Donnelly, R. (2014). The book-to-market ratio, optimism and valuation. _{Journal of Behavioral} and Experimental Finance, ₄, 14-24.

Engel, J. S. (2015). Global clusters of innovation: Lessons from Silicon Valley. _California Management Review, ₅₇(2), 36-65.

Fama, E.F. (1965). The Behavior of Stock-Market Prices. _{The Journal of Business,}₃₈(1), 34-105.

Fama, E. F. (1991). Efficient Capital Markets: II. _{The Journal of Finance,}₄₆(5), 1575-1617. Lakonishok, J., Shleifer, A., & Vishny, R. W. (1994). Contrarian investment, extrapolation,

(22)

Mlonzi, V.F, Kruger J. & Nthoesane, M.G (2011). Share price reaction to earnings

announcement on the JSE-ALtX : a test for market efficiency. _{Southern African Business} Review, 15(3),142 - 166.

Money Tree Report Q4 2017(pp. 50-53, Rep.). (2018). PwC. Retrieved January 15, 2018,

from

https://www.pwc.com/us/en/moneytree-report/assets/MoneyTree_Report_Q4_2017_FINA L_1_10_18.pdf

Skinner, D. J., & Sloan, R. G. (2002). Earnings surprises, growth expectations, and stock returns or don't let an earnings torpedo sink your portfolio_{. Review of Accounting Studies,} 7(2), 289-312.

Patenting In Technology Classes Breakout by Origin, U.S. Metropolitan and Micropolitan Areas. (n.d.). Retrieved January 15, 2018, from

(23)

Appendix

Table 2.0 - Descriptive Statistics Quarterly Mean Abnormal Returns.

Mean Abnormal Return

Q1 Q2 Q3 Q4 Event Time 2015 2016 2017 2015 2016 2017 2015 2016 2015 2016 -10 0.0022 0.0022 0.0042 0.0020 -0.2424* 0.0035 -0.0010 -0.0002 0.0038 0.0003 -9 0,0029 0.0029 0.0025 0.0005 0.3386 -0.0002 0.0005 -0.0028 0.0034 -0.0005 -8 0.0034 0.0034 -0.0004 0.0013 0.4797 -0.0033 0.0020 0.0051 0.0075 -0.0012 -7 0,0016* 0.0017 -0.0024 -0.0042 -0.2877* 0.0057** -0.0029 0.0014 0.0013 -0.0016 -6 0.0000 0.0000 -0.0006 0.0039 -0.3102 0.0022 -0.0038 0.0002 -0.0010 -0.0063* -5 -0.0055 -0,0055 0.0030 0.0011 -0.0048 0.0034 0.0004 0.0023 -0.0068** -0.0003 -4 -0,0001** -0.0001 0.0026 -0.0063 0.3464 -0.0029 -0.0024 0.0034 0.0008 0.0046 -3 -0.0006 -0.0006 -0,0045 0.0008 -0.7320_{** -0.0008 -0.0064 0.0013 -0.0021 -0.0053*} -2 0.0083 0,0083* 0,0062** -0.0090* -0.0944 0.0007 -0.0011 0.0044 -0.0032 0.0004 -1 -0.0039 -0.0039 -0.0010 -0.0035 -0.0546 -0.0016 -0.0065 -0.0011 -0.0019 -0.0009 0 0.0081 0.0081 0,0057* 0.0046 -0.3975 -0.0041* 0.0013 -0.0019 0.0013 0.0047 1 -0.0062 -0,0061** -0.0037 -0.0059** -0.3880 -0.0064** -0.0023 -0.0034 -0.0122_** 0.0001 2 -0.0009 -0.0009 -0.0029 -0.0006 -0.4775 -0.0025 0.0028 -0.0006 -0.0095 0.0008 3 0,0010* 0.0010 -0.0004 0.0038 -0.3404 0.0025 0.0040 -0.0007 0.0069 -0.0022 4 0.0017 0.0017 0,0060** -0.0072 0.1523 0.0053** 0.0056 -0.0007 -0.0165* 0.0030 5 -0.0006 -0.0006 0.0050 0.0034 0.0001 -0.0011 0.0060* 0.0022 -0.0123** 0.0061* 6 -0,0010* -0.0010 0.0019 -0.0026 0.0371 0.0077** -0.0055 0.0027 -0.0005 0.0022 7 -0,0004 -0.0004 0.0075 -0.0026 0.0809 0.0022 0.0027 -0.0015 -0.0021 -0.0032 8 -0.0083 -0,0082** 0.0021 -0.0057 -0.3627 -0.0039 -0.0044 -0.0051* -0.0027 0.0015 9 0.0104 0,0104* 0.0017 -0.0030 -0.0048 -0.0033 -0.0028 0.0082 0.0059 -0.0074** 10 0.0042 0.0042 0.0021 -0.0019 0.0361 0.0027 0.0036 -0.0036 -0.0010 -0.0040

(24)

Table 3.0 - Descriptive Statistics Quarterly Mean Cumulative Abnormal Returns.

Mean Cumulative Abnormal Return

Q1 Q2 Q3 Q4 Event Time 2015 2016 2017 2015 2016 2017 2015 2016 2015 2016 -10 0.0022* 0.0022 0.0042** 0.0020 -0.0024 0.0035 -0.0010 -0.0002 0.0038** 0.0003 -9 0.0051* 0.0051 0.0068** 0.0025 0.0010 0.0033 -0.0005 -0.0030 0.0072** -0.0002 -8 0.0084* 0.0084 0.0064** 0.0038 0.0058 0.0000 0.0015 0.0021 0.0146** -0.0014 -7 0.0101* 0.0101 0.0039** -0.0003 0.0029 0.0056 -0.0014 0.0035 0.0160** -0.0030 -6 0.0101* 0.0101 0.0034** 0.0036 -0.0002 0.0079 -0.0052 0.0037 0.0150** -0.0093 -5 0.0046* 0.0046 0.0064** 0.0047 -0.0003 0.0113 -0.0048 0.0060 0.0082** -0.0096 -4 0.0045* 0.0045 0.0090** -0.0016 0.0032 0.0084 -0.0072 0.0094 0.0090** -0.0050 -3 0.0039* 0.0039 0.0045** -0.0008 -0.0041 0.0075 -0.0136 0.0107 0.0068** -0.0103 -2 0.0122* 0.0122 0.0106** -0.0098 -0.0051 0.0082 -0.0147 0.0151 0.0036** -0.0099 -1 0.0083* 0.0083 0.0096** -0.0133 -0.0056 0.0066 -0.0212 0.0140 0.0017** -0.0108 0 0.0164* 0.0164 0.0154** -0.0087 -0.0096 0.0024 -0.0199 0.0121 0.0030** -0.0061 1 0.0103* 0.0103 0.0116** -0.0146 -0.0135 -0.0040 -0.0223 0.0088 -0.0093** -0.0060 2 0.0093* 0.0093 0.0087** -0.0152 -0.0182 -0.0064 -0.0195 0.0081 -0.0188** -0.0052 3 0.0103* 0.0103 0.0084** -0.0114 -0.0216 -0.0039 -0.0155 0.0074 -0.0118** -0.0074 4 0.0120* 0.0120 0.0144** -0.0186 -0.0201 0.0014 -0.0099 0.0067 -0.0283** -0.0044 5 0.0115* 0.0115 0.0194** -0.0152 -0.0201 0.0003 -0.0038 0.0089 -0.0406** 0.0017 6 0.0104* 0.0104 0.0213** -0.0179 -0.0198 0.0080 -0.0093 0.0115 -0.0411** 0.0039 7 0.0100* 0.0100 0.0288** -0.0205 -0.0189 0.0101 -0.0066 0.0100 -0.0432** 0.0007 8 0.0018* 0.0018 0.0309** -0.0262 -0.0226 0.0062 -0.0111 0.0049 -0.0459** 0.0023 9 0.0121* 0.0121 0.0326** -0.0292 -0.0226 0.0029 -0.0139 0.0131 -0.0400** -0.0051 10 0.0163* 0.0163 0.0347** -0.0311 -0.0223 0.0056 -0.0103 0.0095 -0.0410** -0.0091

Note: Results with * are rejected at 5%, and ** at 1% significance level.

How do quarterly earnings announcements affect the prices of stocks of Silicon Valley companies? : a comparison of Silicon Valley growth and value companies