1 - Duke University



Behavioral vs. Fundamental Finance:

An Analysis of Macroeconomic Indicators Effects on Stock Prices[1]

Allison Keane

Fall 2008

Introduction

An article in the Wall Street Journal on September 6, 2008 stated, “When the jobless report first came out Friday, the Dow Jones Industrial Average slid more than 150 points… The Dow’s nearly 345 point fall on Thursday was largely attributed to anticipation of grim job news.” The recent market turmoil and sensitivity to information would make one assume that macroeconomic new announcements, such as the non-farm payrolls report, when released, would cause fluctuations in the market. However, questions arise regarding the legitimacy of these assumptions. Does the market always react to the economic measures upon release each month, or does it do so currently only because the economy is suffering? If the economy is prospering, and the unemployment rate is released at a higher value than the market expected, would stock prices still suffer a shock?

The buying and selling of stocks depends on the individual investor and his or her reaction to outside factors. The state of the economy, current market fads, an individual’s personal wealth and willingness to take on risk are all factors affecting how one might invest. A risky technology stock might be appealing to an investor following the crowd, but passed up by a risk-averse individual. An investor with a family member who dies of lung cancer may decide to sell a cigarette company stock in their portfolio for personal reasons, while many people would choose to buy such a company’s stock during a recession. The release of a new cancer treatment may attract a rush of investors to a pharmaceutical company. Many factors play into market conditions and reactions to different events, especially media attention. The release of an economic indicator may be overshadowed in the media by a political scandal or natural disaster, causing investors to ignore the indicator. During times of economic prosperity, the indicators are not the “hot news” to report. The Dow Jones Industrial Index appeared to have fallen last September due to the jobless report as the Wall Street Journal described, but during the technology bubble in the late 1990s were investors even paying attention to the jobless rate?

Speculative bubbles occur when stock prices rise to an overvaluation according to the fundamental valuation technique. Malkiel’s “Castle-in-the-air theory” of investing described in his book, A Random Walk Down Wall Street, assumes people invest based on how they believe the crowd will behave in the future. From this viewpoint, a stock is worth as much as someone else is willing to pay, regardless of the company’s finances or strategy. As a result, investors buy at any price as long as they expect to be able to sell at a higher price. The increase in prices results in increased demand, which leads to even higher prices, and the cycle runs continually until the bubble pops. Shiller calls this repetitive process the feedback theory. “In the most popular version of feedback theory, one that relies on adaptive expectations, feedback takes place because past price increases generate expectations of further price increases” (69). The “castle-in-the-air theory” and “feedback theory” help explain the reoccurrence of market bubbles throughout history.

As described by Chancellor (14-20), the emergence of modern economic bubbles began with the Tulip Craze in the 17th century, when the Dutch Republic was introduced to the tulip. The flower quickly grew in popularity and the more colorful the flower was, the more valuable it became. Merchants cross-pollinated the colors to produce increasingly more expensive goods and create more market participation. The tulip became a sign of wealth and an assumed safe investment. Prices rose so high that people decided they would rather trade the flower for the money, and the tulip bubble crashed. In the end, the flower was worthless, and people who had traded goods for tulips were left with nothing but perishable flowers.

Such bubbles and investor actions are not only confined to the past. A recent example is the technology bubble in 1999. New technology was developed and considered a new business opportunity that caught on as a market obsession. Through the feedback loops described in Shiller’s theory, technology company stocks soared in prices. “Cisco was selling at triple-digit multiple of earnings… If Cisco returned 15% per year for the next twenty-five years and the national economy continued to grow at 6% over the same period, Cisco would have been bigger than the entire economy (Malkiel 80). Clearly, these prices and estimates were outrageous and unrealistic, which caused analysts to create “new metrics” for evaluating technology companies, such as the number of people visiting the websites. Following the fad, non-technology companies changed their names to include a web orientation, such as “.com”, to increase the stock prices. “Companies that changed their names enjoyed an increase in price… that was 125 percent great than that of their peers…even when the companies core business had nothing whatsoever to do with the Net” (Malkiel 82). The technology craze also increased the financing of ridiculous startup companies; for example Digiscents, which offered a computer plug-in to make web sites smell, gained financial support (Malkiel 83). As is common in all economic bubbles, an oversupply and the realization of the absurd ventures caused a burst.

Bubbles have possibly occurred many times within the US stock market as a result of the imperfections of human behavior which cause investors to act irrationally and make errors in judgment. Overconfidence in their own abilities, bias towards certain stocks, aversion to loss, and herding are four systematic ways people deviate from rational investment behavior (Malkiel). The study of stock prices as a result of human psychology and estimates of future actions is known as Behavioral Finance. Historic bubbles and market deviations from intrinsic values lend support to this branch of economics. The major underlying assumption of behavioral finance is that investors do not act rationally, and as a result, their decisions can only be determined by accounting for the underlying human psychology.

On the contrary, the market efficiency theory assumes all investors act rationally and the intrinsic value of a company is reflected as the price. The market may deviate from this value at times, possibly for an extended period as in economic bubbles, but will return to the “correct” price in the long-run. If the price is too high, investors will sell, and if the price is too low, investors will buy. The market gravitates towards the intrinsic value. The market efficiency theory assumes that prices reflect and correctly interpret available public information.

Malkiel discusses the “firm-foundation theory” to oppose his “castle-in-the-air theory” in which the value of a company is evaluated based on the present conditions and future company prospects. The dividends and earnings are the basis for the valuation, and the investments should be based on comparison of the actual price and the computed price. If it is trading below the fundamental value, the stock is a buy, but if trading above, the stock is a sell. The prices are affected by announcements and news that provides information regarding future returns and growth. The “firm-foundation theory” bases investments on estimating the values of the rate of return and the growth rate, two variables in the fundamental value calculation, as opposed to predicting the crowd behavior.

This paper investigates the fundamental viewpoint and behavioral finance viewpoint by analyzing the impact macroeconomic news announcements have on stock returns. A clear relationship would support the fundamentalist theory that investors react rationally to available news. Lack of a relationship would support that investors are acting on their own accord and ignoring the economic indicators, the behavioral view. The availability of high-frequency data has enabled the use of more precise measurements in the investigation. The overnight return can be calculated using prices from a specific time in the morning and the closing price of the previous day. The intraday returns can be determined and used to calculate variance. The data allows the development and implementation of the local variance, a new standardization technique. Therefore, the returns can be standardized by a value more accurately reflecting the volatility of the time when the return occurs.

This paper is organized as follows; section 2 describes the contrasting pricing models; section 3 describes the return models; the regressions for analysis are provided in section 4; the data is explained in section 5; section 6 describes the results, followed by the conclusion in section 7.

Pricing Models

The fundamental pricing and behavioral finance viewpoints utilize different valuation models as support their theories. Both techniques are explained in the following section.

1 Fundamental Stock Valuation

Market efficiency theory assumes rational behavior from investors and that stock prices accurately incorporate all available, public information. By combining the public information, such as macroeconomic announcements and company financials, the stock will be correctly priced via a mathematical model. Two different models of the fundamental value of a stock are discussed here, the Dividend Discount Model and Company Comparables. The latter is used when the firm being valued does not pay a dividend or pays a very small dividend.

1 Dividend Discount Model (DDM)

The DDM is a mathematical model used to calculate the price of a stock based on the dividend a company pays to its shareholders and the discount rate. The discount rate, r, is the expected rate of return an investor requires in order to buy the stock. The riskier the stock is, the higher the rate of return will be; the higher return is the payoff for taking on more risk. The discount rate is calculated via the Capital Asset Price Model (CAPM),

[pic]. (1)

Here, rriskfree is the return of a riskless asset, commonly the 10 year Treasury bond return, rmarket is the market return, and β is a measure of the riskiness of the stock. The discount rate is affected by economic activity via rriskfree and rmarket, and thus responds to macroeconomic announcements. As shown in the equation, the higher the β value (the riskier the stock), the higher the rate of return r.

The DDM takes the rate of return from the CAPM and the company’s dividend, which can be found on the company balance sheet, to calculate the current price. A balance sheet lists the assets and liabilities a company holds and is the record of the company’s financial position at a specific point in time. The pricing model,

[pic], (2)

assumes that a stock is worth the sum of the future dividends, discounted to today’s value. P0 is the price today, Dt is the dividend at time t, and r is the rate of return calculated from CAPM.

A simplified version of the DDM is the Gordon Growth Model which assumes that a company’s earnings will grow at a constant rate g and the company will retain a constant dividend payout ratio d. The expected dividend at time t then becomes

[pic]. (3)

N0 and D0 are the current earnings and dividend values. The DDM simplifies to

[pic]. (4)

The value of the constant growth rate g can be estimated based on historical growth, analyst estimates, or an accounting method

[pic]. (5)

ROE is the company’s Return on Equity, a statistic found on the company’s balance sheet. The return on equity is a ratio defining how efficiently a company is using its equity. It is calculated by dividing the net income by the total equity. The estimate for g used will alter the stock price determined by the fundamental valuation.

Based on the market efficiency value and the assumption that investors act rationally, the r and g values used in the DDM should be affected by economic indicators, and consequently, the price of the stock P0 should be affected. For example, the theory assumes that if a high unemployment rate is released representing the onset of a recession, the values of r and g will decrease in the future. If r decreases by more than g, then the price P will increase, but if g decreases by more than r, then the price P will decrease. The DDM assumes the investors incorporate the information such as announcements into this valuation.

2 Company Comparables

The DDM does not provide a valuation technique for companies that do not issue dividends. A company may decrease dividends to fund a new project or reinvest the money into company operations to add value as opposed to issuing a dividend. According to the DDM, such companies would be worthless, which is misleading. An alternative method, using comparable companies to value a stock, can be used in such situations.

This model uses ratios to value comparable assets, but begins by first determining companies that are considered to be comparable to the company in question. The choices are based on comparing size, industry, market share, and structure. Once suitable companies are found, statistics of these companies, excluding the company in question, are analyzed. For example, the price P0 of each company, the Book Value B0 of each company, and the ROE of each company are noted. These three values can be found on each company’s balance sheet. Using the statistics, a regression,

[pic], (6)

is performed to determine a value for [pic]. Once [pic] is found, it is multiplied by the ROE of the company J in question and the Book Value BJ of company J to determine the price PJ,

[pic]. (7)

The model can be applied using other statistic values instead of ROE and the Book Value, but those are chosen at the discretion of the analyst.

A short-coming of this valuation technique is its performance during bubbles or times of poor performance. If companies are overvalued and prices are being inflated, then the statistics of the similar companies being used would be inflated as well. For example, during the technology boom, the tech stocks were overvalued and the prices high. Using such prices in a valuation will skew the result towards the higher end. On the other hand, if on average a sector is poorly performing, then using these prices and statistics to value a prospering company within the sector is misleading. For these reasons, careful consideration of the companies used in the valuation must be exercised.

3 Microstructure Noise

The observed market price of a stock may differ from the fundamental price determined from the Gordon Growth Model or comparable companies’ model; this difference is known as microstructure noise

[pic]. (8)

[pic] is the fundamental price, and [pic] is the observed price. The fundamental model does not hold for every minute of trading and the price will fluctuate. Due to the bid-ask bounce throughout trading days, the price will not remain at a constant value. When using high-frequency data, such as that used in this paper, the microstructure noise can not be ignored. The noise affects the return volatility calculations and is accounted for by using five-minute intraday returns. Microstructure noise affects the overnight returns in this paper as well. The price at a specific time, such as 10:00am, is used to calculate overnight returns. If the price at this time is affected by microstructure noise, it could skew the return value.

2 Behavioral Finance

The equations above calculate the fundamental value of an asset, but sometimes the prices that stocks are traded at differ greatly from these values. The valuations above are based upon the impression that investors act rationally to optimize their returns, but a contrary view has begun to be investigated in behavioral finance. Based on common human errors in investment decisions, stocks may not be priced according to their fundamental value, but subject to whatever price someone will pay. “Such price behavior is consistent with common models of an irrational market in which stock prices take long temporary swings away from fundamental values” (Fama, French 1988).

Economic bubbles can be explained through behavioral finance theories and are thoroughly discussed in Shiller’s Irrational Exuberance. The formation and propagation of bubbles is explained by the “feedback loop theory”. An increase in prices will create greater investor demand, which in turn causes a further rise in prices. Feedback that occurs because past increases generate expectations of even more price increases is a result of adaptive expectations. Past price increases can also boost investor confidence, thus causing growing demand and further price increases. There are two specific types of feedback that can cause price inflation experienced in bubbles, price-to-price and price-to-GDP-to-price. In the first, an increase in the price creates feedback that directly increases the same price again. The latter is a process in which the stock prices increase, which raises personal wealth and optimism. This wealth effect causes people to invest more. The feedback theory propagates assuming the price will continue to rise and others will buy, but when investors begin to think the prices will stop rising, everyone wants to sell and the bubble bursts. Often such large increases in prices during bubbles are followed by equal or larger decreases during the price declines (Shiller 68-71).

The media plays an interesting role in the propagation and burst of an economic bubble. By directing attention toward certain stocks or a specific industry, the media enhance the feedback loop. Announcing estimates that the prices will stop rising, may pop a bubble and instigate or exacerbate the price decline. The media, serving as a mass communication channel, reaches out to the public and influences them on which stocks are trendy and which are passé. The media tends to create a story about the stocks, and embellish the details making the news more interesting. These stories appeal to people and tend to influence investment. The effects the stories have, according to Shiller, are based more on how the public feels about the news than the underlying implications for the fundamental value. For example, if people are happy that the rate of return increases, prices may increase as well, even though the intrinsic value calculation predicts a decrease in the price.

Decisions people make are influenced by suggestions, and Shiller presents two “psychological anchors” investors use to justify prices they are paying. The first is a quantitative anchor where the price is based on a numerical value. The price of the stock the previous day or week will serve as a starting point for determining a price. Prices of similar companies, either geographically or industrially, may also serve as anchors. A third numerical anchor may be analyst estimates. If a company exceeds estimates, investors see this as a strong sign, but if they are below estimates, they are seen as underperforming. This basis is only relative to what the estimate had been set at, not on the underlying reasons for the numerical value. An alternative anchor used by investors is an emotional anchor. People are more interested and likely to invest in stocks with an accompanying vivid story. These stories provide meaning and significance to company events or changing prices, when in reality the events are purely random.

There are certain characteristics of human nature that are unconsciously used by people to justify their investment decisions. Four main deviations from rational behavior are of greatest focus in research: bias judgment, overconfidence, loss aversion, and herding behavior. Individual investors often have difficulty remaining unbiased when evaluating their portfolios. If they advocated for a stock, maybe because they found the story behind it inspiring, they will be reluctant to sell even as the price declines. They hold out, hoping that it may rebound long after they should have sold.

People tend to believe they know more than they really do; they are overconfident in their abilities. “More than most other groups, investors tend to exaggerate their own skill and deny the role of chance. They overestimate the risks involved, and exaggerate their ability to control events” (Malkiel 220). Often, people develop types of superstitions where they think a certain action has made them lucky and will continue to, even though the events are random. Common among investors is the intuitive sense about the future course of the market. Shiller explains the role overconfidence plays in the market, “if people were completely rational, then half the investors should think that they are below average in their trading ability and should therefore be unwilling to do speculative trades with the other half, who they think will probably dominate them in trading” (154). He uses the existence of overconfidence as evidence against the rational assumption of fundamental theoretical pricing.

Psychologists have found that people unequally value gain and loss, resulting in risk-averse investors. Tversky and Kahneman (1981) state “because the value function is steeper for losses than for gains, a difference between options will loom larger when it is framed as a disadvantage.” This unbalanced view of risk causes the framing of an investment opportunity to be important. For example, there are two choices A and B such that:

A: If chosen, 100 people will survive.

B: If chosen, there is a 1/3 probability 300 people live, and 2/3 chance the people die.

Most people will choose A. On the other hand, suppose they are worded

A: If chosen, 200 people will die

B: If chosen, there is a 1/3 probability 300 people live, and 2/3 chance the people die.

Program B would most often be chosen. Both scenarios provide the exact same outcome, but the framing influences the decision because the choice explicitly stating the loss will be avoided.

The characteristic considered the most responsible for bubbles and the largest deviations from intrinsic value is herding behavior. There is a reassurance in numbers felt by investors. The mind set that “it must be true if everyone believes it” applies to stock prices. An individual will react to a judgment the way a large group does, even if it contradicts what they originally believed or obvious logic. As mentioned earlier, Cisco was trading at triple-multiples of earnings during the technology bubble, even though investors knew this to be an overvaluation. The herding behavior is largely a result of the “information cascade” (Shiller). The third investor in line will have the same information the previous two did, but also will know the choices investors one and two made, and will follow suit. The most influential information passage occurs through one-on-one conversations discussing “hot stocks”, scandals, and gossip about businesses. The mathematical finance and statistic ratios of companies are dry, lifeless topics often avoided. As a result, Shiller defined these theories as the “failure of information about true fundamental value to be disseminated and evaluated” (160). The juicy gossip is transmitted most frequently and investors focus attention on others’ decisions and not necessarily on what is numerically correct.

These characteristics of human behavior lead to trades that deviate strongly from fundamental values. The market becomes subject to the errors in human judgment and the market efficiency is put on hold for some time. During times of high speculation and strong herding behavior, the fundamental values of stocks are irrelevant.

Returns, Volatility Measures, and Jump Values

This section first explains the intraday return, volatility, and jump measures based on previous work by Barndorff-Nielson and Shepard (2004, 2005). The intraday measures are altered to develop return, volatility, and jump measures specific for overnight observations, which are used to investigate the effects of announcements that occur when the market is closed. Lastly, in order to compare prices of varying stocks with different volatilities, standardization techniques are outlined.

1 Intraday Returns, Volatility, and Jumps

Consider a price function pt that takes the log of the stock price Pt. The five minute intraday returns are calculated as

[pic], (9)

where j = 1, 2, …M and M = 78, the sampling frequency. Using the five minute intraday returns, the quadratic variance was measured by the Realized Variance (RV)

[pic]. (10)

The RV is used throughout this paper to as a variance measure for the returns. The Bipowered Variation, BV, is also calculated as a measure of quadratic variation,

[pic], (11)

[pic].

These two measures are investigated in the Barndorff-Nielson and Shepard (2004, 2005) papers and found to separately identify two components of the quadratic variation. The Realized Variance combines the variation of the continuous process as well as the variation of the jump process. The Bipowered Variation estimates only the variation of the continuous process by multiplying adjacent returns.

Using Bipowered Variation in conjunction with Realized Variance, the jump components of daily volatility can be isolated via the equation for the relative jump,

[pic], (12)

defined in the Barndorff, Nielson, and Shepard (2005) paper. The jump components calculated were tested for significance using the z-statistic

[pic]. (13)

In equation (13),

,

(14)

is an estimate of integrated quarticity. The quad-powered quarticity, QP, value studentizes the relative jump value and when investigated by Huang and Tauchen in their Monte Carlo simulation, was found it to be an appropriate integrated quarticity estimate. The jump components are tested against the hypothesis that no jumps occur with a 0.999 confidence level. When the hypothesis is rejected for a specific day, a jump is presumed to have occurred at some time throughout that day.

2 Overnight Returns, Volatility, and Jumps

Don May (1992) investigated the effect rate change announcements, made by the Fed, have on stock prices. May confirms “the market’s adjustment to the information contained in a discount rate announcement occurs in the hour of the announcement.” When faced with announcements occurring while the market is closed, May uses “the previous day’s closing price and the opening price as of the next trading day to gauge the market reaction.” A similar, but slightly altered model is used in this paper to calculate the market’s reaction to announcements.

Using the same price function pt as described in 3.1.1, the overnight returns are calculated as

[pic], (15)

where d = 2,3, …, D, for D days in a sample, k is the time of the day, and c is the closing time of the market. For example, the overnight return of day 2 is the difference between the log price p at time k on day 2 and the log price p at the time of closing of day 1. The log of the closing price of a stock would be subtracted from the log of the price of that same stock at 10:00 AM the next morning. The magnitude of the overnight return varies depending on the time k chosen to use for the calculation (9:30AM, 10:00AM, 10:30AM, etc). The opening price is not used in this paper as it was in May’s work because this does not allow ample time for the market to move as a reaction to the announcement. Multiple times are investigated to determine when the market can be considered adjusted from the announcements occurring before the market is open.

The Realized Variance for the overnight returns can not be used as a variance measure due to lack of observations; therefore, an alternative model is used to calculate the overnight volatility. Using the five minute intraday returns, the Bipowered Variation for each day is calculated via equation (11). The moving average of the BV values is calculated as

[pic] , (16)

where λ = 0.95. The moving average is weighted by a scaling factor, L, to determine the overnight volatility. The scaling factor L,

[pic], (17)

calculates the amount overnight variance contributes to total variance. The scaling factor measures the contribution for the entire sample. Scaling the moving average BV by L results in a measure of the overnight variance

[pic]. (18)

Using the overnight variance, the overnight jumps can be defined as

[pic]. (19)

The overnight jumps are tested by a t-test against the hypothesis that no overnight jumps occurred with a significance level of 0.10.

3 Standardization of Returns

The volatility of returns for one stock will differ from that of another stock, and the volatility pattern will vary by stock as well. Standardizing the returns by their own volatility measure allows the returns to be compared across industries and prevents over-calculating jumps for more volatile stocks. The standardized return values are

[pic], (20)

where rt is the overnight return calculated according to equation (15). Three volatility measures, VM1, VM2, VM3, are investigated to determine an appropriate standardization. The realized variance for each day is calculated using the 5 minute intraday returns as explained in equation (9). The first volatility measure is

[pic]. (21)

The square root of each day’s RV value was taken, and then the overnight return was standardized by the previous day’s RV. For example, r2 (the return calculated by subtracting the closing price of day 1 from a morning price of day 2) is divided by the square root of the RV of day 1. The second volatility measure is

[pic]. (22)

The RV values for the preceding 5 days are averaged and then the square root is taken. This is a measure of the average variance of the past week. The standardization occurs such that r6 (the return calculated by subtracting the closing price of day 5 from a morning price of day 6) is divided by a volatility measure calculated from the previous trading week (days 1, 2, 3, 4, and 5, but not including day 6). The third volatility measure, VM3, considered calculates the average volatility of the past trading month instead of week. The previous 22 days RV values are averaged and then the square root is taken. Therefore, r23 is standardized by the volatility measure calculated via the previous 22 days.

The beginning work for this paper focused on investigating the three different volatility measures and determining which value was appropriate to use as a standardization value. Using a variance measure to diversify out the different volatility patterns of individual stocks is an intuitive method, but the variance changes over time, preventing the use of the variance of the entire sample. VM1, VM2, and VM3, are three different estimates of local variance, a day, week, and month estimate. VM1 skews results if one day experiences specifically large or abnormally small variance. Averaging the variance over the month with VM3 presents the same difficulty as using the variance of the entire sample. The VM2 measure allows the return to be standardized by an appropriate measure of the local variance that does not skew results based on one day’s high or low volatility and also allows the value to evolve over time the way the volatility does in reality.

2 Announcements

The announcements used in this paper are quoted in different units and measure varying quantities. In order to compare across all twelve announcements, the announcement surprise S,

[pic], (23)

is calculated following the model of Anderson, Bollerslev, Diebold, and Vega (2007). A is the actual announced value of the kth announcement at time t, E is the expected value of the kth announcement at time t, and σk is the sample standard deviation of the (Ak – Ek) over the entire time period investigated. S is the standardized value of announcement k = 1, 2, …, 13, and t = 1, 2, 3, …, D for D days in the sample. This paper looks only at the difference between the actual and expected announcement, but not at the actual announcement alone. The announcements are considered to be jumps if the magnitude of S is greater than 1.645, equivalent to testing at the 0.10 significance level.

Regression Methods

The following regression methods are used to investigate the effects announcement surprises, announcement values different from the expected values, have on stock prices. Univariate regressions are performed as well as multivariate regressions. The returns used as the dependent variable are either the standardized overnight returns of individual stocks or an equally weighted average of standardized overnight returns of stocks in the same industry.

1 Univariate Regressions

Regression analysis was used to investigate the effect one specific announcement would have on an individual stock and the effect a single announcement would have on a portfolio of returns. The standardized return values are regressed on the standardized announcement values by the model

[pic]. (24)

Here, t represents the day and k corresponds to the announcement. Only data for the days when the kth announcement was released were used for the regression. The regressions estimate the values for βk, and test the coefficient against the hypothesis that β= 0. Regressions producing a p-value < 0.1 are considered statistically significant.

2 Multivariate Regressions

The effect multiple announcements have on an individual stock return or equally weighted portfolio returns was investigated with multivariate regressions. They are performed according to the following equation where the sum over k is taken from 1 to K where K = 2, 3, 4, or 5 corresponding to K different announcements. The regression,

[pic] , (25)

included days where at least one of the K announcements was released. If one announcement was released that day, but the others included in the regression were not, the ones not released were assigned a value of 0. The coefficients, βk, are tested against the hypothesis that all the βk values are zero.

3 Sign - Split Regressions

Anderson, Bollerslev, Diebold, and Vega concluded that “on average, the effect of the macroeconomic news often varies with its sign” (2003). A negative announcement surprise would have a more powerful effect on the stock prices than a positive announcement surprise because investors react more to bad news than good. This corresponds to Tversky and Kahneman’s conclusion that the value of a loss is greater than the value of a gain of equal magnitude. For example, if the hourly earnings announcement was significantly lower than the expected measure, the reaction would be greater than if the announcement had been equally higher than the expected measure. Using an approach similar to Anderson et al. (2003), the regressions are split based on the sign of the announcement,

[pic] (26)

[pic].

If the actual announcement is less than expected, then S ................
................

In order to avoid copyright disputes, this page is only a partial summary.

Google Online Preview   Download