The aim of this report is to analysis the share price and firm size with statistics tools (SPSS) between manufacture and service sector.
Introduction
Investors always pay more attention to share price in a stock market and use many complicated technique to analysis the trend of price. As a result, that may overlook some important factors such as revenue, profit and employees which are fundamental compositions of a company’s size and a potential support to share price, Some researchers state that the distinction between the effect of firm size and share price is important for the direction of future trend(Brockman.1997) while others argue that focusing on sectors is more important than on individual stocks. So, it is necessary to analysis the share price between sectors as well as illustrates some possible relation with size.
Therefore, the aim of this report is to analysis the share price and firm size with statistics tools (SPSS) between manufacture and service sector.
Data Collection
To make the result more reliable, the data collection is a crucial step. In this case, the primary data are required
The data collecting procedure including the following steps:
1. Defining the population of shares number from the stock market. Here, the research focuses on Fortune 500 for they delegate main stream of stock market in the world.
2. Selecting 500 shares of manufacture sector and 500 shares of service factor from Fortune website and input them into tables with SPSS software.
3. Using SPSS to random 20 cases out of 500 companies in manufacture sector as well as in service factor.
4. Searching for share prices and other data related to companies’ size. The sizes are limited to three factors: revenue, profit and the number of employees. These three factors are main criteria for evaluate a company’s scale. Producing these two sectors tables as spreadsheets.
5. Ranking price within these two tabled samples as ascending order.(Figure:1,Figure:2)
There are some assumptions and conditions related to the above procedures:
1. The samples are completely random.
2. The samples selected are representatives of all population.
3. Each score within the sample is independent of all other scores. There are not influences among them.
4. All data are collected in the same time.
Figure 1: Manufacture Data
Figure 2: Service Sector
The Research Methodology
The research procedure will begin with a description of the prices and size with presentation of graphs and Kolmogorov-Smirnov test are used to check the normality of scores. Following that, T-test is used to assess if there is significant difference in share price between two sectors. Then, Chi-Square test can be used to search for some possible relations between the price and firm size. If relations are reported, tests such as the linear correlation and linear regression are necessary for further illustration.
Distribution of Data
To do the following research, it is important to assess if the share price and size variables are approximate normally distributed. The assessment includes following steps:
- Display the data distribution for price and size variables (revenue, profit and employee) with graphs in two sectors. Graphs show as following:
- Manufacture Sector:
- Service Sector:
v
- It can be seen that all graphs have a hump, the distributions are unimodal.
- Both the share price and profit in two sectors show symmetric curves.
This is a preview of the whole essay
- Display the data distribution for price and size variables (revenue, profit and employee) with graphs in two sectors. Graphs show as following:
- Manufacture Sector:
- Service Sector:
v
- It can be seen that all graphs have a hump, the distributions are unimodal.
- Both the share price and profit in two sectors show symmetric curves.
4. Further test of normality(SPSS) show only share prices in two sectors are normal distribution. (Kolmogorov-Smirnov Sig value are more than .o5)
With these analyses, it can be seen that only share prices are normal distribution while three size variables are not normal distribution although they show approximate curves.
Testing and Presentation of Finding
One of the purposes of research is to assess if there is difference in share price between two sectors. This can transform to statistic question: Is there a significant difference in the mean share prices for manufacture and service sectors. Through the above analyses, the independent-samples t-test can be used with the suitable conditions as follows:
- The variables are continuous scale.
- Random sampling.
- Independent observations
- The distribution of scores for each sector is normal.
Meanwhile, some assumptions are violated. Firstly, assumption that the population are normally distributed is not assured. Secondly, sample size is not sufficiently large(around 30). Lastly, sample variances are unequal (eg., largest/smallest=115). However, the t test is a robust test. This means that it can b appropriately performed even when an underlying assumption is violated. (Grimm.1993)
There are seven steps for test using the independent-samples t test:
Step 1: Define the null and alternative hypotheses:
H0: m1=m2 (The share prices means between sectors are the same, there is not a significant difference in the mean share prices for two sectors)
H1: m1 m2 (The share prices mean between sectors are not the same, there is a significant difference in the mean share prices for two sedtors)
Step 2: Set alpha. Alpha is set at 0.05.
For some companies may influence other companies, such as earn the amount of shares of others, the assumption of independence of observations can not be guaranteed absolutely. A more stringent alpha value is acceptable.(Stevens.1996)
Step 3: Computer t score (using SPSS, show the table Figure 4)
Figure 4:
ur
Step 5: Checking assumption.
The Sig.value in Levene’s test is 0.389, larger than 0.05.This means that the assumption of equal variances of share prices has not been violated, the first line of table can be used to report t-value.
Step 6: Assessing differences between two sectors.
The value in the Sig.( 2-tailed) column is above 0.05(0.739), then the null hypothesis can not be rejected. There is not a statistically significant difference in the mean share prices between two sectors.
Step 7: Calculating the effect size
Eta squared=t*t/(t*t+n1+n2-2)=3.36*3.36/(3.36*3.36+20+20-2)=0.229.
It means 22 per cent (>14 per cent, large effect) of the variance in share price is explained by sectors.
Relationship Analysis
---Chi-Square test
In most time, investors care more about the share price and factors that effect it. Many researchers also have explored the effects of share price and firm size. Beedles et al.(1988) who reported that both size and price effects are statistically significant to average return. While found that 1 per cent change in share price will have a greater impact on average return than a 1 per cent change in firm size. These opinions illustrate some relationship may be existed between the share price and firm size.
Firstly, it is suitable to use Chi-Square test to determine if there is a relationship between the price and size. It includes six steps:
Step 1: Changing the share price and size variables into z-scores, this can make it easy for test samples with standard.
Step 2: Sorting the z-scores in share price with one’s score above 0, another below 0 while the three size variables are sorted into three groups equally.
Step 3: Caculating in SPSS software. Table is showed as following:
29
INTRODUCTION
The finance literature on market efficiency presents the size effect as an anomaly in which small firms stocks
tend to earn higher returns than large firms stocks even after adjusting for the higher risks associated with small
stocks. Several research papers attempted to explain this phenomenon but to no avail. In presence of this
unexplained anomaly, the question remains whether other variables were not accounted for in prior analysis.
In this context, several studies tried to explain the size effect proposing several hypotheses. These explanations
include return measurement error [Roll, 1981], tax-loss selling effect [Reinganum, 1983], transaction costs [Lustig
and Leinbach, 1983; Schultz, 1983; and Stoll and Whaley, 1983], macro economic risk factors in an APT framework
[Chen and Hsieh, 1985], skewness preference [Booth and Smith, 1987], share price level [Kross, 1985; and
Bhardwaj and Brooks, 1992a], and the differential information effect [Barry and Brown, 1984; and Elfakhani and
Zabos, 1992]. None of these studies, however, provided a complete explanation of the firm size effect.
This paper proposes an alternative explanation for the documented size effect. It postulates that small firms
suffer from excessive lack of public information and therefore are neglected. Neglect is defined as professional
informed investors expressing less, or no, interests in some stocks, particularly small stocks. Neglect occurs when
there is less professional analysis available on the stock, and therefore less public information. As such, small
(uninformed) investors require additional returns for holding small-neglected stocks.
Thus, we examine the linear relationship between the size effect and the differential information among small
and large stocks as symbolized by analysts’ neglect. Analysts’ neglect is measured twice, (1) by the number of
financial analysts pursuing information and making forecasts about the stock, and (2) by a dummy interaction
variable that relates size effect to firm neglect. Our methodology differs from that of Bhardwaj and Brooks (1992b)
in at least three ways: First we study month to month variation in the size-neglect group, second we provide a
*University of Saskatchewan,
**Indiana State University
The first author acknowledges partial financial support from the Board of Governors and the Dean of the College of Commerce, University of
Saskatchewan. The second author recognizes the financial support of Indiana State University in getting research data. The authors gratefully
acknowledge the contribution of I/B/E/S Inc. for providing forecast data as a broad academic program to encourage earnings expectations
research. The authors also thank the participants of the seventh annual meeting of the Academy of Financial Services for helpful comments. All
remaining errors are the authors’.
Journal of Financial and Strategic Decisions 30
broader examination of these size-neglect groups (10 versus 4), and third we avoid any cross-sectional
heteroscedasticity or any autocorrelation that may have existed in their two-stage regression.
In this context, we inspect the existence of size effect by using White’s time series cross-sectional regression
corrected for heteroscedasticity and autocorrelation.
Our results support the existence of size effect in January, but only for portfolios of large stocks. Interestingly,
the size-January effect is often dominated by neglect effect; i.e., large firms that are less popular among financial
analysts are found to earn higher return premiums than other more popular large firms. This finding suggests that the
market discounts stocks of larger firms when neglected by financial analysts. The results also show that individual
investors could still earn higher returns on stocks that are less pursued by financial analysts during the 1986-1990
period.
The rest of the paper is organized as follows. Next section reviews previous literature on the subject. Section II
presents the research design, hypotheses to be tested, and data sources. Section III describes the results, followed by
conclusions in section IV.
LITERATURE REVIEW
Banz (1981) finds that risk-adjusted stock returns are a monotone decreasing function of firm size. This finding
implies that CAPM is misspecified. Following this evidence, some studies focus on the interaction between size
effect and other anomalies, e.g., P/E ratio effect.1 Other studies emphasize the magnitude of the size effect.2 These
studies, however, fail to explain size effect month-to-month instability (for example, in January), or its existence.
Other studies referred to earlier also attempt to explain the size effect, but no hypothesis provided a satisfactory
explanation of the firm size effect.
Of these hypotheses, the differential information effect seems the most acceptable theory. Using the period of
listing (as a proxy for information availability), Barry and Brown (1984) find no evidence of an independent size
effect.
Another proxy for differential information is the neglected-firm effect. Under this hypothesis, firms neglected by
analysts’ investors, financial analysts, and other investment agencies suffer from lack of information or asymmetric
information [Arbel and Strebel, 1983]. Thus, neglected stocks should earn substantially higher returns to compensate
for this gap of equal access to firm information. This notion is reinforced by Ajinkya and Gift (1985) in that there
may be a connection between analysts’ forecasts, firm size, and stock returns.
The neglect hypothesis is further tested by Arbel (1985). Using the coefficient of variation of the financial
analysts’ forecasts and the number of institutions holding the stock, Arbel reports a strong January effect associated
with neglected firms. On the other hand, Dowen and Bauman (1986) argue that the size effect dominates neglect
(defined as institutional popularity when stocks are less widely held by institutions). Nevertheless, Amihud and
Mendelson (1986 and 1991) report a confounded size-neglected firm effect; i.e., stocks with low liquidity, as
measured by bid-ask spreads, are usually small and less researched, and therefore earn abnormal returns. However,
they could not explain the interaction that exists between size effect and January.
More recently, Bhardwaj and Brooks (1992b) test the neglected firm effect (proxied by the number of financial
analysts following the stock at the end of each year). They find that the neglect effect is strong in January; however,
it is weakened after controlling for share price. Outside January (all 11 months lumped together), they show that
there is a strong neglect effect in the lowest-price group in the 1977-82 period, and in the highest-price group during
1983-88. Thus, neither the neglect effect nor the share price effect is stable over all the sampling periods. Bhardwaj
and Brooks also report a confounded size-neglect effect in January, which results in a no clear dominance of either
effect. Also, they do not completely rule out an independent neglect effect.
In brief, the literature predicts there maybe a possible relationship between firm neglect and size effect.
Therefore, there maybe profit opportunities for individual investors to seek by pursuing news about financial
analysts’ neglect of small stocks.
RESEARCH METHODOLOGY
The testing methodology is implemented in taking two steps. First, a test of the firm size effect is conducted to
verify the superior performance of small stocks in January and non-January months. This step is tested using the
conventional pooled time series regression analysis (in line with Fama and MacBeth [1973]). Second, analysts’
Differential Information Hypothesis, Firm Neglect and The Small Firm Size Effect 31
neglect effect is tested as a possible explanation of the size effect, and as an inducement to individual investors to
benefit from this strategy.
The Proposed Hypotheses
Findings by Jaffe et al. (1989) and Ritter and Chopra (1989) show that the size effect is restricted to January.
Using our methodology, the null Hypothesis 1 is constructed to examine whether size-portfolios formed in January
would display any different behavior from non-January months or from the overall sample.
Hypothesis 1: For each month of the year, there are no excess return premiums related to firm size.
The next step is to investigate whether the size effect is simply a proxy for the lack of information on small
firms. At the same time, we test the ability of individual investors to realize return premiums one month following
the reports of financial analysts’ forecasts. For that purpose, the following hypotheses are formulated:
Hypothesis 2: After controlling for neglect by financial analysts, there are no excess return premiums
related to size effect.
Hypothesis 3 combines the first two hypotheses to examine any seasonality in the joint size-neglect effects.
Hypothesis 3: After controlling for neglect by financial analysts, and regardless of what month of the
year, there are no excess return premiums related to size effect.
The intuition behind the testable hypotheses is that one would suspect less publicized-small stocks to be more
sensitive to negative information, and therefore would be more volatile. Thus, firms with fewer specialized analysts
are expected to observe gradual market adjustment to unexpected corporate news (neglect effect), and therefore
abnormal returns may be realized by informed investors.
Data Sources
The sample includes all firms trading on the New York (NYSE) or American Stock Exchanges (AMEX) that
have the number of financial analysts following the sampled stocks reported on the Institutional Brokers Estimate
System Tape (IBES). Data on stock prices, returns, and number of shares outstanding are obtained from the Centre
for Research in Security Prices (CRSP) tape. Following other size studies (e.g., Fama and French [1992]), financial
institutions such as banks and insurance companies are considered to have their returns generating behavior
influenced by regulations, and therefore can confound the size effect. Thus, these firms are separated from the
sample for future research. The sampling period in this study is five years, extending from January 1, 1986, to
December 31, 1990.3 However, we exclude October 1987 from the sample to avoid any noise caused by the stock
market crash.4
All remaining CRSP firms are ranked in ascending order based on their market values. The monthly market value
of the firm (Sizet) is defined as the number of shares outstanding multiplied by month-end share price. Shares
outstanding are considered the same from one month to another until a change occurs. At the beginning of each
month, we form ten portfolios, with portfolio 1 has the 10% smallest stocks, and portfolio 10 makes up the 10%
largest stocks.5
The monthly mean return (Rt,ip) and the 90-day risk-free rate for the month (RFt) are used to calculate the mean
risk premium for each stock (Rt - RFt) for each stock in portfolio p (p=1, ..., 10). As a result of merging these data
bases, the final number of sampled firms is 972, and the sample points are 47,629 observations in 59 months (1986-
1990). To simplify the hypothesis testing procedure, the overall mean of the monthly risk premiums is calculated for
each of the ten portfolios.
Testing the Hypotheses
We test Hypothesis 1 using the conventional pooled time series cross-sectional regression. The regression model
examines the relationship between size and return premiums after controlling for January:
Journal of Financial and Strategic Decisions 32
Model 1
Yt, i p = a + b1 Size t -1, i p + b2 Dt-1 + b3 (Size t-1, i p ´ Dt-1)
where:
Yt,i p = natural log of one plus monthly return premiums of stock i in portfolio p (p=1, ..., 10)
for month t (t=1, ...,59). Return premiums are defined as Yt,ip = Rt,ip- RFt, where Rt,ip is
the return on stock i in portfolio p during month t, RFt is the 90-day risk free rate for
the same month.
Dt-1 = lagged dummy variable equals to 0 for all eleven months of the year except January,
and 1 for January.
Size t-1, i p × Dt-1 = lagged interaction dummy variable that is equal to 0 for all eleven months of the year
except January, and the product of size t-1, i p and Dt-1 for January.
where the market value of firm i assigned to portfolio p in month t-1 (Size t-1, i p) is calculated as the number of shares
outstanding multiplied by month-end share price.
The lag operator on the explanatory variables is used because portfolios are formulated at the beginning of each
month, and are then liquidated at the end of month t when returns are actually realized. It must be noted that in
January the coefficient for size is the sum of b1 and b3. A significant total coefficient (b1 + b3) would support the
notion that return premiums are high for small stocks in January.
The following regression with the proposed explanatory variables being lagged one month (as in Model 1) is run
in order to indirectly test the neglect effect as to create profit opportunities to individual investors, thus testing
Hypothesis 2:
Model 2
Yt, i p = a + b1Size t-1, i p + b2N t-1,i p + b3I t-1,i p + εt
where:
Yt, i p = natural logarithm of one plus monthly return premiums of stock i in portfolio p for month t (t=1,
...,59) as defined earlier in Model 1.
Size t-1, i p = lagged monthly natural logarithm of market value of stock i for the previous month t-1.
N t-1, i p = lagged monthly number of financial analysts tracing stock i.
I t-1, i p = lagged interaction dummy variable. The dummy variable is equal to 0 if the number of financial
analysts specializing in stock i during the previous month (N t-1, i p) is larger than the median, and is
equal to firm size (Size t-1, i p) otherwise.
In Model 2, neglect effect is tested twice; first, by the number of financial analysts (N t-1, i p), and second by the
interaction term (I t-1, i p) that connects size effect to firm neglect. We run the regression Model 2 twice. The first run
includes all data. The second regression is performed, this time after grouping the data by size, in order to examine
whether there is any seasonal interaction between size effect and the number of financial analysts pursuing the stock.
The interaction between size and seasonality is also reexamined using regression Model 2. For that purpose, the
sample is reclassified by the month of the year. For example, January includes data from January of each year of the
sampling period, 1986-1990.
Implications of the Proposed Model
Under the proposed models, and given the specified dummy variables, the hypothesized signs of the regression
coefficients in Model 2 are: b2<0 (if the neglect effect exists and significant), and b3<0 (if the size effect is restricted
to neglected firms). Statistically significant coefficient estimates on the number of financial analysts would confirm
the notion that stock performance relates to information availability. Note that if differential information exists
among large and small firms, the size effect in Model 2 should disappear. In this case, b1 must have positive sign.
The constant term, a, would capture factors other than neglect effect.
Differential Information Hypothesis, Firm Neglect and The Small Firm Size Effect 33
Econometric Concerns
Since the regression data is time series and cross-sectional, heteroscedasticity and autocorrelation can present
some serious problems. When heteroscedasticity exists, the estimators may diverge. White’s (1980)
heteroscedasticity-consistent covariance matrix estimates are used to correct the estimates for any unknown form of
heteroscedasticity. This technique also picks up a mis-specified mean or any correlation between the error term and
independent variables. Since the residual effect of stock trading is unlikely to disappear and therefore it may affect
later trading, autocorrelation can also be a problem. Furthermore, to improve parsimony and avoid collinearity,
Akaike’s (1969) Final Prediction Error (FPE), also called Amemiya Prediction Criterion, and Schwarz’s (1978)
Criterion (SC) are used to find the best descriptive model.6 The lower FPE or the higher SC, the more parsimonious
the model. Moreover, a violation of normality may make the regression analysis unreliable. Hence, normality is
tested measuring skewness and kurtosis, and by conducting Kolmogorov (D normal statistics) and non-parametric
sign rank tests.7
FINDINGS AND INTERPRETATIONS
Table 1 reports descriptive statistics for each of the ten-size portfolios. In particular, the table shows the portfolio
mean and median return premiums (portfolio return - risk free rate), and portfolio total risk (σ). The results show
that the mean return premiums are negative for all ten portfolios. In fact, only 23,019 of the 47,629 sample points
have monthly returns higher than monthly risk-free rates (which is about 48.33% of the total sample). Of the 24,610
observations that had negative premiums, 22,371 had negative returns (less than zero) over the sampling period. This
pattern can be partially explained by the observation that 54% of return premiums were negative after the October
1987 crash. Moreover, the table reveals that the smallest stocks (portfolios 1-5) had the most negative mean (and
median) premiums, while the largest stocks (portfolios 6, 8, 9, and 10) had least negative (or even positive) return
premiums over the entire sampling period. This evidence suggests that, contrary to the prediction of the size-effect
literature, smaller stocks underperformed larger stocks during our sampling period (1986-1990). Also, smaller
stocks had higher average total risk than larger stocks.