This chapter discusses the type of the research, sources of data, population of research and methodology how to analyze data of the research. In addition, this chapter also describes dependent and independent variables and chooses the model to test hypothesis.
3.1 Research design
This research is an empirical research. Quantitative approach will be used in this research which is mean that this study use number in the data processing
Types of data used in this research:
- Researcher used secondary data. The source of data for this research is taken from Ho Chi Minh Stock Exchange website, www.hsx.vn in the form of the financial statements of the company.
- The data analyzed in this article are quantitative data, systematically indexed for easy analysis and statistics. Data is analyzed based on collection, statistics and treatment figures. Quantitative data in the study was collected through financial statements of real estate firms from HOSE website.
3.2 Sampling design
Categories statistics used is inferential statistics or sampling statistics. The population of this research is all the real estate companies listed on the Ho Chi Minh Stock Exchange (HOSE) period 2007 - 2016 with a total of 35 companies. It is an attempt to make the database of companies as complete as possible. Because some data is insufficient and some companies did not update all data, therefore, the sample consists of about 305 observations.
3.3 Data collection method
A method of collecting data on this study is the technique documentation, by collecting the data from financial statements of the estate enterprises in period 2007 - 2016 that has been published. Data is collected from the official website of the Ho Chi Minh Stock Exchange.
Types of data used are interval data, ratio data and continuous data. The study uses book values of calculated variables.
3.4 Measure of research variables
The dependent variable is ROE ratio. This variable has important information when assessing the profitability of a company.
The independent variables used in this dissertation are based on the work by Boyd et al.
(2007). The definitions of the variables are summarized in Table 3.1 Table 3.1 Description of variables in model testing
Calculation Expected sign
ROE Net Income
LOG ASSETS Liabilities + Shareholders’ Equity +
Because the influence of these variables are not really clear so in this study, the variable will be used in the form LOG (ASSETS) to limit the dispersion of data and does not affect to other variables.
3.5 Data analysis
After all the data completed, researcher will analyze the data using statistical software called as Eview 8.1 for drawing appropriate conclusion. Data analysis was conducted to determine the effect of independent variables on the dependent variable.
After all the data completed, researcher will analyze the data using statistical software called as Eview 8.1 for drawing appropriate conclusion.
Research model should look:
Reliability analysis is used to know the consistency of the instrument, whether the instrument can be relied and still consistent when it is test many times. When choosing Least Squared model, we should test the existence of a unit specific component in the error. In this paper, author will test multicollinearity, autocorrelation and heteroskedasticity to make sure the result is reliable.
Test multicollinearity
Multicollinearity is a circumstance that show the correlation between independent variables and their relationships can be explicit in form of a formula. To identify multicollinearity, Variance inflation factor (VIF) is used. A variance inflation factor (VIF) quantifies how much the variance is inflated. One other way to test multicollinearity is following the correlation table. “According to the rule of thumb test, multicollinearity is a potential problem if the absolute value of the sample correlation coefficient exceeds 0.7 for any two of the independent variables” (Anderson, 2008).
Test Autocorrelation
Autocorrelation is the phenomenon of correlation between observations in the same dataset.
This phenomenon usually occurs with time series.
, 0 (u j)
In this paper, we use Breusch - Godfrey test to check autocorrelation. The Breusch-Godfrey (BG) test is most common test and test for higher order serial correlation, AR (q)
(3.2)
Autocorrelation is usually occurred when data is following over the time, therefore the form of equation is:
. . . (3.3)
H0: ρ1 =...= ρm = 0 (no autocorrelation) H1: ρ1 =...= ρm 0 (autocorrelation)
After running regression, p_value > α (0.05), it is said that H0 is accepted. This model has no autocorrelation
p_value < α (0.05) that means H0 is rejected which exist autocorrelation diagnostic.
Test heteroscedasticity
Heteroskedasticity is said to occur when the variance of the dependent variables has varied levels of change for each value of the independent variable, is not constant.
(3.4) One method to test heteroskedasticity is using White test. According to White test, we conduct regression secondary:
(3.5)
(3.6)
To test for heteroskedasticity, we use hypothesis
H0: | , , … ,
H1: | , , … ,
p_value > α (0.05), H0 is accepted. This model has no heteroskedasticity p_value < α (0.05), H0 is rejected which exist heteroskedasticity diagnostic.
3.5.2 Haussman test
Panel data is used to run data in this study. In the panel data, space-based diagonal units are surveyed over time. In brief, table data has both spatial and temporal dimensions (Juanda, 2012). According to Baltagi (1995), Ajija (2011) and Khadul (2014), panel data contributes several advantages. First, panel data refers to individuals and businesses over time, there should be a distinct (heterogeneous) feature in these units. The table data estimation technique can formally consider that difference by examining individual-specific variables, as discussed below. We use the term individual in the general sense including micro units such as individuals and businesses. Secondly, by combining spatial chronology of spatial observations, table data provides more informative, more diverse, less coherent data between variables, more degrees of freedom, and more effective. It is better because panel data can measure the effects that cannot be observed in pure time series data or in purely spatial data. Finally, by collecting the available data for several thousand units, panel data can minimize the bias that can occur if we aggregate individuals or businesses into aggregate data
According Biørn (2016), General linear regression model:
(i=1,…, n) (3.7)
There are two types of the method of Generalized Least Squares (GLS) which are Fixed Effects Models (FEM) and Random Effects Models (REM) to test incontrollable variables.
The Fixed Effects Model is an extension of the classic linear model and is defined according to the following formula (Allison, 2009)
(3.8)
With i = 1, 2, 3… (Number of companies, which is company A, B, and others) t = 1, 2, 3, 4… (Number of years, which is 2007 - 2016)
are called the fixed effects, and induce unobserved heterogeneity in the model.
The term 'fixed effects' is due to although the original toss may be different for individuals (here are 35 companies), but the toss of each company does not change over time. Meaning immutable over time.
The Random Effects Model is that, Instead of seeing β1i as fixed, we assume it is a random variable with an average value of β1 (without the i symbol here). And the original pitch value for an individual company can be expressed as:
(3.9)
Where the error term is decomposed as
is a random effect ∼ N(0, ) . It is the permanent component of the error term.
is a noise term ∼ N(0, ). It is the idiosyncratic component of the error term.
Also, in order to find which of these models is the most appropriate, the Hausman test can be conducted. Hausman test is an analytical test to choose we use the Fixed Effects Model or Random to analyze the regression data.
: and independent variable uncorrelated : and independent variable correlated
If the null hypothesis is rejected, the appropriate model is the FEM. Likewise if the null hypothesis is not rejected, the appropriate model is REM.