Embedding that target in a general unrestricted model GUM In its simplest acceptable representation which will later be specified in the econometric model, the GUM of is determined to be
Trang 1FACULTY OF INTERNATIONAL ECONOMICS
-o0o -ECONOMETRIC REPORT
Topic: Factors that determine housing prices
Student Name – ID : Nguyen Ha Trang - 1711150066 – 40%
Nguyen Mai Thuy Tien - 1711150064 – 30% Nguyen Thi Lan Huong - 1715150032 – 30% Supervisor : Dr Dinh Thanh Binh
Hanoi, 2018
Trang 2Table of Contents
II Introduction 3
III Literature overview 3
1 Questions of interest 3
2 Procedure and program used 3
IV Economic model 4
1 Specifying the object for modeling 4
2 Defining the target for modeling by the choice of the variables to analyze, denoted x i 4
3 Embedding that target in a general unrestricted model (GUM) 4
V Econometric model 5
VI Data collection 5
1 Data overview 5
2 Data description 5
VII Estimation of econometric model 6
1 Checking the correlation among variables 6
2 Regression run 8
VIII Check multicollinearity and heteroscedasticity 9
1 Multicollinearity 9
2 Heteroskedasticity 10
IX Hypothesis postulated 12
1 The impact of neighborhood factors 12
2 The impact of accessibility factors 13
X Result analysis & Policy implication 14
XI Conclusion 15
XII References 16
Exhibit 1: Definition of variables in the Housing Price model 4
Exhibit 2: Statistic indicators of variables in the Housing Price model 5
Exhibit 3: Correlation matrix 6
Exhibit 4: Scatterplot of variables in the Housing Price model 7
Exhibit 5: Regression model 8
Exhibit 6: Multicollinearity test 9
Exhibit 7: Heteroskedasticity test 10
Exhibit 8: Residual-versus-fitted plot of the Housing Price model 11
Exhibit 9: Correcting heteroskedasticity 11
Exhibit 10: Hypothesis testing of multiple regression model of neighborhood factors 12
Trang 3I Introduction
As much as Economy is a meaningful science that determines the social development in general and national growth in particular, Econometrics is the use of statistical techniques
to understand those issues and test theories Without evidence, economic theories are abstract and might have no bearing on reality (even if they are completely rigorous) Econometrics is a set of tools we can use to confront theory with real-world data
Given the data set, our group, which includes three members: Nguyen Ha Trang, Nguyen Mai Thuy Tien, and Nguyen Thi Lan Huong, follows the methodology of econometric comprising eight steps to analyze the data Note that because of the lack of information on the data set, all inferences of abbreviations and others are based on assumptions and self-research As a result, we hope to have shown clearly our logic and reasoning of analysis
To the extent of purpose and resources, there are still deficiencies in this report, but we look forward to providing readers with a decent view of the overall of the data set given and the knowledge that we have gained through Dr Dinh Thanh Binh’s Econometrics course
II Literature overview
1 Questions of interest
“Why do housing prices differ among locations and regions?” – this is the basic question to which this report targets to find the answer Although there is a variety of factors that might affect housing prices, they are divided into four main categories: structure, neighborhood, accessibility, and air pollution Consequently, elements that represent each of these categories are taken into account to find out whether they do, or at least statistically do have an impact
on housing prices
In following parts, models are going to be built, data are going to be used in order to run the regression model and then the results are going to be analyzed to finally answer the question
of interest above
2 Procedure and program used
Procedure
Step 1: Questions of interest
Step 2: Economic model
Step 3: Econometric model
Step 4: Data collection
Step 5: Estimation of econometric model
Step 6: Check multicollinearity and heteroscedasticity
Step 7: Hypothesis postulated
Step 8: Result analysis & Policy implication
Stata program is primarily used to analyze the data and run the regression
Trang 4III Economic model
As data are provided up front, the economic model used in this report is an empirical one Note that the fundamental model is mathematical; with an empirical model, however, data is gathered for the variables and using accepted statistical techniques, the data are used to provide estimates of the model's values
Empirical model discovery and theory evaluation are suggested to involve five key steps, but for the limitation of purpose and resources, this part of the report only follows three of them: (1) specifying the object for modeling, (2) defining the target for modeling, (3) embedding that target in a general unrestricted model
1 Specifying the object for modeling
As such, this report finds the relationship between housing price, which is the object for modeling, and each of relating factors including structure, neighborhood, accessibility, and air pollution ones
2 Defining the target for modeling by the choice of the variables to analyze, denoted
x i
As mentioned above, there are four main categories that are expected to affect housing prices: structure, neighborhood, accessibility, and air pollution Hence, the choices of x i would be
such variables that constitute them After thorough research, factors have been narrowed down to eight significant ones: (structure) number of rooms, (neighborhood) crimes, property tax, the percentage of people of low status, student-teacher ratio, (accessibility) distances to employment centers, accessibility to radial highways and (air pollution) nitrous oxide
3 Embedding that target in a general unrestricted model (GUM)
In its simplest acceptable representation (which will later be specified in the econometric model), the GUM of is determined to be:
lprice f crim, nox , rooms , dist , radial , proptax , stratio, lowstat
A brief description of each variable is given in Exhibit 1.
Exhibit 1: Definition of variables in the Housing Price model
lprice logarithm of median housing price, $
crime crimes committed per capita
nox nitrous oxide, parts per 100 million square
rooms average number of rooms per house
dist weighted distances to 5 employment centers
radial accessibility index to radial highways
proptax property tax per $1000
stratio average student-teacher ratio
lowstat percentage of people of low status
Trang 5IV Econometric model
To demonstrate the relationship between housing price and other factors, the regression
function can be constructed as follows:
(PRF):
lprice o crime nox rooms dist radial 6 proptax stratio lowstat i
(SRF):
lprice o crime nox rooms dist radial 6 proptax stratio lowstat i
where:
0 is the intercept of the regression model
i is the slope coefficient of the independent variable x i
is the disturbance of the regression model
0 is the estimator of 0
i
is the estimator of i
i
is the residual (the estimator of i )
From this model, this report is interested in explaining lprice in terms of each of the eight
independent variables ( crim, nox,rooms,dist,radio,proptax,stratio )
1 Data overview
This set of data is a secondary one, as they are collected from a given source
Data source: Regression Diagnostics: Identifying Influential Data and Sources of
Collinearity, by D.A Belsey, E Kuh, and R Welsch, 1990 New York: Wiley
The structure of Economic data: cross-sectional data
2 Data description
To get statistic indicators of the variables, in Stata, the following command is used:
sum lprice crime nox rooms dist radial proptax stratio lowstat
The result is shown in Exhibit 2.
Exhibit 2: Statistic indicators of variables in the Housing Price model
Trang 6Obs is the number of observations
Std Dev is the standard deviation of the variable
Min is the minimum value of the variable
Max is the maximum value of the variable
1 Checking the correlation among variables
First of all, the correlation of lprice and nox, rooms, dist, radial, proptax, stratio, lowstat is
checked by calculating the correlation coefficient among these variables The correlation
coefficient r measures the strength and direction of a linear relationship between two variables
on a scatterplot In Stata, the correlation matrix is generated with the command:
corr lprice crime nox rooms dist radial proptax stratio lowstat
The result is shown in Exhibit 3.
Exhibit 3: Correlation matrix
radial -0.4810 0.6254 0.6103 -0.2098 -0.4951 1.0000
proptax -0.5597 0.5828 0.6670 -0.2921 -0.5344 0.9102 1.0000
stratio -0.4976 0.2887 0.1869 -0.3540 -0.2293 0.4642 0.4542 1.0000
lowstat -0.7914 0.4470 0.5856 -0.6096 -0.4956 0.4760 0.5276 0.3654 1.0000
From the matrix, it can be inferred that the correlation between lprice and each of the
independent variable is decent enough to run the regression model Specifically:
- lprice and crime have a moderate downhill relationship
- lprice and nox have a moderate downhill relationship
- lprice and nox have a moderate uphill relationship
- lprice and dist have a weak uphill relationship
- lprice and radial have a moderate downhill relationship
- lprice and proptax have a moderate downhill relationship
- lprice and proptax have a moderate downhill relationship
- lprice and proptax have a strong downhill relationship
The correlation between each pair of variables can be visualized using the scatter
command in Stata
The result is shown in Exhibit 4.
Trang 7Exhibit 4: Scatterplot of variables in the Housing Price model
Trang 82 Regression run
Having checked the required condition of correlation among variables, the regression model
is ready to run In Stata, this is done by using the command:
reg lprice crime nox rooms dist radial proptax stratio lowstat
The result is shown in Exhibit 5.
Exhibit 5: Regression model
From the result, it can be inferred that
crime, nox, rooms, dist, radial, proptax, stratio and lowstat all have statistically significant effects on lprice at the 5% significant level (as all p-values are smaller than 0.05) In
particular, those effects can be specified by the regression coefficients as follows:
- 0 11.1951: When all the independent variables are zero, the expected value of housing
price is 1011.1951
- 0.0112
1
expected value
: When the number of crime committed per capita increases by one, the
of housing price decreases by 1.12%
- 20.0755
expected value
: When nitrous oxide increases by one part per 100 million square, the of housing price decreases by 7.55%
-
: When the number of rooms increases by one, the expected value of housing price decreases by 9.97%
-
: When the distance to 5 employment centers increases by one unit, the expected value of housing price decreases by 4.64%
Trang 9- 5 0.013 : When the accessibility index to radial highways increases by one unit, the expected value of housing price increases by 4.64%
- 6
0.0062 : When the property tax per $1000 increases by $1, the expected value of housing price decreases by 0.62%
- 70.0413 : When the student-teacher ratio increases by 1%, the expected value of housing price decreases by 4.13%
- 8
0.028 : When the percentage of people of lower status increases by 1%, the expected value of housing price decreases by 2.80%
The coefficient of determination Rsquared0.7669 : all independent variables (crime, nox, rooms, dist, radial, proptax, stratio, lowstat) jointly explain 76.69% of the variation
in the dependent variable (lprice); other factors that are not mentioned explain the remaining 23.31% of the variation in the lprice.
Other indicators:
- Adjusted coefficient of determination adj R-squared = 0.7631
- Total Sum of Squares TSS = 84.5822
- Explained Sum of Squares ESS = 64.8619
- Residual Sum of Squares RSS = 19.7203
- The degree of freedom of Model Df m = 8
- The degree of freedom of residual Dfr = 497
Based on the data collected from the table, the sample regression function is established:
SRF : lprice 11.2 0.01crime 0.08nox 0.1rooms 0.05dist 0.01radial 0.01proptax 0.03stratio 0.03lowstat
VII Check multicollinearity and heteroscedasticity
1 Multicollinearity
Multicollinearity is the high degree of correlation amongst the explanatory variables, which may make it difficult to separate out the effects of the individual regressors, standard errors may be overestimated and t-value depressed The problem of Multicollinearity can be detected by examining the correlation matrix of regressors and carry out auxiliary regressions
amongst them In Stata, the vif command is used, which stand for variance inflation factor.
Exhibit 6 shows the result.
Exhibit 6: Multicollinearity test
Trang 10The value of VIF here is lower than 10, indicating that Multicollinearity is not too worrisome
a problem for this set of data
2 Heteroskedasticity
Heteroskedasticity indicates that the variance of the error term is not constant, which makes the least squares results no longer efficient and t tests and F tests results may be misleading The problem of Heteroskedasticity can be detected by plotting the residuals against each of the regressors, most popularly the White’s test It can be remedied by respecifying the model – look for other missing variables In Stata, the imtest white command is used, which
stands for information matric test.
Exhibit 7 shows the result.
Exhibit 7: Heteroskedasticity test
imtest, white
White's test for Ho: homoskedasticity
against Ha: unrestricted heteroskedasticity
Prob > chi2 = 0.0000 Cameron & Trivedi's decomposition of IM-test
At the 5% significance level, there is enough evidence to reject the null hypothesis and conclude that this set of data meets the problem of Heteroskedasticity
Another way to test if Heteroskedasticity exists is to graph the residual-versus-fitted plot, which can be generated using the rvfplot, yline (0) line command in Stata
The result is shown in Exhibit 8.
Trang 11Exhibit 8: Residual-versus-fitted plot of the Housing Price model
In a well-fitted model, there should be no pattern to the residuals plotted against the fitted values - something not true of our model Ignoring the outliers at the top center of the graph,
we see curvature in the pattern of the residuals, suggesting a violation of the assumption that price is linear in our independent variables We might also have seen increasing or decreasing variation in the residuals— heteroskedasticity
To fix the problem, robust standard errors are used to relax the assumption that errors are both independent and identically distributed In Stata, regression is rerun with the robust option, using the command:
reg lprice crime nox rooms dist radial proptax stratio lowstat, robust
Exhibit 9 shows the result.
Exhibit 9: Correcting heteroskedasticity
Robust
Trang 12Note that comparing the results with the earlier regression, none of the coefficient estimates changed, but the standard errors and hence the t values are different, which gives reasonably more accurate p values
VIII Hypothesis postulated
1 The impact of neighborhood factors
The question of interest: In the multiple regression model:
lprice o crime nox rooms dist radial 6 proptax stratio lowstat
(full model) Does the subset of independent variables (crime, proptax, lowstat, stratio) contribute to
explaining/ predicting lprice? Or, would it do just as well if these variables were dropped and we reduced the model to
lprice o nox rooms dist radial
(reduced model).
From this question, the following hypothesis is postulated:
Null Hypothesis: The initial assumption is that the subset does not contribute to
the model's explanatory power Alternative Hypothesis: At least one of the independent variables in the subset is useful
in explaining/predicting lprice
which is expressed as:
H
o : 6
7 0
H : at least one
In Stata, the test statistic F is calculated using the command:
test crime proptax lowstat stratio
The result is shown in Exhibit 10.
Exhibit 10: Hypothesis testing of multiple regression model of neighborhood factors
( 1) crime = 0 ( 2) proptax = 0 ( 3) lowstat = 0 ( 4) stratio = 0
As a result, there is enough evidence to reject the null hypothesis and conclude that at least