DAO2702 Programming for Business Analytics

AY19/20 Semester 2

Team Project

Tutorial Group: A05

Group Name: Caramel
Group Members

Name Matriculation Number

Huang Xin A0201421E

Mariana Chen A0209040W

Wang Qian A0187619M

Ho Zheng Ting A0182688L

Sito Wynice A0191116M

Ke Yiting A0212559A
Wong Kai Zhe Russell A0211865E

1
Contents
1. Introduction

2. Business Problem – A Case for a Data-Driven Approach

3. Data Source and Feature Engineering

3.1 Overview of Original Dataset

3.2 Data Wrangling and Feature Engineering

4. Preliminary Findings

4.1 Statistical Distribution of Resale Prices

4.2 Geographical Distribution and Prices

4.3 Resale Price Trends Over Time

5. An Analysis on Selected Features

Key Attributes

5.1 Floor Area

5.2 Average Storey Level

5.3 Remaining Lease

5.4 Geographical Region

5.5 Flat Type

5.6 Flat Model

Other Attributes

5.7 Block Number

5.8 Distances to Nearest Key Facilities

6. Summary Statistics – Correlation Heatmap Of Attributes

7. Modelling Methodologies

7.1 Assumptions

7.2 Model Construction

7.3 Model Validation – Train-Test Splits and Cross Validation

8. Final Predictive Model – Interpretations

9. Concluding Remarks

10. References And Appendix

2
1. Introduction

HDB flats are the most common type of subsidised public housing in Singapore, currently accommodating an estimated
81 percent of Singapore’s population (HDB, 2019). The demand for resale flats has also experienced an upward trend,
with a 6.7% increase in applicants registered for resale flats from 2018 to 2019, and a further 17.6% increase in 2020
(Yong, 2020).

2. Business Problem – A Case for A Data-Driven Approach

Consider the case for a Real Estate Consulting Firm aiming to provide consulting services to those who intend to buy
or sell resale flats that maximise their benefits based on the valuation of current resale flat properties. HDB resale flats
have been an attractive market for real estate investors and agents, which has been seeing appreciating prices and getting
more attention from investors and agents (OrangeTee, 2019). Like most markets, the HDB resale flat market also sees
fluctuations in prices over time. To profit from market conditions, it is crucial to be able to accurately value properties
and decide if current market prices are suitable for entry or exit. For agents or homebuyers with intention of using real
estate as an investment, it is paramount to predict a fair value for their properties, which depend on a myriad of factors,
ranging from the region or district the flat is in, floor area, flat model, remaining lease period.

As a complex market, the widely available market data on HDB resale flats provides the perfect opportunity for a data-
driven approach to valuation of these real estate properties to provide a concrete, evidence-based metric to complement
investors and agents in their decision-making process, that is, to determine with current market conditions, if a property
is worth buying or selling. Here, we will explore a quantitative way to do so.

3. Data Source and Feature Engineering

3.1 Overview of Data Set

The preliminary dataset resale-flat-price.csv is freely available from data.gov.sg. This dataset includes key attributes
such month, resale price, floor area, flat type and model, remaining lease and the storey level from 2017 to 2020. There
is a total of 68161 data points, with initial variables “month”, “town”, “flat_type”, “block”, “street_name”,
“storey_range”, “floor_area_sqm”, “flat_model”, “lease_commencement_date”, “remaining_lease”, and one variable
“resale_price”, that will be our target variable.

3.2 Data Wrangling and Feature Engineering

We converted the string attribute “remaining_lease” to obtain a new numerical variable “months_remaining_lease”.
Next, we grouped the 26 different towns under “town” according to their geographical region (e.g. North, East etc) with
reference to the dataset “towns_data.csv” and reclassified the HDBs by their regions in “region”. The original data on
storey level is given in a range in string, thus we found the mean storey level (e.g. 10 TO 12 → 11) and stored the
information under “mean_storey_range”. Lastly, we extracted the numerical block numbers from “block” to obtain a
new variable “block_number” by removing the alphabets that follow the block numbers (e.g. 100B → 100). The updated
primary data source is stored in a new dataset file “processed-resale-flat-price.csv”.

However, these data attributes are insufficient for a robust predictive model as it does not include other important factors,
such as proximity to schools, malls or MRTs. To enhance the dataset, we included the proximity to the nearest MRT
station, malls, and schools for each data point. This required the coordinates of each HDB flat, which was retrieved
using OneMap.sg’s free geocoding Application Programming Interface (API). Using these coordinates, we queried for
the nearest MRT based on these coordinates using Google Places API. We also queried the coordinates on a
comprehensive list of malls obtained from Wikipedia, and school data from data.gov.sg. Subsequently, we used these
data to determine the nearest malls and schools and combined all our data into a final dataset, “finalDataSet.csv”.

3
4. Preliminary Findings

To derive a meaningful, robust model, a visualization on factors which may affect valuation must be understood. To
rigorously select these variables and understand their effects individually, a preliminary visualization will help to guide
our analysis.

4.1 Statistical Distribution of Prices

Using Seaborn’s violinplot,
important statistical
information can be visualized –
the median, upper and lower
quartiles, range and the kernel
density function (frequency) by
their resale prices. The Central
Region has the widest range of
prices while Bukit Timah has
the highest median resale price.
Choa Chu Kang also has the
most “centered” resale prices
slightly below $400,000

4.2 Geographical Distribution and Prices

Using Google Maps API, bokeh
and our queried geocoded
coordinates, we can visualize
how HDB resale flats are
distributed geographically, as
well as their relative prices,
marked by their intensity of
colour. Evidently, the more
expensive flats lie in the
Downtown Core, and perhaps
surprisingly, pockets of
expensive HDB flats are also
found in the Ang Mo Kio and
Jurong Region (See circled
points). Please refer to Notebook
3.2 for the interactive plot.
4.3 Resale Price Trends Over Time
Using Seaborn’s new lineplot
method, we can visualize price
trends over time, with its
confidence bands, over the
various regions. The Central
region maintains an exclusive
upper band of prices. The HDB
resale price of units in the East is
the 2nd most expensive on average
in Singapore, and units in the
North as valued the lowest. There
does not seem to be a clear
seasonality in terms of prices based on this visualization.

4
5. Analysis On Selected Features
A full comprehensive visualization can be found in our Notebook. We have selected a few key features based on our
visualizations which have exhibited clear relationships with resale prices, analysed below.

Key Attributes

5.1 Floor Area

Rationale: Widely known, on average, keeping all other factors
constant, the larger the floor area, the more expensive a property as one
would have to pay more to own more land.
Analysis: Using Seaborn’s jointplot, analysis shows a clear positive
linear relationship between floor area and price, pearsonr coefficient of
0.632. Hence, we expect floor area to be an important determinant in
valuating HDB resale flats. We also note from the distribution plots of
floor area that among the floor areas around 90 to 130 square metres,
there is a large range of resale prices, indicating the possibility that there
are other factors which may be significant factors to resale prices
beyond simply the floor area. Potential buyers could also use this
empirical data to decide if an offered price could be negotiated lower.
We also note the presence of 2 points with a floor area of around 250𝑚2 . To investigate this point, we take advantage
of bokeh’s HoverPlot (see Notebook 3.7). These points are revealed to be that of 65 Jln Ma’Mor and 41 Jln Bahagia,
part of HDB’s 1959 project of landed properties, accounting for its high floor area and resale price. Removing these 2
points, we get a new pearsonr coefficient of again 0.632, which still maintains that there is a strong relationship between
these 2 variables. Assuming a linear relationship between these 2 variables, the regression coefficient stands at 4003,
that is, every additional metre square cost $4003. Due to the importance of this factor on resale price, we will be
normalizing resale prices with their respective floor area, that is, their price per square metre, in subsequent analysis.
5.2 Average Storey Level
Rationale: Usually, higher floors are priced higher even by HDB
(HDB, n.d, 1). This could be due to a majority of homeowners
having a preference for better views from higher floors, better air
ventilation, or a increased sense of privacy from passer-by on the
ground level.
Analysis: This plot shows the relationship between the storey level
and the normalized resale price by floor area (price per 𝑚2 ). The
correlation coefficient is found to be 0.47 (see below), indicating a
moderate positive linear relationship between these 2 factors, though it is apparent that at least some of the difference
can be explained by its geographical region, as we observe the rough “bands” of colours in this strip plot.
5.3 Remaining Lease
Rationale: All HDB flats have a lease of 99 years (HDB, n.d, 2),
after which it is returned to HDB. Therefore, the longer the
remaining lease, the more we expect the unit to be valued. At the
same time, a longer lease implies that the unit is newer, and hence
is valued at a higher price. However, there is a conflicting factor of
floor area, that is, that HDB flats have been getting smaller over
time (Chan & Lim, 2019). This would entail paying less for a
smaller land area but paying a premium for a newer flat. This will
be accounted for in our model construction later.
Analysis: The relationship between resale prices and remaining
lease is relatively strong, with a correlation coefficient of 0.325.
After normalizing for floor area, the correlation coefficient drops to
0.258. This suggests that the remaining lease remains an important
factor for valuation, though the lease period may also share a

5
relationship with floor area (see Notebook 5.2). This may be due to financing options from banks or HDB become more
limited when lease becomes shorter (Wong, 2018). Further, flats with less than 60 years’ lease may start to depreciate
as prospective buyers are limited in using Central Provident Funds (CPF) to finance the purchase (Wong, 2018), which
may affect the demand for these resale flats and hence ability to command higher prices. Interestingly, we can also note
this effect from the frequency distribution of remaining lease displayed by the joint plot, with those with extremely low
remaining leases having low volume of transactions as compared to those with longer leases. There also seems to be a
large variation in price at every lease level, which could suggest that there are other factors which also contribute to its
price.

5.4 Geographical Region

Rationale: The location of a property is a key factor in its
value due to its proximity to town, reputation and status.
As a status-conscious society (Paulo & Low, 2018),
property is an important status symbol (Goh, 2005).
Hence, the more “expensive” a region is, the higher the
cost per square foot of the property.
Analysis: As expected, properties in the Central Region
command on average higher prices, but its distribution is
relatively even, that is, there is broad but almost equal
frequency of properties prices from $5K to $8K per 𝑚2 .
A buyer looking for cheaper housing would find most in
the North and West, with most of its values around $3K
to $4K per 𝑚2 . Properties in the East have more
expensive prices but exhibits a tighter spread of prices
compared to the North-East, which exhibits “fatter” tails.
Analysing the price per 𝑚2 over time, unlike the total
resale price examined above, properties in the North-East
seem to be more expensive on average compared to the
East. This may be attributed to the “fat-tails” of the
North-East distribution of prices. It is also likely that properties in the East have larger floor areas on average, accounting
for the higher overall resale price despite having lower per 𝑚2 price than the North-East.
5.5 Flat Type (Number of Rooms)
Rationale: We would expect that the higher the number
of rooms, the higher the resale price.
Analysis: It appears that this assumption seems to be
empirically true, with median resale prices increasing
over time, as well as their interquartile ranges. While this
may also be attributed to the increase in floor area, we
note that the marginal increase in floor area with every
increase in room number diminishes (HDB, n.d, 4).
Hence, we do observe some premium effect with the
increase in number of rooms.

6
 5.6 Flat Model
Rationale: HDB has been constantly upgrading its
designs to improve the perceived quality of its public
housing. Popular new initiatives such as Build-To-Order
aim to provide condominium-like finishings to HDB
homeowners at a fraction of the cost (Ang, 2019). We do
expect that more polished, luxurious designs and quality
be worth a higher price.
Analysis: Evidently, different flat models incorporate
different floor areas (HDB, n.d, 3). We hence look at the
price per 𝑚2 instead of the absolute prices to determine
the effect of flat model itself on prices (Visualization on
absolute prices found in Notebook 3.10). While most of the flat models have around the same median price per 𝑚2 ,
some exceptions are present such as DBSS models, Type S1, S2, and Terrace models. These units are often seen as
“premium units” (Ong, 2017). These flats are usually found in prime areas, and feature unique architectural features
(Choo, 2019), which hence command for the premium visualized here. Hence, the flat models itself, independent of its
floor area, may be an important metric in valuation.

Other Attributes
5.7 Block Number
Analysis: Surprisingly, block numbers played a non-
trivial effect on valuation prices even after accounting
for floor area through normalization, having a
correlation coefficient of -0.339. We note that smaller
block numbers are found disproportionally in the
Central Region, which may account for its higher price
per 𝑚2 . Interestingly, this is also due to HDB’s block
numbering policy, where the first digit in HDB's "3-
digit numbering system" denotes the neighbourhood
(Teo, n.d), while 2-digits usually denotes a Central
area. This may account for the variation in prices due to block number.

5.8 Distances To Nearest Key Facilities

Rationale: Proximity to key facilities such as to MRT stations, malls and schools implies higher convenience, which
would hence be expected to be worth a premium. In fact, MRT is a common mode of transport, with 60% of
Singaporeans taking MRT to work (Lee, 2016). It is therefore no surprise that on average, properties near MRT stations
are valued around 10-15% more than those further away from MRT stations (Navaratnarajah, 2015). We also expect a
property to be more expensive when it is nearer to a school since it is more convenient for families with children. This
would be especially true for parents with children seeking entry to primary schools, where their residential address must
be within a 1km radius to gain priority in balloting phase for a place in the primary school (MOE, n.d).

7
Analysis: Likewise, to remove the effect of floor area on resale prices, we normalized the prices by their floor area.While
our hypothesis held for the first 2 attributes (MRT and malls), with a correlation coefficient of -0.185 and -0.041
respectively, distance to schools showed a positive correlation (0.168). We note that all three exhibit a few points which
are much further away from all 3 facilities than most, which corresponed to properties on Changi Village Road. After
accounting for these, the correlation coefficients are rougly the same, at -0.185, -0.078 and 0.175 respectively. This
may be due to our dataset having only public school data and not private schools (e.g international schools) which may
be in the area. The correlation between schools and prices may also be attributed to the fact that properties near schools
have significantly higher noise pollution and congestion in the morning (Lagman, 2019), which may lead to lower prices
when units are closer to schools. We nonetheless see that these factors are also contributing attributes to the valuation
of HDB resale units.

6. Summary Statistics – Correlation Heatmap Of Attributes

After conducting visualizations on some key
expected vairables, we can summarize our
findings using Seaborn’s pairplot function
(see Notebook 4.1) and Pandas’ correlation
method to visualize quantitatively, the
relationships between attributes and our
target variable, that is, resale price, while also
analysing the matrix to identity potential
attributes which may pose a potential
multicollinearity problem in our subsequent
model construction.
Upon inspection, we observe that floor area,
remaining lease and storey level have higher
positive correlations to resale price, with
values 0.63, 0.32 and 0.37 respectively. These are expected as they are key primary determinants of property prices, as
explained above. Price per 𝑚2 shows the highest positive correlation to resale price at 0.69. Distances to MRT and block
number exhibit a weak correlation with resale price of -0.12. We also observe possible collinear attributes, some of
which are expected. For instance, lease_commence_date and months_remaining_lease has a perfect positive correlation,
as expected since the latter was derived from the former. The mean storey level and remaining lease also exhibit some
positive linear correlation, which is expected, since newer projects by HDB are usually built taller due to increasing
land constraints (Neo, 2020), as well as HDB’s recent endavours to built impressive high-rise buildings such as
Pinnacle@Duxton. Block numbers also show positive linear correlation to remaining lease due to HDB’s relatively
recent “3-digit numbering” system, which was only introduced in the 1970s (Teo, n.d).

7. Modelling Methodology

7.1 Assumptions
• Linearity of Model: Our models use a general linear form of 𝑌 = 𝑏0 + 𝑏1 𝑥1 + 𝑏2 𝑥2 + ⋯ with variables
transformed where appropriate to the model. We assume that the relationship between the attributes (transformed
or otherwise) and the target variable is linear.
• No Perfect Collinearity: This is seen from the correlation matrix that no attribute is perfectly correlated with any
other attribute (less derived attributes mentioned above which will be removed during model construction).
Checking for Multicollinearity
Multicollinearity may influence the interpretation of our coefficients
in the final model (Frost, n.d). We investigate the possibility of
multicollinearity in our model using Variance Influence Factor
(VIF). In practice, a VIF that exceeds 5 would be considered a
problematic level of collinearity. We observe that all our numerical
variables do not exhibit this level of collinearity. We will hence retain
all these variables in our preliminary model construction.

8
7.2 Model Construction
In the preliminary model, we used the baseline model:

𝒓𝒆𝒔𝒂𝒍𝒆 𝒑𝒓𝒊𝒄𝒆 = 𝑐𝑜𝑛𝑠𝑡 + 𝑠𝑡𝑟𝑒𝑒𝑡 𝑛𝑎𝑚𝑒 𝑝𝑟𝑒𝑚𝑖𝑢𝑚 + 𝑡𝑜𝑤𝑛 𝑝𝑟𝑒𝑚𝑖𝑢𝑚 + 𝑓𝑙𝑎𝑡 𝑚𝑜𝑑𝑒𝑙 𝑝𝑟𝑒𝑚𝑖𝑢𝑚
+ 𝑓𝑙𝑎𝑡 𝑡𝑦𝑝𝑒 𝑝𝑟𝑒𝑚𝑖𝑢𝑚 + 𝑚𝑜𝑛𝑡ℎ 𝑝𝑟𝑒𝑚𝑖𝑢𝑚 + 𝑟𝑒𝑔𝑖𝑜𝑛 𝑝𝑟𝑒𝑚𝑖𝑢𝑚 + 𝛽𝑠𝑡𝑜𝑟𝑒𝑦 𝑙𝑒𝑣𝑒𝑙 (𝑠𝑡𝑜𝑟𝑒𝑦 𝑙𝑒𝑣𝑒𝑙)
+ 𝛽𝑓𝑙𝑜𝑜𝑟 𝑎𝑟𝑒𝑎 (𝑓𝑙𝑜𝑜𝑟 𝑎𝑟𝑒𝑎) + 𝛽𝑟𝑒𝑚𝑎𝑖𝑛𝑖𝑛𝑔 𝑙𝑒𝑎𝑠𝑒 (𝑟𝑒𝑚𝑎𝑖𝑛𝑖𝑛𝑔 𝑙𝑒𝑎𝑠𝑒) + 𝛽𝑏𝑙𝑜𝑐𝑘 𝑛𝑢𝑚𝑏𝑒𝑟 (𝑏𝑙𝑜𝑐𝑘 𝑛𝑢𝑚𝑏𝑒𝑟)
+ 𝛽𝑑𝑖𝑠𝑡 𝑡𝑜 𝑚𝑟𝑡 (𝑑𝑖𝑠𝑡 𝑡𝑜 𝑀𝑅𝑇) + 𝛽𝑑𝑖𝑠𝑡 𝑡𝑜 𝑠𝑐ℎ𝑜𝑜𝑙 (𝑑𝑖𝑠𝑡 𝑡𝑜 𝑠𝑐ℎ𝑜𝑜𝑙) + 𝛽𝑑𝑖𝑠𝑡 𝑡𝑜 𝑚𝑎𝑙𝑙 (𝑑𝑖𝑠𝑡 𝑡𝑜 𝑚𝑎𝑙𝑙)

7.2.1 Transformations
In the initial model, using Ordinary Least Squares method, we note a good
fit of 𝐴𝑑𝑗𝑢𝑠𝑡𝑒𝑑 𝑅 2 = 0.935, and an AIC of 1635781.9548. We
conjectured that this may be due to the large range of resale prices. We
hence moved to transform the target variable.

After applying a logarithmic transformation on resale price, our preliminary

model’s fit improved with a new 𝐴𝑑𝑗𝑢𝑠𝑡𝑒𝑑 𝑅 2 = 0.947, with a new AIC
of -158116. We will hence use the natural log of resale prices in subsequent
steps of model construction.

7.2.2 Attribute Selection – Backward Stepwise

Regression
Using our current attribute set, there is a need to
prevent the possibility of overfitting, which may
occur when there are too many attributes such that
the learned hypothesis from the model may fit the
“training set” (here, all the data in our dataset) but cannot be generalized for uses outside of our dataset. We therefore
use a backward stepwise regression method to remove unnecessary variables. We can compare the relative degree of
over or underfitting by comparing the Akaike Information Criterion (AIC) between the candidate models during the
iteration of backward stepwise regression. The smaller the number, the lesser the degree of overfitting. Of course, the
𝐴𝑑𝑗𝑢𝑠𝑡𝑒𝑑 𝑅 2 should not be compromised. We hence used both criteria in attribute selection (see Notebook 6.3). This
results in a final model with a 𝐴𝑑𝑗𝑢𝑠𝑡𝑒𝑑 𝑅 2 = 0.947. We will use the resulting model in our subsequent validation.

7.3 Model Validation – Train-Test Splits and Cross Validation

We use scikit-learn packages to conduct train-test splits, that is, to

separate our dataset into “training data”, which we will feed the model
to “learn” , and use the “test data” to test how “accurate” the model is
able to predict resale prices. As with standard practice, we used a 75-
25% train-test split in this process. We also used cross validation to
ensure that the results of our model are not obtained by chance of a
good set of train-test data splits. This idea is best explained by this
diagram retrieved from scikit-learn.

Cross Validation Results

The final model after using the training data returned an
𝐴𝑑𝑗𝑢𝑠𝑡𝑒𝑑 𝑅 2 = 0.809712, and a
𝑅𝑜𝑜𝑡 𝑚𝑒𝑎𝑛 𝑠𝑞𝑢𝑎𝑟𝑒𝑑 𝑒𝑟𝑟𝑜𝑟 (𝑅𝑀𝑆𝐸) = 0.14211. However, we also
note that the final model included “street_name” as an attribute, which
could be too specific for potential users of our model. We attempted to
remove the attribute and retested the model using the Train-Test splits.
There was a slight improvement in the 𝐴𝑑𝑗𝑢𝑠𝑡𝑒𝑑 𝑅 2 = 0.809715 and
9
𝑅𝑀𝑆𝐸 = 0.14211. We hence have confidence in accepting this final model even if without the street name attribute,
that is, the remaining attributes are sufficient for a reliable valuation of resale prices. To further verify this, we used 5-
fold cross validation, which returned a similar 𝐴𝑑𝑗𝑢𝑠𝑡𝑒𝑑 𝑅 2 = 0.80952, 𝑅𝑀𝑆𝐸 = 0.14213. We can visualize the
difference between the predicted resale price, after removing the street name attribute, and the empirical data in this
scatter plot above, which shows a respectable predicted resale price compared to the empirical data from our dataset.

8. Final Predictive Model – Interpretations

𝒍𝒏(𝒓𝒆𝒔𝒂𝒍𝒆 𝒑𝒓𝒊𝒄𝒆) = 10.4072 + 𝑡𝑜𝑤𝑛 𝑝𝑟𝑒𝑚𝑖𝑢𝑚 + 𝑓𝑙𝑎𝑡 𝑚𝑜𝑑𝑒𝑙 𝑝𝑟𝑒𝑚𝑖𝑢𝑚 + 𝑓𝑙𝑎𝑡 𝑡𝑦𝑝𝑒 𝑝𝑟𝑒𝑚𝑖𝑢𝑚 + 𝑟𝑒𝑔𝑖𝑜𝑛 𝑝𝑟𝑒𝑚𝑖𝑢𝑚
+ 0.0082 ∗ 𝑠𝑡𝑜𝑟𝑒𝑦 𝑙𝑒𝑣𝑒𝑙 + 0.0081 ∗ 𝑓𝑙𝑜𝑜𝑟 𝑎𝑟𝑒𝑎 + 0.0009 ∗ 𝑚𝑜𝑛𝑡ℎ𝑠 𝑟𝑒𝑚𝑎𝑖𝑛𝑖𝑛𝑔 𝑙𝑒𝑎𝑠𝑒
−0.0001 ∗ (𝑏𝑙𝑜𝑐𝑘 𝑛𝑢𝑚𝑏𝑒𝑟) − 0.0008 ∗ (𝑚𝑜𝑛𝑡ℎ 𝑛𝑢𝑚𝑏𝑒𝑟) − 0.07(𝑑𝑖𝑠𝑡 𝑡𝑜 𝑀𝑅𝑇, 𝑘𝑚)
+ 0.0206(𝑑𝑖𝑠𝑡 𝑡𝑜 𝑠𝑐ℎ𝑜𝑜𝑙, 𝑘𝑚) − 0.0639 ∗ (𝑑𝑖𝑠𝑡 𝑡𝑜 𝑚𝑎𝑙𝑙, 𝑘𝑚)
Or approximately,
𝑹𝒆𝒔𝒂𝒍𝒆 𝒑𝒓𝒊𝒄𝒆 = −202196.1733 + 𝑡𝑜𝑤𝑛 𝑝𝑟𝑒𝑚𝑖𝑢𝑚 + 𝑓𝑙𝑎𝑡 𝑚𝑜𝑑𝑒𝑙 𝑝𝑟𝑒𝑚𝑖𝑢𝑚 + 𝑓𝑙𝑎𝑡 𝑡𝑦𝑝𝑒 𝑝𝑟𝑒𝑚𝑖𝑢𝑚 + 𝑟𝑒𝑔𝑖𝑜𝑛 𝑝𝑟𝑒𝑚𝑖𝑢𝑚
+ 4437.4498 ∗ 𝑠𝑡𝑜𝑟𝑒𝑦 𝑙𝑒𝑣𝑒𝑙 + 3726.6492 ∗ 𝑓𝑙𝑜𝑜𝑟 𝑎𝑟𝑒𝑎 + 403.3752 ∗ 𝑚𝑜𝑛𝑡ℎ𝑠 𝑟𝑒𝑚𝑎𝑖𝑛𝑖𝑛𝑔 𝑙𝑒𝑎𝑠𝑒
−51.0047 ∗ (𝑏𝑙𝑜𝑐𝑘 𝑛𝑢𝑚𝑏𝑒𝑟) − 253.9495 ∗ (𝑚𝑜𝑛𝑡ℎ 𝑛𝑢𝑚𝑏𝑒𝑟) − 32769.8752(𝑑𝑖𝑠𝑡 𝑡𝑜 𝑀𝑅𝑇, 𝑘𝑚)
−27985.1580 ∗ (𝑑𝑖𝑠𝑡 𝑡𝑜 𝑚𝑎𝑙𝑙, 𝑘𝑚) + 11665.3730(𝑑𝑖𝑠𝑡 𝑡𝑜 𝑠𝑐ℎ𝑜𝑜𝑙, 𝑘𝑚)
The premiums for each categorical variable can be found in the appendix.

From our validated model, we can estimate that each floor up is worth approximately $4400 and the average.
Convenience seems to be an important factor, with each km nearer to the MRT be worth around $32K and each km
nearer to the mall to be worth $27K. Some towns are also valued more highly than others. For instance, Bedok is worth
approximately a premium of approximately $10K (see appendix), while Bukit Batok enjoys a discount (or negative
premium) of 16K. This is as expected due to different regions and districts having different prices as explain in our
analysis above.

9. Concluding Remarks

While our model has tried, to the best of our ability, to incorporate salient factors of valuation, there may be other factors
which affect the valuation, such as amenities and features the estate offers (Benson et al., 1998). For instance, having a
scenic view amenity tends to add considerably residential property values, these includes amenity view to ocean, lake
and mountains (Benson et al., 1998). More locally, unquantifiable factors may also affect valuation figures. For instance,
Feng Shui could be a significant factor on residential prices (So, 2009). With a predominantly ethnically Chinese
population, these investors may be influenced by their beliefs and traditions, including Feng Shui. Studies proclaim that
bad Feng Shui may arise as a negotiating factor for buyers to negotiate for the final prices of units (So, 2009). Other
factors such as the quality or reputation of nearby primary schools may also affect resale prices of nearby units. However,
this data is not easily obtained. Existing interior designs by current owners and possible historical significance of certain
units may also affect the value of a unit.

These factors, though important, are at this stage, hard to incorporate. Nonetheless, the current model provides a
quantitative method to estimate a benchmark for an acceptable valuation to guide individual buyers or agents to make
their investment or purchase decisions.

As a Real Estate Consulting firm aiming to provide consulting services to those who seek to buy or sell resale flats at
the greatest worthiness in terms of its cost and price, it is imperative that we are able to predict reasonably, the value of
a unit before offering appropriate advice to clients. As with all predictive models, it is important for users to be wary
and cognizant of other peripheral factors affecting the value of real estate property. These include amenities and features
offered by the HDB, ‘Feng Shui’, neighbourhood characteristics as well as ever-changing geopolitical landscape and
changes in the Government's policies. We should also be cognizant of individual preferences which may cause
individuals to value a property differently. Concluding, we should use our predictive models as the beginning of the
valuation, not as an end.

10
10. References
Ang, R. (2019). New HDB flats to come with condo-like fittings. The Straits Times. Retrieved From:
https://www.straitstimes.com/singapore/housing/new-hdb-flats-to-come-with-condo-like-fittings
Chan, J. & Lim, J. (2019). 4 Questions About Owning A HDB Flat That You Probably Always Had In Your Mind.
Retrieved From: https://dollarsandsense.sg/4-questions-owning-hbd-probably-always-mind/
Choo, C. (2019). $1 Million HDB Flats: Here's What You Need To Know. Retrieved From: https://blog.seedly.sg/1-
millon-dollar-hdb-flats-singapore-heres-what-you-need-to-know/
Choo, C. (2019). HDB to build more new flats next year to meet greater demand. Channel News Asia. Retrieved
From: https://www.channelnewsasia.com/news/business/hdb-to-build-more-new-flats-next-year-to-meet-greater-
demand-12188602
Frost, J (n.d). Multicollinearity in Regression Analysis: Problems, Detection, and Solutions. Retrieved From:
https://statisticsbyjim.com/regression/multicollinearity-in-regression-analysis/
Goh, R.B.H (2005). Contours of Culture: Space and Social Difference in Singapore. HK: Hong Kong University
Press. Retrieved From:
https://books.google.com.sg/books?id=ZFD1GxJRx98C&pg=PA150&lpg=PA150&dq=property+is+a+status+symbol
+singapore&source=bl&ots=_HRkdB53e3&sig=ACfU3U32F0z9v08iuNBAXBchQIl8apURAg&hl=en&sa=X&ved=
2ahUKEwiSy4DNxpnpAhV8wTgGHUyXAXMQ6AEwGXoECAoQAQ#v=onepage&q=property%20is%20a%20stat
us%20symbol%20singapore&f=false
Hastie, T., James, G., Witten, D & Tibshirani, R. (2013). An Introduction to Statistical Learning with Applications in
R. UK: Springer
HDB (2019). Key Statistics, HDB Annual Report 2018/2019. Retrieved From:
https://services2.hdb.gov.sg/ebook/AR2019-keystats/html5/index.html?&locale=ENG&pn=9
HDB (n.d, 1) DIFFERING PRICES FOR BTO FLATS IN THE SAME TOWN. Retrieved From:
https://www.hdb.gov.sg/cs/infoweb/hdbspeaks/differing-prices-for-bto-flats-in-the-same-town
HDB (n.d, 2). Do buyers of HDB flats own their flat? Retrieved From:
(https://www.hdb.gov.sg/cs/infoweb/hdbspeaks/hdb-flat-buyers-own-their-flats)
HDB (n.d, 3). Types of Flats. Retrieved From: https://www.hdb.gov.sg/cs/infoweb/residential/buying-a-
flat/resale/types-of-flats
HDB (n.d, 4). Sales Launch. Retrieved From:
https://esales.hdb.gov.sg/bp25/launch/20feb/bto/20FEBBTOSB_page_0687/pricing.html
Lagman, M (2019). Dear Parents, Is Living Near Your Child’s School Really Worth It? Retrieved From:
https://www.99.co/blog/singapore/dear-parents-is-living-near-your-childs-school-really-worth-it/
Lee, P. (2016). More Singaporeans take bus, MRT to work: Government survey. The Straits Times. Retrieved from
https://www.straitstimes.com/singapore/more-singaporeans-take-bus-mrt-to-work-government-survey
MOE (n.d). Understand how balloting works. Retrieved From: https://beta.moe.gov.sg/primary/p1-
registration/understand-balloting/
Navaratnaragjah, R (2015). The MRT Effect: How It Will Affect Your Property's Value. Retrieved From:
https://www.propertyguru.com.sg/property-management-news/2015/2/85301/the-mrt-effect-how-it-will-affect-your-
propertys-value
Neo, X. (2020). Evolution of HDB designs. The Straits Times. Retrieved From:
https://www.straitstimes.com/singapore/housing/evolution-of-hdb-designs
Ong, K.S (2017). Why are DBSS flats achieving high prices in the HDB resale market? Retrieved From:
https://www.99.co/blog/singapore/dbss-flats-high-prices-hdb-resale/

11
Orange Tee (2020). HDB Market Pulse – Real Estate Data Trend Q4 2019. Retrieved From:
https://blog.orangetee.com/market-analysis-news/hdb-market-pulse-real-estate-data-trend-q4-2019/
Paulo, A.D & Low, M. (2018). Class – not race nor religion – is potentially Singapore's most divisive fault line.
Channel News Asia. Retrieved From: https://www.channelnewsasia.com/news/cnainsider/regardless-class-race-
religion-survey-singapore-income-divide-10774682
So, C. F. (2009). An examination of the Effect of Feng Shui on residential property price in HongKong, HongKong:
The University of HongKong.
Teo, A. (n.d). HDB Interesting facts and records. Retrieved From: https://www.teoalida.com/singapore/hdbrecords/
Wong, P. (2018). The Big Read: No easy answers to HDB lease decay issue, but public mindset has to change first.
Channel News Asia Retrieved from https://www.channelnewsasia.com/news/singapore/big-read-hdb-lease-decay-
public-mindset-change-homeownership-10361572
Yong, C. (2020). More HDB resale flats sold in March, prices down 0.3%. The Straits Times. Retrieved From:
https://www.straitstimes.com/singapore/more-hdb-resale-flats-sold-in-march-prices-down-03
Appendix
Ordinary Least Squares Regression Model Report, Target Variable: np.log(resale_price)
Results: Ordinary least squares
=========================================================================================
Model: OLS Adj. R-squared: 0.906
Dependent Variable: np.log(resale_price) AIC: -119379.4787
Date: 2020-05-08 03:20 BIC: -118849.9603
No. Observations: 68161 Log-Likelihood: 59748.
Df Model: 57 F-statistic: 1.147e+04
Df Residuals: 68103 Prob (F-statistic): 0.00
R-squared: 0.906 Scale: 0.010151
-----------------------------------------------------------------------------------------
Coef. Std.Err. t P>|t| [0.025 0.975]
-----------------------------------------------------------------------------------------
Intercept 10.4072 0.0684 152.1924 0.0000 10.2732 10.5412
C(town)[T.BEDOK] 0.2430 0.0021 116.9297 0.0000 0.2390 0.2471
C(town)[T.BISHAN] 0.9216 0.0062 148.6568 0.0000 0.9095 0.9338
C(town)[T.BUKIT BATOK] 0.0840 0.0020 41.0248 0.0000 0.0799 0.0880
C(town)[T.BUKIT MERAH] 0.9214 0.0059 157.0992 0.0000 0.9099 0.9329
C(town)[T.BUKIT PANJANG] -0.0098 0.0021 -4.7341 0.0000 -0.0138 -0.0057
C(town)[T.BUKIT TIMAH] 1.0710 0.0087 122.4992 0.0000 1.0539 1.0882
C(town)[T.CENTRAL AREA] 0.9725 0.0074 131.5329 0.0000 0.9580 0.9870
C(town)[T.CHOA CHU KANG] -0.1280 0.0020 -63.4738 0.0000 -0.1320 -0.1241
C(town)[T.CLEMENTI] 0.3522 0.0026 137.8339 0.0000 0.3472 0.3572
C(town)[T.GEYLANG] 0.7870 0.0061 129.1071 0.0000 0.7750 0.7989
C(town)[T.HOUGANG] -0.1596 0.0027 -58.9058 0.0000 -0.1649 -0.1543
C(town)[T.JURONG EAST] 0.1664 0.0025 65.5025 0.0000 0.1614 0.1713
C(town)[T.JURONG WEST] 0.0044 0.0017 2.6598 0.0078 0.0012 0.0077
C(town)[T.KALLANG/WHAMPOA] 0.8320 0.0060 138.5702 0.0000 0.8203 0.8438
C(town)[T.MARINE PARADE] 1.1631 0.0074 158.0910 0.0000 1.1487 1.1776
C(town)[T.PASIR RIS] 0.0682 0.0024 28.7987 0.0000 0.0636 0.0728
C(town)[T.PUNGGOL] -0.2355 0.0029 -80.7878 0.0000 -0.2412 -0.2298
C(town)[T.QUEENSTOWN] 0.9162 0.0061 150.9034 0.0000 0.9043 0.9281
C(town)[T.SEMBAWANG] 0.0235 0.0024 9.9964 0.0000 0.0189 0.0282
C(town)[T.SENGKANG] -0.3132 0.0027 -115.0341 0.0000 -0.3185 -0.3078
C(town)[T.SERANGOON] 0.0153 0.0034 4.5208 0.0000 0.0086 0.0219
C(town)[T.TAMPINES] 0.1880 0.0020 95.3023 0.0000 0.1842 0.1919
C(town)[T.TOA PAYOH] 0.7957 0.0059 133.7963 0.0000 0.7840 0.8073
C(town)[T.WOODLANDS] 0.0716 0.0019 37.2004 0.0000 0.0679 0.0754
C(town)[T.YISHUN] 0.2130 0.0020 108.1198 0.0000 0.2092 0.2169
C(flat_type)[T.2 ROOM] 0.1000 0.0189 5.3015 0.0000 0.0630 0.1370
C(flat_type)[T.3 ROOM] 0.2511 0.0187 13.4021 0.0000 0.2144 0.2878
C(flat_type)[T.4 ROOM] 0.3327 0.0191 17.4052 0.0000 0.2952 0.3701
C(flat_type)[T.5 ROOM] 0.3534 0.0197 17.9621 0.0000 0.3148 0.3919
C(flat_type)[T.EXECUTIVE] 0.3467 0.0204 16.9729 0.0000 0.3067 0.3867
C(flat_type)[T.MULTI-GENERATION] 0.2873 0.0381 7.5325 0.0000 0.2126 0.3621
C(flat_model)[T.Adjoined flat] 0.1527 0.0719 2.1234 0.0337 0.0118 0.2937
C(flat_model)[T.Apartment] 0.1330 0.0715 1.8601 0.0629 -0.0071 0.2731
C(flat_model)[T.DBSS] 0.2202 0.0714 3.0831 0.0020 0.0802 0.3602
C(flat_model)[T.Improved] 0.0632 0.0713 0.8860 0.3756 -0.0766 0.2031
C(flat_model)[T.Improved-Maisonette] 0.3812 0.0767 4.9722 0.0000 0.2309 0.5314
C(flat_model)[T.Maisonette] 0.1746 0.0715 2.4416 0.0146 0.0344 0.3147
C(flat_model)[T.Model A] 0.0684 0.0713 0.9585 0.3378 -0.0714 0.2082
C(flat_model)[T.Model A-Maisonette] 0.2419 0.0720 3.3595 0.0008 0.1008 0.3830
C(flat_model)[T.Model A2] 0.0609 0.0714 0.8534 0.3934 -0.0790 0.2009
12
C(flat_model)[T.Multi Generation] 0.2873 0.0381 7.5325 0.0000 0.2126 0.3621
C(flat_model)[T.New Generation] 0.0996 0.0714 1.3957 0.1628 -0.0403 0.2395
C(flat_model)[T.Premium Apartment] 0.0970 0.0713 1.3596 0.1740 -0.0428 0.2368
C(flat_model)[T.Premium Apartment Loft] 0.1888 0.0798 2.3672 0.0179 0.0325 0.3452
C(flat_model)[T.Premium Maisonette] 0.0536 0.0810 0.6620 0.5080 -0.1051 0.2124
C(flat_model)[T.Simplified] 0.0951 0.0714 1.3317 0.1830 -0.0448 0.2350
C(flat_model)[T.Standard] 0.0764 0.0714 1.0706 0.2843 -0.0635 0.2164
C(flat_model)[T.Terrace] 0.7305 0.0732 9.9825 0.0000 0.5871 0.8739
C(flat_model)[T.Type S1] 0.1439 0.0721 1.9945 0.0461 0.0025 0.2852
C(flat_model)[T.Type S2] 0.1324 0.0727 1.8217 0.0685 -0.0101 0.2748
C(region)[T.EAST] 0.4993 0.0043 115.9612 0.0000 0.4908 0.5077
C(region)[T.NORTH] 0.3082 0.0043 71.4226 0.0000 0.2998 0.3167
C(region)[T.NORTH-EAST] 0.7499 0.0059 126.1594 0.0000 0.7383 0.7616
C(region)[T.WEST] 0.4692 0.0049 96.0172 0.0000 0.4596 0.4788
floor_area_sqm 0.0081 0.0001 113.1330 0.0000 0.0080 0.0083
months_remaining_lease 0.0009 0.0000 182.5345 0.0000 0.0009 0.0009
mean_storey_range 0.0082 0.0001 109.2927 0.0000 0.0081 0.0084
block_number -0.0001 0.0000 -39.9598 0.0000 -0.0001 -0.0001
month_number -0.0008 0.0001 -6.6485 0.0000 -0.0010 -0.0005
dist_to_mrt -0.0700 0.0008 -87.0537 0.0000 -0.0716 -0.0685
dist_to_mall -0.0639 0.0013 -51.0544 0.0000 -0.0664 -0.0615
dist_to_sch 0.0206 0.0024 8.6123 0.0000 0.0159 0.0253
-----------------------------------------------------------------------------------------
Omnibus: 1224.409 Durbin-Watson: 1.195
Prob(Omnibus): 0.000 Jarque-Bera (JB): 2056.949
Skew: 0.160 Prob(JB): 0.000
Kurtosis: 3.788 Condition No.: 10189999488328168
=========================================================================================
* The condition number is large (1e+16). This might indicate strong
multicollinearity or other numerical problems.

Ordinary Least Squares Regression Model Report, Target Variable: resale_price

Results: Ordinary least squares
==========================================================================================================
Model: OLS Adj. R-squared: 0.884
Dependent Variable: resale_price AIC: 1674355.4701
Date: 2020-05-08 03:22 BIC: 1674884.9886
No. Observations: 68161 Log-Likelihood: -8.3712e+05
Df Model: 57 F-statistic: 9134.
Df Residuals: 68103 Prob (F-statistic): 0.00
R-squared: 0.884 Scale: 2.7257e+09
----------------------------------------------------------------------------------------------------------
Coef. Std.Err. t P>|t| [0.025 0.975]
----------------------------------------------------------------------------------------------------------
Intercept -202196.1733 35433.9362 -5.7063 0.0000 -271646.6465 -132745.7002
C(town)[T.BEDOK] 13620.6741 1077.0656 12.6461 0.0000 11509.6268 15731.7215
C(town)[T.BISHAN] 32671.2206 3212.4860 10.1701 0.0000 26374.7518 38967.6893
C(town)[T.BUKIT BATOK] -16907.6449 1060.4560 -15.9437 0.0000 -18986.1374 -14829.1523
C(town)[T.BUKIT MERAH] 25625.3302 3039.0358 8.4321 0.0000 19668.8237 31581.8368
C(town)[T.BUKIT PANJANG] -68863.5048 1068.5106 -64.4481 0.0000 -70957.7843 -66769.2254
C(town)[T.BUKIT TIMAH] 119390.0984 4530.4436 26.3528 0.0000 110510.4344 128269.7624
C(town)[T.CENTRAL AREA] 14018.2740 3831.2728 3.6589 0.0003 6508.9837 21527.5642
C(town)[T.CHOA CHU KANG] -119643.2264 1045.0592 -114.4846 0.0000 -121691.5412 -117594.9115
C(town)[T.CLEMENTI] 109487.8271 1324.1493 82.6854 0.0000 106892.4960 112083.1582
C(town)[T.GEYLANG] -39432.9943 3158.6369 -12.4842 0.0000 -45623.9188 -33242.0698
C(town)[T.HOUGANG] -79197.0279 1403.9957 -56.4083 0.0000 -81948.8577 -76445.1980
C(town)[T.JURONG EAST] 14791.7377 1316.0696 11.2393 0.0000 12212.2428 17371.2326
C(town)[T.JURONG WEST] -54806.4242 861.8989 -63.5880 0.0000 -56495.7450 -53117.1034
C(town)[T.KALLANG/WHAMPOA] -25402.3491 3111.3923 -8.1643 0.0000 -31500.6744 -19304.0239
C(town)[T.MARINE PARADE] 119576.6203 3812.4296 31.3649 0.0000 112104.2627 127048.9779
C(town)[T.PASIR RIS] -75610.3909 1227.0654 -61.6189 0.0000 -78015.4378 -73205.3441
C(town)[T.PUNGGOL] -125668.4287 1510.5328 -83.1948 0.0000 -128629.0712 -122707.7861
C(town)[T.QUEENSTOWN] 23472.9942 3146.2210 7.4607 0.0000 17306.4046 29639.5837
C(town)[T.SEMBAWANG] -97580.3668 1220.3497 -79.9610 0.0000 -99972.2508 -95188.4828
C(town)[T.SENGKANG] -162803.8325 1410.7059 -115.4059 0.0000 -165568.8145 -160038.8506
C(town)[T.SERANGOON] 1630.4105 1748.9870 0.9322 0.3512 -1797.6020 5058.4230
C(town)[T.TAMPINES] -20411.7010 1022.3783 -19.9649 0.0000 -22415.5613 -18407.8407
C(town)[T.TOA PAYOH] -28706.9889 3081.4785 -9.3160 0.0000 -34746.6831 -22667.2947
C(town)[T.WOODLANDS] -66057.1564 997.8192 -66.2015 0.0000 -68012.8808 -64101.4320
C(town)[T.YISHUN] -1592.8498 1021.0023 -1.5601 0.1187 -3594.0130 408.3134
C(flat_type)[T.2 ROOM] 3840.9663 9777.1295 0.3929 0.6944 -15322.1959 23004.1285
C(flat_type)[T.3 ROOM] 25495.6771 9709.2343 2.6259 0.0086 6465.5892 44525.7649
C(flat_type)[T.4 ROOM] 39282.8169 9903.8492 3.9664 0.0001 19871.2840 58694.3497
C(flat_type)[T.5 ROOM] 53030.0603 10194.0649 5.2021 0.0000 33049.7052 73010.4155
C(flat_type)[T.EXECUTIVE] 50899.8275 10584.2289 4.8090 0.0000 30154.7513 71644.9037
C(flat_type)[T.MULTI-GENERATION] 107669.0068 19764.7906 5.4475 0.0000 68930.0406 146407.9731
C(flat_model)[T.Adjoined flat] 81466.1652 37263.8556 2.1862 0.0288 8429.0522 154503.2783
C(flat_model)[T.Apartment] 60869.8084 37039.3376 1.6434 0.1003 -11727.2497 133466.8664
C(flat_model)[T.DBSS] 142741.2368 37008.4061 3.8570 0.0001 70204.8045 215277.6690

13
C(flat_model)[T.Improved] 16321.6452 36971.7402 0.4415 0.6589 -56142.9219 88786.2124
C(flat_model)[T.Improved-Maisonette] 187821.4947 39723.9074 4.7282 0.0000 109962.6831 265680.3063
C(flat_model)[T.Maisonette] 94247.0324 37047.2947 2.5440 0.0110 21634.3786 166859.6862
C(flat_model)[T.Model A] 18587.9631 36964.7010 0.5029 0.6151 -53862.8072 91038.7333
C(flat_model)[T.Model A-Maisonette] 138487.9449 37306.7499 3.7121 0.0002 65366.7591 211609.1306
C(flat_model)[T.Model A2] 26008.7809 37006.7469 0.7028 0.4822 -46524.3992 98541.9611
C(flat_model)[T.Multi Generation] 107669.0068 19764.7906 5.4475 0.0000 68930.0406 146407.9731
C(flat_model)[T.New Generation] 32903.6806 36976.1209 0.8899 0.3735 -39569.4727 105376.8340
C(flat_model)[T.Premium Apartment] 29803.4517 36968.8603 0.8062 0.4201 -42655.4708 102262.3742
C(flat_model)[T.Premium Apartment Loft] 176282.4400 41335.1534 4.2647 0.0000 95265.5882 257299.2918
C(flat_model)[T.Premium Maisonette] 71202.1659 41973.7185 1.6964 0.0898 -11066.2729 153470.6046
C(flat_model)[T.Simplified] 38034.7056 36986.9235 1.0283 0.3038 -34459.6207 110529.0320
C(flat_model)[T.Standard] 30234.8742 36992.0087 0.8173 0.4137 -42269.4193 102739.1676
C(flat_model)[T.Terrace] 373097.0005 37919.9502 9.8391 0.0000 298773.9429 447420.0580
C(flat_model)[T.Type S1] 188682.4312 37377.2069 5.0481 0.0000 115423.1499 261941.7126
C(flat_model)[T.Type S2] 232502.1731 37653.6828 6.1748 0.0000 158700.9993 306303.3469
C(region)[T.EAST] -82401.4178 2231.0386 -36.9341 0.0000 -86774.2508 -78028.5848
C(region)[T.NORTH] -165230.3730 2236.1107 -73.8919 0.0000 -169613.1474 -160847.5986
C(region)[T.NORTH-EAST] -59835.3523 3080.2548 -19.4255 0.0000 -65872.6480 -53798.0566
C(region)[T.WEST] -135941.2355 2532.0842 -53.6875 0.0000 -140904.1175 -130978.3534
floor_area_sqm 3726.6492 37.2567 100.0263 0.0000 3653.6262 3799.6723
months_remaining_lease 403.3752 2.4757 162.9317 0.0000 398.5227 408.2276
mean_storey_range 4437.4498 39.0690 113.5798 0.0000 4360.8746 4514.0250
block_number -51.0047 1.0572 -48.2432 0.0000 -53.0769 -48.9325
month_number -253.9495 58.8779 -4.3132 0.0000 -369.3501 -138.5488
dist_to_mrt -32769.8752 416.8423 -78.6146 0.0000 -33586.8855 -31952.8648
dist_to_mall -27985.1580 648.5906 -43.1476 0.0000 -29256.3948 -26713.9212
dist_to_sch 11665.3730 1238.4814 9.4191 0.0000 9237.9509 14092.7950
----------------------------------------------------------------------------------------------------------
Omnibus: 5053.864 Durbin-Watson: 1.105
Prob(Omnibus): 0.000 Jarque-Bera (JB): 7962.797
Skew: 0.586 Prob(JB): 0.000
Kurtosis: 4.195 Condition No.: 10189999488328168
==========================================================================================================
* The condition number is large (1e+16). This might indicate strong multicollinearity or other
numerical problems.

14