Overview

HDB resale flat prices started to increase rapidly in 2007 as demand for resale flats surged. Besides the introduction of property cooling measures such as raising the Minimum Occupation Period, ABC also increased the supply of BTO flats to channel housing demand away from the resale market. These measures seemed to be effective as HDB resale prices have been steadily declining since 2013.

As a data scientist in ABC, you have been tasked to evaluate the effectiveness of increasing BTO supply on reducing HDB resale prices. Specifically, you are required to develop a case study based on Queenstown, which experienced two major BTO launches (Skyterrace @ Dawson and Skyville @ Dawson) in December 2009.

  1. Start by providing a theoretical explanation of how a BTO launch might lower HDB resale prices in a town.

  2. Provide a clear explanation of your methodology. This should include: (1) listing down the set of variables needed, and (2) providing a clear explanation of your analytical strategy. Specifically, you should explain how your proposed approaches (e.g. statistical, econometric models and machine learning techniques) are able to attain a reliable estimate of the impact of the two BTO launches on HDB resale prices in Queenstown. Furthermore, you should highlight potential caveats and limitations of each proposed model.

Question 3a

Build to Order (BTO) is a Housing and Development Board (HDB) flat allocation system that allows eligible buyers to apply for a new HDB BTO flat from specific sites launched. The BTO system was introduced in 2001. It offers flexibility in timing and location when buying a new flat in Singapore. A tender for construction will only be called if the number of applicants is at least 50% of the number of flats in a specific launch.

Previously, the waiting period for a new flat under the old Registration for Flats System (RFS) was relatively short as flats were built ahead of demand. The new BTO system requires successful applicants to wait for about 3 to 4 years before moving in. Construction will only commence when the BTO project attains about 60% to 70% in sales. This helps to remove the need for the government to make projections ahead of demand.

For existing HDB resale flats, flat owners can sell their flats on the open market to any eligible buyer at a mutually agreed price. The resale price must be declared to the HDB even though it does not regulate these prices. Flat owners can only sell their flat once they have met the Minimum Occupation Period (MOP).

Essentially, the demand and supply for HDB flats can be mitigated by the demand and supply of new BTO flats and existing resale flats. When there is a high demand for HDB flats, prices for resale flats increase as there is a limited supply of new BTO flats. The launch of new BTO flats typically happens every quarterly.

Nevertheless, not every eligible buyer is looking to move in immediately or earlier. Hence, by increasing the supply of BTO flats with an increase in new launches, one can reduce the mid-term to long-term demand for HDB flats. This directly reduces the demand for resale flats, which helps to reduce their prices.

For eligible buyers in urgent need of housing, they can apply for unsold BTO flats online anytime, on a first-come first-served basis. Upon confirmation, they can book their flat by the next working day. They can also ballot for leftover BTO flats that have been reintroduced into the market through the Sales of Balance Flats (SBF) exercise. Otherwise, they will need to consider resale flats.

During the initial launch of BTO flats, the demand was low in the first few years. Projects that did not meet the minimum 70% application ratio were cancelled. Once the demand kicked in, HDB launched about 20 to 40 BTO projects annually. This resulted in a supply of 15,000 to 28,000 units annually, since 2010. By the end of 2015, HDB has launched 251 BTO projects, comprising a total of 168,453 units.

Question 3b

Our Methodology

For this Case Study, we will develop an Econometric Forecasting Model to provide ABC with a reliable estimate of the impact from two BTO launches on HDB resale prices in Queenstown. These two BTO launches were Skyterrace @ Dawson and Skyville @ Dawson in December 2009. The objective of our Forecasting Model is to better inform ABC and the market so as to cater for increasing housing demand.

  • Variables Needed:

    1. Timeline for Launches of New HDB Flats in Queenstown, from 1999 to 2009.
    2. Prices for Transactions of Resale HDB Flats in Queenstown, from 1999 to 2009.
    3. Prices for Transactions of Resale HDB Flats in Queenstown, from 2010 to 2014.

  • Analytical Strategy:

    For this Case Study, we will use a Vector Autoregression (VAR) Model as our Forecasting Model. It is an Econometric Forecasting Model used to identify underlying factors that might influence the variable that is being forecasted. After we built our Forecasting Model using the variables mentioned above, we will carry out tests for statistical quality control. One of them is a Cumulative Sum Control Chart (CUSUM) Test, which detects Structural Breaks in our model.

    In Econometrics, a Structural Break is an unexpected change over time in the parameters of regression models, which can lead to huge forecasting errors and reduced reliability of the model. Structural stability is the time-invariance of regression coefficients and a central issue in all applications of linear regression models. The lack of structural stability of coefficients frequently caused forecast failure and should be tested for routinely. The CUSUM Test is a sequential analysis technique that ensures our model is reliable and robust.

    Essentially, we will make forecasts of HDB resale prices in Queenstown from 2010 to 2014, using data on time series from 1999 to 2009. We will then use these projected HDB resale prices and compare them with the actual prices derived from 2010 to 2014. This allows us to make a comparison between projected and actual HDB resale prices in the next 5 years since the two BTO projects were launched in Queenstown, in December 2009.

Advantages and Disadvantages of Vector Autoregression (VAR) Models

  • Advantages:

    1. Commands needed to execute a VAR Model can be implemented in virtually any programming language.
    2. VAR Models use a relatively small number of variables and are easy to update as needed.

  • Disadvantages:

    1. Multicollinearity is where two or more explanatory variables in a Multiple Regression Model are highly linearly correlated. It is a problem for data on time series where variables are highly correlated with their previous values, along with past and present values from other time series. As the model expands, it can be difficult to separate the effects of explanatory variables.
    2. Parameter estimates can become highly sensitive to the combination of variables used in the model. A high degree of multicollinearity can make it difficult to determine which explanatory variables are significant. The standard errors of coefficient estimates can be large and will affect the accuracy of forecasts.
    3. as the number of variables in a VAR Model increases, the number of parameters to be estimated increases. This increases the number of coefficients relative to the amount of data available. Systematic relationships and random past events become increasingly reflected in the coefficients. Random future events that differ from these shocks in the past may result in less accurate forecasts.

Other Prediction Models

For this Case Study, we can also use Classification Prediction Models to make short-term forecasts. They include Random Decision Forests, Classification Tree, and Generalized Boosted Models. We can use the metric of Accuracy as our evaluation criterion. It is defined as the total number of correct classifications (true positives + true negatives) divided by the total number of documentations. It allows us to determine the degree of closeness of measurements derived from our predictions to their true values.

Random Decision Forests Prediction Model

For Random Decision Forests, there are no formal assumptions as random forests are non-parametric. This prediction model can handle skewed, multi-modal, and categorical data that are ordinal or non-ordinal.

  • Advantages: High levels of predictive accuracy. Only need a few control parameters. Strong in classification prediction. More resistant to overfitting of data. Trains rapidly with thousands of potential predictors. No need to select predictors for analysis. Can handle missing data. Provides estimates for the importance of predictors.

  • Disadvantages: Can overfit data with noisy classification or regression tasks. Classifications can be difficult to interpret as splits are not listed in results. can be biased in favour of predictors with more levels. Predictor importance scores may not be reliable.

Conclusion

In conclusion, we hope to develop an Econometric Forecasting Model to provide ABC with a reliable estimate of the impact from two BTO launches on HDB resale prices in Queenstown. Our model can provide forecasts of HDB resale prices for the next 5 to 10 years after the two BTO projects were launched in Queenstown, in December 2009. This allows us to make a comparison between projected and actual HDB resale prices since the two BTO projects were launched. Our Forecasting Model can be utilised to better inform ABC and the market so as to cater for increasing housing demand.