JAMAICA RESIDENTIAL

PROPERTY PRICE INDEX

METHODOLOGY
Application of the Hedonic Price Approach

Bank of Jamaica
July 2016
Table of Content

1. Overview……………………………………………………………………………………………………………………………..1

2. Data Description…………………………………………………………………………………………………………….......2

 Characteristics………………….…………………………………………………………………………………………2

 Data Preparation…………………………………………………………………………………………………………3

3. Methodology……………………………….………………………………………………………………………………….…..4

 Model Specification…………………………………………………………………………………………………….4

 Time Dummy Method – Rolling Window…………………………………………………………………….5

Appendix
1.0 Overview
This document summarizes the collaborative effort between the Bank of Jamaica (BOJ) and the
National Housing Trust (NHT) in the production of residential property price indices with the
intended purpose of measuring and monitoring changes in house prices in Jamaica over time.
Separate indices have been created that gauge movement in house prices island-wide and as well
for the major regional markets of St. Catherine, St. James and Kingston & St. Andrew.1 Table 1
summarizes the essential elements in the production of the indices.

Table 1: Summary Elements of the Jamaica Residential Property Price Index

Index Measure  Residential property
 Island-wide
 By region
Region 1. Kingston & St. Andrew
2. St. Catherine
3. St. James
Frequency  Quarterly

 Hedonic Price Econometric Approach

Method of calculation
- Rolling window time dummy method

 Sales price
Data  Property characteristics
 Geographic location

Figure 1 previews the time trend of the All-island residential property price indices over time. The
Figure presents two all-island indices derived using the time dummy hedonic pricing approach
and compares the time trend in each house price index to the time trend in the consumer price
index (CPI). Specifically, the indices were produced through the use of a rolling window time
dummy approach. Both are described in detail in section 3.0.

1
The house price indices derived by including the sample of all property price data will be referred to as the “All-
island Index”

Page | 1
 Figure 1: All-island Residential Property Price Indices for Jamaica
180.0

160.0
Index Points

140.0

120.0
2008Q4 = 100

100.0

Time Dummy: 8-Quarter CPI Time Dummy: 4-Quarter

Rolling Window Rolling Window

2.0 Data Description

A house is a heterogeneous good whose inherent characteristics such as floor area, number of
bedrooms, number of bathrooms, number of floors, existence of a garage, and geographical
factors including location and size of lot are linked to its market price. The hedonic estimation
approach utilized in the production of the house price index is underpinned by the assumption
that the market price of a housing unit depends on its physical properties and geographic
location. As such, the applied econometric procedure uses a list of characteristics that are known
to mainly influence the market value of a residential property.

Characteristics
 postal code
 lot size
 floor area
 year of construction
 type of dwelling
 number of floors
 number of bedrooms
 number of bathrooms
 number of laundry rooms
 number of car ports/garages
 existence of a water tank

Page | 2
Data Preparation
The NHT is a statutory body missioned with increasing the housing stock in Jamaica and providing
financial assistance to its contributors. The Trust is the largest provider of residential mortgages
in Jamaica accounting for over 50 percent of the total market share. The BOJ receives mortgage
transactions data from the NHT on a quarterly basis that include property sale price and the
above mentioned characteristics as at the loan disbursement date. In order to maintain
consistency and reliability of the estimated indices the data on property characteristics is cleaned
according to the following requirements:
Removals
1. Missing values - transactions with missing values in any of the property characteristics.
2. Outliers - transaction records with a sale price far exceeding the stated market value.2
3. Other seemingly erroneous data.
Adjustments
4. Range values - the midpoint value is used for characteristics provided with a range.
5. Half values - housing units recorded as having a ½ bath or ½ bedroom are adjusted to full.

Table 2: Sample totals after data scrubbing (2008Q4 – 2016Q1)

No. of
No. of Usable Data
Parish Observations
Observations (%)
after Scrubbing
Trelawny 1 031 899 87.2
St. James 1 196 1 056 88.3
St. Catherine 6 503 6 173 94.9
Kingston & St.
3 025 2 840 93.9
Andrew
Manchester 158 143 90.5
Clarendon 364 318 87.4
Portland 58 51 87.9
St. Thomas 120 90 75.0
St. Ann 233 217 93.1
Hanover 35 26 74.3
Westmoreland 116 92 79.3
St. Mary 132 122 92.4
St. Elizabeth 137 112 81.8
TOTAL 13 108 12 139 92.6

2
A sale price exceeding twice the market value was used as the benchmark for data removal.

Page | 2
3.0 Methodology
A hedonic pricing approach is employed by the BOJ to specify the econometric models estimated
for producing the house price indices.3 The approach is based on the premise that the price of a
house is determined by its internal and external characteristics. In the context of the housing
market, these characteristics refer to aspects of the physical structure as well as the location of
the property. The employed econometric models therefore provide price indices that control for
changes in the characteristics of the housing stock sold over time.4

Model Specification
To generate the house price index, it is assumed that the price 𝑝𝑛𝑡 of property n in period t is a
𝑡
function of a fixed number of k characteristics measured by ‘quantities’ 𝑧𝑛𝑘 . For T+1 time periods,
going from base period 0 to period T, price can be represented as a function:
𝑡 𝑡
𝑝𝑛𝑡 = 𝑓(𝑧𝑛1 , … , 𝑧𝑛𝐾 , 𝜀𝑛𝑡 ) (1)

Here 𝜀𝑛𝑡 is a random error term. While there are several possible functional forms for hedonic
specifications, selecting an appropriate functional form for the hedonic model is important for
minimizing any bias in the estimated coefficients and, by extension, the property price index. To
estimate the marginal contributions of the characteristics equation (1) is first specified as a
logarithmic-linear (semi-log) parametric model.

𝑡 𝑡
ln 𝑝𝑛𝑡 = 𝛽0𝑡 + ∑𝐾 𝑡
𝑘=1 𝛽𝑘 𝑧𝑛𝐾 + 𝜀𝑛 (2)
Equation (2) represents the base specification on which a time dummy rolling window method is
utilized to construct and update Jamaica’s property price index.5 Here 𝛽0𝑡 and 𝛽𝑘𝑡 are the
characteristic parameters to be estimated. The parameters 𝛽𝑘𝑡 are allowed to change over time

3
The methodology employed is largely in keeping with the Handbook on Residential Property Prices Indices (2013)
and mirrors the approach used to compute official housing price indices in other jurisdictions such as Ireland, France,
Germany, Austria, Norway, Finland and the United Kingdom.

4
Shadow prices of property characteristics cannot be independently observed. However, the demand and supply
for properties implicitly determine the characteristics’ marginal contribution to the prices of the properties. The
hedonic regression analysis values these marginal contributions.

5
All property characteristic variables, except for floor area, are included in each regression in dummy variable
format. The dummy variable coding for each characteristic is provided in Table 6 of the appendix. The appendix also
includes summary statistics for housing data over the period 2008Q4 to 2016Q1.

Page | 3
based on the premise that varying housing market conditions determine the marginal
contributions of these characteristics. Furthermore, when demand and supply conditions change
there is no a priori reason to expect that the contributions are constant.6

Time Dummy Method – Rolling Window

The time dummy method augments equation (2) by including a set of time dummy variables 𝐷𝜏 .

𝑡
ln 𝑝𝑛𝑡 = 𝛽0𝑡 + ∑𝐾 𝑇 𝜏 𝜏 𝑡
𝑘=1 𝛽𝑘 𝑧𝑛𝑘 + ∑𝑡=1 𝛿 𝐷 + 𝜀𝑛 (3)

Equation (3) is estimated using Ordinary Least Squares regression on a rolling sample of 4 and 8
quarters separately. The time dummies account for each quarter over the sample window.7 The
coefficients, 𝛿̂ 𝑡 , form the basis for estimating the price index using this approach. These
coefficients estimate the proportionate change in price arising from the progress in time, having
controlled for changes in the property characteristics.

First, an initial estimate of equation (3) is obtained by including all property characteristics 𝑧𝑘
over the sample window. Following this, a stepwise approach is used to obtain an estimate of
equation (3) that is derived from including only those property characteristics that are statistically
significant at the 90 percent confidence level. This stepwise procedure is repeated for each rolling
sample index estimate.

The index value for each quarter of the initial sample window, M0, is calculated as 100 × 𝑒𝑥𝑝(𝛿̂ ).
After which, the index value representing the last quarter of rolling window M t+1 is calculated
based on a linking approach to the index value of the last quarter of rolling window Mt. To do so,
the implied rate of price growth for the last two (2) periods of rolling window Mt+1 is calculated
𝑡
̂𝑀+1
𝑒𝑥𝑝(𝛿 )
as ̂ 𝑡−1 . This price growth is then applied to the last quarter index value of sample window
𝑒𝑥𝑝(𝛿𝑀+1 )

Mt. This is done in order to link the new index value to the previous period index value.

6
Pakes, A. (2003), “A Reconsideration of Hedonic Price Indexes with an Application to PCs”, American Economic
Review 93(5), 1576-93.

7
The initial quarter of each rolling sample is used as the reference dummy such that D takes a value of either 0 or 1
in all subsequent quarters.

Page | 4
The following illustrates the calculation mathematically:

̂𝑡
exp(𝛿 )
𝐻𝑃𝐼𝑡+1 = exp(𝛿̂𝑀+1
𝑡−1 ) × 𝐻𝑃𝐼𝑡 (4)
𝑀+1

The table below further illustrates the procedure using a four quarter rolling window as an
example.

Table 3: Demonstration of the 4 quarter rolling window approach

i) First 4 quarter rolling window - Q1 Q2 Q3 Q4 HPI Q5=

ii) Derived index - HPI Q1 HPI Q2 HPI Q3 HPI Q4 [exp(δ5)/exp(δ4)]*HPI Q4

iii) Second rolling window - Q2 Q3 Q4 Q5

---
iv) Used to derive the index for Q5 based δ3 δ4 δ5
on time dummy coefficient estimates δ.

This step is repeated for each subsequent period, dropping the data for the earliest quarter in
rolling window M and adding the data for the latest period in rolling window M+1. Two indices
are created following the approach outlined based on a four quarter and an eight quarter rolling
window. 8

8
The BOJ also produces property price indices using a hedonic imputation method as an alternative approach. The
index is a Laspeyres-type price index which obtains the predicted prices for a reference stock of property units. These
index values therefore reflect prices over time associated with a fixed set of housing characteristics. In the absence
of repeat sales data, this approach allows for the ability to gauge the price of a property sold in the reference year,
in all subsequent years. A geometric means approach is applied by taking the log of each predicted house value,
averaging over the sample reference stock, and then taking the exponent. Following this, the reference period stock
characteristics are valued using the estimated marginal contributions in the subsequent time periods. The index
values are derived by finding the change in the value of the reference period stock characteristics relative to the
subsequent periods. This method is not presented, however, due to its susceptibility to extreme swings in the value
of the index in years for which the number of observations of specific characteristics are small.

Page | 5
 Appendix

Data Summary
As at March 2016, data on residential mortgages issued by the NHT across Jamaica consisted of
13 108 observations, comprised of sale price and other primary characteristics. Summary
statistics are provided in the table below. The sample statistics for the final data set showed an
average price of $6.3 million for a housing unit with a standard deviation of $3.1 million and a
maximum of $40.9 million.9 Additionally, the average floor area was 102.6 square meters with a
minimum of 10.9 square meters and a maximum of 8 492.0 square meters. Furthermore, on
average, housing units had one floor, two and a half bedrooms, one and a half bathrooms and no
water tank.

Table 4: Summary statistics for final data set

Mean Standard Deviation Minimum Maximum
Purchase Price (J$000) 6 364.0 3 101.7 50.0 40 903.2
2
Floor Area (m ) 102.6 164.5 10.9 8 492.0
No. of Bathrooms 1.5 0.8 1.0 10.0
No. of Bedrooms 2.5 1.2 1.0 14.0
No. of Floors 1.1 0.3 1.0 4.0
No. of Laundry Areas 0.2 0.4 0.0 3.0
No. of Carports 0.2 0.4 0.0 4.0
No. of Water Tanks 0.0 0.2 0.0 4.0

The production of the price index utilizes Ordinary Least Squares regression estimates which are
subject to standard validity and robustness checks before the coefficients are included in the
compilation of the indices. One such assessment is the evaluation of multicollinearity of
regressors. Partial correlation statistics are provided in the table below that show no concern for
multicollinearity.

9
Given the NHT’s mandate of providing affordable housing for needy contributors, the sample data may not provide
full representation of the Jamaica housing market. However, considering the right tailed skew in the distribution of
the population Jamaican residential properties, the NHT sample will reflect the pricing trends inherent in the
majority of properties.

Page | 6
The number of bathrooms showed the highest correlation with price (32.4 per cent), followed by
the number of floors (24.4 per cent) and the number of carports (20.7 per cent). A strong positive
pair-wise correlation was found between the number of bedrooms and bathrooms (74.1 per
cent).

Table 5: Correlation of characteristics relative to prices

No. of
Purchase Floor No. of No. of No. of No. of
Laundry
Price Area Bathrooms Bedrooms Floors Carports
Areas
Floor Area 0.18
No. of Bathrooms 0.32 0.34
No. of Bedrooms 0.15 0.33 0.74
No. of Floors 0.24 0.13 0.27 0.17
No. of Laundry Areas 0.20 0.15 0.38 0.28 0.19
No. of Carports 0.21 0.20 0.38 0.38 0.03 0.23
No. of Water Tanks 0.09 0.08 0.12 0.09 0.07 0.07 0.08

As stated earlier, the hedonic characteristics are coded as dummy variables for use in the
regression analysis. Table 6.0 presents the codification of these variables and as well the
prevalence of each characteristic in the data set as at March 2016. Detached houses account for
64.0 per cent of total dwelling types. Semi-detached houses is the second most frequently
occurring dwelling type, accounting for 18.1 per cent. Other most frequently occurring
characteristics include: one floor (90.5 per cent); housing unit constructed between 2010 and
2015 (30.3 per cent); two bedrooms (54.7 per cent); one bathroom (65.2 per cent); no carport
(78.0 per cent); no laundry area (76.9 per cent); and no water tank (96.7 per cent). The parishes
of St. Catherine, Kingston & St. Andrew and St. James were the most frequently occurring
locations, accounting for 50.9 per cent, 23.4 per cent and 8.7 per cent, respectively.

Page | 7
 Table 6: Description of data format
Variables Description Proportion (%)
Type of Dwelling
detached Used as the reference characteristic 64.0
apartment = 1, if unit is an apartment; = 0, otherwise 7.8
attached = 1, if unit is attached; = 0, otherwise 5.4
semi-detached = 1, if unit is semi-detached; = 0, otherwise 18.1
townhouse = 1, if unit is a townhouse; = 0, otherwise 4.7
Number of Baths
1 Used as the reference characteristic 65.2
2 = 1, if unit has two bathrooms; = 0, otherwise 25.9
3 = 1, if unit has three bathrooms; = 0, otherwise 6.7
4 or more = 1, if unit has four+ bathrooms; = 0, otherwise 2.3
Number of Bedrooms
1 Used as the reference characteristic 12.1
2 = 1, if unit has two bedrooms; = 0, otherwise 54.7
3 = 1, if unit has three bedrooms; = 0, otherwise 19.0
4 = 1, if unit has four bedrooms; = 0, otherwise 8.2
5 or more = 1, if unit has five+ bedrooms; = 0, otherwise 6.0
Number of Laundry Areas
0 = 1, if unit has no laundry room; = 0, otherwise 76.9
1 Used as the reference characteristic 22.5
2 or more = 1, if unit has two laundry rooms; = 0, otherwise 0.5
Number of Water Tanks
0 Used as the reference characteristic 96.7
1 or more = 1, if unit has one or more water tanks; = 0, otherwise 3.3
Number of Carports
0 Used as the reference characteristic 78.0
1 = 1, if unit has one carport; = 0, otherwise 21.0
2 or more = 1, if unit has two or more carports; = 0, otherwise 0.9

Page | 8
 Table 6: Description of data format (cont’d)
Variables Description Proportion (%)
Year of Construction
1950 Used as the reference characteristic 1.0
= 1, if unit was constructed between 1960 & 1969; = 0,
1960 otherwise. (note: for any sample with no units built in 1950, 5.2
1960 is used as reference)
= 1, if unit was constructed between 1970 & 1979; = 0,
1970 15.2
otherwise
= 1, if unit was constructed between 1980 & 1989; = 0,
1980 14.1
otherwise
= 1, if unit was constructed between 1990 & 1999; = 0,
1990 17.6
otherwise
= 1, if unit was constructed between 2000 & 2009; = 0,
2000 16.6
otherwise
= 1, if unit was constructed between 2010 & 2015; = 0,
2010 30.3
otherwise
Number of Floors
1 Used as the reference characteristic 90.5
2 = 1, if unit has two floors; = 0, otherwise 9.3
3 or more = 1, if unit has three or more floors; = 0, otherwise 0.3
Location (Parishes)
Kingston & St.
Used as the reference characteristic 23.4
Andrew
Clarendon = 1, if unit is located in Clarendon; = 0, otherwise 2.6
Hanover = 1, if unit is located in Hanover; = 0, otherwise 0.2
Manchester = 1, if unit is located in Manchester; = 0, otherwise 1.2
Portland = 1, if unit is located in Portland; = 0, otherwise 0.4
St. Ann = 1, if unit is located in St. Ann; = 0, otherwise 1.8

St. Catherine = 1, if unit is located in St. Catherine; = 0, otherwise 50.9

St. Elizabeth = 1, if unit is located in St. Elizabeth; = 0, otherwise 0.9
St. James = 1, if unit is located in St. St. James; = 0, otherwise 8.7
St. Mary = 1, if unit is located in St. Mary; = 0, otherwise 1.0
St. Thomas = 1, if unit is located in St. Thomas; = 0, otherwise 0.7
Westmoreland = 1, if unit is located in Westmoreland; = 0, otherwise 0.8
Trelawny = 1, if unit is located in Trelawny; = 0, otherwise 7.4
Other Variables
Purchase price purchase price for housing unit
Floor area surface area of a housing unit in square meters

Page | 9