Calculating a constant quality price index for the stock of residential housing

Transcription

1 Calculating a constant quality price index for the stock of residential housing Huw Dixon (Cardiff Business School), Rhys Lewis (ONS), Tim Marshall (ONS), Kishan Rana (ONS) Acknowledgements: Many people helped us. We had very useful conversations with Yoel Finkel, Doron Sayag and Larisa Fleishman at the Israeli Central Bureau of Statistics in Jerusalem. At the Nationwide s Swindon headquarters, Robert Gardner and Andrew Harvey were most helpful, as were Martin Ellis and Nitesh Patel at the Halifax. Last but not least, Philippe Guiblin at the Valuation Office Agency and Tristan Carlyon at the National Housing Federation were able to guide us through some complex data issues. Faults remain our own.

2 The Need for a Stock-weighted House Price The current UK HPI is a measure of the level of house prices based on the flow of sales to private households. The houses sold in a particular year represent about 4.2% of the stock of housing. They are not a representative sample of the stock: some houses are more likely to be sold than others (due to house type, location etc.). Bernoulli probability by house characteristic. If all houses are equally likely to be sold, then B i is the same across types i (That is B i = B = 4.2% for all types i). We look at Bernoulli probabilities across three characteristics: English-Welsh LA areas, ACORN groups and house types. Clear frequency bias.

3 The Frequency Bias B is the aggregate Bernoulli probability, Bi the probability for characteristic i, the share of i in the population. The proportion of i in the flow of houses sold is then:

4 The Frequency Bias Local Authorities

5 The Frequency Bias ACORN Categories

6 The Stock-weighted House Price Index What we mean by the stock of housing: a. The stock of all housing (owner occupied, social housing and private rental); b. The stock of owner occupied housing. This paper focusses on the first, an index for all housing. However, I will also discuss how to construct a stock of owner occupied housing.

7 How is the current HPI constructed? Prices: 80, ,000 per month Characteristics: House type, Floor space, Location HM Land Registry Hedonic Regression Valuation Office Agency

8 How is the current HPI constructed? In effect, each characteristic of the house is given a value which contributes to the overall house price (the proportional contribution, since in logs). The HPI is a geometric mean. Because the regression is log-linear, the geometric mean depends only on the aggregate distribution of characteristics Q i, valued by their hedonic coefficients. HPI=exp[InHPI]

9 Weighting The weights for each characteristic Q i represent the proportions of houses having that characteristic in the transactions over a particular period. In the UK, the weights reflect the transactions in the previous calendar year and are updated annually. Because the transaction in a calendar year are not representative of the stock of housing (the frequency bias), for a SHPI we need to use different weights which reflect the distribution of characteristics across the whole stock of housing. We use hedonic coefficients from monthly HPI to value stock. Replace flow weights with stock weights.

10 Dwelling Stock Estimates by Tenure

11 Is it valid to use prices from HPI to value stock? HPI transactions are based on HM Land Registry Price Paid Data. This consists of transactions at full market price (excluding Right to Buy). Basically, mostly from owner occupied part of stock. The number of sales of social housing is small. In a total of 21,992 sales were recorded (2-3% of transactions in a year). However, whilst there are few sales of social housing, there are a lot of ex-social housing in owner-occupier and private rental stock: since 1980 two million social houses have been sold. Private rental sector is very diverse, but includes many houses that are similar to owner occupied.

12 Comparison of Stock with Flow Since the prices (coefficients on hedonic regression) are the same, the HPI and SHPI will only differ due to the differences in the weights. Weights capture the distribution of characteristics. Some examples. Flow: 2015, Stock: 2015.

13 Comparing weights House type

14 Comparing weights Local Authorities

15 Comparing weights London boroughs 0.70% 0.60% 0.50% 0.40% 0.30% 0.20% 0.10% 0.00% Stock Flow

16 Price Levels ( )

17 Growth Rates ( )

18 Main causes of differences: 2015 Levels:

19 Decomposition of differences 2015 Contributions ACORN: 45% Rooms: 18% Floor space: 16% House type: 13% Location: 8%

20 Comparing weights ACORN category

21 Comparing weights Habitable rooms

22 Comparing weights Floor space

23 Weights over time: House type

24 Weights over time: ACORN categories

25 Weights over time: Habitable rooms

26 Conclusions and recommendations It is possible to construct a SHPI using the same methodology as the HPI, by extending the number of addresses used from the set of all transactions to the set of all domestic addresses or random sample thereof. An owner-occupied SHPI can be constructed using the list of addresses from national housing surveys or LFS. More research on social housing in terms of housing values. Also the private rental sector. Both in conjunction with other organisations. Include full market price social housing sales in hedonic regressions for SHPI. The SHPI and beyond: I recommend ONS to construct a register of housing based on addresses to link together several datasets including the Census, EHS and other national housing surveys, monthly transactions data for HPI etc., LFS and no doubt others. Using the postal address as the key field you can link together a lot of data about households and houses and track how they evolve over time.