Nationwide Vs Land Registry

[TheCountOfNowhere]

On one hand we have the land registry index of ACTUAL SALES has fallen 8 months in a row for many areas and on the other hand we have the Nationwide index of mortgage approvals increasing for the last 5 months.

Why the difference ?

The ACTUAL sales figures give me more confidence than some government back B.S. who has a vested interest in not seeing house prices collapse, am I just being silly.

[Milton]

The Land Registry do not include the sales of repossessed properties to be included in their data, used to determine the average house price.

They have no satisfactory reason why they do not.

The Land Registry are therefore NOT an impartial body.

They are a vested Interest.

At every level the game is rigged to keep you in debt slavery.

Edited July 29, 2011 by Milton

[TheCountOfNowhere]

The Land Registry do not include the sales of repossessed properties to be included in their data, used to determine the average house price.

They have no satisfactory reason why they do not.

The Land Registry are therefore NOT an impartial body.

They are a vested Interest.

At every level the game is rigged to keep you in debt slavery.

That goes without saying. if the nationwide included loans to buy repossessed houses their index would be halved. The question is why the difference between the indexes, not which is most "fixed"

[neil324]

A house i know sold at auction last Sept fpr £75K, falling down nearly. Back on the market yesterday for £170K, done up to a very high standard.

It will sell, but most likely the vendor will need to drop his price some what. The first sale will not be in any indexes. The second will.

[TheCountOfNowhere]

A house i know sold at auction last Sept fpr £75K, falling down nearly. Back on the market yesterday for £170K, done up to a very high standard.

It will sell, but most likely the vendor will need to drop his price some what. The first sale will not be in any indexes. The second will.

How does that help explain the divergence of the Nationwide and Land Registry indexes ?

[Pent Up]

The only real way to judge what prices are doing is not make notes of properties that go STC and then look up the prices they made.

I'm seeing (in Essex) that there are many 2007 priced houses which very rarely sell. Then there are the cheaper ones, who obviously actually want to sell, not too many but of these but they are normal around 2007 - 10% these ones sell quite often for big chunks off asking. Then you repos which get offers for anything back to 2003 prices.

One I've been watching last sold for £217,500 in 06. Land registry data puts it peak value around £235k. Came on the Market last year at £217k. Which is not bad considering asking prices around here. Didn't sell and about three months reduced to £205k went SSTC about two months later. Then came available again at £199k. if I liked it more I would have put a low bid on that, £170k or so. Unfortunately these are quite rare.

[Statistician]

On one hand we have the land registry index of ACTUAL SALES has fallen 8 months in a row for many areas and on the other hand we have the Nationwide index of mortgage approvals increasing for the last 5 months.

Why the difference ?

The ACTUAL sales figures give me more confidence than some government back B.S. who has a vested interest in not seeing house prices collapse, am I just being silly.

Go here to Radio 4 Prog Moneybox on the UK Housing scam

http://news.bbc.co.uk/1/hi/programmes/moneybox/9538647.stm

[zebbedee]

The only real way to judge what prices are doing is not make notes of properties that go STC and then look up the prices they made.

I'm seeing (in Essex) that there are many 2007 priced houses which very rarely sell. Then there are the cheaper ones, who obviously actually want to sell, not too many but of these but they are normal around 2007 - 10% these ones sell quite often for big chunks off asking. Then you repos which get offers for anything back to 2003 prices.

One I've been watching last sold for £217,500 in 06. Land registry data puts it peak value around £235k. Came on the Market last year at £217k. Which is not bad considering asking prices around here. Didn't sell and about three months reduced to £205k went SSTC about two months later. Then came available again at £199k. if I liked it more I would have put a low bid on that, £170k or so. Unfortunately these are quite rare.

You could use propertybee to obtain the list of sold and the price at which it was last listed (at least if theres someone like me who runs a macro daily on the same search in the area) then cross reference with zoopla/RM data. I'll have to have a look when I get home.

[TheCountOfNowhere]

SO basically, no one has any ideas why the Nationwide indexes are diverging so much.

Is everyone on holiday ?

[bendy]

That goes without saying. if the nationwide included loans to buy repossessed houses their index would be halved. The question is why the difference between the indexes, not which is most "fixed"

Do repossessed 'sales' go to a loan though. I get the idea most are CB's?

I'm on the fence whether including them is the right or wrong thing to do - it may wake up the masses that they are asking stoopid prices yet I think the repossessions would still be unobtainable for those only able with a mortgage.

[Pent Up]

Do repossessed 'sales' go to a loan though. I get the idea most are CB's?

I'm on the fence whether including them is the right or wrong thing to do - it may wake up the masses that they are asking stoopid prices yet I think the repossessions would still be unobtainable for those only able with a mortgage.

You don't need to be a cash buyer. Anyone can buy them. It would be advisable to have mortgage, solicitors and everything ready to go before getting an offer accepted though because some insist on a 28 day exchange.

[sarahleyburn]

SO basically, no one has any ideas why the Nationwide indexes are diverging so much.

Is everyone on holiday ?

Dunno, but happy to think aloud.

The obvious answer is that 1) Nationwide apparently only accounts for 7% of the market and low transaction volumes on such a small sample render it fairly meaningless 2) it has a southern bias 3) the cash buyers who now make up 40% of the market may be driving harder bargains than people taking out mortgages, who already feel pretty smug about their rates.

It could also be that large numbers of chains are collapsing / prices are being negotiated down after lenders' surveyors have done their valuations. The sales that actually go through may be the more 'realistically' priced ones, which ends up being reflected in Land Reg figures but not Haliwide ones.

But, IIRC, in 2008, the Haliwide indices diverged from the LandReg in the opposite way - showing big falls when the Land Reg didn't. That may have reflected the fact that bottom end properties were falling much faster than top end ones (the average house price of a Haliwide customer is much less than the national average, IIRC) so maybe now we're seeing heavier falls at the top end of the market and more stability at the bottom? But that wouldn't explain why Halifax and the Land Reg are diverging less than Nationwide and the Land Reg (I think) - the only thing I can see that explains that is the North / South divide.

In short: I dunno

[Monkey]

does the nationwide figures include remortgages and MEWing. as its bassed on mortgages.

so its not just house purchases/sales but refinancing of that house as well at the end of the mortgage deal

[Pent Up]

does the nationwide figures include remortgages and MEWing. as its bassed on mortgages.

so its not just house purchases/sales but refinancing of that house as well at the end of the mortgage deal

No.

[NEO72]

Dunno, but happy to think aloud.

The obvious answer is that 1) Nationwide apparently only accounts for 7% of the market and low transaction volumes on such a small sample render it fairly meaningless 2) it has a southern bias 3) the cash buyers who now make up 40% of the market may be driving harder bargains than people taking out mortgages, who already feel pretty smug about their rates.

It could also be that large numbers of chains are collapsing / prices are being negotiated down after lenders' surveyors have done their valuations. The sales that actually go through may be the more 'realistically' priced ones, which ends up being reflected in Land Reg figures but not Haliwide ones.

But, IIRC, in 2008, the Haliwide indices diverged from the LandReg in the opposite way - showing big falls when the Land Reg didn't. That may have reflected the fact that bottom end properties were falling much faster than top end ones (the average house price of a Haliwide customer is much less than the national average, IIRC) so maybe now we're seeing heavier falls at the top end of the market and more stability at the bottom? But that wouldn't explain why Halifax and the Land Reg are diverging less than Nationwide and the Land Reg (I think) - the only thing I can see that explains that is the North / South divide.

In short: I dunno

What she says. They have all tracked each other pretty well in the past but now we have this apparent divergence between London and elsewhere, my guess is that NW does have a southern bias (although they claim not to) - when London was moving in line with the rest of the country this bias would have effectively been 'hidden'.

[The Knimbies who say No]

This is Free Trader's domain, but a couple of thoughts:

The methodology of the indices is likely different, and of course Nationwide's is approvals and given the distinction it is safe to say that approval does not equal sale. Land Reg uses a 'repeat sales regression' - ie looking for houses that have sold before, so I guess that excludes any new builds from appearing. Nationwide's methodology is not really clear to me, but obviously they have a sample bias to start with as it is obviously only people who approach Nationwide and pass Nationwide's current criteria in relation to the applicant(s), the property and the mortgage. Seems like a lot to account for- someone may have been refused this month who would have been approved last month etc etc. A real pickle to account for, and they may not take it too seriously. Then of course there's the mix adjustment to iron out bias in the sample, I guess that weights by type of property, but again how much detail to the go into? Does a three-bed semi count the same as another three-bed semi, even if one has a much larger garden, or a driveway, or is in Yorkshire instead of Devon?

There are a multitude of statistical sins hidden within the indices and the methodologies used will themselves introduce bias and systematic errors which likely dwarf the odd sub-percentage point month-on-month change.

I don't think that it is inconsistent to have modest rises in Nationwide/falls in the land Reg- seems like there are too many variables to properly account for month to month. If one was showing strong trends and the other was not, that would be different, but in this low volume situation I think the indices are likely to be at their least accurate.

The bottom line in statistics is: "The empirical basis for scientific truth is x +/- y"

The indices rarely publish the 'y' part and as such it is impossible to take them seriously. As someone with a bit of stats training, I'd be surprised if the total errors were not of the order of a few percentage points, so it is perfectly possible that the indices are consistent with each other even if the central figures appear to be moving away from each other.

Of course, it doesn't make for a good press release to say "Nationwide index for July suggests that there is a 68% chance that the average house price is in the range £163,546 +/- £5,391". The public and media cannot handle statistical uncertainties and as such they are simply not provided. But it makes the figure meaningless without them.

Edited July 29, 2011 by cheeznbreed

[TheCountOfNowhere]

This is Free Trader's domain, but a couple of thoughts:

The methodology of the indices is likely different, and of course Nationwide's is approvals and given the distinction it is safe to say that approval does not equal sale. Land Reg uses a 'repeat sales regression' - ie looking for houses that have sold before, so I guess that excludes any new builds from appearing. Nationwide's methodology is not really clear to me, but obviously they have a sample bias to start with as it is obviously only people who approach Nationwide and pass Nationwide's current criteria in relation to the applicant(s), the property and the mortgage. Seems like a lot to account for- someone may have been refused this month who would have been approved last month etc etc. A real pickle to account for, and they may not take it too seriously. Then of course there's the mix adjustment to iron out bias in the sample, I guess that weights by type of property, but again how much detail to the go into? Does a three-bed semi count the same as another three-bed semi, even if one has a much larger garden, or a driveway, or is in Yorkshire instead of Devon?

There are a multitude of statistical sins hidden within the indices and the methodologies used will themselves introduce bias and systematic errors which likely dwarf the odd sub-percentage point month-on-month change.

I don't think that it is inconsistent to have modest rises in Nationwide/falls in the land Reg- seems like there are too many variables to properly account for month to month. If one was showing strong trends and the other was not, that would be different, but in this low volume situation I think the indices are likely to be at their least accurate.

The bottom line in statistics is: "The empirical basis for scientific truth is x +/- y"

The indices rarely publish the 'y' part and as such it is impossible to take them seriously. As someone with a bit of stats training, I'd be surprised if the total errors were not of the order of a few percentage points, so it is perfectly possible that the indices are consistent with each other even if the central figures appear to be moving away from each other.

Of course, it doesn't make for a good press release to say "Nationwide index for July suggests that there is a 68% chance that the average house price is in the range £163,546 +/- £5,391". The public and media cannot handle statistical uncertainties and as such they are simply not provided. But it makes the figure meaningless without them.

So, what you are saying is...the nationwide is a load of boll-ocks ?

Edited July 29, 2011 by TheCountOfNowhere

[rantnrave]

So, what you are saying is...the nationwide is a load of boll-ocks ?

Only when it's rising...

[The Knimbies who say No]

So, what you are saying is...the nationwide is a load of boll-ocks ?

pretty much

Although so is the land reg, but for different reasons.

The main point is that it is hard to have confidence when they are not open about the method and uncertainties.

[koala_bear]

Good post, I'm not doing any more actual stats this week, (taught it to Ugrad engineers will doing PhD) but:

When I had a good look at Nw vs Hx ( I have certainly had a few discussions with FT on the subject on a couple of occasions) I came to the conclusion I trusted Hx more for transparency as is referenced back to their 1983 approvals mix factors included location, home type, bedroom, garage (y/n), central heating (y/n) etc. The real mix will have changed radically since '83 but there is a solid benchmark for comparsion. It is hedonical adjusted to take account of the improvement in housing quality with time (i.e. more homes have central heating than 1983 and it regresses that out). Hx is similar to the CPI hedonic adjustment in that respect.

Nw on the other hand the mix adjustment changes gradually with time and is effectively much less hedonically adjusted as improvement in housing quality just results in price increases... Similar to RPI calculation in that respect.

Hence if only the best quality stuff is selling (i.e. non repo and gets past a NW survey (and in the SE!!!) prices could easily be going up as Nw won't fish out the improvement in "quality".

Also have certain ranges to exclude properties if they are outside their definition i.e. a detached house can only be over 400sqft otherwise it is excluded etc.

NW CEO quoted their market share as 7% in Dec'10 so assuming the CML latest figures of 41,500 mortgages per month that leaves NW with 2905 mortgages a month as there are 140 categories that a home can fit in (assuming none excluded which a good chunk are by NW admission) then that is an average of 20.75 homes per category per month. For accuracy purposes I would prefer 40+ per category or the errors bounds start to get horrific when every thing is added together. There will obviously be variation in this average some categories with lower numbers homes (i.e. 3) and some higher (i.e. 37) so unless they are v careful things could get easily skewed. Or if they have an exclusion if the number of homes per category goes under 10 for example (to maintain accuracy) large areas of the country or home types could be excluded to ensure accuracy is maintained...

Will post Nw and Hx methodology in a separate post.

This is Free Trader's domain, but a couple of thoughts:

The methodology of the indices is likely different, and of course Nationwide's is approvals and given the distinction it is safe to say that approval does not equal sale. Land Reg uses a 'repeat sales regression' - ie looking for houses that have sold before, so I guess that excludes any new builds from appearing. Nationwide's methodology is not really clear to me, but obviously they have a sample bias to start with as it is obviously only people who approach Nationwide and pass Nationwide's current criteria in relation to the applicant(s), the property and the mortgage. Seems like a lot to account for- someone may have been refused this month who would have been approved last month etc etc. A real pickle to account for, and they may not take it too seriously. Then of course there's the mix adjustment to iron out bias in the sample, I guess that weights by type of property, but again how much detail to the go into? Does a three-bed semi count the same as another three-bed semi, even if one has a much larger garden, or a driveway, or is in Yorkshire instead of Devon?

There are a multitude of statistical sins hidden within the indices and the methodologies used will themselves introduce bias and systematic errors which likely dwarf the odd sub-percentage point month-on-month change.

I don't think that it is inconsistent to have modest rises in Nationwide/falls in the land Reg- seems like there are too many variables to properly account for month to month. If one was showing strong trends and the other was not, that would be different, but in this low volume situation I think the indices are likely to be at their least accurate.

The bottom line in statistics is: "The empirical basis for scientific truth is x +/- y"

The indices rarely publish the 'y' part and as such it is impossible to take them seriously. As someone with a bit of stats training, I'd be surprised if the total errors were not of the order of a few percentage points, so it is perfectly possible that the indices are consistent with each other even if the central figures appear to be moving away from each other.

Of course, it doesn't make for a good press release to say "Nationwide index for July suggests that there is a 68% chance that the average house price is in the range £163,546 +/- £5,391". The public and media cannot handle statistical uncertainties and as such they are simply not provided. But it makes the figure meaningless without them.

[koala_bear]

Nationwide methodology:

http://www.nationwide.co.uk/hpi/Nationwide_HPI_Methodology.pdf'>http://www.nationwide.co.uk/hpi/Nationwide_HPI_Methodology.pdf

The text apologies as table and graphs will be a bit weird or missing:

Methodology

Introduction

There are several methods that could be used to calculate the trend in house prices, ranging from a simple average of purchase price to a statistical method of averaging. Then there is the matter of making sure that the different mixture of properties sold in each month does not give a false impression of the actual change in house prices. The next few sections explain the way we do this as well as providing some background to the Nationwide house price series and the current methodology that we employ to calculate average house prices.

Background to Nationwide House Price Information

Nationwide Building Society has a long history of recording and analysing house price data and has published average house price information since 1952. The following provides a short chronology of publish series and developments in Nationwide’s methodology of calculating average house prices:

•

1952 – Annual publication of house price data

•

1974 – Quarterly data is published for the first time

•

1989 – Development of new house price methodology. A statistical ‘regression’ technique was introduced under guidance of ‘Fleming and Nellis’ (Loughborough University and Cranfield Institute of Technology)

•

1993 – The house price system was further improved following publication of the Census 1991 data. Frequency for UK series increased to monthly.

The monthly figure measures the mix adjusted average house price for all houses in the UK. Every quarter the Nationwide also publishes a more detailed breakdown of house prices. These include both UK and 13 regional estimates for:

•

4 types of property (Detached houses, semi-detached houses, terraced houses and flats/apartments)

•

2 types of buyer (First time buyer and Former owner occupiers)

•

3 property ages (New, modern and old)

This makes a total of 140 separate series, all of which are published quarterly on our internet site http://www.nationwide.co.uk/hpi/

Data - source, cleaning and sample size

Source

All house price information is derived using Nationwide mortgage data. This data is extracted monthly for mortgages that are at the approvals stage and after the corresponding valuation report has been completed. Approvals data is used as opposed to mortgage completions since it should give an earlier indication of current trends in prices in the housing market.

Cleaning

Nationwide house price series utilise only owner occupied property information. In addition, properties that are not typical and may distort the series are also removed from the data set. Therefore, the following criteria are used to select which properties to include:

•

House purchases - remortgages and further advances are excluded

•

Owner occupied properties – buy to let properties are excluded

•

Properties sold at true market prices - right to buy sales at discounted price are excluded

•

Floor size has to be within specified limits for a give type of property - e.g. a detached house has to have at least 400 sq ft floor area

Sample Size

The number of cases that are used to calculate the average price for a given month will depend on the volume of monthly mortgage activity and out of these the cases that meet the criteria in the cleaning process. The monthly sample size will therefore vary from month to month.

Nationwide has sufficient sample size to produce a representative house price series. N.B. Net lending figures quoted at our half yearly and annual results are not a guide to our sample size. Sample size is based on the number of new loans we write i.e. the amount of gross lending for house purchase (remortgage cases are excluded).

The Nationwide Building Society is currently the 3rd largest mortgage lender in the UK. Our share of the gross house purchase market has averaged c10% over the last 3 years. This allows us to be confident that the series based on Nationwide mortgage data is representative of the whole house market. The quarterly UK series for all houses uses 3 months of data and hence a much larger sample than at the month. The samples sizes for the other quarterly series will depend on what it is they are measuring, for example the series for first time buyers only considers properties being brought by first time buyers and hence this will have a smaller sample size than that used for the whole of the UK. It is for this reason that detailed breakdown of house prices are produced quarterly.

Mix Adjustment Process

Why Mix Adjust?

The purpose of mix adjustment is to simply isolate pure prices changes. The simple example below illustrates how the changes in the mixture of properties sold each month could give a misleading picture of what is actually happening to house prices. The set of properties sold from month to month will vary by location and design etc. and some adjustment is necessary to make sure all of these do not give a false impression of the actual changes to house prices. A mix-adjusted or ‘standardised’ index is not affected by such changes because the relative weight given to each characteristic of a property in the ‘mix’ (or ‘basket’, to use an analogy with retail prices) is fixed from one period to the next.

Simple example - Benefits of mix adjustment

050,000 100,000 150,000 200,000 250,000 300,000 350,000 P1P2P3P4P5Price of FlatsPrice of DetachedSimple AverageMix Adjusted AverageMix adjustment example

Suppose that the price of both detached houses and flats increased at the same rate for five periods, with flats being cheaper.

Further suppose that the proportion of flats and detached sold in each period varied considerably, as the table below shows.

Time period

P1

P2

P3

P4

P5

% Flats

50

30

70

30

70

% Detached

50

70

30

70

30

The simple average of both kinds of properties will be influenced by the proportion of each property sold. In periods 3 and 4 the simple average shows a decrease, whereas the actual prices of both increased!

The mix adjusted average uses a consistent measure of the proportion of each type of property and is able to better reflect the true change in prices.

The mix-adjusted price represents the price for an average or ‘typical’ house. This should not be confused with the average price of all houses. The latter is usually higher because even though there are fewer more expensive houses sold, their price is such that they bias the simple average to be greater than the price of the typical house.

Calculating the price of a typical house

Calculation

The price of a property will depend on the characteristics of the property. These characteristics could include physical properties of the house, like its design, but other aspects such as the type of neighbourhood the house is located in will also contribute to the price someone is willing to pay. Using mortgage data, the Nationwide house price system can relate all the observed combinations of these factors and relate them to the price of which the house was sold for. From this, the model can estimate how much on average a house would cost given a set values for these characteristics, in particular a set of characteristics that describes the ‘typical’ house. This typical house does not physically exist, it is an ‘average’ house across all the characteristics that the model uses. This method is repeated on data sets at different points in time and changes in the price of this typical house reflect only the price changes over the same time periods, and not the mixture of properties sold in the current or pervious periods.

Factors that affect the price of a house

The following are the items that are used to describe the characteristics of a property. There is no set order that these contribute most to the price of the house, although UK location, the type of neighbourhood and house size are consistently the three most important followed by the design of the house.

•

UK Location, i.e. part of country.

•

Type of neighbourhood. The Nationwide index uses an established demographic system that classifies areas in the UK into 54 categories based on the type of people that live there; two examples include retirement and council areas.

•

Floor size

•

Property design (detached house, semi-detached house, terraced house, bungalow, flat, etc.)

•

Tenure (freehold/leasehold/feudal) except for flats, which are nearly all leasehold

•

Number of bathrooms (1 or more than 1)

•

Whether property has central heating or not

•

Type of garage (single garage, double garage or none)

•

Number of bedrooms (1, 2, 3, 4 or more than 4)

•

Whether property is new or not

Seasonal Adjustment

House prices are slightly seasonal - that is, prices are higher at certain times of year irrespective of the overall trend. This tends to be in spring and summer, when more buyers are in the market and hence sellers do not need to discount prices so heavily, in order to achieve a sale. The effect on prices over the year is of the order of +/- 1.5%; however this is much smaller than the change in volume of property transactions. The seasonal effect is estimated each month using established statistical methods.

For the monthly house price index where changes can be as little as 0.1%, seasonal factors are important. The Nationwide therefore produce a seasonally adjusted series for UK house prices which seeks to remove this effect so that the overall trend in prices is more readily apparent.

Seasonal adjustment shows that July is generally the strongest month for house prices (raw prices are 1.2% above their SA level) and February is the weakest (raw prices are 1.5% below their SA level). Nationwide House Price Index Methodology

[South Lorne]

On one hand we have the land registry index of ACTUAL SALES has fallen 8 months in a row for many areas and on the other hand we have the Nationwide index of mortgage approvals increasing for the last 5 months.

Why the difference ?

The ACTUAL sales figures give me more confidence than some government back B.S. who has a vested interest in not seeing house prices collapse, am I just being silly.

...mortgage approvals and house sales are not the same animal ...the lead in to a sale could take many months or years while an approval

which may never be used is more instant...

[koala_bear]

How does that help explain the divergence of the Nationwide and Land Registry indexes ?

NW say that LR is higher because the price of high end homes is excluded form the NW index as this creates a long tail that is unrepresentative and skews the average (similar to the difference between the mean and median...)

Edited July 29, 2011 by koala_bear

[koala_bear]

Halifax methodology download from link on this page:

The Halifax House Price Index

Background

The Halifax House Price Index is the UK's longest running monthly house price series covering the whole country from January 1983. The UK Index is derived from the mortgage data of the country's largest mortgage lender, which provides a robust and representative sample of the entire UK market.

There are a number of national indices covering different categories of houses (all, new and existing) and buyers (all, first-time buyers and home-movers). These indices are also adjusted to allow for seasonal variations. The most commonly used and quoted Halifax Index is the UK seasonally adjusted index covering all houses and all buyers. Regional indices for the 12 standard planning regions of the UK are produced on a quarterly basis.Methodology

The indices calculated are 'standardised' and represent the price of a typically transacted house. The need for 'standardisation' arises because no two houses are identical and may differ according to a variety of characteristics relating to the physical attributes of the houses themselves or to their locations.

In summary, prices are disaggregated into their constituent parts using a commonly used statistical technique called multivariate regression analysis. This allows values to be attributed to the various qualitative characteristics (type of property, region, etc.) and quantitative characteristics (age of property, number of habitable rooms, garages, bathrooms, etc.) of a property.

As a result, the technique allows us to track the value of a 'typical' house over time on a like-for-like basis (i.e. with the same characteristics). This prevents the possibility of short-term changes in the set of properties sold from month to month (for example, shifts in the regional complexion of the market or a change towards more large properties being sold) giving a misleading impression of the change in the price of a 'typical' house.

Analyses of house prices based on simple arithmetic average prices (as, for example, is the case with the Land Registry) do not compare like-for-like.

Data

The Halifax House Price Indices are derived from information on the following house characteristics:

• Purchase price.
• Location (region).
• Type of property: house, sub-classified according to whether detached, semi-detached or terraced, bungalow, flat.
• Age of the property.
• Tenure: freehold, leasehold, feudal.
• Number of rooms: habitable rooms, bedrooms, living-rooms, bathrooms.
• Number of separate toilets.
• Central heating: none, full, partial.
• Number of garages and garage spaces.
• Garden.
• Land area if greater than one acre.
• Road charge liability.

Although one hundred per cent coverage of all house purchase transactions financed by the Halifax is obtained, those transactions that do not constitute a fully consistent body of data for the purpose of house price analysis are excluded from the Indices. These exclusions primarily cover property sales that are not for private occupation and those that are likely to have been sold at prices which may not represent 'free' or 'normal' market prices, for example, most council house sales, sales to sitting tenants, etc. Only mortgages to finance house purchase are included; remortgages and further advances are excluded.

The data refer to mortgage transactions at the time they are approved rather than when they are completed. Whilst this may cover some cases which may never proceed to completion, it has the important advantage that the price information is more up-to-date as an indicator of price movements and is on a more consistent time-base than completions data (such as the ODPM Index) given the variable time lags between approval and completion. The monthly indices cover transactions during the full calendar month and the regional quarterly indices cover transactions over the entire quarter.

Properties over £1 million have been included since December 2002 to reflect the increasing number of this hitherto tiny market segment.

Seasonality

Houses prices are seasonal with prices varying during the course of the year irrespective of the underlying trend in price movements. For example, prices tend to be higher in the spring and summer months when more people are looking to buy. We therefore produce seasonally adjusted series to remove this effect and to allow us to concentrate on the underlying trend in house prices. These seasonal factors are updated monthly.

The Halifax House Price Index is prepared from information that we believe is collated with care, but we do not make any statement as to its accuracy or completeness. We reserve the right to vary our methodology and to edit or discontinue the indices at any time for regulatory or other reasons. Persons seeking to place reliance on the indices for their own or third party commercial purposes do so at their own risk. Full Technical Details are available free on request - contact the Housing Economics Help-line on 0845 604 5404.

Edited July 29, 2011 by koala_bear

[koala_bear]

original technical paper from 1983 download from link in post above

THE HALIFAX

HOUSE PRICE INDEX

Technical Details

Professor M C Fleming M.A., Ph.D

And

Professor J G Nellis, B.Sc(Econ), M.A. Ph.D

Logo

Background

When first introduced in 1984, the Halifax House Price Index represented a major advance in the measurement of house price changes throughout the country. Unlike earlier series, and house price statistics produced by other institutions, the new figures issued by the Halifax were standardised rather than based on simple price averages. Given the variety of houses in the United Kingdom, simple averages are not comparable: they do not compare like with like. However, by allowing for the influence of the different characteristics of houses on their prices, using a database especially established by the Halifax for this purpose, the new series placed the measures on a truly comparable footing, thereby providing a more accurate indication of house price movements than was previously possible.

The research work on which the new series of Halifax price index numbers was based was carried out by Professor M C Fleming and Professor J G Nellis. This paper provides a short technical description of the methodology, and presents the results for that first year: 1983.

For further information contact:

Martin Ellis

Halifax, Trinity Road, Halifax HX1 2RG

Telephone 01422 333916

1. Introduction

This paper, describing the background to the development of the standardised indices of house prices launched by the Halifax in April 1984, provides a short account of the methodology and presents the series for 1983. The methodology is applied to produce a number of standardised indices covering different categories of houses (all, new and existing) and of buyers (first-time buyers and former owner-occupiers). Future results are to be published in regular monthly bulletins for the United Kingdom and quarterly bulletins giving regional analyses.

The need for "standardisation" arises out of the fact that no two houses are alike: they may differ according to a variety of quantitative and qualitative characteristics relating to the physical attributes of the houses themselves or to their locations. Thus analyses of average house price differences between one region and another, or of changes in average prices over time, are not based on the comparison of like with like if the "characteristics-mix" of houses traded is not standardised. Until recently, little or no attention was paid to this problem: both official and unofficial analyses were based on simple averages of prices and were thus beset by the problem of non-comparability. A review of the problems and of the series available up to 1981 is to be found in Fleming and Nellis (1981).

The problem of comparability cannot be tackled without information about the characteristic, as well as the price, of each house sold. Given the great variety of combinations of characteristics possessed by houses, and given also a desire to measure their influence at regional, as well as at national, levels, it is necessary to establish a data-capturing system large enough to provide representative coverage of all house transactions in each region of the UK. As the country's leading lender it will be appreciated that the Halifax is in an ideal position to obtain large-scale representative data. A comparative study of the Halifax and other building societies*, carried out by the authors, has confirmed that the Halifax data does provide information on price movements that is representative of all transactions financed by building societies in the UK.

*This paper was written prior to the Halifax's conversion to a public limited company in1997. Any references to the Halifax as a building society should now be regarded as the "Company".

The data themselves are described fully in the next section. This is followed in Section 3 by an explanation of the methodology and its application. Section 4 reports the monthly and quarterly results for the different categories of houses and of buyers in 1983.

2. The data on house characteristics

The database established by the Halifax from the beginning of 1983 has two notable merits from the point of view of this study. First, the size of the database is exceptionally large because the number of house-purchase transactions financed by the Halifax each month is very large and all of these are covered in the statistical reporting system. As a consequence, the analytical procedures followed in this study permit much more reliable estimation than would otherwise be the case. Secondly, the scope of the data collected about house characteristics is more extensive than anything available hitherto in the United Kingdom. This again helps to improve the reliability of the statistical analyses. Incidentally, these two facts also mean that the database is larger and more detailed than that obtained in the official Five Per Cent Sample Survey of Building Society Movements conducted by the Department of Environment.

Information is obtained about the following house characteristics:

• Purchase price

• Location (region)

• Type of property: house, sub-classified according to whether detached, semi-detached or terraced, bungalow, flat

• Age

• Tenure: freehold, leasehold, feudal

• Number of rooms: habitable rooms, bedrooms, living-rooms, bathrooms

• Number of separate toilets

• Central heating: none, full, partial

• Number of garages and garage spaces

• Garden

• Land area if greater than one acre

• Road charge liability

The use of the information on the characteristics defined above in the analyses undertaken here is explained in the next section dealing with the methodology and its application.

Although one hundred per cent coverage of all house purchase transactions financed by the Halifax is obtained, they are not all allowed to enter the analyses because they do not constitute a fully consistent body of data for the purpose of house price analysis. Therefore, certain properties are excluded, namely those which are not for private occupation and those that are likely to have been sold at prices which may not represent "free" or "normal" market prices, for example, council house sales, sales to sitting tenants, etc. After editing in this way, the database covers around 12,000 house purchase transactions per month.

A final point to note at this stage is that the data refer to mortgage transactions at the time they are approved, rather than completed. This has the disadvantage of covering some cases which may never proceed to completion. On the other hand, it has the important advantage that the price information is more up-to-date as an indicator of price movements and is on a more consistent time-base than completions data given the variable time lags between approval and completion.

3. The methodology and its application

The methodology is based on the "hedonic" approach to price measurement in which goods are valued not for themselves as such but for the set of attributes which they possess (see Lancaster 1966, 1971, Griliches 1971 and Triplett 1971). Thus in the case of housing, prices will reflect the valuation placed by purchasers on the particular set of locational and physical attributes (or characteristics) possessed by each house. The difficulty facing the analyst, of course, is that the implicit "price" placed by a purchaser on each characteristic is not observed because transactions take place in terms of a single total price. Therefore, in order to remove that part of price variation due to changes in the mix of house characteristics over time, and so to measure the variation caused by inflationary factors, it is necessary to disaggregate prices into their constituents statistically. This is done using multivariate regression analysis. On this basis it is possible, given data on the prices and the attributes of the houses sold in different time periods, to estimate the change in average price, from one time period to another, on a standardised basis (that is, keeping the mix of housing attributes or characteristics constant). An obvious analogy is with the standard "basket" of goods in the retail price index. It is of interest to note that the technique has been employed by the United States Bureau of the Census in generating a price index of new one-family houses sold each quarter in the USA (see US Department of Commerce, Bureau of the Census 1981).

In relation to the present study, a set of house prices, Pi (i= 1,2, ….n), may be observed in any time period (t) in which each house (i) is sold. Given the supply and demand conditions in the housing market, such houses may be priced differently due to differences in qualitative characteristics (such as the type of property, the availability of certain amenities, the regional location of the property etc), and to differences in quantitative characteristics (such as the age of the property, the number of habitable rooms, garages, bathrooms etc.). Thus, for each house i, we can write Pi as some function of these various characteristics, Xj, together with a group of unmeasured factors (assumed to be randomly distributed) which are specific to each house but for which data are not available, ei. In general terms the relationship may be expressed as follows:

Pi= b0+b1X1i+B2X2i+…..bjXji+ei

Where b1, b2 …bj are the regression coefficients corresponding to the qualitative and quantitative variables (Xj).

Given the nature of the data employed in this study, qualitative characteristics can only be represented by "dummy variables" which take the value of one or zero depending upon the presence of absence of a particular attribute. Definitions of the variables for each set of characteristics and the codings used are listed in Table 1 overleaf. The technique of ordinary least squares allows us to estimate the coefficients bj pertaining to each of the explanatory variables Xj for any set of houses. These coefficients indicate the relative importance of the variables in explaining the variation of house prices in any one time period t.

Table 1

Definitions and Code Names of Variables Included in the Analyses

House Characteristic Code Definition

House Type:

Detached DH

Semi-detached SDH

Terraced TH

Bungalow BUNG

Flat FLAT

Five dummy variables taking the value of 1 if the property corresponds to a particular type. Otherwise 0

Number of bathrooms NOBATHS Actual number of bathrooms

Number of separate toilets NOTOILET Actual number of separate toilets

Number of garages NOGARAGE Actual number of garages

Number of garage spaces NOGSPACE Actual number of garage spaces

Presence of a garden GARDEN Dummy variable taking the value of 0 if the property has a garden. Otherwise 1.

Number of acres A1 Dummy variable taking the value of 1 if the property has one acre or more. Otherwise 0.

Central heating;

Full CHF

None CHO

Partial CHP

Three dummy variables taking the value of 1 according to central heating provision. Otherwise 0.

Freehold FH Dummy variable taking the value of 1 if the property is freehold. Otherwise 0.

Location:

(Standardised Statistical Region – formerly Economic Planning Region)

North EPR1

Yorkshire & Humberside EPR2

North West EPR3

East Midlands EPR4

West Midlands EPR5

East Anglia EPR6

Wales EPR7

South West EPR8

South East EPR9

Greater London EPR10

Northern Ireland EPR11

Scotland EPR12

Twelve dummy variables taking the value of 1 according to the region in which the property is located. Otherwise 0.

Road Charge Liability ROADCHRG Dummy variable taking the value of 1 if the property is liable to a road charge. Other wise 0.

Number of habitable rooms NOHABS Actual number of habitable rooms.

Age of property PROPAGE Actual age of property in years

Having obtained estimates of the coefficients, bj, it will be appreciated that the average price of any set or sub-set of houses in any period depends on the number of observations on each characteristic in that period. Therefore, standardisation to allow for the varying mix of characteristics between one time period and another may be accomplished by applying a standard "representative" set of weights corresponding to the numbers of each characteristic observed in a chosen period. It is common to adopt, as a standard, the set of characteristics that pertained in a base period and this is the practice adopted here, the year 1983 being chose for this purpose. Thus the index numbers we calculate represent the movement in average prices for houses possessing the same characteristics as those bought in 1983. The index numbers themselves are computed by comparing the weighted (i.e. mix-adjusted) prices in each current period with the weighted average price in the base period.

Before proceeding to the results of these calculations, however, we would draw attention to the nature of the underlying statistical analysis because the reliability of the results depends upon the reliability of the basic price-estimating equations. We comment briefly on the most important matters here and place a more detailed exposition of the analysis in an Appendix. Two matters demand special attention in establishing the appropriate estimating equations.

Firstly, it is necessary to ensure that the explanatory variables used in the equations are sufficiently independent of one another to allow their relative importance as determinants of prices to be reliably estimated. Although in principle it may be thought desirable to use all the information available about all explanatory variables, in practice certain variables may be so correlated one with another that it may be impossible to measure particular coefficients in any one set or sub-set of the data without the problem of "multicollinearity". Where this is severe the regression coefficients for the variables affected reflect not only the relative importance of those variables but also that of other variables with which they are correlated. As a consequence it is necessary to conduct appropriate statistical tests to examine the possible existence of this problem and, if necessary, to take the appropriate remedial action. It will be appreciated that it is not possible to measure all the characteristics that may influence prices or to measure them satisfactorily in every case. In particular, qualitative factors relating to the standards of repair of existing (non-new) houses, the quality of workmanship, the nature of fixtures and fittings, environmental quality of the neighbourhood etc., are not reflected in our equations except in so far as they may be correlated with the variables which are measured. Consequently it is not possible to explain all of the variation in prices that is observed. However, the characteristics used in the equations in this study generally explain around 70 per cent of this variation in the UK and 55-80 per cent at the regional level, depending on the particular sub groupings of houses. Explanatory power of this order is generally held to be very satisfactory indeed in studies of this kind.

Secondly, it is necessary to determine the appropriate form of the functional relationship between the variables. In this respect the dummy variable technique is particularly useful because it allows the incorporation of variables with which price may be related non-linearly, without the necessity of specifying the nature of the non-linearity. This, of course, is not the case for those variables such as age, number of habitable rooms etc., which are not dichotomous variables and these are incorporated into the estimation procedure as integer values. However, preliminary analyses of polynomial transformations of these variables showed that such transformations provided no significant addition to the explanatory power of the regression equations. But, in contrast, transformation of the dependent variable, price, proved to be statistically significant and this was confirmed by the use of Box-Cox tests (see Appendix): these showed that the semi-logarithmic functional form (with the dependent variable Pi measured in natural logarithms) was to be preferred. The specifications of the final regression equations used to generate the standardised index numbers, are shown in Table 2 overleaf. Reference should be made to the Appendix for a fuller account of these and other matters relating to the statistical analyses.

Table 2

Final Regression Specification (Dependent Variable Inpi)

Variables*  denotes variables included. [X] denotes variables excluded.

Table 2

Regression Equations For:

All

Houses New

Houses Exisiting

Houses First Time

Buyers Former Owner

Occupiers All

EPRs+

DH

Omitted Dummy

SDH      

TH      

BUNG      

FLAT      

NOBATHS      

NOTOILET      

NOGARAGE      

NOGSPACE      

GARDEN  X    X

A1      

CHF

Omitted Dummy

CHO      

CHP      

FH  X    X

EPR1      

EPR2      

EPR3      

EPR4      

EPR5      

EPR6      

EPR7      

EPR8      

EPR9

Omitted Dummy

EPR10      

EPR11      

EPR12      

ROADCHRG   X   

NOHABS      

PROPAGE  X    

* Variable code names are defined in Table 1.

+ The Regional regressions for new and existing houses only include variables common to both the region and sub-sample specifications.

 Variables omitted for computational purposes. One variable from each dummy variable sey must be excluded in order to avoid the problem of indeterminancy of the ordinary least squares normal equations.

We are now in a position to apply the methodology to derive index numbers. The methodology is applied here to produce base-weighted standardised house-price index numbers, whereby a weighted average of the estimated regression coefficients is calculated (each coefficient being regarded as an implicit characteristics-price). It will be appreciated that weights other than those appropriate to the base period may be adopted.

The steps involved may be summarised as follows:

(i) calculate the weights, Qj1983: the proportions of the qualitative variables and the means of the quantitative variables presenting the chosen base period (i.e 1983);

(ii) with price recorded in natural log form, use the technique of ordinary least squares to estimate the regression coefficients bj for the j explanatory variables, in both the base period (i.e. bj1983) and for every subsequent time period t (bjt);

(iii) calculate a base-weighted (Laspeyre's type) index for the current period (It) as follows:

It = antilog bjtQj1983 X100

antilog bj1983Qj1983

Summation is carried out, of course, over all variables included in each regression.

4. Results

For the United Kingdom as a whole, the procedure outlined above is used to compute five monthly series of base-weighted index numbers covering all, new and existing houses, and houses bought by first –time buyers and former owner-occupiers respectively. The results for 1983 are printed in Table 3 overleaf.

The increases over the period from January to December 1983 are always in the range 7 – 8 per cent but with some variation according to house type and house buyer. The index for all houses, based on 1983 = 100, rises from 94.8 in January to 102.2 in December: an increase of 7.8 per cent.

Table 3

Price Movements in the UK by Category of House and Buyer

Monthly Series for 1983

(Index 1983 = 100)

House Buyer

Month All Houses New Houses Existing Houses First-Time Buyers Former Owner-Occupiers

January 94.8 96.1 94.7 95.6 94.9

February 95.7 96.3 95.9 96.8 95.5

March 96.9 97.0 97.0 97.4 96.7

April 98.5 99.2 98.4 98.5 98.4

May 100.3 101.1 100.3 99.8 100.3

June 100.9 100.6 100.7 100.5 100.9

July 102.1 100.9 102.2 101.8 102.0

August 102.3 100.4 102.5 101.4 102.6

September 102.2 101.9 102.3 101.6 102.6

October 102.5 102.2 102.7 102.5 102.4

November 102.4 102.4 102.4 102.4 102.2

December 102.2 102.9 101.9 102.8 102.1

In addition, index numbers for each of the twelve standard (statistical) regions for all, new and existing houses are computed quarterly. It should be appreciated that, as the regional analyses are inevitably based on smaller sample sizes than the analyses for the UK, the regional index numbers are subject to larger margins of error. It is for this reason that they are computed on a quarterly rather than a monthly, basis.

The regional results for 1983 are given in Table 4 overleaf. They show that price changes at the regional level during 1983 were much more variable than at the national level, ranging from a low of 2.4 per cent to a high of 10.4 per cent, both relating to existing houses in the West Midlands and in Scotland respectively. For all houses the increases ranged from 3.2 per cent to 9.7 per cent, again in the West Midlands and Scotland respectively. In the case of new houses, the increases ranged from 2.8 per cent (Greater London) to 7.9 per cent (South East).

The UK and regional series are updated in monthly and quarterly bulletins respectively. These are available from the Halifax on request.

Table 4

Regional Price Movements. Quarterly Series for 1983

(Index 1983 = 100)

All Houses New Houses Existing Houses

Region Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4

North 96.5 98.8 102.4 102.6 96.2 99.8 102.4 100.2 96.6 98.6 102.4 102.9

Yorkshire & Humberside 96.2 100.2 101.7 102.2 96.3 100.7 101.0 102.1 96.3 100.0 101.8 102.1

North West 96.5 100.2 101.8 101.5 96.2 101.7 100.3 101.4 96.8 100.0 102.0 101.4

East Midlands 95.5 100.4 102.5 102.0 95.6 99.5 102.4 102.6 95.6 100.5 102.3 102.1

West Midlands 97.1 101.2 101.7 100.2 97.2 98.0 100.6 104.2 97.1 101.5 102.1 99.4

East Anglia 95.1 99.2 101.7 104.2 95.7 99.2 100.5 103.2 95.2 99.3 102.0 104.1

Wales 96.4 100.8 102.2 100.4 96.4 103.8 98.1 101.9 96.6 100.3 103.0 100.3

South West 96.5 100.2 101.8 101.7 97.7 98.9 102.2 102.0 96.4 100.3 101.6 101.8

South East 95.6 99.2 102.8 103.3 95.9 100.2 101.8 103.5 95.5 99.1 103.0 103.3

Greater London 95.7 99.2 102.6 103.2 97.1 101.9 101.8 99.8 95.7 99.1 102.6 103.3

Northern Ireland 96.3 99.5 102.5 101.4 95.9 101.5 100.8 101.5 96.1 99.0 103.3 102.3

Scotland 94.4 98.9 102.9 103.6 98.0 99.2 100.0 103.3 93.9 99.0 103.4 103.7

Appendix

In regression analysis, four types of interrelated problems are commonplace, and are especially relevant in the present context, namely:

(i) the choice of functional form for the estimating equation,

(ii) the degree of correlation between the explanatory variables themselves (i.e. multicollinearity),

(iii) the possibility of the error term, ei, exhibiting a non-constant variance (i.e. the presence of heteroskedasticity), and

(iv) errors in the recording of data and the inclusion of extreme or unusual observations ("outliers").

Considerable attention was devoted to each of these subjects in arriving at the estimating equations; we comment on each of them in turn below.

(i) Functional Form

The first step in regression analysis is to determine an appropriate functional form for the estimating equation. A potentially serious source of bias in hedonic price and other regression-based studies may be associated with functional form mis-specification. In principle, many functional forms are possible, but unfortunately there is no theoretical guidance as to which form is the most appropriate to particular models. The solution to this problem, therefore, reduces to an empirical one. Box and Cox (1964) have developed a statistical test for the functional form providing the "best fit" based on likelihood ratio tests and the procedure they suggest is adopted here. The results showed that the semi-logarithmic functional form (with the dependent variable Pi measured in natural logs) was to be preferred. The specifications of the final regression equations which are used to generate the standardised index numbers, are shown in Table 2 above.

(ii) Multicollinearity

Multicollinearity refers to the situation in which some or all of the explanatory variables are very highly correlated and are therefore not independently distributed. The existence of marked interrelationships between explanatory variables can cause problems with respect to the following aspects of regression analysis:

(a) estimated regression coefficients may not be uniquely determined,

( estimates of the coefficients from sample to sample may fluctuate markedly, and

© less reliability may be placed on the relative importance of variables as indicated by the partial regression coefficients.

It should be appreciated, however, that multicollinearity is inevitably present to some degree in most multiple regression analyses. It can rarely be eliminated completely and the aim of the researcher therefore is to minimise its influence as much as possible.

The procedure adopted for identifying such interrelated variables was partly by examination of the correlation matrices for the explanatory variables but mainly by a stepwise regression procedure in which the relative explanatory power of each variable is determined in turn. At each step of the procedure the interdependence between each variable being considered as a candidate for inclusion is compared against the group of variables already selected. The measure used in this procedure is referred to as the "tolerance" level of a variable, measured by (1 – R2j) where R2j is the squared multiple correlation when the Jth explanatory variable is considered as the dependent variable and the regression equation between it and the other independent variables is calculated. If the variable has a large R2 – or equivalently a small tolerance – when it is predicted from the other independent variables, then the presence of multicollinearity (to a degree) can be assumed. As a consequence, the variance of the estimators will be inflated and computational problems can occur.

The classic way of dealing with the problem of multicollinearity is to discard variables from the regression analysis. If two variables are very highly correlated, use of either one in the regression (rather than both) can capture the effect of both. This is the approach adopted in the present study. By setting tolerance limits at an acceptable level, as well as gleaning the evidence from the relevant correlation matrix, it is possible to identify the most significant interrelationships between variables. Three alternative variable-selection procedures are possible: "forward selection", "backward elimination" and "stepwise selection". These are described fully in Nie et al. (1975), pp 345-347 and Norusis (1982), pp. 118 –121. Each procedure will not necessarily give identical results. In the present study, however, all three methods were employed and they were consistent in leading to particular conclusions concerning the variables that should be included in the final regression specifications.

(iii) Heteroskedasticity

In regression analysis, the error terms, ei, are assumed to be normally and independently distributed with a mean of 0 and a constant variance of o-2. However, when dealing with data of the kind used in the present study, the constant-variance assumption may be violated, and the appropriate model may in fact be one with a so-called "heteroskedastic" error term – for a detailed discussion of this problem see Kmenta (1971), pp. 249-269.

A simple and direct method of detection is by visual inspection of the residuals (i.e. deviations of observed prices from those estimated from the regression equation) plotted against the estimated values of the dependent variable or against the independent variables. As most of the latter in this study are dichotomous, our attention was focused on the former. If the spread of the residuals increases or decreases with estimated values, it may be taken as an indicator of heteroskedasticity in the data set. The total data set for a sample month was used to test for the existence of this problem and the resultant scatterplot of residuals was observed to be relatively uniform across the range of estimated values. It is concluded therefore that the constant-variance assumption appears not to be seriously violated with respect to the sample data set. There is no reason to believe this should not hold for the data sets relating to other time periods. It should be appreciated, however, that even if heteroskedasticity exists, it would only increase the confidence intervals of the indices; it would not affect their unbiasedness.

(iv) Data-recording (Coding) Errors and Outliers

Cross-tabulations of the date were used to spot potential coding errors and outliers. Examination of the residual plots, referred to above, also helped to identify outliers, as they are cases with very large positive or negative residuals. Particular attention was devoted to the recorded number of habitable rooms (NOHABS) as this is a key variable and it was finally decided to restrict the variable for habitable rooms to the range from one to twenty inclusive, all properties having values outside this range being excluded from any subsequent analyses. Certain restrictions are also placed on other variables.

References

Box, G.E.P and Cox, D.R. (1964), An Analysis of Transformations, Journal of the Royal Statistical Society, 26, Series B, pp 211-243.

Fleming, M.C. and Nellis, J.G. (1981), The Interpretation of House Price Statistics for the United Kingdom, Environment and Planning A, 13, pp 1109-24.

Griliches, Zvi (1971), Price Indexes and Quality Change, Harvard University Press, Cambridge.

Kmenta, J. (1971),Elements of Econometrics, Macmillan, New York.

Lancaster, K. J. (1966), A New Approach to Consumer Theory, Journal of Political Economy, 74,pp 132-57.

Lancaster, K.J. (1971), Consumer Demand, A New Approach, Columbia University Press, New York and London.

Nie, N.H., Hull, C.H., Jenkins, J.G., Steinbrenner, K and Bent, D.H. (1975), SPSS: Statistical Package for he Social Sciences, 2nd Edition, McGraw-Hill, New York.

Norusis, M.J. (1982), SPSS: Introductory Guide, McGraw-Hill, New York.

Triplett, Jack E (1971), The Theory of Hedonic Quality Measurement and its Use in Price Indexes, US Department of Labor BLS Staff Paper 6, US Government Printing Office, Washington DC.

US Department of Commerce, Bureau of the Census (1981), Construction Reports. Price Index of New One-Family Houses Sold, First Quarter 1981, US Government Printing Office, Washington DC.