The Scottish Government

« Previous | Contents | Next »

Listen

DEVELOPING A METHODOLOGY TO CAPTURE LAND VALUE UPLIFT AROUND TRANSPORT FACILITIES

6. CONCLUSIONS FROM THE REVIEW AND THE NEXT STEPS FOR RESEARCH ON THE LAND AND PROPERTY EFFECTS

6.1 Introduction

In this review a wide range of evidence has been presented, together with some commentary on the main issues that have emerged. In this extended final section, three main objectives are set. First, some of the main findings and preliminary conclusions are presented on the literature that has covered the 150 recent references on land values and public transport. Secondly, this experience is updated with the evidence from the very recent study carried out on Croydon Tramlink (ARW et al, 2003), not in terms of what was found, but a commentary on the methodological issues and the implications for further research. In the final part, the next steps in terms of the development of the research methodology to capture land value uplift around transport facilities are presented.

6.2 Findings from the Literature Review

1. Much of the analytical and empirical research comes from the USA and Canada, and this has mainly concentrated on the commercial property market. It seems that the quality and availability of data from North America is good and allows the use of more sophisticated methods, including regression analysis and hedonic pricing.

2. The evidence from the UK and Europe is more varied, but does include individual case studies and comprehensive reviews. There seem to be less time series or repeated cross sectional data available for analysis. The use of more simple indicators such as transactional price analysis, growth assessment and projected rateable values has been explored.

3. Until recently, UK and European research has tended to concentrate on the transport impacts in terms of more traditional changes in demand patterns and the switching of travel to the new mode. There has been less interest in the property market effects. But this has now changed, with much of the recent work from key researchers and the property market sector itself being concentrated on the land and property value uplift effects of transport. This work has concentrated on the residential sector, where there is reasonable quality data from the land registry, the Valuation Office and other agencies.

4. From the extensive literature cited, 18 key references were selected for more in depth analysis ( see Annex B). It seems that the expected effect on both the residential and commercial property markets is positive, but the range of impacts is very variable - from marginal to over 100% in the commercial sector from the North American evidence. In the UK the impact is seen as being positive, but there has been less emphasis put on exact amounts. However, some of this uplift may be due to the optimism of the markets rather than actual effects. Where possible, it is important to use transactional data rather than valuation data.

5. One conclusion here is that each study is different and the bringing together of results in the review has been difficult. It requires a greater depth of investigation that looks at data, definitions, methods and actual cases to unravel what effects can be attributed to the transport investment. This means that knowledge must be built up from a series of carefully constructed case studies

6. The treatment of time is important in all studies. This is because changes will take place in land and property values in advance of the completion of the transport investment as developers and house builders will invest in the expectation of improvements in the transport infrastructure. Effects might also be expected immediately after the transport investment is opened, and further in the future as the full benefits are recognised. Ideally, data should be available from before the decision to build was taken and immediately after opening, as well as downstream. A continuous database is ideal, but data are needed for at least these three points in time.

7. Catchment areas are also important. The impact area for residential developments seems to be wider than those for commercial developments. Depending on the investment, residential impacts could extend to 1000m, whilst those for commercial developments are likely to be concentrated in a 800m radius. There is also some evidence that residential property prices might be depressed immediately around the transport investment or station.

8. Most studies seem to take a series of key thresholds as inputs to the study of where different types of impacts might be found. Only a few used the data to define the range of distances away from the transport investment at which impacts might be found.

9. Impacts are more easily identified for tram and metro investments than for bus investments. Most of the research has concentrated on urban rail systems. In the first instance, rail investments offer the best opportunities to test for the property market effects.

10. Although the analysis has produced variable results, there is also the question of attribution of impacts. It seems that the contextual situation is important and should be seen as an input to any analysis. Similar transport investments will have different impacts in locations where there is a vibrant local economy and where the economic conditions are less advantageous. The key question then becomes, what other actions are needed in an area apart from the transport investment to make a measurable impact in terms of value uplift?

11. Value uplift has tended to be looked at in the literature in a narrow way, mainly through changes in property and land values. We would suggest that, where possible, a wider range of measures should be used. These would include changes in accessibility, ownership patterns for land and property, site consolidations, numbers of transactions and yields, as well as the use of composite measures such as density of development.

To develop a robust methodology requires commentary on the contextual situation, an appreciation of the data requirements and limitations, a clear understanding of the issues relating to attribution, and some means by which the numerical results can be interpreted. This suggests a mixed quantitative and qualitative approach - this was the approach used in the Croydon Tramlink study (ARW et al, 2003).

6.3 Evidence from the Croydon Tramlink Study (CTLS)

The logic of the approach used in Croydon was that a major investment in transport infrastructure would lead to significant improvements in accessibility to the transport system as a whole, and that this would result in increases in local demand for land and property. Prices would rise (value uplift), with pressures for new development, which would in turn result in further increases in demand. The purpose of the CTLS was to develop a methodology to measure the scale of that value uplift (ARW et al, 2003).

The analysis process is presented in brief in Table 7. It consisted of three main elements with most effort going into the quantitative analysis. The methodology was attempting to measure change over space (the CTL corridor) and over time (1996-2002), where small local effects needed to be isolated within a very heterogeneous residential property market - there was little spatial autocorrelation in the data.

Table 7: The Structure of the CTLS

Qualitative Analysis

The analysis was carried out in two stages, one concentrating on the South London residential property market and the other on a two kilometre corridor along the actual CTL route. In the South London study, the overriding impact of the Central London property market was clear as price increases cascaded for the centre towards Croydon. There were no discernable patterns of change in the corridor itself. The localised corridor study did produce spatial variation, but not where expected. For example, in the New Addington spur where the largest increases in accessibility took place, residential property value changes were consistently below average.

The greatest increases in all four property types (detached, semi detached, terraced, and flats and maisonettes) took place near to Beckenham. There were also increases in the semi detached market around Central Croydon and Mordon/South Wimbledon. Increases in flats and maisonettes were again found around Mordon/South Wimbledon and Mitcham. In the terraced market, the increases were found just north of Central Croydon and around Crystal Palace, which is further away from the line.

It can be concluded that residential property values in the four market sectors demonstrated very variable patterns of price change, which reflected the heterogeneity of the market and the very localised impacts of market changes. It was very difficult to isolate any CTL effect.

Of greater interest in this study has been the development of the GIS Mapping approach to the measurement of value change. This seems to have been successful. The GIS Mapping approach does not use zones, but each data point is used to produce data surfaces so that land and property values across a landscape can be mapped. All the data are fully used, and the software uses spatial interpolation methods to draw the maps. As it uses the individual data points at the unit postcode level, each transaction is located to the side of a particular street - typically, unit post codes cover 10-15 delivery points. This means that changes in property values can be mapped at any distance from the transport network. In the CTL study, thresholds of one kilometre and two kilometres were used to identify change, and the level of change was related to price changes as a whole within each location - the variation around the mean price levels. This method seems robust and very flexible.

The data source used was the Land Registry data for the period 1996-2002, for each of the four housing types (detached, semi detached, terraced, and flats and maisonettes). The limitations are that there is no other data available for a more complete understanding of the property market (e.g. floorspace, bedrooms, plot size etc). Hence there is a strong assumption that price is dependent only on location and type of property. If the Valuation Office data were available, then additional details can be added to the analysis (e.g. floorspace, type and age of property are recorded in about half of the entries), and a similar analysis can be carried out on the commercial sector. The availability of high quality geocoded data over time is essential to this type of analysis.

Various modifications had to be carried out on the data to ensure that the comparison was taking place of like with like. For example, spatial bias in the data has been controlled by only including unit postcodes that had a transaction in both time periods (e.g. Q2 1996 and Q2 2002). This meant that the size of the data set was reduced to 4016 common entries out of the 12,265 transactions in Q2 1996 and the 19,072 transactions in Q2 2002. Test were also carried out to establish whether the four subsectors could be aggregated, but the analysis of variance and t tests clearly demonstrated that each submarket was significantly different to the others. Similarly, there was a problem with outliers where individual properties were at a much higher (or lower) price than its neighbours - these outliers were also removed from the database. Finally, the data were normalised to allow for more confidence with the comparison between locations, and to control to a limited extent for local variability. This was achieved through a density variable being used to normalise the transaction price. The argument here is that the higher the density, the lower the expected price for a given housing type (and vice versa).

Much of this data manipulation was designed to address the substantial heterogeneity within the market (much more so than in the US), and to ensure that comparisons made over time and location were robust. This process takes time, but it is an essential part of the research quality design. It also demonstrates the need for more variables to work with (as with the Valuation Office data). A question emerged as to the appropriate timescale over which property value effects might be expected to take effect. The CTL had been open for two years (2000-2002), and the data set extended back to 1996 to control for any potential value uplift before the opening. Two years after opening may be too short a time for the effects to work their way through into the housing market. In the Croydon situation, a five year timescale may be more appropriate (to 2005), as this was the timescale for measurable change in the Sheffield Supertram corridor.

The GIS Mapping approach permits a range of different methods to be used. The Inverse Distance Weighting (IDW) method was used in the CTL study (ARW et al, 2003), where the distance decay function is preset so that the surfaces are drawn on the basis of the property transaction costs and the distance between the properties. Here it is assumed that the degree of spatial autocorrelation declines with distance. Other methods were tested (e.g. Kriging where the data itself is used to determine the distance decay function).

One promising approach is Geographically Weighted Regression (GWR) which combines an explicit spatial dimension within a multiple regression approach, so that price can be seen as a function of earlier prices, density and accessibility, as well as a range of other variables relating to the property (if available). A limited version of GWR was applied to the CTL data, with promising results. There seems to be considerable potential for further use of this method if more data were available.

The main alternative to the GIS Mapping approach is Hedonic Pricing models. These have not been used in the CTL study, but have been used to analyse the price variations in complex goods, and they have been applied to the housing market with mixed success. Hedonic pricing is a form of regression modelling that tries to estimate how a wide range of independent variables can be capitalised into land values or property prices (Rosen, 1974). Simpler versions have been used, such as the matched pair method, where all variables are held constant except for the one under study. But as with GIS Mapping and Hedonic Price, the matched pair method depends on good quality data and even stronger assumptions about the control variables. It is also very difficult to validate the output.

Mathematical/econometric models simplify the factors and interactive forces at work in a system in pursuit of a reductionist mathematical description of behaviour. They blend together sets of data using appropriate statistical weightings to track the behaviour of the dependent variable of interest.

Generally speaking they are correlational and predictive in nature, although it can be argued that curve fitting, by implication, may add an explanatory aspect to the model. Thus some analysis is required in order to make a judgement about which explanatory variable to include, the functional form of the equation, how the statistical fit of the model should be interpreted and how useful the resulting model is for forecasting or explanatory purposes. Numbers, however, dominate the output and they may say little on process and the dynamics of change.

Such models, by considering processes in play at a given time may appear to offer deterministic predictions covering the future, assuming different policies or exogenous events. However, they can only be correct for as long as the qualitative structure of the system to which they relate remains unchanged. In summary, they offer an indication of those factors that are important. They do not necessarily explain matters accurately or completely, and are not well able to accommodate change. Nevertheless, they are a commonly applied approach that does offer some understanding on underlying relationships.

Hedonic Pricing requires data on individual properties that reflect as many of the differences as possible in terms of the characteristics of the properties themselves, location, local facilities, and even preferences. As noted above, the Hedonic Pricing method then estimates the contribution of each of the independent variables to explaining the dependent variable (house price). It is not difficult to represent the Hedonic Pricing approach within a GIS framework as there are many spatial variables used. The GWR approach might be one way to link the two methods together formally. The key difference between the Hedonic Pricing and the GIS Mapping as outlined here is data. For Hedonic Pricing, it is necessary to collect primary data, but the GIS Mapping approach has only used available geocoded data There may be a substantial additional cost involved in collecting data for Hedonic Pricing.

As one example of the use of Hedonic Pricing, Cheshire and Sheppard (2003) have estimated the influence of quality of local schools (primary and secondary) on house prices, calculating a maximum uplift of 18.7% for secondary schools and 33.5% for primary schools (if the households moved from the worst to the best possible school area). The estimates were made from a sample of 490 houses sold in the Reading area (1999-2000: a 17 month period), with about 25 independent variables including performance at Key Stage 2 and GCSE examinations being the two quality variables for schools. The data used had been assembled from a wide variety of sources and supplemented by direct measurement. In this way it differs to the CTL method, which relied exclusively on secondary data for the quantitative analysis. Again, data is key to the quality of the research, but Hedonic Pricing combined with GIS Mapping allows for land values (and property prices) to be seen as a function of location, accessibility (including shops, schools and amenities), the property itself, and the structure of the market.

The clear conclusion here is that the methodology used in the CTL study is robust and suitable for the measurement of land value uplift. It is the best available. It can be extended through GWR and Hedonic Price to include more variables, so that the impact of other variables can be better controlled, and the particular aspects of interest isolated. The two key determining factors are the availability and quality of the data available, and the time period over which measurable impacts should actually be identified.

6.4 The Next Steps

This extensive review of the literature on land values and transport investment has pointed to the many weaknesses in the methods and data used. While accepting that the issues being discussed are complex, this does not mean that simplistic analysis should be carried out, and it does not mean that research is not needed. In this section, the main elements of the debate are reiterated and the next steps are identified.

Transport investments are expensive, and it is increasingly difficult to justify these investments on transport criteria alone. Even the most significant new links in Scotland will have a limited effect on the total levels of accessibility. But they will have a major effect on local communities, both in terms of new development pressures on those areas, and in terms of opportunities created for residents by new investment. All transport links have two-way effects as they can encourage movement in and out of local areas. Analysis should attempt to measure these movements over time in a holistic way, so that the benefits (or costs) to the community can be captured. At present the interest is more on the net effects rather than the dynamics of change.

One might expect the land value uplift from a transport investment to relate to the size of that investment. But here again there is little evidence from the literature, as studies have tended to concentrate on a particular system rather than to compare impacts. As already commented (in the notes to Table 5 in Section 3), there are substantial difficulties in making comparisons between schemes. For example, there is no compatibility between definitions, the measurement of distance (or time), whether the price changes are relative (to what base) or absolute, the time over which change is monitored, and the difficulties of cross national price comparisons. It would seem that the best means to progress the research agenda would be to build up a series of internally consistent case studies in the first instance. At a later stage, it may be possible to carry out comparative analysis.

The case study approach is needed to understand the complex range of interactions going on at the local level over time. Most evaluation (in transport) is carried out ex ante to determine whether an investment is worthwhile in transport terms. Little attention is given to monitoring outcomes to see whether they match up with expectations, and this has been seen as a weakness in the approaches to demand forecasting and cost estimation (Flyvbjerg et al, 2003).

But more important in the context of land value change is the acceptance that these variations take place over time and that there are differential effects on existing and new developments. Ex post evaluation allows a better understanding of the processes at work, and through monitoring it is possible to allow feedback to improve these forecasting methods. The new generation of evaluation methods must evolve with a clear dynamic element to them. We would expect value uplift to take place at the time of the initial announcement of the project, when it opened (either partially or in its entirety), and over time as the longer term adjustment process takes place.

The importance of this issue cannot be underestimated, as it provides the main mechanisms to understand cause and effect (attribution). The transport investment is the trigger to change, but that action does not happen at one point in time, so the traditional before-and-after studies are of limited value. The process of change is dynamic and occurs over time, but one would expect the key moments in time to be highlighted through a greater than expected movement in property values.

As well as identifying the inputs (triggers) and the outputs (property value changes), it is necessary to place a change in transport within the wider contextual situation so that the effects of other actions such as the property market cycles, local inward investment, and skill levels can all be controlled for. In dynamic analysis, one is accepting the complexity of situations and the importance of time, but one must also be realistic about other changes taking place. These can be complementary and support the impact of transport investment, but they can also work against that investment.

Returning to the issue of scale of investment and expected impacts, there are potentially two types of effects. With the large scale investment, one might expect most market adjustments to take place prior to the opening of the new transport scheme in anticipation of change. With smaller scale investments, one might expect little market response prior to the opening of the scheme, but a greater impact after it has opened. Such a set of hypotheses could be investigated through a series of case studies.

The methods that can be used are dependent on the data that are available. On data, it is important to have local level data on actual transactions rather than valuations. For the residential sector, this is available from the Land Registry (LR), and for the residential and commercial sectors the best source is the Valuation Office Agency (VOA) in England and Wales. Scotland also has the Register of Sassines. In addition, there are other sources, relating to the non-residential building stock on vacancy levels and measures of quality, and commercial sources (e.g. Investment Property Databank Ltd, Scottish Property Network) on capital values, returns and revenue flows.

The key limitation here is the scale at which that data are available. The impacts for transport investments are highly localised, usually within about 800 metres of the station for commercial developments and up to 1km for residential developments. To make real progress on linking land value change with transport investment requires a high quality data set to be made available at the necessary level of disaggregation, often down to the individual property level - or the unit post code level. Even though this may raise issues of confidentiality, progress is contingent upon that information being available. However, the results of any research would be aggregated so not to reveal individual transactions.

In addition to the transaction data, supplementary information could be assembled over time on vacancy rates, vacant and derelict land, and interviews could be carried out with developers and agents on the broader market changes taking place, covering such topics as yields and the image of the area. The key elements here are that a commitment is made to collect the relevant information consistently over a period of time.

Once the database is available, then analysis can be undertaken on a consistent basis to cover all the methods highlighted in Section 3 and earlier in Section 6. The main reason why more sophisticated analysis has been possible in the USA (Cervero and Duncan, 2002a and b), France (Poupard, undated) and the UK (Cheshire and Sheppard, 2003) has been the availability of a few high quality data sets that allow regression analysis and hedonic pricing to take place. Even though the conclusion reached from this review would be that these two methods are the most appropriate to understand a scale of the impacts and to attribute cause and effect, the levels of explanation are still modest (up to 50 per cent).

In addition to the multivariate analysis, there is a requirement to use the data in other ways, including contextual analysis and to establish simpler measures for indicator analysis. They would cover the transport and non-transport impacts - including travel times, accessibility levels, rent levels, land values, ownership patterns and land availability. It is suggested that a series of indicators of change are developed to monitor these factors. Included here are:

• New business and residential locations - movers in (and out);
• Benefits to existing residents and businesses - capital uplift;
• Land price and rental levels (and changes);
• Changing patterns of land ownership, with consolidation of sites for development;
• Investment yields on existing commercial property and on newly completed schemes;
• Accessibility levels and changes (including travel times);
• Land availability and vacant land (and buildings);
• Development starts, including planning permissions.

To combine elements of the changes in the system with those relating to the property market, a series of quantitative relationships could be established:

• Correlations between location (near to public transport) and price, land values and rents;
• Correlations in the changes in land price and accessibility by public transport (with perhaps the need to keep car accessibility levels constant). This could be carried out on an area wide basis with weighted travel times;
• Regression on the price of land (or rent levels) and accessibility by public transport, including an analysis of the signs on the equations derived as one would expect location to be positively influenced by the availability of land and accessibility;
• Regression to test the notion that the amount of new build is positively related to changes in accessibility and the amount of vacant land is negatively related to accessibility.

The interpretation of the quantitative data is important and needs to be matched with qualitative interviews with developers and key agents. Rather than just assuming that the statistical relationships and indicators demonstrate the impacts of transport investment on land value, these interviews are intended to unpack the relationships through discussion. So the quantitative analysis is being used to inform the qualitative analysis, which in turn feeds back into unravelling the relationships between land values and transport. Such a combined approach provides the greatest potential for understanding the complexity and the question of attribution of cause and effect over time.

Supplementary research is required on the area over which the impact might be expected. This can be carried out in two ways. One would be to use a continuous approach that assumes the impact declines with distance and the objective is to find the point (distance of time) at which there is no impact. The other would be to set discrete limits (distance or time) at which different scales of impact would be expected. These could be initiated by using those from other studies, but they could be modified to take account of local circumstances. The review evidence seems to suggest that catchment areas are different for residential uses (up to 1000 metres) than for commercial uses (up to 800 metres). This topic is not seen as important as the others discussed here, as local circumstances may be the key determinant for a catchment area. For example, catchment areas can be extended through having high quality feeder services, such as good local bus services or park-and-ride facilities.

The final element of the research agenda would complement the quantitative and qualitative analyses, and it relates to the wider context. It seems to be accepted that even if very similar transport investments take place in different locations, different outcomes would be obtained in terms of the property market effects. This conclusion relates to the uniqueness of location. The contextual analysis needs to record these differences in terms of the local conditions (including institutional and organisational issues), market conditions more generally, individual location constraints, and quality factors (including development and neighbourhood quality, the image of the area and environmental quality).

This review has highlighted many important issues related to land values and transport investment, as it relates to conceptual approaches, methods of analysis and data requirements. There seems to be a huge interest in this research both in the UK and in the USA and Europe. There is a realisation that the transport benefits from many investments in Scotland will not be sufficient to justify the scale of public investment on their own. Typically, the discounted travel time savings and other benefits will cover between 90 to 100 per cent of the costs of the scheme. To give a higher benefit cost ratio often requires additional non transport benefits to be included in the cost benefit analysis

Provided that these additional benefits can be measured, it then becomes possible to determine who were the main beneficiaries and to look at the most appropriate means to either get them to contribute to the cost of the scheme in the first instance (e.g. Section 75 or LGA Agreements), or to recoup the cost downstream once investment has been opened (e.g. through the rating and the local tax system, or through a freehold levy to reflect the benefit from the capital gain generated in the value of the existing property stock from the infrastructure projects).

In summary the following seven key points emerge from the literature review:

1. Analysis needs to be carried out along the whole corridor under study, as effects are expected to be variable in their scale and location. In addition there may be network effects that need to be considered, particularly where the existing network is sparse or where the new investment links two unconnected networks;

2. Case studies should be concentrated on larger scale investments with limited access points as this provides the greatest potential for identification of land value uplift - an accessibility analysis should be carried out to determine the scale of change brought about by the investment;

3. Property market effects are expected to be concentrated over 800m and 1000m distances from the stop, station or interchange, depending on the type of development;

4. Data are required before the announcement of the investment, prior to opening, immediately after the opening and some 5-10 years later;

5. There seems to be some convergence here between the two most effective methods, probably with GIS methods having the edge as they are explicitly spatial and make less assumptions than HP methods for time series analysis;

6. Effective analysis depends on transaction data at the unit post code level (i.e. the full post code);

7. A database of changes in other related factors is required for the local area, and the broader area within which it is embedded for the time period under study.

« Previous | Contents | Next »