The Scottish Government

« Previous | Contents | Next »

Listen

DEVELOPING A METHODOLOGY TO CAPTURE LAND VALUE UPLIFT AROUND TRANSPORT FACILITIES

APPENDIX A: REVIEW OF LAND-USE AND TRANSPORT INTERACTION MODELS

1. Current approaches to modelling

This review is based upon a range of previous reviews carried out by DSC over a long period. Recent work has included a paper prepared for DfT in early 2003 and a short presentation to the 2004 Transportation Research Board Annual Meeting.

The state of play in LUTI modelling as it relates to present or likely future UK practice can be summed up by identifying two streams of modelling:

• numerous application of aggregate modelling, i.e. models which work in terms of variables which measure the number of households of each kind in each zone, etc;
• a smaller number of applications of microsimulation modelling, i.e. models which work in terms of lists of households with details of their characteristics, including the zone in which they live, etc.

The aggregate models are represented mainly by the MEPLAN, TRANUS and DELTA packages, plus the MENTOR system which is a subset of MEPLAN adapted to work with different transport models; all of these ultimately have their origins in Cambridge, England. The economic/employment components of the TLUMIP2 model also fall into this category.

The microsimulation models are currently represented by the IRPUD, UrbanSim and the household components of the TLUMIP2 models. IRPUD was developed in Dortmund, Germany (and has only ever been applied there); UrbanSim has been, and TLUMIP2 is being, developed in North America. Note that there are numerous connections between these various modelling streams: MEPLAN and TRANUS originated in the same research projects; DELTA specifically sought to integrate aspects of MEPLAN/TRANUS and IRPUD; IRPUD also had a major influence on UrbanSim, and TLUMIP2 is being developed out of experience with UrbanSim and TRANUS. All the key developers are known to one another, and many are long-standing friends (most of the time).

We have also considered the MUSSA model, which exists in both aggregate and disaggregate forms. MUSSA is a specific model application for Santiago, Chile, but it is of wider interest because the model design gives special attention to the formation of rent values, and the approach adopted could, in principle, be applied in other models such as DELTA.

The following two sections provide summaries of the aggregate and microsimulation models respectively in terms of the ways in which they forecast rents or prices. The differences in the natures of the two groups of the models mean that the descriptions are themselves different in nature. An additional section outlines the relevant aspects of the MUSSA model.

2. Aggregate, deterministic models

2.1 The MEPLAN and TRANUS models

From the mid-1960s to the late 1970s, research in the Martin Centre at Cambridge University spawned a family of interactive land use and transportation models known as the Martin Centre Model. These models are now represented mainly by the MEPLAN ands TRANUS packages (de la Barra, 1989; Hunt and Simmonds, 1993; Williams, 1994). The land-use model is basically a spatial input-output model. It is assumed that activities compete for real estate, resulting in equilibrium prices. The location of activities, generated by the transport system, is influenced by generalized costs. At the heart of the land-use system is an input-output model to predict the change in demand for space. The coefficients for floorspace densities can be elastic with respect to prices and incomes. A spatial system is used to allocate the demand to spatial zones, using random utility concepts. The utility function includes the expected price or rent of the new stock, the price of the stock being replaced, demolition costs, building costs, and so on. Spatial choices link production to consumption, generating the demand for transport. A supply model is used to simulate the expected behaviour of land and floor space developers, and an equilibrium model is derived by solving all the equations, subject to constraints.

The transport model predicts modal split and assignment, with adjustment for times for capacity constraints, given transport demand by type and flow. Again, random utility concepts are used in the transport model. Information about costs, travel times due to congestion, etc are fed back into the land-use economic model to provide time-lagged measures of accessibility between zone pairs. This feedback, however, does not occur instantly in the same time period, but is lagged. The distribution of activities is directly influenced by the transport outputs from the previous modelled year - nearly always five years earlier in contrast with DELTA, where a change in transport conditions in one year directly affects activities in each of a sequence of subsequent years. The indirect effects of changes in transport on floorspace supply (which also influences activities) may take several periods to consolidate. The results of the MEPLAN and TRANUS models consist of a set of flow matrices, ie the results of the distribution model, and the final characteristics of each sector: production, consumption and prices.

The economic/employment system in TLUMIP2 is in effect a generalisation of the MEPLAN/TRANUS approach.

2.2 The DELTA package

This model was developed by David Simmonds Consultancy, MVA Consultancy and the Institute of Transport Studies, Leeds. It is not an integrated package, but a link of separate models. The overall aim of DELTA is to allow the development of land-use models which, in combination with appropriate transport models, enable users to study the future effects of both land-use and transport policies, singly or in combination, on both the land-use and transport markets (Simmonds, 2001). The processes modelled predict changes in

• the quality and quantity of floorspace available for occupation;
• household transitions and employment;
• the property market;
• the employment status of individuals;
• car ownership.

Figure 2.1 Urban and regional submodels in one forecast period

DELTA consists of six urban and three regional sub-models. The basic linkage of these in modelling one period of change is illustrated in Figure 2.1. The location or relocation model is the main locus of interactions both between activities and space and between land use and transport. Its main function is to predict the location of those activities that are mobile in this period, taking accounts of changes in accessibility, transport-related changes in local environment, area quality and the rent of space. In the model it is assumed that potentially relocating activities will remain in the same place, and newly formed or incoming households or jobs will locate in proportion to the existing distribution of the same activity unless there are changes in one or more variables listed above.

The operation of the model involves adjusting the prices until all the locating households and all the available housing are accounted for. Similar equations are applied to employment and commercial floorspace. The amount of housing left vacant may change as a result of changes in demand. The intersection of supply and demand yields a level of floorspace prices as well as a short-run stock of vacant units. The price of each type of floorspace in each zone has to be adjusted until all floorspace is accounted for:

The ratio of floorspace to activity (ie the inverse of density), the location of activities and the amount of floorspace left vacant are all sensitive to the rent. It is assumed that households are utility-maximising whilst employment is cost-minimizing.

The typical DELTA model involve:

• Many household types competing for housing;
• small number of employment types (sectors) competing for each of the rental, office and industrial floorspace types.
• Other uses (eg sports and leisure) are generally not modelled.

The MEPLAN and TRANUS models

• are different models to locate activities;
• locate all activities every 5 years;
• the rent adjustments are similar to those in DELTA.

3. Disaggregate, microsimulation models

3.1 Introduction

The basic feature of the microsimulation approach is the identification and representation of the individuals with some dynamic, adaptive behaviour producing individualized response on endogenous and exogenous stimuli. The focus shifts from sectors of the economy to the individual decision making units. Knowledge about individual behaviour and other actors is integrated into the model and the consequences of many individuals' behaviour or responses to external impacts are investigated.

Microsimulation models usually contain both deterministic and stochastic relations. Deterministic relations mirror unavoidable rules or strong logical or structural constrains and obviously produce the same result in every simulation with identical initial conditions. The error term of the regression equation can be used as an estimate of the undeterminable part of individuals' behaviour, and it is the focal point of microsimulation that this error is maintained in the microsimulation. The unknown part is replaced with a random number generator in the simulation. In such a stochastic model, the result of each simulation is different. Replicating the model execution many times with different random seeds gives information about overall prediction error. Impact of parameter changes can be directly compared with this unavoidable random variation.

The salient features of microsimulation packages such as UrbanSim, TLUMIP and IRPUD are an explicit accounting of land, structures and occupants and microsimulation of mobility and location choices of individual households and jobs, land development and redevelopment; and disaggregation of household and employment location choices. The package components usually include economic and demographic transition models, household and employment mobility models, household and employment location choice models, real estate development models, and accessibility and land price models. The microsimulation models incorporates a dynamic disequilibrium approach, representing adjustment processes that occur at different rates, unlike the cross-sectional equilibrium approach adopted by models such as MEPLAN and TRANUS - which can be difficult to sustain when considering the complex interactions among urban housing, labour and transportation markets, since equilibrium may not be realistic in a complex and evolving urban system.

The level of microsimulation for different packages is different. For example, UrbanSim considers land prices at a very fine cell level; TLUMIP considers rents only at the coarser zone level (with a separate, non-market process to allocate activities to cells); the older IRPUD model worked only at zone level, but the latest version of IRPUD works on cells.

3.2 UrbanSim

The initial design of the UrbanSim model was funded by the Oahu Metropolitan Land-Use Model as part of a larger effort to undertake the development of new travel models. The model was further developed in 1996 when the Oregon Department of Transportation launched the Transportation and Land Use Model Integration Project (TLUMIP) to develop analytical tools to support land-use and transportation planning.

Figure 3.1 Structure of UrbanSim ¹⁰ (Waddell, 1998).

UrbanSim is a software-based system designed to be used for integrated planning and analysis of urban development, incorporating the interactions between land use, transportation, and public policy. Figure 3.1 shows the key components of the model system. The model components represent the behaviour of households, businesses, developers, and governments, interfaced through the land market (Waddell, 1998). The model draws on random utility theory for its theoretical foundation, and builds on techniques of disaggregate choice modelling widely used in mode choice models. The model simulates land market clearing by adjustment prices to reconcile the competing demands for locations and structures among households and businesses against the supply of space in each zone. It is assumed that businesses are making individual choices about the location of each job, and are not constrained to moving the entire firm (Waddell et al).

The floorspace market clearing handles the assignment of moving businesses and households to their highest utility alternative that is available, and adjust land prices over time according to the ratio of demand to supply in each zone. The solution is based on an expectation of incomplete information and nontrivial transactions and search costs, so that movers obtain the highest satisfactory location that is available, and prices respond at the end of the year to the balance of demand and supply at each location.

The form of the price adjustment is:

Total demand is estimated from three components, total latent demand, current occupied space, and total vacant space (including potential vacancy to be created by out-movers):

3.3 TLUMIP/Oregon2

TLUMIP is an integrated transportation, land use and economic model for use in transportation planning and policy analyses at the regional and statewide levels. The model starts with an input-output economic model of commodities, the amounts correspond to the production and consumption of goods and services and trading relationships between sectors of the economy in the study area. Given economic activities and their location within the study area, the model then generates the travel required to support production and consumption of activities. These trips are assigned to travel the system via the most cost-effective (time and distance) path.

The Oregon2 model, the second generation of TLUMIP model, represents the behaviour of the land use transport system in the State of Oregon using a set of 7 connected modules that cover different components of the full system. The principal components of the model are shown in Figure 3.2. Note that in many respects the Oregon2 model combines the microsimulation approach to household change/location developed in UrbanSim with the aggregate approach to economic location/interaction developed in TRANUS. (The first generation of TLUMIP models involved applications of UrbanSim and TRANUS.)

Figure 3.2. The principal components of Oregon2 model (Upton, 2003).

The household allocations module is a fully disaggregate representation uses an agent -based microsimulation of each household and each component person to simulate the transitions and choices made by these agents in two-month steps. Almost all submodels are probabilistic, in that they have equations which assign a probability to various events, based on the household and person characteristics and the characteristics of the place where the household lives and where household members work. The probabilities are calculated and a random number generator is used to sample an outcome from the possible outcomes.

The residential space price update is applied to each non-zero quantity of each type of residential floorspace in each zone. It adjusts the unit price in reaction to the vacancy rate. The following equation is used:

The unit price for a given category of floorspace will stabilize at a non-zero value only when the vacancy rate equals the reference vacancy rate. If PFS_r,z( t) is greater than PFS_r then U( t) is proportional to PFS_r,z( t). If PFS_r,z( t). is less than PFS_r then U( t) is independent of PFS_r,z( t). (subject to zero constraint).

3.4 IRPUD

The IRPUD model was developed for the city of Dortmund by Michael Wegener and his colleagues (Wegener, 1982). A macromodel of economic and demographic change is used to simulate employment change by industrial sector and demographic changes by age, gender, and nationality within a set of labour market regions. Then, a mesoscopic spatial model is used to simulate intra-regional location of decisions of industry, residential developers and households. Finally, a microscopic model of land use development within statistical tracts is used to allocate the demand generated by the mesoscopic model.

Floorspace prices are adjusted as a function of the demand for floorspace expressed by the proportion of vacant units:

The function f (•) is an inverted S-shaped elasticity curve entered exogenously resulting in a reduction of floorspace prices if there is a large percentage of vacant floorspace of that type not occupied in the previous housing market simulation , and in a price or rent increase if there are no or only few vacant floorspace left. No attempt is made to determine equilibrium floorspace prices. The price adjustment model reflects price adjustment behaviour by landlords. If they reduce or increase prices too much, this will be corrected in the subsequent simulation period (usually every two years).

4. MUSSA

he MUSSA model is a disaggregate equilibrium model with an aggregate version also available. In MUSSA, which was developed by Martínez (2000, 2001), the spatial allocation of land uses is handled using a bid function. The model is not a fully integrated model of land-use, urban economy and transport, but can accept as input the total demand (growth) from households and firms and a transport model. Central to the model is to predict the location of households and firms and the resulting rents. To that effect a bid function is used, which is specified as a function of property attributes, zone attributes, transport attributes and consumer clustering variables. This approach is called the bid-choice approach, and is analytically derived from specifying the choice approach in terms of the consumers' willingness-to-pay. Operationally, a multinomial logit model is assumed.

The willingness-to-pay function

represents the maximum value the consumer or household h is willing to pay for a location ( v,i), described by z _vi to obtain a utility level U⁰ given a fixed income I _h and an exogenous set of prices for other goods P. The utility function V _h is adjusted to each consumer. The bids are expected to be lower than willingness-to-pay, that is B _hvi = WP _hvi-w _h, with w _h =0. The speculative term w _h is equal for all location options (which, we think, is not always the case in reality). It is assumed that the bids are given by B _{hvi =} B _hvi +e _hvi, where B _hvi is the deterministic component and e _hvi is a random term. It is assumed that the stochastic terms are independent and identically distributed Gumbel with scale parameter . Then the expected rent at location ( v,i) is

with y the Euler's constant (approximately 0.577). Note that this is not the direct formula to calculate the expected rent but an iterative process of solving the location problem. The published papers do not give all the equations used to carry out the iterative process or to calculate the willingness-to-pay function, but we suspect that the iterative process may be one of maximising the utility levels U ⁰ across all households.

There is a fuller version of the model (Martínez, 2003) which extends the assumption of stochastic behaviour to the real estate market, the supply side, including production economies of scope and scale. This represents a more sophisticated, longer-term equilibrium in which the supply of buildings is variable.

« Previous | Contents | Next »