Infrastructure and firm dynamics: a micro simulation approach

Abstract

Scope and research objective
The construction of significant new transport infrastructure projects, such as the A4-extension from Delft to Rotterdam, has long-term spatial economic effects that are difficult to understand. In general, the planning of investments in new transport infrastructure and the development of new urban areas are important for the quality of our future urban environment. First, urban planning involves decisions that have long term impacts. Next to that, cities are often faced with a scarcity of resources, both in terms of money and in space. The consequences of this difficult task are obvious such as housing shortages and congested infrastructures.
The long term impact of urban plans and the scarcity of resources should be a motivation for urban planners to make a balanced evaluation of alternative urban planning strategies. Integrated land use and transport models can help identify effective or sustainable policy solutions for urban development and are becoming increasingly disaggregate, both in the spatial environment and the behaviour of agents (Timmermans, 2003). The effect of transport infrastructure on firm dynamics is a specific but important issue in the evaluation of spatial policy alternatives.
The design of simulation models that can describe this effect is far from trivial and a number of issues can be identified. First, the existing simulation models that describe infrastructure effects on firm location do not represent individual firm behaviour. Second, the build up of agglomeration economies and spatial externalities are under represented in existing urban simulation models. Third, the geographic detail mostly is too limited to evaluate policy measures at a satisfactory level of detail.
This thesis tries to make a contribution to the development of simulation models that can make more valid projections of the effect of spatial policy options on firm dynamics and firm behaviour. The following objectives are formulated:
1. Develop a simulation model that adequately describes the effects of spatial and transport infrastructure plans on firm dynamics. The model is based on the following assumptions:
a. Firm-specific behaviour should be implemented allowing for heterogeneity in preferences and responses;
b. Accurate representation of spatial externalities by the inclusion of spatial attributes that are specified from urban economic theories;
c. Geographic detail allowing the specification of detailed location attributes and the implementation of detailed zoning schemes.
2. Support the assumptions made in objective 1. with empirical analyses. Many uncertainties exist about the influence of accessibility and agglomeration on the performance and preference of firms, and in particular how these influences relate to internal organisational factors. The driving factors behind firm dynamic developments need to be identified in an extensive set of micro-data on the firm population.
To achieve these objectives, the thesis has develop a dedicated theoretical framework based on a review of most recent theories, and established operational models estimated using empirical data from a large region in The Netherlands.
Approach
This research applies a dynamic micro simulation approach that quantifies the effects of different spatial and transport planning scenarios on the firm population and mobility. The Spatial Firm-demographic Micro-simulation (SFM) model simulates transitions and events in the firm population. These include firm relocation, firm growth, firm dissolution and firm formation. In each time step of a model run, the firm population is processed through the separate components of the firm demographic micro-simulation. Such a firm demographic micro-simulation approach is motivated by the need to improve the representation of firm behaviour in spatial economic models. This improvement is the result of three distinctive features of the spatial firm demographic simulation. First, it represents firm specific behaviour and heterogeneity in responses. Firm behaviour is influenced both by firm internal processes and attributes of the firm’s location. For example, the most important reason for firms to relocate is firm growth (Brouwer et al., 2004; Louw, 1996). Second, it allows distinctive accessibility measures as explanatory variables for each event, such as firm relocation, firm performance, formation or firm dissolution. For instance, some firms perform better in the proximity of motorway on-ramps (Hilbers et al., 1994), while in a relocation decision, accessibility is evaluated in a different way. A third advantage is the possibility to account for path dependency between events. For example, firm growth – e.g., triggered by a new motorway opening – might induce a firm relocation in the following years. By keeping record of the developments at the individual firm level, the life-cycle of a firm and the causality between subsequent events can be modelled endogenously.
The quality of locations is measured by advanced accessibility and agglomeration measures. First, accessibility is measured by proximity to infrastructure access points. These measures are directly observed by decision makers and identified in many empirical studies as significant location factors. Second, accessibility is measured by logsums of business and commuting trips from a transport model. These logsums are used to measure the accessibility to the labour and business markets. They measure the value of a choice set and are well founded in utility theory. Agglomeration measures are based on the theories from urban economics and integrate travel time range bands into measures of specialisation and diversification. Specialisation is included as a location factor to test if Marshallian locational externalities are significant, e.g. labour market pooling, input sharing or knowledge sharing within the same industry sector. Diversification is included to test if Jacobs location externalities are significant, in other words if firms benefit from knowledge spill-overs from ideas or innovations in other industry sectors.
The SFM-model features a firm location choice model that accounts for interdependencies between location alternatives that arise from the spatial dimension of the choice context. This spatial dimension distinguishes itself from a-spatial choice in several ways, such as that each decision maker has a spatial location, there are usually a large number of spatial alternatives and alternatives show a complex interdependency following from the continuous spatial dimension (Fotheringham, 1989; Guo, 2004). Therefore, the location choice model is based on a Competing Destinations model (Fotheringham, 1989) that accounts for spatial cluster membership of competing alternatives. It applies individual choice sets with a number of feasible firm locations.
Case study
To test the assumptions in the model, it is applied to a case study for the province of South Holland in The Netherlands. The area contains a firm population of approximately 90.000 firms, which are distributed across 70.000 6-character zip code locations. The SFM model has been developed in three stages: model design, model estimation, and model-validation. The model is estimated on an extensive longitudinal dataset with the full firm population from 1990 to 1996, the LISA data (National Information System of Employment). The model has been validated on data for the succeeding period from 1996 to 2004.
Various accessibility and agglomeration attributes are computed for each year of the calibration and validation periods, and linked to the location of each firm observation. The distance measures to transport infrastructure access points are computed form coordinate information. The logsum are derived from the Dutch National Modelling System (NMS, see Hague Consulting Group, 2000). The agglomeration attributes for diversification and specialisation are computed with travel times from the NMS and the location of all firms in the LISA data.
Estimation results
The calibration of the SFM-model consists of the separate estimation of all firm demographic sub models. The purpose of these estimations is to provide an empirical foundation of the parameters in the SFM model and to gain more insight into the different components of spatial economic development, and the possible interactions between these components.
The relocation probability can mostly be explained from the firm attributes, which is in line with firm demographic literature: bigger firms are less likely to relocate (Carroll and Hannan, 2000; Brouwer et al., 2004) and firms with relatively large growth rates are more likely to relocate (Carroll and Hannan, 2000; Pellenbarg, 1996; Louw, 1996). The relocation probability varies across industry sectors, but accessibility appears to have a limited influence. Agglomeration, however, does have an effect: firms at diverse locations are more likely to relocate. This is interpreted as a pattern of successful firms that leave their breeding areas.
When firms relocate and search for a new location, they have a significant preference for locations in the proximity of their original location. This is interpreted as evidence for keep-factors: a relocating firm strives to maintain its existing spatial network. Moreover, the spatial clustering of location alternatives proves to have a significant influence on the choice behaviour of firms: alternatives that are clustered in space, individually have a smaller choice probability. This is in line with findings by Pelligrini and Fotheringham (2002). Furthermore, outspoken differences in location preference between industry sectors are measured for highway and/or train station proximity. Moreover, some industry sectors have a preference for locations with a good labour market accessibility. The estimations provide strong evidence that firms prefer locations that have a relatively high representation of firms from their own industry sector. This is interpreted as evidence for the existence of Marshall externalities and consistent with the findings of Duranton and Puga (2000) and Holl (2004b).
The firm growth model revealed a dominant influence of the growth pattern in the previous years on the expected firm size, consistent with firm demographic literature (Dunne and Hughes, 1994; McCloughan, 1995; Audretsch et al., 2002; Van Wissen and Huisman, 2003). The estimated autoregressive coefficients were smaller than unity for all industry sectors, implying that large firms are expected to grow more slowly than small firms. All autocorrelation coefficients that are estimated have a negative sign, implying a negative correlation between firm growth in subsequent years. In other words, a firm with a relatively substantial growth in one year, is expected to grow less quickly the next year. Infrastructure proximity has a significant effect on firm performance for various industry sectors. Most estimated coefficients for diversity are negative, indicating relatively less growth at diversified locations. The estimation results for the specialisation coefficient reveal a positive influence of specialisation on the expected growth.
Firm dissolution can be explained for the largest part by the firm attributes: age and size. Firm dissolution is high among young firms. This confirms the liability of newness hypothesis (Stinchcombe, 1965). Larger firms have smaller dissolution probabilities, as do firms that are growing in size. Both findings are in line with empirical literature (Hannan and Freeman, 1977; Ekamper, 1996; Van Wissen, 2000; Brouwer, 2004). Next, structural developments within industry sectors are significant as well. Firm dissolution appears to be higher among firms at locations with higher diversity indices. Firm dissolution appears to be a little bit higher in the proximity of highway on-ramps (γ-locations), probably indicating higher dynamics at industrial sites at such locations.
In conclusion, the estimates confirm a complex interdependency between firm demographic events and confirm a significant influence of accessibility and agglomeration throughout the various events. In general, the estimations provide a consistent pattern that can be understood from an agglomeration and life-cycle perspective on firms, similar to Duranton and Puga (2000). The simulation approach that is presented provides a behavioural methodology to better describe and understand the effect of infrastructure and spatial planning scenarios on firm dynamics, taking into account theories on urban economics and industrial organisations.
Validation
To analyse the validity of the estimated models, the coefficients are implemented in the SFM model, and validated with firm data for a succeeding period. Other input to the validation runs consists of the firm population in the base year, the economic development and the real estate development scenario. In base year 1996, the study area hosted 95 thousand firms. The validation runs are based on the observed regional economic development in the study area between 1996 and 2004. The macro economic development is input to the firm growth model and forms constraints to the simulation and influences the outcomes at micro level. The supply of industrial real estate determines the constraints to what locations are available to a firm. Changes in this industrial real estate supply are implemented through synthesized real estate data for the corresponding period.
The validation of the SFM-model across different levels of aggregation yielded interesting results that lead to more questions as well. It is shown that the simulated firm population size follows the trend of the observed population size. However it averages out a more dynamic growth pattern over the years. A sample of microscopic results showed the effect of the stochastic elements in the firm-demographic models. These stochastics prove to lead to very distinct developments at micro level that are representative for the element of coincidence in firm-demographic events. At neighbourhood level, the results indicate a reasonable match between observed and simulated developments, but a systematic flaw was identified in small vs. large neighbourhoods. It is concluded that the presented approach provides reliable estimates of future firm location but further enhancement of the models is desirable. Despite the question raised in this validation, it is argued that this is no reason to reject the approach: the disaggregate nature of the approach raises questions but provides many ways of analysing these issues further.
Final conclusions
The spatial firm demographic micro-simulation model that is presented makes some contributions to the further improvement of models of firm behaviour. First, the approach is based on industrial organisation theory. The approach features individual firm characteristics in the models, and the representation of distinct events that are each differently influenced by accessibility. A second distinctive feature is the inclusion of urban economic theories and advanced agglomeration measures that until now have only been applied marginally in projection models. A number of disaggregate agglomeration attributes are included that measure the level of specialisation or diversification in the composition of the local firm population. The transport dimension is included in these measures through travel times from a transport model.
Policy makers should account for the composition of the firm population in their region: the future demand for firm location is a function of the existing firm population, and complicated developments that can be explained from the perspective of organisational and urban economic theories. In such a way mal-investments can be avoided and optimal urban plans can be identified. In such a way, the regional economy is supported by developing firm locations that better serve the (future) demand of the local firm population.