Agent-Based Firmographic Models: A Simulation Framework for the City of Hamilton

1
Agent-Based Firmographic Models A Simulation
Framework for the City of Hamilton

• By
• Hanna Maoh and Pavlos Kanaroglou
• Workshop on modelling and microsimulating firm
  demography
• University College London (UCL), London, July
  2nd, 2005

2
Research Objectives

• Objective Simulate the evolution of business
  establishment population between t and t 1
• Improve upon existing methods used to model firms
  and jobs in conventional Land use and
  transportation models
• Adopt the agent-based approach to develop an
  agent-based microsimulation model
• Apply concepts from firm demography to model the
  evolutionary process

3
The Demography of Firms

• Firm demography is dedicated to the study of
  processes that relates to
• Formation of new firms (birth or entry)
• Failure of existing firms (death or exit)
• Migration of existing firms (local and regional)
• Growth and decline of firms
• It is concerned with identifying and quantifying
  the causes associated with firmographic processes

4
Evolutionary Process of Business Establishment
Population
5
Modeling Framework
6
Real Estate Market

• A market for industrial and commercial floor
  space at the parcel level drives the framework
• This market is influenced by
• The firmographic events
• Demand for floor-space is generated by the newly
  formed, relocating and in-migrating
  establishments
• Failure, departing current location
  (out-migration or intra-urban migration) free up
  floor space
• Growth, decline, merging and splitting also
  contributes to change in floor space
• Development and redevelopment practices influence
  the available floor-space

7
(No Transcript)
8
(No Transcript)
9
(No Transcript)
10
Firm Micro-Data Statistics Canada Business
Register (BR)

• Maintain annual information about business
  establishments in Canada since 1990
• Confidential and can only be used to conduct
  statistical analysis
• Attributes Establishment size, location (postal
  code and SGC), SIC code and Establishment Number
  (EN)
• BR provides the life trajectory of business
  establishments over space and time
• BR can be used to measure firmographic events
  such as the formation, migration, location
  choice, failure, growth and decline of business
  establishments

11
Small and Medium (SME) Size establishments

• SME with less than 200 employees is the target of
  our analysis
• Account for more than 94 of establishments in
  1990, 1996 and 2002
• Extracted population was constrained to
  self-owned single establishments
• Establishments that are part of a chain were not
  included in the model!
• However, the extracted sample is deemed
  appropriate
• Around 80 of SME are with less than 10
  employees, 93 of which are single owned
  establishments

12
Modeling Methods

• Use discretetime hazard duration models to
  explain the failure process
• Use multinomial logit models to explain the
  mobility (stay, relocate or out-migrate) of
  business establishments
• Use multinomial logit models to explain the
  location choice behavior of intra-urban mobile,
  newly formed and in-migrating establishments
  (maximum utility and Bid-rent concepts)
• Use multivariate regression models to explain the
  growth/decline process of business establishments

13
Failure SubmoduleExploring and modeling
survival and failure of establishments
14
Exploring Survival

• We follow the life trajectory of 1990 and 1996
  small and medium size establishments till 2002
• We determine the duration of survival and time of
  failure
• We explore variation in establishment survival by
  size, age, industry and geography
• Non-parametric survival curves suggests that
  size, age, industry and geography has an
  influence on the survival rates

15
(No Transcript)
16
Survival rates of the 1991 SME cohort by size
class
17
(No Transcript)
18
(No Transcript)
19
Failure Model

• We follow the life trajectory of 1996 SME cohort
  till 2002 to model the failure process of SME
  with less than 50 employees via a discrete time
  hazard duration model
• Pit(f) 1/(1 exp(-?t? xit))

• Geography specific variables
• Local Competition (ve)
• Agglomeration economies (-ve)
• Location dummies

• Firm specific variables
• Age (ve)
• Size (-ve) and Size-squared
• Growth (-ve)
• Relocation (-ve)

• Macro economic variables
• Unemployment rate (ve)
• Average total income (-ve)

• Industry specific variables
• Average size of industry (ve)
• Industry dummies

20
Estimation Results

• Firm specific variables
• Young and small establishments are more
  susceptible to failure
• Growing establishments are more likely to remain
  in business
• Relocation signals a superiority in performance
  either because it is undertaken to expand or as a
  reaction to location stress
• Geography specific variables
• Market power (competition) has a positive
  influence on failure
• Market share (agglomeration) has a negative
  influence on failure
• Suburban establishments are less likely to fail
  compared to those located in the core

21
Estimation Results

• Macro economic variables
• Economic downturn or low demand for services and
  goods lead to higher rates of failure
• High levels of demand for services and goods
  (purchase power) in the city decrease the
  propensity of failure
• Industry specific variables
• Small establishments in large industries are more
  likely to fail
• Failure vary by industry (Health and Social
  Services have the lowest rates of failure
  finance insurance services have the highest rates
  of failure)

22
Conclusions on the failure model

• Firm, geography, macro-economy and industry
  specific factors can explain failure with firm
  and macro-economic being the most influential
• The BR can be useful in developing agent-based
  firm demographic models
• Extension of the modeling framework to study the
  failure by economic sector may have a value added

23
Mobility SubmoduleExploration and modeling of
mobility trends
24
Mobility Trends

• 7 and 2 of 1996 SME establishments relocated
  and out-migrated by 1997, respectively
• 12 and 3 of 1996 total establishment population
  relocated and out-migrated by 2002, respectively
• Mean employment size of relocating establishments
  is 15 and mean relocating distance is 5
  kilometres (1996 2002)
• 50 of moves happened at short distance within
  the same municipality
• 91 of out-migrating establishments moved within
  a radius of 100 kilometres around Hamilton
  between 1996 and 2002
• 57 of out-migrants moved to close by location in
  the Greater Toronto Area

25
Establishment Mobility Model

• Objective Determine if an individual
  establishment will choose to Stay (S) at its
  current location, Relocate (R) to a different
  place within the city or will Leave (L) the city
  between 1996 and 1997
• We use a MNL model to predict probabilities P(S),
  P(R) and P(L)
• Mobility is modeled by main economic sector

26
Utility Specification for establishment i

• Establishment internal factors and location
  factors are used in the specification of the
  Stay, Relocate and move utilities
• Internal factors included Size, Age, Growth rate
  and dummies for type of industry industry_d
• Location factors included Geography dummies, a
  measure for agglomeration economies (Agglom),
  distance between old and new location (Dod) and a
  measure for location competition (Lcomp)

27
Overview of Results

• Mobility is more prominent among very small and
  very large establishments as depicted by the Size
  and Size2 parameters
• The Age parameter suggests that young
  establishments are more likely to relocate or
  out-migrate
• The need to grow as suggested by the Growth
  parameter push manufacturing establishments to
  relocate

28
Overview of Results

• The Growth parameter in retail and wholesale
  models appear as a proxy for performance since
  growing establishments were less mobile
• The location dummies suggest decentralization and
  suburbanization of establishments in Hamilton
• Mobility is more pronounced among the Central
  Business District (CBD) establishments

29
Overview of Results

• Agglomeration increases the propensity of
  inertia. This effect is more prominent among
  retail and service industry establishments
• The increase in local competition (location
  pressure) will push the establishment to move
  long distance
• Mobility vary by the type of industry as
  discerned from the specified industry dummies

30
Conclusions on the mobility model

• Mobility is not common place in the urban context
• Firm internal factors and location factors are
  important determinants of mobility
• The research emphasizes the value in using data
  from Statistics Canada Business Register to study
  firmography in the urban context
• More work need to be done to investigate the role
  of organizational structure on mobility
• Future research is still needed to thoroughly
  scrutinize the relation between public policy and
  establishment mobility behavior in the urban
  context Therefore, enhancing the attributes of
  existing firm micro data is required

31
Location choice submodule

• A micro-analytical modeling framework

32
Modeling Approach

• Establishments search the city for the location
  that will maximize their profit
• Searching pool relocating, new born and
  in-migrating establishments
• Measuring firmographic events
• Continuer establishments if for two consecutive
  years, the establishment has the same EN and the
  Hamilton SGC
• Relocating establishments if a continuer
  establishment has a different postal code address
  or coordinates between two consecutive years
• Newborn establishments if the establishment has
  an EN number in year t 1 which did not exist in
  year t
• In-migrating establishments Those with the same
  EN in two consecutive years, but with a different
  SGC, and an SGC in Hamilton for the later year

33
Representing space

• Bidding process
• Establishments in the pool will out-bid each
  other for a particular location which will be
  assigned to the highest bidder
• The bidding and maximizing profit processes can
  be modeled using discrete choice models
  (Martinez, 1992)
• Space at the micro-level
• Use boundaries of developed land parcels but
  postal code addresses has a one-to-many
  relationship with parcel
• Alternatively
• Divide the city into grid cells of 200 x 200
  meters extract grid cells that correspond to
  developed commercial and industrial land uses to
  create the set of alternative locations

34

• We employ a MNL model to handle the location
  choice decisions
• Creation of Choice set
• Grid cells resulted into a large choice set of
  2635 and 2855 alternatives (cells) in the two
  periods 1996-1997 and 2001 2002, respectively
• Therefore
• Random sample of alternatives (McFadden, 1978)
  9 randomly selected cells (locations) in addition
  to the chosen cell (location)
• Linear in parameter systematic utility Vni is a
  function of
• Location characteristics and establishment
  attributes

We model the location choice problem by major
economic sectors
35
Model Specification

• Model specification is based on information we
  gathered from the urban economic literature and
  the available data
• Location specific factors included
• Distance to CBD (CBDPRO)
• Main road and highway proximity (MRHWYPRO)
• Regional Mall proximity (MALLPRO)
• Measures of Agglomeration economies (AGGLOn) n
  is economic sector
• Geography classification Inner suburbs
  (MOUNTAIN) and outer suburbs (SUBURBS)
• Density of new residential development
  (NEWDEVELOP)
• Density of old residential development
  (OLDDEVELOP)
• Population density (POPDENS) and Household
  density (HHLDDENS)
• Household income density (HHLDINCDENS)
• Average Housing value density (AVGDWELLVAL)
• Percentage of a particular land use at a given
  location (LANDUSEk) k is type of land use
• Firm specific factors included
• Dummies to reflect firmographic event (NEWBORN)
  and type of industry the firm belongs to
  (INDUSTRYsic) SIC is 2-digit or 3-digit SIC code

36
Estimation Results

• Most firms in Hamilton prefer locating on land
  far away from the CBD
• Central location is important for
• Printing, publishing, and allied manufacturing
  firms (SIC 28),
• Communication and utilities firms SIC(48 49)
• Food, beverage drug and tobacco wholesale firms,
• Finance insurance, business services,
  accommodation food and beverages and other
  services
• New born manufacturing firms (i.e incubation
  plant hypothesis)
• Main road and highway proximity is important for
  all firms except for
• All construction firms except for Electrical work
  firms (SIC 426)
• Other product wholesale trade firms (SIC59)
• NEWBORN Health and social services AND
  accommodation food and beverage firms favor land
  in close proximity to main roads and highways
• Land in proximity to Regional Malls attracts
  retail trade firms specialized in food, beverage
  and drugs (SIC60), apparel, fabric and yarn (SIC
  61) and general retailing stores (SIC 65).
• Construction, communication and transportation
  firms avoid land in close proximity to regional
  malls

37

• Agglomeration economies is prominent in the city
  of Hamilton. All firms seems to appreciate the
  externalities associated with clustering in the
  local market
• All Construction firms except for electrical work
  firms (SIC 426) favor locating in the inner
  suburbs above the escarpment. Other services
  firms show affiliation of location in the inner
  suburbs area
• Wholesale trade and retail trade firms show
  evidence of suburbanization. This is true for
  all firms except for food stores (SIC 601),
  gasoline service station firms (SIC 633), motor
  vehicle repair shops (SIC 635) and general
  merchandize stores (SIC641)
• Construction, wholesale trade, retail trade, real
  estate, businesses, and accommodation food and
  beverage firms favor locations with new
  residential development
• Construction and retail show evidence of avoiding
  the location with old residential development

38

• Manufacturing firms avoid highly populated areas
• High Income Locations are attracting services and
  retail trade firms except for firms specialized
  in selling shoe, apparel fabric and yarn (SIC
  61), household furniture, appliances and
  furnishing retail (SIC 62) and automotive
  vehicles parts and accessories sales and services
  (SIC 63)
• Land use variables suggest that
• Construction, communication and transportation
  firms locate predominantly on open space land
• Manufacturing and communication firms favor
  locations with resource and industrial land use.
• Retail trade and services firms show high
  affiliation with commercial land use
• General merchandize Stores (SIC 64) and SIC(65)
  show affiliation with residential land use areas
  (i.e population oriented)
• Service firms also show affiliation with
  governmental and open space land uses

39
Location behavior over time An Example from the
retail sector
Estimation results of the 2001 2002 models
Suggest a consistency in the location choice
Behavior over time
40
Conclusion on the location choice model

• The research was successful in extending the
  conventional firm location modeling approach to
  study location choice behavior at the micro-level
• Results suggest a variation in the location
  choice behavior among firms from the different
  sectors
• The modeling approach was able to account for the
  heterogeneity in location choice behavior
• Results are consistent with the urban economic
  literature

41
Future Research

• Model implementation Creating a synthetic list
  of business establishments (BE) to use as a base
  year population for any simulation
• Develop a dynamic Geodatabase data model to
  store, maintain and update the BE list during
  simulations. Utilize Unified Modeling Language
  (UML) as the basis for the development
• Implement a real estate development model to
  predict the change in industrial and commercial
  floor-space
• Simulate the inter-play between the local economy
  and urban form in Hamilton

42
Acknowledgments

• We would like to thank Statistics Canada for
  supporting this research through their (2003
  2004) Statistics Canada PhD Research Stipend
  program.
• We would like to John Baldwin, Mark Brown and
  Desmond Beckstead for providing office space and
  access to the BR data. Also I am thankful to
  them for their useful discussions, input and
  assistance.
• We are grateful to Social Sciences and Humanities
  Research Council of Canada (SSHRC) for financial
  support through a Standard Research Grant and a
  SSHRC doctoral fellowship