Modelling Freight Mode Choice Behaviour Accounting for Supply Chain Structures and Taste Heterogeneity

Transcription

1 Modelling Freight Mode Choice Behaviour Accounting for Supply Chain Structures and Taste Heterogeneity Kriangkrai Arunotayanun Wednesday, 28 January :00 Location: Room 609, Skempton (Civil Eng) Bldg, Imperial College London

2 Modelling Freight Mode Choice Behaviour Accounting for Supply Chain Structures Mr Kriangkrai ARUNOTAYANUN Transport Studies Department of Civil and Environmental Engineering Imperial College London CTS Seminar, 28th January 2009 Transport Studies

3 Outline Background Motivation Models Data Description Analysis and Results Supply Chain Variables Supply Chain Structures Conclusions Transport Studies 02/23

4 Background (1) Freight transport activities Key factors influencing the complexity of freight mode choice decisions Characteristics of freight Characteristics of firms Supply chains linking all vendors, service providers and customers Transport Studies 03/23

5 Background (2) Supply chain relationships OEM OEM OEM OEM OEM Arm s Hierarchical length Structure relationship Network Structure Full vertical integration Transport Studies 04/16 04/23

6 Background (3) Supply chain operations Shippers/ Manufacturers Correlation Retailers/ Customers Carriers Freight Forwarders Transport Studies 3PLPs logistical services logistical services logistical services 05/23

7 Motivation Existing modelling approaches Ignore the influence of supply chain and logistics concepts Rely on conceptual and methodological approaches developed in the passenger sector Conventional logit models Transport Studies 06/23

8 Models Generalised Extreme Value (GEV) P( a C) = e G ( e µ G( e V ( a) V (1) a V (1),..., e,..., e V ( n) V ( n) ) ) = e V ( a) + lng ( e V ( h) + lng h e C a V (1) h ( e,..., e V (1) V ( n ) ),..., e V ( n ) ) Car Bus Rail µ 1 =1.0 µ Car Bus Rail µ µ α 1 α 2 Transport Studies Car Bus Rail Multinomial logit (MNL) = h C G( Y 1,..., Y n ) Y Nested logit (NL) G( Y G( Y K µ h ( ) µ µ k µ,..., Y n = Y 1 ) k = 1 h C Cross-nested logit (CNL) K k h k ( ( ) ) µ µ µ k k 1 µ α,..., Y n = Y 1 ) k = 1 h C k hk h 07/23

9 Data Description (1) 2004 French shipper survey (ECHO) Transport Studies 08/23

10 Data Description (2) Establishment characteristics Pre-interview (postal mail) Establishment (face-to-face) 2,935 shippers Shipment characteristics (3 shipments per establishment) Operator (telephone) Shipment (face-to-face) Reconstitution of organisational and physical chains 10,462 shipments 9,738 chains Leg of journey (telephone) Transport Studies 09/23

11 Data Description (3) Level-of-service attributes 4 alternative land transport modes 542 shippers (38 business sectors) 1,095 shipments (1,080 completed chains) Variables relating to shipper, shipment and flow characteristics Transport Studies Transport cost (%) Delay (%) Own account road Rail Travel time (hour) For-hire road Combined road-rail 10/23

12 Analysis and Results Supply chain variables Unobserved correlation amongst different modes Degree of closeness (e.g. type of contract) Supply chain structures Unobserved correlation along two choice dimensions: transport mode and supply chain OEM Transport Studies 11/23

13 Supply Chain Variables Basic specification - CNL leads to an improvement in terms of model fit over MNL - There is a significant amount of correlation amongst alternatives - MNL underestimates the value of cost-time trade-off Transport Studies Attributes For-hire road Rail Combined Cost (%) Time (hr) Delay (%) µ µ fixed Cost-time trade-off Estm. Parameters Value MNL RO RH CB RL t-test CNL α=0.5 µ 1 µ 2 RO RH CB RL Value Final LL Adjusted ρ t-test /23

14 Supply Chain Variables Detailed specification - 20 additional explanatory variables are all statistically significant - There is no statistical difference in model fit between MNL and CNL - Unobservable correlation now becomes observable Transport Studies 13/23

15 Level-of-service attributes Shipper characteristics RH,RL,CB CB RO RL, CB RH,RL,CB RO CB RH Attributes For-hire road constant Rail constant Combined road-rail constant Cost (%) Time (hour) Delay (%) Annual tonnage shipped Combined used in the last 12 months No. of own truck; 3.5 tons Use of warehouse Contract type (long-term=2, equal=1, occasionally=0) Transport organised by shippers Transport organised by providers Access to domestic parking area Value MNL t-test Centre RHfor Zone type of parking; specific to freight Transport Studies 14/ Value CNL t-test

16 Attributes MNL CNL Value t-test Value t-test Shipment & flow characteristics RO Time of departure CB Time of departure RO Distance RH Fragile products RL, CB Fragile products RO Bulky products RL, CB Weight of shipment RH Shipment is a part of journey CB Shipment is a part of journey RH RFID or electronic labels CB RFID or electronic labels µ µ fixed Cost-time trade-off Estimated Parameters Final LL Adjusted ρ Transport Studies 15/23

17 Analysis and Results Supply chain variables Unobserved correlation amongst different modes Degree of closeness (e.g. type of contract) Supply chain structures Unobserved correlation along two choice dimensions: transport mode and supply chain OEM Transport Studies 16/23

18 Supply Chain Structures Shippers using Own account road (ShpRO) Directly contacted Operators (DOp) Large freight forwarders (Lff) Transport Studies Chaining Operators (COp) 17/23

19 Supply Chain Structures Supply chains ShpRO DOp COp Lff Transport modes ShpRO RH CB RL Transport modes RH CB RL RH CB RL RH CB RL Supply chains DOpCOp Lff DOpCOp Lff DOpCOp Lff Supply chains Transport modes ShpRO DOp COp Lff RH CB RL All α = 0.5 Centre DOpRH for DOpCB DOpRL COpRH COpCB COpRL LffRH LffCB LffRL Transport Studies 18/23

20 Supply Chain Structures - Both NL models do not statistically outperform the MNL model - Correlations exist amongst alternatives within the nests of Directly Contacted Operators chain and Combined road-rail mode - CNL provides the best model fit - MNL overestimates the value of cost-time trade-off Structural Parameters (µ) ShpRO Cost-time trade-off Estimated Parameters Final LL Adjusted ρ Transport Studies 19/23 DOp COp Lff RH CB RL MNL 24 Value CNL t-test fixed fixed CNL Re-estimated Value t-test fixed 2.65 fixed fixed fixed 3.47 fixed

21 Supply Chain Structures Choice elasticities - Shippers with complicated supply chains are more sensitive to time than shippers with simple chains - For-hire road: more competitive to the other modes in the same chain than the same mode in the other chains - Rail and Combined road-rail: more competitive to the same modes in the other chains than the other modes in the same chain Transport Studies 20/23

22 Conclusions (1) Apart from cost and time, logistical and supply chain attributes are also the major determinants of demand Unobserved correlation amongst mode and/or supply chain alternatives must be properly taken into account The study offers greater insight into shippers choice behaviour with respect to modes and supply chains Failure to properly account for these observed and unobserved supply chain influences leads to Degraded explanatory power of freight demand models Increased risks of misinterpreted results and violated policy implications Transport Studies 21/23

23 Conclusions (2) Future work Interdependent choice process amongst agents Models accounting for inter-alternative correlation and inter-agent taste heterogeneity simultaneously Correlation Taste heterogeneity Shippers/ Manufacturers Retailers/ Customers Carriers Freight Forwarders logistical services logistical services logistical services Transport Studies 3PLPs 22/23

24 Contact Information Mr Kriangkrai ARUNOTAYANUN Transport Studies Dept. of Civil & Environmental Engineering Skempton Building Imperial College London Exhibition Road London SW7 2AZ address: Transport Studies 23/23