Leveraging Transaction Data for Online Pricing and

13 downloads 0 Views 3MB Size Report
Jun 3, 2016 - Training data – 30 respondents (270 obs.); ... Rkj = {-1/(βPrice + βPrice-ME)} * (βLoaded Miles*Loaded Milesj + βOne-Day Lead*1-Day Lead ...
Leveraging Transaction Data for Online Pricing and Sourcing of Carrier Capacity in the Truckload Spot Market: Models and Application Hani  S.  Mahmassani  

Rotterdam School of Management Erasmus University June 3, 2016

OUTLINE   1.  Mo.va.on:    Value  of  informa.on  in  vola.le  environments–    The   Transporta.on  Spot  Market   Based on PhD Dissertation 2.  3PL  Broker  Model   of Christopher Lindsey 3.  Online  Pricing  Engine  

4.  5.  6.  7. 

• 

Model  Framework  

• 

Behavioral  Model  Components  

Part I conducted in collaboration with Diego Klabjan, and ECHO Global Logistics

Applica.on  to  Actual  3PL  Data   Improving  Sourcing  Process  through  Bundling   Data:        State  Choice  Experiment   Model  Es.ma.on  and  Results   • 

Reserva.on  Price  Calcula.ons  

8.  Takeaways  and  Q&A  

The  Broader  Context   §  The so-called “Data Revolution” -  Analytics, big data, and data mining (among other associated terms) have become pervasive in the practices of a number of businesses.

§  Though common in many industries, logistics has been slow to catch up. This is now changing. §  Logistics companies have access to substantial databases of transactional and behavioral data. Happening at a company near you?

The  Transporta2on  Spot  Market  

§  The  spot  market  exists  to  facilitate  unfulfilled  demand  due  to  the   uncertainty  in  shipper-­‐carrier  rela.onships  for  dedicated  services.   -  Supply  chain  uncertainty   -  Shipper-­‐Carrier  contract  structure   1

Kirkeby, K. 2007. Transportation: Commercial, Industry Surveys. Standard & Poor’s, New York. Tsai, M.-T. et al. 2011. Valuation of freight transportation contracts under uncertainty. Transportation Research Part E 47 (6). Caplice, C., Sheffi, Y. 2006. Combinatorial auctions for truckload transportation. In: Cramton, P., Shoham, Y., Steinberg, R. (Eds.), Combinatorial Auctions, Cambridge: MIT Press (Chapter 21). 2 3

Supply  Chains  and  Shipper-­‐Carrier  Contracts   •  Supply chain uncertainty is the difference between perception (forecast demand) and reality (actual demand) 1 •  The freight hauled by a carrier may differ significantly from what was awarded in the contract, because it was based on a forecast 2 Contract  Structure   •  Contracts  are  op#ons  that   give  shippers  the  right  but  not   the  obliga#on  to  a  carrier’s   services  2   –  Volume  commitments  are   frac.ons  of  traffic  flow  

•  Carriers  may  refuse  a  frac.on   of  a  shipper’s  requests   –  Typically  70-­‐80%   1 2

Chopra, S., Meindl, P. 2001. Supply Chain Management: Strategy, Planning, and Operation. Upper Saddle River, NJ: Prentice-Hall. Caplice, C., Sheffi,Y. 2006. Combinatorial auctions for truckload transportation. In: Cramton, P., Shoham,Y., Steinberg, R. (Eds.), Combinatorial Auctions, Cambridge: MIT Press (Chapter 21).

Characteristics of the Spot Market Characteristic

Description

Purpose

A mechanism by which fluctuations in the trucking market are facilitated. It exists to serve urgent or unfulfilled demand. 1

Physical form

Electronic marketplaces (i.e., load boards, shipper-carrier matching sites)

Typical transactions

Spot market contracts are on a load-by-load basis, as opposed to the lane-by-lane contracts common to transportation auctions for dedicated services.

Supply and Demand Characterized by capacity uncertainty and price volatility Size

1 Tsai, M.-T. et

•  In 2010, 15% to 20% of total truck tons was moved on the spot market 2 •  In 2012, it accounted for $141.8 billion of the 3PL sector; of that, non-asset based providers (i.e., brokers) accounted for $45.1 billion. 3

al. 2011.Valuation of freight transportation contracts under uncertainty. Transportation Research Part E 47 (6). TransCore Freight Solutions. The Spot Market: A Primer, for Shippers. http://www.dat.com/~/media/81E66B9B31424E4888BD48067619DFA5.ashx. Accessed 10/25/2013. 6 3 Armstrong & Associates. U.S. 3PL Market Size Estimates. http://http://www.3plogistics.com/3PLmarket.htm. Accessed 10/2/2013. 2

Role  of  Non  Asset-­‐Based  3PLs  in  the  Spot  Market   §  Non  asset-­‐based  3PLs  own  no  physical  assets   related  to  the  physical  distribu.on  of  freight   -  i.e.,  rolling  stock,  warehouses,  etc.   §  Match  supply  (carriers’  capacity)  with  demand   (shippers’  shipments)  for  a  price   §  Brokers  exist  in  order  to  facilitate  the  spot  market   and  provide  a  degree  of  trust  and  accountability   §  Increasingly  rely  on  technology  and  informa.on  to   remain  compe..ve   §  Example  companies   -  Coyote  Logis.cs   -  Echo  Global  Logis.cs   -  CH  Robinson  

3PL  Broker  Business  Model   §  Quote  prices  to  shippers  before  securing  capacity  from  carriers   §  Consider  the  following  example:   The broker ends the search and selects the carrier expected to generate the highest profit - $20.

Carrier #1 - $105

Shipper

Broker quotes the shipper $100

Carrier #2 - $80

Broker

Carrier #3 – $85 Time

Research  Mo2va2on  &  Objec2ves   Mo2va2on   §  Pricing  shipments  and  sourcing  capacity  in  the  spot  market  are  difficult   tasks;  3PLs  some.mes  make  unprofitable  deals.   -  Spot  market  is  highly  vola.le   -  Rela.vely  short  opera.ng  .me  frames  

-  Uncertainty  on  both  sides  of  the  market  

§  However,  the  3PL  has  informa.on  to  use  to  its  advantage   -  Firsthand  knowledge  of  both  sides  of  the  market   -  Tendencies  of  shippers  and  carriers  through  transac.onal  data  

§  How  can  that  informa.on  be  used?   Objec2ve   §  Develop  methodological  frameworks  that  improve  the  outcomes  of  3PL   pricing  and  sourcing  decisions  for  truckload  shipments  using  transac.onal   data  and  relevant  transporta2on  and  logis2cs  theories.  

Online  Pricing  Engine:     Shipper  Acceptance  and  Carrier  Ranking  

Online  Pricing  Engine:     Shipper  Acceptance  and  Carrier  Ranking  

• Shipper acceptance decreases with increasing cost while carrier acceptance increases with increasing price. • Probabilities are estimated using the variables created in the data mining process.

Objec2ves   Predic.ve  analy.cs  for  real-­‐.me  3PL  broker  decisions  of:   -  Pricing  –  recommend  to  3PL  brokers  shipper  quotes  that  are   reasonable  and  poten.ally  profitable   -  Sourcing  –  determine  the  poten.al  carriers  that  give  the  best   opportunity  for  a  profitable  spot  market  transac.on  in  real  .me  

Background   §  Sparse  literature  on  3PL  spot  market  pricing  and  sourcing   -  Pricing   •  Cheng  and  Qi  (2011)  –  Op.mal  3PL  pricing  decision  in  a  single   supply  chain  

-  Sourcing   •  Caplice  &  Sheffi  (2003)  and  Sheffi  (2004)  –  benefits  of  economies   of  scope  and  non-­‐price  variables  in  auc.ons   •  Song  &  Regan  (2003)  –  Combinatorial  auc.on  for  simultaneous   bidding  over  several  shipping  lanes   •  Figliozzi  (2004)  –  Sequen.al  auc.ons  to  model  an  ongoing   transporta.on  spot  market  

Background  (cont.)   §  Non-­‐asset  3PL  spot  market  procurement  process  is   fundamentally  different   -  Search  process  rather  than  a  tradi.onal  auc.on  

§  Proposed  framework  func.ons  in  real  .me   §  Non-­‐asset  3PLs  are  an  important  industry  segment  that  is   largely  unexplored  

Conceptual  Framework:     Behavioral  Models  

• Shipper acceptance decreases with increasing cost while carrier acceptance increases with increasing price. • Probabilities are estimated using the variables created in the data mining process.

Data   §  Data  is  from  a  U.S.-­‐based  3PL  provider   §  Historical  shipments  with  informa.on  on:   -  Origin  and  des.na.on   -  Price   -  Equipment  type   -  Number  of  stops   -  Hazardous  material  status   -  Etc.  

§  The  data  is  processed  and  enters  the  Pricing/Sourcing   framework  

Conceptual  Framework:  Model  Development:   Behavioral  Models   Yi = {0,1}

Rejection  or   acceptance

Y = f ( Shipper / Carrier, Lane, Market, etc )

Choice   attributes

N

1−Yi

L(b ) = ∏ PiYi [1 − Pi ] i =1

N

LL ( b) = log !" L ( b)#$ = ∑Yi log Pi + (1− Yi ) log (1− Pi ) i=1

exp( X i , β ) Pi = Pr (Yi = 1) = 1 + exp( X i , β )

Likelihood   function Log-­‐likelihood   function

Logit  model

… with further specifications to account for the choice-restricted nature of the data.

Conceptual  Framework:  Model  Development:   Profit  Maximiza2on  

Conceptual  Framework:  Model  Development:   Profit  Maximiza2on   z = max pS ≥0 E [ Profit | pS ]

• Based on the behavioral curves we calculate the expected profit. • Expected profit takes into account the shipper price and a distribution of simulated minimum carrier prices.

Variables included in shipper and carrier acceptance choice models

Valida2on  Results   §  The  pricing  tool  is  validated  for  both  shippers  and  carriers   -  Shipper  prices  are  validated  using  historical  revenue  data   -  Carrier  prices  are  validated  using  average  market-­‐to-­‐market   rates  from  a  third  party  

Valida2on  Results:  Carrier  Prices  

The simulated carrier prices produced by the tool are comparable to both the third-party and historical prices, especially for shipment-types with many observations.

Valida2on  Results:  Carrier  Prices   Histogram of the difference between the model and third party carrier rates

Price Difference = Third party carrier rate – Model-estimated rate

Valida2on  Results:  Shipper  Prices  

The shipper prices produced by the tool are comparable to historical prices, though more tightly distributed about the mean.

Profit  Analysis   §  We  compare  the  profits  the  Pricing  Tool  could  have  generated   to  historical  profits   §  Each  metric  –  pricing  tool  profits  and  historical  profits  –  is   weighted  by  the  likelihood  of  the  deal  occurring  

Profit  Analysis:  Difference  Between  Pricing  Tool   and  Historical  Profits  

In general, the tool produces profits that are slightly higher than historical profits.

Weighted Profit Difference = Weighted Ideal Profit – Weighted Historical Profit

(

)(

WeightedIdeal Pr ofit = Pr Shipper Pr iceOptimal ⋅ Shipper Pr iceOptimal − Carrier Pr ice Historical

(

)(

)

WeightedHistorical Pr ofit = Pr Shipper Pr ice Historical ⋅ Shipper Pr ice Historical − Carrier Pr ice Historical

)

Valida2on  Results:  Profit  Analysis   25th75thStd. Percentile Percentile Deviation

Profit Measure Mean Profit per Mile of Profitable Shipments 8.76% Loss per Mile of Unprofitable Shipments 74.76%   Total Change   Profit Measure Profitable Shipments -5.41% Unprofitable Shipments 5.41% Total Profit 4.90%

56.49%

15.39%

-41.99%

88.97%

-70.79%

79.45%

 

 

Results based on a sample of 54,805 shipments. Model calibrated to produce prices and acceptance rates comparable to what was historically observed.

Valida2on  Results:  Profit  Analysis  

Profit Measure Profit per Mile of Profitable Shipments Loss per Mile of Unprofitable Shipments

Profit Measure Total Profit

Mean

25th75thStd. Percentile Percentile Deviation

$0.51

$0.31

$0.66

$0.30

-$0.52   Total Change   +$581, 876

-$0.67

-$0.10

$0.70

   

3PL Broker Business Model §  Quote prices to shippers before securing capacity from carriers §  Consider the following example: The broker ends the search and selects the carrier Carrier #1 - $105 expected to generate the highest profit - $20.

Shipper

Broker quotes the shipper $100

Carrier #2 - $80

Broker

Carrier #3 – $85 Time

29

Opportunities for Improvement §  The search for capacity could be improved if: -  Brokers have a good guess of the minimum price each potential carrier would demand before the search for capacity begins -  Brokers could source multiple shipments simultaneously, taking advantage of the economies of scope present in the spot market 1 Load #1 CHI

Load #2

Load #1

CHI

CHI

ATL BHM Chicago à Atlanta, Birmingham à Chicago 30

Load #2

CHI

ATL Chicago à Atlanta + Birmingham à Chicago

BHM

Caplice, C., Sheffi, Y. 2006. Combinatorial auctions for truckload transportation. In: Cramton, P., Shoham, Y., Steinberg, R. (Eds.), Combinatorial Auctions, Cambridge: MIT Press (Chapter 21). 1

Framework for Sourcing Capacity on the Spot Market Reservation Price Function

Set of Shipments to be Sourced

Shipment / Bundle attributes Shipment Assignment

Set of Available Carriers

• Optimal Shipment Assignments • Expected Reservation Prices / Profits

Carrier attributes

Time

§  The framework is as powerful as a user’s ability to segment carriers according to the likely levels of their reservation prices §  The rest of the discussion will focus on demonstrating how a meaningful reservation price function may be developed

31

Spot Market Stated Choice Experiment Because real-world data was not known to exist, a hypothetical field experiment was performed You have two trucks that are each available to make a move. Currently, Truck #1 is located in Rockford, IL and Truck #2 is in Dayton, OH. A shipper has just contacted you to inquire about transporting the following loads: Shipment 1 2

Origin

Des2na2on

Pick-­‐up  Date Empty  Miles  (Travel  Time)   Loaded  Miles  (Travel  Time)   to  Origin to  Des2na2on Chicago,  IL Cincinna.,  OH Today 90  mi  (2hrs.)  for  Truck  #1 295  mi  (5  hrs.)  for  Truck  #1 Columbus,  OH Atlanta,  GA Tomorrow 70  mi  (1  hr.)  for  Truck  #2 568  mi  (9  hrs.)  for  Truck  #2

You may choose to accept one, both, or none of the shipments offered. If you choose both, it is possible to use Truck #1 for both moves. Shipment Empty  Miles  (Travel   Time)  to  Origin  1 90  mi  (2hrs.) 1&2

Loaded  Miles  (Travel  Time)   Empty  Miles  (Travel   Loaded  Miles  (Travel   to  Des2na2on  1 Time)  to  Origin  2 Time)  to  Des2na2on  2 295  mi  (5  hrs.) 107  mi  (2  hrs.) 568  mi  (9  hrs.)

Below are your options for which shipment(s) and at what payment (for the linehaul portion of the trip) you choose to accept from the shipper. Please choose the best available option. Choose only one. a)  Shipment (1) only for $1000 ($3.39 per mile) – using Truck #1 for the move; b)  Shipment (2) only for $600 ($1.06 per mile) – using Truck #2 for the move; c)  Shipments (1) and (2) for $1520 (Combined $1.76 per mile) – using either Truck #1 for both moves, or both Trucks #1 and #2 individually; d)  None. I will pass on this shipper and wait for something better.

§  Submitted to approximately 400 truckload carriers with a response rate ≈ 11% §  Each respondent evaluated 9 pricing scenarios 32

-  N = 45; J = 9;

N x J = 405 observations

Randomized Survey Design Pricing Scenario Data

Carrier Submits Preliminary Information

Database Query via Web Service

Base Shipment Data v Origin-Destination of Shipments 1 & 2 v Current truck locations v Loaded and empty miles Random Shipment Data v Price of Shipments 1 & 2 v Bundle discount v Lead time

Carrier Completes Randomized Pricing Scenarios

Randomized Pricing Scenarios

• Respondent enters preliminary information • Web service places a call to a relational database consisting of base pricing scenarios • Base pricing scenarios are shipment O-Ds and current locations of the carrier’s hypothetical trucks

• Web service randomizes shipment attribute data for all scenarios and returns them to the online survey tool Survey administered using Qualtrics (www.Qualtrics.com)

End Time 33

Variables Derived from Survey Results Variable

Definition

Shipment Variables Loaded Miles

Number of miles shipment is actually transported

Empty Miles

Number of miles associated with a shipment not related to its actual transport

Cross Empty Miles

Number of miles from the drop-off location of one shipment and the delivery location of the next shipment

Demographic Variables

34

Individual experience

Number of years a respondent has been in the logistics industry

Firm size

Size of the carrier by number of employees

Firm fleet size

Number of vehicles under control by the carrier

Firm tenure

Number of years a firm has been in the logistics industry

Survey Summary Statistics Single Shipment Offers Variable Loaded Miles Empty Miles Price per Mile Price

Lead Time

Min. 300 0 0.49 212

Bundle Offers

Mean

Max.

397 31.9 1.39 547

Same Day 177

Variable

577 126 2.76 1,417

1 Day 344

2 Days 163

Min.

Loaded Miles   Cross Empty Miles   Price per Mile   Price  

600   0   0.73   587  

Mean 793   112   1.33   1,049  

Same Day   169

Lead Time  

Definition and Frequency by Category Variable Experience Firm Size Firm Fleet Size Firm Tenure

35

Low

Moderate

High

Less than 7 years

7 to 10 years

More than 10 years

Less than 250 employees

250 to 500 employees

More than 500 employees

Less than 250 trucks

250 to 1,000 trucks

More than 1,000 trucks

Less than 7 years

7 to 10 years

More than 10 years

Variable

Low

Individual Experience

0.103

0.342

0.526

Firm Size

0.237

0.237

0.526

Firm Fleet Size

0.579

0.211

0.158

0.0263

0.395

0.579

Firm Tenure

Moderate

Max.

High

1,090   213   2.31   2,120  

1 Day   2 Days   163   --  

Characterizing Carrier Behavior §  Reservation price is the least price at which carrier k is willing to sell a transportation service, in our case capacity k m j

denotes  the  pricing  scenario

Pm

denotes  the  price  of  alternative  m

p

Compensation  offered  to  motor  carrier  k  to  provide  transportation  service  S  

denotes  the  Carrier Alternative  available  to  carrier  k  as  part  of  the  offer

(1) U mk (S, p) −U ( 0, 0) ≡ 0

The  reservation  price,  Rk(p),  is  the  price  at  which  motor  carrier  k     sells  the  transportation  service  and  nets  a  zero  gain  in  utility

{

}

1 (2) p − R k ( p ) ≥ max p − R k (P1 ),..., PM − R k (PM )

(3) p ≥ R k (p )

{

Offer  price  must  exceed  the  reservation  price

}

(4) max p1 − R k (P1 ),..., PM − R k (PM ) < 0

36

Surplus  must  be  greater  than  all  other   alternatives

Carrier  will  refrain  from  all  alternatives  if  they  do   not  offer  more  utility  than  the  ‘No  choice’  option

Jedidi, K. and Zhang, Z.J. (2002) Augmenting Conjoint Analysis to Estimate Consumer Reservation Price. Management Science 48 (10).

Random Parameters Mixed Logit for Panel Data U kmj = Vkmj + ukmj

Utility  yielded  to  motor  carrier  k  from  the  surplus  of  alternative  m

ukmj ~ i.i.d . Gumbel

Random  portion  of  utility  is  distributed  iid  Gumbel

Jk

Conditional  likelihood  of  motor  carrier  k  selecting   among  the  four  alternatives  in  price  scenario  Jk

Lk | β = ∏ Qkmj j =1

Qkij = e

β `xkij

∑e

β `xkmj

Logit  probability  of  the  i  th  is  selected

m

Lk = ∫ (Lk | β )⋅ f (β )dβ β

2 ⎛ c11 Ω = C C = ⎜⎜ ⎝ c11c12 T

c11c12 ⎞ ⎟ 2 2 ⎟ c12 + c 22 ⎠

1 Train, K. (1998)

37

Unconditional  likelihood  of  the  observed  choice   sequence  for  carrier  k • Allowing β to vary across the population induces correlation across alternatives and scenarios • Additionally, we allow for correlation among the random parameters • Coefficients for Price, and for Loaded Miles specified as random

Recreation Demand Models with Taste Differences Over People. Land Economics 74 (2). Estimation of multinomial logit models in R: The mlogit Packages.

2 Croissant, Y. (2013)

Variable Types and Definitions Variable

Type

Model 1

Model 2

Single

Indicator

ü

ü

ü

Bundle

Indicator

ü

ü

ü

Price

Continuous

ü

ü

ü

Loaded Miles

Continuous

ü

ü

ü

Cross Empty Miles

Continuous

ü

ü

ü

Lead Time

Categorical (Same day, 1 day, 2 days)

ü

ü

ü

Experience

Categorical (Low, Moderate, High)

ü

ü

ü

Firm Size

Categorical (Small, Moderate, Large)

ü

Fleet Size

Categorical (Small, Moderate, Large)

Tenure

Categorical (Low, Moderate, High)

ü

Note: Underlined categories denote the reference levels in the discrete choice analysis. 38

Model 3

ü

Deterministic Utility Specifications Model 1

Model 3

Variable 1-Shipment Bundle Price Price * Low Experience Price * Mod. Experience Loaded Miles

Parameter βSingle βBundle βPrice βPrice-LE βPrice-ME βLoaded Miles

Parameter βSingle βBundle βPrice βPrice-LE βPrice-ME βLoaded Miles

βSingle βBundle βPrice βPrice-LE βPrice-ME βLoaded Miles

Loaded Miles * Mod.-Sized Firm Loaded Miles * Large-Sized Firm Loaded Miles * Mod.-Size Fleet Loaded Miles * Large-Size Fleet

βLoaded Miles-MF βLoaded Miles-LF ---

--βLoaded Miles-FM βLoaded Miles-FL

-----

Loaded Miles * Mod. Firm Tenure --

--

βLoaded Miles-TM

Loaded Miles * High Firm Tenure --

--

βLoaded Miles-TL

Cross Empty Miles * Bundle Lead Time 1-Day Lead Time 2-Days 39

Model 2

Parameter

βCross Empty Miles βCross Empty Miles βCross Empty Miles βOne-Day Lead βOne-Day Lead βOne-Day Lead βTwo-Day Lead βTwo-Day Lead βTwo-Day Lead

Discrete Choice Analysis Results Model 1

Model 2

Model 3

Significance

Significance

Significance

**** ****

* **** ****

****

****

Loaded Miles * Mod.-Sized Firm

--

--

LoadedMiles * Large-Sized Firm

--

--

**

--

Variable 1-Shipment Bundle Price Price * Low Experience Price * Mod. Experience Loaded Miles

**** * *** ****

LoadedMiles * Mod.-Size Fleet

--

Loaded Miles * Large-Size Fleet

--

Loaded Miles * Mod. Firm Tenure

--

--

Loaded Miles * High Firm Tenure Cross Empty Miles * Bundle Lead Time 1-Day Lead Time 2-Days Log-Likelihood LRI

--

--

-263.91 0.360

-262.94 0.363

§ 

Results statistically significant at the 5% level or better

§ 

Indicate that carriers are most influenced by offered payment and amount of loaded miles -  Single shipment elasticities for ‘Price’ and ‘Loaded Miles’ are 0.0011 and -0.0022 -  Bundled shipment elasticities are 0.0002 and -0.0004

-****

§ 

Results are intuitive -  Offer price reflects a carrier’s revenue for transporting a shipment

-330.64 0.199

Significance codes: [0, 0.001) = ‘****’, [0.001, 0.01) = ‘***’, [0.01, 0.05) = ‘**’, [0.05, 0.1) = ‘*’. 40

-  Distance is the primary driver of operational costs.

Discrete  Choice  Analysis  Results   Model  1   Variable  

Model  2  

Coefficient   Std.  Error  

1-­‐Shipment   Bundle   Price  

Sig.  

   

Coefficient  

Model  3  

Std.  Error   Sig.  

Coefficient  

Std.  Error   Sig.  

0.269  

0.895  

0.503  

0.904  

-­‐0.363  

0.886  

1.24  

1.68  

1.59  

1.68  

3.13  

1.64  

0.0108  

0.00146  

***  

0.0224  

0.00226   ***  

0.00385  

0.00099  

***  

-­‐0.00367  

0.00104   ***  

0.000427  

0.00069  

0.0105  

0.00143   ***  

Price  *  Low  Experience  

0.00188  

0.00097  

.  

Price  *  Mod.  Experience  

0.00204  

0.00071   **  

0.00076  

0.00056  

Loaded  Miles  

-­‐0.0203  

0.00331   ***  

-­‐0.0217  

0.00349  

LoadedMiles  *  Mod.-­‐Sized  Firm  

0.00154  

0.00104  

-­‐-­‐  

-­‐-­‐  

-­‐-­‐  

-­‐-­‐  

-­‐-­‐  

-­‐-­‐  

LoadedMiles  *  Large-­‐Sized  Firm  

-­‐0.00131  

0.00105  

-­‐-­‐  

-­‐-­‐  

-­‐-­‐  

-­‐-­‐  

-­‐-­‐  

-­‐-­‐  

-­‐-­‐  

-­‐-­‐  

-­‐-­‐  

-­‐-­‐  

-­‐-­‐  

-­‐-­‐  

***  

-­‐0.0398  

0.00436   ***  

LoadedMiles  *  Mod.-­‐Size  Fleet   -­‐-­‐  

-­‐-­‐  

-­‐-­‐  

-­‐0.00217  

0.00095  

Loaded  Miles  *  Large-­‐Size  Fleet  -­‐-­‐  

-­‐-­‐  

-­‐-­‐  

0.00096  

0.00107  

LoadedMiles  *  Mod.  Firm   Tenure  

-­‐-­‐  

-­‐-­‐  

-­‐-­‐  

-­‐-­‐  

-­‐-­‐  

-­‐-­‐  

0.00910  

0.00164   ***  

Loaded  Miles  *  High  Firm   Tenure  

-­‐-­‐  

-­‐-­‐  

-­‐-­‐  

-­‐-­‐  

-­‐-­‐  

-­‐-­‐  

0.00205  

0.00143  

Cross  Empty  Miles  *  Bundle  

*  

.  

-­‐0.00320  

0.00299  

-­‐0.00326  

0.00298  

-­‐0.00071  

0.00306  

Lead  Time  1-­‐Day  

-­‐0.0939  

0.299  

-­‐0.115  

0.302  

0.0318  

0.290  

Lead  Time  2-­‐Days  

0.0262  

0.435  

-­‐0.0194  

0.435  

0.532  

0.434  

cor(Price)   cor(Price,  Loaded  Miles)   cor(LoadedMiles)   Log-­‐Likelihood   LRI  

-­‐0.00722  

0.00188   ***  

-­‐0.00705  

0.00173  

***  

0.0564  

0.00829   ***  

0.0141  

0.00316   ***  

0.015  

0.003  

***  

-­‐0.102  

0.0144   ***  

0.00626  

0.00093   ***  

0.006  

0.001  

***  

-­‐0.00786  

0.00108   ***  

-­‐263.91  

   

-­‐262.94  

   

-­‐330.64  

0.360  

   

0.363  

   

0.199  

Significance codes: [0, 0.001) = ‘***’, [0.001, 0.01) = ‘**’, [0.01, 0.05) = ‘*’, [0.05, 0.1) = ‘.’.

Calculating Reservation Price §  Recall that the primary goal is to estimate reservation prices -  Let W define the set of non-price parameters, where w = 1 … W. -  Let zw denote the non-price variables, and β Price and βW denote the price and non-price parameters, respectively Rkmj

W

⎞ β z = ⎛⎜ − 1 w w β Pr ice ⎟⎠∑ ⎝ w

»  Reservation prices are inferred by taking the sum of the non-price parameters and multiplying by the reciprocal of the appropriate price parameter(s).

42

Reservation Price Validation Procedure §  Decision Rule: Carriers select the option that yields the highest utility (i.e. surplus) §  Criterion: Reproduce observed choice frequencies over several hundred simulations §  Validation procedure 1.  Delineate Training and Holdout data sets of carrier choice scenarios 1.  Training data – 30 respondents (270 obs.); Holdout data – 15 respondents (135 obs.)

43

2. 

Draw values for random parameters and error terms from their respective distributions

3. 

Calculate utility for all choice alternatives presented to carriers in the Holdout data set

4. 

Determine the preferred alternative based on the decision rule

5. 

Repeat steps 2-4 several hundred times

6. 

Calculate the simulated choice frequencies

Reservation Price Validation Results Shipment 1

Shipment 2

Bundle

None

Observed Choice Frequencies 0.0815

0.189

0.211

0.519

Simulated Choice Frequencies Model 1

0.0900

0.151

0.170

0.588

Model 2

0.0820

0.148

0.185

0.585

Model 3

0.132

0.174

0.344

0.350

Note: Density of choice frequencies for Model 1 depicted in the figures.

44

Primary Implications: Carrier-segmented Reservation Price Functions

2000   1800  

§  LE  

Price  ($)  

1600  

ME  

1400  

LE-­‐MF  

1200  

LE-­‐LF   ME-­‐MF  

1000  

ME-­‐LF  

800  

MF  

600  

LF   HE-­‐SF  

400   300  

375  

450  

525  

600  

675  

750  

825  

§  §  §  §  §  §  §  § 

Model 1 Carrier Segments LE = Low Experience (Not Mod.- or LargeSize Firm) ME = Mod. Experience (Not Mod.- or LargeSize Firm) LE-MF = Low Experience + Mod.-Size Firm LE-LF = Low Experience + Large-Size Firm ME-MF = Mod. Experience + Mod.-Size Firm ME-LF = Mod. Experience + Large-Size Firm MF = Mod.-Size Firm (Not Low or Mod.Experience Respondent) LF = Large-Size Firm (Not Low or Mod.Experience Respondent) HE-SF = High Experience + Small-Size Firm

900  

Loaded  Miles  

Carrier Segment

Reservation Price Function

LE

Rkj = {-1/(βPrice + βPrice-LE)} * (βLoaded Miles*Loaded Milesj + βOne-Day Lead*1-Day Lead Timej)

ME

Rkj = {-1/(βPrice + βPrice-ME)} * (βLoaded Miles*Loaded Milesj + βOne-Day Lead*1-Day Lead Timej)

LE-MF

Rkj = {-1/(βPrice + βPrice-LE)} * (βLoaded Miles*Loaded Milesj + βLoaded Miles-MF*Loaded Milesj * Mod. Firm Sizej + βOne-Day Lead*1-Day Lead Timej)





45

Primary Implications: Bundled Offers may be Less Expensive 2500  

§ 

Reservation price functions for bundles are similarly developed

§ 

Generally exhibit lower pershipment reservation prices

§ 

May be an important source of cost savings for shippers while offering benefits to carriers

LE  

2000  

Price  ($)  

ME   LE-­‐MF  

1500  

LE-­‐LF   ME-­‐MF  

1000  

ME-­‐LF   500  

MF   LF  

0   300  

375  

450  

525  

600  

675  

750  

825  

900  

General  

Loaded  Miles  

Carrier Segment

Reservation Price Function

LE

Rkj = {-1/(βPrice + βPrice-LE)} * (βLoaded Miles*Loaded Milesj + βOne-Day Lead*1-Day Lead Timej)

ME

Rkj = {-1/(βPrice + βPrice-ME)} * (βLoaded Miles*Loaded Milesj + βOne-Day Lead*1-Day Lead Timej)

LE-MF

Rkj = {-1/(βPrice + βPrice-LE)} * (βLoaded Miles*Loaded Milesj + βLoaded Miles-MF*Loaded Milesj * Mod. Firm Sizej + βOne-Day Lead*1-Day Lead Timej)





46

Key Findings and Limitations §  The speculative nature of the spot market has considerable behavioral dynamics that are important to capture §  3PLs can develop effective revenue management strategies based on carriers’ different valuations of spot market shipments. §  Offering multiple shipments simultaneously on the spot market (bundling) can create cost savings for 3PLs. §  Limitations -  Ability to capture real-world, as opposed to experimental reservation prices •  Because of the spot market’s volatility, the framing of the experiment changes in substantial and un-modeled ways over time.

-  Magnitude: A larger field experiment with many more carriers would yield richer insights and the ability to improve carrier segmentation 47

We Love Feedback   Ques2ons/Comments   Email:  [email protected]  

  Follow  Me   Twiner  @b_ra.onal  

Connect  with  NUTC  

48

Elasticities Single Shipment   Variable Price Respondents with low Price * Low Experience experience among the Price * Mod. Experience most price-sensitive. Loaded Miles Loaded Miles * Mod.-Sized Firm Loaded Miles * Large-Sized Firm Loaded Miles * Mod.-Size Fleet Loaded Miles * Large-Size Fleet Loaded Miles * Mod. Firm Tenure Loaded Miles * High Firm Tenure   Variable Price Price * Low Experience Firms with moderate Price * Mod. Experience tenure among the mostLoaded Miles Loaded Miles * Mod.-Sized Firm distance-sensitive. Loaded Miles * Large-Sized Firm Loaded Miles * Mod.-Size Fleet Loaded Miles * Large-Size Fleet

Model 1 Elasticity 0.0011 0.0031 0.0034 -0.0022 -0.0034 -0.0014

49

Model 2 Elasticity 0.0010 0.0065 0.0015 -0.0019

   

---

--  Model 1 Elasticity 0.0002 0.0016 0.0019 -0.0004 -0.0012 -0.0002

---

   

---0.0001 -0.0005

---

---0.0032  

-0.0134 -0.0017

 

Model 2 Elasticity 0.0001 0.0058 0.0003 -0.0003

---

Model 3 Elasticity 0.0005 0.0001 0.0006 -0.0008 -----

-0.0009 -0.0026

---

Bundle

Loaded Miles * Mod. Firm Tenure Loaded Miles * High Firm Tenure Cross Empty Miles * Bundle

   

-0.0033  

   

Model 3 Elasticity 0.0002 0.000003 0.0003 -0.0003 -----0.0279 -0.0014 -0.0007

Probability Curves & Reservation Price Functions Probability Curves   Calculated using the following assumptions »  1 day of lead time and mean values of all variables »  Alternatives treated as separate binary choices (e.g. zero utility values for the unselected alternatives)

Reservation Price Functions

§  Let W define the set of non-price parameters, where w = 1 … W. §  Let zw denote the non-price variables, and β Price and βW denote the price and non-price parameters, respectively rkmj

50

W

⎞ β z = ⎛⎜ − 1 w w β Pr ice ⎟⎠∑ ⎝ w

Reservation prices are inferred by taking the sum of the non-price parameters and multiplying by the reciprocal of the appropriate price parameter(s).

Single Shipment Probability Curves

1  

0.8  

LE  

0.7  

ME  

0.6  

LE-­‐MF  

0.5  

LE-­‐LF  

0.4  

ME-­‐MF  

0.3  

ME-­‐LF  

0.2  

MF  

0.1  

LF   900  

870  

840  

810  

780  

750  

720  

690  

660  

630  

600  

570  

540  

510  

480  

450  

420  

390  

360  

HE-­‐SF   330  

0  

300  

Probability  of  Acceptance  

0.9  

Model 1 Carrier Segments v LE = Low Experience (Not Mod.- or LargeSize Firm) v ME = Mod. Experience (Not Mod.- or LargeSize Firm) v LE-MF = Low Experience + Mod.-Size Firm v LE-LF = Low Experience + Large-Size Firm v ME-MF = Mod. Experience + Mod.-Size Firm v ME-LF = Mod. Experience + Large-Size Firm v MF = Mod.-Size Firm (Not Low or Mod.Experience Respondent) v LF = Large-Size Firm (Not Low or Mod.Experience Respondent) v HE-SF = High Experience + Small-Size Firm

Offer  Price  ($)  

Carrier Segment

Deterministic Utility

LE

Vkj = βSingle*Singlej + βPrice*Pricej + βPrice-LE*Pricej * Low Experiencek + βLoaded Miles*Loaded Milesj + βOne-Day Lead*1-Day Lead Timej

ME

Vkj = βSingle*Singlej + βPrice*Pricej + βPrice-ME*Pricej * Mod. Experiencek + βLoaded Miles*Loaded Milesj + βOne-Day Lead*1-Day Lead Timej

LE-MF

Vkj = βSingle*Singlej + βPrice*Pricej + βPrice-LE*Pricej * Low Experiencek + βLoaded Miles*Loaded Milesj + βLoaded Miles-MF*Loaded Milesj * Mod. Firm Sizej + βOne-Day Lead*1-Day Lead Timej

51