This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Transcript
The Pennsylvania State University
The Graduate School
Smeal College of Business
SERVICE REFUSALS, INFORMATION SHARING, AND
COMMITMENTS: EMPIRICAL ESSAYS IN FOR-HIRE TRUCKING
Powell 1996). Because a shipper’s demand for service is usually derived demand – that is,
dependent on someone else’s demand and therefore not completely under a shipper’s control –
forecasting the demand for TL service is imperfect and often wildly inaccurate. In addition to
service requests from shippers at contract prices, carriers also can service spot loads. Because
there is no widely-used centralized market in TL transportation (Tsai, Regan, and Saphores
2009), the availabilty of spot loads is an additional element of randomness that a carrier must
consider. In the “pickup and delivery problem” (Berbeglia et al. 2010, p. 8), a carrier decides
2
which service requests to accept or reject in real-time as requests for service arise randomly from
contracted shippers and shippers needing spot capacity.
Shippers contract with carriers at static prices, and it is typically assumed that carriers
perfectly live up to these prices. But, carriers strategically accept or reject load service requests.
Despite this clear mismatch in assumptions, very little research analyzes how uncertain contract
carriers affect transportation costs for shippers. This research fills much of this void.
Specifically, I address the following questions:
1) Do carriers reject a significant number of service requests? If yes, when and why are
they likely to do so?
2) When contract carriers reject service requests, do shippers typically pay a high
penalty? How much higher are spot prices than contract prices?
3) What cause spot prices to be high or low? Can a “market price” be inferred from a
large amount of spot prices and, if so, how can it be calculated?
4) How well do shipper governance mechanisms – the commitment of future business,
performance monitoring, and more explicit contracts – ensure better carrier
performance?
5) Do governance mechanisms perform differentially as the spot market changes?
I adopt an empirical lens in this dissertation. My primary reason is due to the lack of
concrete evidence in the extant research. I use several large industry datasets and perform
rigorous econometric analyses to address the research questions. This dissertation proceeds as
follows. Chapter 2 addresses the first two questions – do carriers often reject service requests,
when and why do they do so, and are spot prices punitive to shippers. Chapter 3 addresses
questions 2 and 3 – what causes high or low spot prices, and can a market status be inferred from
3
a large collection of spot prices. Finally, chapter 4 examines how governance mechanisms
ensure good performance from carriers and how these mechanisms perform as prices in the spot
market change. Each chapter is self-contained, with suggestions for future research at the end.
4
CHAPTER 2: SERVICE REFUSALS IN SUPPLY CHAINS: DRIVERS
AND DETERRENTS OF FREIGHT REJECTION
Contracts in the for-hire trucking industry are unusual in that, while they establish prices for
different services, there is typically no legally-binding obligation or penalty for either party to
offer or accept a load. When a load is rejected by all contract carriers, shippers must turn to the
spot market, which can significantly increase supply chain costs. Because these transactions
occur between private parties, data on load acceptances/rejections and contract/spot prices have
not been available to academic researchers, leaving the freight rejection problem largely
unexplored. We are able to examine this problem using a detailed transactional dataset of a large
national shipper. We estimate that spot prices for truckload services average about 62% higher
than contract rates. We find key operational and economic factors to be drivers of freight
rejection and the shipper-carrier relationship to be a deterrent to freight rejection. We also find
that primary and secondary carriers respond differently to these operational and economic
factors. We discuss how these insights could be used by a shipper to get better performance and
lower cost from their carrier base.
2.1 Introduction
Trucking has been described as “the lifeblood of the US economy” (Costello 2014). Hauling
70% of all domestic tonnage in the United States and with revenues valued at $681.7 billion, it
touches every industry that makes or ships goods (Corridore 2014). For-hire truckload
transportation is one of the largest trucking subsectors, with annual revenues of $298.6 billion in
2012 (Corridore 2014). For-hire truckload transportation consists of manufacturers and
distributors (i.e., shippers) who hire service providers (i.e., carriers) or third-party intermediaries
(i.e., brokers) to haul loads, which occupy a full trailer, from one point to another. With a large
number of shippers, carriers, and brokers and a homogeneous product, truckload transportation is
best characterized as a perfectly competitive market.
5
Carriers continuously decide whether to accept or reject potential loads as their trucks
move from one location to another. This problem is known as the dynamic pickup and delivery
problem, and it has been an active research area for decades (e.g., Powell 1987; Berbeglia et al.
2010). A related area of research is the procurement of contract rates for shippers from carriers
(Moore et al. 1991; Caplice and Sheffi 2003; Caplice and Sheffi 2005). Rates are most
commonly established for origin-destination combinations (i.e., lanes) via annual reverse
auctions, where a shipper invites carriers to bid competitively on the right to haul their freight on
a given lane. The standard agreements between shippers and carriers are highly flexible, best
described as relational contracts (Baker et al. 2002). In these agreements, the winning carrier
(“primary carrier”) gets the first-right-of-refusal for loads on a lane at the bid price based on the
outcome of the auction. The shipper generally holds the primary carrier accountable for some
level of service (e.g., percentage of freight accepted) (Caplice 2007). Interestingly, the
enforcement of these agreements are not legal sanctions but the promise of future business –
poor performance can result in a loss of business for a carrier. As a Senior Vice President at a
carrier describes it: “the ‘rate agreements’ and ‘load commitments’ for the most part have no
contractual obligation or penalties on either party” (Taylor 2011, p. 2).
Given static rates, highly flexible contracts, and carriers that strategically accept or reject
freight, shippers are often faced with a “freight rejection problem.” That is, when a shipper
needs service on a lane, what happens when it is offered to a contract carrier? What factors
affect freight rejection and thus transportation costs? Despite the ubiquity of the problem, the
causes of freight rejection has received little research attention.
We are able to provide insight into the issue of freight rejection because we have been
granted access to a two-year transactional dataset from our industry partner, a large national
shipper. We estimate when and why a shipper is required to utilize the spot market to secure
transportation service. In doing so, we make several contributions to the extant research. We
quantify the incremental expense due to freight rejections and show that it can be extremely high.
Specifically, based on an analysis using tens of thousands of spot transactions over a two-year
6
period, we observe that truckload spot prices are, on average, about 62% higher than their
corresponding contract rates. It is remarkable how widely average spot prices can vary
throughout the year; at times, spot prices are comparable to contract rates, while at other times
spot prices average more than double their respective contract rates.
We analyze the drivers of and deterrents to freight rejection by carriers. Carriers are
influenced by three primary factors: (1) operational effects, such as upswings and uncertainty in
demand; (2) economic effects, such as when the transportation market is tight (i.e., demand
exceeds supply); and (3) relationships, such as the volume of business transacted between the
two parties. We find general support for our conjectures and thus contribute to a better
understanding of freight rejection and its consequences.
In further analysis, we investigate whether the level of interorganizational commitment
between shippers and carriers elicits different behavior. Primary carriers are influenced
relatively more by operational factors, whereas carriers that are offered a load only after the
primary carrier has rejected it (“secondary carriers”) are influenced more strongly by economic
factors. Moreover, secondary carriers are not influenced by relationship factors, such as the
value of business transacted between the two parties.
A final contribution of our paper is that it informs models that include transportation
costs. Operations management and transportation models typically assume either a
transportation cost that is linear in the number of units shipped or with a fixed charge and linear
component (e.g., Hirsch and Dantzig 1954; Roberti et al. 2014). Our data and results suggest
that in some cases the underlying transportation costs for a shipper are non-linear in volume due
to contract carrier rejections and the necessity to use the spot market.
2.2 Transportation Industry Background
There are three important areas that pertain to our study. First, we discuss the ownership and
contract structures that exist in the transportation industry, focusing particularly on the most
common: for-hire truckload contracts. Second, we briefly review the extensive research that has
7
been done on the dynamic pickup and delivery problem. Finally, we review the scant literature
that discusses carrier performance from a shipper’s perspective.
2.2.1 Shipper-Carrier Contracts
In truckload transportation, there are four primary contract types: private fleets, dedicated fleets,
contract freight moves, and spot market freight moves. In terms of asset ownership, private
fleets are owned by shippers whereas the assets are owned by carriers in the last three.
Private fleets are observed frequently in industry, with a total value of services almost
equivalent to that of for-hire services, $292.0 billion versus $298.6 billion, respectively
(Corridore 2014). When choosing a private fleet, the shipper sacrifices cost efficiencies (for
example, Caplice (2007) estimates that the percentage of empty miles for private fleets average
about 24%, whereas for-hire fleets average between 6% and 12%) for service and the residual
right of control over the assets (Baker and Hubbard 2003). The residual right of control gives the
shipper authority to direct the truck assets as they see best; when they do not own the asset, they
must negotiate with carriers. Having a private fleet, however, does not preclude a shipper from
utilizing for-hire carriers. On the contrary, shippers, such as Wal-Mart and Sysco, who own
large private fleets are also large buyers of for-hire transportation services (Caplice 2007).
Two hybrid contract structures are common in the industry, dedicated fleets and for-hire
contract freight. Dedicated fleets are owned and operated by a carrier, but the shipper has full
control over how the assets are used. In essence, it is a private fleet for the shipper that takes
advantage of a carrier’s operating expertise.
For-hire “contract” freight is the structure under which most carriers operate (Corridore
2014). For-hire fleets are advantageous from a cost point-of-view because they can aggregate
supply and demand across a variety of shippers and lanes, referred to as economies of integration
(Keeler 1989) and economies of scope (Caplice and Sheffi 2003, Özener, Ergun, and Savelsbergh
2011).
Because contracts between shippers and carriers in the contract freight segment specify
prices but do not require load offers or acceptances, these agreements cause tension in the
8
relationship when there are market imbalances on either the supply or demand side. For
instance, in a case study with Hershey, a large confectionary company based in the United States,
Zsidisin et al. (2007, p. 15) state that “carriers must be allocated enough freight during periods of
slack transportation capacity to maintain their long-term commitment during periods of
constrained capacity.” J.B. Hunt Transport, one of the largest truckload carriers in the U.S.,
implicitly acknowledged this tension in a 2005 earnings release (J. B. Hunt Transport, Inc. 2005,
p. 3): “the number of loads moving under spot quotes increased significantly from earlier in the
year, which indicates that capacity availability in the truckload market remains extremely
fragile. Current spot activity in October remains at third-quarter levels indicating continued
tightness in the market. Despite this significant increase in spot demand, we continue to honor
our base customer commitments [emphasis added].” In an industry report, Taylor (2011, p. 3)
articulates the tension succinctly: “generally, there are no volume guarantees, nor financial
penalties, so essentially when load acceptance rates fall, a lot of yelling and hollering is what
happens. The same occurs when loose capacity emerges [emphasis added].”
The pure “buy” contract is the spot market. When the spot market is used, a shipper
searches for short-term capacity for a load or series of loads, with no further commitments on
either side. Brokers, who act as middlemen in the for-hire trucking market, are often used as a
source of spot capacity because they specialize in maintaining contacts with a large number of
carriers. Not much is known publically about the size and magnitude of the spot market.
Caplice (2007) estimates it at between 5% and 10% of the overall market, and Kirkeby (2013)
concurs with an estimate of “less than 10%” of most carriers’ business. Tsai et al. (2011, p. 925)
acknowledge the dearth of public information on spot prices: “reliable truckload spot price data
can be challenging” and they find no published papers “that analyzed industry data.” Lindsey et
al. (2015a) fill some of this gap with an analysis of observed spot rates from a large broker.
Their study attempts to explain the underlying costs of a spot carrier. Scott (2015) studied a
year’s worth of private spot transactions and estimated the value of lead time for carriers. Our
9
study differs from previous studies because we address how changes in prevailing spot prices
affect carrier decision-making.
2.2.2 The Pickup/Delivery Problem
The carrier’s dynamic pickup and delivery problem has been thoroughly studied for decades
(Powell 1987). Berbeglia et al. (2010) surveys the literature for this class of problems. In
accordance with the contracts discussed in Section 2.1.1, a carrier faces random demand with
short lead-times as its trucks move from delivery to delivery. When a truck becomes empty after
a delivery, the carrier must decide whether to relocate the empty truck (referred to as
“deadhead”) from one location to another in search of more profitable freight or wait for offers to
appear in the current location. Faced with a highly uncertain environment, the carrier decides
which freight is most profitable to accept and which s/he should reject.
2.2.3 Freight Rejections from a Shipper’s Perspective
Freight rejections from the shipper’s perspective has received little research attention. Zsidisin
et al. (2007) is the closest published study that we found pertaining to our topic. Through a case
study with Hershey, the authors explore the impact that the strength of the relationship between a
shipper and carrier has on carrier performance, as measured by on-time deliveries and percentage
of freight rejected. Freight rejections are a major problem for Hershey because they delay
shipments, which can result in missed delivery appointments and therefore cause stockouts at the
retailer level. Furthermore, the search for transportation capacity becomes labor-intensive.
Zsidisin et al. (2007) do not specifically study the impact of spot freight moves, but do claim that
“arms’-length” carriers have higher freight rates than “partnership” carriers during periods of
tight transportation capacity, but lower rates during loose markets. Their primary findings are
that carriers with whom Hershey invests more in the relationship, via frequent communications,
high-level inter-firm interactions, and the willingness to allocate freight at above-market rates
perform better from a freight rejection standpoint.
In related research, Lindsey et al. (2015b) consider how a broker dynamically sources
capacity in an online environment. In their scenario, they include the likelihood of a carrier to
10
accept freight at an offered price. Caplice (2007) does not specifically study freight rejections,
but mentions that 74% of loads were accepted in their dataset. Furthermore, Master’s theses by
Harding (2005), Kafarski and Caruso (2012), and Kim (2013) discuss the issues of carrier
capacity and freight rejection. In an analysis of two shippers’ data, Harding (2005) reports that
13% and 20% of loads were rejected. Harding hypothesizes that rejections and volume are
related to one another and provides correlational evidence of this assertion. Kim (2013)
examines a large dataset where 19.8% of loads were rejected at least once and finds that volume
volatility on a lane increases freight rejection, but geography, distance, and price are not
significant factors.
Anecdotal evidence of this problem can be found in weekly survey reports from Morgan
Stanley, a large investment bank. For example, in a report on January 21st, 2015, a carrier stated:
“Extraordinarily tight market. We turn down 40% of what's offered to us each week” (Greene et
al. 2015, p. 5). In a report on October 7th
, 2015, a carrier reported: “Market is still very tight.
We're turning down 25% of what's offered to us and anticipate even higher demand in Q4”
(Vecchio et al. 2015, p. 18). It seems evident that contract carrier freight rejection is a
widespread issue for shippers in the for-hire trucking industry.
2.3 Freight Profitability from a Carrier’s Perspective
We consider the salient costs that a carrier should consider when accepting or rejecting freight.
Caplice and Sheffi (2005) discuss a model with respect to carrier considerations during a
truckload auction. While the underlying costs of the model we propose are similar, the
individual components of our model differ.
The situation we consider is thus: when a shipment is required on a lane, the shipper’s
transportation management system (TMS) offers it to the primary carrier at previously negotiated
rates. The primary carrier has a rough time window (e.g., about 90 minutes) to decide whether to
accept or reject the shipment. If the load is rejected by the primary carrier, the TMS offers it to
secondary carriers, also at previously negotiated prices, generally in order of cheapest to most
expensive (although this need not always be the case). This process continues through
11
contracted carriers until either a) a carrier accepts the load, or b) all contracted carriers reject the
load, in which case the load is offered to the spot market. Spot market capacity and prices are
procured using a first-price, sealed-bid auction (Milgrom 1989) in which numerous carriers and
brokers are invited to participate; the lowest bid at the close of the auction wins the right to haul
the load. The carrier that accepts the load for either the contract or spot price picks up the freight
and delivers it to the destination at the specified price.
A carrier must evaluate whether it is in their interest to haul a given load. Previous
research (e.g., Powell 1996) has recognized that a carrier faces four major costs: (1) the direct
cost to haul a load from one point to another; (2) the relocation cost associated with assigning a
particular truck to a particular load; (3) the opportunity cost due to the truck being exclusively
assigned to a particular load for the duration of the delivery; and (4) the network cost associated
with moving a truck from one region to another, where each region may be differentially
desirable due to “next load” possibilities (Özener, Ergun, and Savelsbergh 2011). Furthermore, a
carrier must consider the revenue for each load and the goodwill cost associated with a possible
load rejection – that is, the “yelling and hollering” discussed in Section 2.1.1. Given these
considerations, we propose the following carrier profit function:
* min( )i i i
i Ir dc oc rc nc gw
(1)
where I is the set of trucks that could possibly service the load. The revenue r is decided upon
during the annual auction and is generally viewed as non-negotiable ex post. The direct cost, dc,
is the cost incurred to haul the load from the origin to the destination. The opportunity cost, oci,
is the lost profit from not being able to serve other freight by committing truck i to the accepted
load. The cost incurred when deadheading truck i to the pickup location and after the delivery to
its next pickup location is the relocation cost, rci. The network cost, nci, which may be positive
or negative, is the value associated with removing a truck from the origin region to the
destination region. For example, some regions may be more or less profitable, on average, for a
12
carrier to have a truck located in, based on the availability of follow-on loads. Finally, the
goodwill in the relationship, gw, derives from living up to these informal agreements.
Equation (1) states that, if a truck is available and the load is profitable, a carrier will
assign the profit-maximizing truck. If the set I is empty (i.e., the carrier has no trucks available)
or if the load is unprofitable (considering goodwill), the carrier will reject the load. There are
three general categories of factors that influence a contract carrier’s decision to accept or reject a
load: (1) operational factors; (2) economic factors; and (3) relationship factors. Below we
discuss how each factor affects a carrier’s decision, using equation (1) as a framework for
discussion.
2.3.1 Operational Drivers
Operational drivers that affect a contract carrier’s decision to accept or reject a load include
empty repositioning cost, the value of having assets in particular regions due to varying
profitability in different regions, and the certainty of freight (Figliozzi et al. 2007; Powell 1987;
Powell 1996; Yang et al. 2004). For every load that is accepted by a carrier, a truck has to be
repositioned inbound to pick it up. A carrier who assigns the most profitable truck to pick up a
load will see decreasing profitability in volume, ceteris paribus. In solving the dynamic pickup
and delivery problem, a carrier that plans based on an average volume from a shipper will
experience higher relocation costs on loads that are offered above the average volume.
Furthermore, for a carrier with a “balanced network” – that is, where freight flows into
and out of regions are roughly equal to one another – large volumes of freight into a particular
region in a short time frame will increase network costs. If unexpectedly large volumes occur,
the carrier faces the prospect of having too many trucks in a region, resulting in follow-on loads
with decreasing profitability. Similar to above, carriers that plan around average volumes from a
shipper will experience higher network costs on loads that are offered above the average volume.
Likewise, predictability is important for a carrier; the more predictable the outbound
freight, the more incentive a carrier has to establish a network that complements the shipper’s
freight. On average, the relocation cost for predictable freight should be lower than the
13
relocation cost for less predictable freight. Higher relocation costs, as shown in (1), means that
freight will be less profitable for a carrier. Hence we expect that load offers that occur in a less
predictable setting will be rejected more frequently than load offers in a more predictable setting.
H1a: Load offers that are above the average demand on a lane are rejected more
frequently than load offers that are below the average demand
H1b: Load offers that are increasingly above the average demand on a lane are rejected
with an increasing frequency
H1c: Load offers that occur in a less predictable setting are rejected more frequently
than load offers in a more predictable setting
2.3.2 Economic Drivers
Economic drivers that influence a contract carrier’s decision to accept or reject a load capture the
highest-valued alternative use of the asset. Industry publications have suggested that carriers
balk at contract rates when spot rates are high. For instance, one of the largest freight brokers in
the industry, C.H. Robinson, claim that “when equipment becomes scarce, carriers may shift
their equipment to transactional customers who will pay higher spot market rates” (C.H.
Robinson 2013, p. 7).
When the market is out of balance in a carrier’s favor, the carrier must decide between
competing requests for the same truck. To the extent that shippers value getting the product to
their customers more than the transportation cost to get it there, shippers will be willing to pay a
premium to get spot capacity. With spot rates averaging more than double contracted rates in
some markets at some times, the carrier’s opportunity cost can be quite high if it accepts a
contract load. During periods of high spot prices, it behooves the carrier to allocate more
capacity to the spot market than in normal and slow periods.
H2: Load offers that occur when spot market prices are high are rejected more frequently
than when spot market prices are low
2.3.3 Relational Deterrents
Relationship deterrents that influence a contract carrier’s decision to accept or reject a load
measure the ongoing relationship between a shipper and a carrier that is important for both
14
parties. Shippers rely on carriers for consistent and high quality service at predictable prices
during times of scarce capacity and volatile transportation costs. Carriers invest in networks of
shippers with complementary freight to achieve economies of scope; losing a single shipper’s
business removes a link in the network, necessitating either costly search to find a replacement
shipper or the reorganization of their interrelated network of freight.
Yet, interestingly, industry practices present conflicting evidence concerning the value of
shipper and carrier relationships. The prevalence of annual auctions that require carriers to
compete anew suggests that shippers consider carriers to be good substitutes for one another, at
least over time horizons as long as a year. On the other hand, business is not always awarded
simply to the carrier with the lowest price (Caplice 2007). Both Thompson (2013) and C.H.
Robinson (2013) stress the importance of long-term relationships with carriers. For instance, one
supply chain manager works with “asset-based providers to build long-term relationships” to,
“most importantly […] guarantee capacity” (Thompson 2013, p. 1). While we recognize (and
respect) the former arguments, our conjectures are consistent with the latter (i.e., that
relationships affect freight rejection).
Thus, consistent with the findings of Zsidisin et al. (2007), we predict that a stronger
relationship, stemming from dependence, results in better performance (i.e., fewer rejections)
from the carrier. To capture this, we adopt the value of business transacted between the shipper
and carrier prior to a load offer as a measure of relational influence. As the value of business
increases, the loss of a shipper’s business is more acute and we predict that a carrier will provide
better service.
Long-term commitments often have aspects that leave one or both transacting parties
vulnerable to opportunism (Sako and Helper 1998). We adopt the length of the relationship
between the shipper and carrier as a measure of relationship strength. This measure has been
used elsewhere in the operations management literature (Krause et al. 2007), and is consistent
with the finding that trust and communications are important to shippers and carriers (Zsidisin et
al. 2007).
15
H3a: Load offers to carriers with whom more business has been transacted in the recent
past are rejected less frequently
H3b: Load offers to carriers with whom the shipper has a longer relationship are
rejected less frequently
2.4. Data and Specification
To analyze why carriers reject freight, we adopt an empirical approach using a detailed industry
dataset. Given the lack of studies addressing our research question, using observations from a
large industrial dataset provides a good basis for understanding carrier behavior. We test the
hypotheses outlined in Section 2.2 with data from a national shipper, which we will call Acme.
In this section, we describe our approach at a high level and provide context on the source of our
data. We then define our variables, describe our data cleaning process, and specify our
regression model.
2.4.1 Approach and Data Source
The outcome in our study is whether a contracted carrier hauled a load or not. Because this is a
binary outcome, we adopt a probit specification for our analysis. Furthermore, the decision to
accept a load and the spot market price is potentially endogenous – a carrier could conceivably
reject a load at the contracted price and then offer their services at a higher price, perhaps
through a broker to disguise their behavior. Indeed, this is precisely the behavior described by
C.H. Robinson in Section 2.2.2. To correct for endogeneity, we adopt an instrumental variables
approach (Angrist and Pischke 2009), which we describe in more detail in Sections 2.3.2 and
2.3.4.
Transportation is one of Acme’s largest expenditures, and they rely solely on for-hire
truckload carriers. Their product is relatively low value to weight and essentially non-perishable.
They have production facilities in every region of the United States, giving us national coverage
of truckload markets. Truckloads usually flow directly from their plants to a customer’s
distribution centers.
16
The dataset contains considerable transaction-level details for every truckload shipment
between January 1, 2012 and December 31, 2013. For each load we observe the origin,
destination, date and time of pickup, the carrier that moved the load, the rate paid (broken down
into categories such as line haul, fuel, accessorials, etc.), and whether the load was moved at a
contracted rate or a spot rate. We also have access to less detailed data for 2011 that allows us to
better estimate the length of the relationships. Because the 2011 data do not indicate whether the
load was moved under a contract or spot rate, we cannot use it to test our hypotheses.
Also included in the dataset is a “fair contract price” index distributed by a firm that
collects transportation cost data from a large number of shippers. The index provides the
average contract price paid from origins to destinations, aggregated at a five-digit zip code level,
for each year. This is useful because it allows us to compare observed spot prices on different
lanes; the index controls for heterogeneity caused by origins and destinations.
2.4.2 Unit of Analysis and Variables
Unit of Analysis
The unit of analysis for our model is at the lane-day-“rank” level. Lanes are pairs of origins and
destinations. An origin can contain multiple physical locations, but all locations in the same
origin are in the same geographic area. Destinations are typically all locations within the same
5-digit zip code, but in some cases individual ship-to locations are grouped separately (e.g., for a
large customer). We define the rank of a load as the order in which it was picked up for a given
lane on a given day. For example, the first load picked up on a day for a lane is assigned the rank
of one and the fifth load picked up on the same day for the same lane is assigned the rank of five.
Dependent Variable
Spot Load. Our primary interest is understanding and explaining why a contract carrier might
reject a load offered by a shipper. For each load, we observe whether it was moved at a contract
rate or at the spot market rate. We code a load as 0 if it moved under contract and 1 if the spot
market was used. We note a contract rate is used when any contract carrier accepts the freight.
On some lanes there is one contract carrier, on others there are multiple. Unfortunately, we do
17
not observe the actual offer and subsequent rejection; we infer load rejection from the necessity
to use the spot market. This assumption was confirmed by Acme.
Independent Variables
There are five broad categories of independent variables: operational, economic, relational,
instrumental, and controls. The independent variables should be calculated at the point when the
carrier evaluated the load offer and not when the load was moved. While the exact timing of the
offer is unknown, Acme confirmed that most freight is offered three days in advance and the
time window for rejection is 90 minutes. We therefore calculate the variables for a load three
days before the load is moved. For example, the Spot Premium for a load hauled on day t is
calculated with respect to day t - 3.
Operational
As the number of loads offered on a lane increases in a short time frame, two things happen that
can increase load rejections. First, the set of possible trucks that could service the load is reduced
resulting in restricted capacity. Second, the relocation, network, and opportunity costs increase,
reducing the profitability of accepting the load.
Low Volume, Medium Volume, High Volume, Very High Volume. To determine whether
increases in load volume results in more load rejections, we calculate the average daily volume
(number of loads on a lane on a given day) for the 30 days prior to each load. We then use the
rank of the load and compare it to the average load volume on each lane. If the rank is above the
average but below the average plus one standard deviation, we classify it as Medium Volume
and include it as an indicator variable. The assumption of 30 days was made to capture recent
behavior on a lane; we test this for robustness in Section 2.4.3, assuming 15 days and 45 days. If
the rank of the load is above the average plus one standard deviation but less than the average
plus two standard deviations, then we classify it as High Volume and include it as an indicator
variable. If the rank of the load is above the average plus two standard deviations, then we
classify it as Very High Volume and include it as an indicator variable. The excluded category is
18
below average load volume. Consistent with H1a and H1b, we predict that rejections will be
positively affected by these variables at an increasing rate.
Log of Standard Deviation of Load Volume. One measure of uncertainty on a lane is the
variability in daily load volume. To capture this, we calculate the standard deviation for daily
load volume for the 30 days prior to the load offer consideration. We use the log to reduce the
positive skewness of the standard deviation.
Days Since Previous Load. Another measure of uncertainty is the number of days since the
previous contract load was moved. For each load, we find the date of the previous contract
shipment on the lane and calculate the number of days since the previous contract load.
Economic
Spot Premium. Truckload carriers usually serve a wide variety of customers; there is little
specialization, particularly in the dry van segment that Acme utilizes. Trucks also have
exclusivity and a lack of scalability: when one is used to haul a load, it cannot be used to haul
anything else for the duration of the move. This means that a carrier’s opportunity cost of
accepting a contract load depends on what other shippers are willing to pay at that point in time.
Due to the many decentralized geographic markets and the proprietary nature of industry data,
there is no public, timely, and accurate measure of truckload spot prices. Our data present a
unique opportunity to construct an index of spot market prices specific to shipping regions and
times. Specifically, we are able to capture the tightness of truckload markets and hence the
opportunity cost of accepting a contract load. To measure the status of the market, we use the
shipper’s own spot prices, which are generated by a competitive bidding process involving
numerous carriers and brokers. Given the competitiveness of truckload markets, with many
shippers and carriers, we feel that our data are reasonably representative of prevailing market
conditions.
To construct a spot market index, we first define an Origin Area as a single location or
combination of locations that are geographically close to each other (i.e., always within the same
city). We group locations together that are part of the same general geographic market. This
19
approach provides more consistent and continuous spot prices, as not all locations have spot
loads every day.
For each Origin Area, we compare observed spot prices and fair contract prices
(described in Section 2.3.1) to estimate market status. Specifically, for every load in the dataset,
we sum the price paid for every spot load from the same Origin Area for seven days prior to each
load offer. We also sum the fair contract price for those same spot loads. The ratio of the two
sums gives us a “Spot Premium.” For example, if an Origin Area had 50 spot loads in a seven
day period with a total spot cost of $25,000 and a fair contract price of $17,000, then the Spot
Premium would be 1.471.
Industry experts suggest that this is a valid method for measuring the status of the local
spot market. For instance, Kirkeby (2013) says: “S&P believes pricing trends in the spot market
provide insight into the general availability of capacity and demand for that capacity.” Figure 2-
1 illustrates the average daily national spot price divided by the fair contract price over the
sample period, alongside the freight rejection rate, illustrating the extent to which the two are
related.
Figure 2-1. Rejection percentage and average spot price premium by month. Rejections
and spot prices appear to be highly correlated and seasonal.
0%
60%
130%
Month
ly S
pot
Pri
ce P
rem
ium
0%
5%
10%
15%
20%
Month
ly R
eje
ction P
erc
enta
ge
January 2012 July 2012 January 2013 July 2013 December 2013
Monthly Rejection % Monthly Spot Price Premium
20
To test the validity of our measure of market conditions, we compare it to a “truckload
supply and demand sentiment index” calculated by Morgan Stanley, a large US-based
investment firm, and found at the website of a third-party logistics provider (Transplace 2015).
The Morgan Stanley index surveys a broad array of shippers, carriers, and brokers to assess
overall supply and demand sentiment. We expect that our measure of spot prices, if spot prices
reflect supply and demand in the market, positively correlate with the Morgan Stanley index. To
compare the two, we first converted the Morgan Stanley index into daily digital points. We
calculated the Spot Premium measure nationally (i.e., across all regions as opposed to region-
specific). Finally, we compared the national Spot Premium variable with the Morgan Stanley
index.
As seen in Figure 2-2, the two measures largely align: the correlation between the two
time series is 77.3%. Thus, we conclude that our measure of market conditions is valid. Our
measure is advantageous to the Morgan Stanley index for two reasons: 1) it measures actual
price swings, instead of an arbitrary “sentiment”; and 2) it can be calculated regionally, capturing
important variation (e.g., the Florida harvest season is well known to cause truckload demand
and supply imbalances).
Relationship
To capture the complexity of relationships, we use three measures: two revenue-based measures
and one time-based measure. We note that we don’t observe the actual rejections in the data.
Instead, we have to look at active carriers on a lane, which we define as those that hauled a
contract load within the past 30 days. We calculate each measure at the carrier level and use a
weighted average for the lane-day.
Log of Carrier Revenue. The more business transacted between the two parties, the higher the
value of the business is to the carrier. At the firm level, we capture the rolling 30-day revenue
between the shipper and carrier on all lanes. We use the log of revenue to reduce rightward
skewness.
21
Figure 2-2. Morgan Stanley sentiment index versus the spot premium index.
Log of Lane Revenue. A lane with higher revenue is more valuable to a carrier. To measure this,
we calculate the rolling 30-day revenue on a lane prior to each load. We use the log of revenue
to reduce rightward skewness.
Length of Relationship. We measure the length of the relationship between the two parties in
days, and classify them into buckets for every one hundred days (e.g., 0-100 days is 1, 101-200
days is 2). We include these as dummy variables.
Instrumental Variables
Using instrumental variables to correct for endogeneity is a common approach used in
econometrics (Angrist and Pischke 2009). There are two requirements for an instrumental
variable: the inclusion restriction and exclusion restriction. The inclusion restriction requires
that an instrumental variable is related to the endogenous independent variable – in our case,
Spot Premium. The exclusion restriction requires that an instrumental variable is not directly
22
related to the dependent variable – that is, any relationship with the dependent variable must be
through the endogenous variable. Below we describe our instrumental variable and how it
satisfies these conditions.
Other Region Spot Premium. Short-term spot prices in regions that are geographically far away
from the origin and destination of a particular load should not affect a carrier’s decision because
the truck on the contracted load cannot be used to service these spot loads. However, because
the US economy has a strong seasonal component to it, we expect spot prices across regions to
be positively correlated with one another. Hence, using a spot price index in regions other than
the origin and destination of a load meets the requirements for an instrumental variable.
Furthermore, the method of using prices in other geographical regions as an instrument for prices
in a specific region has been used in other studies (Hausman, Leonard, and Zona 1994; Pinkse,
Slade, and Brett 2003).
To calculate the Other Region Spot Premium instrument, we first classify the origin and
destination region of every load, using the region classification system of the US Census Bureau
(US Census Bureau 2015). The US Census Bureau classifies the United States into four regions
– Northeast, South, Midwest, and West. Given the origin and destination region for a given load,
we observe the regions at which the load does not originate or end in. Then, for every load we
calculate a variable in a similar manner to the Spot Premium variable, but instead of using the
origin location, we use the other geographic regions. We include this variable as an instrument
for the Spot Premium variable.
Controls
Lane fixed effects. We control for time-invariant heterogeneity across the 1,129 lanes in our data
using lane fixed effects. These control for differences in the lanes that are constant through time
such as the distance between the origin and destination, population densities at the two locations,
and road quality (as long as there are no major changes in road quality during our sample).
Contract Carriers. For various reasons, the shipper has a different number of carriers under
contract on different lanes, even at different times within lanes. To control for this, we include
23
the number of unique contract carriers that are active on the lane. We calculate this as the
number of carriers that have hauled a load at a contracted rate within the previous 30 days.
Month-year fixed effects. We include temporal fixed effects to control for economy-wide factors
that may vary over time, such as changes in the broad U.S. economy, different weather
conditions, and changes to regulations governing the industry (e.g., new Hours of Service rules
were implemented on July 1, 2013, restricting work-hours for drivers between required breaks).
We include fixed effects for the month-year-region as a robustness check in Section 2.4.3.
Day-of-week fixed effects. We include day-of-week fixed effects to control for the different
behavior by the day of week. Specifically, it is conceivable that carriers behave differently early
in the week versus late in the week or on weekends.
Hour-of-day fixed effects. We include hour-of-day fixed effects to control for the possibility that
carriers behave differently during different times of the day. For instance, carriers may prefer
loads to be picked up early in the day versus late in the day.
2.4.3 Data
Our primary dataset contains 452,846 observations on lanes that have both spot and contract
loads on them. For various reasons we are not able to use the entire dataset. First, some variables
are calculated as 30-day running averages. These variables cannot be calculated for the first 30
days of the dataset. Excluding these observations removes 12,116 loads (2.67%). Second, several
variables can only be calculated when there was a load on the same lane within the previous 30
days. For example, the standard deviation of volume on a lane in the previous 30 days cannot be
calculated if there are no loads carried on the lane in the previous 30 days. This removes 21,449
loads (4.74%). Third, the spot premium can only be calculated when there have been spot loads
from an origin in a given 7-day period, which is not always the case. Excluding observations for
which a reliable spot premium cannot be calculated removes 4,303 loads (0.95%). Fourth, to
minimize bias (which is discussed in more detail in Section 2.3.4), we include only lanes that
have at least 20 loads on them, which removes 6,565 loads (1.45%). Finally, after applying the
above filters there are some lanes that no longer have variation in them (i.e., some lanes have all
24
spot loads or all contract loads). Excluding these observations removes 7,454 loads (1.65%),
leaving 400,959 records in our dataset, which is 88.5% of the original. Summary statistics of the
final dataset and a description of the variables are presented in Table 2-1.
2.4.4 Model Specification
In our model, carrier profitability is a latent variable, and we observe the actual accept or reject
decision. The dependent variable in our model, , is coded 0 if the shipment is moved by any
contracted carrier and 1 if the spot market is used (i.e., all of the contracted carriers rejected the
freight). Our observed dependent variable is related to the latent variable as such:
(2)
where is determined by (1).
Table 2-1. Variable descriptions and summary statistics. NR means Not Reported due to
confidentiality agreements with Acme. N = 400,959.
Because we have a binary outcome, we adopt a probit model with instrumental variables
for our analysis. The unobserved lane-specific characteristics necessitate the use of fixed effects
in our model. It is well known that nonlinear models with fixed effects are inconsistent due to
the “incidental parameters problem” (Neyman and Scott 1948). However, when T (in our case,
*
*
0 if 0,
1 if 0,
*
Variable names Definitions Mean St Dev Min Max
Spot Load 0 if the load moved at a contract rate; 1 if it moved via the spot market 0.109 0.311 0 1
Spot Premium 7-day rolling average of the spot prices divided by the fair contract prices
at the origin region 1.489 0.412 0.572 4.004
Log of Standard Deviation of Load Volume Natural log of 30-day rolling standard deviation of the number of loads on
the lane 0.697 0.963 -1.698 3.105
Medium Volume 1 if the rank of the load is above the 30-day rolling average number of
daily loads but less than 1 standard deviation above the average, 0
otherwise 0.287 0.452 0 1
High Volume 1 if the rank of the load is between 1 and 2 standard deviations above the
average number of loads, 0 otherwise 0.189 0.392 0 1
Very High Volume 1 if the rank of the load is more than 2 standard deviations above the
average number of loads, 0 otherwise 0.120 0.324 0 1
Log of Lane Revenue Natural log of 30-day weighted revenue for carriers on the lane; revenue
measure at the lane level NR NR NR NR
Log of Carrier Revenue Natural log of 30-day weighted revenue for carriers on the lane; revenue
measure at the firm level NR NR NR NR
Length of Relationship ('00s) 30-day weighted length of relationship for carriers on the lane 7.423 2.103 1 11
Days Since Previous Load Days since the previous load, capped at 30 2.375 3.370 1 30
Contract Carriers The number of contracted carriers seen on the lane within 30 days of
shipment 2.636 1.697 1 14
25
the number of loads on a lane throughout time) is reasonably large, bias in the coefficients is
small (Greene 2004). Katz (2001) and Coupé (2005) show that bias is negligible when T is
greater than or equal to 20; hence, we restrict our analysis to lanes with 20 or more loads over
two years. Our model is:
ldr l ldr my w h ldrx (3)
where ldrx are the independent variables as described in Table 2-1 for lane l on day d with rank
r, denotes lane fixed effects, denotes month-year fixed effects, w denotes the day-of-
week fixed effects, h denotes the hour-of-day fixed effects, and is the idiosyncratic error
term. We instrument Spot Premium with Other Region Spot Premium and use the maximum-
likelihood estimator.
We use clustered standard errors to allow for heteroskedasticity and correlation of errors
in the variance-covariance matrix. One potential option is to cluster errors at the lane level but
this does not allow correlated errors for two lanes that are geographically similar, i.e., two lanes
may originate and terminate in the same general area but errors clustered at the lane level would
not allow for any correlation in errors corresponding to these two lanes. To allow for these types
of correlations we instead cluster errors at the origin-destination state so that errors on any lanes
with the same origin and destination states can be correlated. This level of error clustering
results in 174 of cluster groups which is larger than the threshold of 50 suggested by Wooldridge
(2003). Alternative error clustering (such as clustering errors at the origin level) would be less
restrictive than clustering errors at the origin-destination state, but would result in too few cluster
groups.
2.5 Results
In this section, we first discuss the results of the main model, which are presented Table 2-2. We
then run separate models for primary and secondary carriers to investigate whether accept/reject
decisions are affected differently by the operational and economic drivers and relationship
exp( )Pr( 1 ) ,
1 exp( )
ldrldr ldr
ldr
x
l my
ldr
26
deterrents. These results are presented in Table 2-3. We then analyze the margins from our
models and discuss the cost of rejections for shippers, with the results shown in tables 2-5 and 2-
6. Following that we perform a number of robustness checks, presented in Table 2-6, to ensure
the conclusions are not the result of a choice we made.
Table 2-2. Main results.
2.5.1 Main Effect
Operational Drivers
We find considerable support for the hypothesis that shippers’ operations significantly impact
load rejection. As load volume increases, so do rejections as evidenced by the positive
coefficients for the Medium Volume, High Volume, and Very High Volume variables. We find
evidence that the likelihood of rejection increases as load volumes increase. High Volume loads
Variables
Spot Premium 1.864***
(0.221)
Log of Standard Dev. of Load Volume 0.428***
(0.058)
Medium Volume 0.0904***
(0.017)
High Volume 0.174***
(0.024)
Very High Volume 0.246***
(0.028)
Log of Lane Revenue -0.277***
(0.039)
Log of Carrier Revenue -0.0555**
(0.022)
Days Since Previous Load 0.0128***
(0.003)
Constant -4.830***
(0.33)
Observations 400,959
Number of Lanes 1,129
Lane FE Yes
Month-Year FE Yes
Day-of-Week FE Yes
Hour-of-Day FE Yes
Standard errors clustered at the origin-destination state
in parentheses. ***p<0.01, **p<0.05, *p<0.10.
27
are rejected more frequently than Medium Volume loads (p-value<0.01), and Very High Volume
loads are rejected more frequently than High Volume loads (p-value<0.01). Hence, Hypothesis
1a and Hypothesis 1b are supported. This result means that increases in load volumes in a short
period of time requires increased utilization of the spot market, which in turn results in non-
linear transportation costs.
Hypothesis H1c is also supported. The log of the standard deviation and the number of
days since the previous load both increase the likelihood of rejection. This indicates that a
carrier is less willing to service loads that are less predictable, perhaps because finding follow-on
loads for less predictable loads is more difficult. This, in essence, negatively affects their
economies of scope.
Economic Drivers
Economic factors are clearly important in the carrier’s decision-making process. The coefficient
for Spot Premium is large and significant, supporting Hypothesis 2. This supports the assertion
that carriers trade off contracted loads with spot market loads. When the market turns in a
carrier’s favor, they are less willing to live up to previously contracted prices
Relationship Deterrents
Both national and lane-based measures of the strength of a relationship are deterrents to
rejections. The local relationship (Log of Lane Revenue) appears
to have a larger deterring effect than the national relationship (Log of Carrier Revenue). This is
likely due to the fact that the lane is more valuable to a carrier, which in theory will be
interconnected with other shippers’ freight. Hence, Hypothesis 3a is supported.
Interestingly, the length of relationship does not show a meaningful trend in the indicator
variables. Maintaining a long relationship with a carrier does not appear to result in better (or
worse) performance. This is consistent with the industry practice of annual freight auctions; if
longer relationships mattered, then annual auctions would be less likely to be the standard.
Hypothesis 3b is not supported.
28
2.5.2 Differences in Commitment between Primary and Secondary Carriers
In the main analysis, we did not attempt to differentiate among contract carriers; we have
assumed that primary carriers behave the same as secondary carriers. There is potential
heterogeneity in the relationship, however, to the extent that the primary carrier feels a higher
level of interorganizational commitment to the shipper than secondary carriers. For instance, a
primary carrier is given the first right of refusal on a lane but is also expected to maintain some
load acceptance ratio, whereas secondary carriers are only expected to provide service at their
convenience (Caplice, 2007).
Unfortunately, we do not observe the order in which load offers are rejected; we only
observe if a load is rejected by all contract carriers (i.e., it goes to the spot market). We therefore
do not observe which carrier was designated primary when a load was moved. Moreover, the
primary designation can change throughout the year for various reasons, such as poor carrier
performance or willing opt-outs by either party.
Upon discussion with Acme, we found that it is possible to estimate the primary carrier
on a lane in most cases. Primary carriers move a majority of the freight, about 72% based on our
calculations. If they do not, then they are replaced by a carrier that will. To determine the
primary carrier, we looked at the number of contract loads hauled on each lane for each month-
year. If a carrier hauled most of the contract loads, then we designated them as the primary
carrier for that month-year and all of the other carriers as secondary. For 90% of the loads, the
designation is clear because a carrier hauled more than half of the total; in the rest, the primary
carrier was designated as the carrier that hauled the most loads. While less than perfect, we are
confident that we are able to identify the primary carrier for a vast majority of the loads.
After classifying primary and secondary carriers, we create two sub-samples. The first
includes all observations in which the accept-or-reject decision is made by the primary carrier.
The second includes only accept-or-reject decisions made by any contract carrier after the
primary carrier has rejected the load. For the two sub-samples, we repeat the analysis of Section
2.4.1.
29
Table 2-3 shows the results of the analysis. We find that while rejection ratios for the two
sub-samples are similar (primary carriers reject 28% of loads offered to them and secondary
carriers reject 31%), their decisions are driven by very different factors. Interestingly, primary
carriers are more heavily influenced by operational and relationship factors and relatively less by
economic factors (Column 1). Secondary carriers on the other hand are strongly influenced by
economic factors but much less so by operational factors and not at all by relationship factors
(Column 2). This result indicates that primary carriers behave less opportunistically to external
market. It is consistent with primary carriers expecting and integrating a certain volume of
freight from this shipper into their overall pickup/delivery framework and considering the
longer-term profitability of the relationship rather than the very short-run profits from diverting a
single truck. Conversely, secondary carriers make decisions based on whether it is the most
profitable load for them given the status of the market. Since the shipper has made no
commitment to turn to them first with a load, there apparently is not a feeling of commitment in
return if a more profitable load is available. These results make it clear that the extent of
interorganizational commitment moderates opportunistic behavior by carriers.
2.5.3 Primary Carrier and Secondary Carrier Margins Analysis
Having estimated the key drivers and deterrents of truckload rejections, we now perform a
margins analysis to illustrate how primary carriers and secondary carriers respond to different
conditions. For the following analysis, we use the model estimates for primary and secondary
carriers, shown in Table 2-3. To calculate margins, we need to use specific inputs – lanes,
months, market conditions, and volume conditions. We define the “Average Lane” as follows.
First, we sort the lanes in ascending order based on their associated fixed effects. We then
choose the lane for which half of the loads traveled on a lane with a smaller fixed effect and half
of the loads traveled on a lane with a larger fixed effect. We define a “Well-performing Lane”
and “Poorly-performing Lane” in a similar manner, where we use the 25th
percentile of fixed
effects for the former and the 75th percentile of fixed effects for the latter.
30
We also define the “Average Market” condition as that at which the Spot Premium
variable is at its average value. When the Spot Premium variable is one standard deviation above
the average, we define these as “Above Average” market conditions. We define “Below
Average” market conditions in a similar manner, except as one standard deviation below the
average. The volume conditions that we use are defined in Section 2.3.2. Month 8 in our 24-
month period was used because it was the closest to the “average month,” as determined by the
fixed effects for months. Unless otherwise noted, covariates were assigned to their means.
Table 2-3. The effect of interorganizational commitment.
Tables 2-4 and 2-5 show the differential behavior of primary and secondary carriers
under a variety of conditions. Table 4 shows the impact that volume has on carrier behavior as it
Variables (1) (2)
Spot Premium 1.213*** 1.956***
(0.263) (0.284)
Log of Standard Dev. of Load Volume 0.362*** 0.283**
(0.038) (0.144)
Medium Volume 0.0961*** 0.0614***
(0.023) (0.018)
High Volume 0.176*** 0.0845***
(0.03) (0.028)
Very High Volume 0.236*** 0.187***
(0.034) (0.039)
Log of Lane Revenue -0.236*** -0.124
(0.032) (0.11)
Log of Carrier Revenue -0.0511** 0.00825
(0.021) (0.027)
Days Since Previous Load 0.008*** -0.0007*
(0.003) (0.001)
Constant -3.323*** -2.786**
(0.253) (1.12)
Observations 385,362 95,238
Number of Lanes 1,094 483
Lane FE Yes Yes
Month-Year FE Yes Yes
Day-of-Week FE Yes Yes
Hour-of-Day FE Yes Yes
R-squared N/A N/A
Standard errors clustered at the origin-destination state in
parentheses. ***p<0.01, **p<0.05, *p<0.10. Primary carriers are
shown in column 1 with a probit specification. Secondary carriers
are shown in column 2 with a probit specification.
31
varies from Low Volume to Very High Volume. For example, on the Average Lane, primary
carriers are more likely to reject loads offered at Very High Volume conditions 8% more
frequently than loads offered at Low Volume, in average market conditions. For the same
calculation, secondary carriers are 6.4% more likely to reject a load. Overall, primary carriers
are about 2.3% more responsive to volume conditions than secondary carriers.
On the other hand, secondary carriers are much more responsive to market conditions, as
shown in Table 2-5. On the average lane and at Low Volume conditions, secondary carriers are
47% more likely to reject loads offered during above average market versus below average
markets. This is more than two and half times the primary carrier response rate differential of
18.4%. Overall, secondary carriers are about 2.3 times more responsive to market conditions
than primary carriers.
When performing the margins analysis, we used the month with the average fixed effect
– month 8. Our month fixed effects control for factors not captured in our other variables – e.g.,
weather disruptions. However, the month fixed effects necessarily consume some of the
variation that can be explained by changing market conditions over time. To estimate the impact
of different months in our time period, we ran a similar analysis to the above while using the 25th
(“good month”) and 75th
percentiles (“bad month”) of the month fixed effects. Secondary
carriers, as expected, are impacted more by the month fixed effects. For example, secondary
carriers are 13.9% more likely to reject loads in the bad month than in the good month, whereas
primary carriers are only 4.7% more likely to reject loads in the bad month than in the good
month.
32
Table 2-4. Differential carrier response rates by volume conditions.
Table 2-5. Differential carrier response rates by market conditions.
Market Status Volume Differential Primary Secondary Primary Secondary Primary Secondary
Below Average Very High Volume - Low Volume 6.2% 3.0% 3.7% 1.0% 8.5% 4.9%
Average Very High Volume - Low Volume 8.0% 6.4% 5.4% 3.5% 9.5% 7.4%
Above Average Very High Volume - Low Volume 9.2% 7.4% 7.2% 6.8% 9.7% 6.1%
Note: This table shows the differential response rates for primary and secondary carriers. The percentages reported represent a carrier's mean response under Very High
Volume conditions minus a carrier's mean response under Below Average Volume conditions. Three different market conditions are shown: below average market
conditions, average market conditions, and above average market conditions. Three lanes are reported: an average lane, a well-performing lane, and a poorly-performing lane.
Average Lane Poorly-performing LaneWell-performing Lane
Volume Status Market Differential Primary Secondary Primary Secondary Primary Secondary
Low Volume Above Average - Below Average 18.4% 47.0% 11.9% 27.2% 23.2% 55.4%
Medium Volume Above Average - Below Average 19.6% 48.6% 13.3% 29.1% 23.8% 56.0%
High Volume Above Average - Below Average 20.7% 49.1% 14.4% 29.8% 24.2% 56.2%
Very High Volume Above Average - Below Average 21.4% 51.4% 15.4% 33.0% 24.4% 56.6%
Note: This table shows the differential response rates for primary and secondary carriers. The percentages reported represent a carrier's mean response during above average
market conditions minus a carrier's mean response during below average market conditions. The four different volume conditions are shown. Three lanes are reported: an
average lane, a well-performing lane, and a poorly-performing lane.
Average Lane Poorly-performing LaneWell-performing Lane
33
Finally, we issue a note of caution with these results. Trucking and shipper operations
are extremely dynamic, and while our dataset has significant detail, we are not able to observe
everything perfectly. As mentioned earlier in the paper, we do not observe the actual accept and
reject decisions – we observe the contract or spot outcome. So it is possible that some of the spot
loads were not offered to contracted carriers (e.g., perhaps due to lead time constraints). In that
scenario, our estimates of carrier responses to market conditions will be overstated. With regards
to volume conditions, we have assumed that the order of pickup is consistent with the order of
offering and accepting. When this is not the case, our estimates of carrier response to volume
conditions will be understated. A conservative estimate to bound both of these responses is to
divide the response to market conditions by two, and to multiply the response to volume
conditions by two. Despite these imperfections, our estimates provide a rough estimate of carrier
marginal responses under various settings.
2.5.4 Robustness
We performed a number of specification checks to test the robustness of our results.
Specifically, we 1) include only lanes with at least 20 loads, 2) measure recent volumes over the
previous 30 days, 3) measure local market conditions by spot premiums in the past 7 days, and 4)
assumed a probit specification. Table 2-6 shows that the results are consistent when we alter
these decisions. In column 1, we show the main results reported in Table 2-2. In column 2, we
report the main model without correcting for endogeneity. While this model is mis-specified, the
results stay directionally the same. Our results are unaltered when we change the number of
loads on a lane to be included in the analysis from 20 to 10 (Column 3) and 30 (Column 4), use a
linear probability model specification (Column 5), use the Morgan Stanley sentiment index
instead of our measure of market conditions (Column 6), calculate the Spot Premium using local
market conditions in the previous 4 days (Column 7) and 10 days (Column 8) as opposed to 7
days, and use a 15-day (Column 9) and 45-day (Column 10) window rather than a 30-day
window for calculating some variables. The robustness of the results to alternative variable
34
definitions and specifications makes us confident that the conclusions are not due to any
decisions we made.
2.6 Discussion
2.6.1 Contributions
Shippers who contract with for-hire truckload carriers often have their freight rejected at
previously established prices and thus have to engage the spot market, which can drive up costs
considerably. Despite its importance, the problem of freight rejection by contracted carriers is
largely unstudied in the literature and thus the magnitude of its effect is unclear. By examining a
two-year transactional dataset, we contribute to the understanding of this problem in several
important ways.
First, while it is expected that truckload spot prices are higher than contract prices, few studies
analyze actual truckload spot prices. We observe that spot prices are surprisingly high relative to
contract prices, with an average premium of 62%; more than 25% had a cost at least 100%
higher than the corresponding average contract price. Given that the for-hire truckload
transportation space is about $299 billion and that spot moves comprise an estimated 5% to 10%
of the total market (Caplice 2007; Kirkeby 2013), we estimate that shippers spend between $15
to $29.9 billion on spot transportation at an incremental expense between $5.7 and $11.4 billion
a year industry-wide. To put that in perspective, the estimated cost of the options considered for
the contentious changes to the truck driver Hours of Service rules that went into effect in 2013
was substantially lower, between $470 million and $2.3 billion (Department of Transportation
2011). On the face of it, this suggests that managers have a strong incentive to adopt strategies
to avoid the spot market in times of tight capacity. As we discuss below, our findings show that
there are indeed factors that drive and deter the need to engage the spot market.
Standard errors clustered at the origin-destination state in parentheses. ***p<0.01, **p<0.05, *p<0.10. Column 1 shows the main result. Column 2 shows the model estimation without
correcting for endogeneity. Columns 3 and 4 vary the minimum number of loads on a lane to 10 and 30, respectively. Column 5 uses the main model with a linear probability model
specification instead of probit. Column 6 uses the Morgan Stanley Index instead of the Spot Premium variable. Columns 7 and 8 vary the measures of Spot Premium from 7 days to 4 days
and 10 days, respectively. Columns 9 and 10 vary the time window for the calculation of several variables from 30 days to 15 days and 45 days, respectively.
36
Second, a shipper’s operations significantly affect carrier rejections. Demand volatility
causes rejections and therefore higher transportation costs. While shippers may have limited
options, there are two potential strategies that could be invoked. One strategy is a flexible
pricing mechanism that responds to upswings in demand; this could reduce rejections by primary
carriers, thereby reducing the need for the spot market. Another strategy that a shipper can adopt
is to smooth their demand for truckload services by utilizing pricing incentives for customers,
multi-sourcing, or pre-positioning inventory in a hub and spoke network.
Third, economics drives decision-making for truckload carriers. When spot market
prices are high, carriers are less willing to accept freight at contracted prices. Similar to above, a
more flexible pricing mechanism that responds to market conditions might mitigate some of this
behavior. Counterintuitively, this indicates that the price that a shipper is paying might be too
little, meaning that they should willingly pay more, in exchange for more capacity. Of course,
this would likely require a change in mindset for many shippers and would therefore have to
overcome inertia.
Fourth, we confirm that, as suggested by previous literature (Zsidisin et al. 2007),
relationships matter. Carriers with whom the shipper does more business accept loads more
frequently than carriers with whom the shipper does less. The relationship is significant at both
the local and national level, but more so at the local level. Hence, this study confirms the value
of core carrier programs and the need to keep the number of carriers to a manageable size.
Fifth, carrier behavior is influenced by the interorganizational commitment between the
carrier and shipper. Primary carriers respond to relatively more to operational and relationship
factors, while secondary carriers respond mostly to economic factors. Because primary carriers
are less sensitive to short-run economic conditions than secondary carriers, strategies to elicit
more capacity during times when spot prices are high could be beneficial for shippers.
Finally, our results can enhance operations management and transportation models. For
example, it is common in the operations management and transportation literatures to model
transportation costs at a constant rate in volume, either per truck or per unit. While this is useful,
37
in a for-hire truckload environment, shippers often face transportation costs that are non-linear in
volume due to the possibility of contracted carrier rejections and the need to use the spot market.
2.6.2 A Flexible Pricing Mechanism?
Our results suggest that flexible pricing could be beneficial for shippers in the following
scenario. Suppose that spot market prices are high, the shipper needs to ship more loads on a
lane than a primary carrier is willing to haul at the contracted price, and that some of these loads
will end up on the spot market. If there is additional capacity that the primary carrier is willing
to provide at a price higher than the contracted price but less than spot market price, then the
shipper could reduce their transportation costs by eliciting more capacity out of their primary
carrier.
Demonstrating a willingness to renegotiate prices, however, could have negative
consequences on carrier behavior. As a large broker advises: “shippers do need to be cautious
about how often they renegotiate rates – too often, and they’re just playing the market” (C.H.
Robinson 2013, p. 6). Hence, a shipper would need to design a pricing mechanism that
incentivizes additional capacity from primary carriers in expensive markets without encouraging
balking or frequent renegotiations.
Thus, the design of a flexible pricing contract in for-hire trucking is worthy of future
research. Accordingly, to provide a starting point, we propose an initial direction that may be
enlightening. Suppose, for example, that a shipper knows: (1) how many loads a carrier is
willing to haul on a lane next week; (2) how many total loads they are going to ship on a lane
next week; (3) how many of these loads will require spot capacity; and (4) the expected cost of
spot capacity. To incentivize extra capacity, the shipper could offer quantity-conditional price
incentives to the carrier, calculated based on expected spot prices. There are several positive
aspects of such an incentive: it would require the primary carrier to meet their volume targets to
receive the additional revenue; provide a higher price on loads that are more expensive or more
undesirable for a carrier to haul; and could vary based on spot prices. For example, if spot prices
are low, the shipper could offer small or no incentives, and when spot prices are high they could
38
offer larger incentives.
2.6.3 Future Research and Limitations
Findings from this study suggest future research possibilities. First, considering a shipper’s cost
function and a carrier’s profit function, a game theoretic model could be used to analyze different
contractual forms between shippers and carriers. This approach could consider different pricing
and contracting schemes and analyze how different policies affect shipper-carrier welfare.
Another modeling approach could address the price-service tradeoff for carriers. That is,
what is the optimal price that a shipper should establish on a lane with a carrier, assuming that
price affects the level of service (i.e., number of loads accepted) from the carrier? Insurance
models from economics could be helpful when analyzing this question (e.g., Crocker and Moran
2003). Finally, despite our best efforts, there are limitations to our study. Our findings and
consequent contributions need to be considered in light of these limitations. The primary
limitation is that our data comes from a single firm over a two-year period. While we do not
believe that there is anything particularly unique about the operations of our industry partner or
the years that we have used – particularly given the significant anecdotal evidence that other
shippers face this problem – it would be useful to validate our findings across other companies
and years. This would, of course, be a difficult task as transactional data sources for researchers
have remained elusive. Also, due to the empirical nature of our study, it is impossible to rule out
all alternative explanations for some of the behavior that we observe. Despite these caveats, we
believe our research makes a significant step forward in this important, yet understudied area.
39
CHAPTER 3: THE VALUE OF INFORMATION SHARING FOR
TRUCKLOAD SHIPPERS
This study explores the potential value to shippers of sharing load offers with carriers and
obtaining carriers’ responses in advance of the scheduled pickup date. Using a private
transactional dataset from a large national shipper, we find that truckload spot prices increase
considerably as the lead time before pickup decreases. As an extension of this empirical
analysis, we develop a method to estimate near-real-time market prices, which does not currently
exist in the truckload industry. A key insight is that market prices persist through time, meaning
that current prices are good predictors of future prices.
3.1 Introduction
Why should truckload shippers share load information in advance with contracted carriers?
Previous studies have shown the value of advance load information for truckload carriers
because it allows them to plan more efficiently (Tjokroamidjojo, Kutanoglu, and Taylor 2006;
Zolfagharinia and Haughton 2012). The incentive for shippers to provide such information is not
so clear. Modeling studies past and present (e.g., Hirsch and Dantzig, 1968; Roberti, Bartolini,
and Mingozzi 2014, among many others) frequently assume that costs for shippers are simply
linear in volume. Contract rates for carriers are typically negotiated in advance with a duration
of a year or more (Caplice 2007), so it would seem that a shipper only benefits from sharing load
information if it reduces the likelihood that the load is rejected.
Despite the significant breadth of research on information sharing for supply chain
partners (e.g., Gavirneni, Kapuscinski, and Tayur 1999; Lee, So, and Tang 2000; Sahin and
Robinson 2002; Angulo, Nachtmann, and Waller, 2004), including in a truckload transportation
40
context (Powell 1996; Tjokroamidjojo, Kutanoglu, and Taylor 2006; Zolfagharinia and
Haughton 2012), we are unaware of any studies that address information sharing from a
shipper’s perspective in a shipper-carrier relationship. This is surprising given the size ($300
billion in 2013; Corridore 2014) and importance (approximately 67% of freight by weight in the
United States moves by truck; Costello 2014) of the trucking industry to the economy of the
United States.
In this study, we empirically examine a year of spot transactions to explore whether and
to what extent shippers benefit from the advance sharing of load information. We find that spot
market truckload prices increase considerably as the time between load offer and pickup
decreases. This finding, combined with the right-of-refusal in industry-standard shipper-carrier
contracts in for-hire trucking (Scott, Parker, and Craighead 2015), means that shippers are
naturally incentivized to offer loads in advance and receive timely responses from their carriers
to minimize spot prices.
We estimate the value of lead time for truckload shippers. Because the estimates result
from an analysis of tens of thousands of bids in a competitive spot-bidding process from
numerous brokers and carriers spread throughout the country, these results are likely
generalizable to other shippers in the United States. Knowledge of the value of lead time is
important for shippers because they regularly decide how far in advance to offer freight to
contract carriers, with some probability of rejection, or offer it on the spot market.
Further, we provide other insights into the truckload spot market, a market that is
virtually unstudied despite its economic significance.1 While there is not much publicly-
available empirical information about the spot market, there is anecdotal evidence that it is
1 In a study of truckload futures contracting, Tsai, Saphores, and Regan (2011) acknowledged that finding “reliable
truckload spot price data can be challenging” and that no published papers were found “that analyzed industry data.”
This is consistent with our review of the literature.
41
important for carriers. For instance, in a 2012 earnings release, J.B. Hunt Transport stated that
“operating income increased 27% compared to 2011” due to “favorable changes in freight mix,
2012). Our analysis of the data supports J.B. Hunt’s emphasis on spot pricing, because the spot
prices we observe have significantly higher revenue than loads moving at previously-contracted
rates.
Finally, we propose a method to estimate market prices for for-hire trucking in the United
States. Due to the highly private nature of and lack of a centralized market in for-hire trucking
(Tsai, Regan, and Saphores 2009), there is no near-real-time index that adequately captures
market prices (Bignell 2013). A novel insight from our analysis is that market prices have
significant serial correlation. Knowing that prices persist from week to week impacts decision-
making for both shippers and carriers. For example, a carrier who observes high spot prices
might be better off accepting fewer contract loads in the near future and allocating more capacity
to spot customers. Likewise, a shipper might benefit from providing price incentives above
contract rates to encourage more capacity from contract carriers in expensive markets.
The rest of the paper proceeds as follows. Section 3.2 reviews the relevant literature. We
propose our hypotheses and discuss factors that affect carrier pricing in Section 3.3. Section 3.4
discusses our methodology and data. Results and robustness checks are discussed in Section 3.5.
Section 3.6 discusses the major findings of the study and implications for truckload shippers.
3.2 Literature review
3.2.1 Information sharing
Information sharing has become central to supply chain research over the past couple of decades
(e.g., Cooper, Lambert, and Pagh 1997; Gavirneni, Kapuscinski, and Tayur 1999), and is still
active (e.g., Özer, Zheng, and Ren 2014). Chen (2003) and Sahin and Robinson (2002) provide
42
thorough reviews of the significant problems considered in information sharing. Most of this
research has focused on the interplay of suppliers, manufacturers, and retailers. A fundamental
insight is that information sharing between partners can reduce inventory costs and improve
forecast accuracy.
With regards to the transportation industry, the value of information sharing has been
studied from a truckload carrier’s perspective by Powell (1996), Tjokroamidjojo et al. (2006),
and Zolfagharinia and Haughton (2012). The insight from these studies is that carriers are able
to plan more efficiently with knowledge of their future loads.
Lindsey et al. (2015) provided the only study that we are aware of that included a
measure of the impact of lead time on truckload prices. These authors analyzed truckload spot
prices observed at a broker. They included a dummy variable to capture lead time as “greater
than 8 days” (p. 11). For the large national shipper used in our study, spot market lead times are
much shorter than 8 days and the impact of lead time appears to drop off quickly after the second
day.
The lack of research attention is due to the fact that finding information about truckload
spot prices is challenging (Tsai et al. 2011) because these transactions occur between private
parties. We have overcome this problem by receiving access to a full year’s worth of spot
transactions from a large national shipper (henceforth, “Acme”).
3.2.2 Dynamic pickup and delivery problem
The dynamic pickup and delivery problem is central to transportation research, and has been for
decades (e.g., Powell 1987). Berbeglia, Cordeau, and Laporte (2010) provide an excellent
review. For the current study, the essence of the problem is as follows. A shipper contracts with
for-hire truckload carriers who operate fleets of trucks and service demand (loads) from many
43
customers. From a carrier’s perspective, load offers are received at random geographic locations
at random times; hence planning how to most efficiently operate a fleet is a very challenging
problem. Quite often, carriers will reject freight at previously contracted rates2. If all contracted
carriers reject a load offer from a shipper, the shipper must utilize the spot market.
Within the dynamic pickup and delivery problem, future knowledge of potential loads is
valuable for a carrier because it allows them to plan more efficiently. Interestingly,
Zolfagharinia and Haughton (2012) claimed that the value of knowledge decreases considerably
after the second day. For example, they estimated that the second day of advance information
makes a carrier 22% more profitable on average, but the third additional day makes the carrier
only 6% more profitable.
3.2.3 Truckload contracts
Truckload contracts between shippers and for-hire carriers in the United States differ from other
types of contracts because they specify a price but not a legally-enforceable service obligation
(Caplice 2007; Scott et al. 2015). When a shipper needs a load hauled from an origin to a
destination (“lane”), the load is offered to a contracted carrier at the previously agreed-upon
price. The carrier has the right to refuse a load at the time of execution; in fact, the carrier can
even accept a load but “push it back” (i.e., reject it) at some later time.
The carrier’s right-of-refusal has evolved due to the relatively low value of the
transactions involved. Truckload moves rarely cost more than a couple of thousand dollars, and
the rejection of a few loads does not justify the cost of legal enforcement. This contractual
flexibility introduces an interesting dilemma for carriers when market prices change. After a
price has been agreed upon and if the market turns in their favor, carriers must balance the
conflicting objectives of maintaining a good relationship with their customers by hauling loads
2 Scott et al. (2015) provide an example where spot market usage can, at times, exceed 20% of freight for a shipper.
44
at the contracted price and moving spot loads with significantly higher revenue. C.H. Robinson,
a large transportation broker, acknowledged a carrier’s conflict: “when equipment becomes
scarce, carriers may shift their equipment to transactional customers who will pay higher spot
market rates” (C.H. Robinson 2014, p. 6). To counteract opportunism, shippers do not sue but
threaten to remove future business.3
Masten (2009) examined a class of contracts between drivers and carriers; these exist one
level below the shipper-carrier relationship. In these contracts, carriers and drivers agree on
terms of pay via a rate-per-mile or revenue-sharing agreement for the duration of the
relationship. Because loads are heterogeneous in the space-time continuum, Masten argued that
drivers and carriers pre-specify prices to save on haggling costs which would occur every time a
new load arises.
While similarities exist between shipper-carrier and carrier-driver contracts, there are also
significant differences. First, shipper-carrier contracts pre-specify prices only on particular
lanes; in this sense, they are less general. Second, they often have general volume commitments4
associated with them. Despite these differences, Masten’s explanation that pre-specified prices
reduce negotiation costs seems reasonable in the context of shipper-carrier contracts.
3.2.4 The truckload spot market
The truckload spot market, much like shipper-carrier contracts, is unusual. There is no
centralized exchange (Tsai et al. 2009). Instead, when a shipper needs short-term capacity, there
are at least three mechanisms that can be used: 1) negotiate directly with a carrier in their
3 In a related context, Masten (2009, p. 83) stated in a study examining contracting practices between drivers and
carriers, “Given…the expense of invoking legal sanctions, parties are likely to prefer self-help (such as termination) to court ordering in dealing with transgressions.” This argument is valid with regards to contracts between shippers
and carriers as well as drivers and carriers. 4 Through this research, it is clear that volume commitments are at best rough guidelines for both parties. Shipper’s
demand often falls well below or above the communicated commitment, and carrier’s capacity for loads also often
falls well below or above the communicated commitment.
45
network for capacity; 2) contact a broker and pay it to find capacity (Lindsey et al. 2015); or 3)
utilize an online mechanism to allow carriers and brokers to bid competitively on a load.
Despite the estimation that the spot market is between 5% and 10% of the $300 billion
for-hire trucking market (Caplice 2007; Corridore 2014), there has been little academic analysis
of it. Lindsey et al. (2015) fill some of this void by analyzing the factors that affect the price per
mile that a broker observes. They found that distance, number of stops, lead time, and type of
equipment used are important drivers of spot transportation cost. Scott et al. (2015) also provide
some insight into the spot market. In their study, they estimate that spot prices are on average
62% higher than corresponding contract prices. However, the focus of their study was not on the
spot market, but the causes of contract freight rejection.
Our study adds to this emerging research stream in several ways. First, we empirically
estimate the value of lead time, which has been theoretically analyzed in the literature and
estimated using simulation but not examined using actual industry data. Second, we evaluate
and estimate the factors that affect spot prices in a short-term competitive bidding process,
including day-of-week effects, the impact of weather, and the effect of regional expertise. Third,
we propose a simple-to-implement method to estimate market prices, which shows that price
conditions persist in the short-term.
3.3 Hypotheses
For-hire trucking has been described as a perfectly competitive industry (Belzer 2000) due to the
large number of buyers and sellers, the low barriers to entry, and the fact that a truck and trailer
can be used to haul a wide variety goods. With plenty of lead time and a large number of
carriers, shippers can expect to pay rates roughly equal to the marginal cost of hauling a load.5
5 In this case, “large numbers” bargaining conditions exist, and price will be driven to marginal cost (Williamson
1975, p. 27).
46
Not surprisingly, carrier profit margins are typically less than 10% (Clancy et al. 2008) and in
unfavorable markets, carrier mortality rates are high (Silverman, Nickerson, and Freeman 1997).
Furthermore, the marginal revenue to a shipper of a fulfilled order is generally much
higher than the marginal cost of transportation. Hence, if a shipper faced a monopoly supplier of
transport services, that carrier could extract significantly more revenue than the marginal cost to
haul a load. In time-sensitive spot market settings, we expect prices to exceed contract prices.
We do not formulate this as a formal hypothesis as this is obviously true directly from our
dataset.
We adopt a cost-based pricing lens for our hypotheses, as this is the most prevalent
pricing strategy in industry (Diamantopoulos 1991; Noble and Gruca 1999). In the cost-based
pricing paradigm, the “primary consideration is the internal costs of the firm” (Noble and Gruca
1999, pp. 438-439). Previous transportation research has found that less lead time increases the
operational costs of a carrier (Tjokroamidjojo et al. 2006; Zolfagharinia and Haughton 2012).
This leads to the following hypotheses:
H1A: Carriers bid higher as the time between the auction and load pickup decreases
H1B: Carriers bid higher, at an increasing rate, as the time between the auction and load pickup
decreases
Furthermore, our empirical analysis will allow us to estimate the time profile of spot rates as lead
time diminishes.
Economies of scope (Keeler 1989; Caplice and Sheffi 2003) impact carrier costs. Before
picking up a load, a carrier must relocate a truck to a shipper’s location; likewise, after delivering
a load, the carrier must relocate the truck to another pickup location. Ceteris paribus, a carrier
with a denser network of interconnected freight is able to operate more efficiently than a carrier
47
with a less dense network. We define a “region of expertise” as a geographic region in which
carriers have more domiciled drivers, equipment, and shippers that they service; that is, a denser
network. We predict that a carrier is able to service loads at lower cost in their region of
expertise. This gives the following hypothesis:
H2: Carriers bid lower on loads that are in their region of expertise
3.4 Setting and model specification
3.4.1 Setting
We use a detailed transactional dataset from a large shipper (Acme) in the United States, which
includes all 133,271 bids from its 2014 spot market auctions. Acme operates plants in every
region of the United States. Its loads occupy full trucks in the dry van segment (as opposed to
refrigerated or flatbed trucks). Due to the density of the product, trailers always hit their weight
limits (i.e., they “weigh out” instead of “cube out”). Their product is essentially homogenous
and low value with no practical shelf-life constraints; hence, all loads are basically the same and
well within typical carrier insurance limits, facts well known to all potential carriers.
When an order arrives from a customer, which is usually 5 to 7 days in advance, Acme
plans a load to meet the customer’s requested delivery time window. The load is offered
(“tendered”) to the primary carrier on the lane, with the pickup and delivery time communicated
at the time of offering. The primary carrier has roughly 90 minutes to respond; if they do not
accept the load, it is offered to the next contracted carrier on the lane (i.e., “backup carriers”). If
the load is not accepted by any carrier with 3 days of lead time remaining, it is offered on the
spot market. These spot market transactions are the focus of this paper.
For Acme, spot loads are entered into a software tool that emails carriers in their network
that a load is up for bid. Carriers are given 90 minutes to respond with a bid price, and the load
48
is awarded to the lowest bidder. Hence, the auction process is a first-price sealed-bid auction
(Milgrom and Weber 1982). Sometimes (about 11% in the dataset) loads are not awarded to any
bidder because no prices were acceptable. When this happens, Acme either individually
negotiates capacity with a carrier or plans the load for another time.
This dataset is ideal for estimating the value of lead time because many difficult-to-
observe factors are either absent or controlled for. First, long-term commitments between
shippers and carriers are difficult to observe or measure; in the spot auction, there is no long-
term commitment between the two parties. Second, communication between the parties is
homogenous: they all receive the same emails and see the same auction screen. Finally, there is
no relationship consideration: carriers are not held to a load-acceptance standard and are not
evaluated with respect to their spot behavior.
3.4.2 Variables
In the current study, the unit of analysis is the load-level. Every load that is auctioned in the spot
market has a set of characteristics uniquely associated with it, such as the lane, day of the week,
and hour of day. Variable names and definitions are shown in Table 3-1, along with descriptive
statistics.
3.4.2.1 Dependent Variable
Price premium. The variable of interest, Price premium, is defined as the premium of a spot load
relative to its corresponding contract price on the associated lane. Price premium is calculated as
the observed spot price for a load divided by the 90-day rolling average contract rate on the lane.
For example, if Acme observes a bid of $1,000 on a particular load but the 90-day rolling
average contract rate on the lane was $500, then the Price premium for the load is 2. We test the
90-day rolling average assumption for robustness in Section 3.5.
49
We measure the dependent variable in this manner for two reasons. First, shippers plan
their transportation budgets around contracted rates. Knowing the spot market premiums relative
to contract rates helps shippers understand transportation costs. Second, measuring the premium
as a percentage of contract rates allows comparison across lanes with different characteristics.
Table 3-1. Summary statistics.
3.4.2.2 Independent Variables
Lead time. The difference between the time when a load is made available on the spot market
and the time it is supposed to be picked up is included as an independent variable. It is
calculated as follows. Every load has three timestamps associated with it. The timestamp
indicates the time at which a load is entered into the auction tool (INSERT_DATE). Because the
auction is open for only 90 minutes, this timestamp accurately captures the time at which the
load carriers and brokers consider the load. The second timestamp shows the estimated pickup
Variable Definition N Mean St. Dev
All Bids
Price premium Bid price divided by the average contract rate on a load 117,158 2.372 1.155
Lead time Amount of time, in days, from bid until estimated pickup 117,158 1.501 0.676
Lead time squared The square of Lead time 117,158 2.711 2.203
Same region 1 if the carrier's primary region is in the same region as
the load origin; 0 otherwise
117,158 0.296 0.456
Only Wins
Price premium (above) 24,372 1.969 0.965
Lead time (above) 24,372 1.474 0.703
Lead time squared (above) 24,372 2.666 2.256
Same region (above) 24,372 0.319 0.466
Controls (all bids reported)
Number of invitees Number of carriers invited to auction 117,158 45.566 19.454
Number of bids Number of bids placed during auction 117,158 5.868 2.714
Average temperature Dummy variable for the average temperature at time of
pickup, separated into 10-degree buckets
117,158 59.242 16.986
Lane Lane upon which load travels
Day of week Day of week that load is to be picked up
Hour of day Hour of day that load is to be picked up
Bid day of week Day of week that load is bid upon
Bid hour of day Hour of day that load is bid upon
Carrier Carrier that placed the bid
Calendar week Week during which load is to be picked up
Notes: The table presents the summary statistics for the data in the study.
50
date displayed to the participants of the auctions (PICKUP_DATE). The second timestamp
minus the first timestamp gives the expected lead time for the load. A third timestamp indicates
when the load actually leaves Acme’s dock (START_DATE); we use this as a robustness check
because it serves as a proxy for the estimated pickup date, which is not always entered correctly.
For example, if PICKUP_DATE – INSERT_DATE is 1 hour and 15 minutes, then lead time
takes on a value of 1.25. Thus, lead time is a continuous variable.
Lead time squared. To allow the relation between the spot price premium and lead time to be
nonlinear, the square of lead time is included in the model. A positive coefficient indicates that
the value of lead time decreases in time.
Same region. A dummy variable is included to capture the effect of regional expertise of
carriers. We classify the origin pickup of each load into the general region of the United States
(e.g., Northeast, Midwest). We also have the self-reported primary region of expertise from
most of the carriers. For the few carriers for whom we do not have this information, we used the
region of their headquarters, gleaned from their website. If the load originated in the same
region of the carrier’s primary region, we coded a dummy variable as 1. Because a vast majority
(~80%) of the loads in this study are less than 500 miles, the origin region captures the entire
operating region of a load.
3.4.2.3 Control Variables
Several other factors could affect the cost of a load and hence the spot market price premium.
These are (1) the lane upon which the load travels; (2) the day of week and hour of day that a
load is to be picked up; (3) day of week and hour of day that a load is bid upon; (4) the carrier or
broker placing the bid; (5) the number of bidders invited to the auction and the number of
51
carriers who placed bids; (6) local weather conditions; and (7) the calendar week of the year
during which the load is to be picked up.
Lane. Lane fixed effects (i.e., dummy variables) are included to control for time-invariant
heterogeneity across lanes in the data. These control for observable and unobservable differences
(to the econometrician) that are constant through time such as the distance between the origin
and destination, population densities at the two locations, origin and destination effects, local and
regional competitive factors (e.g., carriers may compete more heavily on some lanes than
others), and average traffic patterns and road quality.
Day of week. The day of the week when a load is to be picked up is important because, for
instance, weekends are likely to be more costly to service because fewer drivers may be
available. Day-of-week fixed effects control for this.
Hour of day. Loads shipped during rush hour may be more costly to serve than loads picked up
in off-peak periods. Hour-of-day fixed effects control for this.
Bid day of week. The day of week when a load is bid upon is likely to impact which firm bids.
Some carriers might have busier days than others and be more or less likely to bid. To control
for this, we include day-of-week fixed effects for the bidding day.
Bid hour of day. The hour of day when a load is bid upon is also likely to impact who
participates in the auction. Fixed effects for the hour of day when each load is bid upon are
included to control for this.
Number of invitees. The number of carriers invited to each auction is not constant across
auctions. We include categorical dummy variables for the number of invitees to control for this
and to allow the estimated relationship to be nonlinear.
52
Number of bids. The number of invited carriers who place bids is also not constant across
auctions. To control for the amount of participation in each auction and allow for nonlinearity,
we include categorical dummy variables for the total number of bids per auction.
Average temperature. Above and beyond the season of the year, outside temperature affects the
operating cost of a carrier. For example, extremely hot or cold temperatures require trucks to
“idle” more when stopped to use air conditioning or heating to keep the driver in a comfortable
environment. Werner, a large truckload carrier, stated in a press release that “severe winter
weather in the first two months of 2014…caused significant freight disruption and weather-
related costs” (Werner 2015, p.1). To control for this, we collected daily high and low
temperatures for 2014 for every origin location in our dataset. These were retrieved from the
United States National Oceanic and Atmospheric Administration National Climatic Data Center
website. We calculated the average of the observed daily high and low temperatures.
To capture non-linear effects of temperature, we include dummy variables for
temperature ranges. The temperature ranges are between 30 and 40 degrees Fahrenheit (°F) (-
1.1 to 4.4 degrees Celsius (°C)), between 40 and 50 °F (4.4 to 10.0 °C), between 50 and 60 °F
(10.0 to 15.5 °C), between 60 and 70 °F (15.5 and 21.1 °C), between 70 and 80 °F (21.1 and 26.7
°C), between 80 and 90 °F (26.7 and 32.2 °C), and above 90 °F (32.2 °C). The excluded variable
is below 30 °F (-1.1 °C).
Carrier. Carrier fixed effects control for differences across carriers, such as size or profit targets.
3.4.2.4 Estimating market prices
Calendar week. The balance between supply and demand is not known with any great precision
or timeliness in truckload transportation (see Bignell 2013 for a discussion of the lack of an
effective truckload index). Hence, the current study provides a textbook example of an implicit
53
market (Rosen 1974), because market prices are not centralized or public. There are many
buyers and sellers of truckload transportation services, so it is a thick market. Prices adjust
quickly to balance supply and demand. As a result, the prices paid by any individual shipper will
generally be a good indicator of local supply and demand conditions. Hence, we propose that
observing carrier and broker bids over time and controlling for all relevant operational factors
provides a good representation of supply and demand conditions. This method resembles the
operation of the only transportation exchange operated, the BIFFEX, where a group of eight
brokers submitted prices daily on a set of 13 ocean routes (Denning, Riley, and Delooze 1994).
Weekly dummy variables included in the regression model capture the relative magnitude
of national market prices. This approach is commonly used to estimate prices in implicit markets
(Diewert, Heravi, and Silver 2007). The proposed method allows for the granular estimation
(e.g., at the regional level) of market prices in near-real-time. Such analysis currently does not
exist publicly, so our results may be useful for shippers, carriers, and financial investors.
4.3 Data
Access was granted from Acme to a dataset that included all bids for spot loads in calendar year
2014. In all, there were 133,271 bids and 31,218 offered loads, for an average of 4.27 bids per
offered load. 27,802 of the offered loads were awarded to a bidder. As with any large industry
dataset, some records were clearly inaccurate and needed to be removed. Per discussions with
Acme, all spot loads that go through the normal spot process are offered between 0 and 3 days.
Removing loads that showed a lead time less than 0 days and greater than 3 days resulted in
117,924 bids and 24,506 awarded loads. The removed records are likely the result of incorrectly
entered pickup appointments; for example, a pickup appointment before the load offer is clearly
inaccurate. Second, some bid prices were entered erroneously or disingenuously by the bidder.
54
For example, some Price premiums are quite large, more than 8 times the corresponding contract
price. We removed all Price premiums that were less than 0.5 or greater than 8. This resulted in
117,158 bids and 24,372 awarded loads, or 87.9% and 87.6% of the original dataset,
respectively. The decisions to remove these observations are tested for robustness in Section 5.
3.4.4 Model specification
The level of analysis is at the load level – all variables are calculated for every load. The
controls included in the model capture the relevant operational and market factors that are likely
to affect the bid price for a load. The model is specified as follows:
0 1 2 3
Price Premium Lead time Lead time squared Same region Lane
Day of week Hour of day Bid day of week Bid hour of day
Number of invitees N
umber of bids Average temperature
Carrier Calendar week
(2)
Each i is the coefficient for the variables defined in Section 3.4.2, is the idiosyncratic error
term for each load, and each categorical variable and fixed effect is represented by its name in
(1).
We use clustered standard errors to allow for heteroskedasticity and correlation of errors
in the variance-covariance matrix. Clustering at the carrier level allows for correlation within a
carrier’s spot bids; this seems likely as the same person at a carrier may place the bids. This
results in 57 clusters, which is above the threshold of 50 suggested by Wooldridge (2003).
3.5 Results
3.5.1 Main results
Table 3-2 shows the results of the main analysis. Hypotheses 1a and 1b are both supported, as is
hypothesis 2. The control variables generate reasonable results, such as higher observed prices
55
during extreme low and high temperatures and higher prices on weekends. Truckload market
prices, as measured by the weekly dummy variables, exhibit high serial correlation.
Lead time has a significant economic impact on truckload spot prices. The impact of lead
time is decreasing in time, consistent with previous findings from the carrier’s point-of-view.
Therefore, hypothesis 1A and 1B are supported. It is important to be clear on the interpretation
of the coefficients. Given that the coefficient in column 1 for Lead time is -0.202 and Lead time
squared is 0.030, the estimated value of one day of lead time versus 0 days of lead time is -0.202
* 1 + 0.030 * 12 = 0.172, or 17.2% of the corresponding contract price. For example, if the
contract price on a lane is $500, then the estimated effect of one day of Lead time and Lead time
squared is -0.172 * $500 = -$86 for a load on that lane.
For the following we focus on winning bids, as shown in column “Wins only” of Table 3-
2. Results are similar for winning bids and all bids, but because the shipper only pays winning
bids, they are more relevant for the following analysis. The mean Price premium for winning
bids is 1.969, meaning that the average winning bid is 96.9% higher than the corresponding
contract price. When the lead time increases from zero days to one day, the price premium
declines by 17.2 percentage points; or, on average, from 1.969 to 1.797. When lead time
increases from one to two days, the price premium further declines by 11.2 percentage points, or
on average from 1.797 to 1.685. A third day of lead time causes a further 5.2 percentage point
decline. Hence, a load with three days of lead time has a Price premium about 33.6 percentage
points lower than a load with no lead time; a load with three days of lead has a Price premium
that is 16.4 percentage points lower than a load with one day of lead time. Even so, with three
days of lead time, a spot load still has a Price premium of 1.633, or 63.3% higher than the
56
associated contract price. Figure 3-1 shows the reduction of Price premium as the days of lead
time increase.
Table 3-2. Effect of lead time on spot market bid premium.
For a shipper that spends $100 M on transportation at contract rates and with a spot percentage
of 10% (these numbers are for illustration and not necessarily representative of Acme), the
difference in cost given zero days of lead time versus three days of lead time is significant. In
the first case, spot transportation would cost $19.69 M, while with three days of lead time the
cost is $16.85 M, a difference of $2.84 M. In this scenario, the value of the first day of lead time
is $1.72 M, the second day is $1.12 M, and the third day is $520,000. Depending on the cost of
Dependent Variable:
Wins only All bids
Lead time -0.202*** -0.235***
Lead time squared 0.030** 0.046***
Same region -0.063** -0.160***
Intercept 2.588*** 3.460***
Lane Yes Yes
Day of week Yes Yes
Hour of day Yes Yes
Bid day of week Yes Yes
Bid hour of day Yes Yes
Number of invitees Yes Yes
Number of bids Yes Yes
Average temperature Yes Yes
Carrier Yes Yes
Calendar week Yes Yes
Observations 24,372 117,158
R-squared 0.732 0.642
Notes: The table displays the effect on lead time, measured in days
(continuous), on Price premium. Column "Wins only" includes only
winning bids. Column "All bids" includes all bids. Errors are clustered
by carrier.
Price premium
* denotes a 10% significance level, ** denotes a 5% significance level,
*** denotes a 1% significance level
Included?
57
advance notification, there are tangible and significant benefits for a shipper to share load
information in a timely manner, and to access the spot market as soon as a load rejection occurs.
Figure 3-1. Price premium decrease versus days of lead time, for winning bids. The curve
clause actually provides is minimal. Based on a report from Acme, the “maximum accept-up-to”
limit was hit less than 3% of the lane-days, affecting an even smaller percentage of load offers.
Nonetheless, we adopt strategies to control for unobservable carrier preferences to select into this
additional clause.
Output monitoring
Primary carriers are expected to accept most loads and their performance is scrutinized by output
monitoring (Heide et al. 2007); if they do not perform adequately, they are replaced by other
carriers. Backup carriers, on the other hand, are not held to load-acceptance performance
standards (Caplice 2007) and are not monitored. In essence, they only accept loads that are in
their own objective interest to haul without considering long-term ramifications. Acme’s
systems automatically monitor carrier load-acceptance performance of primary carriers and they
have regular reviews with carriers.
Market opportunism in FHTL
Carriers reject load offers a significant percentage of the time. Caplice (2007) analyzed a large
dataset and found that load offers were rejected 26% of the time. There are two main reasons
why they do so (Scott et al. 2016): (1) if the market shifts in the favor of the carrier, at which
point previously negotiated prices are no longer attractive, or (2) if the shipper has a spike in
demand, at which point the carrier either does not have assets to service the demand or it is
prohibitively expensive to reposition assets to do so.
C.H. Robinson, a large buyer and seller of truckload services, describes market
opportunism in FHTL: “when equipment becomes scarce, carriers may shift their equipment to
transactional customers who will pay higher spot market rates” (C. H. Robinson 2014, p. 6).
This causes major problems for shippers, including higher costs and poor service levels to their
78
customers (Zsidisin, Voss, and Schlosser 2007). The tension caused by load offer rejections due
to static contract prices and high spot prices is palpable in the words of a vice president at a
carrier: “generally, there are no volume guarantees, nor financial penalties, so essentially when
load acceptance rates fall, a lot of yelling and hollering is what happens” (Taylor 2011, p. 3).
4.4 Empirical Strategy and Data Source
The data to test our hypotheses come from Acme, a large shipper with plants throughout the
United States. We focus only on Acme’s FHTL carriers, who haul > 95% of all of their goods.
Because Acme and their carriers endogenously select into the form of governance, we use a
method to account for selection bias suggested by Vella (1998) and Tucker (2010) and used by
other business researchers (e.g., Main and Reilly 1993, Holloway and Parmigiani 2014). Below
we discuss our treatment and control groups, how we measure FHTL market conditions (and
hence the payoff for market opportunism), the data used in the analysis, the variables and model
specification, our identification strategy, and our clustering strategy.
Treatment and control groups
Our control group is the set of backup carriers. They are not governed by contractual clauses,
monitoring, or commitments – hence, they are an excellent control group because they provide a
glimpse into how a carrier would behave in the absence of the governance mechanisms. One
alternative treatment group is the set of “implicit carriers”; their treatment is output monitoring
and the commitment of a primary carrier. The other alternative treatment group is the set of
“explicit carriers”; their treatment is output monitoring, the commitment of a primary carrier, and
the additional “accept-up-to” clause in the contract. To be clear, the actual business entity that is
a carrier can be a backup carrier on some lanes, an implicit carrier on others, and an explicit
79
carrier on other lanes. Table 4-1 shows the treatment and control groups and their corresponding
forms of governance.
Table 4-1. Control and treatment groups.
Detecting FHTL market conditions
A widely-used centralized marketplace in FHTL does not exist (Tsai, Regan, and Saphores 2009)
– hence, detecting market conditions in near-real-time is challenging. One solution is to identify
the relevant operational variables describing a load, include a time-period dummy variable to
capture latent market conditions (Diewert, Heravi, and Silver 2007), and regress these variables
on time-varying prices. The included dummy variables capture time-period effects in the
varying prices, thus generating a price index indicative of market conditions.
Two elements of data are necessary for this strategy – time-varying prices and the
operational aspects of the loads associated with the prices. We use carrier bids in online spot
auctions conducted by Acme for our purposes. In the spot auctions, a large number of carriers
(typically 50 or so) are invited to bid on one-off loads in a first-price sealed-bid format. They are
displayed the relevant load information, such as the origin and destination locations, number of
miles, expected weight of the load, and pickup time. Auctions are open for 90 minutes.
To estimate market conditions, we adopt a specification suggested by transportation
researchers (Scott 2015). We provide a brief discussion of the approach here; for a detailed
Group Type No governance Mutual
commitment
Output
monitoring
Extra contractual
clause
Backup carriers Control x
Implicit carriersAlternative
treatmentx x
Explicit carriersAlternative
treatmentx x x
80
justification, see Scott (2015). First, given contract prices9 on a lane and a bid price on a spot
load on the same lane, the ratio of the bid and contract price provide a measure of the “premium”
associated with a particular bid. For example, if the contract price on a lane is $500 and a spot
bid is $1,000, then the premium would be 2. All else equal, a higher premium indicates that
supply and demand conditions are more favorable to carriers, whereas low premiums indicate the
opposite10
. The advantage of converting raw price data to a ratio is the ability to compare across
loads, most of which have different characteristics.
The spot premium for a load can be broken down into cost components. For example, the
cost of a load depends on the distance, lead time, the day of week of pickup, origin and
destination attributes, and even the temperature of the external environment (e.g., it is costlier to
operate a truck in extremely cold weather). The model specification is
0 1 2 3
Price Premium Lead time Lead time squared Same region Lane
Day of week Hour of day Bid day of week Bid hour of day
Number of invitees N
umber of bids Average temperature
Carrier Number of origin loads Calendar week
(1)
which is the same specification as Scott (2015) except that we have included fixed effects for the
number of loads at each origin for each day to account for any capacity depletion effects. Table
4-2 describes the variables included and associated summary statistics, which are described in
detail in chapter 3.
9 Contracted prices are close to the marginal cost to operate a load due to the highly competitive nature of the
industry and the “large numbers” bargaining conditions that exist at the time of price negotiations (Williamson 1975,
p. 27). 10 In a transportation industry report by Standard & Poor’s, Kirkeby (2013, p. 7) says: “S&P believes pricing trends
in the [trucking] spot market provide insight into the general availability of capacity and demand for that capacity.”
81
Table 4-2. Data description and summary statistics for price index analysis.
Our dataset contains 133,271 bids in the calendar year 2014. We remove bids that are
less than 50% or greater than 800% of the corresponding contract prices on a lane, because some
bids were clearly data entry errors (e.g., $9,999; $0). This removes 878 observations, less than
0.6% of our sample.
To capture regional variation, we assigned the origin and destination region for each load
using the regional classifications from the U.S. Census Bureau (Northeast, Midwest, South,
West). For example, if a load begins in Arizona and ends in Indiana, it would be classified as
originating in the West and ending in the Midwest. We then ran regression model (1) for each
region combination, where all loads that either begin or end in a region (or both) are included:
within Northeast, within Midwest, within South, within West, Northeast-Midwest, Northeast-
South, Northeast-West, Midwest-South, Midwest-West, and South-West. For example, a bid on
a load that begins in the West region and ends in the Midwest region would be included in the
Midwest-West price index; a load that begins in the Midwest and ends in the Midwest would be
included in the within Midwest, Northeast-Midwest, Midwest-South, and Midwest-West price
index. The 10 resultant indexes are shown in Figure 4-1.
Variable Definition Mean St. Dev
Price premium Bid price divided by the average contract rate on a load 2.382 1.165
Lead time Amount of time, in days, from bid until estimated pickup 1.764 1.026
Lead time squared The square of Lead time 4.165 7.142
Same region 1 if the carrier's primary region is in the same region as the
load origin; 0 otherwise
0.296 0.456
Number of invitees Number of carriers invited to auction 45.852 19.387
Number of bids Number of bids placed during auction 5.834 2.695
Average temperature Dummy variable for the average temperature at time of
pickup, separated into 10-degree buckets
5.953 1.696
Lane Lane upon which load travels N/A N/A
Day of week Day of week that load is to be picked up N/A N/A
Hour of day Hour of day that load is to be picked up N/A N/A
Bid day of week Day of week that load is bid upon N/A N/A
Bid hour of day Hour of day that load is bid upon N/A N/A
Carrier Carrier that placed the bid N/A N/A
Number of origin loads Number of loads bid upon on that day from that location 12.385 9.49
Calendar week Week during which load is to be picked up N/A N/A
82
Figure 4-1. The ten indexes over 2014 (Northeast = NE, Midwest = MW, South = S, West =
W).
There are a couple of interesting observations about the price indexes. First, there is
significant variation in the spot price of FHTL services over the year. Given the mean spot
premium of a bid of 238% (shown in Table 4-2), the average spot premium varies from the
minimum to maximum by about 150 percentage points. This means that spot prices can, at
times, approximately equal contract prices, while at other times average more than 2.5 times the
contract price. Second, the price indexes are highly correlated. The average correlation is about
84% and the least correlated series, “within Midwest” and “within West”, have a positive
correlation of 65.8%. This is expected because of seasonal nation-wide economic output and if
imbalances in a particular region were to persist over time, a truck could be relatively easily
repositioned to the region with high prices.
Despite the fact that our spot price dataset consists of more than a hundred thousand spot
prices from a large number of carriers, including bids from large brokers who themselves have
-150%
-100%
-50%
0%
50%
100%
150%
NE-NE NE-MW NE-S NE-W MW-MW
MW-S MW-W S-S S-W W-W
-150%
-100%
-50%
0%
50%
100%
150%
NE-NE NE-MW NE-S NE-W MW-MW
MW-S MW-W S-S S-W W-W
JAN FEB MAR APR MAY JUN JUL AUG SEP OCT NOV DEC
Pri
ce P
rem
ium
Re
lati
ve to
We
ek
1
83
connections with a large number of carriers, it is possible that these spot prices do not represent
conditions in the overall market. For validation, we compared a nationalized version of our spot
price index (i.e., model (1) not subdivided into regions) with a biweekly “supply and demand
sentiment” index, published by Morgan Stanley, a large U.S.-based investment bank, which can
be found at the website of Transplace, a third-party logistics provider (Transplace 2016). We
converted the “sentiment” index into weekly data points and compared it to our national price
index. The correlation between the time series is 68%, which further validates the representation
of our spot price index of the broader market. The time series are shown in Figure 4-2.
Figure 4-2. The national spot price index versus the Morgan Stanley sentiment index.
Market opportunism
We use our regional indexes of market conditions to classify when the market is favorable or
unfavorable for a carrier. For our main analysis, we classify the market into thirds: unfavorable
market, moderate market, and favorable market market conditions. To test this classification for
robustness, we also run our main analysis including the market segmented into halves and with a
linear and quadratic term instead of discrete classifications.
0
1
2
3
4
5
6
7
8
9
10
-100%
-80%
-60%
-40%
-20%
0%
20%
40%
60%
80%
100%
Spot Price Index Morgan Stanley Sentiment IndexJAN FEB MAR APR MAY JUN JUL AUG SEP OCT NOV DEC
Pri
ce P
rem
ium
Re
lati
ve to
We
ek
1
0
1
2
3
4
5
6
7
8
9
10
-100%
-80%
-60%
-40%
-20%
0%
20%
40%
60%
80%
100%
Spot Price Index Morgan Stanley Sentiment Index
Sup
ply / D
em
and
Sen
time
nt
84
For each time series, we rank each weekly observation by magnitude and classify each
week into one of the three categories. For example, in week 45 of the “within Midwest” time
series, the index value is -129%, which is the lowest value in the series; week 50 has an index
value of -90%, in the middle third of series values; and week 13 has an index value of 42%,
which is the highest in the series. Hence, week 45 is classified as an unfavorable market for a
carrier, week 50 is a moderate market, and week 13 is a favorable market. We use these
classifications as dummy variables in our analysis. If a load offer occurs in the “within
Midwest” region in, say, week 50, then this load offer will have occurred during a moderate
market – i.e., the moderate market dummy variable will be 1 and the unfavorable market and
favorable market dummy variables will be 0. Table 4-3 shows an example of our classification
system, where we also show classification of the market into halves.
Data
Acme’s loads typically travel from a plant-warehouse to a customer’s distribution center. Due to
the density of the product, trailers always hit their weight limits. For practical purposes, their
product is homogenous, low value, and with no shelf-life constraints; hence, all loads are
basically the same and well within typical carrier insurance limits, facts well known to carriers.
Because negotiations with carriers took place early in 2014, our analysis is based on load
offers from March 1st to December 31
st, 2014. We use several datasets for our analysis, listed in
Table 4-4. The first dataset contains all EDIs between Acme and their carriers. From this
dataset, we observe whether the carrier accepted or rejected the load offer and a timestamp of
each transmission. A second dataset contains general load information for all loads hauled in
2014 – origin, destination, carrier, price paid, time of pickup, whether it was at a contracted or
85
spot price, and other data not used in our analysis. A third dataset indicates whether the carrier at
the
86
Table 4-3. Example of market status classification, using the “within Midwest” series.
Region Week Price Index Unfavorable Moderate Favorable Bottom Half Top Half
within Midwest 45 -129% 1 0 0 1 0
within Midwest 46 -122% 1 0 0 1 0
within Midwest 47 -121% 1 0 0 1 0
within Midwest 37 -121% 1 0 0 1 0
within Midwest 44 -121% 1 0 0 1 0
within Midwest 39 -120% 1 0 0 1 0
within Midwest 33 -117% 1 0 0 1 0
within Midwest 40 -112% 1 0 0 1 0
within Midwest 38 -111% 1 0 0 1 0
within Midwest 31 -111% 1 0 0 1 0
within Midwest 34 -110% 1 0 0 1 0
within Midwest 32 -110% 1 0 0 1 0
within Midwest 51 -110% 1 0 0 1 0
within Midwest 41 -109% 1 0 0 1 0
within Midwest 43 -105% 1 0 0 1 0
within Midwest 42 -103% 0 1 0 1 0
within Midwest 35 -102% 0 1 0 1 0
within Midwest 49 -99% 0 1 0 1 0
within Midwest 30 -96% 0 1 0 1 0
within Midwest 50 -90% 0 1 0 1 0
within Midwest 36 -78% 0 1 0 1 0
within Midwest 29 -71% 0 1 0 1 0
within Midwest 48 -69% 0 1 0 0 1
within Midwest 23 -53% 0 1 0 0 1
within Midwest 18 -46% 0 1 0 0 1
within Midwest 22 -43% 0 1 0 0 1
within Midwest 21 -40% 0 1 0 0 1
within Midwest 19 -38% 0 1 0 0 1
within Midwest 20 -38% 0 1 0 0 1
within Midwest 16 -27% 0 1 0 0 1
within Midwest 28 -26% 0 0 1 0 1
within Midwest 26 -24% 0 0 1 0 1
within Midwest 17 -19% 0 0 1 0 1
within Midwest 25 -8% 0 0 1 0 1
within Midwest 27 -1% 0 0 1 0 1
within Midwest 15 -1% 0 0 1 0 1
within Midwest 52 0% 0 0 1 0 1
within Midwest 24 4% 0 0 1 0 1
within Midwest 14 15% 0 0 1 0 1
within Midwest 12 28% 0 0 1 0 1
within Midwest 10 30% 0 0 1 0 1
within Midwest 9 33% 0 0 1 0 1
within Midwest 11 38% 0 0 1 0 1
within Midwest 13 42% 0 0 1 0 1
87
time of load offer was a primary carrier or a backup carrier. A fourth dataset contains
contractual “accept-up-to” limits (i.e., explicit commitments). A fifth dataset contains bids in
spot auctions from 2014, which we use to estimate market conditions. A sixth table contains
load information in 2013.
Table 4-4. Data tables.
Our dataset has 178,643 load offers, representing well over $100 million in offered
business. We took several steps to ensure the records were accurate. First, a small number of
records had extraordinarily small prices associated with them. Removing load offers showing
prices less than 50% of other contract prices on a lane removes 741 (0.4%) of the records. Logit
and probit estimation requires variation on lanes and carriers – i.e., some carriers rejected no load
offers, and on some lanes there were no rejections. Removing these takes out 8,617 (4.8%) of
the records. We removed all lanes that had less than 20 load offers on them to minimize bias
when estimating the logit and probit models with fixed effects (Katz 2001). This removed 44 (<
0.1%) of the records. Finally, we have to include data from 2013 for reasons discussed shortly.
Removing lanes that did not have loads moved on them in 2013 removed 4,127 (2.3%) of the
records. This leaves a total of 165,114 (92.4%) of the original 178,643 records.
To check for potential bias due to the removal of these records, we compared the
rejection rate of the load offers. In the dropped records, 20.5% of load offers were rejected; in
Table number Name Description Useful fields
1 EDIsLoad offers and carrier responses in 2014, including
timestamps of each transmissionShipment ID, carrier ID, carrier response, timestamps
2 Load information - 2014 Information on every load shipped in 2014Shipment ID, origin, destination, carrier ID, price paid, time of
pickup, whether it was at a contracted or spot price
3 Primary or backup Primary and backup carriers on each lane in 2014 Shipment ID, origin, destination, carrier ID, primary or backup
4 Explicit commitmentsCarriers and their contractual daily "accept-up-to"
commmitments in 2014Origin, destination, carrier ID, day of week, maximum up-to limit
5 Spot bids Bids in spot auctions in 2014Shipment ID, carrier ID, origin, destination, bid price, winner or
not flag, bid timestamp
6 Load information - 2013 Same as (2), but for 2013Shipment ID, origin, destination, carrier ID, price paid, time of
pickup, whether it was at a contracted or spot price
88
the kept records, 26.4% of the load offers were rejected. We compared the percentage of load
offers that went to explicit, implicit, and backup carriers. In the dropped records, these
percentages are 50.0%, 30.0%, and 20.0%, respectively. In the kept records, the percentages are
58.2%, 22.1%, and 19.7%, respectively. Because less than 10% of our records were dropped and
there are no glaring disparities in any of these numbers, we are unconcerned with bias from their
removal.
Self-selection
Acme and its carriers endogenously select into the backup, implicit, and explicit carrier
agreements. Without correcting for endogeneity, estimates will be biased and inconsistent
(Greene 2008). In our case, there are two endogenous carrier preferences: preferences for the
lanes that are up for bid in the auction and preferences to adopt each contract type.
We include two variables to control for lane preferences. First, we include the carrier-
lane prices11
associated with each load offer. Because carrier-lane prices are generated via a
competitive, first-price sealed-bid annual procurement auction in which many carriers
participate, the bid prices closely reflect the carrier’s private valuation of a load on a lane12
.
Second, we include lane fixed effects to control for average lane preferences across carriers. For
example, if one lane is particularly unattractive for all contracted carriers, the lane fixed effect
will capture this fact.
To control for the endogenous selection into backup, implicit, and explicit contracts, we
adopt a strategy discussed by Vella (1998) and Tucker (2010) that is used when selection
11 We transform these into “price premiums” in our model for reasons discussed shortly. 12 Technically, a first-price sealed-bid auction is not a demand revealing auction. However, the optimal strategy is
to bid between the bidder’s own valuation and the expected valuation of the other bidders. As the number of bidders
increases, the optimal bid strategy gets arbitrarily close to the bidder’s private value (Kagel and Levin 1993).
Because there are close to a hundred carriers invited to the annual auctions, the bidders bid very close to their own
private values.
89
involves more than two groups. The type of contract is ordered in the number of governance
mechanisms – backup carriers have fewer than implicit carriers who have fewer than explicit
carriers. Hence, we create a contract type variable with backup carriers as 1, implicit carriers as
2, and explicit carriers as 3. The selection into the contract type depends on the projected lane
volume by Acme, which is common in FHTL procurement auctions (Caplice 2007), and
individual carrier-region preferences, where we define region the same as the U.S. Census
Bureau. After discussions with Acme, we included the lane volumes from 2013 as the
projections for 2014, and we included carrier-region preferences as dummy variables. We then
estimated an ordered probit model where the dependent variable is contract type and the
independent variables are the lane volume projections and carrier-region dummies. From the
selection model, we can then retrieve selection correction factors (selection correction) to
include in the main model. The selection model is identified by the inclusion of lane volumes
from 2013; these are exogenous to the decision to accept or reject a load offer, except through
the selection into the governance mode. The procedure and results from the selection model are
discussed in detail below. As expected, the higher the projected load volume, the more likely
Acme and carriers are to select into more governance – the explicit contract.
Selection correction procedure
As discussed in Section 3.4, we have a variable, contract type, that takes on three values and is
ordered in the number of governance mechanisms. Acme suggested that they and the carriers
generally select into more governance based on projected load volumes. Hence, we model the
selection process of Acme and the carriers using projections of load volumes for every lane and
unobserved carrier-region preferences. We include dummy variables for carrier-region
preferences, allowing the selection model the maximum amount of flexibility. If we had more
90
than one shipper, we would include dummy variables to allow for their unobserved preferences;
since we only have one, a dummy variable would be redundant. To estimate projected load
volumes, we use volumes from 2013 (2013 load volumes) on the same lanes and check this for
robustness using actual load volumes from 2014. Our results are consistent using either.
The procedure is to regress contract type on 2013 load volumes and the carrier-region
dummy variables using an ordered probit model. Two thresholds are calculated from this model,
and we calculate the selection correction using the equation discussed in Vella (1998, p. 147,
Section VI A). Main and Reilly (1993) and Holloway and Parmigiani (2014) discuss this
approach in more detail. Our Stata code is below.
*contract_type is contract type, lane_loads_2013 is 2013 load volumes
Note: *** p<0.01, ** p<0.05, * p<0.1. Standard errors are clustered at the carrier-census region-week level and are reported in parentheses. Column
(1) reports the linear probability model (LPM) with the main effects but without the selection correction factor, (2) reports the LPM with the main effects
and the selection correction factor, (3) reports the LPM and the interactions between governance form and market conditions, (4) reports a probit
specification, (5) reports a logit specification, and (6) reports the LPM but with standard errors clustered by carrier-census region-month instead of
carrier-census region-week. Lane, carrier, hour-of-day, day-of-week, month, and days of lead time fixed effects are included in all models. For the
probit and logit models, the constant term is included and explicit carriers * unfavorable market is the omitted category. Pseudo R-squared values
are reported for the probit and logit models.
97
carriers, and explicit carriers reject loads 12.3% less frequently (p-value < 0.01) than backup
carriers. The difference in behavior is more apparent when market conditions are included, as
shown in Figure 4-3. In unfavorable market conditions, a Wald test that implicit carriers behave
the same as backup carriers (i.e., H0: implicit carriers*unfavorable market = backup
carriers*unfavorable market) is not rejected, but in favorable markets implicit carriers reject
loads 11.6% (p-value < 0.01) less frequently than backup carriers . Explicit carriers reject loads
8.1% (p-value = 0.057) less frequently than backup carriers in unfavorable markets, but 16.4%
(p-value < 0.01) less frequently in favorable markets. Commitment and monitoring protect
against short-run opportunistic behavior.
Carriers with explicit contracts provide better service on average than carriers with
implicit contracts, supporting the increased coordination and enhanced monitoring that more
explicit contracts entail. On average, explicit carriers reject loads 7.9% (p-value < 0.01) less
frequently than implicit carriers, and better performance persists across market conditions.
Hence, H1 is supported.
Our second hypothesis postulates that the effectiveness of a more explicit contract
weakens relative to commitment and monitoring as the payoff to market opportunism increases.
To analyze this, we look at the rejection rate differential for each type of carrier across market
conditions. Figure 4-3 and column 3 of Table 4-8 show marginal rejection rates by market
condition and mode of governance.
98
Figure 4-3. Marginal response rates by governance mode and market conditions.
First, all three carrier types increase in their likelihood to reject load offers as the market
becomes more favorable for them. The marginal rejection rate for backup carriers is 23.7% in
unfavorable markets but 41.7% in favorable markets; the difference between the two, 18.0%, is
significant (p-value < 0.01). Likewise, implicit carriers are 5.6% (p-value < 0.01) more likely
and explicit carriers are 9.7% (p-value < 0.01) more likely to reject load offers when market
conditions change in their favor. None of the governance mechanisms completely restrain
market opportunism.
We are interested in testing the response differential of the carrier types. A Wald test on
the hypothesis that implicit carriers respond the same as backup carriers to changes in market
conditions from unfavorable to favorable (i.e., H0: implicit carriers*favorable market conditions
N 165,114 165,114 165,114 165,114 165,114 144,949 165,114 165,114
Note: *** p<0.01, ** p<0.05, * p<0.1. Standard errors are clustered at the carrier-census region-week level unless otherwise noted. Lane, carrier, hour-of-day, day-of-week, month, and
days of lead time fixed effects are included in all models unless otherwise noted. All models use the LPM specification. Column (1) reports the main model except with month-region fixed
effects instead of month fixed effects, (2) uses carrier-region fixed effects instead of carrier fixed effects, (3) clusters standard errors at the carrier-region level instead of the carrier-region-
week level, (4) clusters standard errors at the carrier-week level instead of the carrier-region-week level, (5) omits load offers with a lead time of less than 3 days, (6) uses 2014 as the lane
selection criteria instead of 2013, (7) segments the market into halves instead of thirds, and (8) uses a linear- and quadratic-term instead of dummy variables, where the linear term is in the
"unfavorable market" rows and the quadratic term is in the "moderate market" rows.
accept or reject
102
For every same region spot price index, we calculated an other region spot price index
using the same model specified in Section 4.4. We then run a two-stage least squares (2SLS)
estimation procedure including the instrument and run the Wu-Hausman test post-estimation.
The null hypothesis is that the same region spot price index is exogenous. The test fails to reject
the null hypothesis by a large margin (p-value = 0.6569). Further, estimates from the OLS
model with the same region spot price index as exogenous and estimates from the 2SLS
procedure are nearly identical, shown in Table 4-11. We conclude that spot prices are exogenous
to a particular carrier.
Table 4-11. Regression using OLS and 2SLS.
4.7 Discussion
The constellation of governance mechanisms enacted between a buyer and supplier significantly
affects exchange outcomes. Some governance mechanisms are fundamentally complements of
Dependent variable:
(1) (2)
same region price index 0.109*** 0.096***
(0.0111) (0.0287)
implicit carriers -0.043 -0.044
(0.0267) (0.0267)
explicit carriers -0.121*** -0.122***
(0.0375) (0.0374)
above average volume 0.073*** 0.073***
(0.0057) (0.0057)
price premium -0.347*** -0.346***
(0.0232) (0.0232)
selection correction -0.077*** -0.077***
(0.0146) (0.0145)
constant 0.311*** 0.318***
(0.0640) (0.0658)
R2
0.288 0.288
N 165,114 165,114
accept or reject
Note: *** p<0.01, ** p<0.05, * p<0.1. Standard errors are clustered at the carrier-census
region-week level. Lane, carrier, hour-of-day, day-of-week, month, and days of lead time
fixed effects are included for both models. Column 1 includes the same region price index as an
exogenous variable. Column 2 uses 2SLS with the other region price index as an instrument.
103
one another – e.g., contracts that are more explicit about a supplier’s output enhances the
monitoring of such output (Heide et al. 2007). Others are substitutes – e.g., strong-arming a
supplier to provide service at static prices is naturally at odds with providing dynamic price
incentives for performance. Buyers should consider these interactions when deciding which
governance mechanisms to negotiate and adopt with their suppliers.
To complicate matters, the legal enforceability of the environment affects the
performance of governance mechanisms individually and in relation to one another (Zhou and
Poppo 2010). One can imagine a spectrum of legal enforceability, with “fully and clearly legally
enforceable” on one extreme and “completely unenforceable” on the other. Where a particular
business environment falls on this spectrum affects the behavior of suppliers.
In this study, we consider an environment closer to “completely unenforceable” than
“fully and clearly legally enforceable.” Contracts are signed and exchanged, but the threat of
legal enforcement is not credible. We find that monitoring a carrier’s output, combined with the
promise of future business, improves performance and restrains opportunism. More explicit
contracts improve coordination between shipper and carrier and enhance output monitoring; we
observe this via better average performance by carriers who are subject to more explicit
contracts.
Previous researchers have made the important observation that the form of opportunism
affects the performance of governance mechanisms (Wathne and Heide 2000, Seggie et al.
2013). We argue that not only does the form matter, but that the payoff to a supplier to act
opportunistically does as well. This is rarely addressed or discussed by researchers. We find
that, when the payoff to opportunism is large, all three of the governance mechanisms considered
in this study – commitment, monitoring, and more explicit contracts – “break down” to some
104
degree. All carriers are more likely to reject load offers when the market is highly in their favor
than when it is not. However, commitment and monitoring are relatively more effective than
explicit contracts when the payoff to opportunism is high. We argue that the high cost associated
with the removal of future business keeps carriers from reneging completely on their
commitments even when short-term payoffs are high; the relatively lower cost of the shipper
“yelling and hollering” at the carrier to fully live up fully to their explicit contract loses its
effectiveness when the payoff to opportunism becomes too large. Thus, when market conditions
become highly favorable for suppliers, buyers might be better off utilizing other mechanisms,
such as price incentives.
Limitations and future research
Our study has several limitations. First, we use data from one company in one particular
industry for ten months for our analysis. While we do not believe that Acme’s operations are
particularly unique and many other truckload shippers operate and contract with carriers in a
similar manner (Caplice 2007), we cannot completely rule out idiosyncrasies. Our time frame,
2014, was a relatively strong year for carriers in terms of market supply and demand (Transplace
2016). While this makes carrier decision making more interesting as they balance short-term and
long-term profitability, if this analysis were performed in a less favorable time frame for carriers,
we might observe different behavior. Moreover, FHTL is a large, well-established industry, with
its own norms and practices. Thus, the generalizability of our findings to other industries is
unclear.
From a study-design perspective, the effectiveness of the primary carrier commitments
and output monitoring cannot be distinguished from one another because they go hand-in-hand –
there are no primary carriers who are not monitored and there are no monitored carriers who are
105
not primary. Also, the coordinating advantages of more explicit contracts along with the
enhancement of output monitoring that more contract explicitness brings cannot be separated
from one another. We believe studies that try to tease out these individual effects are worthwhile
pursuits. Despite these drawbacks, we believe our high level of institutional detail, usage of
longitudinal transactional datasets, observation of actual contracts, and ability to focus on
causality makes a significant contribution to the literature.
Opportunities for future research are ample. A significant number of studies have
focused on the effectiveness of governance mechanisms; however, we are unaware of a
comprehensive review of all of the different governance mechanisms observed in practice, their
characteristics and interactions with one another, and how they affect supplier performance and
opportunism. There are few studies that consider the legal enforceability of the business
environment, and fewer still that consider how mechanisms perform as the supplier’s payoff to
act opportunistically changes. Finally, few studies focus on causality by using longitudinal data.
Research that addresses these issues would be valuable to our understanding of buyer-supplier
interactions.
106
REFERENCES
Achrol, R.S., Gundlach, G.T. 1999. Legal and Social Safeguards Against Opportunism in
Exchange. Journal of Retailing 75(1) 107-124.
Adelman, D., Mersereau, A.J. 2013. Dynamic Capacity Allocation to Customers Who
Remember Past Service. Management Science 59(3) 592-612.
Anderson, S.W., Dekker, H.C. 2005. Management Control for Market Transactions: The
Relation between Transaction Characteristics, Incomplete Contract Design, and Subsequent