- Email: [email protected]
The impact of bullwhip on supply chains: Performance pathways, control mechanisms, and managerial levers
Accepted Manuscript Title: The Impact of Bullwhip on Supply Chains: Performance Pathways, Control Mechanisms, and Managerial Levers Author: Alan W. Mackelprang Manoj K. Malhotra PII: DOI: Reference:
S0272-6963(15)00017-0 http://dx.doi.org/doi:10.1016/j.jom.2015.02.003 OPEMAN 899
To appear in:
OPEMAN
Received date: Revised date: Accepted date:
17-2-2014 9-1-2015 25-2-2015
Please cite this article as: Mackelprang, A.W., Malhotra, M.K.,The Impact of Bullwhip on Supply Chains: Performance Pathways, Control Mechanisms, and Managerial Levers, Journal of Operations Management (2015), http://dx.doi.org/10.1016/j.jom.2015.02.003 This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and review of the resulting proof before it is published in its final form. Please note that during the production process errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.
The Impact of Bullwhip on Supply Chains: Performance Pathways, Control
cr
ip t
Mechanisms, and Managerial Levers
M
an
us
Alan W. Mackelprang * Assistant Professor of Operations Management College of Business Administration Georgia Southern University P.O. Box 8151 Statesboro, GA 30460-8151 [email protected]
Ac ce p
te
d
Manoj K. Malhotra Jeff B. Bates Professor and Chair Department of Management Science Moore School of Business University of South Carolina Columbia, SC 29208 Phone: (803) 777-2712; Fax: (803) 777-6876 [email protected]
March 3, 2015
* Corresponding Author
1 Page 1 of 54
The Impact of Bullwhip on Supply Chains: Performance Pathways, Control Mechanisms, and Managerial Levers
ip t
Abstract Even though few empirical studies have tried to actually explicate the relationship between the
cr
bullwhip effect and performance of the supplier firm, there exists a common perception for over
us
30 years among both practitioners and academics that the bullwhip effect naturally results in decreased firm profitability. Anecdotal evidence further suggests that this decline in profitability
an
arises from a decline in operational performance. However, the results of our study, which empirically examines the bullwhip effect across supply chain partners through an analysis of 383
M
actual customer base-supplier dyads, challenges this commonly held position by suggesting that
d
while traditional bullwhip often yields reduced ROA, it ultimately has no relationship with the
te
firm’s operating margin. Additionally, our results also call into question whether or not production coordination between customers and suppliers can minimize the need for inventory
Ac ce p
and capacity buffers, which are the two commonly utilized methods for battling the bullwhip effect. Thus the relationship between bullwhip and firm performance is far more nuanced and complicated than previously believed. We also show how the managerial bullwhip levers of coordinating production across supply chain partners, or deploying inventory and capacity buffer control mechanisms, can help maximize a firm’s performance along different dimensions. Keywords: Bullwhip Effect, Supplier Performance, Empirical, and Supply Chain
2 Page 2 of 54
The Impact of Bullwhip on Supply Chains: Performance Pathways, Control
ip t
Mechanisms, and Managerial Levers
1. Introduction
cr
It is commonly perceived that bullwhip reduces a firm’s performance by wreaking
us
operational havoc. We challenge this conventional wisdom by suggesting that while traditional bullwhip generally does result in lower return on assets (ROA), it has no relationship with the
an
firm’s operating margin, a primary indicator of operational performance. Such a finding suggests that how bullwhip actually impacts ROA is not as straightforward as previously thought.
M
This study dives deeply into the bullwhip-performance relationship by proposing and
d
evaluating a new dimension of bullwhip, which captures the acceleration (or deceleration) of
te
bullwhip between supply chain partners. Such an evaluation allows us to gain performance insights into not only the level of bullwhip a firm faces, but also the extent to which an
Ac ce p
acceleration (or deceleration) of the bullwhip impacts that firm’s performance. Results from this analysis will also challenge the conventional wisdom that customers and suppliers in a supply chain that synchronize/coordinate their production can minimize the need for inventory and capacity buffers that represent traditional mechanisms utilized to combat bullwhip. As such, we will show that with respect to the bullwhip effect, leaning too heavily upon what many would consider to be “already known” or “well understood” information may not always yield contextually accurate insights. At its most fundamental level, the bullwhip effect increasingly distorts the pattern of actual end customer demand to upstream supply chain partners (Lee, Padmanabhan and Whang,
3 Page 3 of 54
1997a, 1997b; Zhang and Burke, 2011), who are then often forced into “boom or bust” production cycles. Such extreme production cycles resulting from bullwhip have been shown to significantly increase supply chain costs and lower performance (Sterman, 1989). Likewise, the
ip t
Economic Theory of Production Smoothing (ETPS) suggests that firms have an economic incentive to avoid such extreme production cycles and instead should smooth production (Holt et
cr
al., 1960; Blanchard, 1983; Blinder 1986; Miron and Zeldes, 1988). Numerous researchers over
us
several decades have highlighted a number of countermeasures for combatting the bullwhip. Despite logical evidence that bullwhip behavior should be avoided and recommendations
an
on how to minimize it, there remains significant evidence that bullwhip still continues to occur in about 67% of firms (Bray and Mendelson, 2012; Shan et al., 2013). If firms are incentivized to
M
smooth production to avoid the “boom and bust” production cycles yet most do not do so, it
d
suggests from a rational perspective that these firms must have a larger incentive not to smooth
te
production, or from an irrational perspective, the majority of firms are subject to the same widespread behaviorally driven irrational actions. This apparent contradiction between theory
Ac ce p
and practice raises an interesting research question. Does the bullwhip effect actually result in reduced firm performance? If so, what are the specific pathways through which this decreased firm performance occurs? Finally, are there appropriate control mechanisms or managerial levers that can be applied to ameliorate the bullwhip effect?
2. Bullwhip Definitions and Assessment Before developing the theoretical model, it is important to first establish definitions and foundational concepts underlying this research. We subscribe to the traditional conceptual definition of bullwhip as described by Lee, Padmanabhan and Whang (1997a) as “the phenomenon where orders to the supplier tend to have larger variance than sales to the buyer 4 Page 4 of 54
(i.e., demand distortion), and the distortion propagates upstream in an amplified form (i.e., variance amplification).” Building upon this conceptual definition of bullwhip, it is clear that bullwhip
ip t
fundamentally consists of not only a level of demand distortion, but also the extent of variance amplification. In other words, while it is important to know if a firm is experiencing demand
cr
distortion, it is also important to know if such demand distortion is accelerating (or decelerating)
us
along the supply chain. For the purpose of this study, we describe these two aspects of bullwhip simply as first and second order bullwhip effects. First order bullwhip effect is the absolute level
an
of demand variation amplification for a given firm relative to the variance of actual demand for the furthest downstream partner (e.g. demand distortion); while second order bullwhip captures
M
the rate of change (e.g. variance amplification) in demand distortion (first order bullwhip)
d
between tiers in the supply chain. When evaluated together, first and second order bullwhip not
te
only capture the level of demand distortion at a given node in the supply chain, but also the extent to which demand distortion is accelerating (or decelerating) between tiers in the supply
Ac ce p
chain. Thus, it is possible to capture all possible bullwhip supply chain configurations by simultaneously utilizing these two aspects of bullwhip. We will first provide the mathematical equations we utilize to assess first and second order bullwhip effects, then we will provide a numerical example to further illustrate these calculations.
First order bullwhip for supplier s in year y is calculated as shown in Equation 1,
where CV (Psy) is the coefficient of variation of production of supplier firm s in the supply chain across the four quarters in year y, and CV (Dcy) is the coefficient of variation of demand in the supplier’s customer c across the four quarters in year y. (1) 5 Page 5 of 54
It is important to point out that in line with extant literature (Bray and Mendelson, 2012; Shan et al. 2013), both demand and production are calculated at the quarterly level. As configured in Equation 1, the first order bullwhip measure represents the percentage change in
ip t
upstream production variation relative to downstream demand variation. As such, positive values indicate an increase in the absolute level of demand variation or amplification, while a
cr
negative number indicates a decrease in variation with respect to downstream demand variation
us
or an absolute dampening. Given that first order bullwhip is consistent with how bullwhip has been examined in extant literature (Cachon et al., 2007; Bray and Mendelson, 2012), we will
an
interchangeably use the terms bullwhip and first order bullwhip.
Whereas first order bullwhip captures the production variability of the supplier with
M
respect to end customer demand variability, second order bullwhip captures the supplier’s
d
production variability as compared to the production variability of their customer. By doing so,
te
it is possible to determine how the firm is behaving on both an absolute (first order) and relative (second order) basis regarding production variability. Second order bullwhip for supplier s and
Ac ce p
customer c in year y is calculated as shown in Equation 2, where CV (Psy) is the coefficient of variation of production of supplier firm s in the supply chain across the four quarters in year y, CV (Dcy) is the coefficient of variation of demand of the supplier’s customer c across the four quarters in year y, and CV (Pcy) is the coefficient of variation in production for customer c across the four quarters in year y.
(2) When this variable is positive, the supplier’s first order bullwhip is larger than their customer’s first order bullwhip, thus indicating that the supplier is accelerating variance
6 Page 6 of 54
amplification relative to their customer. When this measure is negative, the supplier is decelerating variance amplification with respect to their customer. Equations 1 and 2 make it possible to distinguish all combinations of first and second order bullwhip effects, such as firms
ip t
that may be amplifying (first order) but at a decelerating rate (second order) and vice versa.
We use a numerical example with contrived data in Table 1 to illustrate the calculation of
cr
first and second order bullwhip using Equations 1 and 2. The first order bullwhip for the
us
customer is 0.4 as shown in Table 1, or in other words the customer’s production variance is 40% larger than the customer’s demand variance. First order bullwhip for the supplier is 0.6, or
an
in other words their production variance is 60% larger than the customer’s demand variance. Second order bullwhip is 0.50, indicating that the supplier is amplifying at a 50% higher rate
M
than which the customer is amplifying, thus confirming that the bullwhip is accelerating between
d
the customer and the supplier.
te
---------Insert Table 1 about here-----3. Recent Perspectives on Bullwhip
Ac ce p
In this section we treat the most recent developments related to the bullwhip. In addition, we refer interested readers to Giard and Sali (2013) for a more comprehensive literature review on research related to the bullwhip effect. While much of the historical bullwhip research has been centered on the causes and remedies of the bullwhip effect, a new research stream is emerging that seeks to empirically verify the existence of the bullwhip effect. For example, Cachon et al. (2007) using industry level data were unable to find evidence of a traditional bullwhip effect, but rather reported that industries within the wholesale sector had higher demand amplification levels than either retail or manufacturing industries. This result is counter to previous industry level studies that found amplifying in retail industries, but not in wholesale
7 Page 7 of 54
industries (Baganha and Cohen 1998). These disparities in recent bullwhip empirical findings (e.g. Baganha and Cohen, 1998) prompted Chen and Lee (2012) to differentiate between how bullwhip was originally measured by Lee et al. (1997a) from how other empirical studies have
ip t
largely measured bullwhip. They suggest that these prior empirical studies have been examining “the existence or magnitude of ‘aggregated’ bullwhip in material flow” because these measures
cr
capture material flow, aggregated products, and aggregated time periods. Furthermore, Chen and
us
Lee (2012) show mathematically that as product or time aggregation increases, the aggregated bullwhip ratio converges to one, meaning that production variation is equal to demand variation
an
and that neither amplifying nor dampening is occurring at the aggregate level. They further suggest that the conflicting results of both Baganha and Cohen (1998) and Cachon et al. (2007)
M
may be a result of the use of excessively aggregated data across both products (industry level)
d
and time (monthly). In an effort to enhance “cross-sectional granularity” rather than temporal
te
granularity, Bray and Mendelson (2012) and Shan et al. (2013) recently attempted to overcome some of the data related shortcomings found in Cachon et al. (2007) by analyzing quarterly-firm
Ac ce p
level data, which indicated that approximately two-thirds of firms experience a bullwhip effect. However, the data utilized by Cachon et al. (2007), Bray and Mendelson (2012), and Shan et al. (2013), does not account for the actual relationships between supply chain partners. As such, it is difficult to draw conclusions from these studies about the extent to which amplification flows from customers to suppliers because of the nature of the data utilized. An issue that emerges clearly from the extant literature is that while the causes of the bullwhip effect have been well covered in the seminal works of Forrester (1958, 1961), Sterman (1989), and Lee et al. (1997a) among others, empirically verified performance implications of the bullwhip effect have been limited. In terms of performance implications, Behavioral Beer
8 Page 8 of 54
Game experimental studies have found in general that as the amplification of bullwhip increases, total costs also increase dramatically (Sterman, 1989). Metters (1997) in his simulation study suggests that if the bullwhip effect could be eliminated entirely, firm profits would increase
ip t
between 15% and 30%. What is missing from this discourse however is a large-scale empirical validation and theorizing of how bullwhip affects firm performance in an actual supply chain
us
cr
relationship, which is what we focus on next as we develop our related hypotheses. 4. Development of Hypotheses
an
We are interested in examining the effects of both first order bullwhip and second order bullwhip on a supplying firm’s performance in a supply chain context. We draw upon the Economic
M
Theory of Production Smoothing (ETPS) and Organizational Information Processing Theory (OIPT) in developing our hypotheses. Since these two bullwhip effects are related but
d
conceptually separate from one another, we will treat each individually and accordingly develop
te
the related hypotheses in the following two sub-sections.
Ac ce p
4.1 First Order Bullwhip Effect and Performance Firms that experience bullwhip are commonly subjected to “boom and bust” production cycles with oscillating periods where demand can greatly exceed capacity and vice versa. Generally, such cycles produce high levels of inventory followed by increasing stock outs and backorders coupled with inefficient and expensive over-production, followed by sub-optimal and low utilization rates (Shan et al., 2013). While it is widely accepted that firms which are subjected to extreme bullwhip conditions generate lower performance, the theoretical underpinnings to explicate such a relationship have remained far less clear in extant literature. Since the 1960’s, economists have put forth ETPS which contends that firms have an economic incentive to smooth production when facing variable demand (Holt et al. 1960; 9 Page 9 of 54
Blanchard, 1983; Blinder 1986; Miron and Zeldes, 1988). Three primary assumptions underlie ETPS (1) demand is variable, (2) production costs are convex and (3) inventories can exist. The intuition behind ETPS is that profit maximizing firms should seek to minimize the joint costs of
ip t
production and holding inventory. Therefore, when production costs are convex, the firm has an incentive to produce at a stable rate, thereby building inventory when demand is low and
cr
satisfying high levels of demand through a combination of both production and inventory. If the
us
firm’s marginal cost of production is linear, there is no incentive to smooth production or to accumulate inventory. Instead demand should be met by adjusting production.
an
Viewing the bullwhip-performance relationship through the lens of ETPS helps to formalize the underlying behavior that has been widely visible in firms that experience bullwhip.
M
Specifically, ETPS suggests that profit maximizing behavior dictates that the least costly and
d
therefore the most profitable method (inventory and/or capacity) should be utilized to satisfy
te
demand. The condition mentioned above where marginal production costs are linear suggests that no penalties are incurred when manufacturing flexibility is invoked. However, it is widely
Ac ce p
documented that exceeding the limits of efficient production by invoking short term capacity options such as over-time and sub-contracting generally increases the marginal cost of production (Koste and Malhotra, 1999). Facing flexibility induced convex production costs, ETPS suggests that the firm has increasingly greater incentives to smooth production via relatively less costly inventory. Such a linkage implies that firms exhibiting increasingly high levels of first order bullwhip are not able to adequately smooth production, though they are incentivized to do so. Firms in this situation are increasingly likely to yield sub-optimal performance as actual behavior (utilizing increasingly more expensive flexibility options) diverges from profit maximizing behavior. Therefore, we suggest that as first order bullwhip
10 Page 10 of 54
increases, and firms diverge from the profit maximizing behavior suggested by ETPS, firm performance will decline. The benefits of stable production are well documented. For example, the concept of
ip t
level-loading or heijunka within Lean/Just-in-Time manufacturing has been widely adopted and shown to improve performance (Liker and Morgan, 2006). Likewise, stable production situations
However, it is possible for firms to diverge from the profit
us
(Liker and Morgan, 2006).
cr
enable process standardization and learning effects to take hold, further enhancing performance
maximizing behavior of ETPS by overly smoothing production. Such instances occur when the
an
firm continues to smooth production even though the relative cost of invoking a flexible response is less than smoothing production via inventory. Greater levels of inventory allow the firm to
However, as the level of inventory increases so do total
d
dampening the bullwhip effect.
M
absorb increasing levels of demand variability, while keeping production stable; thereby
te
inventory-holding costs. ETPS suggests that at some point, the firm’s marginal benefit of holding inventory will be surpassed by the marginal cost of doing so (Arrow, Karlin and Scarf,
Ac ce p
1958). If the firm continues to utilize inventory when the relative cost to do so exceeds the relative cost of invoking flexibility, ETPS suggests that firm performance is likely to be suboptimal as the firm diverges from profit maximizing behavior. Thus while on one hand, there are predicted negative performance implications associated with excessive amplification, there are also predicted negative performance repercussions associated with extreme smoothing. This reasoning suggests that in a quadratic fashion, extreme amplifying or dampening will result in lower levels of firm performance relative to when the firm is effectively neither dampening nor amplifying. Therefore based on ETPS, we argue that firm performance will have a negative quadratic relationship with the first
11 Page 11 of 54
order bullwhip, such that the firm performance will be highest when the first order bullwhip approaches zero (when there is essentially no bullwhip effect). H1: Firm performance improves as the first order bullwhip converges to zero
ip t
4.2 Second Order Bullwhip and Performance
cr
Whereas first order bullwhip captures the behavior of the firm with respect to end customer demand, second order bullwhip captures such behavior relative to the firm’s immediate
us
downstream partner (its own customer) in the supply chain. As such, second order bullwhip evaluates the extent to which the supplying firm is accelerating or decelerating amplification
an
relative to its customer. When second order bullwhip is zero, the supplying and customer firms
M
have identical first order bullwhip effects, suggesting that the supplier and customer are producing in tandem and thereby generating identical levels of production variability. Firms that
d
are producing in “tandem” are emblematic of coordinating firms.
te
The issue of sharing information between supply chain partners has been a consistent and major theme within bullwhip research (Lee, So and Tang, 2000). The general consensus is that
Ac ce p
sharing information across supply chain partners can reduce the distortion of demand information and also reduce the bullwhip.
Organizational Information Processing Theory
(OIPT) (Galbraith, 1973; Daft and Lengel, 1986) is especially informative about how the flow of information across supply chain partners can impact performance via the bullwhip. Within OIPT, tactics for managing uncertainties depend upon the magnitude of the uncertainty. OIPT suggests that low levels of uncertainty can be managed through information processing capabilities which are typically routine based mechanisms (Galbraith, 1974; Tushman and Nadler, 1978). However, in high uncertainty environments, the firm must increase their own information processing capabilities or employ slack resources as a way to keep the level of 12 Page 12 of 54
uncertainty within the current scope of the firm’s information processing capabilities (Swink and Schoenherr, 2014; Galbraith, 1974). In addition to uncertainty, OIPT indicates that equivocality can be another driver of needed information processing capabilities. Equivocality occurs when a
ip t
common interpretation does not emerge as a result of processing the gathered information. For example, during the 2008 global recession, firms were often unable to reach a consensus with
cr
respect to market demand, and therefore were also challenged in making strategic production
us
decisions (Bloom, 2013). Whereas uncertainty typically results from a lack of information
Lengel, 1986; Goodhue, Wybo and Kirsch, 1992).
an
quantity, equivocality is seen as resulting from low levels of information richness (Daft and
Within the context of second order bullwhip, the extent to which information is not
M
passed from customer to supplier can be viewed from the lens of OIPT as a source of uncertainty
d
for the supplier. Additionally, equivocality may also be expected when the level of coordination
te
between supply chain partners is low as it is unlikely that the supplying firm will have a high degree of confidence that their chosen level of production is what it would have been with
Ac ce p
perfect information. Slack resources as well as buffers can be utilized to combat the impacts of both uncertainty and equivocality by reducing the need for information processing (Swink and Schoenherr, 2014). However, slack resources can be costly (Modi and Mishra, 2011). Coordinating firms support processes that collect and share more accurate supply and demand information (Stank et al., 1999). Therefore, within the OIPT framework, coordinating firms are more likely to obtain not only a greater quantity of accurate information thus reducing uncertainty, but also the obtained information is more likely to have a higher level of detail and/or richness, thus also suggesting a lower level of equivocality (Daft and Lengel, 1986). By reducing both uncertainty and equivocality, the need to increase information processing
13 Page 13 of 54
capabilities or to utilize capacity and inventory buffers also decreases as do their associated costs. This situation is similar to the concept of “virtual integration” whereby the need to hedge against demand uncertainty is reduced as greater levels of in-depth high quality information is
ip t
shared between supply chain partners and information is effectively substituted for buffering mechanisms (Wang, Tai and Wei, 2006).
cr
Therefore, anchored by OIPT, we suggest that supplier performance should be highest
us
when customer and supplier firms are producing in tandem thereby reducing the information processing requirements of the supplier and also reducing the need for additional information
an
processing capabilities or capacity and inventory buffers. This argument suggests a quadratic relationship between second order bullwhip and supplier performance, because as firms diverge
M
from coordinated production (second order diverges from zero), performance should decline as
te
arguments, we posit hypothesis 2.
d
the firm’s uncertainty and equivocality increases. Based upon the preceding OIPT based
Ac ce p
H2: Firm performance improves as the second order bullwhip converges to zero 5. Data and Measurement
One of the major weaknesses of prior bullwhip empirical research has been an inability to simultaneously evaluate a number of different supply chains.
Prior empirical research has
generally been conducted with data being at the industry level (e.g. Cachon et al., 2007) or at the firm level (Bray and Mendelson, 2012), neither of which explicitly captures interactions across specific supply chain dyads. This issue is an important one because at the heart of the bullwhip effect is the reaction of a supplier to the incoming variability signals from their customers, which results in the transmission of the bullwhip effect from the customer to the supplier.
14 Page 14 of 54
In order to overcome this limitation and evaluate firm-level supply chains, we create and study matched dyadic relationships between buyers and suppliers by utilizing publically available secondary data between the years 1990 to 2008. Since 1976, publically traded firms in
FAS 131 (1997) and FAS 14 (1976).
ip t
the U.S. have been required to report their major customers under the accounting regulations Major customers are defined by these accounting As a result, researchers
cr
regulations as customers buying at least 10% of the selling firm’s sales.
us
are starting to utilize the customer segment data located within the CompuStat database to examine supply chain dyads (see for example Hertzel et al. 2008). Using this data, it is possible
an
to identify matched buyer-supplier dyads. However, the form in which the data is reported requires that the data must first be systematically cleaned and validated before any dyads can be
M
identified. It entails significant effort, which is not unexpected in studies focused at empirically
d
isolating phenomenon of interest at the firm level in supply chains.
te
The data in the Compustat segment files related to major customer relationships is selfreported. Customers are commonly listed as government agencies, foreign corporations, or
Ac ce p
privately held corporations. In many cases, even though the major customer is identified, the supplier indicates the sales to that customer as $0. The inability to identify the magnitude of the relationship makes those particular observations unusable.
Likewise, firms commonly disguise
the identities of their large customers, so instead of identifying customers by name, they are simply identified as “Customer A” or “Top 5 customers.” Those relationships are unusable as well, and must be systematically discarded. Because there is no standard method for firms to report the names of their customers, the same customer can be identified with slightly different abbreviations or spellings. For instance, Wal-Mart Stores Inc. might be identified as Wal-Mart, Wal-Mart Corp, Walmart, or WMT in the
15 Page 15 of 54
database. Therefore, unless these discrepancies are resolved, much of the customer relationship data is unusable.
Hertzel et al. (2008) utilized a text-matching program which helped them to
correctly consolidate the customer abbreviations provided by suppliers. While such programs
ip t
can aid researchers in narrowing down the potential matches, the number and complexity of the customer abbreviations also make it likely that the text matching programs would likely discard a
cr
number of potentially usable relationships. Therefore, in order to preserve as many relationships
data by hand instead of using text-matching programs.
us
and real world supply chain dyads as possible, we opted for the tedious process of cleaning the
an
Further refining is necessary because of the existence of multiple types of buyer-supplier relationships. For instance, the supplier may be providing equipment, raw material, services
M
(consulting, financial etc), supporting goods (such as software), logistics support, contract
d
manufacturing, or energy. Furthermore, the buying firms commonly hold distinct roles in the
te
supply chain, such as retailer, wholesaler, manufacturer, service provider etc. While previous research has utilized all types of supply chains (Lanier et al. 2010), it is desirable from a control
Ac ce p
perspective that the data is somewhat homogeneous in nature. Hence we opted to include only manufacturing firms. Finally, as per the recommendations of Bray and Mendelson (2012), we limited our sample to only those firms that had positive revenues and positive inventory value.
5.1 Bullwhip Transmission
Once buyer-suppler dyads are identified, in order to accurately evaluate bullwhip across these firms, it is necessary to ensure that (1) any production variance occurring at the supplier firm is in response to their actual customer’s demand variance and (2) customer demand and production variance is with respect to the inputs that the supplier provides. While other dyadic research (Lanier et al. 2010) has suggested that only the largest or strongest buyer-supplier relationship 16 Page 16 of 54
should be evaluated, this recommendation becomes tenuous since it is unlikely that a supplier’s largest customer possesses bullwhip characteristics that are representative of the supplier’s entire customer base. Furthermore, unless a firm’s largest customer is essentially a single customer
ip t
(representing 100% of the supplier’s sales), and that same supplier provides 100% of the customer input materials; it becomes difficult to suggest that full transmission occurs whereby
cr
the supplier’s amplification is a result of or in response to this single customer’s demand
us
amplification characteristics. Thus unlike the Beer game, in real-world supply chains there is no assumed one-to-one supply chain relationship guaranteeing full bullwhip transfer between a
an
customer and a supplier. While it is important to capture the inter-firm effects, it is equally important to ensure that we are also capturing the transfer of bullwhip between customer and
M
supplier. So in order to overcome these challenges, we have taken several safeguards that make
d
it more likely that true bullwhip transmission is indeed being captured within our dataset.
te
First, we limit our sample firms to only manufacturing (SIC 20-39) suppliers and customers and exclude retailers because in the context of firm level bullwhip, what is unique
Ac ce p
about retailers is that their demand variation could be a result of any of the items sold. For example, Wal-Mart’s demand variation could be a result of selling milk, tires or shirts. Clearly, a single supplier would not supply all of these items. If we limit our sample to manufacturing firms only, it is much more likely that the demand variance at the customer is related to a product for which our identified supplier provides an input. Second, instead of evaluating only single buyer-supplier dyads, which in most cases would only capture a small portion of the total sales for the supplying firm, we choose to aggregate across our identified buyer-supplier dyads, generating customer-base dyads for each supplier. Doing so not only accurately reflects the bullwhip behavior of the supplier, but also captures the inter-firm dynamics. In order to balance
17 Page 17 of 54
the contrasting desires of possessing significant statistical power while also capturing the highest proportion of sales accounted for in the identified customer base, we choose to only evaluate those firms whose identified customer-base accounted for at least 50% of the supplying firm’s
ip t
revenues. Our represented customer bases accounted for 69.1% of their supplier’s total revenues on an average, and therefore captured the dominant behavior of the firm’s entire customer base.
cr
Finally, we limited our sample to “major” suppliers only. Hertzel et al. (2008) define major
us
suppliers as providing at least 1% of their customer’s cost of goods sold (COGS). The average supplier accounted for 6.2% of their customer-base’s COGS in our sample, suggesting that any
an
customer demand/production variance is related to products for which this major supplier has provided inputs. Cumulatively these safeguards allow us to examine a real-world dataset that
M
reflects the behavior of actual supply chains.
d
5.2 Bullwhip Measurement
te
While we have already outlined our assessment of first and second order bullwhip effects in Equations 1 and 2, this section provides additional details as to how we applied these
Ac ce p
measurements to our study. A common measure used by researchers to operationalize the bullwhip effect is the difference between upstream production variance to downstream demand variance (Bray and Mendelson, 2012; Cachon et al. 2007). Because our dataset contains firms of significantly differing sizes, the difference of downstream and upstream variance could show amplification simply because of size differences. Therefore, in line with previous bullwhip research, we accommodate size differences among the firms in a given supply chain by evaluating the coefficient of variation rather than the variance itself (Fransoo and Wouters, 2000; Dejonckheere et al., 2004). Additionally, since we utilize publically available financial data, we use a proxy for firm production by utilizing the method of Cachon et al. (2007) outlined in 18 Page 18 of 54
Equation 3. Production for firm f in quarter t is the cost of goods sold plus any changes in the inventory level of the firm quarter-over-quarter, where Cft and Ift are the cost of goods sold and inventory respectively for firm f in quarter t. (3)
ip t
Pft = Cft + (Ift – Ift -1)
cr
The sales and demand figures are adjusted for profit margin in order to more accurately reflect actual production and demand numbers (Cachon et al., 2007). This measure is intended to
us
broadly capture production in dollars rather than in units, while cost of goods sold is intended to
an
capture demand. If firms draw from inventory, then production will be lower than demand, while if firms are building inventory, production will exceed demand.
M
Since the production and demand data are quarterly snapshots, the coefficient of variation is calculated across all 4 quarters in the calendar year. In line with Cachon et al. (2007), we do
d
not de-seasonalize our data because firms must meet actual demand and not the de-seasonalized
te
demand. Given the Chen and Lee (2012) results, which suggest that time aggregation hides
Ac ce p
variance amplification; this approach represents a more conservative evaluation of amplification. Finally, since we aggregate across customer-supplier dyads to generate customer basesupplier dyads, it is necessary derive customer base measures that can be evaluated in place of the single customer measures shown in Equations 1 and 2. To do so, we follow a weighted average approach to compute customer base measures as shown in Equation set 4, where the weighted average customer base measure c for supplier-year observation s; number of identified firms in the customer base for supplier i; customer base firm j;
is is the
is the weight given to
is the percentage of total sales purchased from supplier i by customer
19 Page 19 of 54
base firm j; and
is the variable of interest for customer base firm j. All variables are from
the same time period.
ip t
(4)
us
cr
given that
an
In line with other empirical bullwhip studies such as Bray and Mendelson (2012), we assume that the amplification identified at the customer base through utilizing firm level data is
M
not systematically different than what would have occurred at the product level. The final sample thus obtained consists of 383 supplier-customer base dyadic observations from 147
te
supplier industries.
d
unique supplier firms. Table 2 provides details of the sample at it relates to different years and
-----Insert Table 2 about here-----
Ac ce p
5.3. Performance Variables
This study evaluates the impact of first and second order bullwhip effects on supplier performance. Return on Assets or ROA (Net Income/Total Assets) is a traditional measure of a firm’s financial performance. Because ROA is an aggregated measure, we choose to further “decompose” ROA into the profit and asset side of the ratio (Swink and Jacobs, 2012) in order to gain additional insights into the performance pathways by which bullwhip may ultimately impact ROA. On the income side, we evaluate Operating Margin ((Sales-COGS)/Sales) (Swink and Jacobs, 2012) and SG&A Expense (SGA Costs/Sales) (Dehning, Richardson, and Zmud, 2007). On the asset side of ROA, we evaluate Fixed Asset Turnover (Sales/Average Fixed Asset Value) (Tang and Liou, 2010) and Inventory Turnover (Sales/Average Inventory Value) (Swink and 20 Page 20 of 54
Jacobs, 2012). As per the recommendations of Dehning, Richardson, and Zmud (2007), all the dependent performance variables were Winsorized at the 99% and 1% level. 5.4. Control Variables
ip t
Given that we are evaluating a number of different types of performance outcomes, a number of common performance-related control variables were included in the analysis to ensure that any
cr
identified relationships were not spurious (Dehning, Richardson and Zmud, 2007; Hendricks,
us
Singhal and Zhang, 2009). Related to the supplying firm, these control variables include Supplier industry (2-digit SIC), Firm Size (Natural Log of Total Assets) and Market Share (Firm
an
Sales/Total Industry Sales 4-digit SIC). Customer-base controls include Firm Size and Market Share. Both customer-base control variables were calculated utilizing the weighted average
M
methodology provided in Equation 4. We elected to control for both firm size and market share
d
in order to fully account for any power dynamics in the customer-supplier relationship that could
te
potentially influence the performance outcomes of the supplier. Finally, time was controlled for by utilizing a continuous variable, with zero equating to the first year of the data. Correlations
Ac ce p
and descriptive statistics for all the performance and control variables are provided in Table 3. Additionally, a summary of how our empirical sample divides into the four quadrants related to the first and second order bullwhip effects is shown in Table 4. Approximately 62.4% of the firms in our sample show an amplifying first order bullwhip effect, and only 60.3% of the firms show an accelerating second order bullwhip.
These numbers confirm that the traditional
bullwhip pattern where demand variance amplification progressively increases as it travels upstream in the supply chain is common, but certainly not universal, and also in line with the findings of other researchers as well. Equally interesting is the large proportion of firms that exhibit dampening and decelerating patterns.
21 Page 21 of 54
-----Insert Tables 3 and 4 about here-----
6. Analysis
ip t
The primary purpose of this study is to evaluate the nature of the relationships between first and second order bullwhip effects and supplier performance. To accomplish this, we utilized OLS
cr
regression techniques coupled with Seemingly Unrelated Estimation (SUE).
Given that
us
quadratic effects have been hypothesized for both first and second order bullwhip, these are tested by evaluating Equation 5, where Z is the performance variable, X is the first order
an
bullwhip effect, and Y is the second order bullwhip effect.
(5)
M
Z= β0 + β1X + β2Y + β3X2 + β4Y2 + ε
Equation 5 is specified such that that the first and second order bullwhip effects are
d
treated as additive. However, it is possible that the first and second order effects actually have a
te
multiplicative relationship with each other. Given that we have no a priori indication as to the
Ac ce p
nature of the joint relationship between first and second order bullwhip, we propose to further explore this possibility with the method outlined by Cortina (1994) by examining Equation 6, where Z is the performance variable, X is the first order bullwhip effect, and Y is the second order bullwhip effect.
Z= β0 + β1X + β2Y + β3X2 + β4Y2 + β5XY +ε
(6)
Even though it is possible that there is a non-linear interaction between first and second order bullwhip, Cortina (1994) suggests that evaluating linear interactions can provide evidence of a non-additive relationship. Furthermore, by only evaluating a linear interaction, we are able to avoid severe multicollinearity problems associated with testing non-linear interactions of nonlinear effects (e.g. quadratic-quadratic interaction). We will conclude that the first and second 22 Page 22 of 54
order bullwhip relationship is multiplicative rather than additive if the interaction term is significant, and if the Equation 6 model fits the data better than the Equation 5 model. Our study evaluates five interrelated aspects of performance. As such, it is necessary to
ip t
simultaneously estimate all models via SUE, which controls for any covariance that may exist between regressions of bullwhip-performance outcomes. Additionally, since it is possible that
cr
the same supplying firm is in the data multiple times across the sample years, our data cannot be
us
considered independent. Therefore, we utilize two-way cluster robust errors (Peterson 2009, Cameron et al. 2011; Thompson, 2011) which cluster around the both the supplying firm as well
an
as time. This approach allows the errors to correlate within a supplying firm and year, but not across supplying firms or years. As such the independence assumption underlying OLS
M
regression is satisfied.
d
Equation 7 shows the procedure developed by Thompson (2011) that we used to calculate ,
and
are the estimated
te
our two-way clustered variance estimator, where
Ac ce p
variances of supplier firm, estimated variances of time, and the common heteroskedasticityrobust OLS variance (White, 1980) respectively. (7)
Since each customer base is idiosyncratic with respect to the specific supplying firm, we assume that there is no bias across customer base observations. We therefore do not control for specific customer firms or industries. Multicollinearity was not deemed to be a significant problem because all Variance Inflation Factors (VIFs) were less than 7.5, well below the common threshold of 10 (Visnjic and Van Looy, 2013).
No significant correlations were
identified (α=0.1) between independent variables and residuals from each equation, thereby
23 Page 23 of 54
suggesting that endogeneity is not a concern in this study. The first and the second-order bullwhip variables were standardized to aid in the interpretation of results.
ip t
7.0 Results The reported results and figures are from the best fitting and most parsimonious model between
cr
analyzing Equations 5 and 6. Table 5 details the results of our analysis with respect to ROA. First order bullwhip has a significant (α<0.05) negative linear relationship with a positive
us
quadratic trend with ROA (-0.08X +0.06X2). While H1 suggests that performance would peak
an
as first order converges to zero, as shown in Figures 1a and 1b, ROA is actually near its lowest level when first order bullwhip is near zero. Additionally, the highest level of ROA occurs when
M
first order bullwhip is severely dampened. Therefore, H1 is not supported with respect to ROA performance.
d
Examining H2 shows that the behavior of first and second order bullwhip effects is
te
almost opposite regarding ROA. Second order bullwhip has a significant (α<0.05) negative
Ac ce p
quadratic relationship with ROA (-0.06X2). As such, we fully support H2 that peak ROA performance occurs as second order bullwhip converges to zero.
Finally, while not
hypothesized, we explored the nature of the joint relationship between first and second order bullwhip and ROA performance. We do find evidence of a significant positive multiplicative relationship, suggesting synergistic effects between first and second order bullwhip effects relating to ROA performance. Taken together, the combined interaction and quadratic effects of first and second order bullwhip generate a “saddle” shaped relationship, where the highest ROA occurs when the first order bullwhip is greatly smoothed or dampened and second order bullwhip nears zero. The lowest ROA occurs in the opposite condition where the first order is greatly amplifying and the second order is divergent from zero. 24 Page 24 of 54
-----Insert Table 5 and Figures 1a and 1b about here-----
7.1 Profit Performance
ip t
Results related to Operating Margin and SGA Expenses are shown in Table 6 and Figures 2 and
cr
3. These results further help to explicate how first and second order bullwhip impacts ROA. While the first order bullwhip is shown to have no significant relationship with Operating
us
Margin, it does have a significant (α<0.05) positive linear relationship with a negative quadratic trend (0.07X-0.06X2) with SGA Expense. Given that H1 suggests peak performance will occur
an
as the first order bullwhip converges to zero, H1 is not supported for either Operating Margin or
M
SGA Expense.
The second order bullwhip is shown to have a significant positive linear relationship with
d
a negative quadratic trend (0.12X-0.10X2) with Operating Margin and a negative quadratic (-
te
0.03X2) with SGA Expense, thus supporting H2 for Operating Margin and failing to support H2 regarding SGA Expense. Exploring the nature of the joint relationship between the first and the
Ac ce p
second order bullwhip suggests that for both the Operating Margin and SGA Expense, the first and the second order bullwhip effects are additive rather than multiplicative. -----Insert Table 6 and Figures 2a, 2b, 3a, 3b about here----7.2 Asset Performance When evaluating the impact of the first order bullwhip on asset performance, the results shown in Table 7 indicate a significant (α<0.05) positive linear relationship with a negative quadratic trend (0.27X-0.19X2) for fixed asset turnover and a significant (α<0.05) negative linear relationship (-1.82X) for inventory turnover. Both results provide partial support of H1. These results are shown graphically in Figures 4 and 5. In contrast, second order bullwhip is shown to have a significant negative linear relationship with (-0.17x) with fixed asset turnover and no 25 Page 25 of 54
relationship with inventory turnover. As such, H2 is partially supported with respect to Fixed Asset Turnover and not supported with respect to Inventory Turnover. Finally, based on our
these aspects of asset performance is additive rather than multiplicative.
ip t
analysis, the nature of the joint relationship between the first and second order bullwhip and
cr
-----Insert Table 7 and Figures 4 and 5 about here----7.3 Robustness Checks
To ensure that our results were robust and not influenced by our chosen method and clustered
us
error structure, we also conducted the analysis utilizing a variety of differing methods and
an
clustering options. These included (1) double clustering on supplier and year, but not utilizing SUE, (2) single clustering on supplier only with SUE, and (3) single clustering on supplier In all cases, the pattern of significant results did not change. Therefore we
M
without SUE.
8.0 Discussion
d
conclude that our results are robust with respect to the chosen method of analysis.
te
The primary motivation for this study was to empirically investigate the bullwhip-performance
Ac ce p
relationship. By doing so we hoped to uncover evidence as to the various performance pathways whereby bullwhip ultimately impacts supplier performance. However, to fully evaluate bullwhip in a supply chain context, demand distortion as well as variance amplification should be examined.
Therefore, in addition to testing the traditional measure of bullwhip demand
distortion (first order bullwhip), we also proposed and tested a measure of variance amplification (second order bullwhip). Throughout this section, we discuss the key findings of this study related to (1) providing descriptive evidence with respect to the preponderance of first and second order bullwhip within a supply chain, (2) explicating potential performance pathways whereby bullwhip impacts performance and (3) identifying potential managerial levers and control mechanisms that may be utilized to influence the bullwhip-performance relationship. 26 Page 26 of 54
The primary insights and key findings stemming from this study are further discussed in the following sub-sections and also highlighted in Table 8. -----Insert Table 8 about here-----
ip t
8.1 How Common is Bullwhip?
The configuration of our sample suggests that what is viewed as traditional bullwhip behavior
cr
(increasingly amplifying demand distortion) is not a universal occurrence. We found that with
us
respect to first order bullwhip, 62.4% of our sample amplified while 37.6% dampened. Our results are consistent with the recent empirical work of Bray and Mendelson (2012) and Shan et
an
al. (2013) that approximately two-thirds of firms exhibit bullwhip. Thus including our study, three separate sets of researchers have triangulated and independently shown that 60% to 67% of
M
firms amplify. Additionally, within our sample, approximately 60.3% have an accelerating
d
second order bullwhip effect, while 39.7% have a decelerating second order effect. When
te
viewing the first and second order effects jointly, about half the sample (50.1%) falls into what is considered to be the traditional bullwhip condition (amplifying first order and accelerating
Ac ce p
second order), thus suggesting that while traditional bullwhip occurs commonly, it is clearly not a universal occurrence.
When evaluating the dispersion of observations across the four
quadrants in Table 4, it is clear that there is no universal or dominant strategy with respect to combinations of first and second order bullwhip. The implication is that firms may be actively managing variability in line with their chosen production strategy and specific incentives, without a prevailing smoothing or amplifying bias. 8.2 Bullwhip Performance Pathways A key contribution of this study is in evaluating the impact of bullwhip on not only ROA but also its key components. By doing so, we are able to identify more precisely the performance pathways bullwhip takes (income and/or assets ) to impact ROA. This study provides evidence 27 Page 27 of 54
that while both the first and second-order bullwhip effects ultimately impact supplier ROA, each does so through distinctly different pathways. Thus, while our results do support conventional wisdom that bullwhip can negatively impact supplier performance; we provide evidence that
ip t
suggests that the bullwhip-performance relationship is far more complicated than what is commonly perceived. The following sub-sections detail more specifically insights gleaned with
cr
respect to the differing performance pathways of both first and second order bullwhip.
us
8.2.1 Bullwhip-ROA Relationship
Holding second order bullwhip constant at zero, a dampened (<0) first order bullwhip yields
an
positive supplier ROA, while an amplified (>0) first order bullwhip generally yields negative ROA. Counter intuitively, these first order bullwhip results suggest that there is virtually no
M
ROA downside to dampening, even to extreme levels. The marginal costs of utilizing bullwhip
d
control mechanisms appear to be more than offset by performance benefits of smoothing
te
production. Additionally, counter to our proposed hypothesis, supplier ROA is near its lowest and most negative level when first order bullwhip is near zero. This interesting and also
Ac ce p
counterintuitive finding suggests, ceteris paribus, that it may not be in the best financial interests of upstream firms to be first order bullwhip neutral (i.e. first order bullwhip of zero). When examining second order bullwhip, holding first order constant at zero, we find almost the exact opposite relative relationship. Supplier ROA is highest, albeit negative, when second order is near zero. Thus we find evidence that with respect to ROA, suppliers benefit from coordinating with their immediate customers, but not with entities further downstream. In fact, we find that the highest ROA for a supplier occurs when there is “coordinated dampening” of first order bullwhip (first order <0 and second order =0). Another interesting finding is that the lowest ROA for a supplier occurs when first order bullwhip is extremely amplified and
28 Page 28 of 54
second order is extremely decelerated (first order >0 and second order <0). Such a bullwhip combination suggests that while the supplier is severely amplifying, they are nevertheless trying to “tame” the bullwhip passed to them by their customer by amplifying less than their customer
ip t
did. While an extreme amplification of first order bullwhip generally produces negative ROA, it is far worse for the supplier to do so by either accelerating or decelerating the first order
cr
bullwhip of their customer.
us
8.2.2. Bullwhip-Income Performance Relationship
Turning our attention to the income side of the ROA decomposition analysis, we find that
an
Operating Margin (OM) is only impacted by a large quadratic second order effect, while SGA expense is impacted by both first and second order quadratic effects. The finding that OM is not
bullwhip
primarily
impacts
performance
through
generating
operational
d
traditional
M
impacted by first order bullwhip is very surprising; given the conventional wisdom that
te
inefficiencies. Instead, our results suggest that supplier operating margin is directly influenced by accelerating or decelerating the bullwhip effect from their customers (second order), rather
Ac ce p
than through their own distortion of end customer demand (first order). It is important to note that Operating Margin, which is calculated from sales and COGS, only captures direct labor costs and material purchases related to what was produced in that period. It therefore reflects only the effects of overproduction, purchasing excessive amounts of new material, or using expensive short-term capacity options. These practices are only likely to occur as a result of responding to the behavior of their customer’s production rather than to further downstream demand signals. It is interesting to note that our results also suggest that firms have largely adapted their operations to their given level of first order bullwhip, such that any OM impacts are
29 Page 29 of 54
primarily due to the acceleration (deceleration) of bullwhip relative to their customers rather than due to the absolute level of the supplier’s first order bullwhip. Focusing on SGA Expenses, we show that both first and second order bullwhip produce
ip t
quadratic effects where the highest SGA Expenses occur as each measure approaches zero. When both the first and second order bullwhip is zero, the supplier and customer both perfectly
cr
match the variability of end customer demand suggesting a condition of full supply chain
us
coordination. Coordinating production across supply chain partners is often very expensive (Schoenherr and Swink, 2012), especially in the context of B2B manufacturing relationships
an
where salespeople are critical to maintaining buyer-supplier relationships and for gathering the necessary information visibility required to coordinate production (Lynch and de Chernatony, Thus it is not unexpected that SGA Expense increases as additional resources are
M
2007).
d
required to gather production information from customers along with end customer demand
te
information from even further downstream.
8.2.3. Bullwhip-Asset Performance Relationship
Ac ce p
The performance impacts of bullwhip have often been controlled through engaging capacity or inventory buffers. Therefore, in evaluating the asset side of ROA, it is possible to gain insights on the extent to which such control mechanisms are related to the bullwhip-performance relationship. However, before discussing the asset-based implications of bullwhip, it is first necessary to understand how firms account for such costs. Capacity buffers, often indicated by a low level of fixed asset turnover (Hendricks, Singhal and Zhang, 2009), have ongoing costs that are typically represented as opportunity costs rather than as increased operating costs. As such, any costs of underutilized production capacity are not likely to be captured in OM via COGS, but will only serve to increase the asset base of ROA. While inventory purchase costs (building
30 Page 30 of 54
inventory) are included in COGS, any subsequent inventory holding costs (costs of holding existing inventory) are typically included in Operating Expense. As such, costs associated with holding extra inventory directly impacts net income rather than OM, while costs associated with
sales constant).
ip t
increasing inventory levels directly translate into higher COGS thereby lowering OM (holding When evaluating the bullwhip-asset relationship, we find starkly differing
cr
results from those on the income side of ROA. First order bullwhip has significant relationships
us
with both fixed asset turnover and inventory turnover. However, counterintuitively when first order bullwhip is extremely dampened, inventory turnover is at its highest and fixed asset
an
turnover is at its lowest; indicating a relatively small inventory buffer and a relatively large capacity buffer. This result suggests that, counter to conventional wisdom, firms appear to
M
dampen first order bullwhip primarily through means other than through inventory-based control
d
mechanisms.
te
Conversely, firms that are experiencing amplified first order bullwhip possess higher relative inventory levels and generally lower capacity buffers.
However, we do find that
Ac ce p
capacity buffers begin to increase at very high levels of first order amplification, indicating that firms start to employ both inventory and capacity buffers at extremely high levels of amplification. These findings suggest that increased inventory levels are not automatically capable of or required for smoothing production or dampening first order bullwhip effects. High inventory levels that accompany high first order bullwhip levels may be symptomatic of previous supply and demand mismatches that have yielded excess inventories of items that while in stock are currently not demanded, and are thus ineffective in smoothing current production. Interestingly, while first order bullwhip was found to impact fixed asset turnover and inventory turnover, these operationally oriented effects are not significant enough to impact OM.
31 Page 31 of 54
While amplifying (dampening) firms tend to hold higher (lower) levels of inventory, their linkage to a non-significant OM result is less clear. For example, holding sales constant, OM would not be impacted by higher (lower) levels of inventory unless the level of inventory was Alternatively, since we have no way to
ip t
growing (shrinking) thereby impacting COGS.
determine the extent of lost sales, it is possible that increased COGS due to growing inventory
cr
levels are offset by increased revenues from the firm having fewer lost sales. However, since
us
first order bullwhip is shown to ultimately have no relationship with OM, the reduction of lost sales or growth of inventory is not significant enough to impact OM.
Conversely, while second
an
order bullwhip does impact OM, it was shown to have no relationship with inventory and only a small effect on fixed asset turnover, thus suggesting that second order bullwhip impacts
M
operational performance through means other than through asset or inventory performance.
d
Taken together, these results indicate that utilizing capacity and/or inventory bullwhip control
te
mechanisms ultimately has little impact on operating performance (OM). 8.2.4 Summative thoughts on Bullwhip-Performance Pathways
Ac ce p
Regarding first order bullwhip, while the capacity and inventory levels are likely to be generally offsetting in terms of increasing (decreasing) the size of the asset base by which ROA is calculated, lower inventory levels have an additional positive impact on net income as holding costs are also reduced. However, unlike inventory buffers, when capacity buffers are reduced, they mainly only impact the size of the asset base since unused capacity is typically considered as an opportunity cost. When coupled with the impact of SGA expenses, first order bullwhip approaches its highest ROA levels as SGA expenses decline and inventory holding costs are reduced due to holding a lower level of inventory. Thus it appears that first order bullwhip
32 Page 32 of 54
impacts ROA primarily through inventory and SGA expenses rather than through directly impacting operational performance or influencing the size of the firm’s asset base. Second order bullwhip effects however, appears to impact ROA primarily through OM
ip t
and lower inventory holding expenses. While we do find that second order effects are linked to increased SGA expense as the second order approaches zero, these increased costs simply act to
The absence of a relationship between second order bullwhip and inventory
us
relationship.
cr
shift the second order-ROA relationship curve lower rather than change the shape of the
turnover, and only a small relationship with fixed asset turnover, further suggests that OM is
an
impacted by second order bullwhip primarily through influencing the level of revenue (e.g. more or less lost sales) or through impacting the non-inventory aspects of COGS rather than through
M
impacting the firm’s asset base.
d
8.3 Supply Chain Production Coordination as Managerial Lever of Bullwhip
te
Bullwhip is inherently intertwined with the concept of supply chain coordination. If all supply chain partners perfectly coordinated their production not only with each other but also with end
Ac ce p
customer demand, bullwhip would not occur and all production variability would be equal to demand variability. Our two measures of bullwhip provide some insights as to the extent to which production coordination along the supply chain informs the bullwhip-performance relationship. As either first or second order bullwhip measures diverge from zero, lower levels of coordination between the supplier and their downstream supply chain partners is indicated. Based on our results, supplier production coordination seems to be most relevant with respect to ROA, OM and SGA performance outcomes. As suggested previously, coordination across supply chain partners can be very expensive, therefore it is not unexpected that the highest SGA costs would occur when full downstream coordination is indicated (both first and second
33 Page 33 of 54
order bullwhip effects approaching zero). When evaluating OM, additional interesting insights are gleaned. For example, OM performance is at its highest level when second order bullwhip approaches zero, indicating that the supplier is coordinating with its customer. Given that OM is
ip t
sensitive to both revenue and direct costs (COGS), this result suggests that when the supplier and customer are coordinating their production, few lost sales occur and the supplier is not forced to
cr
utilize expense short-term capacity adjustment options, such as overtime or lay-offs.
us
Additionally, given the costs associated with coordination, this result suggests that such additional coordination costs are more than offset by the benefits of avoiding lost sales and short
an
term capacity alternatives. However, we find no significant relationship between OM and first order bullwhip, suggesting that with respect to OM, there are few benefits of the supplier
M
coordinating beyond the production of their immediate customers.
d
Finally, while first order effects generally indicate if ROA will be positive or negative;
te
firms nevertheless can benefit significantly by optimizing second order bullwhip so as to generate the highest possible ROA for a given level of first order bullwhip. For example, the
Ac ce p
highest positive ROA occurs when suppliers are extreme dampeners of their first order bullwhip and also highly coordinated with their customers. Likewise, at high levels of first order bullwhip where ROA is generally negative, a high level of coordination can even result in slightly positive ROA. Such a finding suggests that second order effects can be seen as a managerial lever that can potentially magnify positive ROA first order effects, or limit the damage from negative ROA first order effects. Therefore, it becomes critical to evaluate both the first and second order effects together rather than separately because otherwise one could erroneously conclude that smoothing always yields a highly positive ROA outcome if only the first order bullwhip is evaluated. However, if
34 Page 34 of 54
the supplying firm is smoothing, but much less so than their customers, then their ROA would actually only approach the breakeven point. Additionally, relying only upon first order results would suggest that extreme amplification would yield negative results. When evaluated together
ip t
with second order effects, it is clear that if the supplier is fully coordinating with their customers, amplifying first order effects can result in breakeven or even slightly positive ROA outcomes.
cr
This finding is surprising given that conventional wisdom suggests that firms should benefit
us
from trying to “tame” the bullwhip effects by decelerating an amplifying first order bullwhip effect; however, we find the opposite actually occurs. These insights suggest that supplying
an
firms are always better off working in tandem/coordinating with their immediate customers. However such benefits should not suggest that suppliers should blindly follow their customer’s
M
bullwhip strategy since the general direction (positive or negative) of supplier’s ROA is largely
d
influenced by their own first order bullwhip behavior rather than through the level of
9.0 Conclusion
te
coordination via second order bullwhip.
Ac ce p
The bullwhip effect has been intensely researched since the 1950’s, and its impact on firm performance is generally accepted as resulting in significant negative supplier performance. However, despite this rich research history, and seemingly universal understanding of its performance implications, the results of this study suggest that the performance pathways whereby bullwhip impacts supplier performance are far more complex and nuanced than previously described in extant literature.
Additionally, our evaluation of both the demand
distortion (first order bullwhip) and rate of amplification (second order bullwhip) provide a more granular empirical examination of bullwhip than previously seen. Despite its long and rich
35 Page 35 of 54
history, we suggest that there remains much to understand about how the ubiquitous bullwhip effect actually impacts firm performance. When building upon our work to further the above-mentioned goal, it is important to
ip t
recognize that to our knowledge, this is the first study to empirically show the existence of a second order bullwhip effect, and to also empirically evaluate the performance implications of
cr
both the first and second order bullwhip effects. Apart from its theoretical contributions, our
us
results provide several insights and recommendations that are relevant to practicing managers. 9.1 Implications for Bullwhip Theory Building
an
This study helps to advance the understanding of not only the extent and pathways whereby bullwhip may impact performance, but also does so by evaluating more specific bullwhip
M
dimensions than has been done before. While our results do generally confirm the conventional
d
wisdom that higher levels of first order bullwhip results in lower ROA performance, we are also
te
able to provide additional insights suggesting that such a negative relationship is driven mainly through increased assets, rather than reduced operating margins.
This result is counter to
Ac ce p
conventional wisdom that poor performance related to bullwhip is mainly due to poor operational performance.
Additionally, in this study we have sought to further clarify the bullwhip-performance relationship by introducing and testing second order bullwhip. The evaluation of second order bullwhip provides additional theoretical depth to this relationship, by showing that not only does the level of amplification matter, but so does the rate of bullwhip acceleration (deceleration). When examining the performance implications of both first and second order bullwhip effects, a complex pattern of results emerges. This high level of complexity indicates that the performance pathways whereby bullwhip impacts performance are far more nuanced that generally believed,
36 Page 36 of 54
suggesting great need for further testing and theory development efforts regarding this relationship. Taken together, we suggest that the bullwhip-performance relationship is deserving of additional research that can further drive the theory development efforts illuminating this
ip t
important relationship. 9.1.1 Implications for ETPS and OIPT
cr
Although the bullwhip effect is heavily researched, much of this deep body of research has
us
focused on its causes and potential remedies. Thus far, anecdotal evidence, case studies and behavioral simulations have served as the primary source of theory building related to the
an
bullwhip-performance relationship. Theorizing based upon empirically grounded analysis, large scale or otherwise, has been conspicuously absent. Leveraging the perspectives of ETPS and
M
OIPT, this study seeks to initiate a more thorough theory development process of the bullwhip-
d
performance relationship that is grounded upon real world empirical data. As such, this study
te
not only seeks to inform us about the generally positive value of ETPS and OIPT in predicting the bullwhip-performance relationship, but it also provides granular evidence that may be
Ac ce p
utilized in formulating a unique and specific theoretical perspective of bullwhip. ETPS suggests that performance will diverge from optimal as firm behavior regarding inventory and flexibility also diverges from optimal or profit maximizing. This study provides some supporting evidence that smoothing production generally results in improved ROA. However, we are unable to support the ETPS based prediction that ROA should be lower at extreme levels of smoothing or that extreme smoothing is accomplished through holding high levels of inventory. As such, our ROA results suggest that a boundary condition exists in the applicability of ETPS to the bullwhip-performance relationship, especially at extreme levels of
37 Page 37 of 54
dampening. Further research is necessary to more fully explicate the means and methods which firm’s employ such that extreme dampening seemingly has no performance downside. Regarding OIPT, our results indicate that ROA always improves as production is more
ip t
highly coordinated (second order = 0). In fact, even when first order bullwhip is severely amplifying, and ROA is typically negative, the supplying firm’s ROA can actually be slightly
cr
positive if customer-supplier coordination is high enough. Such a result suggests that the ability
us
of the supplier to reduce their uncertainty and equivocality through coordinating production with their customer can be a powerful managerial lever in enhancing bullwhip related performance.
an
However, evaluation of the ROA decomposition results do not indicate clear support that fully coordinating production decreases inventory or capacity buffers, as suggested by OIPT. For
M
example, counter to the predictions of OIPT, we find no relationship between inventory and
d
second order bullwhip. While OIPT based arguments do predict our ROA results, the exact
te
mechanisms whereby customer-supplier coordination impacts ROA seem not to be those mechanisms (inventory and capacity buffers) suggested within OIPT. Taken together, our results
Ac ce p
indicate that while OIPT is generally applicable in evaluating the second order bullwhip-ROA relationship, a boundary condition seems to exist. Additional research is necessary to identify alternative mechanisms for how the second order bullwhip-ROA relationship is achieved. 9.2 Implications for Practice
First, while it was expected that bullwhip would impact ROA, this study demonstrates that the relationship is highly sensitive to the joint effect of both first and second order bullwhip effect and can often be the difference between positive and negative ROA. Secondly, while both first and second order bullwhip effects impact various types of performance outcomes, managers should recognize that significant dampening of first order bullwhip almost always yields positive
38 Page 38 of 54
ROA, regardless of second order effects. However, supplier’s that engage the managerial lever of coordinating production with their customers always experience higher levels of ROA than when not coordinating. Thus this study expands our current understanding of the bullwhip effect
ip t
by suggesting that both first and second order bullwhip effects should be managed simultaneously in order to achieve the highest levels of ROA. Third, because ROA implications
cr
of first order bullwhip are largely driven through inventory levels and SGA expenses, managers
us
could seemingly insulate their firm’s ROA from first order bullwhip implications if they could control the growth of both SGA and inventory. Finally, these results suggest that managers
an
should not underestimate their own power to influence their firm’s bullwhip-performance relationship via coordination (managerial lever) and bullwhip control mechanisms (inventory and Firms that actively manage the bullwhip-performance relationship by
M
capacity buffers).
d
appropriately using these managerial levers and control mechanisms may be able to gain
te
competitive advantage over their peers who are unable to do so. 9.3 Limitations and Future Research
Ac ce p
As is the case for all empirical research, there are a number of limitations associated with this study. First, we recognize that bullwhip is inherently a product level issue and that decisions related to its control are generally conducted at the plant level rather than the firm level, thus while we are confident that our firm level aggregation approach is conservative given the findings of Chen and Lee (2012), we cannot guarantee that the bullwhip-performance relationship behaves the same at the product/plant level of analysis. While firm performance and bullwhip inherently occur at different theoretical levels (firm versus plant or product), we utilize aggregated measures of bullwhip such that both bullwhip and performance are evaluated at the firm level. Should product level supply chain data become available, we encourage future
39 Page 39 of 54
research to extend our model to the product level. Additionally, we infer that coordination is occurring between firms if first or second order bullwhip effects are zero, meaning the supplier has the same production variance as end customer demand variance (first order) or the supplier’s
ip t
production variance is the same as their customer’s production variance (second order). However, we are unable to measure explicitly if coordination is occurring primarily due to
cr
information sharing, VMI arrangements, or other explicit coordination mechanisms. We do
us
believe however that it is unlikely for supplier and customer production variance to equalize randomly without coordination/communication between firms.
an
Despite some of the limitations that arise naturally in conducting such a complex study with real world data that we assiduously put together, we believe that it both confirms as well as
M
opens up new areas of research. First, the identification of second order bullwhip effect is a
d
prime avenue for future researchers to explore. Such research could include identifying the
te
mechanisms for influencing or controlling second order bullwhip beyond various forms of coordination. In particular, does second orders bullwhip impact performance more severely in
Ac ce p
certain contexts, for example intensely competitive markets? Is the second order bullwhip more or less easy to influence than the first order bullwhip by the dominant player in the supply chain? How does second order bullwhip impact ROA, if it is seemingly not through substituting inventory and capacity for information?
Additionally, while we have primarily evaluated
inventory and capacity dampening mechanisms, it may be worthwhile to explore the impacts of alternative dampening mechanisms. Finally, the empirical finding that there is no universal expression of traditional bullwhip (Table 4), suggests that there may be rational reasons for firms to choose their bullwhip levels. Future research should evaluate what characteristics contribute
40 Page 40 of 54
to firms being better (worse) off by possessing various combinations of first and second order
ip t
bullwhip.
cr
REFERENCES
Ac ce p
te
d
M
an
us
Arrow, K. J., Karlin, S. and Scarf, H. (1958). Studies in the mathematical theory of inventory and production (No. 1). Stanford University Press. Baganha, M., and Cohen, M. 1998. The stabilizing effect of distribution center inventory. Operations Research, 46, 72–83. Blanchard, O. 1983. The production and inventory behavior of the American automobile industry. Journal of Political Economy, 91, 365-400. Blinder, A. 1986. Can the production smoothing model of inventory behavior be saved? Quarterly Journal of Economics, 101, 431-453. Bloom, N. (2013). Fluctuations in uncertainty (No. w19714). National Bureau of Economic Research. Bray, R. L., and Mendelson, H. 2012. Information transmission and the bullwhip effect: An empirical investigation. Management Science, 58(5), 860-875. Cachon, G.P., Randall, T., and Schmidt, G.M. 2007. In search of the bullwhip effect. Manufacturing and Service Operations Management, 9 (4), 457-479. Cameron, A. C., Gelbach, J. B., & Miller, D. L. 2011. Robust inference with multiway clustering. Journal of Business & Economic Statistics, 29(2), 238-249. Chen, L., & Lee, H. L. 2012. Bullwhip effect measurement and its implications. Operations Research, 60(4), 771-784. Chen, F., and Samroengraja, R. 2004. Order volatility and supply chain costs. Operations Research 52(5): 707–722. Cortina, J. M. (1994). Interaction, nonlinearity, and multicollinearity: Implications for multiple regression. Journal of Management, 19(4), 915-922. Daft, R. L., & Lengel, R. H. 1986. Organizational information requirements, media richness and structural design. Management science, 32(5), 554-571. Dejonckheere, J., Disney, S.M., Lambrecht, M.R. and Towill, D.R. 2004. The impact of information enrichment on the Bullwhip effect in supply chains: A control engineering perspective. European Journal of Operations Research, 153 (3), 727-750. Dehning, B., Richardson, V. J., & Zmud, R. W. (2007). The financial performance effects of ITbased supply chain management systems in manufacturing firms. Journal of Operations Management, 25(4), 806-824. Financial Accounting Standards Board. 1997. Statement of Financial Accounting Standards No. 131: Disclosures about Segments of an Enterprise and Related Information. Norwalk, CT. Financial Accounting Standards Board. 1976. Statement of Financial Accounting Standards No. 14: Financial Reporting for Segments of a Business Enterprise. Norwalk, CT. Forrester, J. 1958. Industrial Dynamics—a major breakthrough for decision makers. Harvard Business Review, 36 (4), 37-66 41 Page 41 of 54
Ac ce p
te
d
M
an
us
cr
ip t
Forrester, J. 1961. Industrial Dynamics. MIT Press, Cambridge, MA. Fransoo, J.C. and Wouters, M.J.F. 2000. Measuring the Bullwhip Effect in the supply chain. Supply Chain Management: an international journal, 5, 78-89. Galbraith, J. R. 1973. Designing complex organizations. Boston, MA: Addison-Wesley. Galbraith, J. R. 1974. Organization design: An information processing view. Interfaces, 4(3), 2836. Giard, V., & Sali, M. 2013. The bullwhip effect in supply chains: a study of contingent and incomplete literature. International Journal of Production Research, 51(13), 3880-3893. Goodhue, D. L., Wybo, M. D., & Kirsch, L. J. (1992). The impact of data integration on the costs and benefits of information systems. MiS Quarterly, 293-311. Hendricks, K. B., V. R. Singhal, and Zhang, R. (2009). "The effect of operational slack, diversification, and vertical relatedness on the stock market reaction to supply chain disruptions." Journal of Operations Management 27(3): 233-246. Hertzel, M.G., K. Li, M.S. Officer, and K.J. Rodgers. 2008. Inter-firm linkages and the wealth effects of financial distress along the supply chain. Journal of Financial Economics, 87 (2), 374-387. Holt, C.C., Modigliani, F., Muth, J. and Simon, H. 1960. Planning Production, Inventories and the Work Force, Prentice-Hall, NJ. Koste, L. L., & Malhotra, M. K. (1999). A theoretical framework for analyzing the dimensions of manufacturing flexibility. Journal of operations management, 18(1), 75-93. Lanier Jr, D., W.F. Wempe, and Z.G. Zacharia. 2010. Concentrated Supply Chain Membership and Financial Performance Chain and Firm Level Perspectives. Journal of Operations Management, 28 (1), 1-16. Lee, H.L., Padmanabhan V. and Whang, S. 1997a. Information distortion in a supply chain: The bullwhip effect. Management Science, 43 (4), 546–548. Lee, H. L., Padmanabhan, V., & Whang, S. 1997b. The bullwhip effect in supply chains1. Sloan management review, 38(3), 93-102. Lee, H. L., So, K. C., & Tang, C. S. (2000). The value of information sharing in a two-level supply chain. Management science, 46(5), 626-643. Liker, J. K., & Morgan, J. M. (2006). The Toyota way in services: the case of lean product development. The Academy of Management Perspectives, 20(2), 5-20. Lynch, J., & deChernatony, L. 2007. Winning hearts and minds: business-to-business branding and the role of the salesperson. Journal of Marketing Management, 23(1-2), 123-135. Metters, R. (1997). Quantifying the bullwhip effect in supply chains. Journal of operations management, 15(2), 89-100. Miron, J. and Zeldes, J.S., 1988. Seasonality, cost shocks, and the production smoothing model of inventories. Econometrica, 56, 877-908. Modi, S. B., & Mishra, S. (2011). What drives financial performance–resource efficiency or resource slack?: Evidence from US Based Manufacturing Firms from 1991 to 2006. Journal of Operations Management, 29(3), 254-273. Petersen, M. A. 2009. Estimating standard errors in finance panel data sets: Comparing approaches. Review of Financial Studies, 22(1), 435-480. Schoenherr, T., & Swink, M. 2012. Revisiting the arcs of integration: Cross-validations and extensions. Journal of Operations Management, 30(1), 99-115. Shan, J., Yang, S., Yang, S., & Zhang, J. 2013. An Empirical Study of the Bullwhip Effect in China. Production and Operations Management. 42 Page 42 of 54
Ac ce p
te
d
M
an
us
cr
ip t
Stank, T., Crum, M. and Arango, M. 1999. Benefits of Interfirm Coordination in Food Industry Supply Chains. Journal of Business Logistics, 20 (2), 21–41. Sterman, J. 1989. Modeling managerial behavior: Misperceptions of feedback in a dynamic decision-making experiment. Management Science, 35, 321-339. Swink, M. and B. W. Jacobs (2012). Six Sigma adoption: Operating performance impacts and contextual drivers of success. Journal of Operations Management 30(6): 437-453. Swink, M. and T. Schoenherr (2014). "The Effects of Cross-Functional Integration on Profitability, Process Efficiency, and Asset Productivity." Journal of Business Logistics: DOI: 10.1111/jbl.12070 Tang, Y.-C. and F.-M. Liou (2010). "Does firm performance reveal its own causes? The role of Bayesian inference." Strategic Management Journal 31(1): 39-57. Thompson, S. B. 2011. Simple formulas for standard errors that cluster by both firm and time. Journal of Financial Economics, 99(1), 1-10. Tushman, M. L., & Nadler, D. A. (1978). Information Processing as an Integrating Concept in Organizational Design. Academy of management review, 3(3), 613-624. Visnjic, I., & Van Looy, B. 2013. Servitization: Disentangling the impact of service business model innovation on manufacturing firm performance. Journal of Operations Management. Wang, E. T., Tai, J. C., & Wei, H. L. (2006). A virtual integration theory of improved supplychain performance. Journal of Management Information Systems, 23(2), 41-64. White, H. 1980. A Heteroskedasticity-Consistent Covariance Matrix Estimator and a Direct Test for Heteroskedasticity. Econometrica, 48, 817-838. Zhang, X., & Burke, G. J. 2011. Analysis of compound bullwhip effect causes. European journal of operational research, 210(3), 514-526.
43 Page 43 of 54
Table 1: Numerical Illustration of Bullwhip Calculations using Contrived Data Supplier (Tier 1) §
Known Information
Market Demand Variance=5 Production Variance of Customer = 7
Production Variance of Supplier = 8
First Order Bullwhip*
0.4 = (7-5)/5) (Production Variance of Customer is 40% higher than Market Demand Variance)
0.60 =(8-5)/5) (Production Variance of Supplier is 60% higher than Market Demand Variance)
N/A
0.50= ((8-5)-(7-5))/|7-5| (Supplier first order bullwhip is 50% higher than their Customer's First order Bullwhip indicating that bullwhip is accelerating between customer and supplier
cr
us
Second Order Bullwhip**
ip t
Customer §
an
§ Both customers and suppliers are manufacturers in our study * Positive values indicate amplifying, while negative values indicate dampening ** Positive values indicate accelerating bullwhip,, while negative values indicate decelerating bullwhip
M
Table 2: Data Characteristics
4
1.04
1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 Total
3 10 11 15 15 21 21 17 16 23 26 37 30 28 35 32 26 13 383
0.78 2.61 2.87 3.92 3.92 5.48 5.48 4.44 4.18 6.01 6.79 9.66 7.83 7.31 9.14 8.36 6.79 3.39 100
Ac ce p
te
1990
2b. Industry Distribution of Supplying Firms Supplier Frequency Percent 2-Digit SIC 20 8 2.09 24 1 0.26 25 17 4.44 26 4 1.04 28 62 16.19 29 1 0.26 30 7 1.83 32 1 0.26 33 1 0.26 34 10 2.61 35 44 11.49 36 79 20.63 37 73 19.06 38 69 18.02 39 6 1.57 Total 383 100
d
2a. Yearly Distribution of Quarterly Data Year Frequency Percent
44 Page 44 of 54
1
8.33
49.38
0.26
-0.01
0.20
-0.03
0.12
1
0.19
0.62
0.03
0.05
0.31
1
0.22
0.20
1.92
1.23
Supplier Inventory Turnover Control Variables 8. Supplier Firm Size 9. Supplier Market Share 10. Customer Base Firm Size
ip t 6
7
8
9
10
0.22
0.06
-0.22
0.58
1
0.08
-0.03
0.21
0.17
-0.04
1
11.01
-0.15
-0.04
0.02
-0.11
-0.26
-0.05
1
5.51
1.98
-0.21
-0.09
0.03
-0.01
-0.48
-0.22
0.22
1
0.06
0.14
-0.08
-0.04
0.05
0.02
-0.25
-0.03
0.21
0.38
1
9.08
2.30
-0.18
-0.08
0.05
0.03
-0.46
-0.08
0.18
0.82
0.28
0.10 4.58
0.02 -0.08
0.01 -0.07
0.02 -0.05
0.02 0.06
-0.09 0.03 -0.02 0.002
-0.11 -0.04
0.04 0.23
11. Customer Base Market Share 0.08 12. Time (Years) 10.96 Significant correlations are in Bold
11
12
1
13.54
Ac
7.
5
4.65
ed
Supplier Fixed Asset Turnover
4
1.52
2.
6.
3
us
2
ce pt
Second Order Bullwhip Dependent Variables 3. Supplier ROA 4. Supplier Operating Margin 5. Supplier SGA Expense
1
M an
Independent Variables 1. First Order Bullwhip
S.D.
cr
Table 3: Descriptive Statistics Mean
1
-0.01 0.07 -0.11 0.37
1 0.01
1
45 Page 45 of 54
Decelerating (<0) Accelerating (>0)
First Order Bullwhip Dampening (<0) Amplifying (>0) 105 47 (27.4%) (12.3%)
152 (39.7%)
39 (10.2%)
192 (50.1%)
231 (60.3%)
144 (37.6%)
239 (62.4%)
383 (100%)
cr
Total
Total
ip t
Second Order Bullwhip
Table 4: Distribution of Sample across First and Second Order Bullwhip Effects
us
Table 5: Regression Results for Financial Performance
Equation 6 -0.05
-0.01** -0.01 0.11 0.002 0.08
-0.01** -0.004 0.10 0.003 0.09
Included
Included
Included
-
-0.08***
-0.08***
Second Order Bullwhip (B)
-
0.10**
0.07
A2
-
0.07***
0.06***
-
-0.07*
-0.06**
AxB
-
-
0.03***
Centered-R2
0.05
0.11
0.14
Δ Centered-R2 F
5.70 (14)
0.06 5.02 (18)
ΔF Observations Number of Suppliers
-
0.68 (4)**
0.03 3.11 (19) 1.91 (1)***
B2
M
d
Ac ce p
First Order Bullwhip (A)
-0.003 -0.007 0.10 0.003 0.04
te
Intercept Controls Time Supplier Firm Size Supplier Market Share Customer Base Firm Size Customer Base Market Share Supplier Industry (Dummy Variables) Dependent Variables
Base Model -0.04
an
ROA Equation 5 -0.07
383 147
All dependent variables are standardized. p<0.05*, p<0.01** p<0.001***
46 Page 46 of 54
ip t
Base Model 0.23**
First Order Bullwhip (A)
Base Model 0.23***
Equation 5 0.24***
Equation 6 0.24***
0.002
0.003
0.001
0.001
0.001
-0.01 0.10 0.04 0.03 Included
-0.01 0.11 0.04 0.02 Included
0.002 -0.16** -0.01 -0.14 Included
0.003 -0.15 -0.01 -0.13 Included
0.002 -0.15** -0.01 -0.13 Included
-
-0.02
-0.02
-
0.07*
0.07**
-
0.12*
0.13**
-
0.03
-0.12
-
0.01
0.01
-
-0.06*
-0.06*
-
-0.10**
-0.10**
-
-0.03**
-0.03**
-
-
-0.01
-
-
-0.001
0.10
0.11
0.11
0.49
0.52
0.53
22.16 (14) -
0.01 2.81 (18) 19.35 (4)*** 383 147
0 2.78 (19) 0.03 (1)
25.35 (14) -
0.03 23.40 (18) 1.95 (4)* 340 128
0.0 20.14 (19) 3.26(1)***
0.003 -0.01 0.10 0.04 -0.001 Included
ce pt
Second Order Bullwhip (B) A2 B2
Ac
AxB Δ Centered-R2 F ΔF Observations Number of Suppliers
SGA Expense
Equation 6 0.20***
M an
Time Supplier Firm Size Supplier Market Share Customer Base Firm Size Customer Base Market Share Supplier Industry (Dummy Variables) Dependent Variables
Equation 5 0.21***
ed
Intercept Controls
Centered-R2
us
Operating Margin
cr
Table 6: Regression Results Related to Profit Performance
All dependent variables are standardized. p<0.05*, p<0.01**. p<0.001***
47 Page 47 of 54
ip t
cr
Table 7: Regression Results Related to Asset Performance
First Order Bullwhip (A) A2 B2 AxB Centered-R2
Ac
Δ Centered-R2 F ΔF Observations Number of Suppliers
Inventory Turnover Base Model Equation 5 Equation 6 15.13*** 15.30*** 15.21***
us
0.01 -0.41*** 1.06 0.21*** 0.24
0.01 -0.41*** 1.09 0.21*** 0.24
-0.15 0.39 11.66* 0.78 -14.11***
-0.16 0.34 11.59* 0.72 -13.37***
-0.15 0.31 11.87* 0.72 -13.59***
Included
Included
Included
Included
M an
Included
Included
-
0.27*
0.27*
-
-1.82*
-1.83*
-
-0.17*
-0.12
-
1.06
1.34
-
-0.19*
-0.17
-
1.07
1.18
-
0.07
0.06
-
-1.03
-1.08
-
-
-0.06
-
-
-0.37
0.25
0.26
0.26
0.15
0.16
0.17
5.42 (14) -
0.01 7.43 (18) -2.01 (4) 383 147
0.0 7.74 (19) -0.31 (1)
24.60 (14) -
0.01 17.69 (18) 6.91 (4)*** 383 147
0.0 41.26 (19) -23.57 (1)
ce pt
Second Order Bullwhip (B)
0.01 -0.41*** 1.08 0.21*** 0.35
ed
Intercept Controls Time Supplier Firm Size Supplier Market Share Customer Base Firm Size Customer Base Market Share Supplier Industry (Dummy Variables) Dependent Variables
Fixed Asset Turnover Base Model Equation 5 Equation 6 1.90*** 2.01*** 1.99***
All dependent variables are standardized. p<0.05*, p<0.01**. p<0.001***
48 Page 48 of 54
ip t
Key Insights
1. Dampening yields increasingly positive ROA.
1. ROA declines as second order diverges from zero.
1. Positive or negative ROA mainly determined by first order bullwhip.
2. Amplifying generally results in negative ROA.
2. ROA is always negative.
3. First order bullwhip impacts ROA predominantly through expenses rather than through operational performance or asset levels.
3. Second order bullwhip impacts ROA predominantly through operational performance and SGA expenses rather than through asset levels.
us
Second Order Bullwhip
2. Second order bullwhip can enhance benefits of dampening, and help to overcome detrimental effects of amplifying.
ce pt
No relationship
Operating margin generally declines as second order bullwhip diverges from zero.
SGA Expense
SGA Expenses are near their highest when first order bullwhip is near zero
SGA Expenses are highest when second order bullwhip is near zero.
Inventory Turnover
Inventory levels decrease as first order bullwhip is reduced.
No relationship.
Fixed Asset Turnover
Dampening is associated with much larger capacity buffers than amplifying.
Small negative effect.
Operating Margin
Ac
Asset Side of ROA
Income Side of ROA
ed
ROA
First Order Bullwhip
M an
Evaluated Aspect of Performance
cr
Table 8: Summary of Key Findings
The level of demand distortion does not impact operational performance, but accelerating or decelerating it does.
Negligible net change in asset levels by either first or second order bullwhip.
Directions of Further Research 1. Is there a downside to dampening? 2. Are there benefits to amplifying that are not captured by ROA? 3. Are there conditions under which second order bullwhip effects dominate the first order effects? 4. To what extent can firms influence their own bullwhip performance pathways? 1. How have firms insulated their operating margins from first order bullwhip effects? 2. Do the different ways to coordinate production impact the extent to which second order bullwhip impacts operating margins? 1. What are the different strategies related to utilizing capacity and inventory jointly to impact bullwhip? 2. Are other bullwhip control mechanisms more effective?
49 Page 49 of 54
d
M
an
us
cr
ip t
Figure 1a: Plot of ROA Results
Ac ce p
te
Figure 1b: Cross-sectional View of ROA Results (First Order =X, Second Order =Y)
50 Page 50 of 54
M
an
us
cr
ip t
Figure 2a: Plot of Operating Margin Results
Ac ce p
te
d
Figure 2b: Cross-sectional View of Operating Margin Results (First Order =X, Second Order =Y)
51 Page 51 of 54
d
M
an
us
cr
ip t
Figure 3a: Plot of SGA Expense Results
Ac ce p
te
Figure 3b: Cross-sectional View of SGA Expense Results (First Order =X, Second Order =Y)
52 Page 52 of 54
M
an
us
cr
ip t
Figure 4a: Plot of Fixed Asset Turnover Results
Ac ce p
te
d
Figure 4b: Cross-sectional View of Fixed Asset Turnover Results (First Order =X, Second Order =Y)
53 Page 53 of 54
M
an
us
cr
ip t
Figure 5a: Plot of Inventory Turnover Results
Ac ce p
te
d
Figure 5b: Cross-sectional View of Inventory Turnover Results (First Order =X, Second Order =Y)
54 Page 54 of 54
S0272-6963(15)00017-0 http://dx.doi.org/doi:10.1016/j.jom.2015.02.003 OPEMAN 899
To appear in:
OPEMAN
Received date: Revised date: Accepted date:
17-2-2014 9-1-2015 25-2-2015
Please cite this article as: Mackelprang, A.W., Malhotra, M.K.,The Impact of Bullwhip on Supply Chains: Performance Pathways, Control Mechanisms, and Managerial Levers, Journal of Operations Management (2015), http://dx.doi.org/10.1016/j.jom.2015.02.003 This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and review of the resulting proof before it is published in its final form. Please note that during the production process errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.
The Impact of Bullwhip on Supply Chains: Performance Pathways, Control
cr
ip t
Mechanisms, and Managerial Levers
M
an
us
Alan W. Mackelprang * Assistant Professor of Operations Management College of Business Administration Georgia Southern University P.O. Box 8151 Statesboro, GA 30460-8151 [email protected]
Ac ce p
te
d
Manoj K. Malhotra Jeff B. Bates Professor and Chair Department of Management Science Moore School of Business University of South Carolina Columbia, SC 29208 Phone: (803) 777-2712; Fax: (803) 777-6876 [email protected]
March 3, 2015
* Corresponding Author
1 Page 1 of 54
The Impact of Bullwhip on Supply Chains: Performance Pathways, Control Mechanisms, and Managerial Levers
ip t
Abstract Even though few empirical studies have tried to actually explicate the relationship between the
cr
bullwhip effect and performance of the supplier firm, there exists a common perception for over
us
30 years among both practitioners and academics that the bullwhip effect naturally results in decreased firm profitability. Anecdotal evidence further suggests that this decline in profitability
an
arises from a decline in operational performance. However, the results of our study, which empirically examines the bullwhip effect across supply chain partners through an analysis of 383
M
actual customer base-supplier dyads, challenges this commonly held position by suggesting that
d
while traditional bullwhip often yields reduced ROA, it ultimately has no relationship with the
te
firm’s operating margin. Additionally, our results also call into question whether or not production coordination between customers and suppliers can minimize the need for inventory
Ac ce p
and capacity buffers, which are the two commonly utilized methods for battling the bullwhip effect. Thus the relationship between bullwhip and firm performance is far more nuanced and complicated than previously believed. We also show how the managerial bullwhip levers of coordinating production across supply chain partners, or deploying inventory and capacity buffer control mechanisms, can help maximize a firm’s performance along different dimensions. Keywords: Bullwhip Effect, Supplier Performance, Empirical, and Supply Chain
2 Page 2 of 54
The Impact of Bullwhip on Supply Chains: Performance Pathways, Control
ip t
Mechanisms, and Managerial Levers
1. Introduction
cr
It is commonly perceived that bullwhip reduces a firm’s performance by wreaking
us
operational havoc. We challenge this conventional wisdom by suggesting that while traditional bullwhip generally does result in lower return on assets (ROA), it has no relationship with the
an
firm’s operating margin, a primary indicator of operational performance. Such a finding suggests that how bullwhip actually impacts ROA is not as straightforward as previously thought.
M
This study dives deeply into the bullwhip-performance relationship by proposing and
d
evaluating a new dimension of bullwhip, which captures the acceleration (or deceleration) of
te
bullwhip between supply chain partners. Such an evaluation allows us to gain performance insights into not only the level of bullwhip a firm faces, but also the extent to which an
Ac ce p
acceleration (or deceleration) of the bullwhip impacts that firm’s performance. Results from this analysis will also challenge the conventional wisdom that customers and suppliers in a supply chain that synchronize/coordinate their production can minimize the need for inventory and capacity buffers that represent traditional mechanisms utilized to combat bullwhip. As such, we will show that with respect to the bullwhip effect, leaning too heavily upon what many would consider to be “already known” or “well understood” information may not always yield contextually accurate insights. At its most fundamental level, the bullwhip effect increasingly distorts the pattern of actual end customer demand to upstream supply chain partners (Lee, Padmanabhan and Whang,
3 Page 3 of 54
1997a, 1997b; Zhang and Burke, 2011), who are then often forced into “boom or bust” production cycles. Such extreme production cycles resulting from bullwhip have been shown to significantly increase supply chain costs and lower performance (Sterman, 1989). Likewise, the
ip t
Economic Theory of Production Smoothing (ETPS) suggests that firms have an economic incentive to avoid such extreme production cycles and instead should smooth production (Holt et
cr
al., 1960; Blanchard, 1983; Blinder 1986; Miron and Zeldes, 1988). Numerous researchers over
us
several decades have highlighted a number of countermeasures for combatting the bullwhip. Despite logical evidence that bullwhip behavior should be avoided and recommendations
an
on how to minimize it, there remains significant evidence that bullwhip still continues to occur in about 67% of firms (Bray and Mendelson, 2012; Shan et al., 2013). If firms are incentivized to
M
smooth production to avoid the “boom and bust” production cycles yet most do not do so, it
d
suggests from a rational perspective that these firms must have a larger incentive not to smooth
te
production, or from an irrational perspective, the majority of firms are subject to the same widespread behaviorally driven irrational actions. This apparent contradiction between theory
Ac ce p
and practice raises an interesting research question. Does the bullwhip effect actually result in reduced firm performance? If so, what are the specific pathways through which this decreased firm performance occurs? Finally, are there appropriate control mechanisms or managerial levers that can be applied to ameliorate the bullwhip effect?
2. Bullwhip Definitions and Assessment Before developing the theoretical model, it is important to first establish definitions and foundational concepts underlying this research. We subscribe to the traditional conceptual definition of bullwhip as described by Lee, Padmanabhan and Whang (1997a) as “the phenomenon where orders to the supplier tend to have larger variance than sales to the buyer 4 Page 4 of 54
(i.e., demand distortion), and the distortion propagates upstream in an amplified form (i.e., variance amplification).” Building upon this conceptual definition of bullwhip, it is clear that bullwhip
ip t
fundamentally consists of not only a level of demand distortion, but also the extent of variance amplification. In other words, while it is important to know if a firm is experiencing demand
cr
distortion, it is also important to know if such demand distortion is accelerating (or decelerating)
us
along the supply chain. For the purpose of this study, we describe these two aspects of bullwhip simply as first and second order bullwhip effects. First order bullwhip effect is the absolute level
an
of demand variation amplification for a given firm relative to the variance of actual demand for the furthest downstream partner (e.g. demand distortion); while second order bullwhip captures
M
the rate of change (e.g. variance amplification) in demand distortion (first order bullwhip)
d
between tiers in the supply chain. When evaluated together, first and second order bullwhip not
te
only capture the level of demand distortion at a given node in the supply chain, but also the extent to which demand distortion is accelerating (or decelerating) between tiers in the supply
Ac ce p
chain. Thus, it is possible to capture all possible bullwhip supply chain configurations by simultaneously utilizing these two aspects of bullwhip. We will first provide the mathematical equations we utilize to assess first and second order bullwhip effects, then we will provide a numerical example to further illustrate these calculations.
First order bullwhip for supplier s in year y is calculated as shown in Equation 1,
where CV (Psy) is the coefficient of variation of production of supplier firm s in the supply chain across the four quarters in year y, and CV (Dcy) is the coefficient of variation of demand in the supplier’s customer c across the four quarters in year y. (1) 5 Page 5 of 54
It is important to point out that in line with extant literature (Bray and Mendelson, 2012; Shan et al. 2013), both demand and production are calculated at the quarterly level. As configured in Equation 1, the first order bullwhip measure represents the percentage change in
ip t
upstream production variation relative to downstream demand variation. As such, positive values indicate an increase in the absolute level of demand variation or amplification, while a
cr
negative number indicates a decrease in variation with respect to downstream demand variation
us
or an absolute dampening. Given that first order bullwhip is consistent with how bullwhip has been examined in extant literature (Cachon et al., 2007; Bray and Mendelson, 2012), we will
an
interchangeably use the terms bullwhip and first order bullwhip.
Whereas first order bullwhip captures the production variability of the supplier with
M
respect to end customer demand variability, second order bullwhip captures the supplier’s
d
production variability as compared to the production variability of their customer. By doing so,
te
it is possible to determine how the firm is behaving on both an absolute (first order) and relative (second order) basis regarding production variability. Second order bullwhip for supplier s and
Ac ce p
customer c in year y is calculated as shown in Equation 2, where CV (Psy) is the coefficient of variation of production of supplier firm s in the supply chain across the four quarters in year y, CV (Dcy) is the coefficient of variation of demand of the supplier’s customer c across the four quarters in year y, and CV (Pcy) is the coefficient of variation in production for customer c across the four quarters in year y.
(2) When this variable is positive, the supplier’s first order bullwhip is larger than their customer’s first order bullwhip, thus indicating that the supplier is accelerating variance
6 Page 6 of 54
amplification relative to their customer. When this measure is negative, the supplier is decelerating variance amplification with respect to their customer. Equations 1 and 2 make it possible to distinguish all combinations of first and second order bullwhip effects, such as firms
ip t
that may be amplifying (first order) but at a decelerating rate (second order) and vice versa.
We use a numerical example with contrived data in Table 1 to illustrate the calculation of
cr
first and second order bullwhip using Equations 1 and 2. The first order bullwhip for the
us
customer is 0.4 as shown in Table 1, or in other words the customer’s production variance is 40% larger than the customer’s demand variance. First order bullwhip for the supplier is 0.6, or
an
in other words their production variance is 60% larger than the customer’s demand variance. Second order bullwhip is 0.50, indicating that the supplier is amplifying at a 50% higher rate
M
than which the customer is amplifying, thus confirming that the bullwhip is accelerating between
d
the customer and the supplier.
te
---------Insert Table 1 about here-----3. Recent Perspectives on Bullwhip
Ac ce p
In this section we treat the most recent developments related to the bullwhip. In addition, we refer interested readers to Giard and Sali (2013) for a more comprehensive literature review on research related to the bullwhip effect. While much of the historical bullwhip research has been centered on the causes and remedies of the bullwhip effect, a new research stream is emerging that seeks to empirically verify the existence of the bullwhip effect. For example, Cachon et al. (2007) using industry level data were unable to find evidence of a traditional bullwhip effect, but rather reported that industries within the wholesale sector had higher demand amplification levels than either retail or manufacturing industries. This result is counter to previous industry level studies that found amplifying in retail industries, but not in wholesale
7 Page 7 of 54
industries (Baganha and Cohen 1998). These disparities in recent bullwhip empirical findings (e.g. Baganha and Cohen, 1998) prompted Chen and Lee (2012) to differentiate between how bullwhip was originally measured by Lee et al. (1997a) from how other empirical studies have
ip t
largely measured bullwhip. They suggest that these prior empirical studies have been examining “the existence or magnitude of ‘aggregated’ bullwhip in material flow” because these measures
cr
capture material flow, aggregated products, and aggregated time periods. Furthermore, Chen and
us
Lee (2012) show mathematically that as product or time aggregation increases, the aggregated bullwhip ratio converges to one, meaning that production variation is equal to demand variation
an
and that neither amplifying nor dampening is occurring at the aggregate level. They further suggest that the conflicting results of both Baganha and Cohen (1998) and Cachon et al. (2007)
M
may be a result of the use of excessively aggregated data across both products (industry level)
d
and time (monthly). In an effort to enhance “cross-sectional granularity” rather than temporal
te
granularity, Bray and Mendelson (2012) and Shan et al. (2013) recently attempted to overcome some of the data related shortcomings found in Cachon et al. (2007) by analyzing quarterly-firm
Ac ce p
level data, which indicated that approximately two-thirds of firms experience a bullwhip effect. However, the data utilized by Cachon et al. (2007), Bray and Mendelson (2012), and Shan et al. (2013), does not account for the actual relationships between supply chain partners. As such, it is difficult to draw conclusions from these studies about the extent to which amplification flows from customers to suppliers because of the nature of the data utilized. An issue that emerges clearly from the extant literature is that while the causes of the bullwhip effect have been well covered in the seminal works of Forrester (1958, 1961), Sterman (1989), and Lee et al. (1997a) among others, empirically verified performance implications of the bullwhip effect have been limited. In terms of performance implications, Behavioral Beer
8 Page 8 of 54
Game experimental studies have found in general that as the amplification of bullwhip increases, total costs also increase dramatically (Sterman, 1989). Metters (1997) in his simulation study suggests that if the bullwhip effect could be eliminated entirely, firm profits would increase
ip t
between 15% and 30%. What is missing from this discourse however is a large-scale empirical validation and theorizing of how bullwhip affects firm performance in an actual supply chain
us
cr
relationship, which is what we focus on next as we develop our related hypotheses. 4. Development of Hypotheses
an
We are interested in examining the effects of both first order bullwhip and second order bullwhip on a supplying firm’s performance in a supply chain context. We draw upon the Economic
M
Theory of Production Smoothing (ETPS) and Organizational Information Processing Theory (OIPT) in developing our hypotheses. Since these two bullwhip effects are related but
d
conceptually separate from one another, we will treat each individually and accordingly develop
te
the related hypotheses in the following two sub-sections.
Ac ce p
4.1 First Order Bullwhip Effect and Performance Firms that experience bullwhip are commonly subjected to “boom and bust” production cycles with oscillating periods where demand can greatly exceed capacity and vice versa. Generally, such cycles produce high levels of inventory followed by increasing stock outs and backorders coupled with inefficient and expensive over-production, followed by sub-optimal and low utilization rates (Shan et al., 2013). While it is widely accepted that firms which are subjected to extreme bullwhip conditions generate lower performance, the theoretical underpinnings to explicate such a relationship have remained far less clear in extant literature. Since the 1960’s, economists have put forth ETPS which contends that firms have an economic incentive to smooth production when facing variable demand (Holt et al. 1960; 9 Page 9 of 54
Blanchard, 1983; Blinder 1986; Miron and Zeldes, 1988). Three primary assumptions underlie ETPS (1) demand is variable, (2) production costs are convex and (3) inventories can exist. The intuition behind ETPS is that profit maximizing firms should seek to minimize the joint costs of
ip t
production and holding inventory. Therefore, when production costs are convex, the firm has an incentive to produce at a stable rate, thereby building inventory when demand is low and
cr
satisfying high levels of demand through a combination of both production and inventory. If the
us
firm’s marginal cost of production is linear, there is no incentive to smooth production or to accumulate inventory. Instead demand should be met by adjusting production.
an
Viewing the bullwhip-performance relationship through the lens of ETPS helps to formalize the underlying behavior that has been widely visible in firms that experience bullwhip.
M
Specifically, ETPS suggests that profit maximizing behavior dictates that the least costly and
d
therefore the most profitable method (inventory and/or capacity) should be utilized to satisfy
te
demand. The condition mentioned above where marginal production costs are linear suggests that no penalties are incurred when manufacturing flexibility is invoked. However, it is widely
Ac ce p
documented that exceeding the limits of efficient production by invoking short term capacity options such as over-time and sub-contracting generally increases the marginal cost of production (Koste and Malhotra, 1999). Facing flexibility induced convex production costs, ETPS suggests that the firm has increasingly greater incentives to smooth production via relatively less costly inventory. Such a linkage implies that firms exhibiting increasingly high levels of first order bullwhip are not able to adequately smooth production, though they are incentivized to do so. Firms in this situation are increasingly likely to yield sub-optimal performance as actual behavior (utilizing increasingly more expensive flexibility options) diverges from profit maximizing behavior. Therefore, we suggest that as first order bullwhip
10 Page 10 of 54
increases, and firms diverge from the profit maximizing behavior suggested by ETPS, firm performance will decline. The benefits of stable production are well documented. For example, the concept of
ip t
level-loading or heijunka within Lean/Just-in-Time manufacturing has been widely adopted and shown to improve performance (Liker and Morgan, 2006). Likewise, stable production situations
However, it is possible for firms to diverge from the profit
us
(Liker and Morgan, 2006).
cr
enable process standardization and learning effects to take hold, further enhancing performance
maximizing behavior of ETPS by overly smoothing production. Such instances occur when the
an
firm continues to smooth production even though the relative cost of invoking a flexible response is less than smoothing production via inventory. Greater levels of inventory allow the firm to
However, as the level of inventory increases so do total
d
dampening the bullwhip effect.
M
absorb increasing levels of demand variability, while keeping production stable; thereby
te
inventory-holding costs. ETPS suggests that at some point, the firm’s marginal benefit of holding inventory will be surpassed by the marginal cost of doing so (Arrow, Karlin and Scarf,
Ac ce p
1958). If the firm continues to utilize inventory when the relative cost to do so exceeds the relative cost of invoking flexibility, ETPS suggests that firm performance is likely to be suboptimal as the firm diverges from profit maximizing behavior. Thus while on one hand, there are predicted negative performance implications associated with excessive amplification, there are also predicted negative performance repercussions associated with extreme smoothing. This reasoning suggests that in a quadratic fashion, extreme amplifying or dampening will result in lower levels of firm performance relative to when the firm is effectively neither dampening nor amplifying. Therefore based on ETPS, we argue that firm performance will have a negative quadratic relationship with the first
11 Page 11 of 54
order bullwhip, such that the firm performance will be highest when the first order bullwhip approaches zero (when there is essentially no bullwhip effect). H1: Firm performance improves as the first order bullwhip converges to zero
ip t
4.2 Second Order Bullwhip and Performance
cr
Whereas first order bullwhip captures the behavior of the firm with respect to end customer demand, second order bullwhip captures such behavior relative to the firm’s immediate
us
downstream partner (its own customer) in the supply chain. As such, second order bullwhip evaluates the extent to which the supplying firm is accelerating or decelerating amplification
an
relative to its customer. When second order bullwhip is zero, the supplying and customer firms
M
have identical first order bullwhip effects, suggesting that the supplier and customer are producing in tandem and thereby generating identical levels of production variability. Firms that
d
are producing in “tandem” are emblematic of coordinating firms.
te
The issue of sharing information between supply chain partners has been a consistent and major theme within bullwhip research (Lee, So and Tang, 2000). The general consensus is that
Ac ce p
sharing information across supply chain partners can reduce the distortion of demand information and also reduce the bullwhip.
Organizational Information Processing Theory
(OIPT) (Galbraith, 1973; Daft and Lengel, 1986) is especially informative about how the flow of information across supply chain partners can impact performance via the bullwhip. Within OIPT, tactics for managing uncertainties depend upon the magnitude of the uncertainty. OIPT suggests that low levels of uncertainty can be managed through information processing capabilities which are typically routine based mechanisms (Galbraith, 1974; Tushman and Nadler, 1978). However, in high uncertainty environments, the firm must increase their own information processing capabilities or employ slack resources as a way to keep the level of 12 Page 12 of 54
uncertainty within the current scope of the firm’s information processing capabilities (Swink and Schoenherr, 2014; Galbraith, 1974). In addition to uncertainty, OIPT indicates that equivocality can be another driver of needed information processing capabilities. Equivocality occurs when a
ip t
common interpretation does not emerge as a result of processing the gathered information. For example, during the 2008 global recession, firms were often unable to reach a consensus with
cr
respect to market demand, and therefore were also challenged in making strategic production
us
decisions (Bloom, 2013). Whereas uncertainty typically results from a lack of information
Lengel, 1986; Goodhue, Wybo and Kirsch, 1992).
an
quantity, equivocality is seen as resulting from low levels of information richness (Daft and
Within the context of second order bullwhip, the extent to which information is not
M
passed from customer to supplier can be viewed from the lens of OIPT as a source of uncertainty
d
for the supplier. Additionally, equivocality may also be expected when the level of coordination
te
between supply chain partners is low as it is unlikely that the supplying firm will have a high degree of confidence that their chosen level of production is what it would have been with
Ac ce p
perfect information. Slack resources as well as buffers can be utilized to combat the impacts of both uncertainty and equivocality by reducing the need for information processing (Swink and Schoenherr, 2014). However, slack resources can be costly (Modi and Mishra, 2011). Coordinating firms support processes that collect and share more accurate supply and demand information (Stank et al., 1999). Therefore, within the OIPT framework, coordinating firms are more likely to obtain not only a greater quantity of accurate information thus reducing uncertainty, but also the obtained information is more likely to have a higher level of detail and/or richness, thus also suggesting a lower level of equivocality (Daft and Lengel, 1986). By reducing both uncertainty and equivocality, the need to increase information processing
13 Page 13 of 54
capabilities or to utilize capacity and inventory buffers also decreases as do their associated costs. This situation is similar to the concept of “virtual integration” whereby the need to hedge against demand uncertainty is reduced as greater levels of in-depth high quality information is
ip t
shared between supply chain partners and information is effectively substituted for buffering mechanisms (Wang, Tai and Wei, 2006).
cr
Therefore, anchored by OIPT, we suggest that supplier performance should be highest
us
when customer and supplier firms are producing in tandem thereby reducing the information processing requirements of the supplier and also reducing the need for additional information
an
processing capabilities or capacity and inventory buffers. This argument suggests a quadratic relationship between second order bullwhip and supplier performance, because as firms diverge
M
from coordinated production (second order diverges from zero), performance should decline as
te
arguments, we posit hypothesis 2.
d
the firm’s uncertainty and equivocality increases. Based upon the preceding OIPT based
Ac ce p
H2: Firm performance improves as the second order bullwhip converges to zero 5. Data and Measurement
One of the major weaknesses of prior bullwhip empirical research has been an inability to simultaneously evaluate a number of different supply chains.
Prior empirical research has
generally been conducted with data being at the industry level (e.g. Cachon et al., 2007) or at the firm level (Bray and Mendelson, 2012), neither of which explicitly captures interactions across specific supply chain dyads. This issue is an important one because at the heart of the bullwhip effect is the reaction of a supplier to the incoming variability signals from their customers, which results in the transmission of the bullwhip effect from the customer to the supplier.
14 Page 14 of 54
In order to overcome this limitation and evaluate firm-level supply chains, we create and study matched dyadic relationships between buyers and suppliers by utilizing publically available secondary data between the years 1990 to 2008. Since 1976, publically traded firms in
FAS 131 (1997) and FAS 14 (1976).
ip t
the U.S. have been required to report their major customers under the accounting regulations Major customers are defined by these accounting As a result, researchers
cr
regulations as customers buying at least 10% of the selling firm’s sales.
us
are starting to utilize the customer segment data located within the CompuStat database to examine supply chain dyads (see for example Hertzel et al. 2008). Using this data, it is possible
an
to identify matched buyer-supplier dyads. However, the form in which the data is reported requires that the data must first be systematically cleaned and validated before any dyads can be
M
identified. It entails significant effort, which is not unexpected in studies focused at empirically
d
isolating phenomenon of interest at the firm level in supply chains.
te
The data in the Compustat segment files related to major customer relationships is selfreported. Customers are commonly listed as government agencies, foreign corporations, or
Ac ce p
privately held corporations. In many cases, even though the major customer is identified, the supplier indicates the sales to that customer as $0. The inability to identify the magnitude of the relationship makes those particular observations unusable.
Likewise, firms commonly disguise
the identities of their large customers, so instead of identifying customers by name, they are simply identified as “Customer A” or “Top 5 customers.” Those relationships are unusable as well, and must be systematically discarded. Because there is no standard method for firms to report the names of their customers, the same customer can be identified with slightly different abbreviations or spellings. For instance, Wal-Mart Stores Inc. might be identified as Wal-Mart, Wal-Mart Corp, Walmart, or WMT in the
15 Page 15 of 54
database. Therefore, unless these discrepancies are resolved, much of the customer relationship data is unusable.
Hertzel et al. (2008) utilized a text-matching program which helped them to
correctly consolidate the customer abbreviations provided by suppliers. While such programs
ip t
can aid researchers in narrowing down the potential matches, the number and complexity of the customer abbreviations also make it likely that the text matching programs would likely discard a
cr
number of potentially usable relationships. Therefore, in order to preserve as many relationships
data by hand instead of using text-matching programs.
us
and real world supply chain dyads as possible, we opted for the tedious process of cleaning the
an
Further refining is necessary because of the existence of multiple types of buyer-supplier relationships. For instance, the supplier may be providing equipment, raw material, services
M
(consulting, financial etc), supporting goods (such as software), logistics support, contract
d
manufacturing, or energy. Furthermore, the buying firms commonly hold distinct roles in the
te
supply chain, such as retailer, wholesaler, manufacturer, service provider etc. While previous research has utilized all types of supply chains (Lanier et al. 2010), it is desirable from a control
Ac ce p
perspective that the data is somewhat homogeneous in nature. Hence we opted to include only manufacturing firms. Finally, as per the recommendations of Bray and Mendelson (2012), we limited our sample to only those firms that had positive revenues and positive inventory value.
5.1 Bullwhip Transmission
Once buyer-suppler dyads are identified, in order to accurately evaluate bullwhip across these firms, it is necessary to ensure that (1) any production variance occurring at the supplier firm is in response to their actual customer’s demand variance and (2) customer demand and production variance is with respect to the inputs that the supplier provides. While other dyadic research (Lanier et al. 2010) has suggested that only the largest or strongest buyer-supplier relationship 16 Page 16 of 54
should be evaluated, this recommendation becomes tenuous since it is unlikely that a supplier’s largest customer possesses bullwhip characteristics that are representative of the supplier’s entire customer base. Furthermore, unless a firm’s largest customer is essentially a single customer
ip t
(representing 100% of the supplier’s sales), and that same supplier provides 100% of the customer input materials; it becomes difficult to suggest that full transmission occurs whereby
cr
the supplier’s amplification is a result of or in response to this single customer’s demand
us
amplification characteristics. Thus unlike the Beer game, in real-world supply chains there is no assumed one-to-one supply chain relationship guaranteeing full bullwhip transfer between a
an
customer and a supplier. While it is important to capture the inter-firm effects, it is equally important to ensure that we are also capturing the transfer of bullwhip between customer and
M
supplier. So in order to overcome these challenges, we have taken several safeguards that make
d
it more likely that true bullwhip transmission is indeed being captured within our dataset.
te
First, we limit our sample firms to only manufacturing (SIC 20-39) suppliers and customers and exclude retailers because in the context of firm level bullwhip, what is unique
Ac ce p
about retailers is that their demand variation could be a result of any of the items sold. For example, Wal-Mart’s demand variation could be a result of selling milk, tires or shirts. Clearly, a single supplier would not supply all of these items. If we limit our sample to manufacturing firms only, it is much more likely that the demand variance at the customer is related to a product for which our identified supplier provides an input. Second, instead of evaluating only single buyer-supplier dyads, which in most cases would only capture a small portion of the total sales for the supplying firm, we choose to aggregate across our identified buyer-supplier dyads, generating customer-base dyads for each supplier. Doing so not only accurately reflects the bullwhip behavior of the supplier, but also captures the inter-firm dynamics. In order to balance
17 Page 17 of 54
the contrasting desires of possessing significant statistical power while also capturing the highest proportion of sales accounted for in the identified customer base, we choose to only evaluate those firms whose identified customer-base accounted for at least 50% of the supplying firm’s
ip t
revenues. Our represented customer bases accounted for 69.1% of their supplier’s total revenues on an average, and therefore captured the dominant behavior of the firm’s entire customer base.
cr
Finally, we limited our sample to “major” suppliers only. Hertzel et al. (2008) define major
us
suppliers as providing at least 1% of their customer’s cost of goods sold (COGS). The average supplier accounted for 6.2% of their customer-base’s COGS in our sample, suggesting that any
an
customer demand/production variance is related to products for which this major supplier has provided inputs. Cumulatively these safeguards allow us to examine a real-world dataset that
M
reflects the behavior of actual supply chains.
d
5.2 Bullwhip Measurement
te
While we have already outlined our assessment of first and second order bullwhip effects in Equations 1 and 2, this section provides additional details as to how we applied these
Ac ce p
measurements to our study. A common measure used by researchers to operationalize the bullwhip effect is the difference between upstream production variance to downstream demand variance (Bray and Mendelson, 2012; Cachon et al. 2007). Because our dataset contains firms of significantly differing sizes, the difference of downstream and upstream variance could show amplification simply because of size differences. Therefore, in line with previous bullwhip research, we accommodate size differences among the firms in a given supply chain by evaluating the coefficient of variation rather than the variance itself (Fransoo and Wouters, 2000; Dejonckheere et al., 2004). Additionally, since we utilize publically available financial data, we use a proxy for firm production by utilizing the method of Cachon et al. (2007) outlined in 18 Page 18 of 54
Equation 3. Production for firm f in quarter t is the cost of goods sold plus any changes in the inventory level of the firm quarter-over-quarter, where Cft and Ift are the cost of goods sold and inventory respectively for firm f in quarter t. (3)
ip t
Pft = Cft + (Ift – Ift -1)
cr
The sales and demand figures are adjusted for profit margin in order to more accurately reflect actual production and demand numbers (Cachon et al., 2007). This measure is intended to
us
broadly capture production in dollars rather than in units, while cost of goods sold is intended to
an
capture demand. If firms draw from inventory, then production will be lower than demand, while if firms are building inventory, production will exceed demand.
M
Since the production and demand data are quarterly snapshots, the coefficient of variation is calculated across all 4 quarters in the calendar year. In line with Cachon et al. (2007), we do
d
not de-seasonalize our data because firms must meet actual demand and not the de-seasonalized
te
demand. Given the Chen and Lee (2012) results, which suggest that time aggregation hides
Ac ce p
variance amplification; this approach represents a more conservative evaluation of amplification. Finally, since we aggregate across customer-supplier dyads to generate customer basesupplier dyads, it is necessary derive customer base measures that can be evaluated in place of the single customer measures shown in Equations 1 and 2. To do so, we follow a weighted average approach to compute customer base measures as shown in Equation set 4, where the weighted average customer base measure c for supplier-year observation s; number of identified firms in the customer base for supplier i; customer base firm j;
is is the
is the weight given to
is the percentage of total sales purchased from supplier i by customer
19 Page 19 of 54
base firm j; and
is the variable of interest for customer base firm j. All variables are from
the same time period.
ip t
(4)
us
cr
given that
an
In line with other empirical bullwhip studies such as Bray and Mendelson (2012), we assume that the amplification identified at the customer base through utilizing firm level data is
M
not systematically different than what would have occurred at the product level. The final sample thus obtained consists of 383 supplier-customer base dyadic observations from 147
te
supplier industries.
d
unique supplier firms. Table 2 provides details of the sample at it relates to different years and
-----Insert Table 2 about here-----
Ac ce p
5.3. Performance Variables
This study evaluates the impact of first and second order bullwhip effects on supplier performance. Return on Assets or ROA (Net Income/Total Assets) is a traditional measure of a firm’s financial performance. Because ROA is an aggregated measure, we choose to further “decompose” ROA into the profit and asset side of the ratio (Swink and Jacobs, 2012) in order to gain additional insights into the performance pathways by which bullwhip may ultimately impact ROA. On the income side, we evaluate Operating Margin ((Sales-COGS)/Sales) (Swink and Jacobs, 2012) and SG&A Expense (SGA Costs/Sales) (Dehning, Richardson, and Zmud, 2007). On the asset side of ROA, we evaluate Fixed Asset Turnover (Sales/Average Fixed Asset Value) (Tang and Liou, 2010) and Inventory Turnover (Sales/Average Inventory Value) (Swink and 20 Page 20 of 54
Jacobs, 2012). As per the recommendations of Dehning, Richardson, and Zmud (2007), all the dependent performance variables were Winsorized at the 99% and 1% level. 5.4. Control Variables
ip t
Given that we are evaluating a number of different types of performance outcomes, a number of common performance-related control variables were included in the analysis to ensure that any
cr
identified relationships were not spurious (Dehning, Richardson and Zmud, 2007; Hendricks,
us
Singhal and Zhang, 2009). Related to the supplying firm, these control variables include Supplier industry (2-digit SIC), Firm Size (Natural Log of Total Assets) and Market Share (Firm
an
Sales/Total Industry Sales 4-digit SIC). Customer-base controls include Firm Size and Market Share. Both customer-base control variables were calculated utilizing the weighted average
M
methodology provided in Equation 4. We elected to control for both firm size and market share
d
in order to fully account for any power dynamics in the customer-supplier relationship that could
te
potentially influence the performance outcomes of the supplier. Finally, time was controlled for by utilizing a continuous variable, with zero equating to the first year of the data. Correlations
Ac ce p
and descriptive statistics for all the performance and control variables are provided in Table 3. Additionally, a summary of how our empirical sample divides into the four quadrants related to the first and second order bullwhip effects is shown in Table 4. Approximately 62.4% of the firms in our sample show an amplifying first order bullwhip effect, and only 60.3% of the firms show an accelerating second order bullwhip.
These numbers confirm that the traditional
bullwhip pattern where demand variance amplification progressively increases as it travels upstream in the supply chain is common, but certainly not universal, and also in line with the findings of other researchers as well. Equally interesting is the large proportion of firms that exhibit dampening and decelerating patterns.
21 Page 21 of 54
-----Insert Tables 3 and 4 about here-----
6. Analysis
ip t
The primary purpose of this study is to evaluate the nature of the relationships between first and second order bullwhip effects and supplier performance. To accomplish this, we utilized OLS
cr
regression techniques coupled with Seemingly Unrelated Estimation (SUE).
Given that
us
quadratic effects have been hypothesized for both first and second order bullwhip, these are tested by evaluating Equation 5, where Z is the performance variable, X is the first order
an
bullwhip effect, and Y is the second order bullwhip effect.
(5)
M
Z= β0 + β1X + β2Y + β3X2 + β4Y2 + ε
Equation 5 is specified such that that the first and second order bullwhip effects are
d
treated as additive. However, it is possible that the first and second order effects actually have a
te
multiplicative relationship with each other. Given that we have no a priori indication as to the
Ac ce p
nature of the joint relationship between first and second order bullwhip, we propose to further explore this possibility with the method outlined by Cortina (1994) by examining Equation 6, where Z is the performance variable, X is the first order bullwhip effect, and Y is the second order bullwhip effect.
Z= β0 + β1X + β2Y + β3X2 + β4Y2 + β5XY +ε
(6)
Even though it is possible that there is a non-linear interaction between first and second order bullwhip, Cortina (1994) suggests that evaluating linear interactions can provide evidence of a non-additive relationship. Furthermore, by only evaluating a linear interaction, we are able to avoid severe multicollinearity problems associated with testing non-linear interactions of nonlinear effects (e.g. quadratic-quadratic interaction). We will conclude that the first and second 22 Page 22 of 54
order bullwhip relationship is multiplicative rather than additive if the interaction term is significant, and if the Equation 6 model fits the data better than the Equation 5 model. Our study evaluates five interrelated aspects of performance. As such, it is necessary to
ip t
simultaneously estimate all models via SUE, which controls for any covariance that may exist between regressions of bullwhip-performance outcomes. Additionally, since it is possible that
cr
the same supplying firm is in the data multiple times across the sample years, our data cannot be
us
considered independent. Therefore, we utilize two-way cluster robust errors (Peterson 2009, Cameron et al. 2011; Thompson, 2011) which cluster around the both the supplying firm as well
an
as time. This approach allows the errors to correlate within a supplying firm and year, but not across supplying firms or years. As such the independence assumption underlying OLS
M
regression is satisfied.
d
Equation 7 shows the procedure developed by Thompson (2011) that we used to calculate ,
and
are the estimated
te
our two-way clustered variance estimator, where
Ac ce p
variances of supplier firm, estimated variances of time, and the common heteroskedasticityrobust OLS variance (White, 1980) respectively. (7)
Since each customer base is idiosyncratic with respect to the specific supplying firm, we assume that there is no bias across customer base observations. We therefore do not control for specific customer firms or industries. Multicollinearity was not deemed to be a significant problem because all Variance Inflation Factors (VIFs) were less than 7.5, well below the common threshold of 10 (Visnjic and Van Looy, 2013).
No significant correlations were
identified (α=0.1) between independent variables and residuals from each equation, thereby
23 Page 23 of 54
suggesting that endogeneity is not a concern in this study. The first and the second-order bullwhip variables were standardized to aid in the interpretation of results.
ip t
7.0 Results The reported results and figures are from the best fitting and most parsimonious model between
cr
analyzing Equations 5 and 6. Table 5 details the results of our analysis with respect to ROA. First order bullwhip has a significant (α<0.05) negative linear relationship with a positive
us
quadratic trend with ROA (-0.08X +0.06X2). While H1 suggests that performance would peak
an
as first order converges to zero, as shown in Figures 1a and 1b, ROA is actually near its lowest level when first order bullwhip is near zero. Additionally, the highest level of ROA occurs when
M
first order bullwhip is severely dampened. Therefore, H1 is not supported with respect to ROA performance.
d
Examining H2 shows that the behavior of first and second order bullwhip effects is
te
almost opposite regarding ROA. Second order bullwhip has a significant (α<0.05) negative
Ac ce p
quadratic relationship with ROA (-0.06X2). As such, we fully support H2 that peak ROA performance occurs as second order bullwhip converges to zero.
Finally, while not
hypothesized, we explored the nature of the joint relationship between first and second order bullwhip and ROA performance. We do find evidence of a significant positive multiplicative relationship, suggesting synergistic effects between first and second order bullwhip effects relating to ROA performance. Taken together, the combined interaction and quadratic effects of first and second order bullwhip generate a “saddle” shaped relationship, where the highest ROA occurs when the first order bullwhip is greatly smoothed or dampened and second order bullwhip nears zero. The lowest ROA occurs in the opposite condition where the first order is greatly amplifying and the second order is divergent from zero. 24 Page 24 of 54
-----Insert Table 5 and Figures 1a and 1b about here-----
7.1 Profit Performance
ip t
Results related to Operating Margin and SGA Expenses are shown in Table 6 and Figures 2 and
cr
3. These results further help to explicate how first and second order bullwhip impacts ROA. While the first order bullwhip is shown to have no significant relationship with Operating
us
Margin, it does have a significant (α<0.05) positive linear relationship with a negative quadratic trend (0.07X-0.06X2) with SGA Expense. Given that H1 suggests peak performance will occur
an
as the first order bullwhip converges to zero, H1 is not supported for either Operating Margin or
M
SGA Expense.
The second order bullwhip is shown to have a significant positive linear relationship with
d
a negative quadratic trend (0.12X-0.10X2) with Operating Margin and a negative quadratic (-
te
0.03X2) with SGA Expense, thus supporting H2 for Operating Margin and failing to support H2 regarding SGA Expense. Exploring the nature of the joint relationship between the first and the
Ac ce p
second order bullwhip suggests that for both the Operating Margin and SGA Expense, the first and the second order bullwhip effects are additive rather than multiplicative. -----Insert Table 6 and Figures 2a, 2b, 3a, 3b about here----7.2 Asset Performance When evaluating the impact of the first order bullwhip on asset performance, the results shown in Table 7 indicate a significant (α<0.05) positive linear relationship with a negative quadratic trend (0.27X-0.19X2) for fixed asset turnover and a significant (α<0.05) negative linear relationship (-1.82X) for inventory turnover. Both results provide partial support of H1. These results are shown graphically in Figures 4 and 5. In contrast, second order bullwhip is shown to have a significant negative linear relationship with (-0.17x) with fixed asset turnover and no 25 Page 25 of 54
relationship with inventory turnover. As such, H2 is partially supported with respect to Fixed Asset Turnover and not supported with respect to Inventory Turnover. Finally, based on our
these aspects of asset performance is additive rather than multiplicative.
ip t
analysis, the nature of the joint relationship between the first and second order bullwhip and
cr
-----Insert Table 7 and Figures 4 and 5 about here----7.3 Robustness Checks
To ensure that our results were robust and not influenced by our chosen method and clustered
us
error structure, we also conducted the analysis utilizing a variety of differing methods and
an
clustering options. These included (1) double clustering on supplier and year, but not utilizing SUE, (2) single clustering on supplier only with SUE, and (3) single clustering on supplier In all cases, the pattern of significant results did not change. Therefore we
M
without SUE.
8.0 Discussion
d
conclude that our results are robust with respect to the chosen method of analysis.
te
The primary motivation for this study was to empirically investigate the bullwhip-performance
Ac ce p
relationship. By doing so we hoped to uncover evidence as to the various performance pathways whereby bullwhip ultimately impacts supplier performance. However, to fully evaluate bullwhip in a supply chain context, demand distortion as well as variance amplification should be examined.
Therefore, in addition to testing the traditional measure of bullwhip demand
distortion (first order bullwhip), we also proposed and tested a measure of variance amplification (second order bullwhip). Throughout this section, we discuss the key findings of this study related to (1) providing descriptive evidence with respect to the preponderance of first and second order bullwhip within a supply chain, (2) explicating potential performance pathways whereby bullwhip impacts performance and (3) identifying potential managerial levers and control mechanisms that may be utilized to influence the bullwhip-performance relationship. 26 Page 26 of 54
The primary insights and key findings stemming from this study are further discussed in the following sub-sections and also highlighted in Table 8. -----Insert Table 8 about here-----
ip t
8.1 How Common is Bullwhip?
The configuration of our sample suggests that what is viewed as traditional bullwhip behavior
cr
(increasingly amplifying demand distortion) is not a universal occurrence. We found that with
us
respect to first order bullwhip, 62.4% of our sample amplified while 37.6% dampened. Our results are consistent with the recent empirical work of Bray and Mendelson (2012) and Shan et
an
al. (2013) that approximately two-thirds of firms exhibit bullwhip. Thus including our study, three separate sets of researchers have triangulated and independently shown that 60% to 67% of
M
firms amplify. Additionally, within our sample, approximately 60.3% have an accelerating
d
second order bullwhip effect, while 39.7% have a decelerating second order effect. When
te
viewing the first and second order effects jointly, about half the sample (50.1%) falls into what is considered to be the traditional bullwhip condition (amplifying first order and accelerating
Ac ce p
second order), thus suggesting that while traditional bullwhip occurs commonly, it is clearly not a universal occurrence.
When evaluating the dispersion of observations across the four
quadrants in Table 4, it is clear that there is no universal or dominant strategy with respect to combinations of first and second order bullwhip. The implication is that firms may be actively managing variability in line with their chosen production strategy and specific incentives, without a prevailing smoothing or amplifying bias. 8.2 Bullwhip Performance Pathways A key contribution of this study is in evaluating the impact of bullwhip on not only ROA but also its key components. By doing so, we are able to identify more precisely the performance pathways bullwhip takes (income and/or assets ) to impact ROA. This study provides evidence 27 Page 27 of 54
that while both the first and second-order bullwhip effects ultimately impact supplier ROA, each does so through distinctly different pathways. Thus, while our results do support conventional wisdom that bullwhip can negatively impact supplier performance; we provide evidence that
ip t
suggests that the bullwhip-performance relationship is far more complicated than what is commonly perceived. The following sub-sections detail more specifically insights gleaned with
cr
respect to the differing performance pathways of both first and second order bullwhip.
us
8.2.1 Bullwhip-ROA Relationship
Holding second order bullwhip constant at zero, a dampened (<0) first order bullwhip yields
an
positive supplier ROA, while an amplified (>0) first order bullwhip generally yields negative ROA. Counter intuitively, these first order bullwhip results suggest that there is virtually no
M
ROA downside to dampening, even to extreme levels. The marginal costs of utilizing bullwhip
d
control mechanisms appear to be more than offset by performance benefits of smoothing
te
production. Additionally, counter to our proposed hypothesis, supplier ROA is near its lowest and most negative level when first order bullwhip is near zero. This interesting and also
Ac ce p
counterintuitive finding suggests, ceteris paribus, that it may not be in the best financial interests of upstream firms to be first order bullwhip neutral (i.e. first order bullwhip of zero). When examining second order bullwhip, holding first order constant at zero, we find almost the exact opposite relative relationship. Supplier ROA is highest, albeit negative, when second order is near zero. Thus we find evidence that with respect to ROA, suppliers benefit from coordinating with their immediate customers, but not with entities further downstream. In fact, we find that the highest ROA for a supplier occurs when there is “coordinated dampening” of first order bullwhip (first order <0 and second order =0). Another interesting finding is that the lowest ROA for a supplier occurs when first order bullwhip is extremely amplified and
28 Page 28 of 54
second order is extremely decelerated (first order >0 and second order <0). Such a bullwhip combination suggests that while the supplier is severely amplifying, they are nevertheless trying to “tame” the bullwhip passed to them by their customer by amplifying less than their customer
ip t
did. While an extreme amplification of first order bullwhip generally produces negative ROA, it is far worse for the supplier to do so by either accelerating or decelerating the first order
cr
bullwhip of their customer.
us
8.2.2. Bullwhip-Income Performance Relationship
Turning our attention to the income side of the ROA decomposition analysis, we find that
an
Operating Margin (OM) is only impacted by a large quadratic second order effect, while SGA expense is impacted by both first and second order quadratic effects. The finding that OM is not
bullwhip
primarily
impacts
performance
through
generating
operational
d
traditional
M
impacted by first order bullwhip is very surprising; given the conventional wisdom that
te
inefficiencies. Instead, our results suggest that supplier operating margin is directly influenced by accelerating or decelerating the bullwhip effect from their customers (second order), rather
Ac ce p
than through their own distortion of end customer demand (first order). It is important to note that Operating Margin, which is calculated from sales and COGS, only captures direct labor costs and material purchases related to what was produced in that period. It therefore reflects only the effects of overproduction, purchasing excessive amounts of new material, or using expensive short-term capacity options. These practices are only likely to occur as a result of responding to the behavior of their customer’s production rather than to further downstream demand signals. It is interesting to note that our results also suggest that firms have largely adapted their operations to their given level of first order bullwhip, such that any OM impacts are
29 Page 29 of 54
primarily due to the acceleration (deceleration) of bullwhip relative to their customers rather than due to the absolute level of the supplier’s first order bullwhip. Focusing on SGA Expenses, we show that both first and second order bullwhip produce
ip t
quadratic effects where the highest SGA Expenses occur as each measure approaches zero. When both the first and second order bullwhip is zero, the supplier and customer both perfectly
cr
match the variability of end customer demand suggesting a condition of full supply chain
us
coordination. Coordinating production across supply chain partners is often very expensive (Schoenherr and Swink, 2012), especially in the context of B2B manufacturing relationships
an
where salespeople are critical to maintaining buyer-supplier relationships and for gathering the necessary information visibility required to coordinate production (Lynch and de Chernatony, Thus it is not unexpected that SGA Expense increases as additional resources are
M
2007).
d
required to gather production information from customers along with end customer demand
te
information from even further downstream.
8.2.3. Bullwhip-Asset Performance Relationship
Ac ce p
The performance impacts of bullwhip have often been controlled through engaging capacity or inventory buffers. Therefore, in evaluating the asset side of ROA, it is possible to gain insights on the extent to which such control mechanisms are related to the bullwhip-performance relationship. However, before discussing the asset-based implications of bullwhip, it is first necessary to understand how firms account for such costs. Capacity buffers, often indicated by a low level of fixed asset turnover (Hendricks, Singhal and Zhang, 2009), have ongoing costs that are typically represented as opportunity costs rather than as increased operating costs. As such, any costs of underutilized production capacity are not likely to be captured in OM via COGS, but will only serve to increase the asset base of ROA. While inventory purchase costs (building
30 Page 30 of 54
inventory) are included in COGS, any subsequent inventory holding costs (costs of holding existing inventory) are typically included in Operating Expense. As such, costs associated with holding extra inventory directly impacts net income rather than OM, while costs associated with
sales constant).
ip t
increasing inventory levels directly translate into higher COGS thereby lowering OM (holding When evaluating the bullwhip-asset relationship, we find starkly differing
cr
results from those on the income side of ROA. First order bullwhip has significant relationships
us
with both fixed asset turnover and inventory turnover. However, counterintuitively when first order bullwhip is extremely dampened, inventory turnover is at its highest and fixed asset
an
turnover is at its lowest; indicating a relatively small inventory buffer and a relatively large capacity buffer. This result suggests that, counter to conventional wisdom, firms appear to
M
dampen first order bullwhip primarily through means other than through inventory-based control
d
mechanisms.
te
Conversely, firms that are experiencing amplified first order bullwhip possess higher relative inventory levels and generally lower capacity buffers.
However, we do find that
Ac ce p
capacity buffers begin to increase at very high levels of first order amplification, indicating that firms start to employ both inventory and capacity buffers at extremely high levels of amplification. These findings suggest that increased inventory levels are not automatically capable of or required for smoothing production or dampening first order bullwhip effects. High inventory levels that accompany high first order bullwhip levels may be symptomatic of previous supply and demand mismatches that have yielded excess inventories of items that while in stock are currently not demanded, and are thus ineffective in smoothing current production. Interestingly, while first order bullwhip was found to impact fixed asset turnover and inventory turnover, these operationally oriented effects are not significant enough to impact OM.
31 Page 31 of 54
While amplifying (dampening) firms tend to hold higher (lower) levels of inventory, their linkage to a non-significant OM result is less clear. For example, holding sales constant, OM would not be impacted by higher (lower) levels of inventory unless the level of inventory was Alternatively, since we have no way to
ip t
growing (shrinking) thereby impacting COGS.
determine the extent of lost sales, it is possible that increased COGS due to growing inventory
cr
levels are offset by increased revenues from the firm having fewer lost sales. However, since
us
first order bullwhip is shown to ultimately have no relationship with OM, the reduction of lost sales or growth of inventory is not significant enough to impact OM.
Conversely, while second
an
order bullwhip does impact OM, it was shown to have no relationship with inventory and only a small effect on fixed asset turnover, thus suggesting that second order bullwhip impacts
M
operational performance through means other than through asset or inventory performance.
d
Taken together, these results indicate that utilizing capacity and/or inventory bullwhip control
te
mechanisms ultimately has little impact on operating performance (OM). 8.2.4 Summative thoughts on Bullwhip-Performance Pathways
Ac ce p
Regarding first order bullwhip, while the capacity and inventory levels are likely to be generally offsetting in terms of increasing (decreasing) the size of the asset base by which ROA is calculated, lower inventory levels have an additional positive impact on net income as holding costs are also reduced. However, unlike inventory buffers, when capacity buffers are reduced, they mainly only impact the size of the asset base since unused capacity is typically considered as an opportunity cost. When coupled with the impact of SGA expenses, first order bullwhip approaches its highest ROA levels as SGA expenses decline and inventory holding costs are reduced due to holding a lower level of inventory. Thus it appears that first order bullwhip
32 Page 32 of 54
impacts ROA primarily through inventory and SGA expenses rather than through directly impacting operational performance or influencing the size of the firm’s asset base. Second order bullwhip effects however, appears to impact ROA primarily through OM
ip t
and lower inventory holding expenses. While we do find that second order effects are linked to increased SGA expense as the second order approaches zero, these increased costs simply act to
The absence of a relationship between second order bullwhip and inventory
us
relationship.
cr
shift the second order-ROA relationship curve lower rather than change the shape of the
turnover, and only a small relationship with fixed asset turnover, further suggests that OM is
an
impacted by second order bullwhip primarily through influencing the level of revenue (e.g. more or less lost sales) or through impacting the non-inventory aspects of COGS rather than through
M
impacting the firm’s asset base.
d
8.3 Supply Chain Production Coordination as Managerial Lever of Bullwhip
te
Bullwhip is inherently intertwined with the concept of supply chain coordination. If all supply chain partners perfectly coordinated their production not only with each other but also with end
Ac ce p
customer demand, bullwhip would not occur and all production variability would be equal to demand variability. Our two measures of bullwhip provide some insights as to the extent to which production coordination along the supply chain informs the bullwhip-performance relationship. As either first or second order bullwhip measures diverge from zero, lower levels of coordination between the supplier and their downstream supply chain partners is indicated. Based on our results, supplier production coordination seems to be most relevant with respect to ROA, OM and SGA performance outcomes. As suggested previously, coordination across supply chain partners can be very expensive, therefore it is not unexpected that the highest SGA costs would occur when full downstream coordination is indicated (both first and second
33 Page 33 of 54
order bullwhip effects approaching zero). When evaluating OM, additional interesting insights are gleaned. For example, OM performance is at its highest level when second order bullwhip approaches zero, indicating that the supplier is coordinating with its customer. Given that OM is
ip t
sensitive to both revenue and direct costs (COGS), this result suggests that when the supplier and customer are coordinating their production, few lost sales occur and the supplier is not forced to
cr
utilize expense short-term capacity adjustment options, such as overtime or lay-offs.
us
Additionally, given the costs associated with coordination, this result suggests that such additional coordination costs are more than offset by the benefits of avoiding lost sales and short
an
term capacity alternatives. However, we find no significant relationship between OM and first order bullwhip, suggesting that with respect to OM, there are few benefits of the supplier
M
coordinating beyond the production of their immediate customers.
d
Finally, while first order effects generally indicate if ROA will be positive or negative;
te
firms nevertheless can benefit significantly by optimizing second order bullwhip so as to generate the highest possible ROA for a given level of first order bullwhip. For example, the
Ac ce p
highest positive ROA occurs when suppliers are extreme dampeners of their first order bullwhip and also highly coordinated with their customers. Likewise, at high levels of first order bullwhip where ROA is generally negative, a high level of coordination can even result in slightly positive ROA. Such a finding suggests that second order effects can be seen as a managerial lever that can potentially magnify positive ROA first order effects, or limit the damage from negative ROA first order effects. Therefore, it becomes critical to evaluate both the first and second order effects together rather than separately because otherwise one could erroneously conclude that smoothing always yields a highly positive ROA outcome if only the first order bullwhip is evaluated. However, if
34 Page 34 of 54
the supplying firm is smoothing, but much less so than their customers, then their ROA would actually only approach the breakeven point. Additionally, relying only upon first order results would suggest that extreme amplification would yield negative results. When evaluated together
ip t
with second order effects, it is clear that if the supplier is fully coordinating with their customers, amplifying first order effects can result in breakeven or even slightly positive ROA outcomes.
cr
This finding is surprising given that conventional wisdom suggests that firms should benefit
us
from trying to “tame” the bullwhip effects by decelerating an amplifying first order bullwhip effect; however, we find the opposite actually occurs. These insights suggest that supplying
an
firms are always better off working in tandem/coordinating with their immediate customers. However such benefits should not suggest that suppliers should blindly follow their customer’s
M
bullwhip strategy since the general direction (positive or negative) of supplier’s ROA is largely
d
influenced by their own first order bullwhip behavior rather than through the level of
9.0 Conclusion
te
coordination via second order bullwhip.
Ac ce p
The bullwhip effect has been intensely researched since the 1950’s, and its impact on firm performance is generally accepted as resulting in significant negative supplier performance. However, despite this rich research history, and seemingly universal understanding of its performance implications, the results of this study suggest that the performance pathways whereby bullwhip impacts supplier performance are far more complex and nuanced than previously described in extant literature.
Additionally, our evaluation of both the demand
distortion (first order bullwhip) and rate of amplification (second order bullwhip) provide a more granular empirical examination of bullwhip than previously seen. Despite its long and rich
35 Page 35 of 54
history, we suggest that there remains much to understand about how the ubiquitous bullwhip effect actually impacts firm performance. When building upon our work to further the above-mentioned goal, it is important to
ip t
recognize that to our knowledge, this is the first study to empirically show the existence of a second order bullwhip effect, and to also empirically evaluate the performance implications of
cr
both the first and second order bullwhip effects. Apart from its theoretical contributions, our
us
results provide several insights and recommendations that are relevant to practicing managers. 9.1 Implications for Bullwhip Theory Building
an
This study helps to advance the understanding of not only the extent and pathways whereby bullwhip may impact performance, but also does so by evaluating more specific bullwhip
M
dimensions than has been done before. While our results do generally confirm the conventional
d
wisdom that higher levels of first order bullwhip results in lower ROA performance, we are also
te
able to provide additional insights suggesting that such a negative relationship is driven mainly through increased assets, rather than reduced operating margins.
This result is counter to
Ac ce p
conventional wisdom that poor performance related to bullwhip is mainly due to poor operational performance.
Additionally, in this study we have sought to further clarify the bullwhip-performance relationship by introducing and testing second order bullwhip. The evaluation of second order bullwhip provides additional theoretical depth to this relationship, by showing that not only does the level of amplification matter, but so does the rate of bullwhip acceleration (deceleration). When examining the performance implications of both first and second order bullwhip effects, a complex pattern of results emerges. This high level of complexity indicates that the performance pathways whereby bullwhip impacts performance are far more nuanced that generally believed,
36 Page 36 of 54
suggesting great need for further testing and theory development efforts regarding this relationship. Taken together, we suggest that the bullwhip-performance relationship is deserving of additional research that can further drive the theory development efforts illuminating this
ip t
important relationship. 9.1.1 Implications for ETPS and OIPT
cr
Although the bullwhip effect is heavily researched, much of this deep body of research has
us
focused on its causes and potential remedies. Thus far, anecdotal evidence, case studies and behavioral simulations have served as the primary source of theory building related to the
an
bullwhip-performance relationship. Theorizing based upon empirically grounded analysis, large scale or otherwise, has been conspicuously absent. Leveraging the perspectives of ETPS and
M
OIPT, this study seeks to initiate a more thorough theory development process of the bullwhip-
d
performance relationship that is grounded upon real world empirical data. As such, this study
te
not only seeks to inform us about the generally positive value of ETPS and OIPT in predicting the bullwhip-performance relationship, but it also provides granular evidence that may be
Ac ce p
utilized in formulating a unique and specific theoretical perspective of bullwhip. ETPS suggests that performance will diverge from optimal as firm behavior regarding inventory and flexibility also diverges from optimal or profit maximizing. This study provides some supporting evidence that smoothing production generally results in improved ROA. However, we are unable to support the ETPS based prediction that ROA should be lower at extreme levels of smoothing or that extreme smoothing is accomplished through holding high levels of inventory. As such, our ROA results suggest that a boundary condition exists in the applicability of ETPS to the bullwhip-performance relationship, especially at extreme levels of
37 Page 37 of 54
dampening. Further research is necessary to more fully explicate the means and methods which firm’s employ such that extreme dampening seemingly has no performance downside. Regarding OIPT, our results indicate that ROA always improves as production is more
ip t
highly coordinated (second order = 0). In fact, even when first order bullwhip is severely amplifying, and ROA is typically negative, the supplying firm’s ROA can actually be slightly
cr
positive if customer-supplier coordination is high enough. Such a result suggests that the ability
us
of the supplier to reduce their uncertainty and equivocality through coordinating production with their customer can be a powerful managerial lever in enhancing bullwhip related performance.
an
However, evaluation of the ROA decomposition results do not indicate clear support that fully coordinating production decreases inventory or capacity buffers, as suggested by OIPT. For
M
example, counter to the predictions of OIPT, we find no relationship between inventory and
d
second order bullwhip. While OIPT based arguments do predict our ROA results, the exact
te
mechanisms whereby customer-supplier coordination impacts ROA seem not to be those mechanisms (inventory and capacity buffers) suggested within OIPT. Taken together, our results
Ac ce p
indicate that while OIPT is generally applicable in evaluating the second order bullwhip-ROA relationship, a boundary condition seems to exist. Additional research is necessary to identify alternative mechanisms for how the second order bullwhip-ROA relationship is achieved. 9.2 Implications for Practice
First, while it was expected that bullwhip would impact ROA, this study demonstrates that the relationship is highly sensitive to the joint effect of both first and second order bullwhip effect and can often be the difference between positive and negative ROA. Secondly, while both first and second order bullwhip effects impact various types of performance outcomes, managers should recognize that significant dampening of first order bullwhip almost always yields positive
38 Page 38 of 54
ROA, regardless of second order effects. However, supplier’s that engage the managerial lever of coordinating production with their customers always experience higher levels of ROA than when not coordinating. Thus this study expands our current understanding of the bullwhip effect
ip t
by suggesting that both first and second order bullwhip effects should be managed simultaneously in order to achieve the highest levels of ROA. Third, because ROA implications
cr
of first order bullwhip are largely driven through inventory levels and SGA expenses, managers
us
could seemingly insulate their firm’s ROA from first order bullwhip implications if they could control the growth of both SGA and inventory. Finally, these results suggest that managers
an
should not underestimate their own power to influence their firm’s bullwhip-performance relationship via coordination (managerial lever) and bullwhip control mechanisms (inventory and Firms that actively manage the bullwhip-performance relationship by
M
capacity buffers).
d
appropriately using these managerial levers and control mechanisms may be able to gain
te
competitive advantage over their peers who are unable to do so. 9.3 Limitations and Future Research
Ac ce p
As is the case for all empirical research, there are a number of limitations associated with this study. First, we recognize that bullwhip is inherently a product level issue and that decisions related to its control are generally conducted at the plant level rather than the firm level, thus while we are confident that our firm level aggregation approach is conservative given the findings of Chen and Lee (2012), we cannot guarantee that the bullwhip-performance relationship behaves the same at the product/plant level of analysis. While firm performance and bullwhip inherently occur at different theoretical levels (firm versus plant or product), we utilize aggregated measures of bullwhip such that both bullwhip and performance are evaluated at the firm level. Should product level supply chain data become available, we encourage future
39 Page 39 of 54
research to extend our model to the product level. Additionally, we infer that coordination is occurring between firms if first or second order bullwhip effects are zero, meaning the supplier has the same production variance as end customer demand variance (first order) or the supplier’s
ip t
production variance is the same as their customer’s production variance (second order). However, we are unable to measure explicitly if coordination is occurring primarily due to
cr
information sharing, VMI arrangements, or other explicit coordination mechanisms. We do
us
believe however that it is unlikely for supplier and customer production variance to equalize randomly without coordination/communication between firms.
an
Despite some of the limitations that arise naturally in conducting such a complex study with real world data that we assiduously put together, we believe that it both confirms as well as
M
opens up new areas of research. First, the identification of second order bullwhip effect is a
d
prime avenue for future researchers to explore. Such research could include identifying the
te
mechanisms for influencing or controlling second order bullwhip beyond various forms of coordination. In particular, does second orders bullwhip impact performance more severely in
Ac ce p
certain contexts, for example intensely competitive markets? Is the second order bullwhip more or less easy to influence than the first order bullwhip by the dominant player in the supply chain? How does second order bullwhip impact ROA, if it is seemingly not through substituting inventory and capacity for information?
Additionally, while we have primarily evaluated
inventory and capacity dampening mechanisms, it may be worthwhile to explore the impacts of alternative dampening mechanisms. Finally, the empirical finding that there is no universal expression of traditional bullwhip (Table 4), suggests that there may be rational reasons for firms to choose their bullwhip levels. Future research should evaluate what characteristics contribute
40 Page 40 of 54
to firms being better (worse) off by possessing various combinations of first and second order
ip t
bullwhip.
cr
REFERENCES
Ac ce p
te
d
M
an
us
Arrow, K. J., Karlin, S. and Scarf, H. (1958). Studies in the mathematical theory of inventory and production (No. 1). Stanford University Press. Baganha, M., and Cohen, M. 1998. The stabilizing effect of distribution center inventory. Operations Research, 46, 72–83. Blanchard, O. 1983. The production and inventory behavior of the American automobile industry. Journal of Political Economy, 91, 365-400. Blinder, A. 1986. Can the production smoothing model of inventory behavior be saved? Quarterly Journal of Economics, 101, 431-453. Bloom, N. (2013). Fluctuations in uncertainty (No. w19714). National Bureau of Economic Research. Bray, R. L., and Mendelson, H. 2012. Information transmission and the bullwhip effect: An empirical investigation. Management Science, 58(5), 860-875. Cachon, G.P., Randall, T., and Schmidt, G.M. 2007. In search of the bullwhip effect. Manufacturing and Service Operations Management, 9 (4), 457-479. Cameron, A. C., Gelbach, J. B., & Miller, D. L. 2011. Robust inference with multiway clustering. Journal of Business & Economic Statistics, 29(2), 238-249. Chen, L., & Lee, H. L. 2012. Bullwhip effect measurement and its implications. Operations Research, 60(4), 771-784. Chen, F., and Samroengraja, R. 2004. Order volatility and supply chain costs. Operations Research 52(5): 707–722. Cortina, J. M. (1994). Interaction, nonlinearity, and multicollinearity: Implications for multiple regression. Journal of Management, 19(4), 915-922. Daft, R. L., & Lengel, R. H. 1986. Organizational information requirements, media richness and structural design. Management science, 32(5), 554-571. Dejonckheere, J., Disney, S.M., Lambrecht, M.R. and Towill, D.R. 2004. The impact of information enrichment on the Bullwhip effect in supply chains: A control engineering perspective. European Journal of Operations Research, 153 (3), 727-750. Dehning, B., Richardson, V. J., & Zmud, R. W. (2007). The financial performance effects of ITbased supply chain management systems in manufacturing firms. Journal of Operations Management, 25(4), 806-824. Financial Accounting Standards Board. 1997. Statement of Financial Accounting Standards No. 131: Disclosures about Segments of an Enterprise and Related Information. Norwalk, CT. Financial Accounting Standards Board. 1976. Statement of Financial Accounting Standards No. 14: Financial Reporting for Segments of a Business Enterprise. Norwalk, CT. Forrester, J. 1958. Industrial Dynamics—a major breakthrough for decision makers. Harvard Business Review, 36 (4), 37-66 41 Page 41 of 54
Ac ce p
te
d
M
an
us
cr
ip t
Forrester, J. 1961. Industrial Dynamics. MIT Press, Cambridge, MA. Fransoo, J.C. and Wouters, M.J.F. 2000. Measuring the Bullwhip Effect in the supply chain. Supply Chain Management: an international journal, 5, 78-89. Galbraith, J. R. 1973. Designing complex organizations. Boston, MA: Addison-Wesley. Galbraith, J. R. 1974. Organization design: An information processing view. Interfaces, 4(3), 2836. Giard, V., & Sali, M. 2013. The bullwhip effect in supply chains: a study of contingent and incomplete literature. International Journal of Production Research, 51(13), 3880-3893. Goodhue, D. L., Wybo, M. D., & Kirsch, L. J. (1992). The impact of data integration on the costs and benefits of information systems. MiS Quarterly, 293-311. Hendricks, K. B., V. R. Singhal, and Zhang, R. (2009). "The effect of operational slack, diversification, and vertical relatedness on the stock market reaction to supply chain disruptions." Journal of Operations Management 27(3): 233-246. Hertzel, M.G., K. Li, M.S. Officer, and K.J. Rodgers. 2008. Inter-firm linkages and the wealth effects of financial distress along the supply chain. Journal of Financial Economics, 87 (2), 374-387. Holt, C.C., Modigliani, F., Muth, J. and Simon, H. 1960. Planning Production, Inventories and the Work Force, Prentice-Hall, NJ. Koste, L. L., & Malhotra, M. K. (1999). A theoretical framework for analyzing the dimensions of manufacturing flexibility. Journal of operations management, 18(1), 75-93. Lanier Jr, D., W.F. Wempe, and Z.G. Zacharia. 2010. Concentrated Supply Chain Membership and Financial Performance Chain and Firm Level Perspectives. Journal of Operations Management, 28 (1), 1-16. Lee, H.L., Padmanabhan V. and Whang, S. 1997a. Information distortion in a supply chain: The bullwhip effect. Management Science, 43 (4), 546–548. Lee, H. L., Padmanabhan, V., & Whang, S. 1997b. The bullwhip effect in supply chains1. Sloan management review, 38(3), 93-102. Lee, H. L., So, K. C., & Tang, C. S. (2000). The value of information sharing in a two-level supply chain. Management science, 46(5), 626-643. Liker, J. K., & Morgan, J. M. (2006). The Toyota way in services: the case of lean product development. The Academy of Management Perspectives, 20(2), 5-20. Lynch, J., & deChernatony, L. 2007. Winning hearts and minds: business-to-business branding and the role of the salesperson. Journal of Marketing Management, 23(1-2), 123-135. Metters, R. (1997). Quantifying the bullwhip effect in supply chains. Journal of operations management, 15(2), 89-100. Miron, J. and Zeldes, J.S., 1988. Seasonality, cost shocks, and the production smoothing model of inventories. Econometrica, 56, 877-908. Modi, S. B., & Mishra, S. (2011). What drives financial performance–resource efficiency or resource slack?: Evidence from US Based Manufacturing Firms from 1991 to 2006. Journal of Operations Management, 29(3), 254-273. Petersen, M. A. 2009. Estimating standard errors in finance panel data sets: Comparing approaches. Review of Financial Studies, 22(1), 435-480. Schoenherr, T., & Swink, M. 2012. Revisiting the arcs of integration: Cross-validations and extensions. Journal of Operations Management, 30(1), 99-115. Shan, J., Yang, S., Yang, S., & Zhang, J. 2013. An Empirical Study of the Bullwhip Effect in China. Production and Operations Management. 42 Page 42 of 54
Ac ce p
te
d
M
an
us
cr
ip t
Stank, T., Crum, M. and Arango, M. 1999. Benefits of Interfirm Coordination in Food Industry Supply Chains. Journal of Business Logistics, 20 (2), 21–41. Sterman, J. 1989. Modeling managerial behavior: Misperceptions of feedback in a dynamic decision-making experiment. Management Science, 35, 321-339. Swink, M. and B. W. Jacobs (2012). Six Sigma adoption: Operating performance impacts and contextual drivers of success. Journal of Operations Management 30(6): 437-453. Swink, M. and T. Schoenherr (2014). "The Effects of Cross-Functional Integration on Profitability, Process Efficiency, and Asset Productivity." Journal of Business Logistics: DOI: 10.1111/jbl.12070 Tang, Y.-C. and F.-M. Liou (2010). "Does firm performance reveal its own causes? The role of Bayesian inference." Strategic Management Journal 31(1): 39-57. Thompson, S. B. 2011. Simple formulas for standard errors that cluster by both firm and time. Journal of Financial Economics, 99(1), 1-10. Tushman, M. L., & Nadler, D. A. (1978). Information Processing as an Integrating Concept in Organizational Design. Academy of management review, 3(3), 613-624. Visnjic, I., & Van Looy, B. 2013. Servitization: Disentangling the impact of service business model innovation on manufacturing firm performance. Journal of Operations Management. Wang, E. T., Tai, J. C., & Wei, H. L. (2006). A virtual integration theory of improved supplychain performance. Journal of Management Information Systems, 23(2), 41-64. White, H. 1980. A Heteroskedasticity-Consistent Covariance Matrix Estimator and a Direct Test for Heteroskedasticity. Econometrica, 48, 817-838. Zhang, X., & Burke, G. J. 2011. Analysis of compound bullwhip effect causes. European journal of operational research, 210(3), 514-526.
43 Page 43 of 54
Table 1: Numerical Illustration of Bullwhip Calculations using Contrived Data Supplier (Tier 1) §
Known Information
Market Demand Variance=5 Production Variance of Customer = 7
Production Variance of Supplier = 8
First Order Bullwhip*
0.4 = (7-5)/5) (Production Variance of Customer is 40% higher than Market Demand Variance)
0.60 =(8-5)/5) (Production Variance of Supplier is 60% higher than Market Demand Variance)
N/A
0.50= ((8-5)-(7-5))/|7-5| (Supplier first order bullwhip is 50% higher than their Customer's First order Bullwhip indicating that bullwhip is accelerating between customer and supplier
cr
us
Second Order Bullwhip**
ip t
Customer §
an
§ Both customers and suppliers are manufacturers in our study * Positive values indicate amplifying, while negative values indicate dampening ** Positive values indicate accelerating bullwhip,, while negative values indicate decelerating bullwhip
M
Table 2: Data Characteristics
4
1.04
1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 Total
3 10 11 15 15 21 21 17 16 23 26 37 30 28 35 32 26 13 383
0.78 2.61 2.87 3.92 3.92 5.48 5.48 4.44 4.18 6.01 6.79 9.66 7.83 7.31 9.14 8.36 6.79 3.39 100
Ac ce p
te
1990
2b. Industry Distribution of Supplying Firms Supplier Frequency Percent 2-Digit SIC 20 8 2.09 24 1 0.26 25 17 4.44 26 4 1.04 28 62 16.19 29 1 0.26 30 7 1.83 32 1 0.26 33 1 0.26 34 10 2.61 35 44 11.49 36 79 20.63 37 73 19.06 38 69 18.02 39 6 1.57 Total 383 100
d
2a. Yearly Distribution of Quarterly Data Year Frequency Percent
44 Page 44 of 54
1
8.33
49.38
0.26
-0.01
0.20
-0.03
0.12
1
0.19
0.62
0.03
0.05
0.31
1
0.22
0.20
1.92
1.23
Supplier Inventory Turnover Control Variables 8. Supplier Firm Size 9. Supplier Market Share 10. Customer Base Firm Size
ip t 6
7
8
9
10
0.22
0.06
-0.22
0.58
1
0.08
-0.03
0.21
0.17
-0.04
1
11.01
-0.15
-0.04
0.02
-0.11
-0.26
-0.05
1
5.51
1.98
-0.21
-0.09
0.03
-0.01
-0.48
-0.22
0.22
1
0.06
0.14
-0.08
-0.04
0.05
0.02
-0.25
-0.03
0.21
0.38
1
9.08
2.30
-0.18
-0.08
0.05
0.03
-0.46
-0.08
0.18
0.82
0.28
0.10 4.58
0.02 -0.08
0.01 -0.07
0.02 -0.05
0.02 0.06
-0.09 0.03 -0.02 0.002
-0.11 -0.04
0.04 0.23
11. Customer Base Market Share 0.08 12. Time (Years) 10.96 Significant correlations are in Bold
11
12
1
13.54
Ac
7.
5
4.65
ed
Supplier Fixed Asset Turnover
4
1.52
2.
6.
3
us
2
ce pt
Second Order Bullwhip Dependent Variables 3. Supplier ROA 4. Supplier Operating Margin 5. Supplier SGA Expense
1
M an
Independent Variables 1. First Order Bullwhip
S.D.
cr
Table 3: Descriptive Statistics Mean
1
-0.01 0.07 -0.11 0.37
1 0.01
1
45 Page 45 of 54
Decelerating (<0) Accelerating (>0)
First Order Bullwhip Dampening (<0) Amplifying (>0) 105 47 (27.4%) (12.3%)
152 (39.7%)
39 (10.2%)
192 (50.1%)
231 (60.3%)
144 (37.6%)
239 (62.4%)
383 (100%)
cr
Total
Total
ip t
Second Order Bullwhip
Table 4: Distribution of Sample across First and Second Order Bullwhip Effects
us
Table 5: Regression Results for Financial Performance
Equation 6 -0.05
-0.01** -0.01 0.11 0.002 0.08
-0.01** -0.004 0.10 0.003 0.09
Included
Included
Included
-
-0.08***
-0.08***
Second Order Bullwhip (B)
-
0.10**
0.07
A2
-
0.07***
0.06***
-
-0.07*
-0.06**
AxB
-
-
0.03***
Centered-R2
0.05
0.11
0.14
Δ Centered-R2 F
5.70 (14)
0.06 5.02 (18)
ΔF Observations Number of Suppliers
-
0.68 (4)**
0.03 3.11 (19) 1.91 (1)***
B2
M
d
Ac ce p
First Order Bullwhip (A)
-0.003 -0.007 0.10 0.003 0.04
te
Intercept Controls Time Supplier Firm Size Supplier Market Share Customer Base Firm Size Customer Base Market Share Supplier Industry (Dummy Variables) Dependent Variables
Base Model -0.04
an
ROA Equation 5 -0.07
383 147
All dependent variables are standardized. p<0.05*, p<0.01** p<0.001***
46 Page 46 of 54
ip t
Base Model 0.23**
First Order Bullwhip (A)
Base Model 0.23***
Equation 5 0.24***
Equation 6 0.24***
0.002
0.003
0.001
0.001
0.001
-0.01 0.10 0.04 0.03 Included
-0.01 0.11 0.04 0.02 Included
0.002 -0.16** -0.01 -0.14 Included
0.003 -0.15 -0.01 -0.13 Included
0.002 -0.15** -0.01 -0.13 Included
-
-0.02
-0.02
-
0.07*
0.07**
-
0.12*
0.13**
-
0.03
-0.12
-
0.01
0.01
-
-0.06*
-0.06*
-
-0.10**
-0.10**
-
-0.03**
-0.03**
-
-
-0.01
-
-
-0.001
0.10
0.11
0.11
0.49
0.52
0.53
22.16 (14) -
0.01 2.81 (18) 19.35 (4)*** 383 147
0 2.78 (19) 0.03 (1)
25.35 (14) -
0.03 23.40 (18) 1.95 (4)* 340 128
0.0 20.14 (19) 3.26(1)***
0.003 -0.01 0.10 0.04 -0.001 Included
ce pt
Second Order Bullwhip (B) A2 B2
Ac
AxB Δ Centered-R2 F ΔF Observations Number of Suppliers
SGA Expense
Equation 6 0.20***
M an
Time Supplier Firm Size Supplier Market Share Customer Base Firm Size Customer Base Market Share Supplier Industry (Dummy Variables) Dependent Variables
Equation 5 0.21***
ed
Intercept Controls
Centered-R2
us
Operating Margin
cr
Table 6: Regression Results Related to Profit Performance
All dependent variables are standardized. p<0.05*, p<0.01**. p<0.001***
47 Page 47 of 54
ip t
cr
Table 7: Regression Results Related to Asset Performance
First Order Bullwhip (A) A2 B2 AxB Centered-R2
Ac
Δ Centered-R2 F ΔF Observations Number of Suppliers
Inventory Turnover Base Model Equation 5 Equation 6 15.13*** 15.30*** 15.21***
us
0.01 -0.41*** 1.06 0.21*** 0.24
0.01 -0.41*** 1.09 0.21*** 0.24
-0.15 0.39 11.66* 0.78 -14.11***
-0.16 0.34 11.59* 0.72 -13.37***
-0.15 0.31 11.87* 0.72 -13.59***
Included
Included
Included
Included
M an
Included
Included
-
0.27*
0.27*
-
-1.82*
-1.83*
-
-0.17*
-0.12
-
1.06
1.34
-
-0.19*
-0.17
-
1.07
1.18
-
0.07
0.06
-
-1.03
-1.08
-
-
-0.06
-
-
-0.37
0.25
0.26
0.26
0.15
0.16
0.17
5.42 (14) -
0.01 7.43 (18) -2.01 (4) 383 147
0.0 7.74 (19) -0.31 (1)
24.60 (14) -
0.01 17.69 (18) 6.91 (4)*** 383 147
0.0 41.26 (19) -23.57 (1)
ce pt
Second Order Bullwhip (B)
0.01 -0.41*** 1.08 0.21*** 0.35
ed
Intercept Controls Time Supplier Firm Size Supplier Market Share Customer Base Firm Size Customer Base Market Share Supplier Industry (Dummy Variables) Dependent Variables
Fixed Asset Turnover Base Model Equation 5 Equation 6 1.90*** 2.01*** 1.99***
All dependent variables are standardized. p<0.05*, p<0.01**. p<0.001***
48 Page 48 of 54
ip t
Key Insights
1. Dampening yields increasingly positive ROA.
1. ROA declines as second order diverges from zero.
1. Positive or negative ROA mainly determined by first order bullwhip.
2. Amplifying generally results in negative ROA.
2. ROA is always negative.
3. First order bullwhip impacts ROA predominantly through expenses rather than through operational performance or asset levels.
3. Second order bullwhip impacts ROA predominantly through operational performance and SGA expenses rather than through asset levels.
us
Second Order Bullwhip
2. Second order bullwhip can enhance benefits of dampening, and help to overcome detrimental effects of amplifying.
ce pt
No relationship
Operating margin generally declines as second order bullwhip diverges from zero.
SGA Expense
SGA Expenses are near their highest when first order bullwhip is near zero
SGA Expenses are highest when second order bullwhip is near zero.
Inventory Turnover
Inventory levels decrease as first order bullwhip is reduced.
No relationship.
Fixed Asset Turnover
Dampening is associated with much larger capacity buffers than amplifying.
Small negative effect.
Operating Margin
Ac
Asset Side of ROA
Income Side of ROA
ed
ROA
First Order Bullwhip
M an
Evaluated Aspect of Performance
cr
Table 8: Summary of Key Findings
The level of demand distortion does not impact operational performance, but accelerating or decelerating it does.
Negligible net change in asset levels by either first or second order bullwhip.
Directions of Further Research 1. Is there a downside to dampening? 2. Are there benefits to amplifying that are not captured by ROA? 3. Are there conditions under which second order bullwhip effects dominate the first order effects? 4. To what extent can firms influence their own bullwhip performance pathways? 1. How have firms insulated their operating margins from first order bullwhip effects? 2. Do the different ways to coordinate production impact the extent to which second order bullwhip impacts operating margins? 1. What are the different strategies related to utilizing capacity and inventory jointly to impact bullwhip? 2. Are other bullwhip control mechanisms more effective?
49 Page 49 of 54
d
M
an
us
cr
ip t
Figure 1a: Plot of ROA Results
Ac ce p
te
Figure 1b: Cross-sectional View of ROA Results (First Order =X, Second Order =Y)
50 Page 50 of 54
M
an
us
cr
ip t
Figure 2a: Plot of Operating Margin Results
Ac ce p
te
d
Figure 2b: Cross-sectional View of Operating Margin Results (First Order =X, Second Order =Y)
51 Page 51 of 54
d
M
an
us
cr
ip t
Figure 3a: Plot of SGA Expense Results
Ac ce p
te
Figure 3b: Cross-sectional View of SGA Expense Results (First Order =X, Second Order =Y)
52 Page 52 of 54
M
an
us
cr
ip t
Figure 4a: Plot of Fixed Asset Turnover Results
Ac ce p
te
d
Figure 4b: Cross-sectional View of Fixed Asset Turnover Results (First Order =X, Second Order =Y)
53 Page 53 of 54
M
an
us
cr
ip t
Figure 5a: Plot of Inventory Turnover Results
Ac ce p
te
d
Figure 5b: Cross-sectional View of Inventory Turnover Results (First Order =X, Second Order =Y)
54 Page 54 of 54