Operations Risk & Supply Chain Design: An Event...

23
Operations Risk & Supply Chain Design: An Event Study Yanni Papadakis Assist. Professor of Decision Sciences LeBow College of Business, Drexel University 3141 Chestnut Street, Philadelphia, PA 19104, USA tel: 215-895-0225, fax: 215-895-2907,[email protected] October 18, 2002

Transcript of Operations Risk & Supply Chain Design: An Event...

Page 1: Operations Risk & Supply Chain Design: An Event Studyopim.wharton.upenn.edu/risk/downloads/02-27-YP.pdf · 1 Introduction The fundamental dichotomy in supply chain management systems,

Operations Risk & Supply Chain Design:An Event Study

Yanni PapadakisAssist. Professor of Decision Sciences

LeBow College of Business, Drexel University

3141 Chestnut Street, Philadelphia, PA 19104, USA

tel: 215-895-0225, fax: 215-895-2907,[email protected]

October 18, 2002

Page 2: Operations Risk & Supply Chain Design: An Event Studyopim.wharton.upenn.edu/risk/downloads/02-27-YP.pdf · 1 Introduction The fundamental dichotomy in supply chain management systems,

Abstract

This empirical study provides evidence linking supply chain strategy andcompany risk structure. An event study on the stock performance of fourmajor PC producers is performed focusing on the 1999 earthquake in Taiwanand the computer memory price increases that ensued. It is shown that in-vestors associate pull-type supply chains for PCs with lower profitability afterabrupt component price increases. A parallel analysis of push-type producerstock returns does not show similar results. Furthermore, in depth analysisof Dell Computer reveals that after the catastrophe-induced disruption theonset of losses to this major pull-type PC producer was very fast. Far fromcondemning pull-type PC supply chains, earthquake-induced disruptions, likethe one researched, pose manageable risks.

Keywords: Supply Chain Disruptions, Push versus Pull Supply Chains,Supply Chain Risk, Operations Risk Management, Personal Computer SupplyChains.

Notes: Thanks to Paul Kleindorfer for helpful comments. A version of thispaper is forthcoming to Supply Chain Management.

Page 3: Operations Risk & Supply Chain Design: An Event Studyopim.wharton.upenn.edu/risk/downloads/02-27-YP.pdf · 1 Introduction The fundamental dichotomy in supply chain management systems,

1 Introduction

The fundamental dichotomy in supply chain management systems, that be-tween the pull and the push approach, has been acknowledged in theory andpractice. The push system takes full benefit from economies of scale in pro-duction and input acquisition by producing optimal output size and thendistributing to wholesalers and retailers. The pull system is known for itsadaptive efficiency. Inventories of products characterized by highly volatiledemand and technological depreciation need to move fast. According to thepull system, a value adding transformation takes place only after someonedemands it, in a Just In Time (JIT) fashion, thereby inventory risk is mini-mized.

Fisher (1997) has provided a parsimonious comparison of push (physicallyefficient) and pull (market responsive) supply chains according to their fitto product characteristics. At question is the relative performance of supplychain systems under stress, during disruptions in production or delivery. Thepull system in particular requires significant efforts in orchestration, so thatcustomers, producers, and suppliers are all at the right place at the righttime. Even if one achieves a clockwork coordination of a pull type supplychain, the latter appears likely to be sensitive to external disturbances.

This paper investigates the above hypothesis by looking at the perfor-mance of major Personal Computer (PC) manufacturers after the 1999 earth-quake in Taiwan, a region with high concentration in production of PC com-ponents. In particular we focus on the supply chain system pioneered in thePC industry by Dell Computer Corporation, a trend-setting company in sup-ply chain management. Dell uses a pull system that circumvents wholesalersand retailers, whereby it sells directly to consumers via internet and phone.This famous Direct Business model (Dell, 1999) operates in a configure-to-order manner, according to which consumers choose the PC they want andthen the producer manufactures it as well as has it delivered to customer’spremises.

The term Customized product Direct Marketing (CDM) will be utilizedfor this operations strategy. The often-used terms “direct marketing” and“direct sales” are more general and may refer to sales of pre-configured prod-ucts. CDM PC producers have attracted significant attention recently, be-cause they have delivered highly profitable Business-to-Business (B2B) andBusiness-to-Consumer (B2C) internet sales at a time when most internetstart-ups were struggling to prove their business models profitable.

2

Page 4: Operations Risk & Supply Chain Design: An Event Studyopim.wharton.upenn.edu/risk/downloads/02-27-YP.pdf · 1 Introduction The fundamental dichotomy in supply chain management systems,

Researching CDM supply chain systems under stress serves as a test tosupply chain theories and is also of interest to computer industry analysts.The main impetus for this study, however, has been to set a firm basis formanaging the risks of earthquake-induced disruptions in the PC industry.The mechanism of disruption propagation through the supply chain is com-plex, from semiconductor plants in Taiwan, to CDM supply chain dysfunc-tion, and ultimately to unexpected financial results in the stock markets.If it becomes clear, then the risk management process is facilitated. Com-panies would find it easier to design contingency plans and to modify theiroperations strategies so that risk is mitigated. In addition, risk transfer in-struments may be utilized, involving the insurance industry and the financialmarkets.

Section 2 provides a brief exposition to the events after the Taiwan earth-quake and their relation to the PC industry. Subsequently, supply chainmanagement theories pertaining to CDM are briefly reviewed. In the samesection selected financial data from published accounting reports of four ma-jor PC producers and relevant coverage from the business press are put inthe context of supply chain theory. The four PC producers are Dell, Gate-way, Compaq, and IBM. Particular emphasis is placed on Dell and on itsearnings-warning announcement a little less than a month after the disas-ter in Taiwan. Section 3 includes two event studies. The first concernsthe stock market performance of the four PC producers immediately afterTaiwan earthquake. The second focuses on Dell’s stock returns after its earn-ings warning, which attributed worse than expected financial outcomes to theearthquake-induced disruption. In Section 4 the results of the event studiesare discussed in detail. Finally, in Section 5 it is concluded that investorsin CDM companies are wary of the effect earthquake-induced supply disrup-tions may have, but the risks of disruptions like the one observed after theTaiwan disaster are manageable.

2 Expected Performance of PC Producers af-

ter the Taiwan Earthquake

On September 21, 1999 a magnitude 7.6 earthquake struck Chichi, Taiwan.It had devastating consequences. Baum (1999) reports that after the dis-aster more than 2,200 people lost their lives, more than 50,000 buildings

3

Page 5: Operations Risk & Supply Chain Design: An Event Studyopim.wharton.upenn.edu/risk/downloads/02-27-YP.pdf · 1 Introduction The fundamental dichotomy in supply chain management systems,

were destroyed, and the country’s infrastructure systems were left in disar-ray. The same author cites two post-disaster statistics as indications to theenormity of the economic disruption that followed: about 120,000 people be-came unemployed and total industrial production losses were estimated at$1.2 billion. 68 miles North of earthquake epicenter lies Hsinchu IndustrialPark, the home of 28 semiconductor fabrication facilities.

Baum (1999) estimates that at the time of earthquake about 10% of worldconsumption in computer memory chips was produced in Hsinchu. This areafeatures high production concentration in many other computer components,e.g. motherboards (more than two thirds of world consumption in 1999) andnotebook displays. Local producers of computer memory, TSMC and UMCbeing the leading Taiwanese suppliers, lost significant quantities of work inprogress at the time of the earthquake. Sherin and Bartoletti (1999) reportthat production lines could not restart at the first couple of days after theevent as sensitive critical-path equipment had been damaged. Furthermore,production returned to regular throughput rates only two weeks after the dis-aster, due to extensive power grid damage and thus unreliable power supplybefore grid reconstruction.

The world markets of memory chips reacted very fast to this news, assupply was constrained at the last part of 1999. The spot price of memorychips went up fivefold. Veverka (1999) reports that contract prices, a betterindicator of the input cost major PC producers face, went up by 25%. Thisadjustment was in excess of a 50% increase during the summer of 1999. It isimportant to emphasize that price increases in semiconductor products, likecomputer memories, are rare.

Indeed, the historical pattern has been one of decreasing prices. Grimm(1998) reports an average annual decline rate of 20% in a performance-adjusted index of PC memories (DRAMs). This index takes account of dif-ferent capabilities in prevalent models and is expressed in price over kilobits.Notably, the downward trend in the DRAM index is consistent with trends inother PC components. Curry and Kenney (1999) estimate that about 50% oftotal cost in PC inputs follows a “continuous and dramatic downward trajec-tory.” On the other hand there is substantial variance around the reporteddownward trend for PC memory prices. Grimm (1998) reports a DRAMprice index decline of 2.6% for 1995, whereas for 1996 this figure jumps to59.4%. The only appreciable yearly increase, however, of the DRAM indexfrom 1974 to 1996 in Grimm’s data (Table 5) occurred during 1988. It’svalue reached 27.3%. In light of Grimm’s results it becomes clear that the

4

Page 6: Operations Risk & Supply Chain Design: An Event Studyopim.wharton.upenn.edu/risk/downloads/02-27-YP.pdf · 1 Introduction The fundamental dichotomy in supply chain management systems,

price increases in PC memories during 1999 have been extraordinary.Curry and Kenney (1999), in their lucid overview of PC production logis-

tics, explain the various operation strategies utilized by producers adaptingto this environment of steadily declining input and concomitant final goodprices. As far as major US-based PC producers are concerned, supply chainsystems vary between two extremes. On the one end are intelligent versionsof the traditional push system, employed, for instance, by Compaq, IBM,and Hewlett-Packard. On the other end lies the CDM approach, also knownas direct business model (Magretta, 1998), represented mainly by Dell andGateway. The relative performance of these supply chain systems in tacklingprice declines and thus inventory risk is addressed briefly in the followingparagraphs. Then follows a discussion of what should be expected duringthese rare cases when input prices go up, as they did after the Taiwan earth-quake.

In an environment of declining prices push-type producers lose a portionof inventory value in inputs, work in progress, and final goods every day theyhold them. The longer the time span between purchase of inputs and receiptof payment from final good sales, the cash-to-cash cycle, the more the erosionin inventory value. Inventory risk is minimized by reducing the cash-to-cashcycle as the rate of decline in prices is not controllable by PC producers.Davis (1993) describes how Hewlett-Packard minimizes final good inventoryrisk by using channel assembly, shipping only generic products to distribu-tors, who are the ones to make final decisions about product configuration.Thus, under this version of the push-type strategy, higher product varietyis offered with fewer inventories. PC producers utilize various strategiesto minimize intermediate product value erosion, including supplier consol-idation, better coordination with suppliers via electronic data interchangesystems, and strategic location of suppliers close to production/distributionsites.

The founder of Dell Computer Corporation (Dell, 1999), however, ex-plains why the most important strategic risk push-type producers face isholding final good inventories for long. PC producers need to offer newproduct lines very often, catching up with the latest technological trends incomputer components. These innovative offerings carry higher prices at theirintroduction phase, while they are considered top line products. If the supplychain is full of conventional products (the ones with long market presence),then either innovative products have to wait or conventional products areoffered at give-away prices. Final goods must move fast to leave room for

5

Page 7: Operations Risk & Supply Chain Design: An Event Studyopim.wharton.upenn.edu/risk/downloads/02-27-YP.pdf · 1 Introduction The fundamental dichotomy in supply chain management systems,

PULL PUSHg Purchase Inputsg Initiate Productiong Distribute Product

Sell Product g Sell Product

Purchase Inputs gInitiate Production gDistribute Product g

@@ ��

Figure 1: Juxtaposed Cash-To-Cash Cycle Diagrams for Pull (CDM) andPush Supply Chains

fast introduction of high-profit innovative products.In contrast, CDM companies have a negative cash-to-cash cycle. As they

produce configure-to-order products, customers pay them first and produc-tion inputs are purchased later. When PC component prices are declining,CDM producers buy components at prices lower than what is budgeted attime of sale. The delivery time anticipated by customers, the establishedmarket standard, is currently about two to three weeks, but production ofa PC may be executed in less than a week. Curry and Kenney (1999) illu-minate the reasons why direct marketers delay purchasing components untilthe latest possible time. A second strategic advantage for CDM companiesis fast introduction of new product lines as they are unencumbered by finalgoods inventories.

The pattern of declining input and final good prices has characterized thecomputer industry since its inception. Nevertheless, Mann (2000) raises somecautionary remarks about the continuation in the long-term horizon of thedownward trend in computer component prices. In the short and mediumterm, however, it is generally accepted that the complexity of integratedcircuits will continue its geometric progress without substantial offsetting in-creases in the cost of new generation products. Moore (1997) in a reappraisalof his famous law, “Moore’s Law,” estimates that the complexity and per-formance of computer components will continue to increase geometrically in

6

Page 8: Operations Risk & Supply Chain Design: An Event Studyopim.wharton.upenn.edu/risk/downloads/02-27-YP.pdf · 1 Introduction The fundamental dichotomy in supply chain management systems,

the coming decade, driving prices down.Grimm’s (1998) review of historical prices reveals that the pattern of

decline in computer memory prices was interrupted in 1988 and 1994. Whenprices go up, push supply chain companies see the value of their inventoriesin inputs and intermediate products appreciating. They continue, though, toface the strategic problem of longer lead times for new product introduction.Pull-type producers, on the other side of operations strategy spectrum, findinput prices at time of purchase to be higher than budgeted during time ofproduct sale.

An examination of Dell’s and Gateway’s current pricing policies revealsthat the prevalent configure-to-order contract type in the PC industry hasthe producer insuring buyers against price escalation before delivery (no ad-justment clauses are prescribed). During an abrupt component price increasethere is little a CDM producer can do but to accept the loss. During gradualcomponent price increases there is more room for actions that will miti-gate losses. Examples of loss mitigation actions include: a) minimizing timebetween sale and component purchase, b) modifying price structure in antic-ipation of increases, and c) focusing marketing to products utilizing less ofthe expensive components and more of other compensating characteristics.Therefore the rate of change during input price increases is of paramountimportance. It should be noted that natural disasters result in surprisingand abrupt supply chain disruptions.

Financial performance indicators of four major producers during account-ing periods affected by the September 21, 1999 earthquake are depicted inTable I. Both major CDM companies, Dell and Gateway, are included inTable I. The other two PC producers are Compaq and IBM, two of thesales leaders in this industry, following the traditional push supply chain ap-proach. The common indicator differentiating CDM producers is inventoryturn period. Pendery (1998) reports an inventory turn period of 3.5 weeksfor Compaq and 4-6 weeks for IBM. The annual reports of Dell Computer(2000b) and Gateway (2000) put the inventory turn periods of the two CDMproducers to 6 and 9 days respectively.

As Dell’s third quarter ends on October 29, both third and fourth quarterresults for 1999 are shown in Table I. Fourth quarter results only are depictedfor the other PC producers. Dell Computer (1999, 2000a) realized higher rev-enues in 1999 than it did a year earlier, but its earnings ratio was 7% insteadof 8.2% in 1998. Gateway also realized increased sales. Its earnings ratio was5%. Compaq’s (2000) results are shown by customer group, differentiating

7

Page 9: Operations Risk & Supply Chain Design: An Event Studyopim.wharton.upenn.edu/risk/downloads/02-27-YP.pdf · 1 Introduction The fundamental dichotomy in supply chain management systems,

Revenue Change Earnings Earnings Reason$million over 98 $million over cited

Revenue

DELL 3Q99 6,784 41% 483 7% Memorysupply

DELL 4Q99 6,801 32% 486 7% Memorysupply,Y2K

GATEWAY 2,451 6% 126 5% Memory4Q99 supply,Y2KCOMPAQ 4Q99 3,133 -19% -79 -2.5% Y2KCommercialCOMPAQ 4Q99 1,966 24% 69 3.5%ConsumerIBM 4Q99 4,131 -7% -246 -6% Y2KPersonal Systems

Table I: Results of selected PC producers during accounting periods whenthe Taiwan Earthquake affected memory prices.

between commercial customers and single-product consumers. The commer-cial sales division produced a loss of 2.5%, due to reduced sales before year2000 (Y2K) conversion. IBM (2000) markets various types of products andservices. Citing IBM’s overall results would have been misleading. Hence,only the results of IBM’s Personal Systems division are shown. IBM faceda loss of 6% largely due to the Y2K effect and to IBM’s selling mainly tocommercial customers.

The figures in Table I may give the false impression that CDM companiesachieved the best results in their capacity for their shareholders in 1999. ByDell Computer’s (2000a) own admission, however, the former deduction isnot correct. Dell Computer announced on October 18, 1999 that its earningsratio had not been as high as 1998’s 8.2%, because supply problems related tocomputer memories were exacerbated after the Taiwan earthquake. This dis-crepancy in Dell’s earnings ratio surprised investors, as is going to be provenin the following. Gateway’s earnings ratio was 5%, similar to the one in 1998,but Gateway (2000) also announced that its sales were negatively affectedby a constrained memory supply and Y2K. The announcements by the two

8

Page 10: Operations Risk & Supply Chain Design: An Event Studyopim.wharton.upenn.edu/risk/downloads/02-27-YP.pdf · 1 Introduction The fundamental dichotomy in supply chain management systems,

CDM producers confirm the previous analysis indicating that increases incomponent prices do not favor CDM supply chains.

3 Event Study Methodology

To test whether investors find CDM companies more vulnerable to supplydisruptions, the stock price of the four major PC producers in Table I isexamined using the event study framework. A stochastic model for regularbehavior of stock price returns is fitted in each of the four cases. If duringa short period after the supply disruption (the event window) a company’sstock returns cannot be explained by the model of normal stock return be-havior, then one can reasonably assume that the disruption affected its stockprice (market value). Naturally, the effect of the supply disruption is ex-pected to be negative for PC producers, thus the presence of extraordinarilynegative abnormal returns is what we try to discern and establish. Impor-tantly, earthquakes cannot be predicted by financial markets, but becomegenerally known very fast after they occur. Hence, we expect that earth-quake induced disruptions of supply, when they have an effect on investorexpectations, result in sudden –thus easily discernible– changes in stock re-turns.

Additionally, in our study we examine whether the event of interest (sup-ply disruption) affected all companies in the sector the same way or whetherstocks exhibited different behavior depending on their supply chain type.We show that only CDM producers exhibited negative abnormal returnsafter the Taiwan earthquake, thus establishing that financial markets takeaccount of supply chain type when they judge the impact of supply disrup-tions. Regression analysis is the basic technique employed for fitting modelsof normal stock return behavior as a function of general market performance.Our analysis goes in depth utilizing various specification tests, which checkwhether key assumptions are supported by the data analyzed. It is shownthat the distribution of regression residuals is not sufficiently close to thetypically employed Normal (Gaussian) error model. Thus, abnormal returnsare analyzed both using the standard method of Normal cumulative returnsand a non-parametric method only assuming error terms have a continuousdistribution.

Stock market closings for Dell Computer (DELL:NASDAQ), Gateway(GTW:NYSE), Compaq (CPQ:NYSE), and IBM (IBM:NYSE) were obtained

9

Page 11: Operations Risk & Supply Chain Design: An Event Studyopim.wharton.upenn.edu/risk/downloads/02-27-YP.pdf · 1 Introduction The fundamental dichotomy in supply chain management systems,

starting on September 1, 1998, a little over a year before the Taiwan earth-quake (main data source: finance.yahoo.com). In addition, daily values ofStandard and Poor’s 500 index were collected for the same period as a proxyto market performance. Closings were adjusted for stock splits and cashdividends. All four companies examined have a sizable market share andconsequently an increased ability to influence suppliers. Thus, the results ofthis study may not be directly extendable to the many smaller companieswith regional or specialty sales presence. Smaller producers are likely to bemuch more exposed to supply price shocks.

R-square: 0.37 Df : 266 Dell Computer t

mt

Dellt eRR +⋅= 791.1

Residual St. Error : 0.030

β̂ St. Error of β̂ t P(>|t|)

1.791 0.1426 12.56 <2⋅10-16 *** F : 157.7 p-value ≈ 0

R-square: 0.25 Df : 266 Gateway t

mt

GTWt eRR +⋅= 550.1

Residual St. Error : 0.030

β̂ St. Error of β̂ t P(>|t|)

1.550 0.1631 9.50 <2⋅10-16 *** F : 90.3 p-value ≈ 0

R-square: 0.18 Df : 266 Compaq t

mt

CPQt eRR +⋅= 145.1

Residual St. Error : 0.032

β̂ St. Error of β̂ t P(>|t|)

1.145 0.1504 7.61 4.67⋅10-13 *** F : 58.0 p-value ≈ 0

R-square: 0.38 Df : 266 IBM t

mt

IBMt eRR +⋅= 131.1

Residual St. Error : 0.019

β̂ St. Error of β̂ t P(>|t|)

1.131 0.0891 12.69 <2⋅10-16 *** F : 161.0 p-value ≈ 0

Table II: Results of Least Squares Regression Models.

The Sharpe-Lintner (Campbell, Lo, and MacKinlay, 1997, p. 182) modelfor excess returns was utilized for the estimation of the long-term behaviorof stock prices. It takes the following form:

10

Page 12: Operations Risk & Supply Chain Design: An Event Studyopim.wharton.upenn.edu/risk/downloads/02-27-YP.pdf · 1 Introduction The fundamental dichotomy in supply chain management systems,

RPt = βP · Rm

t + εt (1)

Where: R stands for return (the relative difference of current price toprevious trading date price); β is the regression coefficient; t ∈ [1, 267] assubscript denotes time, with 1 corresponding to September 2, 1998 and 267 toSeptember 23, 1999; m and P are superscripts indicating variables correspondto the market and to producer P respectively; P has the following instances:“DELL”, “GTW”, “CPQ”, and “IBM”; finally, εt ∼ N(0, σ) are independentNormal errors with constant variance and zero mean.

3.1 Analysis of Post Earthquake Returns

The results of Least Squares Regression on all four stocks are depicted inTable II. The usual regression fit indicators are included. Note that all p-values in Table II are very low indicating different than 0 slope coefficientsand statistically significant F values. Consequently, Model 2 is an adequatedescription of return behavior for all four PC producers during the eventwindow. The relatively low R2 values are typical in the analysis of stockreturn series. Of particular importance are tests verifying properties of theerror term. To test independence, Box-Ljung tests for autocorrelation lags1-40 were performed on residuals (Cromwell, Labys, and Terraza, 1994). Inno case did a p-value lower than 10% result. It was concluded that theassumption of independent residuals was valid.

To test whether the error term follows the Normal distribution, threestatistics were collected for each set of residuals: median, 3rd moment, and4th moment. Figure 1 depicts smoothed densities of the observed residuals inall four cases (graphed residuals are standardized). These densities are jux-taposed to the density of the Normal distribution, the theoretically expecteddistribution of residuals under the Normality assumption. In addition, thestatistics of interest for all cases were added under the graph of residuals.Recall that for the Standard Normal distribution the median and the 3rd

moment are zero while the 4th moment takes the value 3.The medians of all stock residuals are slightly negative and the 4th mo-

ments are higher than 3. These are not rare phenomena in financial timeseries, but they raise doubts about the validity of usual parametric testsbased on the assumption that the error term follows the Normal distribution.Jarque-Bera tests (Cromwell, Labys, and Terraza, 1994, pp. 20-21), consid-

11

Page 13: Operations Risk & Supply Chain Design: An Event Studyopim.wharton.upenn.edu/risk/downloads/02-27-YP.pdf · 1 Introduction The fundamental dichotomy in supply chain management systems,

D

ell

Gat

eway

Median 3rd Moment

4th Moment

Jarque-Bera p-value

Median 3rd Moment

4th Moment

Jarque-Bera p-value

-0.0590 0.14 3.93 0.006 -0.0565 0.60 4.73 <10-6

Com

paq

IBM

Median 3rd Moment

4th Moment

Jarque-Bera p-value

Median 3rd Moment

4th Moment

Jarque-Bera p-value

-0.1538 -1.39 14.24 <10-6 -0.0131 1.20 9.78 <10-6

Figure 2: Graphs depict smoothed densities of standardized regression residu-als (ragged lines) compared to the theoretically expected Normal distribution.Under each company’s graph observed statistics describing distributional lo-cation and shape are added together with Jarque-Bera test p-values.

ering observed 3rd and 4th moments, rejected the hypothesis that residualsfollow the Normal distribution in all four cases. P-values of the Jarque-Beratests appear in Figure 2. In the following, non-parametric procedures will beutilized, complementing standard approaches, when assessing extreme resid-ual values.

The first event analysis concerns the performance of the four stocks afterthe Taiwan earthquake. The earthquake occurred on 12:47pm Eastern Timeof September 20th, 1999 (It was morning of the 21st in Taiwanese local time).It is assumed there was little time for the markets to react on the sametrading day, a little over four hours before New York Stock Exchange end oftrading on the 20th. Three trading days are included in the Post Earthquakeevent Window (PEW), September 21-23, 1999. The estimation period inregression analysis included PEW. Often the event window is not included in

12

Page 14: Operations Risk & Supply Chain Design: An Event Studyopim.wharton.upenn.edu/risk/downloads/02-27-YP.pdf · 1 Introduction The fundamental dichotomy in supply chain management systems,

the estimation period, as event window returns cause the regression line to tilttheir way. If this approach were followed, normal returns in the event windowwould be calculated as regression forecasts. The approach adhered to in thispaper is marginally more conservative. The benefit of using PEW residualsas Abnormal Returns (ARs) for each producer, P, (∀t ∈ PEW ARP

t =RP

t − R̂t ) arises when non-parametric procedures are employed. Regressionforecasts are mutually correlated, whereas regression residuals were tested tobe independent.

Two tests were performed on abnormal returns. The first considersproducer P’s Cumulative Abnormal Return in PEW ( CARP (PEW ) =∑

t∈PEW ARPt ), assuming independent Normal residuals. According to the

Null Hypothesis of no abnormal returns:

E{CARP (PEW )} =∑

t∈PEW

E{ARPt } = 0 (2)

V {CARP (PEW )} =∑

t∈PEW

V {ARPt }

=∑

t∈PEW

(σP )2 = |PEW | · (σP )2 (3)

The variance of the cumulative abnormal returns is obtained by the resid-ual standard error 1 of each stock’s model:

V {CARP (PEW )} = |PEW | · (sP )2 (4)

The Standardized Cumulative Abnormal Return for producer P in PEW

(SCARP (PEW ) = CARP (PEW ) · (√|PEW | · sP )−1) follows Student’s t-

distribution. The p-values against the alternative hypothesis, that abnormalreturns are less than zero, are depicted in Table III. Additionally, the stan-dardized abnormal returns in PEW for all four cases appear in Table III.

1Alternatively, Studentized residuals could be used in this study. It should be noted,however, that sample size is high and dependent variable values are close to their mean.Thus, Studentized and Standardized residuals are not far apart.

13

Page 15: Operations Risk & Supply Chain Design: An Event Studyopim.wharton.upenn.edu/risk/downloads/02-27-YP.pdf · 1 Introduction The fundamental dichotomy in supply chain management systems,

The second test performed was the two-sample Kolmogorov-Smirnov test(Rohatgi, 1976), which uses two assumptions: a) the mild assumption thatthe distribution of residuals is continuous, and b) the assumption that resid-uals are independent, which was tested true previously. The null hypothesisin this test is that residuals and abnormal returns belong to the same con-tinuous distribution. The alternative hypothesis is that the abnormal returnsin PEW belong to a different distribution, which is stochastically dominatedby the distribution of residuals outside PEW. Kolmogorov-Smirnov statisticsD+ and corresponding p-values are depicted in Table III.

DELL GTW CPQ IBM

Return Sep 21 -0.372 -0.784 0.096 0.033

Residuals Sep 22 -0.449 -1.333 0.003 -0.956

Standardized Sep 23 -0.848 -2.082 -0.012 0.027

Normal Error SCAR -0.964 -2.424 0.05 -0.518

Test p-value 0.168 0.008∗∗ 0.52 0.303

Kolmogorov D+ 0.652 0.841 0.394 0.477

Smirnov Test p-value 0.081· 0.015∗ 0.398 0.259

Table III: Standardized Residuals of Stock Return Models in the Post-Earthquake Event Window for Major PC Producers and 2 Tests (SCAR andKolmogorov-Smirnov) Against the Null Hypothesis of Normal Returns. OnlyCDM Producers Exhibit Abnormal Negative Returns After the Earthquake.

14

Page 16: Operations Risk & Supply Chain Design: An Event Studyopim.wharton.upenn.edu/risk/downloads/02-27-YP.pdf · 1 Introduction The fundamental dichotomy in supply chain management systems,

From the results of Table III we may not reject the null hypothesis ofnormal returns in PEW for the two push-type PC producers: Compaq andIBM. The Kolmogorov-Smirnov test results for the two CDM producers rejectthe null hypothesis in favor of negative abnormal returns at significance levelα=10% for Dell and α=5% for Gateway. The SCAR test fails to reject thenull hypothesis for Dell, but agrees with the Kolmogorov-Smirnov test forGateway at level of significance α=1%. The weak evidence (p-value=8.1%) inassociating poor financial performance for Dell immediately after the Taiwanearthquake is corroborated by Dell’s public statements.

R-square: 0.35 Df : 284 Dell Computer tmt

Dellt eRR +⋅= 792.1

Residual St. Error : 0.030

β̂ St. Error of β̂ t P(>|t|) 1.7915 0.1382 12.41 <2⋅10-16 ***

F : 154.1 p-value ≈ 0

Table IV: Results of Least Squares Regression Model for Dell Computer inEWW.

3.2 Analysis of Dell’s Returns after its Earnings Warn-ing

Dell Computer (1999) by its own admission, with its earnings warning an-nouncement on October 18, less than a month after the Taiwan earthquake,linked below expectations 3rd quarter results to the disaster. Using a sep-arate event study it will be shown that the earnings warning resulted innegative abnormal returns. Thus, magnitude and speed of onset of the post-earthquake losses to Dell surprised investors.

15

Page 17: Operations Risk & Supply Chain Design: An Event Studyopim.wharton.upenn.edu/risk/downloads/02-27-YP.pdf · 1 Introduction The fundamental dichotomy in supply chain management systems,

The Sharpe-Lintner model is now fitted to Dell Computer’s returns fromSeptember 1, 1998 to October 19, 1999. The results appear in Table IV.Not surprisingly the regression fit is very close to the fit in Table II. Againthe slope coefficient and F statistic are statistically different than zero. Thechosen Earnings Warning event Window (EWW) contains October 18 and 19of 1999, day of announcement and following trading day. Results of SCARand Kolmogorov-Smirnov tests appear in Table V. Both tests reject the nullhypothesis of normal returns in favor of negative abnormal returns at signif-icance level α=5%.

DELL

Std. Return Oct 18 -1.484

Residuals Oct 19 -2.612

Normal Error SCAR -2.896

Test p-value 0.002 ∗∗

Kolmogorov D+ 0.096

Smirnov Test p-value 0.026 ∗

Table V: Standardized Residuals of Stock Return Model for Dell Computerin EWW and 2 Tests (SCAR and Kolmogorov-Smirnov) Providing Evidenceof Abnormal Negative Returns after the Warning for Reduced ProfitabilityDue to DRAM Shortages.

4 Discussion of Event Study Results

and Lessons from Dell’s Experience

The analysis of the stock performance of major PC producers in PEW provesthat investors perceive companies operating pull-type supply chain systemsas vulnerable to disruptions in the market of PC components. The value ofDell Computer’s and Gateway’s stocks at the end of PEW was, respectively,4.9% and 13.7% less than what is expected by their long term relation withmarket return. These losses cannot be explained by random variation. Bycontrast, IBM and Compaq, two major push-type PC producers do not showabnormal returns in the same window.16

Page 18: Operations Risk & Supply Chain Design: An Event Studyopim.wharton.upenn.edu/risk/downloads/02-27-YP.pdf · 1 Introduction The fundamental dichotomy in supply chain management systems,

It is instructive to investigate how the losses from higher than expectedinput prices were distributed in Dell’s case. In general, input price increasesmay be passed on to consumers or absorbed by the producer. In the lat-ter case investors will shoulder the burden either in the short term in theform of lower dividends or in the long term as lower cash reserves for thepursuit of growth opportunities. Dell Computer’s (2000b) annual report re-veals that its gross margin in 1999 was 20.7% of net revenue, whereas in1998 it was higher, 22.5%. Moreover, Dell spent $1.1 billion during 1999 inrepurchasing common stock, whereas in 1998 it spent $1.5 billion in stockrepurchases. Therefore, computer memory increases were not passed on toconsumers as higher product prices, but they were absorbed by the companyand were passed on to investors in the form of less stock repurchases. It fol-lows that Dell’s investors were right to be vigilant about component marketdisruptions.

As in the prevalent configure-to-order contract the producer is not com-pensated by the consumer if prices increase after sale and before delivery,CDM companies are left with few options during abrupt increases in com-ponent prices. In Dell’s case the effect of memory price increases emergedremarkably fast in the company’s accounting data. Dell Computer issued itsearnings warning a month after the Taiwan earthquake. It predicted lowerearnings for its third quarter of 1999, ending October 29, a little over a monthafter the disaster. The results of the second event analysis make clear thatthis warning surprised investors, apparently because of the speed of onsetand magnitude of loss the sudden memory price increase caused.

The existence of a clearly articulated risk management plan for disaster-induced supply disruptions hasn’t appeared in Dell’s official announcementsduring the six month period after the event in Taiwan. The inherent supplychain agility of this CDM company, however, offered it several means ofrecourse during the month that followed the disruption. Dell operates on aconfigure-to-order basis, thus the final decision on product configuration restswith Dell’s customer. The moment an input’s price increases customers maymodify their configuration preferences by requesting less of the expensiveinput. Veverka (1999) reports that Dell changed its marketing strategy afterthe Taiwan earthquake in an effort to shift consumer preferences towards lowmemory products.

17

Page 19: Operations Risk & Supply Chain Design: An Event Studyopim.wharton.upenn.edu/risk/downloads/02-27-YP.pdf · 1 Introduction The fundamental dichotomy in supply chain management systems,

A second ingredient of Dell’s supply chain strategy, long term contractswith suppliers, did not deliver steady prices, despite expectations to thecontrary in PC industry press (Deckmyn, 1999). Baljko-Shah (2000) re-ports that Dell was forced to buy regular DRAM memories after the Tai-wan earthquake, while their prices were high. Dell was planning to incorpo-rate in its most innovative product line best-available technology memories(RDRAM). Contrary to earlier announcements, Computer Processor Unit(CPU) suppliers did not make available on time CPUs compatible with newtechnology memories. Dell ended up buying conventional memories duringthe earthquake-induced shortage in order to meet advertised commitmentsto increased memory capability in its innovative products.

It is very difficult to predict whether promising new computer technolo-gies will go from drawing board to the market without delays. Nevertheless,in a decreasing component price environment it is easy to find cheap conven-tional technology substitutes, when the introduction of leading technologycomponents is delayed. Dell Computer (2000a) announced that during thefourth quarter of 1999 it lost $300 million in revenue, due to unrealized salesof primarily newly introduced products. The latter product category pro-duces the highest profit margins, so this was a noteworthy hit to the CDMproducer’s profitability.

5 Conclusions

Investors associate pull-type supply chains for PCs with lower performanceduring an abrupt disruption of supply, like the one ensued after the 1999earthquake in Taiwan. In Dell’s case the loss due to the price increase inmemory components was sizable, showed up fast in the company’s accountsand lead it to issue an earnings warning. This warning surprised investorsand caused a statistically significant decrease in the company’s stock price.Investor uneasiness about a company’s risk exposure is clearly an importantreason for reformulating operation risk policies. Additionally, the very natureof an earthquake-induced disruption in a CDM supply chain, its comingunexpectedly and leaving the company without means of recourse in the shortterm, provides a second major justification for risk management planning.Therefore, supply disruption risks deserve consideration by CDM companies.

18

Page 20: Operations Risk & Supply Chain Design: An Event Studyopim.wharton.upenn.edu/risk/downloads/02-27-YP.pdf · 1 Introduction The fundamental dichotomy in supply chain management systems,

After assessing financial results by Dell and Gateway at the end of 1999(key indicators are depicted in Table I), however, it becomes clear that theperformance of the two CDM PC producers, albeit below expectations, wasnot disappointing compared to competitors’ results. An unwanted outcomeis weighted heavily, if its probability of recurrence or the magnitude of lossesresulting after it are high. And this negative weight should be juxtaposedto the benefits of the strategy that introduces it. As earthquakes are notfrequently recurring events, the risk CDM companies face from a disaster-induced disruption, like the one studied here, is manageable.

Pull-type PC producers do not need to change their successful core oper-ations strategy just to avoid this contingency. Therefore, a risk managementpolicy, if deemed acceptable by investors, may alleviate the adverse impactdisaster-induced disruptions have. Of course, a detailed study of naturaldisaster risk in areas with high concentration of computer component pro-duction may provide a crisper picture of the risks faced by the PC industry.For example, Sherin and Bartoletti (1999) estimate that Hsinchu IndustrialPark, where most of the semiconductor production capacity in Taiwan is sit-uated, may be exposed to disruptions lasting over six months, about twentytimes longer than the one studied here.

Putting to practice supply chain theories in order to bridge supply chainstrategy with company financial performance is a daunting task. Supplychain theory attempts to clarify the complex interconnections among manyactors in supply networks. Yet, it is unclear whether simple formulas forsupply chain performance, encompassing a few variables, will have generalapplication to business practice. In addition, it is difficult to design empiricalstudies that would isolate the effect of supply chain strategy on businessperformance from other company decisions and environmental variables. Thestudy of supply chain disruptions may provide an interesting exception to thelatter restriction, in that disruption impact may test whether supply chainmanagement affects company risk structure, as is accomplished here.

19

Page 21: Operations Risk & Supply Chain Design: An Event Studyopim.wharton.upenn.edu/risk/downloads/02-27-YP.pdf · 1 Introduction The fundamental dichotomy in supply chain management systems,

Supply chain risk management is a decision process often requiring a mul-tidisciplinary approach. Typically, risk mitigation and contingency planningentails skills in operations strategy and supply chains. Crisis managementis handled mainly by the legal and public relations departments of a com-pany. Purchasing, insurance and finance experts are often involved in risktransfer. Therefore, a number of dimensions need to be explored in everyrisk management study fitting the needs of the problem in hand. One mightsurmise that among the two risk shielding strategies a company may opt for,risk mitigation offers more cost-effective alternatives than risk transfer. DellComputer’s doctrinal commitment to minimal inventories, however, is wellknown. Companies with similar strategic commitments are unlikely to beinterested in risk mitigation policies involving emergency inventories alongthe supply chain.

In this case risk transfer is left as the main option to consider, includingcontracts with emergency suppliers and insurance contracts. There is a fastgrowing literature on alternative methods of risk transfer. It would be in-teresting to explore whether the latter methods may shield CDM companiesfrom investor uneasiness after disruptions in component markets. In light ofKunreuther and Bantwal’s (2000) discussion on rigidities in the successful in-troduction of Cat-Bonds, one alternative risk transfer instrument, the lattertask may be challenging, but appears to be worth the effort.

References

Baljko-Shah J. (2000), “DRAM supply chain gets stress tested,” ElectronicBuyer’s News, Manhasset, NJ, January 28.

Baum, J. (1999), “Back on track,” Far Eastern Economic Review, HongKong, November 25.

Campbell, J. Y, Lo A. W., and A. C. MacKinlay (1997), The Econometricsof Financial Markets, Princeton University Press, Princeton, N.J.

Compaq Co. (2000), “Compaq announces fourth quarter, full year 1999results,” Compaq press release, January 25.

Cromwell, J. B., Labys, W. C., Terraza, M. (1994), Univariate tests for timeseries models, Sage Publications, London.

20

Page 22: Operations Risk & Supply Chain Design: An Event Studyopim.wharton.upenn.edu/risk/downloads/02-27-YP.pdf · 1 Introduction The fundamental dichotomy in supply chain management systems,

Curry, J. and Kenney, M. (1999), “Beating the clock: Corporate responsesto rapid change in the PC industry,” California Management review, Vol. 42No.1, pp. 8-36.

Davis, T. (1993), “Effective supply chain management,” Sloan ManagementReview, Summer, pp.35-45.

Deckmyn, D. (1999), “Memory price increase hits Dell”, Computerworld,October 25.

Dell Computer Co. (1999), “Dell posts Q3 per share earnings of 18 cents,ex acquisition charge,” Company announcement, November 11. Availablewww.dell.com.

Dell Computer Co. (2000a), “Dell fourth-quarter revenue, earnings willbe lower than expected,” Company announcement, January 26. Availablewww.dell.com.

Dell Computer Co. (2000b), Annual Report for FY ending January 28, 2000:Form 10-K, Securities and Exchange Commission, Washington, DC.

Dell, M. S.(1999), Direct from Dell, Harper Business, New York. Fisher, M.L. (1997), “What is the right supply chain for your product?” Harvard Busi-ness Review, March-April, pp. 105-116 Gateway Inc. (2000), Annual Reportfor FY 1999: Form 10-K, Securities and Exchange Commission, Washington,DC.

Grimm, B. T. (1998), “Price indexes of selected semiconductors, 1974-1996,”Survey of Current Business, Vol. 78 No. 2, pp. 8-25.

IBM Co. (2000), “IBM announces 1999 fourth-quarter, full-year results,”IBM press release, January 19.

Kunreuther, H. C. and Bantwal V. J. (2000), “A Cat Bond Premium Puzzle,”Journal of Psychology and Financial Markets, Vol. 1.

Magretta, J. (1998), “The power of virtual integration: An interview withDell Computer’s Michael Dell,” Harvard Business Review, March-April, pp.73-84.

Mann, C. C. (2000), “The end of Moore’s law?” Technology Review, Vol.103 No. 3, p. 42.

Moore, G. E. (1997), “The microprocessor: Engine of the technology revolu-tion,” Communications of the Association of Computing

21

Page 23: Operations Risk & Supply Chain Design: An Event Studyopim.wharton.upenn.edu/risk/downloads/02-27-YP.pdf · 1 Introduction The fundamental dichotomy in supply chain management systems,

Machinery, Vol. 40, No. 2, pp.112-114.

Pendery, D. (1998), “Compaq to shorten inventory cycle,” InfoWorld, July20, p.20.

Rohatgi, V. K. (1976), An Introduction to Probability Theory and Mathe-matical Statistics, John Wiley & Sons, New York.

Sherin, B. and S. Bartoletti (1999), Taiwan’s 921 Quake: Effect on the semi-conductor industry and recommendations for preparing future earthquakes,Semiconductor Equipment and Materials International (SEMI), MountainView, CA.

Veverka, M. (1999), “A DRAM shame,” Barron’s, October 25, p. 15.

22