Accepted Manuscript

Forecasting Supply Chain Performance Resilience Using Grey Prediction

Rajesh Rajagopal

PII: S1567-4223(16)30054-0
DOI: http://dx.doi.org/10.1016/j.elerap.2016.09.006
Reference: ELERAP 684

To appear in: Electronic Commerce Research and Applications

Received Date: 21 September 2016

Revised Date: 23 September 2016
Accepted Date: 23 September 2016

Please cite this article as: R. Rajagopal, Forecasting Supply Chain Performance Resilience Using Grey Prediction,
Electronic Commerce Research and Applications (2016), doi: http://dx.doi.org/10.1016/j.elerap.2016.09.006

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.
 FORECASTING SUPPLY CHAIN PERFORMANCE RESILIENCE
USING GREY PREDICTION

Rajesh Rajagopal
Indian Institute of Space Science and Technology
Email: [email protected], [email protected]
Last revised: September 23, 2016
______________________________________________________________________________

ABSTRACT
New digital technologies have empowered companies to make use of tools to produce insights for key
supply chain processes. Resilience, the property of supply chains to respond better to disruptions, can be
identified and evaluated using performance indicators. This research analyzes and predicts the indicators
of firm resilience based on indicators from secondary data. These indicators can be measured on a regular
basis based on performance in flexibility, responsiveness, quality, productivity and accessibility. Since
source information for the data is often unknown, a methodology that is suited for prediction needs to be
used. An improved grey prediction model is proposed in this research for forecasting the periodic indica-
tors of resilience performance. This research shows that error measures ensure the best fit of the data to
achieve strong prediction capability. A prediction model is applied to the supply chain of an Indian elec-
tronics manufacturer to forecast the measures of its resilience. Also, the results obtained were validated
and have practical implications. For the supply chain that we studied, the indicators of flexibility, respon-
siveness and accessibility increased for the future periods, whereas the indicators for quality and produc-
tivity showed slight decreases. We argue that indicators with a negative trend should be given more atten-
tion by the firm. And managers should analyze the predicted values of resilience so they have an empiri-
cal basis for adjusting their strategies.
Keywords: Big data; grey theory; grey prediction; supply chain resilience; resilience indicators
_____________________________________________________________________________________

1
1. INTRODUCTION

Companies have high expectations for the implementation of big data analytics into their supply chain

planning, implementation and monitoring activities, but have difficulties in adopting it. For those compa-

nies that have succeeded in adopting big data analytics strategies, their operations have realized the prom-

ising power of digital technology. The increased demand for digital technology adoptions enabled com-

panies to collect massive amounts of data, but powerful yet reliable decision-making tools are needed to

sense and analyze those data (Tan et al., 2015; Isasi et al., 2015; Hofmann, 2015). In recent years, there is

much more awareness of the value and need for big data; still many companies have not understood how

to apply big data analytics to achieve high supply chain performance.

Despite the stated benefits from big data analysis, most companies are facing difficulties owing to the

large investments required to deploy and use analytics, security issues, privacy issues, lack of business

cases, limited support from executives and lack of in-house capabilities. Many of them have a dedicated

group of people who use a sophisticated tool. This is their in-house capability. They also may have an

independent team of data scientists, who are competent enough for finding, manipulating, managing and

interpreting data to generate insights (George et al., 2014; Sanders, 2014; Wang and Alexander, 2015).

Among the companies that employ big data analysis, a majority makes use of one or more sophisticated

tools that enable day-to-day decision-making using big data or that uses sophisticated tools to produce

insights for key supply chain processes.

Resilience is defined as the adaptive capability of the system to respond better to disruptions or even

gain advantages out of such events (Wieland and Wallenburg, 2013; Brandon‐Jones et al., 2014; Scholten

et al., 2014). It is the property by which a supply chain can reduce, react to and overcome potential risks

and vulnerabilities. The resilience of the firm can be identified and evaluated using resilience perfor-

mance indicators (RPI) (hereafter, shorted to resilience indicators or just indicators). They can be meas-

ured on a regular basis by using several operational level measures. Five major indicators of resilience

were taken into consideration for this study and the data give a reading on the indicators of flexibility, re-

2
sponsiveness, quality, productivity and accessibility. Our problem employs a representative case study

(Yin, 2013) of an Indian electronics manufacturing firm.

Using data analytics tools, the datasets can be aggregated to create a yearly performance measure for

the particular indicator. Irrespective of the units or the measured quantity for performance, a uniform

evaluation scheme needs to be implemented. To measure indicator quality, an aggregation of several per-

formance measures for the period are taken into consideration and the resulting value represents the

measure for the indicator quality for the particular year. The aggregation scheme should follow a similar

pattern the indicator measures during the successive years. The measurement schemes or the aggregation

techniques can vary, but should follow a uniform pattern for the measurement of a particular indicator for

all periodic assessments.

The resulting values represent the raw data for analysis of the prediction model. A grey prediction

model is most appropriate when the available data need to be analyzed despite uncertainty at their source.

It is also possible to test the model for several measures of errors, and to fit the data into the best grey

prediction model. Each resilience indicator is analyzed, predicted and checked for errors, so they are ade-

quate for making effective value predictions in the subsequent years.

2. BIG DATA AND FORECASTING

In this digital era, organizations have access to enormous structured and unstructured data considering

its volume, velocity, variety and veracity. By employing proper analytics tools and big data solutions, or-

ganizations can harvest significant benefits and value out of this data. This enables companies to improve

their decision-making capabilities and to reduce their vulnerabilities. Many industries consider big data to

be a core component for their strategic decision-making. Analytics offers multiple benefits to improve

and enhance products and services so firms will excel in competitive markets (Chen et al., 2012; Waller

and Fawcett, 2013b; Tien, 2013; Salmon, 2015). The key benefits vary with industries, such as for the

security industry, where it enhances risk management and fraud detection. Or for the retail banking indus-

try, where the focus is on cost reduction, whereas in retailing, data analytics help enhance the prediction

3
of customer behavior and preferences. Organizations should plan for maximizing their return on invest-

ments related to big data solutions. A brief literature on the application of data analytics in supply chain

management is presented in Table 1.

INSERT TABLE 1 ABOUT HERE

Data analytics have enormous potential to improve business benefits and have a major impact on

overall operating and financial performance. Enabling the use of data analytics in operations cangenerate

substantial returns (Demirkan and Delen, 2013; Wixom et al., 2014; Chang et al., 2014). The major bene-

fits include shorter order-to-delivery cycle times, improved demand-driven operations, better customer

and supplier relationships, effective decision-making, quicker reaction to issues, optimized inventory and

asset productivity. Operationalization requires the use of the right tools for supporting the right processes

in an appopriate. Predictive models can also be built that rely on data analytics (Waller and Fawcett,

2013; Hazen et al., 2014; Schoenherr and Speier‐Pero, 2015). Building analytical processes for big data

applications into predictive modeling requires evaluation of the data needed for accurate findings.

The ability to handle and analyze all available data is perceived as a corporate asset. The analytics

teams need to weigh the benefits of a full assortment of data at their disposal (Vera-Baquero et al., 2013;

Davenport, 2014; Li et al., 2015). We often wish to know whether feeding more data into an analytical

model will more provide accurate results. Much of the data may not be used as inputs to the model, but

the advantage of data availability is that it will be there when such need arises. This provides an analyst

with the flexibility of testing and validating various models with larger datasets.

The adoption of big data, the implementation of data analytics programs, and the practices associated

with these varies with different industries, but companies should consider data as a major component of

their decision-making processes while considering their vendors and solution providers. By assessing and

documents the details of the available data sources, data scientists will be able to identify their potential

value. Data scientists may examine settings with few production impacts and involve a smaller number of

users, in order to create proof-of-concept tests. And only after achieving benefits at a smaller scale size of

data analytics, they can extend their data to handle a business setting that requires much more. What is

4
required though is the expertise and talents of data scientists, conceptualization specialists, data analysts,

data engineers, systems architects and data stewards (Demirkan and Delen, 2013; Kwon et al., 2014;

Kaynak and Yin, 2015). As companies establish data analytics capabilities, they are emphasizing align-

ment for these business and product line activities.

Many forecasting techniques and models have been employed to handle the increasing variety and

complexity of managerial forecasting problems. Managers and forecasters play critical roles in technique

selection and understanding the range of possibilities in forecasting. The selection of methods depends on

the context of the forecast, relevance and availability of historical data, levels of accuracy needed, period

of forecast, the cost-benefis of the forecast to the firm, and the time available for making the analysis

(Manrique et al., 2013; Zheng et al., 2015; Kim, 2016). The weightings given for the factors should be

assumed only on the basis of relevant evidence and the technique used must make best use of the availa-

ble data. A successful forecasting model should primarily satisfy and address: (1) the purpose of the fore-

cast; (2) the dynamics and components of the system under study; and (3) importance of past data in esti-

mating the future. A brief literature on data and forecasting models is presented in Table 2.

INSERT TABLE 2 ABOUT HERE

3. A FRAMEWORK OF RESILIENCE PERFORMANCE INDICATORS

Turbulent business environments and the increasing complexity of global supply chains pose a seri-

ous threat of unexpected but inevitable risk and disruption. A resilient supply chain has adaptive capabil-

ity to manage disruptions by enabling the supply chain to bend rather than to break. Supply chain resili-

ence is a property that increases the sustainable competitive advantage of firms (Ponis and Koronis, 2012;

Melnyk, 2014; Ambulkar et al., 2015). Supply chain resilience is mainly attributed to six performance

elements: (1) collaboration to increase information sharing; (2) cross-functional teams to deal with risk

events; (3) inventory management using safety stocks and available buffers; (4) contingency plans to re-

duce mistakes; (5) redundancies in the form of multiple suppliers and slack resources; and (6) visibility in

5
the form of early warning indicators and financial monitoring (Johnson et al., 2013; Pereira et al., 2014;

Hohenstein et al., 2015; Scholten and Schilder, 2015).

The level of a firm’s supply chain resilience can be evaluated by certain performance measures. In

this research, five indicators of resilience are considered for determining the actual performance of the

firm and to forecast the indicators in the coming periods by employing grey prediction analysis. These

indicators are determined by several other measures that are operationally quantifiable. Periodic assess-

ment of these measures can be done and it is possible to aggregate them for a long period. The indicators

represent the flexibility, responsiveness, quality, productivity and accessibility of the firm or its supply

chain.

3.1. Flexibility Indicators

Flexibility in a supply chain is the ability to manage changes quickly without undue effort and loss. It

can be related to a system, product or process. System flexibility is the redundancy in capacity, inventory,

sourcing capabilities, and coordination and integration levels. Product flexibility corresponds to modular-

ization and standardization. Process flexibility is related to flexible production and warehousing systems

(Ivanov et al., 2014). Several determinants measure the flexibility of a system:

(1) Stockout rate: This is the percentage of facilities that experienced a stockout of a product that the

site was expected to deliver for a period of time:

   
= !           " × 100
       
(1)

(2) Inventory accuracy rate: This is the ledger balance of stock that matches the inventory on hand:

&'(
')  )  
=

 " × 100
   *    + ,  
!    
(2)

(3) Number of small disruptions managed through flexibility: This counts the operational emergency

incidents handled through process, product, and production flexibility.

(4) Percentage increase in sales from design flexibility: This is the percentage of sales after a product

has been redesigned or modified to suit market requirements due to design flexibility.

6
 -

' .
/'
0
/' 0 1
0 ' 2
0/.' 31
4/5/1/) =

 " × 100
6  7 786  7 7
6  7 7
(3)

3.2. Responsiveness Indicators

For managing risk, it is important for a firm’s supply chain to be coordinated with its partners to im-

prove information sharing and increase responsiveness. Visibility and velocity are the two major indica-

tors of supply chain resilience that contributes to the agility of the supply network. The responsiveness of

the supply chain is the speed of delivery of its products from a customer’s viewpoint (Cantor et al., 2014;

Qi et al., 2014). Several measures can be used for the responsiveness of the supply chain system.

(1) On-time delivery ratio: It is the ratio of number of orders delivered on date to the total number of

orders delivered.

9' − /;
2
1/(
)  / =
   <  
!    <
(4)

(2) Contract issue time: It represents the average amount of time it takes when a decision to order is

made to when the procurement department issues the contract.

(3) Contract approval time: It is the average amount of time it takes when a purchase order was is-

sued to when the contract was actually awarded and signed by a supplier.

(4) Put-away time ratio: It is the ratio of the difference in time for which product has been unloaded

from handling systems or a truck until it is stored in a designated place to the total time for the

product is expected to stock in.

- − = ) /;
 / =

>   ?@ 8>   ?@ 
!     7
(5)

3.3. Quality Indicators

The quality of a product or process is a primary performance indicator for a supply chain. The

quality of a product involves the quality of raw materials, testing and shipping qualities. The quality of

aprocess involves the accuracy of the manufacturing, transformation and assembly processes, including

7
the quality of the machinery. Accuracy in forecasting is a measure of their quality (Rajesh and Ravi,

2015). Key quality measures for a supply chain system are:

(1) Quality of forecasts: Measures the percentage of difference between forecasts previously made

for a year and the actual consumption or other operational aspects for that year.

A 1/) 3 3
 0 = B1 −  "C × 100
 <8  <
 <
(6)

(2) Testing quality: Represents the percentage of individual products per lot per shipment that under-

go quality testing over a period of time.

D
0/'. E 1/) =  !      " × 100
     
(7)

(3) Shipping accuracy: This is the percentage of the products shipped without error or damage to the

total number of units shipped.

ℎ/GG/'.  ) =  " × 100

   <7 *  7 
!    <
(8)

(4) Security measures: This is the percentage of operations following guidelines or standard operat-

ing procedures to prevent delays, loss or leakages.

3.4. Productivity Indicators

In a supply chain, productivity is positively related to customer satisfaction levels. Good communica-

tion systems and continuous training among employees are essential to improve productivity. A produc-

tivity-oriented supply chain will result in maximum savings, improved service levels and an inspired team

of employees (Pettit et al., 2013). The major indicators are:

(1) Order compliance: This is the percentage of orders for which the manufacturer does not have suf-

ficient stocks and have to back order products during a defined period of time.

92
 ;G1/ '
=  " × 100
   *    
!    
(9)

(2) Fill rate: This is the suppliers’ or manufacturers’ ability to fill orders completely in terms of

items and quantity as defined in the contract.

H/11  
=  "
        
! +, 
(10)

8
 (3) Storage space utilization: This is the percentage of total storage space used out of total storage

space available:

 .
0G 
/1/I /' = ! 7  <" × 100
! 7   
(11)

(4) Units moved per person-hour: This is the number of units moved during a defined period of time

per person-hour for each person working during the period:

J'/0 ;(
2 G
 G
0' − ℎ0 = !   8 "
!    <
(12)

3.5. Accessibility Indicators

This indicates the visibility of supply networks and increased accessibilities reduces the vulnera-

bilities and increases trust in supply chains. Supply chain risk increases due to the lack of accessibility

and high supply accessibility guarantee resilience. From a complexity theory perspective, accessibility

increases the interactive complexities of networks, but increases the trust and transparency of operations

(Brandon‐Jones et al., 2014). The major indicators of accessibility in supply chains are:

(1) Dealer accessibility: This represents the percentage of total dealers having direct access to manu-

facturers and suppliers.

(2) Retailer accessibility: It indicates the percentage of total retailers having direct access to manu-

facturers and suppliers.

(3) Customer accessibility: This is a measure of the percentage of total customers having direct ac-

cess to manufacturers and suppliers.

(4) Network intensity: This indicates the number of connections among suppliers, retailers and cus-

tomers.

4. METHODOLOGY FOR GREY PREDICTION

Grey theory and its allied methodologies have wide applications in theory and practice of solving

uncertain and indeterminate problems. Grey theory was developed by Deng (1982; 1989), which has in-

vited attention from various practitioners for dealing with vague and imprecise data sets (Liu et al., 2012;

9
Liao et al., 2013; Tsai, 2015). Prediction for periodic evaluations of any datasets can be done using grey

forecasting methods as it neither depends on the source nor on how the data values are determined (Lee et

al., 2014; Tsai and Lu, 2015; Xia et al., 2015). Grey prediction methods can be well applied to a set of

periodic measures to forecast the values of the coming periods, subject to a test for several measures of

error. Error measures ensure the best fit of data values to achieve effective prediction (Liu and Forrest,

2010; Samvedi and Jain, 2013). In most cases, there may be external disturbances in the data so that the

data values show faster or slower evolution tendencies that may not reflect the system’s behavior. Grey

prediction methods can handle such situations and a series of error checks will ensure the true evolution

of the system is understood. The step-by-step procedure for forecasting using grey prediction and the

evaluation of sequences is shown next.

4.1. Step 1: Estimate the Periodic Indicators of Resilience Performances

The data scientists for the supply chain will extract the signs of the resilience indicators based on the

measures discussed earlier in this article. The datasets provide yearly indicators of the resilience perfor-

mance of the firm for the period of 1 ≤ / ≤ '. Let:


L ?M@ = 4 ?M@ ?/ @"
NO
(13)

If 4 ?M@ ?2@ represents the zero-order yearly indicator for the second period, then it will be a period estimate

for a measure. With the general time sequence:

L = Q4?1@, 4 ?2@, 4 ?3@ … , 4 ?'@U (14)

Then, for example, 4 ?2@ can represent the indicator for the second period.

4.2. Step 2: Apply Strengthening or Weakening Operators

In the periodic sequences, due to uncontrollable shocks, the data that is obtained may evolve too

quickly or slowly, and not actually represent the true changes in the system over time. Conclusions draw

from those sequences may not be usable to build models or to make predictions as a result. In such cases,

to eliminate the interference of shock, it is necessary to apply strengthening sequence operators or weak-

ening sequence operators, so that the accuracy of predictions can be improved.

10
 To illustrate this, we let V be a sequence operator on L such that:

LV = ?4 ?1@2, 4 ?2@2, 4 ?3@2 … , 4 ?'@2@ (15)

A monotonically-increasing function V will be a weakening operator if:

4?@2 ≥ 4 ?@;  = 1, 2, 3 … , ' (16)

and if, for example, if 4 ?2@d ≥ 4 ?2@ for the second period, and likewise for all periods.

V will be a strengthening operator if:

4?@2 ≤ 4 ?@;  = 1, 2, 3 … , ' (17)

for example, if 4?2@2 ≤ 4?2@ for the second period, and similarly for all periods. For a monotonically-

decreasing function, the inverse of Equation 16 and 17 is true. The basic idea is that data in a monoton-

ically-increasing sequence will expand when a weakening operator is applied, and shrink when a

strengthening operator is applied.

Next, we introduce a second-order weakening operator VZ :

L ?M@ V = Q4 ?M@ ?1@2, 4 ?M@ ?2@2, 4 ?M@ ?3@2 … , 4 ?M@ ?'@2U, (18)

where

4 ?M@ ?@2 = 4 ?M@ ?@ + 4 ?M@ ? + 1@ + 4 ?M@ ? + 2@ … , 4 ?M@ ?'@",

O
8 [O
(19)

and

L ?M@ VZ = Q4 ?M@ ?1@2Z , 4 ?M@ ?2@2Z , 4 ?M@ ?3@2Z … , 4 ?M@ ?'@2Z U (20)

where

4 ?M@ ?@2Z = 8 Q4 ?M@ ?@2 + 4 ?M@ ? + 1@2 + 4 ?M@ ? + 2@2 … , 4 ?M@ ?'@2U
O
[O
(21)

] with the sequence of values:

Next, let L ?M@ VZ be represented by a sequence L

] = 4^ ?M@ ?1@, 4^ ?M@ ?2@, 4^ ?M@ ?3@ … , 4^ ?M@ ?'@"

L ?M@ VZ = L (22)

So, for example, 4^ ?M@ ?2@ can represent an indicator for the second period after applying the second-order

weakening operator.

11
4.3. Step 3: Obtain the First Aggregate Generator Operator for the Sequence

] can be represented as:

The first aggregate generator sequence (1-AGO) of L

] ?O@ = 4^ ?O@ ?1@, 4^ ?O@ ?2@, 4^ ?O@ ?3@ … , 4^ ?O@ ?'@"

L (23)

where

4^ ?O@ ?@ = ∑NO 4^ ?M@ ?/ @ ;  = 1, 2, 3 … , ' (24)

For example, 4^ ?O@ ?2@ can represent the indicator for the second period after applying the second-order

weakening operator and the aggregated generator operator.

4.4. Step 4: Obtain the Mean Generated Sequence

] ?O@ is given by:

The mean generated sequence considering consecutive neighbors of L

L` ?O@ = 4` ?O@ ?1@, 4` ?O@ ?2@, 4` ?O@ ?3@ … , 4` ?O@ ?'@" (25)

where

4` ?O@ ?1@ = B0.5 × 4^ ?O@ ?@" + 0.5 × 4^ ?O@ ? − 1@"C ;  = 2, 3, 4 … , ' (26)

And then 4` ?O@ ?2@ can represent the indicator for the second period after applying the second-order

weakening operator, the aggregate generator operator and the mean generator operator.

4.5. Step 5: Simulate the Sequence of Performance Indicators

Simulate the data values using grey prediction, employing matrices Y and B constructed as follows:

4^ ?M@ ?2@ −4` ?O@ ?2@ 1

g ?M@ k g ?O@ k
f4^ ?3@j f−4` ?3@ 1j
with d = f4^ ?M@ ?4@j ; l = f−4` ?O@ ?4@ 1jj,
f j f
f ⋮ j ⋮ ⋮j
(27)
f
e4^ ?M@ ? @
' i e−4` ?'@
?O@ 1i

+ 4^ ?O@ = 5,
^ ?m@

Assuming that (28)

then the estimated values of , 5 are determined using least squares as:

qn = o , 5p! ,
(29)

where n represents sequence of parameters calculated as:

n = ?ol! lp8O l! d@ (30)

12
 After plugging in appropriate values, the following expression results:

g sBtumQ∑wxy ^ ? @×∑wxy ` ? @UC8∑wxy^ ? @×` k

m ?m@?
t ?v@ t ?m@ t ?v@ @""z
fr ?m@? @|y 8 m {∑t y }j
n = f wxy{`
∑t
tum wxy
` ?m@? @| "
j
f j
(31)

B8O Q∑NZ 4^ ?M@ ?@ + ∑NZ 4` ?O@ ?@UC

O
e i

By estimating the parameters , 5, the time response sequence can be constructed for GM (1, 1):

4n ?O@ ? + 1@ = B~4^ ?M@ ?1@ − "

8 + "C ;  = 0, 1,2,3 … , ' − 1
 
(32)

4n ?M@ ? + 1@ = 4n ?O@ ? + 1@ − 4n ?O@ ?@ (33)

From Equation 32 and 33, a simulated sequences can be generated as:

L€ = Q4n?M@ ?@U NO

(34)

For example, xn ?M@ ?2@ can represents the simulated indicator for the second period.

4.6. Step 6: Estimate the Measures of Error

The error estimates are then calculated to check the accuracy of the simulation model. Five indicators

for errors are measured and checked for their values to achieve an effective prediction model for forecast-

ing. The sequence of errors is calculated as:


‚ ?M@ = Q‚ ?M@ ?@U NO
(35)

where

‚ ?M@ ?@ = Q4^ ?M@ ?@ − 4n ?M@ ?@U (36)

with the sequence of relative errors

∆= ?„ @NO (37)

where

…† ?v@? @…
„ = ^ ?v@? @
(38)

with the mean relative error

‡ = O ?∑NO „ @
∆ 
(39)

and the filtering error

13
 ∅ = „ (40)

By checking all these errors relative to their permissible limits of less than 0.01, the accuracy of the

simulation can be determined and must be greater than 0.9. If it is greater than 0.9, the absolute degree

] '2 L€ needs to be calculated as;

?‰@ of grey incidences of L

|0| = ‹Q∑8O ^ ?@ − 4^?1@pU + Z o4^ ?'@ − 4^?1@p"‹

NZ o4
O
(41)

|0̂ | = ‹Q∑8O n?@ − 4n?1@pU + Z o4n?'@ − 4n?1@p"‹

NZ o4
O
(42)

|0̂ − 0| = ‹Q∑8O
NZ?o4
n?@ − 4n?1@p − o4^ ?@ − 4^?1@p@U + BZ ?o4n?'@ − 4n?1@p − o4^ ?'@ − 4^?1@p@C‹ (43)
O

so that

O[||[|̂ |
‰=
O[||[|̂ |[|̂ 8|
. (44)

The level of incidence also should be at Level 1 (> 0.9).

Next is to compute the ratio of mean squared deviations (C):

4̅ =  ∑NO 4^ ?M@ ?@

O
(45)

Z

OZ =  ∑NO{4^ ?M@ ?@ − 4̅ |
O
(46)

and

‚̅ = ∑NO ‚ ?M@ ?@@

O

(47)

Z

ZZ =  ∑NO{‚ ?M@ ?@ − ‚̅|
O
(48)

6
C = 6y, (49)
m

also should be at Level 1 ?< 0.35@.

Then we compute the additional small error probability:

∃= 0.6745 ×
O

(50)

and check that Q-…‚ ?M@ ?@ − ‚̅… < ∃U (51)

should be at Level 1 (> 0.95).

14
4.7. Step 7: Predict the Values for Resilience Performance Indicators for the Coming Periods

With all the error checks satisfied, we can directly apply the grey model as in Equations 19 and 20) to

predictions of values for next ; periods via:

L€ ?M@ = Q4n?M@ ?/@UN

[
(52)

Here, for example, 4n ?M@ ?6@ can represent the predicted indicator for the sixth lagged period in the future.

The methodology is shown as a flow diagram in Figure 1 so the reader will see the steps involved.

INSERT FIGURE 1 ABOUT HERE

5. A CASE EVALUATION FOR A SUPPLY CHAIN IN AN INDIAN ELECTRONIC COMPANY

5.1. Descriptive Background

The prediction model is applied to a firm’s supply chain in India, as a way of evaluating the forecast

of its measures for supply chain resilience performance. The firm is a leading manufacturer of electronics,

especially in the smartphone and tablet sector in India, and is a player in the global markets. Its supply

chain has faced the ill-effects of disruptive events in recent years. This occurred due to shortages from

second-tier suppliers related to raw materials. The firm had contingency plans to recover from the situa-

tion. And because it operates in the international markets and those markets are competitive and relatively

volatile, its products require frequent innovation. In addition, its customers are in need of better services.

The firm has a dedicated team of data scientists to track and analyze data for operational-level disrup-

tions. The yearly data indicators of the firm for five indicators of supply chain resilience were extracted

and taken for analysis. The period of analysis of the data sets was from 2010 to 2015. The policies of the

firm prohibited us from revealing the source of the data or how the firm would have measured resilience

performance from the indicators. Since the data that were available for this research was to show how the

resilience indicators in the data evaluation were built, I used a methodology that was suited to data predic-

tion.

5.2. Background of the Data for Evaluation

15
 Extract, transform and load (ETL) is the most common method for transferring data from the source

to the data warehouse. Extract, load and transform (ELT) was employed instead of ETL, as the volume of

data involving the performance measures and their operational indicators was very large. Also, the data-

base engine was well equipped for processing massive amounts of data. Unlike in ETL, where transfor-

mations are processed by ETL tools, ELT has a target data source to make the required transformations.

This minimizes processing time at the source. Even though there was a lot of disk space needed for the

staging table, ELT was advantageous since the source was a production system that could have had an

impact on the experience of users.

The process involved extracting data from the source to measuring all of the indicators at various stra-

tegic, tactical and operational levels in the supply network. (We discussed some of the key measures ear-

lier.) The indicators were loaded into staging tables in the databases, and then were transformed for load-

ing into the target databases. The process of moving data from the staging tables to data warehouses was

done using T-SQL for ease of transformation and better performance. Indicators at a strategic level were

obtained through incidence checks and by aggregating the quantitative measures contributing to it at the

tactical and operational levels.

The process of tracking data via ELT was done with the aid of data scientists. The detailed infor-

mation on the background processes was not available per the non-disclosure policy of the firm. The ag-

gregate periodic indicators were the only raw data for analysis. They represented the period by period re-

silience performance of the firm as indicators of flexibility, responsiveness, quality, productivity and ac-

cessibility. I had to analyze the data without proper source information. So the use of a grey prediction

models was appropriate to obtain the predicted the indicators for future periods subject to error checks,

and to analyze secondary data when key information is unknown.

5.3. Analyzing the Secondary Data for Prediction

We implemented the methodology described in Section 4 to forecast the supply chain resilience per-

formance indicators of the firm for future periods. The evaluation steps follow.

16
• Step 1: Three data scientists extract periodic performance for the five indicators of resilience for

flexibility, responsiveness, quality, productivity and accessibility (Equation 1-13). The extracted

values indicators from Equations 13 and 14 were for 2010 to 2015. See Table 3.

INSERT TABLE 3 ABOUT HERE

• Step 2: The first-order weakening operator was applied to Equations 15-17 for the indicators to

smooth their values. Then a second-order weakening operator was applied to Equations 18-22 to

further smooth the data values for prediction. See Tables 4 and 5.

INSERT TABLES 4 AND 5 ABOUT HERE

• Step 3: The first aggregate generator sequences were obtained for the indicators from Equations

23 and 24. The data values correspond to the AGO sequences in Table 6.

INSERT TABLE 6 ABOUT HERE

• Step 4: The mean generated sequences of the data were obtained from Equations 25 and 26.

Their values are shown in Table 7.

INSERT TABLE 7 ABOUT HERE

• Step 5: The sequences of indicators for 2010 to 2015 were simulated using Equations 27-34. The

time response sequences and simulated data sequences are shown in Tables 8 and 9.

INSERT TABLES 8 AND 9 ABOUT HERE

• Step 6: The error sequences were estimated using Equations 35-51. The values of the estimated

and relative errors are shown in Tables 10 and 11. The mean relative and filtering errors all were

at the acceptable threshold less than 0.01. The absolute degrees of grey incidences were calculat-

ed and checked for the acceptable limits of greater than 0.90. The ratios of the mean squared de-

viations were calculated and were at the acceptable threshold of less than 0.35. The small error

probabilities were calculated and tested for acceptable limits of greater than 0.95. Since all the er-

rors of prediction lying in the acceptable limits, we can ensure Level 1prediction for the coming

periods. See Tables 12-15 for the results.

17
 INSERT TABLES 10-15 ABOUT HERE

• Step 7: The generated time response sequence and predicted sequence of simulated values of the

resilience indicators from Equation 52, finally, are shown in Tables 16 and 17.

INSERT TABLES 16 AND 17 ABOUT HERE

6. DATA ANALYTICS FRAMEWORK AND MODEL EVALUATION

During the process of contextualization, we made relevant configuration parameters available for the

simulation. Here meta-data and user-data are defined. Cloud computing capabilities, in this case, shielded

the user data from different virtual machines. The cloud infrastructure helped to interpret the user data

and the different entities required to interpret them. The data after the analysis were aggregated to repre-

sent the sum of the indicators of an attribute for a period. Prediction was done based on secondary data

and the results were obtained based on an improved grey prediction model. Primary data were not provid-

ed, and information on source and processing were not disclosed for the secondary data that we used.

I built an evaluation framework recognizing the big data and limited richness of the background in-

formation acquired. So I carried out assessment based on descriptive scales with the aid of data scientists

from the firm. Those who worked on its supply chain for more than five years were selected for assess-

ment. Depending on the value chain processes in the supply chain, I classified the processes in three are-

as: (1) the upstream supply chain, handling supplier relations and data management; (2) the internal sup-

ply chain, handling data within the firm; and (3) the downstream supply chain, handling data related to

customer relations and management. Two data scientists from each area were identified considering their

experience and availability. They rated the supply chain of the firm for its data analytics performance and

the management of its big data based on five dimensions of value. 1

1
These dimensions are attributable to IBM and other industry consultants, who have used it to sell the business value of data
analytics to organizations in services, manufacturing, government and education. Volume the amount of data generated over time.
(2) Velocity refers to the speed of data generation and its movement. (3) Variety is the type of data that are used, including struc-
tured and unstructured data. (4) Veracity refers to the trustworthiness of the data for use in the creation of insights through analyt-
ics. Quality and accuracy are harder to control when the amount of data increases, it flows faster, or it is less structured. Finally,
reliability is the extent to which data need to be checked and validated before they are used for evaluation.

18
 The respondents gave ratings on a five-point scale varying very good to very poor for the different pe-

riods the data were intended to cover. Their evaluations were converted to grey scales using the conver-

sion scale presented in Table 18. The ratings obtained for the the period are shown in Table 19, and were

converted to grey values, as shown in Table 20. The values were aggregated and slightly more weight was

assigned to the internal supply chain processes (0.4) since more data were generated within the plant. The

other processes, the upstream portion and downstream portion of the supply chain, were assigned equal

weightings of 0.3 each. The values were then further converted using mean value whitening (Deng, 1989)

for analysis via the five dimensions of business value. The sequential processes are shown in Tables 21

and 22. By setting a threshold of 0.5 for evaluating the average effectiveness of the data for the periods,

the effectiveness of the use of data analytics appears to have been justified. All of the further converted

values for the evaluation parameters in the periods were above the threshold for acceptable reliability.

INSERT TABLES 18-22 ABOUT HERE

The results from the prediction model from the panel of experts also needed to be validated. They

were given the flexibility to rate the indicators on five-point scales. The scaled responses obtained were

converted into grey values using the conversion in Table 18, and are shown in Table 23. The grey values

were averaged based on the equal weighting of each individual analyst’s responses. The average grey val-

ues were converted by using mean value whitening, and the whitened values are shown in Table 24.

Based on a threshold of 0.5, the indicators seem to be realistic and seem close to the actual values from

the practical viewpoint of the analysts.

INSERT TABLES 23 AND 24 ABOUT HERE

7. RESULTS AND DISCUSSION

There is high demand for innovative data analytics to improve the strategy, operation and perfor-

mance of modern supply chains. Most companies that capture data-at-scale in their supply lack the know-

how to use effective and sophisticated tools to analyze them. Unfortunately, using more data in a deci-

sion-making tool does not always produce more accurate results. Appropriate analysis of the right data

19
indicators can improving the accuracy of decision-making and the exactness of predictions. Supply chain

resilience enables it to handle disruptions and a variety of unpleasant operational events. Operational-

level measures contribute to the potentially large set of data that are needed for aggregation with appro-

priate analysis tools. In addition, aggregate data showing the indicators of supply chain performance resil-

ience performance of the firm represent the raw data for analysis.

Grey theory is appropriate when the available data must be analyzed without refard to their sources or

units. The resilience performance of the firm for future periods can be forecasted using improved grey

prediction methods. I carried out a case analysis of the use of the methodology for an Indian electronics

manufacturing supply chain. Raw data from the past, representing the indicators of resilience perfor-

mance, were used as inputs from 2010 to 2015for predicting the resilience performance of the firm in fu-

ture years. Raw data were analyzed using improved grey prediction methods and the values for the period

were simulated by checking for accuracy for five measures of errors: mean relative error, filtering error,

absolute degree of grey incidences, ratio of mean squared deviations and small error probabilities.

After all these values were confirmed to be acceptable, the simulated data could be considered as an

actual representation of the sequence of periodic data for the indicators. And the grey prediction model

could be assumed to represent the actual evolution of the data in the system. The model was used to pre-

dict the values of the indicators in the future, from 2016 to 2020. The true evolution of the data trends

were analyzed and the firm could use this information to adopt strategies for change, if the predictions di

match the performance expectations of the firm. The indicators that appeared to have a negative path of

evolution could be be given more attention to improve or diminish faltering performance.

Grey prediction models seem to work well when the data show increasing or decreasing trends. Fluc-

tuating data values can be smoothed by applying successive weakening operators on the data. The effect

of smoothing from the application of weakening operators is visible in Figures 2, 3 and 4, as the varia-

tions in the predictions are affected. Also, the slope seem to be reduced when the weakening operators are

applied, as illustrated by the gradual transition from Figure 1 to Figure 4. When the aggregate generator

operator was applied, the data values showed more of a common trend, even when the raw data showed

20
increasing, decreasing or volatile evolution in their time series. This suggests the accuracy of the simula-

tion of the indicator data sequences. The first aggregated generator sequence for trends is shown in Figure

5. The mean generated sequence is obtained to smoothen the values and can be seen in Figure 6.

INSERT FIGURES 1-6 ABOUT HERE

The aggregate response sequence over time also was obtained from simulation. The sequence of sim-

ulated values for the resilience performance indicators is shown in Figures 7 and 8. From Figure 8, the

simulated data sequences appear to resemble the actual data sequences after the second-order weakening

operators were applied. This suggests the prediction model fits the actual values better, with less probabil-

ity of fit errors. Also, the predictions for future values seem to be reliable since the model captures the

true evolution of the time series of resilience performance indicator data in the system. The predicted val-

ues for the resilience indicators are shown in Figures 9 and 10. Figure 10 shows the actual evolution of

the resilience performance sequences for 2016 to 2020 (Periods 6 to 10 on the x-axis).

INSERT FIGURES 7-10 ABOUT HERE

The simulated values for 2010 to 2015 (Periods 1 to 5) were checked in terms of the measures for er-

rors related to the main dimensions of resilience performance indicators (RPI). The error trends and rela-

tive errors are shown in Figures 11 and 12. Note that RPI 4 shows the maximum values for the error

trends, while RPI 2 and RPI 4 show the maximum values for the relative errors. My quantitative checks

of the error measure with respect to the expected maximum limits to achieve Level 1 prediction are

shown in Figures 13-16. One can observe from the graphs that the mean relative and filtering errors fell

less over time based on the check values, and the absolute degrees of the grey incidences were higher

based on the check values. Also, the ratio of the mean squared deviations mostly fell within permissible

limits and the small error probabilities seemed to be satisfactory in terms of the check levels.

INSERT FIGURES 11-16 ABOUT HERE

8. CONCLUSIONS

Data analytics is gaining increased acceptability in modern supply chains. It is essential to identify

21
input data to support obtaining usable outputs. The selection of proper data analysis tools is essential to

deal with big data. This article presents an improved grey prediction model for a case to forecast the resil-

ience performance indicators of a firm. The prediction model is beneficial as the values of prediction nei-

ther depend on the source of data nor on how the system has generated it. The prediction of values de-

pends only on the patterns of the data sequences, which is a major advantage of a grey prediction model.

This article offers two scientific contributions for the practice of supply chain resilience and opera-

tions prediction. One is to identify the operational level indicators of supply chain resilience and to aggre-

gate them as measures for strategic-level indicators of flexibility, responsiveness, quality, productivity

and accessibility. The second is to analyze the measures of performance indices based on secondary data

from big data analytics. Identifying the operational level indicators of supply chain resilience allows man-

agers to analyze the disturbances and these aid them in taking timely and necessary actions. Predictions of

strategic-level performance further enable managers to identify the firm future trend, and help them to

give more attention to those indicators that suggest problematic trends.

Five major indicators of resilience performance of the firm were analyzed in this research. They were

assessed for several operational measures of performances. The results offer implications for managers in

practice. They can identify the performance indicators for future periods. If the indicators do not show

satisfactory performance, proper planning and change strategies can be adopted. Validation of the predic-

tions is important too. For the firm we studied, the indicators of flexibility, responsiveness and accessibil-

ity increased for future periods, whereas the indicators for quality and productivity showed slight decreas-

es.

The firm operates in volatile markets and its products are in need of constant innovations. So flexibil-

ity is the primary need to attain market share and to achieve sustainable competitiveness. But, flexibility

only comes at a cost. These increasing costs suggested diminishing productivity indicators for the firm

going forward. Similarly, responsiveness and accessibility were more preferred attributes from the cus-

tomer view. So these measures deserved attention in the supply chain. We saw this in the increases in

22
their forecast values. Apart from these things, product cost is a major attribute in determining the accepta-

bility of the product.

Constant effort to reduce the cost of the products is needed to sustain the market share, espe-

cially in the Indian context. This is achieved by slightly reducing high-quality components by

substituting good-quality alternatives that do not affect the performance of products. This strate-

gy indicates the reason for the slight dip in the quality indicators, resulting in a larger market

share in the lower-cost segments. For 2016, the flexibility, responsiveness and accessibility indi-

cators showed increasing trends of 2.58%, 5.87% and 1.94% compared to the previous year,

whereas the quality and productivity indicators show dipping trends of 6.89%and 3.99%, as

compared to the 2015 indicators. The results of the analysis are useful for their managerial level

implications for improvements and changes.

The model has some limitations also. The data predictions do not necessarily consider the

sources of the data. So the companies are recommended to do an operational level analysis of the

values of prediction to achieve more insights into the results for practice. Apart from that, it is

possible to implement weighing schemes for various operational level indicators to obtain the

major indicators of resilience performance. So managers may wish to assign importance to those

measures for which the firm and its managers express more interest. The prediction model can be

extended to include sustainability measures also. This can help a firm to develop a sustainable-

resilient supply chain in practice.

REFERENCES
Ambulkar, S., Blackhurst, J., and Grawe, S. (2015). Firm's resilience to supply chain disruptions:
Scale development and empirical examination. Journal of Operations Management, 33,
111-122.
Brandon‐Jones, E., Squire, B., Autry, C. W., and Petersen, K. J. (2014). A contingent re-
source‐based perspective of supply chain resilience and robustness. Journal of Supply Chain
Management, 50(3), 55-73.
23
Cantor, E. D., Blackhurst, J., Pan, M., and Crum, M. (2014). Examining the role of stakeholder
pressure and knowledge management on supply chain risk and demand responsiveness. The
International Journal of Logistics Management, 25(1), 202-223.
Chambers, J. C., Mullick, S. K., and Smith, D. D. (1971). How to choose right forecasting tech-
nique. Harvard Business Review, 49(4), 45.
Chang, R. M., Kauffman, R. J., and Kwon, Y. (2014). Understanding the paradigm shift to com-
putational social science in the presence of big data. Decision Support Systems, 63, 67-80.
Chen, H., Chiang, R. H., and Storey, V. C. (2012). Business intelligence and analytics: from big
data to big impact. MIS Quarterly, 36(4), 1165-1188.
Christopher, M., and Ryals, L. J. (2014). The supply chain becomes the demand chain. Journal
of Business Logistics, 35(1), 29-35.
Davenport, T. (2014). Big Data at Work: Dispelling the Myths, Uncovering the Opportunities.
Harvard Business Review Press, Boston, MA.
Demirkan, H., and Delen, D. (2013). Leveraging the capabilities of service-oriented decision
support systems: putting analytics and big data in cloud. Decision Support Systems, 55(1),
412-421.
Deng, J. L. (1982). Control problems of grey systems. Systems and Control Letters, 1(5), 288-
294.
Deng, J. L. (1989). Introduction to grey system theory. The Journal of Grey Systems, 1(1), 1-24.
Dong, H., Wu, M., Ding, X., Chu, L., Jia, L., Qin, Y., and Zhou, X. (2015). Traffic zone division
based on big data from mobile phone base stations. Transportation Research Part C:
Emerging Technologies, 58, 278-291.
George, G., Haas, M. R., and Pentland, A. (2014). Big data and management. Academy of Man-
agement Journal, 57(2), 321-326.
Goude, Y., Nedellec, R., and Kong, N. (2014). Local short and middle term electricity load fore-
casting with semi-parametric additive models. IEEE Transactions on Smart Grid, 5(1), 440-
446.
Groves, W., Collins, J., Gini, M., and Ketter, W. (2014). Agent-assisted supply chain manage-
ment: Analysis and lessons learned. Decision Support Systems, 57, 274-284.
Guerard, J. B., Rachev, S. T., and Shao, B. P. (2013). Efficient global portfolios: big data and
investment universes. IBM Journal of Research and Development, 57(5), 11-1.
Hazen, B. T., Boone, C. A., Ezell, J. D., and Jones-Farmer, L. A. (2014). Data quality for data
science, predictive analytics, and big data in supply chain management: an introduction to
the problem and suggestions for research and applications. International Journal of Produc-
tion Economics, 154, 72-80.
Hazen, B. T., Boone, C. A., Ezell, J. D., and Jones-Farmer, L. A. (2014). Data quality for data
science, predictive analytics, and big data in supply chain management: an introduction to
the problem and suggestions for research and applications. International Journal of Produc-
tion Economics, 154, 72-80.

24
Hofmann, E. (2015). Big data and supply chain decisions: the impact of volume, variety and ve-
locity properties on the bullwhip effect. International Journal of Production Research, 1-19.
DOI:10.1080/00207543.2015.1061222.
Hohenstein, N. O., Feisel, E., Hartmann, E., and Giunipero, L. (2015). Research on the phenom-
enon of supply chain resilience: a systematic review and paths for further investiga-
tion. International Journal of Physical Distribution and Logistics Management, 45(1/2), 90-
117.
Isasi, G., Khouri, N., Morosini Frazzon, E., and Uriona, M. (2015). Big data and business analyt-
ics in the supply chain: a review of the literature. Latin America Transactions, IEEE
(Revista IEEE America Latina), 13(10), 3382-3391.
Ivanov, D., Sokolov, B., and Dolgui, A. (2014). The Ripple effect in supply chains: trade-off ‘ef-
ficiency-flexibility-resilience’ in disruption management. International Journal of Produc-
tion Research, 52(7), 2154-2172.
Johnson, N., Elliott, D., and Drake, P. (2013). Exploring the role of social capital in facilitating
supply chain resilience. Supply Chain Management: An International Journal, 18(3), 324-
336.
Kardakos, E. G., Alexiadis, M. C., Vagropoulos, S. I., Simoglou, C. K., Biskas, P. N., and
Bakirtzis, A. G. (2013). Application of time series and artificial neural network models in
short-term forecasting of PV power generation. In Power Engineering Conference (UPEC),
2013 48th International Universities (pp. 1-6). IEEE Computing Society Press, Washington,
DC.
Kaynak, S., and Yin, S. (2015). Big data for modern industry: challenges and trends: point of
view. Proceedings of the IEEE, 103(2), 143-146.
Kim, S. (2016). Forecasting short-term air passenger demand using big data from search engine
queries. Automation in Construction, 70, 98-108.
Kwon, O., Lee, N., and Shin, B. (2014). Data quality management, data usage experience and
acquisition intention of big data analytics. International Journal of Information Manage-
ment, 34(3), 387-394.
Lee, Y. C., Wu, C. H., and Tsai, S. B. (2014). Grey system theory and fuzzy time series forecast-
ing for the growth of green electronic materials. International Journal of Production Re-
search, 52(10), 2931-2945.
Li, J., Tao, F., Cheng, Y., and Zhao, L. (2015). Big data in product lifecycle management. The
International Journal of Advanced Manufacturing Technology, 81(1-4), 667-684.
Li, X., Pan, B., Law, R., and Huang, X. (2017). Forecasting tourism demand with composite
search index. Tourism Management, 59, 57-66.
Liao, M. S., Liang, G. S., and Chen, C. Y. (2013). Fuzzy grey relation method for multiple crite-
ria decision-making problems. Quality and Quantity, 47(6), 3065-3077.
Liu, S., and Forrest, J. Y. L. (2010). Grey Systems: Theory and Applications. Springer, Berlin,
Germany.

25
Liu, S., Forrest, J. Y. L., and Yang, Y. (2012). A brief introduction to grey systems theory. Grey
Systems: Theory and Application, 2(2), 89-104.
Manrique, P., Qi, H., Morgenstern, A., Velasquez, N., Lu, T. C., and Johnson, N. (2013). Con-
text matters: improving the uses of big data for forecasting civil unrest: emerging phenome-
na and big data. In 2013 IEEE International Conference on Intelligence and Security Infor-
matics (ISI), (pp. 169-172). IEEE Computer Science Press, Washington, DC.
Melnyk, S. A. (2014). Understanding supply chain resilience. Supply Chain Management Re-
view, 18(1).
Pereira, R. C., Christopher, M., and Lago Da Silva, A. (2014). Achieving supply chain resilience:
the role of procurement. Supply Chain Management: An International Journal, 19(5/6),
626-642.
Pettit, T. J., Croxton, K. L., and Fiksel, J. (2013). Ensuring supply chain resilience: development
and implementation of an assessment tool. Journal of Business Logistics, 34(1), 46-76.
Ponis, S. T., and Koronis, E. (2012). Supply chain resilience: definition of concept and its forma-
tive elements. Journal of Applied Business Research, 28(5), 921.
Qi, J. Y., Qu, Q. X., and Zhou, Y. P. (2014). How does customer self-construal moderate CRM
value creation chain? Electronic Commerce Research and Applications, 13(5), 295-304.
Rajesh, R., and Ravi, V. (2015). Supplier selection in resilient supply chains: a grey relational
analysis approach. Journal of Cleaner Production, 86, 343-359.
Salmon, K. (2015). Creating opportunities using big data and analytics. Risk and Compliance
Magazine, January-March 2015.
Samvedi, A., and Jain, V. (2013). A grey approach for forecasting in a supply chain during in-
termittent disruptions. Engineering Applications of Artificial Intelligence, 26(3), 1044-1051.
Sanders, N. R. (2014). Big Data Driven Supply Chain Management: A Framework for Imple-
menting Analytics and Turning Information into Intelligence. Pearson Education, Upper
Saddle River, NJ.
Schoenherr, T., and Speier‐Pero, C. (2015). Data science, predictive analytics, and big data in
supply chain management: current state and future potential. Journal of Business Logis-
tics, 36(1), 120-132.
Scholten, K., and Schilder, S. (2015). The role of collaboration in supply chain resilience. Supply
Chain Management: An International Journal, 20(4), 471-484.
Scholten, K., Sharkey Scott, P., and Fynes, B. (2014). Mitigation processes–antecedents for
building supply chain resilience. Supply Chain Management: An International Jour-
nal, 19(2), 211-228.
Simmhan, Y., Aman, S., Kumbhare, A., Liu, R., Stevens, S., Zhou, Q., and Prasanna, V. (2013).
Cloud-based software platform for big data analytics in smart grids. Computing in Science
and Engineering, 15(4), 38-47.
Tan, K. H., Zhan, Y., Ji, G., Ye, F., and Chang, C. (2015). Harvesting big data to enhance supply
chain innovation capabilities: an analytic infrastructure based on deduction
graph. International Journal of Production Economics, 165, 223-233.

26
Tien, J. M. (2013). Big data: Unleashing information. Journal of Systems Science and Systems
Engineering, 22(2), 127-151.
Tsai, C. F., and Lu, S. L. (2015). Novel grey models for the trend forecast of Taiwan waste gas
apparatus. International Journal of Environmental Technology and Management, 18(2),
170-184.
Tsai, S. B. (2015). Using grey models for forecasting China’s growth trends in renewable energy
consumption. Clean Technologies and Environmental Policy, 1-9.
Varian, H. R. (2014). Big data: new tricks for econometrics. Journal of Economic Perspec-
tives, 28(2), 3-27.
Vera-Baquero, A., Colomo-Palacios, R., and Molloy, O. (2013). Business process analytics using
a big data approach. IT Professional, 15(6), 29-35.
Waller, M. A., and Fawcett, S. E. (2013). Data science, predictive analytics, and big data: a revo-
lution that will transform supply chain design and management. Journal of Business Logis-
tics, 34(2), 77-84.
Waller, M. A., and Fawcett, S. E. (2013a). Click here for a data scientist: big data, predictive an-
alytics, and theory development in the era of a maker movement supply chain. Journal of
Business Logistics, 34(4), 249-252.
Waller, M. A., and Fawcett, S. E. (2013b). Data science, predictive analytics, and big data: a
revolution that will transform supply chain design and management. Journal of Business
Logistics, 34(2), 77-84.
Wang, L., and Alexander, C. A. (2015). Big data driven supply chain management and business
administration. American Journal of Economics and Business Administration, 7(2), 60.
Wieland, A., and Marcus Wallenburg, C. (2013). The influence of relational competencies on
supply chain resilience: a relational view. International Journal of Physical Distribution and
Logistics Management, 43(4), 300-320.
Wixom, B., Ariyachandra, T., Douglas, D., Goul, M., Gupta, B., Iyer, L., and Turetken, O.
(2014). The current state of business intelligence in academia: the arrival of big da-
ta. Communications of the Association for Information Systems, 34(1), 1.
Wixom, B., Ariyachandra, T., Douglas, D., Goul, M., Gupta, B., Iyer, L., Kulkarni, U. Mooney,
J. G., Phillips-Wren, G., and Turetken, O., and Turetken, O. (2014). The current state of
business intelligence in academia: the arrival of big data. Communications of the Associa-
tion for Information Systems, 34(1), 1.
Xia, T., Jin, X., Xi, L., Zhang, Y., and Ni, J. (2015). Operating load based real-time rolling grey
forecasting for machine health prognosis in dynamic maintenance schedule. Journal of In-
telligent Manufacturing, 26(2), 269-280.
Yin, R. K. (2013). Case Study Research: Design and Methods. Sage, Thousand Oak, CA.
Zheng, Y., Yi, X., Li, M., Li, R., Shan, Z., Chang, E., and Li, T. (2015). Forecasting fine-grained
air quality based on big data. In Proceedings of the 21th ACM SIGKDD International Con-
ference on Knowledge Discovery and Data Mining (pp. 2267-2276). ACM Press, New
York, NY.

27
 Analyze the indicators of resili-
ence performance of companies
and its measures

Estimate the periodic indicators of

resilience performance employing
data analytics

Apply strengthening/ weakening

operators

Construct first aggregation gen-

erator of the sequences

Construct the mean generated se-

quences

Simulate the performance indica-

tor sequences

Calculate the small error probabil- Calculate the mean relative

ities errors
Figure 1: The methodology for grey prediction
Estimate the
measures of error

Estimate the ratio of mean squared Calculate the filtering

deviations errors

Estimate absolute degree of grey

incidences

Ensure a level one prediction model

Estimate the indicator sequences for future periods

29
Figure 2: Series of performance indicators

Figure 3: Series of first order weakening of performance indicators

Figure 4: Series of second order weakening of performance indicators

30
Figure 5: First aggregation generator sequence of performance indicators

Figure 6: Mean generated sequence of performance indicators

Figure 7: Time response sequence of performance indicators

31
Figure 8: Simulated sequence of performance indicators

Figure 9: Predicted time response sequences (cumulative)

Figure 10: Predicted sequences of performance indicators

32
Figure 11: Error trends of simulated sequences

Figure 12: Relative errors of simulated sequences

Figure 13: Mean relative error and filtering error checks

33
Figure 14: Absolute degree of grey incidence checks

Figure 15: Ratio of mean squared error checks

Figure 16: Small error probability checks

34
Table 1: Big data and supply chain management
Sl References Area/ Application Key benefit(s) to supply chains
No.
1 Hazen et al. Data quality for data science, pre- Enhance supply chain performance
(2014) dictive analytics and big data and functions
2 Waller and Management theory and practice of Enhance supply chain design and
Fawcett data science, predictive analytics management functions
(2013b) and big data
3 Waller and Big data, predictive analytics and its Execute the right supply chain
Fawcett relation to changing nature of pro- strategies with greater precision and
(2013a) ducers efficiency
4 Demirkan Tying information technology strat- Enhance agility of supply chain sys-
and Delen egy perspectives to database and tems through service oriented deci-
(2013) design science perspectives sion support systems
5 Yin and Analyze applications, challenges Research directions are proposed
Kaynak and threats for big data in modern for building efficient strategies
(2015) industries
6 Christopher Implication of additive manufactur- Build supply chains running on
and Ryals ing and big data for information lower inventory and fast customer
(2014) management response
7 Wixom et al. Analyze the needs to respond to Enhance responsiveness in supply
(2014) emerging markets using business chains
analytics and business intelligence
8 Kwon et al. Explain acquisition of big data from Expose positive relation of big data
(2014) perspectives of data quality man- adoption to competence of firm for
agement and usage experience quality of corporate data
9 Li et al. Study application of big data in Enhance intelligence and efficiency
(2015) product lifecycle management of design, production, and service
processes
10 Groves et al. Use of agent-based competitive Propose business key performance
(2014) simulation as a tool to develop com- indicators (KPIs) to analyze market
plex decision making strategies dynamics

Table 2: Big data and prediction modeling

Sl References Area Application
No.

35
Sl References Area Application
No.
1 Varian Econometric analysis Use of machine learning techniques for modeling
(2014) of complex relationships to aid prediction models
2 Simmhan et Smart power grids Perform intelligent demand-side management and
al. (2013) relieve peak load using agile demand forecasting
3 Zheng et al. Air quality monitoring Predictive model of a linear regression-based
(2015) system temporal predictor to model the local factors of
air quality
4 Manrique et Social dynamic be- Model for predicting civil unrest events using
al. (2013) havior open source indicators
5 Goude et al. Electricity load con- Semi-parametric approach based on generalized
(2014) trol additive models theory to forecast electrical load
of distribution networks
6 Kim (2016) Air passenger demand Short-term air passenger demand forecasting
model model developed from search queries to identify
short-term fluctuations
7 Guerard et Stock market risk and Expected return models are developed for global
al. (2013) returns equity markets using a certain stock-selection
model
8 Dong et al. Information traffic Improve the analysis of complex travel patterns
(2015) analysis and behaviors for travel demand modeling and
transportation planning
9 Li et al. Tourist demand man- Creating a composite search index uses a general-
(2017) agement ized dynamic factor model and tests the frame-
work in predicting tourist volumes.
10 Kardakos et Photo voltaic power Proposes forecasting methods for electricity gen-
al. (2013) generation eration using time series analysis and artificial
neural networks of grid-connected PV plants

Table 3: Periodic resilience performance indicators of case firm

Sl no. x(0)1 x(0)2 x(0)3 x(0)4 x(0)5
RPI 1 2386 2399 3453 3645 4064
RPI 2 1345 1876 2345 2784 3764
RPI 3 674 567 453 363 278
RPI 4 6784 5647 4563 3743 3345
RPI 5 1763 2376 2763 2983 3216

36
Table 4: First order weakening operator sequence of periodic indicators
Sl no. x(0)1d x(0)2d x(0)3d x(0)4d x(0)5d
RPI 1 3189.4 3390.25 3720.667 3854.5 4064
RPI 2 2422.8 2692.25 2964.333 3274 3764
RPI 3 467 415.25 364.667 320.5 278
RPI 4 4816.4 4324.5 3883.667 3544 3345
RPI 5 2620.2 2834.5 2987.333 3099.5 3216

Table 5: Second order weakening operator sequence of periodic indicators

Sl no. x(0)1d2 x(0)2d2 x(0)3d2 x(0)4d2 x(0)5d2
RPI 1 3643.763 3757.354 3879.722 3959.25 4064
RPI 2 3023.477 3173.646 3334.111 3519 3764
RPI 3 369.083 344.604 321.056 299.25 278
RPI 4 3982.713 3774.292 3590.889 3444.5 3345
RPI 5 2951.507 3034.333 3100.944 3157.75 3216

”?•@ ?•@
“ ”?•@ ?–@
“ ”?•@ ?—@
“ ”?•@ ?˜@
“ ”?•@ ?™@
“
Table 6: First accumulating generator sequence of periodic indicators
Sl no.
RPI 1 3643.763 7401.117 11280.84 15240.09 19304.09
RPI 2 3023.477 6197.123 9531.234 13050.23 16814.23
RPI 3 369.083 713.687 1034.743 1333.993 1611.993
RPI 4 3982.713 7757.005 11347.89 14792.39 18137.39
RPI 5 2951.507 5985.84 9086.784 12244.53 15460.53

“` ?•@ ?•@ “` ?•@ ?–@ “` ?•@ ?—@ “` ?•@ ?˜@

Table 7: Mean generated sequence of periodic indicators
Sl no.
RPI 1 5522.44 9340.978 13260.46 17272.09
RPI 2 4610.3 7864.179 11290.73 14932.23
RPI 3 541.385 874.215 1184.368 1472.993
RPI 4 5869.859 9552.45 13070.14 16464.89
RPI 5 4468.674 7536.312 10665.66 13852.53

š?•@ ?•@
“ š?•@ ?–@
“ š?•@ ?—@
“ š?•@ ?˜@
“ š?•@ ?™@
“
Table 8: Time response sequence of performance indicators
Sl no.
RPI 1 3643.763 7410.197 11273.91 15237.42 19303.29
RPI 2 3023.477 6181.463 9524.683 13064 16810.92
RPI 3 369.083 713.667 1034.506 1333.235 1611.379
RPI 4 3982.713 7739.84 11347.12 14810.54 18135.82
RPI 5 2951.507 5988.942 9085.259 12241.6 15459.13

37
 š?›@ ?•@
“ š?›@ ?–@
“ š?›@ ?—@
“ š?›@ ?˜@
“ š?›@ ?™@
“
Table 9: Simulated sequence of periodic indicators
Sl no.
RPI 1 3643.763 3766.434 3863.713 3963.505 4065.873
RPI 2 3023.477 3157.986 3343.22 3539.319 3746.921
RPI 3 369.083 344.584 320.839 298.729 278.144
RPI 4 3982.713 3757.127 3607.283 3463.413 3325.283
RPI 5 2951.507 3037.435 3096.317 3156.341 3217.528

œ?›@ ?•@ œ?›@ ?–@ œ?›@ ?—@ œ?›@ ?˜@ œ?›@ ?™@
Table 10: Estimated errors
Sl no.
RPI 1 0 -9.08 16.009 -4.255 -1.873
RPI 2 0 15.66 -9.109 -20.319 17.079
RPI 3 0 0.02 0.217 0.521 -0.144
RPI 4 0 17.165 -16.394 -18.913 19.717
RPI 5 0 -3.102 4.627 1.409 -1.528

• – — ˜ ™
Table 11: Relative errors
Sl no.
RPI 1 0 0.002 0.004 0.001 0
RPI 2 0 0.005 0.003 0.006 0.005
RPI 3 0 0 0.001 0.002 0.001
RPI 4 0 0.005 0.005 0.005 0.006
RPI 5 0 0.001 0.001 0 0

Table 12: Mean relative errors and Filtering errors (Level 1 (<0.01))
Sl no. Mean relative errors Filtering Errors
RPI 1 0.001 0
RPI 2 0.004 0.005
RPI 3 0.001 0.001
RPI 4 0.004 0.006
RPI 5 0 0

|ž| ž |žn| žn |žn − ž| Ÿ

Table 13: Absolute degree of grey incidences (Level 1 (>0.90))
Sl no.
RPI 1 875.156 875.156 873.418 873.418 1.738 0.999
RPI 2 1326.588 1326.588 1331.816 1331.816 5.228 0.998
RPI 3 187.881 -187.881 188.567 -188.567 0.686 0.998
RPI 4 1457.315 -1457.32 1449.031 -1449.03 8.284 0.997
RPI 5 570.753 570.753 568.583 568.583 2.17 0.998

38
  
“ ¡–• œ‡ ¡–– ∁
Table 14: Ratio of mean square deviations (Level 1 (<0.35))
Sl no.
RPI 1 3860.818 21829.36 0.16 72.044 0.057
RPI 2 3362.847 67420.45 0.662 206.114 0.055
RPI 3 322.399 1036.282 0.123 0.053 0.007
RPI 4 3627.479 52471.96 0.315 261.873 0.071
RPI 5 3092.107 8568.553 0.281 6.991 0.029

œ?›@ ?•@ − œ‡ œ?›@ ?–@ − œ‡ œ?›@ ?—@ − œ‡ œ?›@ ?˜@ − œ‡ œ?›@ ?™@ − œ‡ Check value P value
Table 15: Small error probability (Level 1 (>0.95))
Sl no.
RPI 1 0.16 9.24 15.849 4.415 2.033 99.656 1
RPI 2 0.662 14.998 9.771 20.981 16.417 175.137 1
RPI 3 0.123 0.103 0.094 0.398 0.267 21.713 1
RPI 4 0.315 16.85 16.709 19.228 19.402 154.506 1
RPI 5 0.281 3.383 4.346 1.128 1.809 62.436 1

š?•@ ?£@
“ š?•@ ?¤@
“ š?•@ ?¥@
“ š?•@ ?¦@
“ š?•@ ?•›@
“
Table 16: Predicted time response sequence for next five periods
Sl no.
RPI 1 23474.175 27752.79 32141.91 36644.39 41263.16
RPI 2 20777.623 24976.99 29422.68 34129.13 39111.65
RPI 3 1870.355 2111.486 2336 2545.042 2739.68
RPI 4 21328.48 24393.81 27336.88 30162.58 32875.58
RPI 5 18739.03 22082.51 25490.81 28965.18 32506.91

š?›@ ?£@
“ š?›@ ?¤@
“ š?›@ ?¥@
“ š?›@ ?¦@
“ š?›@ ?•›@
“
Table 17: Predicted sequence of indicators for next five periods
Sl no.
RPI 1 4170.887 4278.612 4389.119 4502.481 4618.771
RPI 2 3966.7 4199.369 4445.687 4706.453 4982.513
RPI 3 258.976 241.131 224.514 209.042 194.638
RPI 4 3192.661 3065.328 2943.074 2825.696 2712.999
RPI 5 3279.902 3343.484 3408.299 3474.371 3541.723

Table 18: Linguistic assessment and associated grey values

Linguistic assess- Associated grey values
ment
Very poor (VP) (0, 2)
Poor (P) (2, 4)
Medium (M) (4, 6)
Good (G) (6, 8)

39
 Very good (VG) (8, 10)

Table 19: 5 V’s evaluation framework for the case

Volume Upstream Internal Downstream
(Period) DS1 DS2 DS3 DS4 DS5 DS6
1 G G M G M M
2 G G G G M G
3 VG G G G G M
4 G VG VG G G G
5 VG VG G VG G VG
Velocity Upstream Internal Downstream
Period DS1 DS2 DS3 DS4 DS5 DS6
1 G G M G G G
2 M M G M G G
3 G VG G G G VG
4 G VG G VG VG G
5 VG VG G G VG VG
Variety Upstream Internal Downstream
Period DS1 DS2 DS3 DS4 DS5 DS6
1 G G M G M G
2 G M G M M M
3 G G G VG G M
4 VG G G G M G
5 G M G G M M
Veracity Upstream Internal Downstream
Period DS1 DS2 DS3 DS4 DS5 DS6
1 G M G M G M
2 G G G G M G
3 G M M M VG G
4 G VG G VG G VG
5 M G VG VG G VG
Value Upstream Internal Downstream
Period DS1 DS2 DS3 DS4 DS5 DS6
1 G M G G G M
2 VG G M G G M
3 M M G G G M
4 G G M G VG G
5 G G VG G G G

40
Table 20: Grey matrix representing 5 V’s evaluation framework for the case
Volume Upstream Internal Downstream
(Period) DS1 DS2 DS3 DS4 DS5 DS6
1 [6, 8] [6, 8] [4, 6] [6, 8] [4, 6] [4, 6]
2 [6, 8] [6, 8] [6, 8] [6, 8] [4, 6] [6, 8]
3 [8, 10] [6, 8] [6, 8] [6, 8] [6, 8] [4, 6]
4 [6, 8] [8, 10] [8, 10] [6, 8] [6, 8] [6, 8]
5 [8, 10] [8, 10] [6, 8] [8, 10] [6, 8] [8, 10]
Velocity Upstream Internal Downstream
Period DS1 DS2 DS3 DS4 DS5 DS6
1 [6, 8] [6, 8] [4, 6] [6, 8] [6, 8] [6, 8]
2 [4, 6] [4, 6] [6, 8] [4, 6] [6, 8] [6, 8]
3 [6, 8] [8, 10] [6, 8] [6, 8] [6, 8] [8, 10]
4 [6, 8] [8, 10] [6, 8] [8, 10] [8, 10] [6, 8]
5 [8, 10] [8, 10] [6, 8] [6, 8] [8, 10] [8, 10]
Variety Upstream Internal Downstream
Period DS1 DS2 DS3 DS4 DS5 DS6
1 [6, 8] [6, 8] [4, 6] [6, 8] [4, 6] [6, 8]
2 [6, 8] [4, 6] [6, 8] [4, 6] [4, 6] [4, 6]
3 [6, 8] [6, 8] [6, 8] [8, 10] [6, 8] [4, 6]
4 [8, 10] [6, 8] [6, 8] [6, 8] [4, 6] [6, 8]
5 [6, 8] [4, 6] [6, 8] [6, 8] [4, 6] [4, 6]
Veracity Upstream Internal Downstream
Period DS1 DS2 DS3 DS4 DS5 DS6
1 [6, 8] [4, 6] [6, 8] [4, 6] [6, 8] [4, 6]
2 [6, 8] [6, 8] [6, 8] [6, 8] [4, 6] [6, 8]
3 [6, 8] [4, 6] [4, 6] [4, 6] [8, 10] [6, 8]
4 [6, 8] [8, 10] [6, 8] [8, 10] [6, 8] [8, 10]
5 [4, 6] [6, 8] [8, 10] [8, 10] [6, 8] [8, 10]
Value Upstream Internal Downstream
Period DS1 DS2 DS3 DS4 DS5 DS6
1 [6, 8] [4, 6] [6, 8] [6, 8] [6, 8] [4, 6]
2 [8, 10] [6, 8] [4, 6] [6, 8] [6, 8] [4, 6]
3 [4, 6] [4, 6] [6, 8] [6, 8] [6, 8] [4, 6]
4 [6, 8] [6, 8] [4, 6] [6, 8] [8, 10] [6, 8]
5 [6, 8] [6, 8] [8, 10] [6, 8] [6, 8] [6, 8]

Table 21: Average grey matrix representing the evaluation framework for the case
Volume Upstream Internal Downstream
(Period) DS1 DS3 DS5
41
1 [6, 8] [5, 7] [4, 6]
2 [6, 8] [6, 8] [5, 7]
3 [7, 9] [6, 8] [5, 7]
4 [7, 9] [7, 9] [6, 8]
5 [8, 10] [7, 9] [7, 9]
Velocity Upstream Internal Downstream
(Period) DS1 DS3 DS5
1 [6, 8] [5, 7] [6, 8]
2 [4, 6] [5, 7] [6, 8]
3 [7, 9] [6, 8] [7, 9]
4 [7, 9] [7, 9] [7, 9]
5 [8, 10] [6, 8] [8, 10]
Variety Upstream Internal Downstream
(Period) DS1 DS3 DS5
1 [6, 8] [5, 7] [5, 7]
2 [5, 7] [5, 7] [4, 6]
3 [6, 8] [7, 9] [5, 7]
4 [7, 9] [6, 8] [5, 7]
5 [5, 7] [6, 8] [4, 6]
Veracity Upstream Internal Downstream
(Period) DS1 DS3 DS5
1 [5, 7] [5, 7] [5, 7]
2 [6, 8] [6, 8] [5, 7]
3 [5, 7] [4, 6] [7, 9]
4 [7, 9] [7, 9] [7, 9]
5 [5, 7] [8, 10] [7, 9]
Value Upstream Internal Downstream
(Period) DS1 DS3 DS5
1 [5, 7] [6, 8] [5, 7]
2 [7, 9] [5, 7] [5, 7]
3 [4, 6] [6, 8] [5, 7]
4 [6, 8] [5, 7] [7, 9]
5 [6, 8] [7, 9] [6, 8]

Table 22: Crisp matrix representing the evaluation framework for the case
5 V’s Period 1 Period 2 Period 3 Period 4 Period 5
Volume 6 6.7 7 7.7 8.3
Velocity 6.6 6 7.6 8 8.2
Variety 6.3 5.7 7.1 7 6.1
Veracity 6 6.7 6.2 8 7.8

42
 Value 6.4 6.6 6.1 6.9 7.4

Table 23: Validation of the prediction results from analysts for the case
Indicators Analyst 1 Analyst 2 Analyst 3 Analyst 4 Analyst 5
Flexibility VG G VG G VG
Responsiveness VG VG VG G G
Quality VG VG G G G
Productivity G G VG G VG
Accessibility VG VG G VG VG

Table 24: Validation results representing grey and crisp matrices for the case
Indicators Analyst 1 Analyst 2 Analyst 3 Analyst 4 Analyst 5 Grey aver- Crisp
age Value
Flexibility [8, 10] [6, 8] [8, 10] [6, 8] [8, 10] [7.2, 9.2] 8.2
Responsiveness [8, 10] [8, 10] [8, 10] [6, 8] [6, 8] [7.2, 9.2] 8.2
Quality [8, 10] [8, 10] [6, 8] [6, 8] [6, 8] [6.8, 8.8] 7.8
Productivity [6, 8] [6, 8] [8, 10] [6, 8] [8, 10] [6.8, 8.8] 7.8
Accessibility [8, 10] [8, 10] [6, 8] [8, 10] [8, 10] [7.6, 9.6] 8.6

43
Highlights

• An advanced grey prediction model for forecasting resilience performances is proposed.

• Supply chain resilience performance indicators are constructed.
• Periodic performance indicators for flexibility, responsiveness, quality, productivity and
accessibility are evaluated.
• Values are simulated and forecasted for future periods with the estimation of error
measures.
• Managers are recommended to analyze the predicted values to mend or amend strategic
decisions.

44