Assessing measures implemented by export-oriented RMG firms in an emerging economy during COVID-19

Manufacturing firms that continued production activities during the COVID-19 have been taking necessary measures to cope with the risks imposed by the pandemic. This study assesses the measures implemented by the Ready-Made Garments (RMG) sector in Bangladesh. With the increase in COVID-19 cases in Bangladesh, following government order, along with firms in other manufacturing sectors, the RMG firms had to shut-down their production between March 26 and April 25, 2020. Soon after the factories reopened, they had to take necessary actions to ensure employee safety, supply of raw materials, and purchase orders from buyers. Using a semi-structured interview approach, we identify 16 measures that have been implemented in the RMG sector in Bangladesh for the employees, suppliers and buyers. Then, we assess the degree of implementation of these measures using the Bayesian Best-Worst method. We find that providing healthcare safety, bringing previously outsourced activities in-house, and ensuring smooth delivery of existing orders were the three most implemented measures for employees, suppliers and buyers, respectively. On a higher level, the RMG industry professionals prioritised buyer-related measures the most, followed by employee and supplier-related. The analysed measures provide a blueprint for supply chain risk management during future waves of COVID-19 transmission and for other potential large-scale natural disasters.


Introduction
Nowadays global supply chains are spanning over multiple countries across different continents. Managing such supply chains is complex, and disruption in one part of the supply chain often has a significant impact on the overall supply chain. Meanwhile, disruptions are not uncommon in supply chains, and studies on supply chain uncertainty, risk management, resilience, and in general supply chain disruptions (SCDs) have been increasing significantly over the last two decades (Xu, Zhang, Feng, & Yang, 2020). However, majority of the studies deal with typical disruptions such as supplier failure (Ruiz-Torres & Mahmoodi, 2007), transportation disruption (Albertzeth, Pujawan, Hilletofth, & Tjahjono, 2020), demand uncertainty (Gupta & Maranas, 2003), and technological disruption (Ivanov, Dolgui, & Sokolov, 2019). Olivares-Aguila and ElMaraghy (2020) modelled the impact of SCDs on inventory levels, costs, service levels and profit of the supply chain in a multi-tier supply chain setting. Regardless of these efforts, studies focusing on irregular global disruptions such as the 2008 financial crisis or 2020 coronavirus crisis are rare.
Although the nature of the COVID-19 disruption is different from previous global disruptions such as the 9/11 in the United States or the global financial crisis in 2008, learning from those disruptions could be useful for firms in surviving the SCD caused by the COVID-19 pandemic. After 9/11, Kleindorfer and Saad (2005) analysed the global supply chain risks in the context of the chemical industry arising from disruptions caused by activities outside the scope of typical SCDs due to coordination issues in supply chains. One of the main mitigation strategy in such disruptions is to fix issues within one's own firm first before expecting or forcing supply chain partners to do so (Kleindorfer & Saad, 2005). As a pre-disruption measure, firms can utilise emerging technologies and increased access to information to optimise safety stocks to reduce stock-outs during SCDs (Di Nardo, Clericuzio, Murino, & Sepe, 2020).
Usually, a diversified supply chain reduces risk in such disruptions, particularly when a region or country is mainly exposed to the disruption, such as the United States during 9/11. Even having a diversified portfolio of suppliers across the world may not be a useful risk mitigation strategy when the whole world is affected by the disruption, which is evident from the 2008 financial crisis, although to a lesser magnitude compared to the current COVID-19 pandemic. Blome and Schoenherr (2011) studied risk management strategies of European firms in the context of the global financial crisis in 2008. Blome and Schoenherr (2011) argued that companies improved the comprehensiveness of their supply chain risk management approaches and also started to accept risks more in comparison to risk avoidance. Despite such development of supply chain risk management strategies and the growing number of literature on the topic, the COVID-19 pandemic has shaken the global supply chains.
Since the starting of the SCD due to the COVID-19 in March 2020, studies have examined its impact on global supply chain (Ivanov, 2020) and risk management strategies (Chowdhury, Sarkar, Paul, & Moktadir, 2020;Sharma, Shishodia, Kamble, Gunasekaran, & Belhadi, 2020).In this context, de Sousa Jabbour et al. (2020) stated that different industries face different challenges, and accordingly, strategies taken by industry professionals' varies. To the best of the authors' knowledge, an in-depth investigation of the measures implemented by manufacturing firms, particularly, in the context of an export-oriented industry of a developing country, is not evident. Hence, we investigate measures implemented in the Ready-Made Garments (RMG) industry of Bangladesh using a novel Multi-Criteria Decision-Making (MCDM) framework. We develop the MCDM framework based on four interviews with industry experts. Based on the MCDM framework, we collected data from 11 respondents using a complex web-survey and used the Bayesian Best-Worst Method (BWM) to analyse the survey data.
The next section presents the context of the study, design process of the survey, data collection and methodological detail of the Bayesian BWM. Section 3 presents the results, and Section 4 discusses the theoretical and practical implications of the findings. Finally, Section 5 summarizes the study with limitations and future research direction.

Context of the study
This study investigates the measures implemented by the RMG firms in Bangladesh. The RMG sector of Bangladesh not only plays a significant role in the country's economy but also a major supplier for the global fashion industry. The RMG sector accounts for 84% of Bangladesh's total export, about USD 34.13 billion for the 2018-19 fiscal year. Besides, followed by China, Bangladesh is the second-largest global RMG supplier 1 .
Regardless of its significant contribution to the Bangladesh economy, the RMG sector of Bangladesh has been vulnerable to disruptions. Political unrest, labour union strikes and accidents have been part of the industry since the early 1980s. However, since the collapse of Rana Plaza Complex (that hosted several garment factories) in 2013, global brands and retailers joined effort for workers' health and safety. The Accord on Fire and Building Safety in Bangladesh and the Alliance for Bangladesh Worker Safety was established in 2013 initiated by mainly European and North American retailers, respectively. Since then, the number of incidents has been declined significantly as well as the safety in RMG factories improved. Initially, the Accord-Alliance was established for a five-year term to run inspection, remediation and workplace empowerment programs. In 2018, Accord-Alliance terms came to an end and members of both continued their word through locally-formed organizations. For instance, as of June 1, 2020, the RMG Sustainability Council (RSC) took over the functions of the Accord office in Bangladesh 2 .
While the industry has been functioning well in recent years, the COVID-19 pandemic has made it vulnerable again. The Regulation of International Supply Chains (RISC) has already published a report on the early and expected impacts of the COVID-19 pandemic on the Bangladesh RMG sector. In the report, Leitheiser et al. (2020) discuss some of the major challenges the industry is facing such as cancellation of orders by the large multinational brands leadings to financial distress among suppliers further impeding the life of the workers. The RMG industry in Bangladesh employs more than four million workers, who are in financial, health and safety hazards due to the COVID-19. Hence, this study investigates the measures taken by RMG firms in Bangladesh to sustain the pandemic by keeping their business running even though in a lower capacity than regular times.

Survey design
This study includes three stages. In the first stage, we interview RMG experts to identify and finalize a list of measures implement during the COVID-19 pandemic. Such an approach has been implemented in previous MCDM studies as well, for example, Pintelon, Di Nardo, Murino, Pileggi, and Vander Poorten (2021) and Ali, Hossen, Mahtab, Kabir, and Paul (2020). The second stage, includes the BWM survey design and data collection from industry experts. In the final stage, we analyse collected survey data using the Bayesian BWM. The three-stage workflow of the study is depicted in Fig. 1.
To identify measures implemented to keep the factories open during COVID-19, we choose four industry experts for interviews based on purposive sampling approach. The interviews were conducted sequentially. Our goal was to identify implemented measures for the suppliers, employees and buyers. After the first interview, we prepared a list of measures under these three criteria. Then, we proceed with the second interview with the aim of validating the already listed measures and identifying additional measures. The same approach was followed for the four interviews. Table 1 presents an overview of the interview participants. The interviews continued between 45 min to 75 min.
After the fourth interview, the authors reached data saturation, that is, no new measures were introduced. Hence, we discussed the full list of 18 measures with the four participants, which was an iterative process and two measures were removed due to low degree of implementation, i. e. introducing in-house accommodation for workers and mobile banking for salary payment. Following the validation of the identified measures by the four interviewees, we proceed with the BWM survey design. In consultation with the interviewees, we group identified measures as subcriteria under three criteriasupplier-related, employee-related and   buyer-related measures. Table 2 presents the criteria and their respective sub-criteria.

Data collection
Following the development of relevant criteria and sub-criteria, we designed a BWM survey. BWM surveys, and MCDM survey, in general, are different from conventional Likert scale surveys and complex in nature. The survey was distributed digitally to contacts of the authors working in the RMG industry in Bangladesh as well as across social media by means of professional RMG industry related groups in Facebook and LinkedIn. As the country was going through governmentimposed lockdowns and physical meetings were discouraged, webbased digital survey was the appropriate medium to collect data. The survey was active for one month from June 1 to June 30, 2020. In total, 26 industry experts responded in the survey. After removing responses from industries other than RMG (03), duplicates (02), logically wrong inputs (05) and straight lining (05), 11 responses were left for analysis. Table 3 presents the background of the respondents who participated in the survey. The majority of the respondents' firms have ISO 9000 certification (08 out of 11), and are members of the Accord Alliance (08 out of 11). This indicates that firms represented by our sample already have quality management systems in place as well as worker health and safety standards. Data of the Best-to-Others and Others-to-Worst vectors are available as online supplementary material.

Bayesian best-worst method
In order to identify the importance of identified criteria in Table 2, this study takes advantage of a multi-criteria decision-making (MCDM), Bayesian Best-Worst Method (BWM), which is a group-decision making method. In Bayesian BWM, the preferences of multiple experts/decisionmakers (DMs) are aggregated within a probabilistic framework, which allows us to draw conclusions on the final aggregated priorities of criteria.
Let C = {c 1 , …, c n } be a set of criteria that are evaluated by K experts or DMs. The following steps are required for applying the Bayesian BWM: Step 1: First, expert k selects the best (c k B ) and the worst (c k W ) criteria out of C. In this step, each of the experts are asked to choose only the best and the worst criteria from the criteria set C. At this stage, the expert does not make any pairwise comparison between the criteria and only selects the best and the worst. The best is the most implemented, while the worst is the least implemented criterion for expert k, and different experts can opt for different criteria as their best and/or worst.
Step 2: Expert k makes the pairwise comparison between the best (c k B ) and the other criteria in C.
In this step, each of the experts needs to express their preferences of the best criterion over other criteria in C by a number between one and nine. An assignment of one means that the criteria being compared are equally implemented, while an assignment of nine means that c B is extremely more implemented. The set of pairwise comparison for expert k in this stage results in a vector, called "Best-to-Others", shown as A k B , and is defined as: where a k Bj denotes the preference of the best criterion (c k B ) over c j ∈ C for expert k.
Step 3: Expert k makes the pairwise comparison between the worst criterion (c k W ) and the other criteria in C.
In this step, each of the experts needs to identify their preferences of the other criteria over the worst criterion, which is selected in Step 1, by a number between one and nine. An assignment of nine in this step shows that the corresponding criterion is extremely more implemented than the worst, while an assignment of one means that the two criteria are equally implemented. This step results in the "Others-to-Worst" vector, shown as A k W for expert k, and is defined as: where a k jW is the preference of criterion c j ∈ C over the worst criterion for decision-maker k (c k W ).
Step 4: Computing the aggregated weights of all K experts w * = (w * 1 , w * 2 , …, w * n ) and the weight for each of the experts w k , k = 1, …, K by the following probabilistic model of the Bayesian BWM: where multinomial is the multinomial distribution, Dir is the Dirichlet distribution and gamma(0.1, 0.1) is the gamma distribution with the shape parameters of 0.1. The probabilistic model identified by equations in (7) does not bear a closed-form solution. As a result, a Markov-chain Monte Carlo (MCMC) (Gilks, Richardson, & Spiegelhalter, 1995) sampling is required to estimate the solution. The Bayesian BWM is implemented in JAGS (Just Another Gibbs Sampler) (Plummer, 2004), and is freely and publicly available 3 .
After solving the problem by using the Bayesian BWM, it provides S sample from the posterior distributions of the aggregated weight w * . According to the aggregated weight, we can draw the importance of each criteria, as well as the extent to which one criterion is preferred over another based on the preferences of all experts/DMs. To compute such an extent, we need to make an inference in a Bayesian sense. The following definitions and computations would provide such a probabilistic sensing based on the aggregated priorities w * . where • R is the relation between the performance metrics c i and c j , i.e., <,>, or =; • d ∈ [0, 1] is the confidence degree of the relation. We now only need to compute the degree in credal ordering for each pair of criteria c i and c j . For doing so, we take advantage of S samples obtained from JAGS, and compute the degree as: Given this computation, we can make a probabilistic comparison between criteria. In the following sections, we visualize the credal ranking of a set of criteria by using a weighted directed graph, where all the required information about the probabilistic comparison can be readily realized. For robustness, two of the authors analysed the data separately and both yield the same result.

Results
This section presents the ranking of the implemented measures estimated using the Bayesian BWM. For each of the problems, we create a weighted directed graph. The nodes of such a graph represent the criteria as well as their average weight computed as the mean of the w * distribution, and each edge A → d B means that A is more important than B with a confidence degree of d. In other words, each edge in the graph represents a credal ordering, and the whole graph visualizes the credal ranking of a set of criteria.
That being said, we first present the credal ranking of the main criteria as shown in Fig. 2. Among the three criteria levels, buyer-related measures were most implemented (0.38705), followed by employeerelated (0.32628) and supplier-related (0.28667). Further, buyerrelated measures were more implemented with 0.91 confidence than supplier-related, while with 0.78 confidence than employee-related, and employee-related measures were more implemented than supplierrelated with 0.72 confidence. Although it depends on the decision makers perception, for simplicity of the interpretation of the confidence scores, we can use 0.70 threshold to indicate strong confidence. Probability based confidence sores are visible on the edges in the figure.
As buyer-related measures were the most implemented, we first present the local ranking of these measures in Fig. 3. Ensuring smooth delivery of existing orders (0.24543) is the most prioritized measure  followed by offering discounts to existing orders to minimize order cancellation (0.2051) and offering favourable terms for new orders (0.18518). Considering the threshold of 0.70, ensuring smooth delivery of existing orders were significantly more implemented in comparison to all other buyer-related measures. We can also see the inter-relationship structure among the implemented measures as represented by different colours in Fig. 3.
Local ranking of the employee-related measures are presented in Fig. 4. Out of the six measures under this criteria, the three most implemented measure was providing healthcare safety of the workers and employees (0.28338), reducing working/factory hours (0.16337), and reducing employee travel (0.13943). Further, providing healthcare safety was extremely more implemented than all other measures, as evident from confidence scores ranging between 0.99 to 1.00.
Finally, Fig. 5 presents the local ranking of the supplier-related implemented measures. Bringing activities in-house which was earlier outsourced was implemented the most (0.22701). Moreover, RMG manufacturers increased sourcing from local suppliers (0.22459) and started actively searching for alternative suppliers both in local and foreign markets (.20242). Again, considering 0.70 threshold, bringing activities in-house was significantly more implemented than other supply-related measures except for increased sourcing from local suppliers (0.52).
We report the confidence score matrix for each of the figures in Table 4-7 in the Appendix. Furthermore, by multiplying the criteria level weights with their respective sub-criteria level local weights, we get the global ranking of the 16 implemented measures. Fig. 6 depicts the global rankings. We find that ensuring smooth delivery of existing orders is the most implemented (0.0950), and maintaining employee salaries and benefits as before COVID-19 is the least implemented (0.0440).

Discussion
We discuss the findings reported in Section 3 considering studies published on SCDs and the four interviews conducted for the identification of the implemented measures.

Setting priorities
Setting clear priorities is one of the most important SCD mitigation strategies (Kleindorfer & Saad, 2005). This study contributes to the SCD literature by providing a validated MCDM framework for determining priorities during SCDs. We found that RMG firms in Bangladesh are focusing more on buyer-related measures than employee and supplierrelated ones. As reported in Fig. 6, out of the first six most implemented measures five are buyer-related. Although on the criteria level, employee-related criteria ranked second, on the global ranking of implemented measures, employee related ones rank lowest. The implications of this finding are twofold. First, it shows that to keep the business running firms should prioritise their buyers. On the other hand, RMG firms in Bangladesh are paying too little attention to their employees and workers. For an industry that has long been criticised on this dimension, firms should consider improving this situation even during the pandemic.

Negotiating contracts
Many orders were cancelled due to COVID-19. By April 2020, orders worth USD 1.8 billion were cancelled, and a further USD 1.4 billion worth of orders to the RMG industry was put on hold (Leitheiser et al., 2020). In case of cancellation of orders, the best approach is to negotiate a new deal. One of the implemented negotiated deals include securing buyers guarantee to cover cost of goods already produced for their orders, mainly by mean of offering discounts from the suppliers' side. Other negotiation deals includes order postponement or slow delivery. It is essential to bear in mind that business can flourish again with the same buyer after the COVID-19. As evident in our findings, even new business opportunities might arise during the COVID-19. Hence, it might be naïve to force legal power too early based on existing contracts. In future sourcing contracts between buyer and supplier should incorporate disruption risk-sharing clauses (Majumdar, Shaw, & Sinha, 2020).
For remaining orders, it is necessary to ensure smooth delivery to the buyers. Also, all four interviews stated that they were expecting winter orders from their foreign buyers and were starting to prepare.

New product development
Some RMG firms have started to manufacture face masks and personal protective equipment (PPE). By the time of interview, Interviewee 2's firm already started production of PPE and cotton-based musk for buyers located in the USA and Canada. According to the interviewee, those buyers put on hold existing orders and gave new purchase orders for PPE and face masks. One of the largest RMG manufactures of Bangladesh, Beximco Textile, has already invested USD 20 million on a PPE and face mask production plant in Detroit, USA 4 . Besides, the disruption in the production process can be useful for deletion of an existing product, and introduce more sustainable version (Zhu, Shah, & Sarkis, 2018). Further, Rapaccini, Saccani, Kowalkowski, Paiola, and Adrodegari (2020) suggest that manufacturing firms should provide services through digitalisation to enhance resilience toward future disruptions. Such initiatives would require innovation in existing business models leading to new product development (de Sousa Jabbour et al., 2020).

Supplier diversification, but how
The majority of the RMG firms in Bangladesh have been sourcing their materials from China, as reported by all the four interviewees. Companies that already source from multiple suppliers in different countries also realize that they cannot depend on outside suppliers for their core supplies, and thus, are planning to or have already started inhouse production of core supplies (de Sousa Jabbour et al., 2020). As stated by Interviewee 1, even though their firm re-opened, supply of incoming materials, majority sourced from China, were cancelled. Hence, they were producing products of which materials they already had in stock. Meanwhile, they were looking for new suppliers (also, stated by all interviewees). Some firms have initiated increasing level of 4 See https://www.bloombergquint.com/onweb/detroit-to-get-bangladeshiinvestment-for-plant-to-make-masks, accessed on October 14, 2020. safety stock, which is a coping strategy with SCDs as Bueno-Solano and Cedillo-Campos (2014) reported earlier that disruptions due to terrorist attacks can lead to a 600% increase in inventory levels. Riddle et al. (2020) also confirm that SCDs lead to capacity expansion decisions by firms that are not sustainable in the long-run. Hence, we suggest manufacturing firms to stress their supply chain resilience strategies, which are often conflicting with lean supply chain strategies.

Improvement of workers' health and safety
As reported in Fig. 6, out of the five least implemented measures, four are employee-related measures. The same has been reported by Leitheiser et al. (2020), too. This situation needs attention from the RMG firms, their buyer, policymakers, and government agencies. For the COVID-19 pandemic, manufacturing firms in several emerging markets have received financial support from their governments to support their workers (Karmaker et al., 2020). In collaboration with Bangladesh Garment Manufacturers and Exporters Association (BGMEA), the government of Bangladesh has already initiated financial subsidy plans for the RMG industry. The subsidy plan provides financial support at 2% interest rate to the RMG sector to be paid back in 18 monthly instalments after six months grace period (Interviewee 4 stated during interview). The RMG Sustainability Council (RSC) that took over the functions of Accord has recommended the RMG factories to keep the factory workers as safe as possible 5 . The RSC may negotiate support programs with Western retailers for mitigating the impact of COVID-19 on RMG workers' health and safety. Meanwhile, Majumdar et al. (2020) argue that multinational brands should prefer suppliers that provide better working environment for their workers, including permanent employment contracts. In the same vein, we argue that some of the measures taken for workers' health and safety, e.g. ensuring cleanliness of the factory, regular health monitoring, and arraigning food supplies for the workers, should stay after the COVID-19.

Conclusion
COVID-19 has brought many challenges for manufacturing firms. An in-depth study on disruption mitigation strategies adopted by the firms is still unexplored in the literature. Therefore, this study investigates the measures implemented by RMG firms in Bangladesh while continuing production facilities during the pandemic. We investigate implemented measures from the perspectives of suppliers, employees and buyers, as these three stakeholders are the most important for any manufacturing firm. First, we conducted four interviews to form a list of implemented measures. Then, we designed a BWM survey, and distributed it among the RMG industry professionals in Bangladesh.
This study has demonstrated the application of a novel MCDM method. Methodology-wise, we used the Bayesian BWM. For group decision making in the original BWM, we usually analyse preferences of multiple decision makers separately, and then aggregate their priorities using arithmetic mean. As such an approach can be sensitive to outliers, the Bayesian BWM reduces this sensitivity by analysing probabilistic group preferences. We find that buyer-related measures such as ensuring smooth delivery of existing orders, offering discounts to buyers to reduce order cancellation, and revised terms for new orders are most implemented measures in the RMG sector in Bangladesh. As of upstream supply chain, companies are looking into bringing outsourced functions in-house and developing local alternative suppliers. We also found that employee-related measures need further attention. To tackle such disruptions in the future, the RMG firms need to improve their supply chain resilience. Long-term relationship building with buyers, maintaining a diversified portfolio of suppliers including local ones, increased safety stock, organizational and absorptive capability development are essential.
Meanwhile, on a brighter note, the COVID-19 pandemic has forced digitalisation in manufacturing firms. Although located in a developing country, RMG firms have started to look into the application of emerging digital technologies such as cloud manufacturing, blockchain, crowdsourcing and others. Future research should investigate adaption of these technologies in developing countries in the long-run. Also, this pandemic has influenced new product development and internationalisation of some Bangladeshi RMG firms. These topics need further exploration, too.