Designing a country's small and medium-sized enterprise development index using firm-level data: The case of Thailand

Shinozaki, Shigehiro; Miyakawa, Daisuke Working Paper Designing a country's small and medium-sized enterprise development index using firm-level data: The case of Thailand ADB Economics Working Paper Series, No. 785 Provided in Cooperation with: Asian Development Bank (ADB), Manila Suggested Citation: Shinozaki, Shigehiro; Miyakawa, Daisuke (2025) : Designing a country's small and medium-sized enterprise development index using firm-level data: The case of Thailand, ADB Economics Working Paper Series, No. 785, Asian Development Bank (ADB), Manila, https://doi.org/10.22617/WPS250228-2 This Version is available at: https://hdl.handle.net/10419/322378 Standard-Nutzungsbedingungen: Die Dokumente auf EconStor dürfen zu eigenen wissenschaftlichen Zwecken und zum Privatgebrauch gespeichert und kopiert werden. Sie dürfen die Dokumente nicht für öffentliche oder kommerzielle Zwecke vervielfältigen, öffentlich ausstellen, öffentlich zugänglich machen, vertreiben oder anderweitig nutzen. Sofern die Verfasser die Dokumente unter Open-Content-Lizenzen (insbesondere CC-Lizenzen) zur Verfügung gestellt haben sollten, gelten abweichend von diesen Nutzungsbedingungen die in der dort genannten Lizenz gewährten Nutzungsrechte. Terms of use: Documents in EconStor may be saved and copied for your personal and scholarly purposes. You are not to copy documents for public or commercial purposes, to exhibit the documents publicly, to make them publicly available on the internet, or to distribute or otherwise use the documents in public. If the documents have been made available under an Open Content Licence (especially Creative Commons Licences), you may exercise further usage rights as specified in the indicated licence. https://creativecommons.org/licenses/by/3.0/igo/ ASIAN DEVELOPMENT BANK ASIAN DEVELOPMENT BANK 6 ADB Avenue, Mandaluyong City 1550 Metro Manila, Philippines www.adb.org ADB ECONOMICS WORKING PAPER SERIES NO. 785 June 2025 Designing a Country’s Small and Medium-Sized Enterprise Development Index Using Firm-Level Data The Case of Thailand This paper employs probabilistic principal component analysis to develop a new way to quantitatively assess what affects micro, small, and medium-sized enterprise (MSME) development nationally, by using granular firm-level panel data for 49,565 MSMEs in Thailand. The estimation results found a potential disproportionate effect of MSME policy interventions during and after the coronavirus disease pandemic. This underscores the importance of using a focused approach when designing policies for MSME development to facilitate more sustainable, resilient private sector growth. About the Asian Development Bank ADB is a leading multilateral development bank supporting inclusive, resilient, and sustainable growth across Asia and the Pacific. Working with its members and partners to solve complex challenges together, ADB harnesses innovative financial tools and strategic partnerships to transform lives, build quality infrastructure, and safeguard our planet. Founded in 1966, ADB is owned by 69 members—50 from the region. DESIGNING A COUNTRY’S SMALL AND MEDIUM-SIZED ENTERPRISE DEVELOPMENT INDEX USING FIRM-LEVEL DATA THE CASE OF THAILAND Shigehiro Shinozaki and Daisuke Miyakawa ASIAN DEVELOPMENT BANK The ADB Economics Working Paper Series presents research in progress to elicit comments and encourage debate on development issues in Asia and the Pacific. The views expressed are those of the authors and do not necessarily reflect the views and policies of ADB or its Board of Governors or the governments they represent. ADB Economics Working Paper Series Shigehiro Shinozaki and Daisuke Miyakawa No. 785 | June 2025 Shigehiro Shinozaki ([email protected]) is a senior economist at the Economic Research and Development Impact Department, Asian Development Bank. Daisuke Miyakawa ([email protected]) is a professor at Waseda University and chief economist of UTokyo Economic Consulting Inc. Designing a Country’s Small and Medium-Sized Enterprise Development Index Using Firm-Level Data: The Case of Thailand Creative Commons Attribution 3.0 IGO license (CC BY 3.0 IGO) © 2025 Asian Development Bank 6 ADB Avenue, Mandaluyong City, 1550 Metro Manila, Philippines Tel +63 2 8632 4444; Fax +63 2 8636 2444 www.adb.org Some rights reserved. Published in 2025. ISSN 2313-6537 (print), 2313-6545 (PDF) Publication Stock No. WPS250228-2 DOI: http://dx.doi.org/10.22617/WPS250228-2 The views expressed in this publication are those of the authors and do not necessarily reflect the views and policies ofthe Asian Development Bank (ADB) or its Board of Governors or the governments they represent. ADB does not guarantee the accuracy of the data included in this publication and accepts no responsibility for any consequence of their use. The mention of specific companies or products of manufacturers does not imply that they are endorsed or recommended by ADB in preference to others of a similar nature that are not mentioned. By making any designation of or reference to a particular territory or geographic area inthis document, ADB does not intend to make any judgments as to the legal or other status of any territory or area. This publication is available under the Creative Commons Attribution 3.0 IGO license (CC BY 3.0 IGO) https://creativecommons.org/licenses/by/3.0/igo/. By using the content of this publication, you agree to be bound bytheterms of this license. For attribution, translations, adaptations, and permissions, please read the provisions andterms of use at https://www.adb.org/terms-use#openaccess. This CC license does not apply to non-ADB copyright materials in this publication. If the material is attributed toanother source, please contact the copyright owner or publisher of that source for permission to reproduce it. ADB cannot be held liable for any claims that arise as a result of your use of the material. Please contact [email protected] if you have questions or comments with respect to content, or if you wish toobtain copyright permission for your intended use that does not fall within these terms, or for permission to use theADB logo. Corrigenda to ADB publications may be found at http://www.adb.org/publications/corrigenda. Note: In this publication, “B” refers to baht. ABSTRACT Understanding the business environment and structural issues that limit growth is critical when designing an effective national policy framework for private sector development—especially for micro, small, and medium-sized enterprises (MSMEs). Given the limited MSME data available, this paper employs probabilistic principal component analysis to develop a new way to quantitatively assess what affects MSME development nationally, by using granular firm-level panel data for 49,565 MSMEs in Thailand as a case study. The estimation results found a potential disproportionate effect of MSME policy interventions during and after the coronavirus disease pandemic. Government assistance for MSMEs likely helped Bangkok-based firms ease the negative pandemic effects, especially in manufacturing. However, it did not help local MSMEs— regardless of sector—as their operational performance deteriorated both during and after the pandemic. This underscores the importance of using a focused approach when designing policies for MSME development to facilitate more sustainable, resilient private sector growth. Keywords: SME development, access to finance, financial inclusion, SME policy, probabilistic principal component analysis, Thailand JEL codes: D22, G20, L20, L50 1. Introduction Macro and micro policy implementation should be consistent with assistance measure targets. While highly aggregated variables such as gross domestic product (GDP) and the inflation rate have been used as key measures in guiding policy design, policy makers are now paying more attention to granular data such as firm-level financial statements. This reflects the need to consider the heterogenous status and characteristics of policy targets—such as micro, small, and medium-sized enterprises (MSMEs)—so policy implementation is more effective and efficient (Gourinchas et al. 2020; Ebeke et al. 2021). In this paper, we empirically measure summarized business activity using granular data to construct firm activity indexes that account for the heterogeneity among policy targets. MSMEs have a large impact on business activity, job creation, and economic output, helping drive growth across developing Asia and the Pacific. Governments in the region thus use various assistance measures for national MSME development—to promote young and women entrepreneurs, adopt and commercialize technology, expand MSME international market access, develop human capital and worker skills, and boost access to finance. However, constraints on MSME development remain in most countries, raising the question of how governments can design effective policies that better reach MSMEs. Understanding the business environment and structural problems associated with business growth is critical to design an effective national policy framework for MSME development. However, limited MSME data make it difficult to do so. Given the general usefulness of constructing firm activity indexes, various global organizations such as the Organisation for Economic Co-operation and Development, the Economic Research Institute for ASEAN and East Asia, and the International Trade Centre have already begun the process. They propose a qualitative approach using assessment matrices for performance ratings or median comparisons based on available data to evaluate MSME development conditions (ADB 2022). Despite these efforts, the discretion associated with ways to construct measures, the cost of producing reliable measures, and the difficulty of interpreting results are also considered major issues public organizations must address. Preceding studies such as ADB (2022) and Shinozaki et al. (2024) aim to overcome this difficulty by applying an empirical method—probabilistic principal component analysis (P-PCA)—to country-level annual data, which is immune to discretion and easy to implement, thus allowing for straightforward interpretation. One caveat to preceding studies is that constructing activity indexes for MSMEs—such as the Small and Medium-Sized Enterprise Development Index (SME-DI)—do not necessarily consider the heterogeneity among policy targets. This drawback reflects the fact that ADB (2022) and Shinozaki et al. (2024) used aggregate data such as country-level macroeconomic variables to construct an SME-DI that covers the entire (or sub) Asian region(s). In this paper, we follow ADB (2022) and Shinozaki et al. (2024) and further apply the P-PCA method to disaggregated data so we can consider the heterogenous status among policy targets, for example, across various industries and regions in constructing an SME-DI. Here, we use proprietary firm-level panel data obtained from a private data vender in Thailand (Dun & Bradstreet) to apply the P-PCA method to the granular data. Specifically, we conduct several exercises to further develop the SME-DI. First, we apply the P-PCA method to industryregion-year level disaggregated panel data (constructed from firm-year level granular data specifically on Thailand’s MSMEs) to create an index accounting for MSME activities in a single country. This index is further broken down by industryand/or regional-level indicators as well as indicators explicitly accounting for MSMEs’ real and financial activities. Second, using the smaller dataset randomly chosen by the entire granular dataset, we construct industry-region-year level 2 disaggregated panel data and re-run the P-PCA algorithm. Comparing this result with that based on the entire MSME dataset, we examine whether or not the relatively small amount of data can reproduce an SME-DI consistent with that obtained from the entire dataset. This provides useful information practitioners can easily implement—for example, survey analyses to construct a reliable indicator summarizing MSME activities. Third, we further apply the P-PCA method to firmlevel “raw” granular data (MSME data not aggregated by industry-region but used as is) and directly obtain the indicators that account for MSME activities in Thailand. Again, this exercise contributes to our deeper understanding of the value of survey analyses in constructing a reliable indicator summarizing MSME activities. 2. Thailand’s MSME Landscape and Policy Support Measures Like in other countries, MSMEs play a crucial role in driving Thailand’s economy.1 Over time, their numbers have consistently increased, even during the coronavirus disease (COVID-19) pandemic that started in early 2020. MSMEs increased in number by 0.9% in 2020 and 1.4% in 2021. As of end-2023, there were 3.2 million MSMEs, accounting for 99.5% of all enterprises (ADB 2024). By sector, the largest share was in wholesale and retail trade (41.8%), followed by other services (including accommodation and food services, 40.4%) and manufacturing (16.0%). MSMEs were spread across the country with 83.6% in the provinces and 16.4% in the capital city Bangkok. As of end-2023, MSMEs employed 12.9 million workers or 70.4% of the total workforce, which has gradually expanded during the post-pandemic recovery. While GDP growth in Thailand slowed in 2023 (2.0%), MSME output remained relatively robust. Despite the pandemic challenges, GDP rebounded quickly for MSMEs—from a 9.6% decline in 2020 to 4.2% growth in 2021, a V-shaped recovery. As of end-2023, MSME output was B6.3 trillion, accounting for 35.2% of GDP, up 3.6% from 2022. Among MSMEs, other services contributed most (41.2%), followed by manufacturing (30.0%) and wholesale and retail trade (21.8%) (ADB 2024). MSME exports also rebounded quickly from the pandemic impact, growing by a remarkable 22.0% in 2021 after a 17.1% drop in 2020. By 2023, MSME export value reached B1.3 trillion or 13.4% of total export value. MSME exports grew by a robust 24.3% from 2022. Major export destinations included other Southeast Asian countries, the People’s Republic of China (PRC), the United States, and the European Union, which together accounted for nearly 80% of total MSME export value. Nearly all major MSME export markets expanded primarily due to their own economic recovery (Office of Small and Medium Enterprises Promotion [OSMEP] 2022). In 2021, export volumes of gems and jewelry, wood, agricultural produce, iron and steel, and plastic products grew fastest. By contrast, MSME exports such as sugar, automobile parts, and rubber products declined in volume. MSME imports also expanded, reaching B1.4 trillion in 2023, accounting for 14.0% of total import value, a 16.2% increase from 2022 (ADB 2024). The government’s quick policy actions supporting MSMEs and workers during the pandemic likely contributed to the lower pandemic impact on businesses in Thailand. For debt financing, the central bank (Bank of Thailand) provided several liquidity support measures for banks to finance MSMEs hurt by the pandemic. These included capital buffers for banks, deferred principal payments, reduced interest rates, and new soft loans/credit lines for MSMEs. The government 1 In Thailand, MSMEs are defined as either (i) manufacturing or (ii) services and trading, using the number of employees and revenue as criteria (ADB 2024). For manufacturing (including agriculture), an MSME is defined as a firm with up to 200 employees or annual revenue of less than B500 million. For services and trading, an MSME can have up to 100 employees or annual revenue below B300 million. 3 used value added tax (VAT) refunds for domestic entrepreneurs and reduced social security contributions and withholding taxes for businesses. It also provided B5,000 monthly handouts for self-employed and laid-off workers, and partial (50%) salary payouts for those displaced. To help lower business costs, water and electricity payments were suspended (ADB 2020). Given the importance of MSMEs in the Thailand economy, the government developed a mediumterm strategy to promote MSMEs. The current SME Promotion Plan 2023–2027 aims to create a strong and progressive environment to make MSMEs more competitive. It has three strategic pillars: (i) inclusive development across all MSME sectors, (ii) identify market opportunities, and (iii) create a supportive MSMEs ecosystem that improves access to finance, technology, innovation, skills development, big data, the legal framework, and policy information (ADB 2024). 3. Empirical Approaches This study uses the P-PCA empirical approach. As detailed in ADB (2022) and Shinozaki et al. (2024), the idea behind the P-PCA method is basically same as standard principal component analysis (PCA), which extracts common factors shared by multiple time series data. One unique feature of the P-PCA method is its robustness against missing records, which is likely the case in developing countries. This iteration of the PCA calculation and interpolation processes gives us the principal components (and factor loadings) based on the data suffering from missing records. The detailed P-PCA model is explained in Appendix 1.2 The ADB (2022) and Shinozaki et al. (2024) main analysis takes out a limited number of common factors shared by, for example, multiple time-series macroeconomic data accounting for the entire (or sub) Asian region(s). Unlike previous studies, we focus on a single country (Thailand) as a case study, and follow four analytical steps. First, we use granular MSME panel data (aggregating time-series industry-region level granular firm data) to take out the common factors as a baseline analysis. Second, we break down the analysis by industry and region, and use that nexus to take out the common factors. Third, the same procedures are used to analyze the trend of real or financial activities of firms. And fourth, we conduct small sample exercises to see to what extent the dynamics of the extracted principal components are influenced by sample size. 4. Data The firm-level panel data used here come from a private data vender in Thailand (Dun & Bradstreet). The data consist of annual frequency firm-level information over the periods from 2016 to 2023.3 In this firm-level data, we have firm-year unbalanced panel data for 11 variables, consisting of those associated with firms’ real and financial activities (Table 1). The original data cover 52,086 firms with no missing data on employment and total 370,676 samples for 2016– 2023. From these, we extract MSMEs that fit the criteria of firms in manufacturing or agriculture with up to 200 employees and those in services and other industries with up to 100 employees– 49,565 MSMEs with a total of 353,373 samples for 2016–2023, equivalent to 95% of total firm samples.4 This MSME dataset is analyzed in this study. 2 The Appendix is available at http://dx.doi.org/10.22617/WPS250228-2. 3 Throughout this paper, we call a firm’s financial statement data stored in the original database ending in year YYYY as the year = YYYY. For example, if a firm’s financial statement accounts for the accounting period from January 2023 to December 2023, we call this data as year = 2023. 4 A firm is classified as an MSME if it meets the criteria for at least 1 year during the data period. This classification is fit to the employment threshold of the national MSME definition in Thailand (see footnote 1). 4 The industry classification of the data refers to nine categories based on the standard industrial classification system: (i) agricultural, forestry, and fishing; (ii) mining; (iii) construction; (iv) manufacturing; (v) transportation; (vi) wholesale; (vii) retail; (viii) financial services; and (ix) other services.5 The regional classification of the data refers to Bangkok (capital city) and six official regions: Central (except Bangkok), Eastern, North, Northeast, Southern, and Western Thailand. To construct the input data for the P-PCA method, we aggregate each variable in each year using these industry and region classifications. For the small sample exercises, we randomly choose a set of data from the entire MSME dataset. This provides us an experimental environment where we use information on the limited number of surveyed firms to compute the SME-DI. Where the number of randomly chosen firms is limited, we apply the P-PCA directly to those selected firm-level panel data to obtain the SME-DI based on the pseudo survey data.6 In other words, we estimate the factors moving behind those firmlevel data series and a large number of factor loadings for each firm-variable pair. Table 1: MSME Data in Thailand A. Variables Used ID Variable name Definition Real/Financial 1 sls_revn_amt Sales revenue Real 2 emp_cnt Number of employees Real 3 grs_pft_or_lss_amt Gross profit/loss Real 4 bef_tax_net_pft_amt Net profit before taxes Real 5 tot_aset_amt Total assets Real 6 wrkg_capl_amt Working capital Real 7 tot_liab_amt Total liabilities Financial 8 trd_dbtr_amt Trade debtors Financial 9 bnk_loan_od_amt Bank loans outstanding/overdraft Financial 10 trd_cr_amt Trade credit Financial 11 tot_curr_liab_amt Total current liabilities Financial 5 In this industry classification, for example, “Wholesale” is treated as a distinct industry classification from “Retail” while these two industries are often placed in one category. Given that the data contain a sufficient number of observations for these two industry classifications and that we prefer a larger number of industries to determine the number of semiaggregate time series data used in applying the P-PCA, we treat “Wholesale” as a distinct industry classification from “Retail.” 6 The application of the P-PCA method to the entire data of firm-year level granular data requires a certain level of computing power, which we do not have for this paper. Refining the computing process would be one of the most important ways to obtain SME-DI in the future from the raw data (i.e., firm-year level data). Continued on the next page 11 B. Relation Between the Second Factor and Variables C. Relation Between the Third Factor and Variables Source: Authors’ calculations. Bank Loan Overdraft Employees Total Gross Profit/Loss Amount Net Profit Before Taxes Sales Revenue Amount Total Assets Amount Total Current Liabilities Amount Total Liabilities Amount Trade Credit Trade Debtors Working Capital Amount (-1.0, -0.9) 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0130 0.0000 0.0000 (-0.9, -0.8) 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0130 0.0000 0.0260 0.0000 0.0000 (-0.8, -0.7) 0.0000 0.0000 0.0000 0.0000 0.0130 0.0000 0.0000 0.0260 0.0000 0.0130 0.0000 (-0.7, -0.6) 0.0260 0.0130 0.0130 0.0000 0.0130 0.0000 0.0260 0.0000 0.0130 0.0000 0.0000 (-0.6, -0.5) 0.0130 0.0000 0.0000 0.0130 0.0000 0.0000 0.0000 0.0000 0.0130 0.0130 0.0000 (-0.5, -0.4) 0.0000 0.0000 0.0519 0.0130 0.0000 0.0000 0.0130 0.0260 0.0000 0.0130 0.0000 (-0.4, -0.3) 0.0000 0.0000 0.0130 0.0130 0.0000 0.0130 0.0130 0.0000 0.0000 0.0000 0.0130 (-0.3, -0.2) 0.0130 0.0260 0.0000 0.0000 0.0000 0.0260 0.0130 0.0130 0.0000 0.0130 0.0000 (-0.2, -0.1) 0.0000 0.0000 0.0000 0.0000 0.0000 0.0260 0.0000 0.0130 0.0130 0.0000 0.0130 (-0.1, -2.22e-16) 0.0130 0.0000 0.0130 0.0260 0.0000 0.0130 0.0130 0.0130 0.0130 0.0130 0.0000 (-2.22e-16, 0.1) 0.0130 0.0000 0.0000 0.0130 0.0000 0.0130 0.0000 0.0000 0.0000 0.0260 0.0000 (0.1, 0.2) 0.0130 0.0130 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0130 (0.2, 0.3) 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0390 (0.3, 0.4) 0.0000 0.0260 0.0000 0.0130 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 (0.4, 0.5) 0.0000 0.0130 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 (0.5, 0.6) 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 (0.6, 0.7) 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0130 (0.7, 0.8) 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 (0.8, 0.9) 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 (0.9, 1.0) 0.0000 0.0000 0.0000 0.0000 0.0649 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 mean -0.285* 0.032 -0.401* -0.156 0.45 -0.143* -0.455* -0.402* -0.555* -0.269* 0.154 Bank Loan Overdraft Employees Total Gross Profit/Loss Amount Net Profit Before Taxes Sales Revenue Amount Total Assets Amount Total Current Liabilities Amount Total Liabilities Amount Trade Credit Trade Debtors Working Capital Amount (-1.0, -0.9) 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 (-0.9, -0.8) 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 (-0.8, -0.7) 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 (-0.7, -0.6) 0.0000 0.0130 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0130 (-0.6, -0.5) 0.0000 0.0130 0.0000 0.0000 0.0000 0.0000 0.0000 0.0130 0.0000 0.0000 0.0130 (-0.5, -0.4) 0.0000 0.0130 0.0000 0.0000 0.0000 0.0130 0.0130 0.0000 0.0130 0.0000 0.0130 (-0.4, -0.3) 0.0000 0.0260 0.0000 0.0260 0.0000 0.0260 0.0000 0.0130 0.0000 0.0000 0.0000 (-0.3, -0.2) 0.0000 0.0000 0.0000 0.0000 0.0000 0.0130 0.0000 0.0000 0.0000 0.0000 0.0130 (-0.2, -0.1) 0.0130 0.0130 0.0000 0.0130 0.0000 0.0130 0.0260 0.0260 0.0260 0.0260 0.0260 (-0.1, -2.22e-16) 0.0130 0.0000 0.0260 0.0130 0.0000 0.0130 0.0130 0.0000 0.0000 0.0260 0.0000 (-2.22e-16, 0.1) 0.0260 0.0130 0.0000 0.0000 0.0260 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 (0.1, 0.2) 0.0000 0.0000 0.0000 0.0260 0.0000 0.0130 0.0130 0.0390 0.0260 0.0260 0.0000 (0.2, 0.3) 0.0130 0.0000 0.0000 0.0000 0.0649 0.0000 0.0000 0.0000 0.0130 0.0000 0.0130 (0.3, 0.4) 0.0000 0.0000 0.0130 0.0000 0.0000 0.0000 0.0130 0.0000 0.0000 0.0000 0.0000 (0.4, 0.5) 0.0130 0.0000 0.0130 0.0000 0.0000 0.0000 0.0130 0.0000 0.0000 0.0130 0.0000 (0.5, 0.6) 0.0000 0.0000 0.0130 0.0000 0.0000 0.0000 0.0000 0.0000 0.0130 0.0000 0.0000 (0.6, 0.7) 0.0130 0.0000 0.0130 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 (0.7, 0.8) 0.0000 0.0000 0.0130 0.0130 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 (0.8, 0.9) 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 (0.9, 1.0) 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 mean 0.171 -0.365* 0.378* 0.017 0.183* -0.196* 0.035 -0.108 0.04 0.061 -0.299* 12 b. MSMEs operating in Bangkok (Capital City) Figure 3 depicts the top three factors in the case of regional-level indicators such as for the capital city, Bangkok. In this estimation, three factors explain 88% of the data variation. 9 Table 4 summarizes the distribution of the estimated impacts of each variable on three factors for the case of Bangkok. The associations between the three factors and the variables in the case of Bangkok suggest several things. The first factor implies that over the course of the COVID-19 pandemic, many MSMEs operating in Bangkok had a higher debt burden under lower profitability and sales revenue. The second factor suggests that profit shows a “hump” shape over time. This group of MSMEs likely faced profit losses during and after the pandemic. The third factor is negatively correlated with most financial-side variables while showing a positive correlation with sales revenue (no association with other real-side variables of employment and profits). MSMEs located in Bangkok were hurt financially in some industries not well covered by sales revenue. This estimation suggests that Bangkok-based MSMEs hurt by the pandemic by sales and profit were likely to obtain more bank credit. Government assistance measures during the pandemic might have been disproportionately allocated to capital city-based MSMEs in some industries. Figure 3: SME-DI for Bangkok (Capital City) SME-DI = Small and Medium-Sized Enterprise Development Index. Source: Authors’ calculations. 9 The contribution of each estimated principal component (PC) is 65% for PC1, an additional 16% for PC2, and a further 7% for PC3, for a total of 88% explained. -0.8 -0.6 -0.4 -0.2 0 0.2 0.4 0.6 0.8 2016 2017 2018 2019 2020 2021 2022 2023 Comp 1 Comp 2 Comp 3 Weights 13 Table 4: Factor Loadings based on Probabilistic Principal Component Analysis— Bangkok (Capital City) A. Relation Between the First Factor and Variables B. Relation Between the Second Factor and Variables Bank Loan Overdraft Employees Total Gross Profit/Loss Amount Net Profit Before Taxes Sales Revenue Amount Total Assets Amount Total Current Liabilities Amount Total Liabilities Amount Trade Credit Trade Debtors Working Capital Amount (-1.0, -0.9) 0.0000 0.0101 0.0000 0.0101 0.0000 0.0101 0.0101 0.0101 0.0101 0.0000 0.0000 (-0.9, -0.8) 0.0000 0.0101 0.0101 0.0101 0.0000 0.0000 0.0000 0.0000 0.0000 0.0101 0.0101 (-0.8, -0.7) 0.0000 0.0000 0.0000 0.0101 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 (-0.7, -0.6) 0.0000 0.0000 0.0000 0.0000 0.0101 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 (-0.6, -0.5) 0.0000 0.0000 0.0000 0.0000 0.0404 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 (-0.5, -0.4) 0.0000 0.0000 0.0101 0.0000 0.0000 0.0000 0.0000 0.0000 0.0101 0.0101 0.0000 (-0.4, -0.3) 0.0000 0.0000 0.0000 0.0000 0.0202 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 (-0.3, -0.2) 0.0101 0.0101 0.0202 0.0000 0.0000 0.0000 0.0000 0.0000 0.0101 0.0000 0.0000 (-0.2, -0.1) 0.0000 0.0000 0.0303 0.0202 0.0000 0.0000 0.0000 0.0000 0.0000 0.0101 0.0101 (-0.1, -2.22e-16) 0.0101 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0101 0.0000 0.0000 (-2.22e-16, 0.1) 0.0101 0.0000 0.0101 0.0000 0.0101 0.0000 0.0202 0.0000 0.0000 0.0000 0.0000 (0.1, 0.2) 0.0101 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0101 0.0000 (0.2, 0.3) 0.0000 0.0000 0.0101 0.0000 0.0000 0.0000 0.0101 0.0101 0.0000 0.0000 0.0101 (0.3, 0.4) 0.0000 0.0101 0.0000 0.0000 0.0000 0.0000 0.0101 0.0101 0.0101 0.0101 0.0101 (0.4, 0.5) 0.0202 0.0000 0.0000 0.0000 0.0000 0.0101 0.0000 0.0000 0.0202 0.0000 0.0202 (0.5, 0.6) 0.0000 0.0000 0.0000 0.0202 0.0000 0.0202 0.0202 0.0202 0.0101 0.0101 0.0000 (0.6, 0.7) 0.0202 0.0000 0.0000 0.0101 0.0101 0.0101 0.0000 0.0202 0.0000 0.0101 0.0000 (0.7, 0.8) 0.0101 0.0101 0.0000 0.0101 0.0000 0.0101 0.0202 0.0202 0.0101 0.0202 0.0101 (0.8, 0.9) 0.0000 0.0404 0.0000 0.0000 0.0000 0.0303 0.0000 0.0000 0.0000 0.0000 0.0202 (0.9, 1.0) 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 mean 0.306* 0.28 -0.197* -0.034 -0.317* 0.51* 0.242 0.399* 0.096 0.192 0.318 Bank Loan Overdraft Employees Total Gross Profit/Loss Amount Net Profit Before Taxes Sales Revenue Amount Total Assets Amount Total Current Liabilities Amount Total Liabilities Amount Trade Credit Trade Debtors Working Capital Amount (-1.0, -0.9) 0.0000 0.0000 0.0303 0.0101 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 (-0.9, -0.8) 0.0000 0.0000 0.0202 0.0000 0.0101 0.0101 0.0202 0.0000 0.0101 0.0000 0.0101 (-0.8, -0.7) 0.0101 0.0000 0.0101 0.0000 0.0000 0.0000 0.0000 0.0000 0.0202 0.0303 0.0000 (-0.7, -0.6) 0.0101 0.0000 0.0101 0.0000 0.0101 0.0000 0.0000 0.0202 0.0000 0.0000 0.0000 (-0.6, -0.5) 0.0000 0.0000 0.0101 0.0000 0.0101 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 (-0.5, -0.4) 0.0000 0.0101 0.0000 0.0101 0.0000 0.0000 0.0101 0.0101 0.0000 0.0101 0.0000 (-0.4, -0.3) 0.0101 0.0000 0.0000 0.0101 0.0000 0.0000 0.0000 0.0000 0.0101 0.0101 0.0000 (-0.3, -0.2) 0.0000 0.0000 0.0000 0.0101 0.0000 0.0000 0.0000 0.0101 0.0000 0.0000 0.0000 (-0.2, -0.1) 0.0000 0.0101 0.0000 0.0000 0.0000 0.0000 0.0000 0.0202 0.0000 0.0000 0.0000 (-0.1, -2.22e-16) 0.0101 0.0101 0.0101 0.0101 0.0000 0.0000 0.0101 0.0000 0.0101 0.0000 0.0202 (-2.22e-16, 0.1) 0.0101 0.0202 0.0000 0.0000 0.0000 0.0404 0.0202 0.0000 0.0000 0.0000 0.0000 (0.1, 0.2) 0.0000 0.0303 0.0000 0.0000 0.0000 0.0202 0.0101 0.0202 0.0000 0.0101 0.0101 (0.2, 0.3) 0.0000 0.0000 0.0000 0.0000 0.0000 0.0101 0.0000 0.0101 0.0101 0.0202 0.0101 (0.3, 0.4) 0.0202 0.0000 0.0000 0.0202 0.0101 0.0101 0.0202 0.0000 0.0000 0.0000 0.0101 (0.4, 0.5) 0.0000 0.0000 0.0000 0.0101 0.0000 0.0000 0.0000 0.0000 0.0202 0.0101 0.0000 (0.5, 0.6) 0.0000 0.0101 0.0000 0.0101 0.0505 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 (0.6, 0.7) 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0101 (0.7, 0.8) 0.0202 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0101 0.0000 0.0000 (0.8, 0.9) 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0202 (0.9, 1.0) 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 mean 0.038 0.07 -0.71* -0.043 0.117 0.026 -0.14 -0.189 -0.083 -0.208 0.219 Continued on the next page 14 C. Relation Between the Third Factor and Variables Source: Authors’ calculations. c. Manufacturing firms by region Here we investigate how the method we use can highlight the heterogeneity among MSMEs through the industry-region nexus. Figure 4 depicts three factors obtained from the data accounting for manufacturing firms operating in Bangkok, while Figure 5 depicts those in manufacturing outside Bangkok (local areas).10 It should be noted that in this exercise we input the raw firm level data of MSMEs into the model instead of the aggregate MSME data series (Appendix 2A). We can immediately identify that the first factors in each figure—which account for the largest variation of the data among the three factors—show different dynamics. Using the first factors from Figures 4 and 5 as an example, we can depict the firm’s activities in those two cases. From the associations between the first factor in Figure 4 and each variable, we found that a large number of capital city-based manufacturing MSMEs likely reduced their employment with larger borrowing while not necessarily seeing sales revenue and profit deteriorate during and post pandemic, although the estimates were not statistically significant (Appendix 2A1). From the associations between the first factor in Figure 5 and each variable, we found that many local manufacturing MSMEs (non-capital city-based firms) experienced lower profits from the pandemic onset until 2022. Their profits likely improved in 2023 (Appendix 2A2). 10 The contribution of each estimated principal component (PC) for Figure 4 is 72% for PC1, an additional 14% for PC2, and a further 5% for PC3, for a total of 91% explained. For Figure 5, it is 53% for PC1, an additional 20% for PC2, and a further 11% for PC3, for a total of 84% explained. Bank Loan Overdraft Employees Total Gross Profit/Loss Amount Net Profit Before Taxes Sales Revenue Amount Total Assets Amount Total Current Liabilities Amount Total Liabilities Amount Trade Credit Trade Debtors Working Capital Amount (-1.0, -0.9) 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 (-0.9, -0.8) 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 (-0.8, -0.7) 0.0101 0.0101 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 (-0.7, -0.6) 0.0000 0.0000 0.0000 0.0000 0.0000 0.0101 0.0101 0.0101 0.0101 0.0000 0.0000 (-0.6, -0.5) 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0101 0.0303 0.0000 0.0303 0.0000 (-0.5, -0.4) 0.0000 0.0000 0.0101 0.0000 0.0000 0.0101 0.0202 0.0101 0.0202 0.0000 0.0000 (-0.4, -0.3) 0.0101 0.0000 0.0202 0.0000 0.0000 0.0202 0.0202 0.0000 0.0000 0.0101 0.0303 (-0.3, -0.2) 0.0202 0.0000 0.0101 0.0202 0.0000 0.0000 0.0000 0.0101 0.0101 0.0000 0.0101 (-0.2, -0.1) 0.0000 0.0101 0.0101 0.0101 0.0101 0.0101 0.0000 0.0303 0.0202 0.0000 0.0101 (-0.1, -2.22e-16) 0.0202 0.0101 0.0101 0.0101 0.0101 0.0202 0.0101 0.0000 0.0000 0.0101 0.0202 (-2.22e-16, 0.1) 0.0101 0.0000 0.0000 0.0101 0.0000 0.0101 0.0101 0.0000 0.0101 0.0202 0.0101 (0.1, 0.2) 0.0000 0.0000 0.0202 0.0000 0.0101 0.0101 0.0101 0.0000 0.0101 0.0101 0.0000 (0.2, 0.3) 0.0000 0.0202 0.0000 0.0202 0.0101 0.0000 0.0000 0.0000 0.0101 0.0000 0.0101 (0.3, 0.4) 0.0101 0.0303 0.0101 0.0101 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 (0.4, 0.5) 0.0101 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0101 0.0000 (0.5, 0.6) 0.0000 0.0000 0.0000 0.0101 0.0101 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 (0.6, 0.7) 0.0000 0.0000 0.0000 0.0000 0.0404 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 (0.7, 0.8) 0.0000 0.0101 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 (0.8, 0.9) 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 (0.9, 1.0) 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 mean -0.099 0.148 -0.102 0.083 0.373* -0.206* -0.291* -0.374* -0.184* -0.156 -0.138* 15 Figure 4: Manufacturing—Bangkok Source: Authors’ calculations. Figure 5: Manufacturing—Outside Bangkok Source: Authors’ calculations. 16 This estimation suggests that manufacturing MSMEs based in Bangkok cost cuts through layoffs and successfully obtained bank credit during the pandemic, attributed to the lower impact of the pandemic on their sales revenue and profits. Emergency government assistance measures likely facilitated their access to bank credit. By contrast, local manufacturing MSMEs likely saw profits fall during the pandemic, as they were not well covered by government financial assistance (though they improved in 2023). d. Agribusinesses by region A similar exercise was conducted on agricultural MSMEs by region. The first factors were obtained from the data accounting for agribusiness firms operating in Bangkok and agribusiness MSMEs outside the capital (Figures 6 and 7) (Appendix 2B).11 The estimates simply show that profitability deteriorated amid the COVID-19 pandemic. Capital city-based agricultural MSMEs likely deteriorated in sales revenue and gross profit with no effective support from any improved working capital conditions during the pandemic, although estimates were not statistically significant (Appendix 2B1). Similarly, local agricultural MSMEs consistently saw their profits deteriorate over the course of the pandemic, accelerated by a constant slowdown in access to bank credit, even as working capital conditions improved (estimates strongly significant) (Appendix 2B2). Figure 6: Agriculture—Bangkok Source: Authors’ calculations. 11 The contribution of each estimated principal component (PC) for Figure 6 is 69% for PC1, an additional 19% for PC2, and a further 4% for PC3, for a total of 92% explained. For Figure 7, it is 59% for PC1, an additional 16% for PC2, and a further 11% for PC3, for a total of 86% explained. 17 Figure 7: Agriculture—Outside Bangkok Source: Authors’ calculations. Real and financial breakdown of the SME-DI In our baseline analysis, we simultaneously used all variables—categorized as both real and financial. One possible, useful exercise is to use the data accounted for by the real and financial variables separately so we can see the extent to which the baseline result is driven by each type of variable. The top three factors based on the data can be shown separately as real or financial variables (Figures 8 and 9). In both estimations, those three factors explain 87% of the data variation.12 A table on factor loadings is attached (Appendix 2C). The most important fact is that the dynamics of the factors in the two panels are similar. More precisely, the first factors in the two figures are increasing while the second factors are in u-shape. The third factors exhibit a more complicated trend in both the real and financial cases. Findings from these separate exercises by real and financial variables are consistent with the baseline results using all variables (see Table 2). For example, the first factor in real variables shows that MSME sales revenues and profits deteriorated during the pandemic while they likely strengthened working capital and the workforce (Appendix 2C1). The financial variables show that MSME borrowing from banks likely increased with more customers not yet paying for their products and services (Appendix 2C2). Although it is natural to see these links between variables over episodes such as the pandemic, this approach would be potentially useful to see whether or not real and financial variables account for MSME business dynamics, and how they do so. 12 The contribution of each estimated principal component (PC) for Figure 8 is 58% for PC1, an additional 19% for PC2, and a further 10% for PC3, for a total of 87% explained. For Figure 9, it is 62% for PC1, an additional 18% for PC2, and a further 7% for PC3, for a total of 87% explained. 18 Figure 8: SME-DI on Real Variables SME-DI = Small and Medium-Sized Enterprise Development Index. Source: Authors’ calculations. Figure 9: SME-DI on Financial Variables SME-DI = Small and Medium-Sized Enterprise Development Index. Source: Authors’ calculations. 19 Small sample exercises As mentioned, by using the pseudo dataset randomly chosen from all granular MSME data and re-running the P-PCA algorithm, we can see the extent to which the relatively small number of data can produce a SME-DI consistent with that using the entire MSME dataset. This experiment is useful for practitioners conducting, for example, survey analyses to construct a reliable indicator summarizing MSME activities. By doing so, we constructed two sets of SME-DI. Panel A (Panel B) of Figure 10 depicts an SME-DI based on the randomly chosen subsample that accounts for 50% (10%) of the entire MSME data.13 A factor loadings table is attached (Appendix 2D). First, we can check the consistency between Figure 1 and Figure 10A. The SME-DI based on 50% of the dataset basically replicates that based on all MSME data. Over the periods of our analysis, the first factor moved from a low to high level, the second factor exhibited a U-shape transition, with the third factor showing more complicated dynamics. A careful inspection of the association between each factor and each variable suggests that the association is compatible with Figure 1 and Table 2. For example, in the first factor based on a 50% sample, MSME sales revenues and net profits before tax deteriorated over the course of the pandemic while they received working capital financing from banks, which contributed to strengthening their balance sheets: a similar result as in the baseline estimate using all MSME data (Appendix 2D1). Figure 10: Small Sample Exercises A. 50% Sample 13 The contribution of each estimated principal component (PC) for Figure 10A is 44% for PC1, an additional 16% for PC2, and a further 12% for PC3, for a total of 72% explained. For Figure 10B, it is 41% for PC1, an additional 16% for PC2, and a further 12% for PC3, for a total of 69% explained. Continued on the next page 20 B. 10% Sample Source: Authors’ calculations. Second, we can confirm nonetheless that it is not necessarily easy to replicate a SME-DI based on the MSME sample by using randomly chosen and smaller samples. The consistency between the two sets of estimated factors worsens as the size of the subsample becomes smaller. Figure 10B accounts for the three factors based on a randomly chosen 10% size of the entire dataset. It fails to follow the original SME-DI based on the entire dataset. Also, the association between each factor and each variable does not necessarily share the same features as in the full sample (Appendix 2D2). To summarize, the small sample exercise helps us confirm that a certain reduction in sample size does not create any serious problems in estimating the SME-DI. As this critically depends on the target group, it would be interesting to see how this empirical study applies toward other datasets. 5.2. Firm-Year Level Raw Granular Data As a second set of results, we present the SME-DI based on firm-year level “raw” granular data, or MSME data not aggregated at the industry-region level but used as is. Figure 11 shows the top three factors obtained from the P-PCA method applied to 10% randomly chosen data. These three factors explain 62% of the entire variation of the data.14 Table 5 summarizes the distribution of the estimated impacts of each variable on the three factors. Similar to the baseline exercise using industry-region-year level aggregate data, applying the P-PCA yields the same number of estimated factor loadings as that of firms (around 5,000 firms depending on choice of variables). For each pair, the impacts are computed as the multiplication of the sign of the factor loading of each factor and share of the variation of each variable to that of the variation of the corresponding factor. First, we can identify that the first factor in Figure 11 is a mirror image of the first factor in Figure 1. We find that the profit has a positive (negative) association with the first factor in Figure 11 (Figure 1). This suggests that to some extent the 10% randomly chosen raw data reproduce the SME-DI based on more aggregated (but not full) data. 14 The contribution of each estimated principal component (PC) is 43% for PC1, an additional 12% for PC2, and a further 7% for PC3, for a total of 62% explained. ASIAN DEVELOPMENT BANK ASIAN DEVELOPMENT BANK 6 ADB Avenue, Mandaluyong City 1550 Metro Manila, Philippines www.adb.org ADB ECONOMICS WORKING PAPER SERIES NO. 785 June 2025 Designing a Country’s Small and Medium-Sized Enterprise Development Index Using Firm-Level Data The Case of Thailand This paper employs probabilistic principal component analysis to develop a new way to quantitatively assess what affects micro, small, and medium-sized enterprise (MSME) development nationally, by using granular firm-level panel data for 49,565 MSMEs in Thailand. The estimation results found a potential disproportionate effect of MSME policy interventions during and after the coronavirus disease pandemic. This underscores the importance of using a focused approach when designing policies for MSME development to facilitate more sustainable, resilient private sector growth. About the Asian Development Bank ADB is a leading multilateral development bank supporting inclusive, resilient, and sustainable growth across Asia and the Pacific. Working with its members and partners to solve complex challenges together, ADB harnesses innovative financial tools and strategic partnerships to transform lives, build quality infrastructure, and safeguard our planet. Founded in 1966, ADB is owned by 69 members—50 from the region. DESIGNING A COUNTRY’S SMALL AND MEDIUM-SIZED ENTERPRISE DEVELOPMENT INDEX USING FIRM-LEVEL DATA THE CASE OF THAILAND Shigehiro Shinozaki and Daisuke Miyakawa