Good Business Practices Improve Productivity in Myanmar’s Manufacturing Sector

Abstract We investigate the relationship between business practices and enterprise productivity using panel data with matched employer and employee information from Myanmar. The data show that micro, small, and medium-sized manufacturing enterprises in Myanmar typically adopt only a few modern business practices, and the persistence in the use is extremely low. Even so, we find a positive and economically important association between business practices and productivity. Specifically, the empirical results show that a one standard deviation difference in applied business practices (equivalent to applying an additional 4 to 5 of the 20 business practices in focus) is associated with an 8–10 per cent difference in labour productivity. Utilising the employer–employee information to estimate Mincer-type wage regressions, we find that workers receive about half to two-thirds of the productivity gain in higher wages. Overall, our findings support the notion of business practices as a production technology, and we find that workers and managers split the productivity gains evenly.

Even though modern management practices enhance both profitability and firm survival, far from all firms adopt them.Indeed, implementation of good management and business practices is much lower among MSMEs in emerging economies than elsewhere, mostly owing to low awareness of the likely benefits (Bloom, Schweiger, & Van Reenen, 2012;Bloom & Van Reenen, 2010).In this paper, we, therefore, examine the potential benefits of the use of business practices among registered private manufacturing firms in Myanmar by investigating how they relate to firm performance assessed by their association with sales, productivity, and wages.
The formal manufacturing sector in Myanmar mainly consists of micro-enterprises.The main industries are the food industry, textiles and apparel, and the wood industry, mostly producing a single product using hand tools or old second-hand machinery, which does not leave much room for innovation (Berkel et al., 2018).Enterprises face significant constraints, including inadequate access to infrastructure, finance, skilled labour, and technological knowledge (Hansen et al., 2020).While the strains of external factors for business growth in Myanmar are well recognised (Danquah & Sen, 2022;Tanaka, 2020), the role of internal factors, such as specific business factors and practices, remains less understood.
Myanmar is a particularly interesting case in which to analyse the relationship between business practices and firm-level productivity.Formal education and training systems in Myanmar have suffered from decades of underinvestment leaving Myanmar with the largest skills deficiency in South-East Asia (Ministry of Labour, Immigration and Population and International Labour Organization, 2016).Moreover, most enterprises in Myanmar carry out the needed training and skills upgrading 'in-house', with the result that business practices and managerial capabilities are critically important for closing the skills gap in Myanmar (Hansen, Kanay De, Rand, & Trifković, 2023).
To increase the understanding of the application and importance of business practices in the formal manufacturing sector in Myanmar, we use nationally representative panel data gathered in 2017 and 2019, which include officially registered private enterprises.Based on survey questions developed by McKenzie and Woodruff (2017), the Myanmar surveys have information about management quality in the form of 20 business practices related to marketing, buying intermediate supplies, record-keeping, and financial planning.The business practices questions are relatively straightforward with 'yes/no' responses about business activities.We create a simple index that measures the share of the 20 practices each enterprise has implemented up to three months prior to the interview.Following Bloom, Sadun, and Van Reenen (2016) and McKenzie and Woodruff (2017), we use this information to estimate the association between the application of business practices and firm productivity.However, inspired by Bender, Bloom, Card, Van Reenen, and Wolter (2018), we also investigate the relative importance of heterogeneity in management practices for observed differences in both physical capital and labour productivity.To our knowledge, our paper is the first to do this in a developing country context.
The enterprises in our sample use, on average, 27 per cent of the 20 business practices, while the median is 20 per cent and no less than 27 per cent of the firms do not apply any of the practices in 2017 or 2019.Thus, MSMEs in Myanmar appear to have ample room for improvement.Applications of individual practices and the total share are unstable over time.We find a correlation of only 0.17 between 2017 and 2019.Nevertheless, business practices appear to be valuable for those that apply them.Differences in the use of business practices of one standard deviation (23 percentage points) is associated with a difference of 15 per cent in sales per employee, conditional on standard firm and owner/manager characteristics.
We split the association with sales into price and productivity effects by estimating constant elasticity of substitution (CES) production functions.The analysis shows that the price link is very limited whereby the main link is with productivity.We find that a one standard deviation difference in the application of business practices is associated with approximately 8-10 per cent difference in productivity.We test whether there is a difference between labour-and capital-enhancing productivity, and we cannot reject the null hypothesis of equal productivity Good business practices improve productivity 1259 effects from better business practices.This supports a formulation of business practices as a technology, as proposed by Bloom et al. (2016).
The strong link between business practices and labour productivity is confirmed when we use information from employee questionnaires and regress individual workers' wages on the business practice score for the enterprise in which they are employed alongside standard Mincer-type regression controls.Specifically, we find that a one standard deviation difference in business practices is associated with a 4-5 per cent difference in wages.This difference amounts to half of the change in labour productivity estimated from the firm data.
The paper is organised as follows.Section 2 gives a brief overview of the directly related literature, mainly to present the quantitative results obtained for firms in other countries.Section 3 describes our data, with a focus on the application of business practices in our samples of firms.This is followed by a description of the empirical models, clarifying our estimation of CES production functions and Mincer-type wage regressions.We present the empirical results in Section 5 and offer some concluding remarks in Section 6.

A brief overview of related literature
The seminal work of Bloom and Van Reenen (2007) demonstrates that better managerial practices are strongly associated with firm performance.Bloom and Van Reenen distinguish between 'good' and 'bad' management based on 18 different management practices that describe firms' operations, monitoring, targets, and incentives.The proposition is that if companies closely monitor their processes, set measurable targets, and incentivise workers, they will outperform those that do not monitor their operations, have few or vague targets, and do not consider employee performance.The empirical results show that their management performance index is significantly associated with total factor productivity gains.
Bloom and Van Reenen's results spurred interest in investigating the relationship between managerial practices and firm performance across countries.Thus, in a follow-up study, Bloom, Schweiger, et al. (2012) use a subset of the management practices measures from the original study to investigate how management practices affect firm performance in 12 transition economies.They find that a difference in management score from the lowest to the highest quartile is associated with differences in sales of 7-13 per cent.
Equally inspired by Bloom and Van Reenen, McKenzie and Woodruff (2017) aim at measuring business practices thought to be especially beneficial for micro and small firms in emerging economies.They construct an index based on 26 practices related to marketing, buying and stock control, record-keeping, and financial planning activities.The activities are all meant to be practices that can be learned rather than innate entrepreneurial abilities.McKenzie and Woodruff show that use of more of the practices is associated with larger sales in microenterprises.Across seven countries, a one standard deviation (25 per cent) increase in business practices is associated with a 22 per cent increase in sales, on average.Advancing in a different direction, Bender et al. (2018) examine the extent to which management practices contribute to increased productivity through manager and employee ability in middle-sized manufacturing plants in Germany.They find a strong relationship between employee ability and management practices and higher productivity among firms with higher human capital stock.Controlling for employee ability when estimating the production function parameters reduces the association of productivity with management practices by 30-50 per cent, but the scores for management practices remain highly significant, indicating that good practices work through both labour and capital productivity.
Training to improve management practices has also been an active area of both activities and research, and McKenzie (2021) presents a meta-study of 15 experiments reported in 11 studies.
McKenzie finds that most of the studies have low power considering the perceived effect size, indicating that all experiments have wide confidence bounds, and most include zero.However, the weighted average of the experiments shows a statistically significant impact of training on sales of just below 5 per cent.
Among the studies that McKenzie does not include in his overall average is an experiment in Ghana that offers basic management training to firms in industrial clusters.The study shows that such training can improve business practices and some aspects of firm performance (Mano et al., 2012).However, while the experiment demonstrates a strong effect of the training programme on firm survival, evidence of the impact on firm sales and profitability is much weaker.In another large-scale study, Bloom, Eifert, Mahajan, McKenzie, and Roberts (2013) implement a field experiment in which they provide free consulting on management practices to randomly selected large textile firms in India.They find large, positive effects of their intervention.Adopting a bundle of new management practices (increasing the management score by 38 percentage points) raises productivity by 17 per cent in the first year and leads to the opening of more production plants within three years.In a follow-up study in which the plants were revisited nine years after the initial experiment, Bloom, Mahajan, McKenzie, and Roberts (2020) find that about half of the practices implemented in the experimental plants had been dropped.Even so, there is still a large and significant gap in practices between the treatment and control plants.

Data
We use data from the nationally representative Myanmar Enterprise Monitoring Survey (MEMS) conducted in 2017 and 2019 (Berkel et al., 2018;Hansen et al., 2020). 1 The survey includes about 1900 registered enterprises in each round that are representative of about 70,000 privately owned MSMEs in the manufacturing sector in Myanmar.Enterprises were selected using stratified area sampling based on all 15 regions and states in Myanmar.The sampling frame was based on lists of active manufacturing enterprises at the municipality level.In 2017, firms were selected at random from 35 randomly selected townships. 2The response rate was 78 per cent, which was not so much affected by the willingness to participate in the survey (only 33 enterprises refused to participate), but by inaccuracies in the sampling frame, which contained duplicate firm entries (13 cases), inaccurate address or owner details (23 cases), enterprises no longer operating (393 cases) and enterprises switching to the service sector (114 cases).In 2019, all firms still in operation were re-interviewed and a subset of firms was selected from updated municipal lists to replace those firms that stopped operating between 2017 and 2019. 3 The data include information needed for the estimation of production functions, such as revenue, profits, value-added, employment, labour costs, material costs, and the value of total assets for the last financial year before the interview.The data also contain key information about the owner or manager, such as sex, age, experience, and education, and similar information about the interviewed workers, in addition to information about their individual wages.
To get reliable estimates of the production function parameters, we omit firms that do not have wage workers or financial records of sufficient quality. 4After the data cleaning, we have Good business practices improve productivity 1261 1458 firms with consistent information in both years.In the empirical analysis, we check whether results are robust to exclusion of firms that did not have the same respondent in both 2017 and 2019, which reduces the sample by 29 per cent.
Up to five production workers were interviewed in each firm, yielding more than 7000 worker observations, of which only approximately 1400 workers were interviewed in both 2017 and 2019, reflecting in part the large labour turnover rates in Myanmar (Hansen et al., 2022).The selection of workers was done on-site by firm owners or managers.This may create a bias in our wage regressions as managers who apply more business practices may have selected the most productive workers for interviews.However, since nearly half of our firms had five or fewer workers (firms where all workers are interviewed), the possible selection bias would be increasing in firm size and especially concentrated among larger firms in our sample.To control for the possible bias, we include a variable measuring the share of the total employed workers in the firm that were interviewed in 2017 and 2019. 5

Business practices
The surveys include questions from McKenzie and Woodruff (2017) and inquire about specific firm activities related to marketing; buying supplies of intermediates; record-keeping; and financial planning. 6The activities are recorded as 'yes' or 'no', and we code these as 1 and 0. Table 1 gives an overview of the 20 business practices and their two-year evolution.The table shows that modern business practices are not in widespread use in the manufacturing sector in Myanmar.The most common practices, used by about half of the firms, are to negotiate lower prices with suppliers and to compare alternative suppliers.This focus on suppliers makes sense in micro firms in which intermediate inputs make up a large fraction of the variable costs.At the other extreme, only 6 per cent of the firms advertised in any form in 2019 (down from 14 per cent) and only 6 per cent of the firms kept formal accounts in 2017, increasing to 12 per cent in 2019.Keeping a formal account is often related to tax compliance requirements.According to KPMG (2018), the threshold for paying corporate income (capital gains) tax is MMK 10 million, a threshold not to be exceeded for three consecutive years to be exempted from corporate income tax.Approximately 43 per cent of the sample report having over MMK 10 million in annual gross profits (a lower bound for total corporate capital gains), which is well above the share of firms keeping formal accounts.
The share of the firms that chose to apply or not apply the specific business practices (instability) illustrates how the practices are clearly not something owners and managers always or never apply.Considering the two most widespread practices, we find that half of the firms changed to, or from, using this practice from 2017 to 2019.The other activities show less variation over time, but even for the important activity (from a compliance point of view) 'keeping formal accounts', 12 per cent of the enterprises changed practice.Coincidentally, only three per cent changed from keeping accounts to not keeping formal accounts between 2017 and 2019.
Following McKenzie and Woodruff (2017), we create an overall business practice index by adding up the number of practices each firm uses and then dividing the firm totals by the number of possible practices.In this way, the business practices index is in the interval [0; 1] and all individual practices have the same relative importance.Summary statistics for the business practice index are given at the bottom of Table 1 and the marginal distributions in 2017 and 2019 along with a cross-plot of the firm-specific adaptation in the two years and the distribution of changes from 2017 to 2019 are shown in Figure 1. 7  On average, our sample of firms applies 27 per cent of the 20 measured practices while the median firm applies 20 per cent.However, there is substantial variability.As seen from Figure 1, in each year it is most common not to use any practice; 23 per cent of firms in 2017 and 32 per cent of firms in 2019 answered 'no' to all 20 business practice questions.At the other extreme, only five firms report using all 20 practices and none of these are doing so in both years. 8 The trend from 2017 to 2019 is towards fewer practices.Thus, overall, the quality of management in the enterprises declined from 2017 to 2019.Moreover, given the instability in the use of individual practices, a weak correlation between the business practice index in 2017 and 2019 (0.17) is not surprising.The average absolute change in the use of business practices is 26 per cent, indicating that the change over time is of the same order of magnitude as the average use within each year.The cross-plot in the upper right corner and the histogram in the lower left corner of Figure 1 illustrate how the change in the use of business practices almost covers the full range.

Business practices and firm statistics
We show the partial associations between business practices and firm characteristics in Table 2, in which we give means and standard deviations for a set of firm and owner/manager attributes grouped by the firms' applications of business practices. 9Specifically, we have gathered the firm observations into five categories.Firms that do not apply any business practices make up the first category.Although applying no practices is the most common each year, as seen in Figure 1, the number of firm observations is larger in the second category in which the owner or manager applies 1-5 practices.There are also relatively many observations in the third group (6-10 practices), while the two last groups, applying 11-15 and 16-20 practices, have relatively few observations.Table 2 illustrates that several of the firm and owner characteristics are significantly associated with the use of business practices. 10

Good business practices improve productivity 1263
The first five rows in Table 2 give the key production variables.As seen, larger firms tend to apply more business practices than smaller firms and average wages for the firms' production workers are also positively correlated with business practices.In contrast, assets per employee (the capital-labour ratio) is negatively correlated with business practices, while neither revenue, value-added, nor input of intermediates per employee is partially associated with the use of business practices.
Looking at firm types, we find that family firms constitute larger shares of the firms with no or low application of the business practices compared to firms applying many practices.This is consistent with the evidence provided by Bloom and Van Reenen (2010) suggesting that familyoperated firms are less well-managed.Moreover, relatively more of the firms applying no or few practices report having no competitors.These results indicate that business practices may not be (perceived as) particularly useful for small firms with specialised (or niche) products.
There is a strong association between the educational level of the owner/manager and the application of the practices.Half of the firm owners/managers applying at least half of the practices have a higher education, whereas less than one-quarter of the owners/managers who do not use any of the practices have a higher education.

Empirical models
Turning to empirical models of the association between business practices and firms' performance, McKenzie and Woodruff (2017) take a broad view by explaining how good business practices may affect all parts of production decisions, including the quality of supplies and even input and output prices.They, therefore, focus on the association between sales (and profits) and business practices in their empirical analysis.
Looking beyond profitability, it is important to open the black box as some channels may be short-lived and part of zero-sum games in the market (say, some marketing practices), while other channels may lead to lasting increases in factor productivity (say, employing more productive workers).Bloom et al. (2016) develop a structural model in which business practices affect productivity directly as the activities are an independent part of the production function (business practices as technology).Bender et al. (2018) present a different approach as they show how business practices are associated with the human capital of the employees, arguing that better managers hire more productive workers.We combine these approaches by formulating simple empirical models of the association between the application of business practices and firm productivity.

Business practices and production functions
We consider the production decision by firm i at a given time t.The firm's revenue may be given by Good business practices improve productivity 1265 where P it and Q it are the output price and quantity, respectively.For simplicity, we formulate a Leontief function in intermediate inputs M it , while the (value-added) production function is a standard neoclassical production function in capital K it , labour L it , business practices b it , and the general productivity level A it .
As several of the business practices are related to the prices of supplies and to marketing, we think of the output price and the prices of intermediate inputs as functions of the business practices.Specifically, we assume that the two price functions P it ¼ P(b it ) and S it ¼ S(b it ) are continuous and monotonic and, given the business practice activities, firms are price-takers.
Next, we specify a CES production function in which we allow for differential effects of business practices on capital and labour productivity, respectively: Here, q ¼ (rÀ1)/r, is a transformation of the elasticity of substitution (r) between capital and labour inputs; p is the capital intensity parameter, and c j , j ¼ p,k,l, are the impacts of business practices on the output and input prices, capital productivity, and labour productivity, respectively; is the returns to scale parameter, and x it is an unobservable firm-specific productivity process.
When modelling production in value terms, we cannot separate marketing and other intermediate and output price effects from total factor productivity.This is a version of the output price bias discussed by De Loecker and Goldberg (2014).Therefore, we also formulate a valueadded model in 'quantity terms': Comparing the estimates of the two formulations will give an indication of the size of the association between output prices and business practices.
Turning to the associations with capital and labour productivity, we note that if good business practice is a technology, as suggested by Bloom et al. (2016), then we would expect c k !c l , while c k < c l if good business practice is associated with employing higher-quality workers as in Bender et al. (2018). 11In the latter case, the business practice index is acting as a proxy for differences in workforce education or innate worker ability across firms in the production function.The distinction between business practice as a capital-or labour-enhancing technology is important as the actions and activities in our business practice index are not directly measuring human resource management.

Estimation
We estimate linearised versions of the production functions using Kmenta's (1967) Taylor-series approximation.Thus, using small letters to denote log-transformed variables, we estimate the parameters of the value-based production function from the equation Most of the parameters of the production function ( 2) can be derived from the linearised equation.Specifically, b l ¼ (À1), b k ¼ p, and b kk ¼ 1 = 2 vqp(1Àp).The individual productivity impacts of business practices cannot be identified from the linearised equation.However, a weighted sum is identified: b we can test whether business practices enhance the productivity of one input significantly more than the other.
When estimating the parameters of the quantity version of the production function (3), we follow De Loecker, Goldberg, Khandelwal, and Pavcnik (2016) and seek to remove the input price bias using a control function.In our model, the input price bias mainly comes from the capital input, which we measure as the total value of assets.To separate the quality from the quantity of capital, we assume the quality is a function of output prices and unit input costs, conditional on sector and location.Thus, the linearised quantity version of the production function includes a polynomial function of the output price and the unit cost to control for quality differences in output and capital.The polynomial function is indicated by P(p it ): In this regression, we only have 'physical' effects of business practices as the estimated coefficient is the input-share weighted average of the capital and labour productivity effects b b q ¼ (pc k þ (1p)c l ).Hence, from the regression parameters, we can identify the two productivity parameters: ).We estimate the parameters of the production functions under different assumptions about the productivity process x it .If the process is exogenous and unobserved by the decision makers, then ordinary least squares (OLS) will yield unbiased estimates.A more realistic and often-used assumption is that the process is time-invariant, that is, if firms have different time-constant productivity levels, then the first difference (FD) estimation will yield unbiased estimates.Finally, following the ideas in Olley and Pakes (1996), Levinsohn and Petrin (2003) and Ackerberg, Caves, and Frazer (2015), one may assume the productivity process x it varies over time and that it follows an exogenous first-order Markov process.Endogeneity arises when, say, intermediate inputs and labour are flexible in the sense that the volumes of inputs are determined after the realisation of the random productivity level in period t.In this situation, the decision maker knows x it when the decision about the volume of intermediate supplies and labour input is made.It is customary to assume this decision is static, while capital inputs may be determined according to a dynamic decision rule.These assumptions allow for the proxy variable approach popularised by Olley and Pakes (1996) and Levinsohn and Petrin (2003).Wooldridge (2009) show how the parameters of the production functions can be estimated given the above assumptions, using a control function approach and instrumental variables regression.Specifically, we can formulate the regression models as where the dependent variable x it is one of the two value added variables (value or quantity), the vector w it has the flexible inputs (labour), s it has the state variables (capital), while c it are the conditioning variables.Business practices may be considered either a state variable or a flexible input.If business practices are determined before the realisation of the productivity shock, then they can be considered a state variable, while they are flexible inputs if they are determined after the realisation of the shock.In the following, we will estimate the association with productivity under both assumptions.
The theory in Olley and Pakes (1996) and Levinsohn and Petrin (2003) implies that the productivity shock can be identified using a proxy-variable.We follow Levinsohn and Petrin (2003) Good business practices improve productivity 1267 and use intermediate inputs as the proxy.Hence, the productivity shock is identified as a function of the state variables and the intermediate inputs.
Substituting g(.) for the productivity shock in the regression model, we obtain If the business practices are endogenous in the sense of being chosen after the realisation of the productivity shock, then we can approximate h(.) by a polynomial function in capital and intermediate inputs and estimate the association from Equation (8).However, this will only give the composite parameters (b b and b kb ) and, as shown in Ackerberg et al. (2015), the parameters may not be identified.
To estimate all parameters of the production function, with due account of the identification problems illustrated in Ackerberg et al. (2015), we follow the suggestion in Wooldridge (2009) and use a mixed control function and instrumental variable estimator.According to the production model, the conditional expectation of the productivity process is a function of the past realisations the state variables and the intermediate inputs Hence, the process at time t can be given as where a it is the innovation in the process, which is uncorrelated with the state variables, but correlated with the flexible inputs.Inserting the new expression for the productivity process, we obtain a slightly different regression model.
For known functions f(.) and g(.), the production function parameters are identified in this model, but the flexible inputs are correlated with the error term and must be estimated using an IV-estimator.Wooldridge (2009) suggests approximating the functions by polynomials and using lagged observations of the flexible inputs as instruments.In the results section, we estimate the parameters using Equations ( 8) and ( 11) in addition to OLS and first differences of Equation (6).

Worker wages
Our final dimension in the empirical analysis is a Mincer-type regression of the interviewed workers' wages.Thus, we regress the log of the individual workers' wages (w ijt ) on the business practice index while controlling for worker-specific information z ijt , and firm-specific information f it , where the latter includes sector and region fixed effects, and the indicator for same respondent in the two survey rounds.The regression model is formulated as The coefficient on the business practices k b , measures the average wage premium for workers employed in firms that are applying the given share of business practices.As we condition on education and tenure, the association with wages measures either unmeasured worker abilities that the manager can identify or an increased productivity originating from the workplace.

Firm performance
Table 3 reports estimates of the central production function parameters.In all regressions, we condition on the firm attributes shown in Table 2, on state/region and sector fixed effects and on the indicator for having the same respondent in both survey rounds. 12Moreover, we have divided the business practices variable by the standard deviation in 2019 (0.23) such that the business practice parameter estimates illustrate the difference in productivity associated with a one standard deviation difference in business practices in 2019.All regressions are for the panel of 1458 firms and the reported standard errors allows for heteroskedasticity and arbitrary correlation within 290 township/sector clusters, which are the clusters reported in the MEMS surveys.
Regression (1), which has the log of sales per employee as the dependent variable, is shown to facilitate direct comparison with the results of Bloom, Schweiger, et al. (2012) and McKenzie and Woodruff (2017).As seen, a one standard deviation difference in application of business practices is associated with a 0.14 log-points (15 per cent) difference in sales.This is slightly larger than the finding of Bloom, Schweiger, et al. (2012), but less than half of the difference of 35 per cent in sales in the sample of firms from McKenzie and Woodruff (2017).Thus, business practices are much less applied in Myanmar compared to the samples from McKenzie and Woodruff (2017), and there is a notably weaker association with sales.
Regressions (2) and ( 6) in Table 3 report OLS estimates of the value-and quantity-based production functions, respectively, while regressions (3) and ( 7) are first difference (FD) regressions.The four regressions all show significantly decreasing returns to scale, with the FD estimator finding much lower returns to scale than the OLS estimator.These results are on par with many previous results from firm specific production function estimates and they illustrate the endogeneity problems discussed in the literature (see Ackerberg, Benkard, Berry, & Pakes, 2007).As for the estimated productivity effect of business practices, the combined parameter estimate (ĉ ¼ bp =ð bl þ 1Þ) is a 10-13 per cent difference in productivity associated with a one standard deviation difference in use of business practices.
Regressions ( 4) and ( 8) are the control function regressions in which only the parameters on labour and business practices are identified.The control function is a third order polynomial of capital per employee and intermediates per employee.Finally, regressions (5) and ( 9) are control function, iv-regressions, estimated using continuously updated GMM, with one and two year lagged employment as instruments for current employment.In regressions ( 5) and ( 9) we include business practices as a state variable such that it is part of the control function polynomial.
All regressions reveal that the production function is better approximated by a CES than a Cobb-Douglas function, as b kk is significantly different from zero.Another common outcome is that b kp is statistically insignificant, whereby we cannot reject the hypothesis that the productivity parameters are equal implying that the business practices can be modelled as a technology.
Further, comparing the value-based and quantity-based production function parameters, the price effect appears to be small and quite likely zero.As seen, the estimate of the composite effect (ĉ) is often larger in the quantity-based regressions than in the value-based regressions.One notable difference is the control function regressions ( 4) and ( 8).However, the quantitybased production function in regression ( 8) appear to suffer from the identification problem discussed in Ackerberg et al. (2015) as the three parameter estimates are all very close to zero.

Notes:
The business practices variable has been divided by the standard deviation (0.23) such that the parameter estimates illustrate the difference in productivity associated with a one standard deviation difference in business practices.All regressions include State/region and sector fixed effects as well as firm and owner characteristics given by: An indicator for no competitors; the log of the number of competitors if any; an indicator for family firms; firm age; an indicator for female ownership; indicators for the educational level of the owner; an indicator for change in survey respondent between 2017 and 2019.Robust standard errors in parentheses, clustered at the township-sector level.Regressions ( 5) and ( 9 The two GMM regressions have very sensible parameter estimates, with constant returns to scale, capital's share around 0.27-0.28and an elasticity of substitution significantly above one.In the two regressions, the estimated productivity difference is about 8 per cent if firms differ by one standard deviation in their application of business practices. In Appendix A, we report additional production function estimates arising when we change the regression specification or the sample of firms.Table A1 reports estimates when a full set of interactions of state/region and sector fixed effects are included.Table A2 has estimates for the subsample of 1109 firms for which the respondent was the same in both survey rounds.Table A3 reports estimates for the subsample of 1086 micro firms.Finally, Table A4 reports estimates for the subsample of 1056 non-family firms.The appendix tables illustrates that the main results are robust to the reported changes in the specification and the sample variations.Compared with Table 3, the magnitude of the estimated coefficients is larger for the sample with the same respondent in 2017 and 2019 (Table A2), and when family-firms are excluded (Table A4), but smaller for the sample restricted to micro firms (Table A5).

Workers' wages
Turning to the wage regressions, we first show summary statistics of key variables for the interviewed employees in Table 4, using the same structure as in Table 2.The information we have about the employees include the number of years they have worked as wage workers prior to the employment at the current firm (prior experience) and the time they have been employed in the present firm (tenure).Moreover, we have information about their sex, marital status, education, and relation to the firm owner.
All the worker attributes vary significantly with the firms' application of business practices.Specifically, family members are more frequently employed in firms that do not apply many practices, and this is also the case for married workers.In contrast, the share of female workers is higher in firms applying many practices.Workers with higher education are also relatively more frequent in firms that apply many practices.Several of these associations may be driven by the correlation between business practices and firm size shown in Table 2.
Table 5 reports Mincer-type regressions, augmented with firm specific information, including the business practices applied by the firm in which the workers are employed.We seek to Note: Ã indicates a statistically significant association with business practice categories at the 5 per cent level.
Good business practices improve productivity 1271 Notes: The business practices variable has been divided by the standard deviation (0.23) such that the parameter estimates illustrate the difference in productivity associated with a one standard deviation difference in business practices.Robust standard errors in parentheses, clustered at the township-sector level.Regressions (1), (2), and ( 4) are based on employee data from both 2017 and 2019 while regressions (3) and ( 5) are only using employee data from 2019.The estimated parameters for all control variables, except for fixed factors, are reported in the Table .Regressions (4) is estimated using continuously updated GMM.Statistical significance is indicated by Ã p < 0.10, ÃÃ p < 0.05, ÃÃÃ p < 0.01.
partially control for the selection of workers within the firms by including a variable measuring the share of the firms' total employment that are selected for interviews.If owners of larger firms select high paid workers, then the association with wages should be negative, such that a low share (a large firm) is associated with higher wages.As seen, we do find this result.
In regression (1) we estimate the wage model with firm fixed effects.As the application of business practices varies strongly over the two periods, we are estimating the within firm covariance between wages and business practices and regression (1) shows a well determined absence of such an association.Still, this does not mean that worker wages are not affected by business practices, as the time average of the business practices is part of the firm fixed effect.
In regression (2) we omit the firm fixed effects, replacing them with sector/year and region/year fixed effects.This regression illustrates how workers in firms applying more business practices, on average, get higher wages compared to workers in firms that apply fewer practices.In regression (3) we estimate the same model, while we limit the sample to 2019, in accordance with the production function regressions in Table 3.We find an average wage premium of 4 per cent when the application of business practices differs by a standard deviation.This is about half of the productivity estimate.
Finally, in regressions ( 4) and ( 5) we replace the business practice index by the average over the two years (business practice, avg.) and the deviation from the average in each year (business practice, dev.).This transformation emphasises the firm fixed factor in the index and if part of the variation in the index is created by measurement error, then the time average will give a more precise estimate of the association.In regression (4) we estimate the model based on both years, while we restrict the sample to 2019 in regression (5).The latter regression shows that a difference of one standard deviation in the average application of business practices is associated with an average difference in the worker's wages of 5.6 per cent.This is more than twothirds of the productivity estimate.
Overall, the wage regressions lend support for the idea that business practice may be modelled as a management technology.Workers employed in firms applying more practices have higher wages, on average, compared to workers with similar attributes, employed in firms applying fewer practices.However, the workers do not get the full share of the productivity difference.Thus, better business practices also increase the gross profits per employee. 13

Discussion and conclusion
MSMEs are often seen as the engine for inclusive economic growth and development in emerging economies.Therefore, substantial efforts have gone into understanding and promoting survival and growth of MSMEs in the manufacturing sector in almost all countries across the globe.Both research and policies have mainly concentrated on external constraints such as the investment climate, credit, infrastructure, and institutions.However, a parallel literature has stressed the importance of internal factors, including entrepreneurial abilities, management, and business practices.Understanding the influence of business practices on firm performance is important because such practices can be taught and learned, thereby providing an opportunity for governments to directly support MSMEs through business training.
To increase our understanding of the use and importance of business practices in a setting with a severely constrained and underdeveloped private sector, we look at registered firms in the manufacturing sector in Myanmar.In contrast to earlier studies of business practices, the firms are from two rounds of a nationally representative survey.Using closed-ended questions developed by McKenzie and Woodruff (2017), the survey has information about business practices in the form of 20 activities that are relevant for MSMEs.On average, 1458 enterprises interviewed in 2017 and 2019 apply 27 per cent of the 20 business practices, while a large fraction (27 per cent) of the firms do not apply any.Moreover, use of the practices is volatile over time.

Good business practices improve productivity 1273
In line with results of other studies, use of business practices appears to be valuable.A one standard deviation difference in the business practice index-equal to an increase from applying none to applying approximately five of the practices-is associated with a difference in sales per employee of about 15 per cent.
We disentangle this effect on sales into price and productivity effects by estimating CES production functions.We find the price link close to zero, while the stronger association is between business practices and productivity.We estimate the association with labour productivity from the CES production function and find a one standard deviation difference in the business practice index to be associated with a 8-10 per cent difference in labour productivity.
We verify the link between business practices and labour productivity when we estimate Mincer-type regressions of individual workers' wages and include the business practice score for the enterprise in which they are employed.The Mincer regressions show that a one standard deviation difference in the business practice index is associated with a 4-5.5 per cent difference in wages.This is between one-half and two-thirds of the difference in productivity estimated from the firm data.Thus, firms with better business practices have higher labour productivity and at the same time firms adopting more business practices pay higher wages to workers having the same observable characteristics as workers in firms with poorer business practices.These findings are well in line with the idea of a management technology developed in Bloom et al. (2016).Notes 1.In the MEMS, both informal and formal enterprises are interviewed.We focus on registered enterprises for which we have a representative sample.Details about the informal enterprises can be found in Berkel and Tarp (2022).2. The township sampling was based on probability proportional to size within each of the 15 states/regions.3.As described in Berkel and Tarp (2022) and Hansen et al. (2022) the attrition rate was relatively low between 2017 and 2019.Only 9 per cent of the firms interviewed in 2017 closed their business or refused to participate in the survey.BGLW tests (Becketti, Gould, Lillard, & Welch, 1988) (2017).The smaller number of business practices is context-dependent, as some recordkeeping and financial practices do not exist in Myanmar.Furthermore, for the present analysis, we had to exclude a question about stockkeeping because of ambiguity in the translation of the question.
7. The plot in the upper left corner of Figure 1 shows the fraction of enterprises that apply each of the different values of the business practices index in 2019 (20 values from a minimum of 0 to a maximum of 1.The plot in the lower right corner shows the histogram for 2017.The large cross-plot in the upper right corner shows the values of the business practices index for each enterprise in 2017 and 2019.Finally, the plot in the bottom left corner shows the fraction of enterprises for each value of the instability variable measuring the change in the business practices index values between 2017 and 2019. A little random noise (jitter) has been added to the individual observations in the crossplot to increase readability.8.In comparison, McKenzie and Woodruff (2017) find that, on average, firms use 39 per cent of the total number of business practices measured.However, there is large variation within and across countries.While firms in Mexico, on average, applied 30 per cent of all practices measured, firms in Kenya applied 52 per cent and firms in Nigeria 72 per cent, on average.Comparing our averages with those of McKenzie and Woodruff (2017) may not be meaningful as the samples they work with are not nationally representative.For example, their sample from Nigeria comprises young, highly educated entrepreneurs, which could explain the observed high score for firms from this country.9.All monetary values are in 2018 prices.We have used regional CPI deflators such that the values are deflated to 2018 prices for an average over the regions and states in Myanmar.In 2018, US$1 was around 1,430 kyat.Financial values (per employee) can be compared to the national minimum wage in 2018 of 4,800 kyat per day, which is approximately 1.5 million kyat per year.10.We test the hypotheses of independence using ANOVA for continuous and count variables (for example, revenue and employment) and Pearson's v 2 -test of independence for categorical variables (for example, if the business is a family firm).11. c k ¼ c l :¼ c may be considered the pure technology case as the impact of business practices simplifies to exp(qcb it ).12.The sector classification is based on the Myanmar Standard Industrial Classification.
We have firms in 22 different 2-digit level sectors and in 14 state/region categories.Myanmar has 14 regions and states and the Nay Pyi Taw Union Territory.In the sampling of firms for the MEMS survey, Chin State and Rakhine State were joined to have enough townships and firms in the random selection.We keep this stratification in the regression analyses because the cleaned data only include one firm from Chin State.13.This is confirmed in regressions of the gross profit per employee that are similar to the sales regression in Table 3.

Notes:
The business practices variable has been divided by the standard deviation (0.23) such that the parameter estimates illustrate the difference in productivity associated with a one standard deviation difference in business practices.All regressions include State/region and sector fixed effects as well as firm and owner characteristics given by: An indicator for no competitors; the log of the number of competitors if any; an indicator for family firms; firm age; an indicator for female ownership; indicators for the educational level of the owner.Robust standard errors in parentheses, clustered at the township-sector level.Regressions ( 5) and ( 9) are estimated using continuously updated GMM.

Notes:
The business practices variable has been divided by the standard deviation (0.23) such that the parameter estimates illustrate the difference in productivity associated with a one standard deviation difference in business practices.All regressions include State/region and sector fixed effects as well as firm and owner characteristics given by: An indicator for no competitors; the log of the number of competitors if any; an indicator for family firms; firm age; an indicator for female ownership; indicators for the educational level of the owner; an indicator for change in survey respondent between 2017 and 2019.Robust standard errors in parentheses, clustered at the township-sector level.Regressions (5) and ( 9) are estimated using continuously updated GMM.Good business practices improve productivity 1281

Notes:
The business practices variable has been divided by the standard deviation (0.23) such that the parameter estimates illustrate the difference in productivity associated with a one standard deviation difference in business practices.All regressions include State/region and sector fixed effects as well as firm and owner characteristics given by: An indicator for no competitors; the log of the number of competitors if any; firm age; an indicator for female ownership; indicators for the educational level of the owner; an indicator for change in survey respondent between 2017 and 2019.Robust standard errors in parentheses, clustered at the township-sector level.Regressions (5) and ( 9) are estimated using continuously updated GMM.

Figure 1 .
Figure 1.The distributions of business practices in 2017 and 2019.Source: Authors' calculations based on the Myanmar Enterprise Monitoring Survey (MEMS).Notes: The plot in the upper left corner shows the fraction of enterprises that apply each the 20 different values of the business practices index in 2019.The plot in the lower right corner shows the same fractions for 2017.The large cross-plot in the upper right corner shows the values of the business practices index for each enterprise in 2017 and 2019.Finally, the plot in the bottom left corner shows the fraction of enterprises for each value of the instability variable measuring the change in the business practices index values between 2017 and 2019.A little random noise (jitter) has been added to the individual observations in the cross-plot to increase readability.
) are estimated using continuously updated GMM.Statistical significance is indicated by Ã

Table 1 .
Application of the individual business practice components The questions asked start with the statement 'During the last 3 months has your business … '.Confirmation ('yes') is recorded as 1 while negation ('no') is recorded as 0. The numbers given in the table are the fractions of firms confirming they have applied the business practice.Instability is the average absolute change in each of the business practice scores from 2017 to 2019.This is equal to the share of firms that changed the use of the business practice.

Table 2 .
Summary statistics for key firm variables, by business practice categories Source: Authors' calculations based on the Myanmar Enterprise Monitoring Survey (MEMS).Notes: Ã indicates that the variable has a statistically significant association with business practice categories at the 5 per cent level (tested using ANOVA, for continuous variables and Pearson's chi-squared test of independence for categorical variables).All monetary values are in 2018 prices.We have used regional consumer price index deflators such that the values are deflated to 2018 prices averaged over the regions and states in Myanmar.

Table 3 .
Production function estimates

Table 4 .
Summary statistics for key employee variables, by business practice categories Source: Authors' calculations based on the Myanmar Enterprise Monitoring Survey (MEMS).

Table 5 .
Business practices and employee wages Source: Authors' calculations based on the Myanmar Enterprise Monitoring Survey (MEMS).
The surveys include 20 of the 26 business practices from McKenzie and Woodruff

Table A1 .
Production function estimates with full sector and region/state interactions The business practices variable has been divided by the standard deviation (0.23) such that the parameter estimates illustrate the difference in productivity associated with a one standard deviation difference in business practices.Allregressionsincludeinteractions of State/region and sector fixed effects, as well as firm and owner characteristics given by: An indicator for no competitors; the log of the number of competitors if any; an indicator for family firms; firm age; an indicator for female ownership; indicators for the educational level of the owner; an indicator for change in survey respondent between 2017 and 2019.Robust standard errors in parentheses, clustered at the township-sector level.Regressions (5) and (9) are estimated using continuously updated GMM.Statistical significance is indicated by Ã

Table A2 .
Production function estimates: only firms with same respondent in 2017 and 2019 Authors' calculations based on the Myanmar Enterprise Monitoring Survey (MEMS).

Table A3 .
Statistical significance is indicated by Ã Production function estimates: only micro firms Authors' calculations based on the Myanmar Enterprise Monitoring Survey (MEMS).

Table A4 .
Production function estimates: excluding family firms Authors' calculations based on the Myanmar Enterprise Monitoring Survey (MEMS).
Statistical significance is indicated by 1282 P.Falco et al.