Data

26

misallocation based on Wu (2018) methodology. David & Venkateswaran (2019) also propose interesting alternative measures of misallocation to HK¹⁴. However, we are constrained by our data to implement their approaches.

27

frame.¹⁵¹⁶ Survey imperfections such as the selection of units with unequal probabilities cause bias and departure of the sample from the reference population. Thus, sampling weights were constructed to make the survey data representative of our targeted population. It should be noted that the misallocation measure is an aggregate national measure and thus the data should reflect the population of firms in the country. We use weights to construct nationally representative measures of misallocation. Because we followed a two-stage sampling design, base weights were constructed to reflect the selection probabilities at each stage. See Appendix for Chapter 1 for a detailed explanation of the data collection procedure.

It is worth noting that we constructed sampling weights to make the survey data representative of our targeted population (see Appendix for Chapter 1 for details). Because we followed a two-stage sampling design, base weights were constructed to reflect the selection probabilities at each stage. There are two issues that we need to express concerning the use of weights. First, the survey for the informal sector is not representative of the total population of informal manufacturing sectors in Zimbabwe but informal enterprises in the respective geographical areas surveyed. Second, guided by the literature, we used weighted estimations only in the non-parametric analysis, as there is inconclusive econometric debate on the use of weights in regression analysis (Lohr, 2019; Cochran, 2007; Deaton, 1997; DuMouchel & Duncan, 1983).

These studies argue that the use of weights in regression analysis may impose additional noise to the standard errors, leading to inefficient estimates.

Our analyses of the formal and informal sector results are based on industrial sectors that match both sectors, that is, metal; wood; and textile. While this may affect the aggregate representativeness of our data (because we drop other industries) this provides a nuanced comparison between formal and informal sector firms with the same industry category. Since this Chapter is based on non-parametric analysis, weights were used for both formal and informal sector analysis.

15 In the first stage, the two main (or main area where informal production is located in a single area) informal areas for each of the industry strata were selected. Where it is possible or sensible these areas were then divided into blocks (enumerating areas) with roughly equal numbers of firms based on spatial area or building complex.

Blocks were then randomly selected. In the second stage, firms within each of these randomly selected blocks were listed. A random sample of firms was then selected for interviewing purposes from the listed firms in each randomly chosen block.

16 Our sampling procedure rules out a likelihood of the informal firms being a select sample of the informal firms; for example, that they constitute the upper end of the productivity distribution of the population of informal firms.

28

The data contains essential information that allows us to construct the measure of markets distortions and misallocation. The information includes data on sales and production, raw material costs, indirect costs, capital stock and labour inputs among other important information. In our analysis, labour input is measured by the wage bill¹⁷ (rather than employment) to account for differences in human capital and hours worked (as in Hsieh &

Klenow, 2009). The capital stock is measured by netbook and market value of fixed assets, summed across vehicles, machinery and equipment, and land and buildings. Value-added is computed as the difference between sales and cost of raw materials, overhead expenses, and energy costs (electricity, fuel, gas).

To implement the Hsieh & Klenow (2009) framework, we also need information on the elasticity of substitution (𝜎), interest rate (𝑅) and industry level of labour and capital share (𝛼_𝑠).

There is little consensus in the literature on which effective magnitude on this parameter.

Following the HK framework, we fixed this parameter at 𝜎 = 3¹⁸ in our baseline computations as this is the value most used in similar empirical approaches. The value of the elasticity of substation is likely to impact the results of TFP gains, thus we will consider a robustness check using alternative values of 𝜎. We also set 𝑅 = 12.5% drawing from the average interest rate reported in our data for the formal and informal firms. We calculate capital share by subtracting the industry mean of labour expenditure as a share of value-added at firm level from one, that is, 𝑎_𝑠 = 1 − ^{𝑤𝐿𝑠𝑖}

𝑃_𝑠𝑖𝑌_𝑠𝑖). In literature, most studies have widely used the US capital share, which is set at one third. In our sample, we dropped all observations where value-added could not be calculated because of either missing variables or those where value-added was negative (14 firms). Having considered overlapping industries between formal and informal sector, as discussed above, this left a sample of 107 out of 195 formal firms and 128 out of 130 informal sector firms. Table 2.1 below provides some key variables in our analysis.

17The sum of wages, bonuses, and benefits.

18 This parameter does not alter our measures of distortions, but rather their effect on aggregate productivity.

29 Table 2. 1. Summary statistics for key variables

Formal Sector Informal Sector Obs Mean Std. Dev. Obs Mean Std. Dev.

Value added per worker (log) 100 8.44 1.24 122 7.74 0.86 Value added per capital (log) 95 -0.08 1.80 121 2.24 1.33 Capital/Labour ratio (log) 95 8.52 1.37 121 5.51 1.22

Labour costs (log) 95 11.56 1.85 113 8.24 1.03

Firm Size (employment) 100 66.03 92.19 122 3.12 1.50

Firm age 100 34.48 23.23 117 8.55 6.41

Notes: For the formal sector, the summary statistics are only for overlapping industries with the informal sector (Metal, Textile and Wood) for plausible comparisons.

The summary statistics in Table 2.1 shows differences in firm size between the formal and informal sector firms, with the formal sector having average employment of 66 as compared to the informal sector of 3. Firms in the informal sector are on average younger than those in the formal sector. However, value-added per worker (labour productivity) is comparably the same in the two sectors. Noticeably, the formal sector firms have a lower value-added per capital as compared to informal sector firms. This signals that firms in the formal sector produce using substantially higher capital-labour ratios.