• No results found

Export Experience and Firms’ Performance

2.3. Methodology

2.4.1. Data overview

2.4.1.1. Export age

Figure 2.2 shows the export age for all firms that started exporting from 2001 to 2012. The number of observations decreases throughout the export age, implying that the number of firms that export for a long period of time is small relative to the number of firms that start exporting. It is important to note that more than half of the firms in the observations exported only once (export age equal to 1). This shows that many firms tried to export but they could not make a profit and survive in the foreign markets in the first year of exporting. They then stopped exporting. Only 5.2 percent of firms that start exporting in 2001 stayed in the export market for 12 years and 6.4 percent of firms which starting exporting in 2001 and 2002 continued exporting for 11 years (see Table 2A.3, Appendix 2).

Figure 2.2. The export age Source. Statistik Industri, calculated

2.4.1.2. Productivity

To evaluate the learning process of firms, I use TFP as the outcome variable. In this study, I construct the TFP series using the Levinsohn and Petrin (2003) model, which is

an extension of the Olley and Pakes (1996) model.10 The strength of this approach lies

in two innovations. Similar to Olley and Pakes’s estimates, the Levinsohn and Petrin model is preferable to ordinary least squares (OLS) estimates in that it controls for simultaneity bias in the production function that may arise from input variables and unobserved productivity shocks. Since firm-specific productivity is known by the firm but not by the econometrician, a firm may adjust its inputs in response to the productivity shocks. In addition, this method also reduces the selection bias because

10 Many models can be used to construct TFP. See Aswicahyono (1998) to estimate TFP growth for

Indonesia industries for period 1975–93.

1 2 3 4 5 6 7 8 9 10 11 12 No obs. 11,297 5,175 3,250 2,330 1,670 1,003 712 244 204 137 99 32 0 2,000 4,000 6,000 8,000 10,000 12,000 N umber of obs er vat ions

Number of observations by export age

some unproductive firms might leave the industry and be replaced by more productive enterprises. Furthermore, Levinsohn and Petrin could be preferable to Olley and Pakes because the latter uses investment as a proxy for unobservable shocks. This investment variable is most likely to have been derived from the capital and may not smoothly respond to productivity shock. Furthermore, the investment proxy is only valid for firms that report non-zero investment. But then, Levinsohn and Petrin uses intermediate inputs, such as materials and electricity, instead of investment. Therefore, using Levinsohn and Petrin in this study avoids truncating all the zero investment firms.

To estimate the TFP, I start by specifying the production technology of a firm as a Cobb-Douglas function:

𝑦𝑦𝑖𝑖 = 𝛽𝛽0+𝛽𝛽𝑙𝑙𝑙𝑙𝑖𝑖+𝛽𝛽𝑖𝑖𝑘𝑘𝑖𝑖+𝜀𝜀𝑖𝑖, (2.6)

where 𝑦𝑦𝑖𝑖 is the firm’s value added at time t, 𝑙𝑙𝑖𝑖 is the freely variable input labour; and

𝑘𝑘𝑖𝑖 is the state variable capital11—all these variables are logged. If 𝜀𝜀𝑖𝑖 is uncorrelated

with the regressors, the production function can be estimated using OLS. However, there is a possibility that the error term impacts on the choices of inputs that I cannot

observe, but that are observable by the firm; then, 𝜀𝜀𝑖𝑖 may be correlated with the right-

hand side. If this is true, it would lead to a simultaneity bias. Thus I follow Olley and

Pakes (1996) and Levinsohn and Petrin (2003) to decompose the error 𝜀𝜀𝑖𝑖into two

components: 𝜔𝜔𝑖𝑖, the transmitted productivity component that is correlated with input

choices and 𝜂𝜂𝑖𝑖, the error term that is uncorrelated with input choices. Then, I

11 In Levinsohn and Petrin, labour is modelled as a fully flexible variable input and capital is the

predetermined variable, which shifts the mean of the production function but does not affect 𝜔𝜔𝑖𝑖.

have𝜀𝜀𝑖𝑖 =𝜔𝜔𝑖𝑖+𝜂𝜂𝑖𝑖, where 𝜔𝜔𝑖𝑖 is a state variable of productivity that affects the firm’s

decision rules.

In Levinsohn and Petrin (2003), it is assumed that demand for the intermediate

input 𝐸𝐸𝑖𝑖 depends on the firm’s state variables, capital and productivity shocks; so

𝐸𝐸𝑖𝑖 =𝐸𝐸𝑖𝑖(𝑘𝑘𝑖𝑖,𝜔𝜔𝑖𝑖). The demand function is monotonically increasing in 𝜔𝜔𝑖𝑖, which allows

an inversion of the intermediate demand function. I obtain 𝜔𝜔𝑖𝑖= 𝜔𝜔𝑖𝑖(𝑘𝑘𝑖𝑖,𝐸𝐸𝑖𝑖), in which

the unobservable productivity term is now expressed solely as a function of two observed inputs. The production function can now be written as:

𝑦𝑦𝑖𝑖 = 𝛽𝛽𝑙𝑙𝑙𝑙𝑖𝑖 +𝜙𝜙(𝑘𝑘𝑖𝑖,𝜔𝜔𝑖𝑖) +𝜂𝜂𝑖𝑖, (2.7)

where 𝜙𝜙(𝑘𝑘𝑖𝑖,𝜔𝜔𝑖𝑖) =𝛽𝛽0+𝛽𝛽𝑖𝑖𝑘𝑘𝑖𝑖+𝜔𝜔𝑖𝑖(𝑘𝑘𝑖𝑖,𝐸𝐸𝑖𝑖). This equation can be estimated using the

procedures discussed in Petrin, Poi and Levinsohn (2004). This study uses the consumption of material inputs and electricity as the intermediate input.

In Levinsohn and Petrin’s estimation, there are two options of the dependent variable: value added and output (gross) revenue. In the main model, TFP is constructed using the value-added estimation from Levinsohn and Petrin with material and electricity used as proxies for the intermediate input. For completeness and comparison, however, the results using OLS, Olley and Pakes, and Levinsohn and Petrin

-revenue are also reported.12 Table 2A.5 in Appendix 2 compares the parameter

estimates from OLS, Olley and Pakes’s model, Levinsohn and Petrin-revenue and Levinsohn and Petrin-value-added to construct the TFP.

12In the robustness check, I also apply the other estimations of the TFP.

Figure 2.3. TFP by sector groups

Source. Results from TFP estimations using the Levinsohn–Petrin value-added model

Figure 2.3 shows the average values of the constructed TFP for all observations and five different industry groups. On average the productivity of firms in the ELE is the highest among other sectors and that those in the ULI have the lowest productivity. It can be noted that the Indonesian economy is biased towards ULI sectors, so the number of observations of firms for this group is the largest among others, and this contributes to the overall low average TFP (for all firms).

6 6.5 7 7.5 8 8.5 9 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 Ln (T F P)

Figure 2.4. TFP: Exporters vs. non-exporters

Source. Results from TFP estimations using the Levinsohn–Petrin value-added model

Figure 2.4 compares the productivity level between exporting firms and non- exporting firms in Indonesia. As expected, it shows that, on average, exporters’ productivity levels are much larger than the firms that serve only the domestic market. Further, to reflect our focus on the export age, I classify firms based on three categories. They are: ‘incumbents’—firms that have already been exporting since the beginning of the observation period; ‘starters’ or new exporting firms—firms that started exporting in 2001–2012; and ‘never export’—firms that did not export during the observation period. Figure 2.5 presents the average productivity level of these three groups. Exporting firms that are relatively older have higher productivity levels than new ones. Meanwhile, firms that have never exported have much lower productivity levels than the other groups.

6 6.5 7 7.5 8 8.5 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 Ln (T F P ) Exporters Non-exporters

Figure 2.5. TFP by firms’ export classifications

Source. Results from TFP estimations using the Levinsohn–Petrin value-added model