Online Appendix Only Funding forms, market conditions and dynamic effects of government R&D subsidies: evidence from China

Transcription

1 Online Appendix Only Funding forms, market conditions and dynamic effects of government R&D subsidies: evidence from China By Di Guo a, Yan Guo b, Kun Jiang c Appendix A: TFP estimation Firm TFP is measured by employing two different methods to ensure that the conclusions of this study are not driven by a specific TFP measure. First, TFP is a straightforward OLS residual derived from the log-linear transformation of a general Cobb Douglas production function (OLS method). The OLS estimates are convenient to obtain but may be biased once unobservable shocks are correlated with input levels. At the same time, the OLS method lacks dynamic consideration. To overcome these problems, we follow the approach utilized by Olley and Pakes (1996) in the second method (OP method). With OP method, we use investment as a proxy for unobservable production shocks. This semi-parametric method is applied to control for both the simultaneity caused by unobserved productivity and the non-random sample selection induced by the differing exit probabilities for small and large low-productivity firms. A.1 Explanation of the OLS Method The OLS method is straightforward. Based on the OLS regression, TFP is estimated as the error term which denotes the effects in the total output that are not caused by tangible inputs. The equation below demonstrates how TFP is estimated in the OLS method. Y = A K α L β lny = lna + αlnk + βlnl. As the residual of the OLS regression, lna is the TFP that we intend to measure. Firm-level TFP is estimated by considering both year and the two-digit SIC code effects. The robustness of the estimation results are verified by relaxing the year effect for accurate estimation. The disadvantage of the OLS method is that it considers only tangible inputs, such as labor and capital, but not unobservable shocks. As a result, a static model is generated in which all types of inputs are assumed to be exogenous and to have no correlation with the a University of Hong Kong, Faculty of Business and Economics, Pokfulam Road, Hong Kong. guodi2007@gmail.com b Peking University, School of Economics, 5 Yiheyuan Rd, Haidian, Beijing, PR China. guoyan@pku.edu.cn c University of Roehampton, Business School, Roehampton Ln, London SW15 5PU, United Kingdom. kun.jiang@roehampton.ac.uk

2 error term (i.e., TFP). The limitation of the OLS method is obvious, and the associated coefficients are biased when the assumptions are not valid. A.2 Explanation of the OP Method In the presence of selection bias and simultaneity, OP estimation allows for the endogeneity of some of the input factors and the unobserved productivity differences among firms. Such estimation also considers the exit of firms from the market; hence, this method has several advantages over the simple OLS method. The OP method is characterized by the Bellman equation and assumes that the firm constantly maximizes the expected discounted value of future profits; thus, stay-or-quit and investment decisions are formulated in each time period. For estimation purposes, this study uses the Cobb Douglas production function; two forms, namely, gross output and value-added production functions, are adopted for robustness check. The equations below express the two production functions in the OP method. Total Output it = β 0 + β 1 L it + β 2 K it + β 3 M it + W it + ε it Value Added it = β 0 + β 1 L it + β 2 K it + W it + ε it, where Total Output/ValueAdded it is deflated by the producer price index for manufactured products (our results are robust with respect to both measures on Y), L it is the labor input by firm ii at time tt (either the number of employees or the total payment for the employees of a firm can be a proxy for this variable), K it is the capital input by firm ii at time tt and is deflated by the price index of investment in fixed assets, and M it denotes the intermediate inputs by firm ii at time tt and is deflated by the producer price index for purchasing products. The OP method connects the observable characteristics of the firm with its unobserved productivity and specifies a function to estimate the coefficients consistently. YY iiii = FF(LL iiii, KK iiii, WW iiii ). Following the Cobb Douglas production function, yy iiii = β 0 + β 1 l it + β 2 k it + w it + ε it, where yy iiii is the log value of value added for firm ii in period tt, l it is the log value of labor input by firm ii at time tt (either the number of employees or the total payment for the employees of a firm can be a proxy for this variable), k it is the log value of the capital input by firm ii at time tt, w it is the productivity known by a firm when it makes its liquidation decision and investment decision, and ε it is the true error term. Both w it and ε it are unobservable; nonetheless, w it affects a firm s input decision as a state variable in the firm s decision whereas ε it does not.

3 The OP method assumes that the expected productivity is a function of current productivity and capital, that is, [w it+1 w it, K it ]. w it is assumed to follow a first-order Markov process for estimation purposes. Given those assumptions of the model, OLS estimation is biased for two reasons: first, the capital input is correlated with productivity. When a firm observes increased productivity, it will increase investments. Second, survival bias is detected; when capital is related to productivity, larger firms are less likely to exit the market. We conduct our estimation process by following the three steps proposed by Yasar et al. (2008). In the first step, based on the assumption that investment of firm ii at time tt (II iiii ) is strictly positive, we can use the following equation to show the relationship among productivity, investment, and capital stock. ww iiii = II 1 (II iiii, KK iiii ) = h(ii iiii, KK iiii ) And the production function can be rewritten as: yy iiii = β 1 l it + (ii iiii, kk iiii ) + ε it, where (ii iiii, kk iiii ) = β 0 + β 2 k it + h(ii iiii, kk iiii ). We approximate (. ) with a second-order polynomial series in investment and capital, as suggested by Yasar et al. (2008). The partially linear equation described above can be estimated by OLS, and the estimation of ββ 1 is consistent because (. ) controls the unobserved productivity. However, we must perform a second step to control survival bias; this step can later be used to estimate β 2. The probability of survival in period tt depends on the productivity in period tt 1, which is in turn dependent on the capital and investment in period tt 1. The predicted probability of survival is denoted by PP. ıııı In the final step, we estimate β 2. yy iiii β lit 1 = β 2 k it + gg tt 1 β 2 k it 1, PP iiii + ε it, where gg(. ) is approximated by a second-order polynomial in tt 1 β 2 k it 1 and PP ıııı. According to Yasar et al. (2008), the function gg(. )is similar in spirit to the inverse of Mills ratio that is included in two-step sample selection models. In this study, all variables in the equations are in their logarithm forms; moreover, heteroskedasticities are controlled for both the time trend and the two-digit industry. A.3 Data source and limitations The firm-level data of this study are derived from the ASIFP together with associated financial information. As a prerequisite in TFP calculation, real capital stock stimulates discussion and dispute. The lack of firm-level capital stock data complicates the construction

4 of a series of real capital stocks that are comparable across time and firms. The perpetual inventory method (PIM) is applied in this study. The effective capital stock in production is measured as a weighted sum of previous fixed asset investments in constant price terms with this approach. RCS t = τ=0 d τ I t τ, where RCS t is real capital stock in t, d τ is the efficiency of fixed asset in the τth year, and I t τ is the fixed asset investment flow τ years ago. With the additional assumption that d τ declines in a geometric pattern, the PIM equation can be written as follows: d τ = (1 δ) τ RCS t = RCS t 1 + I t δrcs t 1. This study formulates fixed asset growth at the two-digit SIC code level as a recursive step back to the year when a firm was established. This study constructs the series of real capital stocks via the aforementioned PIM method, together with the series of investment deflators obtained from the China Urban Life and Price Yearbook (2009). The year 1978 is set as the starting point of the initial capital stock for series calculation, and 9% is applied as the specific fixed depreciation rate. * All the nominal values are deflated by price indices with a benchmark of 100, as set in There are a few limitations with the calculations of firm TFP caused by the constraints of the data. First of all, in the OP model, the decision-making process of a firm must be clarified, that is, whether or not a firm opts to remain in the market. However, this information is not contained in the dataset used in this study. Accordingly, the panel data are used to verify this exit variable. This research defines the exit of the firm from the market when the observation record of the company is discontinued with the unbalanced panel data in the period of 1996 to The value of the dummy variable exit is one if the firm exits from the market in the current period and is zero otherwise. Another limitation with the calculation of firm TFP is that we do not have information for R&D stock which is assumed to be an important factor affecting TFP. In particular, if Innofund-backed and non-innofund-backed firms differ systematically in terms R&D stock, the TFP we have calculated may be biased. In such case, the variations in TFP of * In this study, we follow Brandt et al. (2012) and Hsieh and Klenow (2009), who apply 9% as the depreciation rate to calculate the TFP of firms in China. The assumed depreciation rate is a chain-linked price deflator calculated by Brandt, Rawski, and Sutton (2008) based on separate price indices for equipment, machinery, and buildings-structures as well as the weights and shares of these items in fixed assets, as reported by China s National Bureau of Statistics. The deflator is a reflection of the price, and the structure of tangible assets of enterprises in China. We use this assumed depreciation rate to ensure that our study is comparable to the existing literature.

5 Innofund-backed firms and their counterparts in the control group we observe may be caused by the different R&D stock rather than Innofund support. Having recognized such potential issues, we utilize two approaches to eliminate such concerns. First, as we have discussed in the paper, we employ two-stage estimations with instrumental variables to deal with the concerns with omitted variables. Second, as we have information for R&D expenditure for years from 2005 to 2007, we utilize such information and construct a new PSM subsample by controlling R&D expenditure for the three years in PSM process. We then exercise the main estimations of Innofund effects based on this newly-built subsample. The results of the estimations based on this newly built subsample are presented in Table B-5. The results are consistent with the main results shown in Table 2 of the paper. Finally, in the present study, we use TFP to approximate firm productivity. Nonetheless, a possible problem of using TFP to measure productivity is that if the innovative firm generates little revenue as a result of the introduction of a new product, TFP may become a biased measure for productivity. We therefore check the revenues and costs of firms. So far, we find that less than 5% of the firms in our sample report annual revenue of less than 1.8 million RMB. Meanwhile, less than 3% of firms cannot break even on average, suggesting a limited chance that this should bias our results (detailed summary is provided in Table B-2 of online appendix). To further test the robustness of our approach, we delete those firms which had negative returns from both the treated and control group and build up a new subsample to see whether such un-breaking even firms are systematically located in either group and consequently affect our estimations. The estimation results based on this newly-built subsample are consistent to those of our baseline examinations based on the full sample (the estimations based on the newly-constructed subsample are presented in Table B-6 of online appendix). Such results suggest that the concern with potential biases in TFP calculation caused by the money losing firms is not a major issue for this study.

6 Appendix B: Additional summary statistics and estimations Table B- 1: Industry and Year Distributions of the Sampled Innofund-backed Firms Panel A: industry distribution Panel B: year distribution of projects Industry description SIC # # # Awarded Innofund-backed % Year Innofund-backed % projects firms (sampled) firms (sampled) (full sample) % Raw Chemical Materials and Chemical Products , Medicines General Purpose Machinery , Special Purpose Machinery Transport Equipment , Electrical Machinery and Equipment , Communication Equipment, Computers, and Other Electronic Equipment , Measuring Instruments and Machinery for Cultural Activity and Office Work , Others , Total 2, , , , , Total 2, ,

7 Table B-2: Distribution of Costs/Revenue Ratio, Intangible assets/ Total Assets Ratio and Fixed Assets/Revenue Ratio of Sampled Firms Panel A Costs, Revenue, Intangible assets, Fixed assets of firms at the time of Innofund being infused Percentiles Costs/Revenue Intangible assets/revenue (lagged) Fixed assets/revenue (lagged) Intangible assets/total assets (lagged) Revenue (thousand RMB) 1% % % % % % % % % Mean Std. Dev # of Observations

8 Panel B Costs, Revenue, Intangible assets, Fixed assets of Firms for the Entire Examination Period Percentiles Costs/Revenue Intangible assets/revenue (lagged) Fixed assets/revenue (lagged) Intangible assets/total assets (lagged) Revenue (thousand RMB) 1% % % % % % % % % Mean Std. Dev Total Obs

9 Table B-3: Balance Tests on Relevant Variables for the PSM Sample Panel A Balance tests on sales of the new products (in natural logarithm format) before- and after-psm Year Sample Treated Controls Difference T-stat P-value 1999 Before *** After Before *** After Before *** After Before *** After Before *** After Before *** After Before *** After Before *** After Panel B Balance tests on volume of exports (in natural logarithm format) before- and after-psm Year Sample Treated Controls Difference T-stat P-value 1999 Before *** After Before ** After Before ** After Before After Before After Before After Before After Before After

10 Panel C Balance tests for the stock of patents before- and after-psm Year Sample Treated Controls Difference T-stat P-value 1999 Before *** After Before *** After Before *** After Before *** After Before *** After Before *** After Before *** After Before *** After Before *** After * p<=0.1, ** p<=0.05, *** p<=0.01

11 Table B-4: Comparison of the TFP and Size of Innofund-backed Firms and PSM Samples Innofund-backed firms Matched Samples Difference T-test (p-value) Wilcoxon rank-sum test (p-value) TFP_ols ** TFP_op TFP_op Firm Size(Lg Number of employees) * p<=0.1, ** p<=0.05, *** p<=0.01

12 Table B-5: Innofund and Firm Productivity (Subsample in which R&D Expenditure is Controlled in PSM) (1) (2) (3) TFP_ols TFP_op1 TFP_op2 InnoAft 0.187*** 0.199*** 0.156*** (0.035) (0.036) (0.036) Firm_age 0.187*** 0.178*** 0.080** (0.036) (0.036) (0.036) State_Shr 0.129*** 0.141*** 0.144*** (0.048) (0.050) (0.051) Lvg_rt 0.264*** 0.220*** 0.195*** (0.043) (0.044) (0.044) Firm_size 0.065*** 0.103*** 0.104*** (0.021) (0.022) (0.021) Export_D 0.055** 0.049** 0.044* (0.022) (0.023) (0.023) FDI_D (0.039) (0.040) (0.038) Contants 0.591*** 3.019*** 2.502*** (0.113) (0.117) (0.114) Year Effect Y Y Y Y Firm Effect Y Y Y Y N adj. R-sq Standard errors in parentheses * p<=0.1, ** p<=0.05, *** p<=0.01

13 Table B-6: Innofund and Firm Productivity (Subsample excluding firms having not broken even ) Panel A: Subsample built based on randomly drawn control group (1) (3) (5) (2) (4) (6) TFP_ols TFP_op1 TFP_op2 TFP_ols TFP_op1 TFP_op2 InnoAft 0.093*** 0.087*** 0.069*** (0.018) (0.019) (0.019) Bfr_Innofund 0.142** 0.172*** 0.127** (0.063) (0.066) (0.063) Sht_Innofund 0.242*** 0.263*** 0.202*** (0.067) (0.069) (0.066) Lg_Innofund 0.150** 0.195*** 0.124* (0.071) (0.074) (0.071) Firm_age 0.148*** 0.174*** 0.068*** 0.147*** 0.173*** 0.067*** (0.020) (0.020) (0.020) (0.020) (0.020) (0.020) State_Shr *** *** *** *** *** *** (0.027) (0.028) (0.028) (0.027) (0.028) (0.028) Lvg_rt *** *** *** *** *** *** (0.025) (0.025) (0.025) (0.025) (0.025) (0.025) Firm_size *** *** *** *** *** *** (0.013) (0.013) (0.013) (0.013) (0.013) (0.013) Export_D 0.051*** 0.049*** 0.048*** 0.052*** 0.050*** 0.049*** (0.014) (0.014) (0.014) (0.014) (0.014) (0.014) FDI_D (0.025) (0.025) (0.025) (0.025) (0.026) (0.025) Constants 0.524*** 2.960*** 2.453*** 0.495*** 2.927*** 2.428*** (0.067) (0.069) (0.067) (0.068) (0.070) (0.068) Year Effect Y Y Y Y Y Y Firm Effect Y Y Y Y Y Y N adj. R-sq Sht_Innofund - Bfr_Innofund 0.100*** 0.091*** 0.075*** (0.018) (0.019) (0.019) Lg_Innofund - Sht_Innofund *** *** *** (0.025) (0.026) (0.026) Standard errors in parentheses * p<=0.1, ** p<=0.05, *** p<=0.01

14 Panel B: Subsample built based on PSM control group (1) (3) (5) (2) (4) (6) TFP_ols TFP_op1 TFP_op2 TFP_ols TFP_op1 TFP_op2 InnoAft 0.074*** 0.079*** 0.032* (0.018) (0.019) (0.019) Bfr_Innofund ** (0.064) (0.066) (0.063) Sht_Innofund 0.173*** 0.232*** 0.111* (0.067) (0.069) (0.066) Lg_Innofund ** (0.071) (0.074) (0.071) Firm_age 0.182*** 0.185*** 0.101*** 0.180*** 0.183*** 0.098*** (0.020) (0.021) (0.020) (0.020) (0.021) (0.020) State_Shr (0.027) (0.028) (0.028) (0.027) (0.028) (0.028) Lvg_rt *** *** *** *** *** *** (0.024) (0.024) (0.024) (0.024) (0.024) (0.024) Firm_size *** *** *** *** *** *** (0.012) (0.012) (0.012) (0.012) (0.012) (0.012) Export_D 0.060*** 0.058*** 0.058*** 0.060*** 0.058*** 0.059*** (0.012) (0.012) (0.012) (0.012) (0.012) (0.012) FDI_D * (0.020) (0.021) (0.021) (0.020) (0.021) (0.021) Constants 0.610*** 3.088*** 2.529*** 0.593*** 3.062*** 2.515*** (0.064) (0.065) (0.064) (0.065) (0.066) (0.065) Year Effect Y Y Y Y Y Y Firm Effect Y Y Y Y Y Y N adj. R-sq Sht_Innofund - Bfr_Innofund 0.083*** 0.084*** 0.042*** (0.018) (0.019) (0.019) Lg_Innofund - Sht_Innofund *** *** *** (0.025) (0.026) (0.026) Standard errors in parentheses * p<=0.1, ** p<=0.05, *** p<=0.01

15 References Brandt, L., Rawski, T. G. and Sutton, J. (2008). China s industrial development. China's great economic transformation, Brandt, L., Van Biesebroeck, J. and Zhang, Y. (2012). Creative Accounting or Creative Destruction? Firm-level Productivity Growth in Chinese Manufacturing. Journal of Development Economics, 97(2), Hsieh, C. T. and Klenow, P. J. (2009). Misallocation and Manufacturing TFP in China and India. The Quarterly Journal of Economics, 124(4), Olley, G. S. and Pakes, A. (1996). The Dynamics of Productivity in the Telecommunications Equipment Industry. Econometrica, 64(6), Yasar, M., Raciborski, R. and Poi, B. (2008). Production function estimation in Stata using the Olley and Pakes method. Stata Journal, 8(2), 221.