2. TOTAL FACTOR PRODUCTIVITY
2.4. A R EVIEW OF THE D ETERMINANTS OF TFP
2.4.13. The Contribution of this Study
The study conducted for this thesis belongs to the literature analysing TFP and its determinants in China at the firm level. Previous studies in this area differ from one another in terms of their aims, estimation methodologies, datasets, determinants and the time periods considered. There are four important studies that analyse Chinese TFP at the firm level (Yao et al., 2007; Li et al., 2010; Brandt et al., 2012; Shen and Song, 2013). The study conducted for this thesis differs from them in four respects.
The first distinction is the use of a more comprehensive set of determinants of TFP in the estimation, providing a better and broader understanding of the potential determinants of total factor productivity in China. Such determinants are included in the estimation of TFP because their omission would produce biased estimates of the production function, and hence biased estimates of TFP. The choice of determinants is also motivated by the empirical results from the literature and the information available in the Chinese National Bureau of Statistics (NBS) dataset from which the sample used in this study has been sourced. Thus, although the previous studies have provided valuable insights, the current study has extended the set of TFP determinants studied to include the following: political affiliation, ownership structure, engagement in exporting, extent of competition, Marshallian (or MAR) spillovers, Jacobian
(or Jacob) spillovers, city spillovers, liquidity, age, R&D expenditure, time trend, and marketing capabilities.
The second major distinction from the studies reviewed above is the analysis of a wider set of 26 industries. In this study, the sample is taken from the yearly accounting reports filed by industrial firms to the Chinese National Bureau of Statistics. Such a sample considers all industrial medium- and large-sized firms, both State-owned and non-State-owned, having annual sales above RMB 5mn. These belong to the entire manufacturing and mining sectors and are located in 31 provinces or municipalities. The estimation of TFP determinants across a wide range of industries allows the accounting for differences in technology, thus avoiding the assumption that firms operate using a standard technology shared across all industries. While most previous studies have used the Olley and Pakes (1996) or Levinsohn and Petrin (2003) methodologies to analyse the determinants of TFP at the firm-level in China, this study adopts SYS-GMM, an approach developed by Arellano and Bond (1991) and Blundell and Bond (1998), and subsequently applied in a production function by Blundell and Bond (2000). SYS-GMM is a system of estimated equations, comprising an equation in first- differences, instrumented by its lagged levels, and an equation in levels, instrumented by its lagged first-differences. The major advantage of this methodology, compared to the widely used semiparametric approaches, is the allowance for firms’ fixed effects. As previous studies have indicated that firms have unmeasured productivity advantages that remain constant over time and that need to be captured, the SYS-GMM approach enables the consideration of such fixed effects. Moreover, SYS-GMM has the advantage of addressing the endogeneity of the right-hand-side variables (including the lagged dependent variable) as well as selection bias by using lagged values of the endogenous variables as instruments in the first differences equation, and first-differences of the same variables as instruments in the levels equation (Blundell and Bond, 1998). SYS-GMM is particularly preferable to the semiparametric methodologies of Olley and Pakes (1996) and Levinsohn and Petrin (2003), as these do not allow for fixed effects and are based on strong and unintuitive assumptions, which generate collinearity problems in the first stage of estimation (Ackerberg et al., 2006). Van Biesebroeck (2007) compared the sensitivity of five different productivity estimators (index numbers, data envelopment analysis, stochastic frontiers, GMM, and semi-parametric estimation) using a Monte-Carlo simulation. Although each method has its own advantages and disadvantages, the system GMM estimator was found to be the most robust technique in presence of measurement errors and technological heterogeneity.
The fourth major distinction from most of the previously mentioned studies is the decomposition of TFP growth using the approach developed by Haltiwanger (1997). These methods separate TFP growth into the contribution provided by the following: a within-firm component representing the impact of the resource reallocation within existing firms, according to their initial shares of output in their related industries; a between-firm component indicating a change in the output share of firms, weighted by the deviation of the firm’s initial productivity from the initial industry index; a covariance component, measuring whether a firm’s increasing productivity corresponds to an increasing market share; an entering component indicating the contribution of entrant firms to their related industry’s TFP growth, measured with respect to the initial industry index; an exiting component indicating the contribution of exiting firms to their related industry’s TFP, measured with respect to the initial industry index. In order to gain an additional understanding into the determinants of TFP growth, this decomposition is also performed at the industry, province and political affiliation/ownership levels. Since Melitz and Polanec (2012) have found this decomposition to be characterized by biases, their approach is also adopted in order to understand which set of results is the most appropriate.
In summary, most of the existing studies on this topic do not use multiple covariates in their models to explain what determines TFP in China, do not include firm-level fixed effects, do not cover the broad range of industries studied in the present paper, and do not decompose TFP growth. Therefore, this study builds on the existing literature by taking these four issues into account, thus distinguishing this study from previous studies on firm-level TFP estimation in China and contributing to the literature in that way. Overall, this study aims to understand what has determined TFP levels and growth rates across Chinese firms during the period of 1998-2007, and how total TFP growth has differed across firms belonging to different industries, based in different provinces, and characterised by different combinations of ownership structures and political affiliations. The results can be used to infer potential microeconomic productivity-enhancing reforms targeting the most relevant determinants of TFP.