如何在缺乏产量数据的条件下计算行业成本加成率?——对DLW模型的修正与改进

How to Calculate Markups in the Absence of Output Quantity Data:Modifications and Improvements to the DLW Model

目前广泛使用的DLW模型(DeLoecker和Warzynski,2012)在估计成本加成率时,需要代入企业“产量”数据。由于企业“产量”数据目前为保密数据,较难获取,上述方法并不适用于我国数据条件。为此,一些研究采用平减后的企业“产值”数据试图代替“产量”数据。本文报告了在DLW模型中采用“产值”数据可能造成的谬误以及DLW模型自身存在的问题。这些问题主要包括:首先,在DLW模型中直接代入“产值”而不是“产量”数据,会导致产品价格外生且固定不变的错误隐含假设,使估计系数收敛到“1”而不是企业真实成本加成率。其次,现有的回归分析技术并不能识别“企业级”的技术参数,这导致企业级的成本加成率实际上无法计算。DLW模型以及随后的一系列改进模型只能得到行业技术参数与个体变量相乘的“伪企业成本加成率”,进而误导采用这一指标的相关研究。在上述工作的基础上,本文构建了新的模型,修正了约束条件和最优化方法,通过估计需求弹性而不是产出弹性避免了DLW模型的识别错误问题,使得新模型可以直接使用常见的工业企业数据库“产值”数据计算,极大地扩展了DLW模型在我国的适用性。最后,本文利用蒙特卡洛模拟验证了新模型的有效性,并利用工业企业数据库数据分行业计算了成本加成率,得到稳健的估计结果。本文模型为一系列基于成本加成率指标的研究提供了重要参考。

The traditional DLW(De Loecker and Warzynski, 2012)model requires firms? "output quantity" data when estimating markups. However, because firms? "output quantity" data is currently confidential in China, the DLW method is not applicable under limited data conditions. For this reason, previous studies have attempted to replace the "output quantity" data with the deflated firm "output value" data. We point out the potential fallacies of using "output value" data in the DLW model and the problems with the DLW model itself as follows. First, using the "output value" rather than the "output quantity" data in the DLW model results in incorrect implicit assumptions that product prices are exogenous and fixed,causing the estimated coefficients to converge to "1" rather than the true markups. Second,existing regression analysis techniques fail to identify firm-level technical parameters. Thus,improved models for the DLW model can only yield a "pseudo-firms? markup" which is calculated by multiplying industrial parameters with individual variables, and may mislead studies that use this indicator. Based on the above work, we improve the identification, modify the constraint conditions and optimization methods, and avoid the misidentification problem by estimating the elasticity of demand rather than the elasticity of output. The new method allows scholars to calculate markups using the Annual Survey of Industrial Firms(ASIF)data that contains output value data, which greatly extends the applicability of the DLW model in Chinese studies. Finally, we verify the validity and robustness of the new model using Monte Carlo simulations, and estimate industry-specific markups using the ASIF data. Our new method provides important support for a series of studies using the markup indicator.