3.1.1. Mean decomposition

The decomposition of output at its mean based on an estimated production function has a long history in economic analysis. It dates back to as early as the pioneer work of Solow (Reference Solow1957), who describe “an elementary way of segregating variations in output per head due to technical change from those due to changes in the availability of capital per head.” Here, we distinguish ourselves from most previous studies that assume log-linear Cobb–Douglas production technologies in terms of decomposition methodologies by considering a flexible relationship between firms’ output and inputs, and therefore allowing for greater heterogeneity.

Taking the logarithm of Equation (3.1), we have

(3.2) \begin{equation} y_{it}=g(k_{it},l_{it},m_{it}) + \omega _{it}+\eta _{it}, \end{equation}

where $g({\cdot}) \equiv \ln G({\cdot})$ , and throughout the paper, we use the lower-case letters to represent the log values of the corresponding upper-case counterparts. Based on the log production function (3.2), firm $i$ ’s expected output growth from period 1 to 2 can be written as:

(3.3) \begin{align} \triangle ^\mu _O \equiv & E\left [y_{i2}\right ]-E[y_{i1}] \nonumber \\ = & \underbrace{E \left [g(k_{i2},l_{i2},m_{i2})\right ]-E[g(k_{i1},l_{i2},m_{i2})]}_{\triangle ^\mu _k} + \underbrace{E[g(k_{i1},l_{i2},m_{i2})]-E[g(k_{i1},l_{i1},m_{i2})]}_{\triangle ^\mu _l} \nonumber \\ & +\underbrace{E[g(k_{i1},l_{i1},m_{i2})]-E[g(k_{i1},l_{i1},m_{i1})]}_{\triangle ^\mu _m}+\underbrace{E[\omega _{i2}]-E[\omega _{i1}]}_{\triangle ^\mu _\omega }. \end{align}

We use $\triangle ^\mu _O$ to represent the growth of output at its mean. We decompose it into contributions of capital, labor, materials, and productivity, denoted by $\triangle ^\mu _k$ , $\triangle ^\mu _l$ , $\triangle ^\mu _m$ , and $\triangle ^\mu _\omega$ , respectively.

To implement the above decomposition for Chinese listed manufacturing firms, we need to replace the mathematical expectation operator with its sample counterpart. During the calculation, we require firm $i$ to be observable in both periods 1 and 2. In other words, the decomposition in (3.3) is conducted based on the same set of firms in periods 1 and 2.Footnote ¹

In Equation (3.3), to isolate the contribution of each input to the output growth, we begin by examine the change in capital from periods 1 to 2, followed by the changes in labor and materials. Consider $\triangle _k^{\mu }$ as an example. It relies not only on the differences in capital values between the two periods but also on the values of labor and materials in period 2. Similarly, $\triangle _l^{\mu }$ and $\triangle _m^{\mu }$ face analogous issues. Consequently, the values of $\triangle ^\mu _k$ , $\triangle ^\mu _l$ , and $\triangle ^\mu _m$ are path-dependent. This means their magnitudes depend on the specific order of { $k$ , $l$ , $m$ } in which the decomposition is conducted. Hence, in the results section, we present the averages of all decomposition outcomes obtained from all possible permutations (3!), ensuring that the reported results are robust and path-independent.

3.1.2. Quantile decomposition

While we employ a nonparametric production function described in Equation (3.1) to account for the diverse production technologies across firms, the decomposition outlined in Equation (3.3) only centers on the mean output of firms. However, it is essential to recognize that the entire output distribution exhibits considerable variation. Firms, varying in production scale, may diverse significantly in terms of output growth and productivity innovation. To delve into this further, we next introduce our quantile decomposition method.

Conducting a quantile decomposition of output growth is more challenging than the mean decomposition, since a direct application of the conventional (conditional) quantile regression will not offer us the desired estimates that show the contributions of production inputs and productivity to the unconditional output quantiles, which show changes of all firms. The challenge here is that the law of iterated expectations does not apply in the case of conditional quantiles. See Fortin et al. (Reference Fortin, Lemieux and Firpo2011) for more discussion. To overcome this hurdle, we follow the approach of Firpo et al. (Reference Firpo, Fortin and Lemieux2009) and Huang et al. (Reference Huang, Zhao and Kumbhakar2022) by employing the RIF method.

Specifically, for the $\tau$ -th quantile of output $y_{it}$ , $q_\tau$ , we can construct its RIF as:

(3.4) \begin{equation} \text{RIF}(y_{it};\, q_\tau ) = q_\tau + \frac{\tau -1\{y_{it}\leq q_\tau \}}{f_{y_{it}}(q_\tau )}, \end{equation}

where $f_{y_{it}}(q_\tau )$ represents the density function of $y_{it}$ evaluated at $q_{\tau }$ , and $1\{\cdot \}$ is an indicator function. For any given dataset, $q_{\tau }$ and $f_{y_{it}}(q_\tau )$ can be estimated empirically. Specifically, we can estimate $q_\tau$ by minimizing the check function (Koenker and Bassett, Reference Koenker and Bassett1978): $\hat{q}_\tau =\arg \min _q\sum _{i,t} (\tau -1\{y_{it}-q \leq 0\})\cdot (y_{it}-q)$ . We can estimate $f_{y_{it}}({\cdot})$ using the local constant kernel density estimator, that is, $\hat{f}_{y_{it}}({q}_\tau )=\frac{1}{NT \cdot h} \sum _{i,t} \mathcal{K}_{y_{it}}\left (\frac{y_{it}-{q}_\tau }{h}\right )$ , where $\mathcal{K}_{y_{it}}\left (\cdot \right )$ is a kernel function, for which we use the $2$ nd order Gaussian kernel, and $h$ is the bandwidth, which can be selected by the leave-one-out cross validation method. Then, $\text{RIF}(y_{it}; q_\tau )$ can be considered as “observed.” Next, we estimate the expectation of $\text{RIF}(y_{it}; q_\tau )$ conditional on firms’ production inputs and productivity to identify their contributions to the RIF of the $\tau$ -th quantile of output:

(3.5) \begin{equation} E\left [\text{RIF}(y_{it};\, q_\tau ) \ | \ k_{it}, l_{it}, m_{it}, \omega _{it}\right ] = f(k_{it}, l_{it}, m_{it}, \omega _{it}). \end{equation}

Note that Equation (3.5) is a regression model, with the dependent variable of $\text{RIF}(y_{it}; q_\tau )$ and independent variables of $k_{it}$ , $l_{it}$ , $m_{it}$ , and $\omega _{it}$ . Similar to the estimation of the production function (3.1), we assume no specific functional forms of $f({\cdot})$ and estimate it using the series estimator based on the second-order polynomial basis.

One desirable property of the RIF is that, for any distributional statistic of a random variable $y$ denoted by $v(F_{y})$ , the expectation of its RIF is equal to itself. Applying this property and the law of iterated expectation to our situation, we have $q_\tau =E\left [ E\left [\text{RIF}(y_{it}; q_\tau ) \ | \ k_{it}, l_{it}, m_{it}, \omega _{it} \right ] \right ]$ . Then, we can decompose output growth at quantiles. Specifically, for any $\tau$ and two periods, denoted by 1 and 2, we have

(3.6) \begin{align} & \bigtriangleup ^{q_\tau }_O \equiv \ q_{\tau 2} - q_{\tau 1} \nonumber \\ = & \underbrace{E \left [f(k_{i2},l_{i2},m_{i2},\omega _{i2})\right ]-E[f(k_{i1},l_{i2},m_{i2},\omega _{i2})]}_{\triangle ^{q_{\tau }}_k} + \underbrace{E[f(k_{i1},l_{i2},m_{i2},\omega _{i2})]-E[f(k_{i1},l_{i1},m_{i2},\omega _{i2})]}_{\triangle ^{q_{\tau }}_l} \nonumber \\ & +\underbrace{E[f(k_{i1},l_{i1},m_{i2},\omega _{i2})]-E[f(k_{i1},l_{i1},m_{i1},\omega _{i2})]}_{\triangle ^{q_{\tau }}_m}+\underbrace{E[f(k_{i1},l_{i1},m_{i1},\omega _{i2})]-E[f(k_{i1},l_{i1},m_{i1},\omega _{i1})]}_{\triangle ^{q_{\tau }}_{\omega }}, \end{align}

where we use $q_{\tau 1}$ and $q_{\tau 2}$ to denote the $\tau$ -th quantile of output in periods 1 and 2, respectively. In Equation (3.6), we decompose the growth of output at quantile $\tau$ into contributions coming from capital ( $\triangle ^{q_{\tau }}_k$ ), labor ( $\triangle ^{q_{\tau }}_l$ ), materials ( $\triangle ^{q_{\tau }}_m$ ), and productivity ( $\triangle ^{q_{\tau }}_{\omega }$ ).

Similar to the mean decomposition, there are several noteworthy points to consider when implementing the quantile decomposition outlined above. First, in our latter application of Chinese listed manufacturing companies, we replace the mathematical expectation operator with its finite sample counterpart. This entails working with a balanced panel across periods 1 and 2.Footnote ² Second, the order of $(k,l,m,\omega )$ in which the decomposition is conducted matters for the estimated contribution of inputs and productivity to output growth. In total, there are twenty-four (4!) different possible permutations for each output quantile. To eliminate the path dependence nature of (3.6), we compute and report the average of decomposition results from all permutations for each quantile in the subsequent section.

3.2.1. Static O-P decomposition

The pioneer work of Olley and Pakes (Reference Olley and Pakes1996) proposes a decomposition method that breaks the industry aggregate productivity, which is defined as the output market-share-weighted average of individual productivity, into two components: an unweighted mean of firm productivity and a covariance between firm’s output share and productivity. The first component captures the technological/productivity progress within firms (when we have a balanced panel across two periods), and the second component measures whether the entire industry produces more outputs by moving resources from less productive companies to more productive ones. The specific formula of this decomposition is

(3.7) \begin{equation} \Phi _t \equiv \sum _{i}s_{it}\omega _{it}=\bar{\omega }_t + \sum _{i}(s_{it}-\bar{s_t})(\omega _{it}-\bar{\omega }_t), \end{equation}

where $\Phi _t$ is the industry aggregate productivity, $s_{it}$ is the output market share, defined as $s_{it}={Y_{it}}/{\sum _{i=1}^{N_t} Y_{it}}$ , where $N_t$ is the number of companies in period $t$ , and for any variable $a$ , $\bar a_t \equiv \frac{1}{N_t} \sum _{i=1}^{N_t} a_{it}$ .

In our subsequent empirical application involving Chinese listed manufacturing companies, we implement the aforementioned decomposition. Our objective is to identify the primary contributor to industry productivity and assess its variation across our sample period, which is characterized by varying economic growth rates.

3.2.2. Dynamic ownership decomposition

The conventional Olley and Pakes (Reference Olley and Pakes1996) decomposition is commonly regarded as a static approach, as it does not account for changes in productivity and reallocation efficiency over time or across different groups of firms. Building on the work of Melitz and Polanec (Reference Melitz and Polanec2015), we further extend our analysis to include a dynamic decomposition of industry aggregate productivity growth among various categories of listed companies.

Since the late 1970s, China has undergone significant market-oriented economic reforms and shifts in ownership structures. State-owned enterprises have encountered intensified market competition and pressure, prompting many, particularly small and less profitable ones, to undergo substantial restructuring and downsizing. This process often involves bankruptcy, sales and auctions, as well as mergers and acquisitions. Simultaneously, China has opened up to the rest of the world and transformed into one of the largest foreign investment destinations.

In acknowledge of the dynamic ownership changes and the pivotal roles played by state and foreign firms in China’s economic growth, we categorize all listed firms into three groups: foreign, state, and private. A firm is classified as a foreign firm ( $F=1$ ) in a given year if it holds foreign equity ( $FHMT\gt 0$ ), including shares from Hong Kong, Macao, and Taiwan. A firm is categorized as a state firm ( $ST=1$ ) if it does not possess foreign equity but holds state equity ( $SOE\gt 0$ ). All other firms are classified as domestic private firms ( $P=1$ ).

For any two consecutive periods 1 and 2, we can write industry aggregate productivity as:

\begin{align*} &\Phi _1 = s_{F_1} \Phi _{F_1} + s_{ST_1} \Phi _{ST_1} + s_{P_1} \Phi _{P_1}, \\ &\Phi _2 = s_{F_2} \Phi _{F_2} + s_{ST_2} \Phi _{ST_2} + s_{P_2} \Phi _{P_2}, \end{align*}

where the subscripts $F$ , $ST$ , and $P$ stand for foreign, state, and private firms, respectively; $s_{G}$ with $G \in \{F, ST, P\}$ is the total market share for firms in group $G$ , that is, $s_{G_t} \equiv \sum _{i\in G} s_{it}$ ; $\Phi _G$ is the market-share-weighted, group-specific aggregate productivity, that is, $\Phi _{G_t} \equiv \sum _{i\in G} \frac{s_{it}}{s_{G_t}} \omega _{it}$ . Then, we can define the industry aggregate productivity growth as the difference between $\Phi _2$ and $\Phi _1$ :

\begin{equation*} \bigtriangleup \Phi \equiv \Phi _2 - \Phi _1 = \sum _{G \in \{F, ST, P\}} \Big ((s_{G_2}-s_{G_1})\Phi _{G_1} + s_{G_2} (\Phi _{G_2}-\Phi _{G_1})\Big ). \end{equation*}

Moving from the first equality to the second, we use the property that $s_{G_2} \Phi _{G_2} - s_{G_1} \Phi _{G_1}=(s_{G_2}-s_{G_1})\Phi _{G_1} + s_{G_2} (\Phi _{G_2}-\Phi _{G_1})$ . To better interpret the above decomposition, we can use the aggregate productivity in the first period as a reference group and compare the group-specific productivity against it. To do that, note that $s_{F_t}+s_{ST_t}+s_{P_t}=1$ , for $t=$ 1 or 2, which suggests $\sum _{G \in \{F, ST, P\}} s_{G_2}\Phi _1 = \sum _{G \in \{F, ST, P\}} s_{G_1}\Phi _1$ . Thus, we can further write the above equation as

(3.8) \begin{equation} \bigtriangleup \Phi = \sum _{G \in \{F, ST, P\}} \Big (\underbrace{(s_{G_2}-s_{G_1})(\Phi _{G_1}-\Phi _1)}_{\text{Reallocation Efficiency}} + \underbrace{s_{G_2} (\Phi _{G_2}-\Phi _{G_1})}_{\text{Productivity Progress}}\Big ). \end{equation}

In Equation (3.8), $s_{G_2} (\Phi _{G_2} - \Phi _{G_1})$ captures the technological/productivity progress within each type of firm, and $(s_{G_2}-s_{G_1})(\Phi _{G_1}-\Phi _1)$ measures the ownership reallocation efficiency. For instance, if state firms are in general becoming more productive over time due to economic reforms, we will find $\Phi _{ST_2} \gt \Phi _{ST_1}$ . Therefore, the productivity progress term $s_{ST_2} (\Phi _{ST_2} - \Phi _{ST_1})$ will be positive. Additionally, if state firms are, on average, less productive than the industry average, that is, $\Phi _{ST_1} \lt \Phi _1$ , an efficient way to reallocate production among different types of firms suggest a decrease in state firms’ market share, which means $s_{ST_2} \lt s_{ST_1}$ . Consequently, the reallocation efficiency measure $(s_{ST_2}-s_{ST_1})(\Phi _{ST_1}-\Phi _1)$ will be positive, indicating an improvement in efficiency.

In our following empirical application, we apply the dynamic decomposition outlined above to study Chinese listed manufacturing companies, aiming at identifying the importance of state firm reforms, foreign investment, and the resilience of private enterprises in driving China’s growth over the past two decades.

5.1.1. Mean decomposition

Table 3 displays the mean output decomposition results over the years in our sample period, while Figure 2 visualizes the contributions of production factors and productivity to the output growth. Based on the differences in economic growth rates and development strategies in China from 2000–2022, we partition our entire sample period into four sub-periods. The first period spans from 2000–2007, characterized by rapid GDP growth and swift industrial expansion. During this period, Chinese listed manufacturing companies on average experience significant growth, with annual growth rates ranging from 12.15–20.74% with the mean of 17.56%. The production technologies and firm productivity of Chinese manufacturing firms in this period have been well examined by previous studies that use the comprehensive Chinese Industrial Enterprises Database (see, e.g., Brandt et al., Reference Brandt, Van Biesebroeck and Zhang2012; Malikov et al., Reference Malikov, Zhao and Kumbhakar2020; Huang et al., Reference Huang, Zhao and Kumbhakar2022; Malikov and Zhao, Reference Malikov and Zhao2023). A closely related study among them is from Huang et al. (Reference Huang, Zhao and Kumbhakar2022), who conduct similar output decomposition analysis for all medium- and large-sized manufacturing firms during this period. They find that the average annual output growth rate of Chinese manufacturing firms from 1999–2007 is 9.27%, and the contributions to output growth coming from firm productivity progress and capital, labor, materials input deepening are 3.18%, 0.38%, 0.05%, and 5.67%, respectively. As discussed in the data section, our listed companies are much larger compared with the firms examined by Huang et al. (Reference Huang, Zhao and Kumbhakar2022). Our firms grow much faster in this period. As to the contributions to the output growth, productivity progress contributes 2.52% to 6.47% with the mean of 4.09%, materials input deepening contributes 7.39% to 16.57% with the mean of 12.68%, and the contributions from other inputs are negligible.

Figure 2.

Output mean decomposition for listed manufacturing firms. Notes: Notations $\omega$ , $k$ , $l$ , and $m$ refer to the contributions from productivity, capital, labor, and materials to firms’ output growth, respectively.

The second period, from 2008–2012, encompasses the global economic crisis and its subsequent recovery. During this timeframe, there is a marked slowdown in output growth, with significant contractions in 2008 and 2009, where growth rates fall to 8.5% and 4.95%, respectively, in these two years. This is followed by a strong rebound in 2010, with output growth surging to 20.88%, before stabilizing at more moderate levels of 13.83% in 2011 and 8.23% in 2012. In 2009, firm productivity temporarily has a negative impact on output growth, likely due to adjustments made in response to the post-crisis economic situation. In the following year, the productivity growth surges to 5.49%, but then gradually declines over time. In the entire period, the average output growth rate is 11.28%, with contributions from firm productivity, capital, labor, and materials inputs being 2.44%, 0.15%, 0.56%, and 8.13%, respectively.

The third period spans from 2013 to 2017, immediately following the 18th National Congress of the Chinese Communist Party. During this five-year period, China transitions from a phase of rapid economic growth to a more moderate one. The economic objective also shifts from “growth at all costs” to “high-quality development.” Consequently, the average annual industrial output growth of Chinese listed companies varies between 8.53% to 15.85%, with an average of 11.56%, slower than the first period and approximately at the same rate as the second period. Upon examining the sources of the output growth, we find that the contribution from firm productivity progress is close to zero, except for year 2017, in which the average productivity growth is 3.05%. In other years, the output growth is almost entirely driven by materials growth. Given that productivity innovations are often considered as the fundamental drivers of long-term economic growth, the absence of evidence for significant productivity progress during this period may raise concerns, particularly for list manufacturing companies.

The last period covers from 2018–2022. It is characterized by a further slowdown in industrial output growth, driven by the compounded effects of the US and China trade conflict and China’s stringent COVID policies. Specifically, the average output growth rate of Chinese listed manufacturing firms ranges from 4.26–7.67%, with an exception in year 2021, which shows a growth rate of 14.93%. During the entire period, the average output growth rate is 7.63%. The contribution of productivity to output growth is relatively higher in the years 2018, 2021, and 2022, at 1.18%, 3.86%, and 1.90%, respectively. The increase in materials usage remains the key source of output growth.

The last column of Table 3 displays the average decomposition over the entire sample period. The dynamic changes of the Chinese economy in the past 23 years have been hidden by the average numbers in this column. Therefore, in the subsequent analysis, similar to Table 3 and Figure 2, we also provide phase-specific interpretations of our empirical results.

In sum, while China’s manufacturing sector has shown resilience in the face of various economic shocks, the diminishing role of productivity growth, especially in the more recent years, signals potential vulnerabilities. To sustain long-term economic growth, it will be crucial for China to reinvigorate productivity through innovation and more efficient resource allocation. The findings suggest that future policy efforts should focus on fostering a competitive environment that supports productivity improvements.

5.1.2. Quantile decomposition

While the above mean decomposition provides valuable insights for firms whose size aligns with the industry average in terms of production and input usage, it may offer less information for large or small firms positioned at the tails of the output distribution. Given the vast number and significant heterogeneity of Chinese manufacturing firms, we are prompted to conduct a quantile decomposition, as outlined in Section 3.1.2, to achieve a more comprehensive understanding of China’s industry expansion.

Specifically, our quantile decomposition of output growth among listed Chinese manufacturing firms focuses on the first, second (median), and third quartiles of the output distribution. They are chosen to represent the small-, median-, and large-sized firms, respectively. The corresponding results are summarized in Table 4 and visualized in Figure 3.

Table 4.

Output quartile decomposition

Notes: We report results from quantile decomposition at the 1st, median, and 3rd quartiles of the output distribution. Column “Total” refers to the average annual output growth rate of listed manufacturing firms in the corresponding period at each quartile ( $\tau$ ), and it equals the sum of the contributions from productivity progress ( $\Delta _\omega ^\tau$ ), capital ( $\Delta _k^\tau$ ), labor ( $\Delta _l^\tau$ ), and materials ( $\Delta _m^\tau$ ) deepening. Row “Average” shows the averages of the entire sample period from 2000 to 2022. Standard errors of cross-industry mean estimates are reported in parentheses.

Similar to the analysis of the mean decomposition, we divide the entire sample period into four sub-periods and report the average annual quartile decomposition for each sub-period in Table 4. Additionally, for detailed insights, we provide the decomposition results for each year in Tables A1 and A2 of the appendix.

There are a few noteworthy findings from our analysis. First, in each period, we observe a decrease in output growth rate as firm size increases, progressing from the first quartile to the median, and finally to the third quartile of the output distribution. The gap of output growth between large and small firms gets narrower over time. Second, within each quartile, we find that production growth rate peaks during the period 2000-2007, followed by 2013–2017, 2008–2012, and 2018–2022, respectively. Third, our analysis, illustrated in Figure 3, highlights significant contributions of productivity progress to output growth before the 2008 financial crisis, which are evident across firms of all sizes. Notably, larger firms exhibit higher productivity growth, which contributes more to their output growth, during the period 2000–2007. However, this trend undergoes a dramatic shift in subsequent periods. Specifically, we observe that only small firms in the first quartile continue to display significant productivity growth in certain years post-2009. For median-sized firms, the contribution of productivity progress to output growth approaches zero after 2009, while for large firms in the third quartile, this contribution even becomes negative.

Overall, our findings indicate a notable shift in output growth dynamics for median and large manufacturing firms over time. Specifically, we observe that their growth is primarily fueled by increases in materials usage rather than advancements in firm productivity in recent years. This shift underscores the need for policies focusing on fostering innovation and creativity, particularly among larger firms, to sustain long-term industrial growth. This dynamic evolution of productivity and input contributions to output growth and the heterogeneity among firms of different sizes constitute a novel contribution to the literature. Previous studies, such as Huang et al. (Reference Huang, Zhao and Kumbhakar2022), which rely on Chinese industrial survey data, have primarily focused on the period before the 2008 financial crisis. Our empirical results demonstrate distinct growth patterns among Chinese manufacturing firms before and after the crisis.

Figure 3.

Output quartile decomposition. Notes: $\omega$ , $k$ , $l$ , and $m$ refer to the contributions from productivity, capital, labor, and materials to output growth, respectively.

5.2.1. O-P decomposition

In this section, we present productivity decomposition for listed manufacturing firms in China over our sample period. We begin our analysis with a simple static (O-P, Olley and Pakes, Reference Olley and Pakes1996) decomposition following Equation (3.7), with summarized results presented in Table 5. For easier interpretation, we normalize the estimated $\{\omega _{it}\}$ series by subtracting an industry-specific constant, ensuring its average in the first year of our sample period (2000) is zero. After this normalization, the values reported in columns “Progress” and “Aggregate” of the table can be interpreted as cumulative productivity growth since 2000. Additionally, in Table 5, we report the last-year value for each of the four sub-periods discussed earlier. For the detailed year-by-year decomposition results, please refer to Table A3 in the appendix.

Table 5.

Industry productivity decomposition

Notes: Industry aggregate productivity decomposition follows Equation (3.7). Column “Reallocation” refers to the covariance between firms’ output share and productivity, column “Progress” refers to the unweighted mean of firm productivity, and column “Aggregate” refers to the industry aggregate productivity, which is the sum of the previous two columns. Standard errors of cross-industry mean estimates are reported in parentheses.

We have the following two key findings from Table 5. First, similar to the pattern of output growth, Chinese listed firms exhibit their fastest productivity growth during the period 2000-2007, with cumulative industry aggregate productivity growth reaching 27.23% over 8 years, translating to an average annual growth rate of 3.40%. In the subsequent 5-year period (2008–2012), encompassing the financial crisis, productivity growth slows considerably, yielding a cumulative industry aggregate productivity growth of 7.98% (35.21–27.23%), with an annual growth rate of 1.60%. Following this, we observe a further deceleration in productivity growth in later years: cumulative industry aggregate productivity growth stands at 2.86% (38.07–35.21%) during 2013–2017 and 3.77% (41.84–38.07%) during 2018–2022, corresponding to annual growth rates of 0.572% and 0.754%, respectively. These results align with the output decomposition results in Tables 3 and 4, which shows a diminishing contribution of productivity progress to overall output growth over time. However, also note that Chinese listed firms exhibit significant variation in productivity growth across individual years, even within each sub-period. We can see the unsmooth pattern from the detailed year-by-year decomposition results in Table A3 of the appendix.

Second, our analysis reveals that across all sub-periods, industry aggregate productivity growth is predominantly driven by the productivity progress term, while the contribution from output reallocation efficiency remains limited. Remember that the productivity progress is defined as the unweighted average of firm productivity, and the reallocation efficiency measures the covariance between firms’ output share and productivity. This result suggests that, despite a notable overall increase in productivity by 41.84% over the past 23 years, the market shares of the most productive firms have not shown significant improvement. This phenomenon can be attributed to two main factors. On the one hand, the continuous influx of new firms into the markets each year intensifies competition and may dilute the market shares of all existing firms. On the other hand, although the entire Chinese manufacturing sector has experienced rapid expansion, the variation in output growth among firms of different productivity levels may not be substantial. Consequently, the overall improvement in reallocation efficiency, as measured by changes in market share, remains limited.

5.2.2. Ownership decomposition

The Olley and Pakes (Reference Olley and Pakes1996) decomposition presented above focuses on the cumulative productivity growth of listed Chinese manufacturing firms over our sample period. However, it offers only indirect evidence of annual productivity growth and does not illustrate the effects of firms’ ownership on productivity. Among state-invested, foreign-invested, and domestic firms, which one has the largest contribution to China’s productivity advancement? Do foreign firms, believed to bring advanced production technologies and better management skills to China, grow faster than domestic firms in alien lands? Do state firms, conventionally regarded as less efficient, slow down China’s growth? To answer these and related questions, we next delve into a more detailed examination of industry aggregate productivity growth decomposition by ownership, following Equation (3.8).

The results are summarized in Table 6 and visualized in Figure 4. In Table 6, we report the average contribution of output reallocation efficiency improvement and productivity progress to the annual industry productivity growth for each type of firms and in each of the four sub-periods. Further sights into the year-by-year decomposition can be found in Table A4 of the appendix. Additionally, Figure 4 provides the year-by-year evidence using bar plots.

Table 6.

Productivity decomposition by ownership

Notes: Industry aggregate productivity decomposition follows Equation (3.8). Column “Reallocation” refers to output reallocation efficiency improvement, column “Progress” refers to unweighted average productivity changes, and column “Sum” refers to the sum of the previous two. Standard errors of cross-industry mean estimates are reported in parentheses.

Figure 4.

Productivity decomposition - state, foreign, private. Notes: State/Foreign/Private Reallocation and State/Foreign/Private Progress refer to the ownership-specific output reallocation efficiency improvement and unweighted average productivity changes, respectively.

The productivity dynamics within China’s manufacturing sector, as highlighted by our analysis, underscore some key trends in the relative contributions of various firm types (state-invested, foreign-invested, and domestic private firms) across distinct phases of economic development. First, consistent with the findings of the above O-P decomposition, the contribution of output reallocation to industry productivity growth is close to zero for all three types of firms and in all periods. This suggests that, compared with their less productive counterparts, the more productive state-invested, foreign-invested, and domestic private firms did not necessarily increase their market shares in China during our sample period.

Second, we observe that in the first period from 2000 to 2007, the average annual productivity growth is 4.09%, and the largest contribution comes from the productivity progress of state-invested firms. This may reflect the effects of the continuation of economic reforms among state firms in China at the beginning of the twenty-first century. During this period, smaller and less productive state firms were privatized, merged, or replaced by more productive private firms. The remaining state firms were larger, crucial for China’s economy, and more productive in general. The faster growth rate of state firms compared with other types of firms in this period has also been documented by previous studies, such as Hsieh and Song (Reference Hsieh and Song2015) and Zhao et al. (Reference Zhao, Qian and Kumbhakar2020). For foreign-invested and domestic private firms, they have similar contributions to industry productivity growth in this period.

Third, in the second period from 2008 to 2012, the importance of state-invested firms to industry productivity growth gradually declines, while the productivity progress of domestic private firms becomes the main contributor. In this period, the average annual industry productivity growth rate remains relatively high at 2.43%. However, there is a significant efficiency loss immediately following the 2008 financial crisis, attributed to a productivity decrease among state-invested firms and a deterioration in reallocation efficiency among private firms. This likely reflects market adjustments toward a new equilibrium after the financial crisis.

Fourth, in the third period spanning 2013–2017, Figure 4 shows two noticeable spikes in industry aggregate productivity growth in 2014 and 2017, which are attributed respectively to the productivity progress of the state-invested and domestic private firms. In other years of this period, industry aggregate productivity growth is low. In the final period from 2018–2022, the productivity progress of domestic private firms continues to be the only significant driver of industry productivity growth. During the 2019–2020 pandemic years, with stringent controls, the industry productivity growth is close to zero.

Overall, these findings highlight the evolution of productivity growth within China’s manufacturing sector, with different firm types taking on varying roles across different economic phases. The limited role of reallocation efficiency, coupled with the declining productivity contributions from state-invested firms and the increasing importance of domestic private firms, reflects broader structural changes in the Chinese economy over the past two decades. These dynamics suggest a need for policy measures that enhance market mechanisms, ensuring that more productive firms can expand and drive industry-wide productivity improvements.