## _wp16249 - Conclusion ## Source details **Canonical URL:** [_wp16249 - Conclusion](https://www.imf.org/-/media/websites/imf/imported-full-text-pdf/external/pubs/ft/wp/2016/_wp16249.pdf) ## Other formats - [Markdown version](/-/media/websites/imf/imported-full-text-pdf/external/pubs/ft/wp/2016/_wp16249.pdf.md) - [Structured JSON version](/-/media/websites/imf/imported-full-text-pdf/external/pubs/ft/wp/2016/_wp16249.pdf.json) --- ### Innovation trends and aggregate context - R&D spending increased by 22 percent per year during 1998-2013, reaching 190 billion USD in 2013. - Domestic applications for invention patents rose from 36 thousand in 1998 to 825 thousand in 2013. - China surpassed Japan and the United States in invention-patent applications in 2011. - Debate: surge in patent filings may reflect real technological progress or be driven by government-set targets and weak IPR; quality and real impact of many patents are questioned. ### Data and methodological approach - Unified dataset linking SIPO patent assignees with China’s Surveys of Industrial Enterprises (SIE) using firm name, address and phone number. - Coverage: more than 260 thousand firms and almost 1.5 million firm-year observations from 1998 to 2007; final merged sample contains 263,111 firms for 1998-2007; matched production-patent data contain 198,414 firm-year observations from 1998 to 2007. - Empirical approaches: negative binomial models for patent counts, within-firm fixed-effects estimations of elasticities, event-study (before-and-after) analysis, and difference-in-difference analysis with Propensity-Score Matching (PSM). ### Distribution of patenting activity across firms and industries - Only 9.08% of all firms in the merged data applied for patent. - Patenting firms account for 37.86% of value added, 42.13% of capital stock and 27.39% of employment. - Among patenting firms, 5.77% engage in innovation in multiple areas and account for 90.82% of overall patents. - Patenting firms tend to be larger, younger (mean age 9.53 vs. 14.44 for nonpatenting), have higher capital-to-labor ratios, and higher share of new products in revenue, but not necessarily higher unconditional productivity. - Industry heterogeneity: Computers industry has about 20 times the patents per firm of the least innovative industry (processing of foods); Medical industry has the highest fraction of firms engaging in patenting at 36%. ### Determinants of patenting - Positive predictors: younger firms, higher R&D investment (lnR&D_{t−1} coefficient 0.054***), larger firms (lnsales_{t−1} coefficient 0.375***), exporters (D_EX_t coefficient 0.239***). - Negative predictors: firm age (age_{t−1} coefficient −0.004***), SOE status (D_SOE_t coefficient −0.068***), new-to-patenting dummy (D_New_t coefficient −0.214*** in cross-section). - Financial constraints (leverage ratio) have a negative effect on firm patenting decisions in some specifications; when financial constraint is relaxed, firms patent less in certain regressions. - R&D expenditure only available for years 2001, 2005, 2006; cannot construct firm-level R&D stock for panel regressions. ### Impact of patent stock and patent scope on firm production performance (within-firm elasticities) - Size elasticities (lnS coefficient, overall panel): - Output: 0.155*** [0.006], R^2 0.93. - Value added: 0.152*** [0.008], R^2 0.87. - Capital stock: 0.142*** [0.007], R^2 0.95. - Employment: 0.137*** [0.006], R^2 0.93. - Productivity elasticities: - Labor productivity: 0.018*** [0.005], R^2 0.85. - Solow Residual: 0.014* [0.007], R^2 0.78. - OLS firm fixed-effects TFP: 0.051*** [0.007], R^2 0.89. - ACF TFP (Translog): 0.028*** [0.006], R^2 0.94. - ACF TFP (Cobb-Douglas): 0.020*** [0.005], R^2 0.91. - New product revenue share: 1.946*** [0.248], R^2 0.73. - Export shipments: 0.402*** [0.034], R^2 0.85. - No observed within-firm changes in factor intensity or markup associated with patent stock increases (markup coefficient 0.044 [0.113], R^2 0.17). - Patent scope (N) increases have stronger associations with real outcomes than patent stock increases; results robust in balanced panel of surviving firms where estimated elasticities are even larger. - Results hold across patent types; elasticities often highest for utility model patents. ### First-time patenting effects (event study and difference-in-difference) - Before-and-After (Switch_it) immediate impacts (φ; Before-and-After specification): - Output: 0.664*** [0.016], R^2 0.87. - Value added: 0.669*** [0.016], R^2 0.80. - Capital stock: 0.463*** [0.018], R^2 0.91. - Employment: 0.216*** [0.013], R^2 0.87. - Labor productivity: 0.448*** [0.014], R^2 0.76. - TFP ACF,Translog: 0.397*** [0.016], R^2 0.85. - New product (share): 3.257*** [0.301], R^2 0.61. - Export shipment: 0.961*** [0.066], R^2 0.78. - Markup: −0.617** [0.241], R^2 0.12. - Difference-in-Difference (matched peers) φ (relative to nonpatenting peers): - Output: 0.176*** [0.011], R^2 0.90. - Value added: 0.170*** [0.014], R^2 0.83. - Capital stock: 0.148*** [0.012], R^2 0.93. - Employment: 0.148*** [0.010], R^2 0.89. - Labor productivity: 0.028*** [0.010], R^2 0.82. - TFP ACF,Translog: 0.043*** [0.009], R^2 0.94. - New product (share): 1.673*** [0.333], R^2 0.63. - Export shipment: 0.490*** [0.061], R^2 0.82. - Markup: −0.501* [0.304], R^2 0.17. - Event-study with lags (φ0 immediate year, φ1 one year after, φ2 two years after, φ3 three years after) shows: - Immediate positive impacts concentrated in year of patenting (φ0) and quickly dissipate; many coefficients become insignificant or negative in years 2–3. - Example output dynamics: φ0 0.165*** [0.010], φ1 0.029*** [0.007], φ2 −0.038*** [0.007], φ3 −0.075*** [0.011], R^2 0.85. - TFP ACF,Translog: φ0 0.034*** [0.009], φ1 0.015** [0.006], φ2 −0.005 [0.006], φ3 0.007 [0.009], R^2 0.94. ### Role of financial constraints - Interaction regressions (lnS × FC) indicate financial constraints significantly and negatively affect the productivity elasticity with respect to patent stock: less constrained firms associate a given count of new patents with higher productivity growth. - FC = Leverage ratio interactions (selected α3 coefficients): - TFP ACF,Translog α3: 0.007* [0.004], R^2 0.94. - OLS-FE α3: 0.011** [0.005], R^2 0.89. - Labor productivity α3: 0.013*** [0.004], R^2 0.85. - FC = Sales (above-median) interactions (selected α3 coefficients): - TFP ACF,Translog α3: 0.023*** [0.005], R^2 0.943. - OLS-FE α3: 0.015** [0.007], R^2 0.888. - Export shipment α3: 0.121*** [0.027], R^2 0.846. - New product (share) α3: 0.415** [0.190], R^2 0.726. ### Role of state ownership (SOEs vs POEs) and SOE reform - Patenting behavior: SOEs and POEs filed similar numbers of patents before 2001; SOEs surpassed POEs starting in 2001. - Baseline heterogeneity regressions (lnS, SOE, lnS×SOE): - lnS positive and significant for size and productivity (e.g., Output 0.166*** [0.007]). - SOE main effect often negative for productivity (e.g., Labor productivity −0.150*** [0.016]; TFP ACF,Translog −0.083*** [0.015]). - lnS×SOE interaction positive and significant for productivity: - Labor productivity λ3: 0.055*** [0.008]. - Solow Residual λ3: 0.041*** [0.010]. - TFP ACF,CD λ3: 0.041*** [0.007]. - TFP ACF,Translog λ3: 0.035*** [0.008]. - Export shipment λ3: −0.130*** [0.043]. - Robustness across samples (benchmark and balanced/constant-ownership subsamples) shows lnS×SOE remains positive and significant in many specifications (e.g., balanced panel lnS×SOE 0.051*** [0.014]). - SOE reform difference-in-difference (Postreform t ≥ 2002) key findings (TFP ACF,Translog): - Postreform coefficient: 0.258*** [0.013]. - SOEjit × Postreformt: 0.110*** [0.024]. - lnSit × SOEjit × Postreformt (φ7): sample-dependent; Full Panel 0.005 [0.012]; Balanced panel 0.029* [0.015]; Balanced & constant-ownership 0.046** [0.020]. - Interpretation: TFP rises post-reform; SOEs’ TFP increased faster than POEs after reform; evidence of improved ability of SOEs to translate patent growth into TFP growth in balanced samples. - Patent-type heterogeneity: - Elasticities of TFP with respect to patent stock are higher for SOEs particularly for utility model and design patents; elasticity not significant for invention patents in some specifications. - Potential mechanisms: - Mean leverage ratio: SOEs 0.69; POEs 0.58. - SOEs receive significantly more subsidies than POEs; gap increases over time, especially after SOE reform. - SOEs operate in sectors with higher Herfindahl index than POEs. ### Robustness and measurement - Patent types defined: invention patents (26% of applications in sample; substantive examination; 3–5 years to prosecute; 20-year term), utility model patents (novelty only; can be obtained within 12 months; 10-year term), design patents (formality exam only; 3–8 months; up to 10 years; design accounted for 43% in sample). - Productivity measures: Labor productivity; Solow residual; OLS-FE residual; ACF (Translog and Cobb-Douglas) TFP measures. - Markup estimation follows De Loecker and Warzynski (2012); no significant association between patents and markups found in main specifications. - Matching for DID uses propensity-score logit on size and age with nearest-neighbor matching; common support enforced. ### Policy-relevant implications and final synthesis - Increases in patent stock are associated with measurable firm-level improvements: output, value added, capital, employment, export performance, TFP, and revenue share of new products. - Associated productivity improvements are larger than those found in prior U.S. studies. - First-time patenting yields contemporaneous gains concentrated in the year of patenting; effects dissipate within 2–3 years on average. - Institutional features matter: financial access, subsidies, and market structure mediate the translation of innovation (patents) into productivity. SOEs, benefiting from easier credit access, larger subsidies, and lower competition, show stronger patent-TFP associations, particularly for utility model and design patents. - Empirical validation: patents and changes in TFP or new product introduction are validated as measures of Chinese innovation in empirical work, but heterogeneity in patent quality and institutions necessitates careful interpretation. *Source: https://www.imf.org/-/media/websites/imf/imported-full-text-pdf/external/pubs/ft/wp/2016/_wp16249.pdf* ### Conclusion __________________________________________________________________ 35 ### _wp16249 - Conclusion ### Innovation trends and aggregate context - R&D spending increased by 22 percent per year during 1998-2013, reaching 190 billion USD in 2013. - Domestic applications for invention patents rose from 36 thousand in 1998 to 825 thousand in 2013. - China surpassed Japan and the United States in invention-patent applications in 2011. - Debate: surge in patent filings may reflect real technological progress or be driven by government-set targets and weak IPR; quality and real impact of many patents are questioned. ### Data and methodological approach - Constructed a unified dataset linking China’s State Intellectual Property Office (SIPO) patent assignees with China’s Surveys of Industrial Enterprises (SIE) using firm name, address and phone number. - Constructed data cover more than 260 thousand firms and almost 1.5 million firm-year observations from 1998 to 2007. - Empirical approaches include negative binomial models for patent counts, within-firm fixed-effects estimations of elasticities, event-study (before-and-after) analysis, and difference-in-difference analysis with Propensity-Score Matching (PSM). ### Distribution of patenting activity across firms and industries - Only 9.08% of all firms in the merged data applied for patent. - Patenting firms account for 37.86% of value added, 42.13% of capital stock and 27.39% of employment. - Among patenting firms, 5.77% engage in innovation in multiple areas and account for 90.82% of overall patents. - Patenting firms tend to be larger, older, have higher capital-to-labor ratios, and higher share of new products in revenue, but not necessarily higher productivity. - Industry heterogeneity: Computers industry has the highest number of patents per firm (about 20 times the least innovative industry — processing of foods); Medical industry has the highest fraction of firms engaging in patenting at 36%. ### Determinants of patenting - Younger firms, firms with higher R&D investment, and larger firms patent more. - SOEs tend to file for fewer patents than POEs. - Exporting firms are more innovative than domestic-only firms. - Firms new to patenting innovate less than incumbents. - Financial constraints, proxied by firm-specific leverage ratio, have a negative effect on firm patenting decisions. ### Impact of patents on firm production performance (within-firm elasticities) - Within-firm increases in patent stock and patent scope are significantly and positively associated with changes in: - Firm size (output, value added, capital stock and employment): elasticities range from 0.13 to 0.15 log points. - Productivity measures: - Labor productivity elasticity: 0.018 log points. - Solow Residual elasticity: 0.014 log points. - OLS firm fixed-effects TFP estimate elasticity: 0.051 log points. - Ackerberg, Caves, and Frazer (2015) TFP measure elasticity: 0.028 log points. - New product revenue share: almost 2% and significant. - Export shipments: positive association. - No observed within-firm changes in factor intensity or markup associated with patent stock increases. - Patent scope increases have even stronger associations with real economic outcomes than patent stock increases. - Results hold when using a balanced panel of surviving firms; estimated elasticities are even larger in that sample. - Findings hold across patent types (invention, utility model, design), with elasticities often highest for utility model patents. ### First-time patenting effects (event study and difference-in-difference) - Before-and-after analysis: firms experience significant increases in size, capital per worker, productivities, new product share and exports, and a fall in markup after applying for a patent for the first time. - Difference-in-difference with PSM-selected controls finds similar changes coincident with first-time patenting (except for the capital-labor ratio). - The positive impact of initial patent application does not appear to be long-lasting. ### Role of financial constraints - Financial constraints significantly and negatively affect the elasticity of productivity with respect to patent stock: less financially constrained firms associate a given count of new patents with higher productivity growth. ### Role of state ownership (SOEs vs POEs) - Contrary to common perception, the positive association between patent application and performance is stronger for SOEs than for POEs in the sample. - This observation is robust to different sample selections (balanced/unbalanced panels, firms that kept or changed ownership). - When patents are divided by type (used as a proxy for quality), SOEs are particularly better at associating patents with lower innovative content with productivity growth (i.e., stronger associations for lower-quality patent types). - SOE reforms: evidence aligns with findings that reforms allowed SOEs to catch up with POEs in TFP and to better associate innovation with productivity growth. ### Relation to existing literature and implications - Findings are broadly consistent with prior work using patents as proxies for innovation (e.g., Hall, Jaffe and Trajtenberg; Balasubramanian and Sivadasan for U.S. firms), but important differences exist (e.g., higher measured productivity elasticities in China). - The results validate using patents and changes in TFP or new product introduction as measures of Chinese innovation in empirical work. - Complementary studies note potential resource misallocations: innovation productivity (elasticity of patents granted with respect to R&D input) is higher for private firms while SOEs receive more government subsidies (Wei, Xie and Zhang (2016)). - SOE reform literature: mixed evidence cited; some work attributes 20 percent of aggregate TFP growth from 1998 to 2007 to SOE reforms (Hsieh and Song (2015)); this paper contributes by highlighting reform effects on innovation-productivity linkage. *Source: https://www.imf.org/-/media/websites/imf/imported-full-text-pdf/external/pubs/ft/wp/2016/_wp16249.pdf* ### Section 6 concludes. ### Section 6 concludes. ### II. Institutional Background, Data and Measurement — China’s Patent System - China’s Patent Law first introduced in 1984; amended several times. - China’s State Intellectual Property Office (SIPO) grants three patent types: invention patents, utility model patents, and design patents. - Invention patents: - Protect technical solutions or improvements relating to products or processes. - Require “Substantive Examination” (novelty, inventiveness and industrial applicability). - Takes approximately 3 to 5 years for an application to complete prosecution. - Duration: 20 years once granted. - 26 percent of total patent applications are submitted for invention patent during the sample period. - Corresponds to the utility patent in the United State. - Utility model patents: - Subject only to novelty test and need to have practical use; lower inventiveness requirement. - Can be obtained as quickly as within 12 months after filing. - Term: 10 years from the application date. - Preferred for structural products with short product life or low technology hurdle. - Design patents: - Do not require “Substantive Examination”; only a formality examination. - Time from filing: generally varies between 3 and 8 months. - Can be granted up to 10 years. - Design applications accounted for 43 percent in the sample. - IP enforcement litigation: - In 2001, 1,597 infringement actions filed in China; by 2010, 5,700 infringement actions filed. - In 2010, United States filed 3,605 patent infringement actions (for comparison). ### II.B Data Description — Patent and Firm-level Data - Patent application data: - Source: China’s National Bureau of Statistics (NBS). - Coverage: universe of approved patents from 1985 until 2011. - Used fields: patent ID, patent title, International Patent Classification (IPC), assignee name and address. - Firm-level data: - Source: China’s Surveys of Industrial Enterprises (SIE), annual from 1998 to 2007. - Coverage: all “above-scale” firms—state-owned (SOEs) and nonstate-owned firms with annual revenue above 5 million RMB (approx. 0.7mil USD)—in industrial sector (mining, manufacturing, public utilities). - Final merged sample (after cleaning and exclusions) contains 263,111 firms for 1998-2007; matched production-patent data contain 198,414 firm-year observations from 1998 to 2007. - Firm-patent matching: - Manual matching by firm name and verified by location (“provincial proxy number”). - 29,284 firms in SIE sample applied for a patent at least once since the patent law establishment in China — labeled “patenting firms” or “patentees”. - Firms with no patents before end of sample period (2007) labeled “nonpatenting firms” or “nonpatentees”. - Representation and coverage: - In 1998 about 2% of firms in SIE sample applied for patents, accounting for 7% of total applications that year. - By 2007, percentage of firms filing patents in SIE sample increased to about 3.4%; matched patents’ share in total nation-wide patent applications rose from 7% in 1998 to 16.5% in 2007. - Two factors for low representation of SIE firms in patent data: (a) majority of patents filed by educational and research institutions not linked to firms; (b) SIE covers only “above-scale” firms. ### II.C Measurement — Innovation, Production, Ownership, Productivity, Markup - Innovation measures: - p^j_{i,t} denotes number of patents filed by firm i in category j in year t. - P_{i,t} = ∑_j p^j_{i,t} is total patents applied by firm i in year t. - Patent stock S_{i,t} = P_{i,t} + S_{i,t−1}. - Patent scope N_t defined as number of technological categories in which firm has filed to date. - Patent status equals one from the year firm filed its first patent; otherwise zero. - Distribution in 2007: among 29,284 firms with positive patent stock, 29.2 percent have only one patent; about 1.38 percent of firms applied for more than 100 patents. - Production and firm variables: - SIE provides sales, value added, export shipment, employment, capital stock, wage, total intermediate inputs, profits, interest cost, and registration date (for firm age). - Industry classification: firms mapped to 4-digit Chinese Industry Classification (CIC); harmonized classification (Brandt et al. 2012) used to ensure comparability across the 2003 revision. - Capital stock construction: - Firms report fixed capital stock at original purchase prices (book values). Use perpetual inventory method to estimate real capital stock. - Annual depreciation rate assumed: 9%. - Investment deflated by investment price deflator (Perkins and Rawski (2008)). - For initial capital stock: - If firm established after 1998: initial nominal capital stock = first reported book value in SIE. - If firm established before 1998: initial capital stock constructed using 1993 annual enterprise survey growth estimates and deflated by investment deflator in that year. - Labor share adjustment: - Sample median labor share of value added ≈ 25%; aggregate manufacturing labor share ≈ 50%. - Follow Hsieh and Klenow (2009): assume nonwage benefits are constant fraction of wage compensation; adjustment factor computed so sum of imputed benefits + wages across firms equals 50% of aggregate value added. - Ownership classification: - Use share of registered capital owned by the state, private person, collective, foreigner, legal person; and “controlling shareholder” type. - Define a state owned firm if: (1) registered capital held by state exceeds 50% or (2) “controlling shareholder” for legal person is the state. - Productivity estimation (four measures): - Labor productivity: real value added per employee. - Solow residual: lnTFP^S_{it} = lnY_{it} − α_{jt} lnL_{it} − (1−α_{jt}) lnK_{it}, where α_{jt} is industry labor share. - OLS-FE residual: residual from OLS regression with firm fixed effects of real value added on capital and employment (logs). - ACF method (Olley-Pakes / Levinsohn-Petrin / ACF 2015 extension): - Translog value added production specification: - y_{it} = β_k k_{it} + β_l l_{it} + β_{kk} k_{it}^2 + β_{ll} l_{it}^2 + β_{kl} k_{it} l_{it} + ω_{it} + ε_{it}. - Use material input m_{it} as proxy; invert m_{it} = f_t(k_{it}, l_{it}, ω_{it}) to recover ω_{it}. - Employ third-order polynomial approximation for f_t^{-1} and second-order polynomial for g(·) in Markov process ω_{it} = g(ω_{it−1}) + ξ_{it}. - Industry-specific β̂ estimated via GMM using moment conditions E(ξ_{it} k_{it}) = 0, E(ξ_{it} l_{it−1}) = 0, E(ξ_{it} k_{it}^2) = 0, E(ξ_{it} l_{it−1}^2) = 0, E(ξ_{it} k_{it} l_{it−1}) = 0. - TFP_{ACF,translog} computed as lnTFP_{ACF,translog,it} = Φ̂_{it} − β̂_k k_{it} − β̂_l l_{it} − β̂_{kk} k_{it}^2 − β̂_{ll} l_{it}^2 − β̂_{kl} k_{it} l_{it}. - Robustness: TFP_{ACF,CD} computed using Cobb-Douglas specification with analogous approximations. - Markup estimation: - Following De Loecker and Warzynski (2012): m_{it} = θ̂^l_{it} (s^l_{it})^{−1}, where s^l_{it} is labor expenditure share w_{it} L_{it} / P_{it} Q_{it}. - With translog production: θ̂^l_{it} = β̂_l + 2 β̂_{ll} l_{it} + β̂_{lk} k_{it}. ### III. Patent Stock and Firm Performance — A. Determinants of Patenting Behavior - Empirical model for patent counts: - Negative Binomial model for patent stock S_{it} with multiplicative feedback and lagged dependent variable: - S_{it} = exp(α_1 D_{it} lnS_{i,t−1} + α_2 D_{it} + β′ X_{i,t−1} + μ_i + τ_t + ε_{i,t}). - D_{it} = 1 if P_{i,t−1} > 0, else 0. - Controls X include ln(R&D), sales, age, export dummies, SOE dummies, “new to patenting” dummy (first patent within five years), leverage ratio (total liabilities / total assets). - R&D expenditure only available for years 2001, 2005, 2006; cannot construct firm-level R&D stock for panel regressions. - Key findings (Table III summary): - Cross-section (2007) results: - lnR&D_{t−1}: coefficient 0.054*** (s.e. [0.002]). - lnsales_{t−1}: coefficient 0.375*** (s.e. [0.006]). - age_{t−1}: coefficient −0.004*** (s.e. [0.001]). - D_EX_t (exporter): 0.239*** (s.e. [0.017]). - D_New_t: −0.214*** (s.e. [0.021]). - D_SOE_t: −0.068*** (s.e. [0.025]). - Panel (1998–2007) within-firm estimates (columns including lagged S_{i,t−1}): - Sales and exporting status retain large, significant positive coefficients. - Firm age negatively associated with patenting. - Once previous patent stock controlled, firms new to innovation are more innovative (positive coefficient). - Financial constraint (leverage) not a robust predictor; when financial constraint is relaxed, firms patent less. - Notes: - Negative Binomial used because variance > mean. - Pre-sample mean scaling approach used to control for firm fixed effects in some regressions. ### III.B Patenting Firms vs. Nonpatenting Firms — Descriptive Statistics (Table IV) - Group means (Patenting firms vs. Nonpatenting firms) — selected variables (means): - Output: Nonpatenting firms mean 265,634 (std dev 1,694,065); Patenting firms mean 66,909 (std dev 452,803); Difference 198,725*. - Value added: Nonpatenting mean 74,104 (std dev 499,630); Patenting mean 17,381 (std dev 100,343); Difference 56,722*. - Capital Stock: Nonpatenting mean 108,754 (std dev 902,112); Patenting mean 20,844 (std dev 143,504); Difference 87,910*. - Employment: Nonpatenting mean 690 (std dev 2,494); Patenting mean 246 (std dev 595); Difference 444*. - Age: Nonpatenting mean 14.44 (std dev 14.95); Patenting mean 9.53 (std dev 10.09); Difference 4.91*. - Capital-labor ratio: Nonpatenting mean 114.343 (std dev 387.318); Patenting mean 85.122 (std dev 239.312); Difference 29.22*. - Labor productivity: Nonpatenting mean 364.20 (std dev 657.37); Patenting mean 361.16 (std dev 876.91); Difference 3.04 (not starred). - Solow Residual: Nonpatenting mean 2.12 (std dev 1.14); Patenting mean 2.22 (std dev 1.12); Difference −0.1*. - TFP_ACF,CD: Nonpatenting mean 3.40 (std dev 1.04); Patenting mean 3.58 (std dev 0.94); Difference −0.17*. - TFP_ACF,Translog: Nonpatenting mean 2.21 (std dev 1.43); Patenting mean 2.60 (std dev 1.43); Difference −0.39*. - New product (share): Nonpatenting mean 9.61 (std dev 22.82); Patenting mean 2.52 (std dev 12.73); Difference 7.09*. - Markup: Nonpatenting mean 1.5548 (std dev 8.061); Patenting mean 1.8120 (std dev 20.47); Difference −0.25*. - Export shipment: Nonpatenting mean 48,511 (std dev 739,047); Patenting mean 15,683 (std dev 239,913); Difference 328,828*. - Number of observations: 198,414 (Nonpatenting firms), 1,263,326 (Patenting firms) — (as presented in table). - Interpretation: - Patenting firms are larger (output, value added, capital, employment), younger (mean age ~9.53 vs. 14.44), higher capital-to-labor ratio, export more, and have higher share of revenue from new products. - No obvious unconditional difference in productivity levels between patenting and nonpatenting firms. ### III.B Within-firm Change in Patent Stock and Firm Performance — Fixed Effects Regressions (equations (9) and (10); Table V) - Regression specification: - lnY_{j,it} = λ lnS_{j,it} + μ_i + γ_{j,t} + ε_{j,it}, where μ_i are firm fixed effects and γ_{j,t} are industry-year fixed effects. - Alternative specification for OLS-FE productivity: lnV_{j,it} = λ_1 lnS_{j,it} + λ_2 lnK_{j,it} + λ_3 lnL_{j,it} + μ_i + γ_{j,t} + ε_{j,it}. - Key elasticities (Overall panel; coefficients for ln(S) with standard errors in brackets; R^2 shown): - Size — Output: ln(S) coefficient 0.155*** [0.006], R^2 0.93. - Size — Value added: 0.152*** [0.008], R^2 0.87. - Size — Capital stock: 0.142*** [0.007], R^2 0.95. - Size — Employment: 0.137*** [0.006], R^2 0.93. - Productivity — Labor productivity: 0.018*** [0.005], R^2 0.85. - Productivity — Solow Residual: 0.014* [0.007], R^2 0.78. - Productivity — OLS-FE: 0.051*** [0.007], R^2 0.89. - Productivity — TFP_ACF,CD: 0.020*** [0.005], R^2 0.91. - Productivity — TFP_ACF,Translog: 0.028*** [0.006], R^2 0.94. - Other — New product (share): 1.946*** [0.248], R^2 0.73. - Other — Markup: 0.044 [0.113], R^2 0.17 (insignificant). - Other — Export shipment: 0.402*** [0.034], R^2 0.85. - Number of observations: 142,717 (overall panel) and 26,310 (patent scope regressions). - Interpretation of magnitudes: - A 10% increase in patent stock implies about 1.5% increase in real output and value added, similar increases in capital and employment, and a 4% increase in export value (as reported in prose). - For productivity measures: - TFP_{ACF,translog}: a 10% increase in patent stock → 0.28% increase in TFP (0.028 coefficient). - TFP_{ACF,CD}: a 10% increase in patent stock → 0.20% increase in TFP (0.020 coefficient). - OLS-FE: a 10% increase in patent stock → 0.51% increase in TFP (0.051 coefficient). - Compared to U.S. studies: Balasubramanian and Sivadasan (2011) report OLS-FE elasticity 0.0152 (about one-third of this study’s OLS-FE estimate). - Robustness checks: - Balanced panel (firms operating throughout sample period) yields similar results; magnitudes for productivity associations even stronger. - No significant association found between changes in patent stock (or scope) and estimated markup — suggests TFP variation not driven by firm-specific price differences. ### III.C Robustness: Different Patent Types (Table VI) - Regressions separate patent stocks by type: Invention, Utility Model, Design. Results report λ coefficients (s.e.). - Patent Stock — Invention / Utility Model / Design (selected coefficients): - Output: 0.127*** [0.015] / 0.170*** [0.008] / 0.139*** [0.010]. - Value added: 0.114*** [0.017] / 0.174*** [0.010] / 0.129*** [0.012]. - Capital stock: 0.113*** [0.015] / 0.144*** [0.009] / 0.137*** [0.010]. - Employment: 0.113*** [0.011] / 0.149*** [0.008] / 0.131*** [0.011]. - OLS-FE productivity: 0.032** [0.014] / 0.066*** [0.009] / 0.031*** [0.012]. - TFP_{ACF,Translog}: 0.019* [0.010] / 0.025*** [0.008] / 0.024*** [0.009]. - New product (share): 2.165*** [0.639] / 2.566*** [0.282] / 0.943*** [0.353]. - Export shipment: 0.318*** [0.073] / 0.502*** [0.041] / 0.346*** [0.053]. - Patent Scope (N) — Invention / Utility Model / Design (selected coefficients): - Output: 0.186*** [0.020] / 0.215*** [0.012] / 0.243*** [0.014]. - OLS-FE productivity: 0.059*** [0.021] / 0.068*** [0.013] / 0.068*** [0.018]. - TFP_{ACF,Translog}: 0.037*** [0.013] / 0.034*** [0.010] / 0.055*** [0.014]. - Interpretation: - Increases in all patent types (invention, utility model, design) are positively and significantly associated with increases in firm size, productivity, revenue share of new products, and exports. - Magnitudes generally smaller for invention patents (the most stringently examined), while utility model and design patent stocks often show larger elasticities for size and export outcomes. - Patent scope increases are associated with better performance across all patent types; coefficients for productivities show no significant differences across types. ### III.D First-time Patenting Firms — Event Study Approach (setup) - Two approaches: - Before-and-after comparison (equation (11)): - Study firms who applied for patents for the first time during the sample period; firm’s pre-patenting own performance serves as control. - Switch dummy equals one in and after year firm first applied for a patent; coefficient φ estimates change associated with first-time patenting. - Difference-in-differences (DID) specification (equation (12)): - Compare post-patenting changes for first-time patenting firms against similar nonpatenting firms in same industry. - Matching: each patenting firm paired with a nonpatenting firm in same industry with similar pre-patenting age and size via Propensity-score Matching. *Source: _wp16249 - Section 6 concludes.* ### Appendix B explains this matching in more detail. Then the post-patenting changes in ### _wp16249 - Appendix B explains this matching in more detail. Then the post-patenting changes in ### Methodology: sample selection and matching - Panel spans just over 10 years. - Include only firms that have at least 3-year observations before and after switching from being a nonpatenting firm to a patenting firm. - Matching: nonpatenting firms are matched to patenting firms (Appendix B explains this matching in more detail). - "The results are largely unchanged when more than one nonpatenting firms are matched with a given patenting firm as control groups." ### Regression specifications used to evaluate changes after switching - Before-and-After specification: - Equation (11): Before-and-After: Y^j_it = φ Switch_it + μ_i + ε^j_{i,t}. - Difference-in-Difference specification: - Equation (12): Difference-in-Difference: Y^j_it = φ Switch_it + μ_i + γ_{j,t} + ε^j_{i,t}. - Definition: - Switch_it = 1 if t ≥ t_0 and t_0 ∈ (1998,2007) is the first year that firm i filed for patent. - Interpretation: - φ captures the magnitude and statistical significance of change in performance after switch (it "captures the changes that accompany the event"). ### Empirical findings (summary from Panel A, Table VII) - Panel A in Table VII summarizes estimates of φ for each outcome variable under the two specifications. - Statistical significance and direction: - Except for capital-labor ratio, φ is estimated to be positive and significant (all at 1% level) under both specifications. - Magnitude: - In the Before-and-After analysis, the increase in output and value added is in the magnitude of - [text ends in source; magnitude not provided in the supplied content] *Source: _wp16249 (excerpt provided)* ### 0.67 log points, while the increases in capital and labor input are smaller (in 0.46 and 0.22 log ### _wp16249 - 0.67 log points, while the increases in capital and labor input are smaller (in 0.46 and 0.22 log ### First-time patenting: immediate impacts (Before-and-After and Difference-in-Difference) - Size effects (Before-and-After φ): - Output: 0.664*** [0.016], R2 0.87 - Value added: 0.669*** [0.016], R2 0.80 - Capital stock: 0.463*** [0.018], R2 0.91 - Employment: 0.216*** [0.013], R2 0.87 - Size effects (Difference-in-Difference φ relative to nonpatenting peers): - Output: 0.176*** [0.011], R2 0.90 - Value added: 0.170*** [0.014], R2 0.83 - Capital stock: 0.148*** [0.012], R2 0.93 - Employment: 0.148*** [0.010], R2 0.89 - Factor intensity: - Capital-labor ratio (Before-and-After): 0.247*** [0.016], R2 0.79 - Capital-labor ratio (Difference-in-Difference): 0.000 [0.012], R2 0.83 - Productivity (Before-and-After): - Labor productivity: 0.448*** [0.014], R2 0.76 - Solow Residual: 0.367*** [0.025], R2 0.63 - OLS-FE: 0.669*** [0.016], R2 0.80 - TFP ACF,CD: 0.439*** [0.013], R2 0.82 - TFP ACF,Translog: 0.397*** [0.016], R2 0.85 - Productivity (Difference-in-Difference relative to peers): - Labor productivity: 0.028*** [0.010], R2 0.82 - Solow Residual: 0.024*** [0.013], R2 0.73 - OLS-FE: 0.170*** [0.014], R2 0.83 - TFP ACF,CD: 0.039*** [0.009], R2 0.90 - TFP ACF,Translog: 0.043*** [0.009], R2 0.94 - Other outcomes (Before-and-After): - New product (share): 3.257*** [0.301], R2 0.61 - Markup: -0.617** [0.241], R2 0.12 - Export shipment: 0.961*** [0.066], R2 0.78 - Other outcomes (Difference-in-Difference relative to peers): - New product (share): 1.673*** [0.333], R2 0.63 - Markup: -0.501* [0.304], R2 0.17 - Export shipment: 0.490*** [0.061], R2 0.82 - Event-study summary narrative: - First-time patenting is associated with increases in output, value added, capital, employment, TFP, new product share, and exports. - Markup drops significantly after the switch. ### Short-run versus long-run dynamics (Event study with lags, regression (13)) - Immediate (φ0) impacts (all positive and significant at 1% except capital-labor ratio and markup): - Output φ0: 0.165*** [0.010], φ1: 0.029*** [0.007], φ2: -0.038*** [0.007], φ3: -0.075*** [0.011], R2 0.85 - Value added φ0: 0.170*** [0.013], φ1: 0.024** [0.001], φ2: -0.057*** [0.011], φ3: -0.098*** [0.015], R2 0.77 - Capital stock φ0: 0.139*** [0.009], φ1: 0.024*** [0.008], φ2: -0.036*** [0.006], φ3: -0.115*** [0.011], R2 0.91 - Employment φ0: 0.130*** [0.009], φ1: 0.029*** [0.006], φ2: -0.008 [0.006], φ3: -0.039*** [0.009], R2 0.88 - Productivity over time: - TFP ACF,Translog φ0: 0.034*** [0.009], φ1: 0.015** [0.006], φ2: -0.005 [0.006], φ3: 0.007 [0.009], R2 0.94 - TFP ACF,CD φ0: 0.033*** [0.009], φ1: 0.010* [0.006], φ2: -0.007 [0.005], φ3: -0.003 [0.008], R2 0.88 - Other outcomes over time: - New product (share) φ0: 1.151*** [0.348], φ1: 0.440 [0.341], φ2: 0.483 [0.309], φ3: 0.667* [0.369], R2 0.58 - Export shipment φ0: 0.421*** [0.059], φ1: 0.039 [0.052], φ2: 0.024 [0.049], φ3: 0.114** [0.058], R2 0.82 - Event-study narrative: - Immediate positive impacts of first-time patenting are concentrated in the year of patenting (φ0). - Impacts quickly dissipate after the first year and often become insignificant or negative in years 2–3. ### IV. The role of financial constraint (regression (14) and Table IX) - Estimation setup: - Regression lnYjit = α1 lnSjit + α2 FCjit−1 + α3 lnSjit × FCjit−1 + μi + γj,t + εjit, with FC measured by leverage ratio or sales (dummy above median). - FC = Leverage Ratio (Panel A highlights α3 where relevant): - TFP ACF,Translog: - α1: 0.025*** [0.006], α2: -0.001 [0.006], α3: 0.007* [0.004], R2 0.94 - TFP ACF,CD: - α1: 0.016*** [0.005], α2: -0.007 [0.005], α3: 0.008** [0.003], R2 0.91 - OLS-FE: - α1: 0.046*** [0.007], α2: -0.047*** [0.008], α3: 0.011** [0.005], R2 0.89 - Other notable α3 coefficients: - Capital stock α3: 0.011*** [0.004], R2 0.96 - Labor productivity α3: 0.013*** [0.004], R2 0.85 - FC = Sales (Panel B highlights α3 where relevant): - TFP ACF,Translog: - α1: 0.004 [0.007], α2: 0.091*** [0.008], α3: 0.023*** [0.005], R2 0.943 - TFP ACF,CD: - α1: 0.002 [0.006], α2: 0.082*** [0.007], α3: 0.017*** [0.005], R2 0.913 - OLS-FE: - α1: 0.036*** [0.008], α2: 0.129*** [0.010], α3: 0.015** [0.007], R2 0.888 - Other notable α3 coefficients: - Export shipment α3: 0.121*** [0.027], R2 0.846 - New product (share) α3: 0.415** [0.190], R2 0.726 - Interpretation in results: - The interactive terms ln(S)×FC on various TFP measures are always significantly positive for both proxies, indicating less constrained firms associate patenting with higher TFP growth. ### V. The role of state ownership (SOEs vs POEs) and SOE reform - Patenting behavior and R&D: - SOEs and POEs filed similar number of patents before 2001; SOEs surpassed POEs starting from 2001. - Regression evidence (Table III referenced) suggests SOEs are less innovative than POEs once size and R&D are controlled for. - Regression specification for heterogeneity (equations (15) and (16)): - lnYjit = λ1 lnSjit + λ2 SOEjit + λ3 lnSjit × SOEjit + μi + γj,t + εjit - lnVjit = α lnKjit + β lnLjit + λ1 lnSjit + λ2 SOEjit + λ3 lnSjit × SOEjit + μi + γj,t + εjit - Key estimates (Table X): - lnS (λ1) generally positive and significant across size and productivity measures (e.g., Output 0.166*** [0.007]). - SOE main effect (λ2) often negative for productivity measures (e.g., Labor productivity -0.150*** [0.016]; TFP ACF,Translog -0.083*** [0.015]). - Interaction lnS×SOE (λ3) positive and significant for productivity measures: - Labor productivity λ3: 0.055*** [0.008] - Solow Residual λ3: 0.041*** [0.010] - TFP ACF,CD λ3: 0.041*** [0.007] - TFP ACF,Translog λ3: 0.035*** [0.008] - Exports and size: Export shipment λ3: -0.130*** [0.043] - Sample selection robustness (Table XI, Dep. Var. TFP ACF,Translog): - Benchmark lnS coefficient: 0.020*** [0.006]; SOE: -0.083*** [0.015]; lnS×SOE: 0.035*** [0.008]; Number of obs 142,717; R2 0.94 - Balanced panel lnS: 0.023* [0.014]; SOE: -0.135*** [0.026]; lnS×SOE: 0.051*** [0.014]; Number of obs 26,310; R2 0.95 - Constant ownership lnS: 0.015** [0.007]; lnS×SOE: 0.054*** [0.014]; Number of obs 121,980; R2 0.95 - Balanced and constant ownership lnS: 0.011 [0.007]; lnS×SOE: 0.089*** [0.027]; Number of obs 19,620; R2 0.95 - SOE reform: difference-in-difference evidence (equation (17), Table XII) - Specification includes Postreform (t ≥ 2002) and triple interactions. - Key coefficients (TFP ACF,Translog dependent): - lnSit: 0.142*** [0.008] (Full Panel) - SOEjit: -0.182*** [0.026] (Full Panel) - Postreformt: 0.258*** [0.013] (Full Panel) - lnSit × SOEjit: 0.036*** [0.011] (Full Panel) - lnSit × Postreformt: -0.015*** [0.006] (Full Panel) - SOEjit × Postreformt: 0.110*** [0.024] (Full Panel) - lnSit × SOEjit × Postreformt (φ7): 0.005 [0.012] (Full Panel) - Sample variations: - Balanced panel φ7: 0.029* [0.015] - Constant ownership φ7: 0.021 [0.014] - Balanced and constant ownership φ7: 0.046** [0.020] - Interpretation: - TFP is higher in the post-reform period. - SOE × Postreform positive: TFP of SOEs increased faster than POEs after SOE reform. - φ7 estimates suggest improved ability of SOEs to transform patent growth into TFP growth particularly in balanced samples. - SOE vs POE by patent quality: - Elasticity of TFP ACF,Translog with respect to patent stock is higher for SOEs particularly for utility model and design patents; elasticity not significant for invention patents. - TFP ACF,CD elasticities with respect to patent stock are significantly higher for SOEs for all three patent types. - Before-and-after ownership-switch analysis indicates switching from state-owned to private-owned is associated with lower TFP-patent elasticity. - Potential mechanisms (descriptive evidence): - Leverage ratio (mean over sample): SOEs average 0.69; POEs average 0.58. - Subsidies: SOEs receive significantly more subsidies than POEs; gap increases over time, especially after SOE reform. - Competition: SOEs (central and local) operate in sectors with higher Herfindahl index than POEs. ### VI. Conclusion (paper summary) - Data and period: merged firm balance sheet and patent application data for 1998-2007. - Main findings: - Increases in patent stock are associated with increases in firm size (output, sales, employment), export performance, firm productivity, and revenue share of new products. - Associated productivity improvements are larger than those found in prior U.S. studies. - Event studies of first-time patentees show contemporaneous performance gains concentrated in the year of patenting. - Contrary to conventional perception, SOEs—though lower in average TFP—show a stronger positive association between patent stock growth and TFP growth compared to POEs, especially after SOE reform; this is concentrated in patents with lower innovative content (utility model and design patents). - Preferential government policies (easier credit access, larger subsidies, and lower competition) for SOEs may help explain higher TFP-patent elasticities. - Policy-relevant implication: - Patenting activity in China is linked to measurable firm-level production improvements, but institutional features (financial access, subsidies, market structure) mediate the translation of innovation into productivity. *Source: IMF working paper content unit _wp16249 (PDF).* ### Appendix A. Constructing Firm-Specific Markups ### Appendix A. Constructing Firm-Specific Markups ### Production technology and assumptions - A firm i at time t produces output using the production technology: - Q_it = Q_it(K_it, L_it, ω_it) (18) - Restriction: Q_it is continuous and twice differentiable with respect to its arguments. ### Cost minimization and first-order condition - Cost-minimizing producers consider the Lagrangian: - Lag(K_it, L_it, λ_it) = r_it K_it + w_it L_it + λ_it (Q_it − Q_it(.)) (19) - First-order condition with respect to labor: - ∂Lag_it/∂L_it = w_it − λ_it ∂Q_it(.)/∂L_it = 0 (20) - The marginal cost of production at a given level of output is λ_it as ∂Lag_it/∂Q_it = λ_it. ### Labor elasticity and markup relationship - Rearranging terms and multiplying both sides by L_it/Q_it yields the labor elasticity, θ_i: - θ_i = ∂Q_it(.)/∂L_it · L_it/Q_it = 1/λ_it · w_it L_it/Q_it (21) - Define markup μ_it as price over marginal cost: - μ_it = P_it/λ_it - Using this definition, equation (21) can be rewritten: - θ_i = μ_it · w_it L_it / (P_it Q_it) (22) ### Constructing firm-level markups - Given an estimate of the labor elasticity θ_it from the production function and a measure of the share of labor costs in total sales w_it L_it / (P_it Q_it), the firm’s markup is constructed as: - μ_it = θ_it · P_it Q_it / (w_it L_it). (23) ### Appendix B. The Analysis of Propensity-Score Matching ### Objective - Ensure comparability between patenting (treatment) and nonpatenting (control) firms for difference-in-difference analysis by making treatment and control groups similar in major firm characteristics using Propensity-Score Matching (PSM). ### Treatment definition and propensity score estimation - Treatment indicator: - D_i = 1 if the firm applies for at least one patent, and zero otherwise. - Propensity score model (logit): - Pr(D_i = 1 | X) = G(size, age, industry dummy, year dummy) - where X = {size, age} and G(z) = exp(z)/(1 + exp(z)). - For firm i in the treatment group, define p_i(x) = Pr(D_i = 1 | X = x). - Common support condition: 0 < p_i(x) < 1. ### Matching procedure - Nearest-neighbor matching criterion: - For treated firm i, pick matched non-treated firm j such that: - ‖p_i − p_j‖ = min_{k ∈ {D = 0}} ‖p_i − p_k‖. *Source: Appendix A and Appendix B from the supplied IMF content unit.* --- _Source: https://www.imf.org/-/media/websites/imf/imported-full-text-pdf/external/pubs/ft/wp/2016/_wp16249.pdf_