*3.3. Variable Measurement*

**Dependent Variables.** To measure firm innovation, we followed common practices in the innovation literature and constructed the metrics, e.g., in [11,13,14,16]. First, we used the patent-based measure to capture the corporate innovation output, that is, the natural logarithm of one plus the number of patents that the firm has filed (*LnPatents*). More precisely, this variable only counts the annual number of filed patent applications that were eventually granted. As China's State Intellectual Property Office (SIPO) does not require the citation of all related patents when applying for a patent, Chinese patent filing has a well-known problem, that is, a lack of patent citation information [27,34,43], Accordingly, we used the natural logarithm of one plus the number of invention patents (*LnInventions*), which are more innovative and original, to reflect the quality of the innovation output.

Following the practices used in previous innovation research [27,34,43], we combined the number of invention and utility patents as the firm's innovation output. We collected the number of patents for all public firms from the CSMAR database. In total, 108,969 patents were filed in the sample period including 48,663 invention patents and 60,306 utility patents.

**Control Variables**. In our regression model, we referred to prior research to control for a vector of firm-level characteristics that may influence innovation [11,16]. All control variables were collected from the CSMAR database. Specifically, we controlled for*firm size*, *firm age*, *return on assets (ROA)*, *leverage*, *cash holdings*, and *R&D intensity*. Firm size is the natural logarithm of total assets. Firm age is the natural logarithm of the number of years since the firm's inception. ROA is the ratio of net income to total assets. The leverage ratio is the total liabilities divided by the total assets. Cash holdings is the ratio of cash to total assets. R&D intensity is calculated as the R&D investment, which is collected from the annual report of each firm, scaled by the total assets. To control for outliers, we winsorize all continuous control variables at the 1st and 99th percentiles of their empirical distributions, i.e., data above (below) the 99th (1st) percentile are set to the 99th (1st) percentile. The definitions of all variables used in this paper are presented in Table A1.
