**2. Method and Data**

## *2.1. Model Design*

This paper used the two-way fixed effect model for estimation, which can make the estimation result control some individual heterogeneity that will not change over time and is difficult to observe as well as reduce the problem of missing variables. The following is the benchmark model of this paper:

$$
tau\_{-it} = \beta\_0 + \beta\_1 sub\_{it} + \beta\_2 X\_{it} + firm\_i + year\_t + \varepsilon\_{it} \tag{1}$$

$$
tau\_{-q\_{it}} = \beta\_0 + \beta\_1 \underline{sub}\_{it} + \beta\_2 X\_{it} + \underline{if} m\_i + \underline{year}\_t + \varepsilon\_{it} \tag{2}$$

where *innov*\_*n* represents the quantity of technological innovation including *tpatent* (the number of total patents); *ipatent* (invention patents); and *upatent* (utility model patents). *innov*\_*q* represents the quality of technological innovation including *width* (patent quality); *iwidth* (invention patent quality); and *uwidth* (utility model patent quality). *sub* represents the government subsidy; *X* represents a series of control variables including capital structure (*lev*), profitability (*roa*), enterprise size (*size*), proportion of fixed assets (*ppe*), proportion of independent directors (*dir*), enterprise age (*age*), enterprise growth ability (*gov*), and enterprise human capital (*hc*). *firm* is the enterprise fixed effect; *year* is the year fixed effect; *ε* is the random disturbance term.
