Tax credits and total factor productivity: firm-level evidence from Taiwan

Abstract

We studied the effect of tax policy on the productivity of Taiwanese manufacturing firms from 2001 to 2008. First, we investigated the determinants of firm decisions on tax credit applications. By employing the dynamic random effect probit model, this study demonstrated that past tax credit use has exerted a causal effect on current tax credit use across various industries. Specifically, there has been a persistence use of tax credit among manufacturing firms in Taiwan. Second, we employed the panel instrumental variable approach to control for tax credit endogeneity and firm heterogeneity in determining firm productivity. The results indicated that firms’ tax credit use produced a positive and statistically significant effect on productivity. Finally, after categorising the firms as either electronics or non-electronics businesses, we observed that the marginal effect of tax credits has been associated with the productivity of electronics firms. This study demonstrated that tax credit use among Taiwanese firms has enhanced their productivity. This indicates that effective tax policy can promote national economic growth.

Appendix: Construction of total factor productivity

To determine whether tax policy and R&D activities are crucial to the productivity of Taiwan’s manufacturing industries, identifying an appropriate measure of productivity is necessary. The measure of productivity used in this paper is a semi-parametric approach that was initially recommended by Olley and Pakes (1996) and subsequently modified by Levinsohn and Petrin (2003). The primary advantage of the Levinsohn and Petrin (2003) method is that it controls for the endogeneity of input selection. Following the Levinsohn and Petrin (2003) technique, we estimated a log-linear transformation of the Cobb-Douglas production function, expressed as

$$ REV_{i,t} = \beta_{0} + \beta_{1} K_{i,t} + \beta_{2} SKL_{i,t} + \beta_{3} NSKL_{i,t} + \beta_{4} ELEC_{i,t} + \varepsilon_{i,t} $$

(4)

where REV _i,t is the logarithm of the revenue of firm i at time t; K _i,t is the logarithm of the firm’s fixed assets; SKL _i,t and NSKL _i,t are the logarithms of skilled labour and non-skilled labour, respectively; and ELEC _i,t is the logarithm of electricity.

According to the Levinsohn and Petrin (2003) method, we considered the problem of endogeneity between ɛ _i,t and the regressors in Eq. (4) by determining ɛ _i,t  = ω _i,t  + η _i,t , where ω _i,t is the transmitted component; hence, this influences the firm’s decision rules, whereas η _i,t is an i.i.d. component that does not affect the firm’s decisions. Therefore, Eq. (4) can be rewritten as

$$ REV_{i,t} = \beta_{0} + \beta_{1} K_{i,t} + \beta_{2} SKL_{i,t} + \beta_{3} NSKL_{i,t} + \beta_{4} ELEC_{i,t} + \omega_{i,t} + \eta_{i,t} $$

(5)

The demand function of intermediate input (ELEC) is given as

$$ ELEC_{i,t} = ELEC_{i,t} (K_{i,t} ,\omega_{i,t} ) $$

(6)

Levinsohn and Petrin (2003) demonstrated that this relationship is monotonically increasing in ω _i,t ; therefore, the intermediate input function can be inverted to obtain ω _i,t  = ω _i,t (K _i,t , ELEC _i,t ). The unobservable productivity term is then expressed solely as a function of the two observed inputs. Thus, the equation for the productivity function is expressed as

$$ REV_{i,t} = \beta_{2} SKL_{i,t} + \beta_{3} NSKL_{i,t} + \varphi (K_{i,t} ,ELEC_{i,t} ) + \eta_{i,t} $$

(7)

where $\varphi (K_{i,t} ,ELEC_{i,t} ) = \beta_{0} + \beta_{1} K_{i,t} + \omega_{i,t} (K_{i,t} ,ELEC_{i,t} )$.

Equation (7) can be estimated using the procedures discussed by Petrin et al. (2004). Therefore, the total factor productivity is defined as

$$ TFP_{i,t} = \exp (REV_{i,t} - \hat{\beta }_{1} K_{i,t} - \hat{\beta }_{2} SKL_{i,t} - \hat{\beta }_{3} NSKL_{i,t} - \hat{\beta }_{4} ELEC_{i,t} ) . $$

(8)