Solar photovoltaic interventions have reduced rural poverty in China.

Since 2013, China has implemented a large-scale initiative to systematically deploy solar photovoltaic (PV) projects to alleviate poverty in rural areas. To provide new understanding of China’s targeted poverty alleviation strategy, we use a panel dataset of 211 pilot counties that received targeted PV investments from 2013 to 2016, and find that the PV poverty alleviation pilot policy increases per-capita disposable income in a county by approximately 7%-8%. The effect of PV investment is positive and significant in the year of policy implementation and the effect is more than twice as high in the subsequent two to three years. The PV poverty alleviation effect is stronger in poorer regions, particularly in Eastern China. Our results are robust to alternative specifications and variable definitions. We propose several policy recommendations to sustain progress in China’s efforts to deploy PV for poverty alleviation.


Supplementary
or not a county was selected for the photovoltaic poverty alleviation policy in a specific year. SECONDGDPR depicts a proportion of the added value of the secondary industry to GDP. PUBEXINR shows the ratio of public expenditure to revenue. LN(AGACRE) examines the land used for facility agriculture facility agriculture land. EDUCATION estimates the ratio of number of secondary school students to the total population. MKTINDEX represents marketization index. LN(SUNHOUR) indicates sunlight exposure time. LN(GDPPROVINCE) is used to investigate the per capita GDP of the province where the county is located. ***, **, and * represent the significance levels of 1%, 5% and 10%, respectively. T-statistics are reported in parentheses. or not a county was selected for the photovoltaic poverty alleviation policy in a specific year. SECONDGDPR depicts a proportion of the added value of the secondary industry to GDP. PUBEXINR shows the ratio of public expenditure to revenue. LN(AGACRE) examines the land used for facility agriculture land. EDUCATION estimates the ratio of number of secondary school students to the total population. MKTINDEX represents marketization index. LN(SUNHOUR) indicates sunlight exposure time.

Supplementary
LN(GDPPROVINCE) is used to investigate the per capita GDP of the province where the county is located. ***, **, and * represent the significance levels of 1%, 5% and 10%, respectively. T-statistics are reported in parentheses.   Chongqing, Guizhou, Yunnan and Guangxi provinces. The dependent variable is the natural logarithm of disposable income of rural people per capita. SEPAP represents whether or not a county was selected for the photovoltaic poverty alleviation policy in a specific year. SECONDGDPR depicts a proportion of the added value of the secondary industry to GDP. PUBEXINR shows the ratio of public expenditure to revenue. LN(AGACRE) examines the land used for facility agriculture facility agriculture land.

Supplementary
EDUCATION estimates the ratio of number of secondary school students to the total population. MKTINDEX represents marketization index. LN(SUNHOUR) indicates sunlight exposure time. LN(GDPPROVINCE) is used to investigate the per capita GDP of the province where the county is located. ***, **, and * represent the significance levels of 1%, 5% and 10%, respectively.
T-statistics are reported in parentheses.
14 Supplementary  LN(GDPPROVINCE) is used to investigate the per capita GDP of the province where the county is located. ***, **, and * represent the significance levels of 1%, 5% and 10%, respectively. T-statistics are reported in parentheses. Notes: The dependent variable is the natural logarithm of disposable income of rural people per capita. SEPAP represents whether or not a county was selected for the photovoltaic poverty alleviation policy in a specific year. SECONDGDPR depicts a proportion of the added value of the secondary industry to GDP. PUBEXINR shows the ratio of public expenditure to revenue. LN(AGACRE) examines the land used for facility agriculture facility agriculture land. EDUCATION estimates the ratio of number of secondary school students to the total population. MKTINDEX represents marketization index. LN(SUNHOUR) indicates sunlight exposure time. LN(GDPPROVINCE) is used to investigate the per capita GDP of the province where the county is located. ***, **, and * represent the significance levels of 1%, 5% and 10%, respectively. T-statistics are reported in parentheses.

Supplementary Note 1. Robustness check-Subsample analysis
PV poverty-alleviation pilot counties are currently concentrated in the western, eastern, and central regions. The average county level GDP per capita, per capita savings deposit-balance, per capita rural disposable income, and other important indicators reflecting the income level of residents (or farmers) in China's three major regions descend from east to west. Thus, the economic development of different regions has a ladder distribution. Poverty alleviation funds distribution discipline may be greater in areas with high income where the system is established and the marketization level is high. Moreover, highincome regions are also characterized by effective data/information sharing and cross-sectoral connectivity that extends to farmers and the agricultural sector. However, precisely because of the high levels of regional economic development, central and local governments may be inclined to support lowincome counties with funds and technical support, thereby reducing the poverty alleviation efforts of high-income counties.
In addition to difference in regional economic development, natural features impact the mode of PV comparative economies of scale, which may have further poverty alleviating effects. The regional impact of targeted PV poverty alleviation policies is therefore complex. The following hypotheses are proposed after the synthesis of this study: H4: The effect of PV poverty-alleviation policy is higher in areas with higher per capita income levels.
H5: The effect of PV poverty-alleviation policies is greater in the eastern parts of China.
We analyze the poverty alleviation effects of PV poverty alleviation policies by subsamples according to regional and economic conditions, in which the regional grouping focuses on comparing the east and the west, and the economic conditions refer to the county GDP per capita. The results in supplementary table 8 show that the coefficient of PV poverty alleviation policy on rural per capita disposable income in the eastern and western regions is positive and statistically significant. The coefficient of PV poverty alleviation policy variables in the east is 0.0716, which is slightly higher than 0.0585 in the western region. According to the per capita GDP value, our sample is divided into two categories: relatively rich areas and poverty-stricken areas. Empirical results reveal that the PV poverty alleviation policy has a greater effect on the poor regions, and the effect amounts to 4.86%. In rich counties, poverty alleviation has no significant results with a coefficient of 0.0013 and fails to pass the significance test. Combining the classifications, relatively poor places are more likely to alleviate poverty from the PV project policy. The reason why the PV policy in the eastern region has a significant effect on poverty alleviation may be that some eastern counties have lower county-level per capita GDP than those in the western region, although the overall economic level of the eastern region is relatively rich.
In addition, the complete establishment of the local system and the rational use of poverty alleviation resources in eastern regions also make the PV poverty alleviation policy effective.

Supplementary Note 2. Robustness check-Measures of dependent variables
The model is re-estimated by re-measuring the dependent variables using local per capita GDP and resident per capita savings deposit balance. However, the impact of PV policy on these two variables was not significant (see supplementary table 9). The possible reason is that the purpose of PV poverty 17 alleviation is to alleviate the poverty of farmers. However, the above two variables reflect the per capita income of the entire county, including urban residents. Therefore the per capita disposable income in rural areas is significantly affected by PV policies but not by these two alternative measures. This result reflects that the poverty reduction effect is due to PV intervention rather than other reasons, as other policy interventions may have a significant impact on the two alternative dependent variables. We also use the absolute value measure of rural per capita disposable income to re-test. Supplementary table 10 provides the robust results: the implementation of PV poverty alleviation policy can increase rural per capita disposable income by 353 yuan/year.

Supplementary Note 3. Robustness check-Endogenous treatment effect
We estimate a linear regression with endogenous treatment effects. The natural logarithm of sunlight hours and per capita GDP of the province where the county is located. These two variables are highly correlated with the selection of the SEPAP in China. Supplementary table 11 presents the relevant results.
In Column (1), the estimated coefficient of SEPAP amounts to 0.1471, which is significant at the 1% level. The economic magnitude of the treatment effect is twice of that in the baseline results without accounting for the endogenous treatment effects, indicating potential underestimate of the effect of SEPAP on rural household income in the baseline model. The result of treatment effects equation shows that sunlight hours are significantly positively correlated with the likelihood of being selected into the SEPAP. Column (2) adds natural logarithm of provincial per capita GDP into the equation for treatment effects, and the coefficient of SEPAP increases to 0.1630, which is significant at the 1% level. The result of treatment effects equation shows that provincial per capita GDP is negatively correlated with the likelihood of being selected into the SEPAP. The result is also consistent with our conjecture since the SEPAP is partially determined by the local economic conditions.