When and How to Support Renewables?—Letting the Data Speak - Green Energy and Efficiency: An Economic Perspective - page 302

Table 5 Results for wind power
Intercept 2.014

continent_dep_wind_lag5: rdd_res_squared_partsum3_lag2 − 0.055

continent_dep_wind_partsum4_lag3: rdd_res_squared_partsum1_lag3 0.012

−

continent_dep_wind_partsum4_lag3: rdd_res_squared_partsum2_lag2 0.068

−

continent_dep_wind_partsum4_lag3: rdd_res_squared_partsum2_lag3 − 0.062

continent_dep_wind_partsum5_lag2: rdd_res_squared_partsum2_lag2 0

continent_dep_wind: rdd_wind_partsum5_lag3 0.009

continent_dep_wind: rdd_wind_rooted 0.062

continent_dep_wind: rdd_wind_rooted_lag2 0.188

continent_dep_wind: rdd_wind_rooted_partsum1_lag1 0.012

continent_dep_wind: rdd_wind_rooted_partsum5_lag3 0.199

dep_total_lag5: continent_dep_wind_partsum2_lag1 − 0.003

dep_total: rdd_wind_partsum2 0.008

−

dep_wind_dwdist: rdd_wind_dwdist 0.016

−

dep_wind: dep_wind_dwdistwces 0.069

pat_solar_rooted_lag1 0.36

pat_solar_rooted_partsum1 0.034

pat_total_logged 0.068

0.045
rdd_wind_rooted_lag5 0.015
rdd_wind_rooted_partsum3_lag2 0.002
rdd_wind_rooted_partsum5_lag2 0.346
Note Model chosen from >47,000 variables based on the lowest mean square error in predicting
the n

rdd_res_squared_lag4: continent_dep_wind_partsum4_lag3

−

' th observation based on n 1 data. Coef cients rounded at the third decimal digit

−

Beyond these three main drivers, there are a number of variables with typically
small negative values that are somewhat dif cult to interpret. We would see them as

correction factors that reduce the aforementioned effects in certain conditions. The
largest is the interaction of RD&D spending on renewables with the deployment of
wind on the continent (continent_dep_wind: rdd_res). One way of interpreting this
is that countries with a lot of non-wind RD&D spending do not bene

t (in terms of
wind patents) as much from the deployment of wind turbines on their continent, as
countries that focus their renewables RD&D on wind (Table 5 ).
The stability of the above-presented results is con

rmed by a plot of the coef-

cients selected by the Lasso for a range of lambdas (Fig. 9 ).
To get some indication of the quality of our results we calculate the share of
variance in the patenting behaviour our model is able to explain (similar to the R 2 ).
The results are displayed in the following table. Given their parsimonious param-
eterisation the

' goodness-of- t ' performance of both models is impressive

(Table 6 ).