# Controle Sintético no R
# Prof. Raphael B. Corbi/ Monitor: Alan Leal

# Leiamos os pacotes necessários para a análise:
if(!require(tidyverse)){install.packages("tidyverse");require(tidyverse)}
if(!require(Synth)){install.packages("Synth");require(Synth)}
devtools::install_github("edunford/tidysynth")
require(tidysynth)

# Lendo agora os dados:
load("/Users/alanleal/Downloads/dataverse_files/data.ME.26.12.21.RData")
data_long <- data_long %>% 
  filter(cntry_code %in% c(
    "BRA",
    "EST",
    "LTU",
    "LVA",
    "SVK",
    "HUN",
    "CHL",
    "TUR",
    "CRI",
    "MEX",
    "COL",
    "IND"))

# Vamos usar um método alternativo usando o pacote tidysynth:
mega_sports_out <-
  
  data_long %>%
  
  # initial the synthetic control object
  synthetic_control(outcome = gdppcgr, # outcome
                    unit = cntry_code, # unit index in the panel data
                    time = year, # time index in the panel data
                    i_unit = "BRA", # unit where the intervention occurred
                    i_time = 2014, # time period when the intervention occurred
                    generate_placebos=T # generate placebo synthetic controls (for inference)
  ) %>%
  
  # Generate the aggregate predictors used to fit the weights
  
  # average log income, retail price of cigarettes, and proportion of the
  # population between 15 and 24 years of age from 1980 - 1988
  generate_predictor(time_window = 1993:2014,
                     industry=mean(industry,na.rm=TRUE),
                     export=mean(export,na.rm=TRUE),
                     lifeexp=mean(lifeexp,na.rm=TRUE),
                     pop=mean(pop,na.rm=TRUE)) %>%
  # Generate the fitted weights for the synthetic control
  generate_weights(optimization_window = 1993:2014, # time to use in the optimization task
                   margin_ipop = .02,sigf_ipop = 7,bound_ipop = 6 # optimizer options
  ) %>%
  
  # Generate the synthetic control
  generate_control()
# Pré e pós-intervenção:
mega_sports_out %>% plot_trends()
# Plotando as diferenças no controle e sintético:
mega_sports_out %>% plot_differences()
mega_sports_out %>% plot_weights()

# 
mega_sports_out %>% grab_balance_table()

mega_sports_out %>% grab_loss() %>% filter(.placebo==0) %>% select(variable_mspe)