var_signal_creation_loop.R

print(paste0(strrep('=',50)))
print(paste0(strrep('=',50)))
print(paste0(strrep('=',50)))
print(paste0('Estimation of basic-VAR forecasts'))

# Check if VAR forecasts have already been estimated
if (length(as.numeric(rownames(tail(subset(df_forecasts, trade_done == 1) ,1))))==0) {
  
  print(paste0(strrep('=',50)))
  print(paste0(strrep('=',50)))
  print(paste0(strrep('=',50)))
  print(paste0('Estimation of VAR forecasts'))
  
  if (length(initial_iloc_to_forecast<nrow(df_forecasts))!=0) {
    # The for loop to estimate the model each day
    for (i in initial_iloc_to_forecast:nrow(df_forecasts)) {
      
      # Set the current iteration date
      iter_date <- df_forecasts[i,'date']
      # Print the date
      print(paste0(strrep('=',50)))
      print(paste0("Date is ",iter_date))
  
      # Select the the 6 years of data to train the model
      data <- subset(var_data, var_data$date<iter_date)[,match(tickers,colnames(var_data))]
  
      # Estimate up to the 15-lag VAR with no constants or trends
      lagselect <- VARselect(data, lag.max = 15, type = "none")
      
      # Choose the best lag order as per the Bayesian Information Criteria
      lag_order <- lagselect$selection[3]
      
      # Estimate the best VAR model
      model <- VAR(data, p = lag_order, type = "none", season = NULL, exog = NULL) 
      
      # Forecast one step ahead
      var_forecast <- predict(model, n.ahead = 1)
      
      # We go long if the forecast return is positive, otherwise, we make no position.
      for (ticker in tickers) {
        df_forecasts[i,paste0(ticker,'_var_signal')] = if (var_forecast$fcst[[ticker]][1]>=0) 1 else next
      }
      
      # Print the signals
      for (ticker in tickers) {
        print(paste0(ticker," var signal is ",df_forecasts[i,paste0(ticker,'_var_signal')]))
      }
    }
  } else {
    print(paste0('Estimation of TVP-VAR-SV forecasts has been completed before...'))
  }
}