server.R

#  ----------------------------------------------------------------------------
#  PRECIOUS METALS PRICE FORECAST
#  File: server.R
#  (c) 2015 - Enrique Pérez Herrero
#  20/Dec/2015
#  GNU GENERAL PUBLIC LICENSE Version 3, 29 June 2007
#  ----------------------------------------------------------------------------

source('helpers.R')

metals.df <- getPreciousMetals(precious_metals, currency_list)

start.date <- head(metals.df$date, 1)
end.date <- tail(metals.df$date, 1)

# Forecast interval parameter
forecast.days <- 90

shinyServer(function(input, output) {
    
    # Metal Prices plot using ggplot2 
    output$prices.plot <- renderPlot({
        tag <- paste(input$metal_id, input$metal_curr, sep = '.')
        ggplot(data = metals.df , aes_string('date', tag)) +
            geom_line() +
            stat_smooth(
                method = input$method, formula = y ~ x, size = 1
            ) +
            xlab('')
        
    })
    # Forecasted data with auto.arima plot 
    output$prediction.plot <- renderPlot({
        tag <- paste(input$metal_id, input$metal_curr, sep = '.')
        selected.metal <- ts(zoo(metals.df[tag],
                                 order.by = metals.df$date))
        arima.fit <- auto.arima(selected.metal,
                                approximation = FALSE,
                                trace = FALSE)
        pred <- forecast(arima.fit, h = forecast.days)
        arimaForecastPlot(pred, start = start.date, ylabel = tag)
        
    })
    
    # Data first differences plot
    output$diff.plot <- renderPlot({
        tag <- paste(input$metal_id, input$metal_curr, sep = '.')
        selected.metal <-
            ts(zoo(metals.df[tag], order.by = metals.df$date))
        
        ifelse(
            input$chcklog,
            selected.metal <- diff(log10(selected.metal)),
            selected.metal <-  diff(selected.metal)
        )
        
        diff.df <- data.frame(date = seq(
            from = start.date + 1, to = end.date , by = "1 day"
        ),
        price = selected.metal)
        
        ggplot(data = diff.df , aes_string('date', tag)) +
            geom_line() +
            xlab('')
        
    })
    
    output$text.sd <- renderText({
        paste0("Start Date: ",
               start.date <- head(metals.df$date, 1))
    })
    
    output$text.ed <- renderText({
        paste0("End Date: ",
               end.date <- tail(metals.df$date, 1))
    })
    
    # ARIMA results from auto.arima
    output$text.arima <- renderPrint({
        tag <- paste(input$metal_id, input$metal_curr, sep = '.')
        selected.metal <- ts(zoo(metals.df[tag],
                                 order.by = metals.df$date))
        arima.fit <- auto.arima(selected.metal,
                                approximation = FALSE,
                                trace = FALSE)
        summary(arima.fit)
    })
    
    
    output$metals.table <- renderDataTable(
        metals.df[, grepl(paste0(input$metal_id, '|date'), names(metals.df))]
    , options = list(pageLength = 10)
    )
    
    output$residuals.plot <- renderPlot({
        tag <- paste(input$metal_id, input$metal_curr, sep = '.')
        selected.metal <- ts(zoo(metals.df[tag],
                                 order.by = metals.df$date))
        arima.fit <- auto.arima(selected.metal,
                                approximation = FALSE,
                                trace = FALSE)
        par(mfrow = c(1, 2))
        acf(ts(arima.fit$residuals), main = 'ACF Residual')
        pacf(ts(arima.fit$residuals), main ='PACF Residual')
    })
    
    forecast.table <- reactive({
        tag <- paste(input$metal_id, input$metal_curr, sep = '.')
        selected.metal <- ts(zoo(metals.df[tag],
                                 order.by = metals.df$date))
        arima.fit <- auto.arima(selected.metal,
                                approximation = FALSE,
                                trace = FALSE)
        pred <- forecast(arima.fit, h = forecast.days)
        pred <- as.data.frame(pred)
        pred <- cbind(date = 0, pred[-1])
        pred$date <-
            seq(from = end.date + 1,
                to = end.date + forecast.days,
                by = "1 day")
        return(pred)
        })
    
    output$table.arima <- renderDataTable({
        forecast.table()
    }, options = list(pageLength = 10))
})