LoanRisk/app.R

# Estimate exposure at risk and create report
# Application developed by LaNubia

library(shiny)
library(dplyr)
library(ggplot2)
library(lubridate)
library(survival)
library(pec)
library(DT)
library(shinycssloaders)
library(imfr)
library(patchwork)
library(tidyr)
library(shinythemes)
library(triangle)
library(DiagrammeR)
library(stringr)


options(shiny.reactlog = TRUE, appDir = getwd())

source("mod_basic.R")
source("panel.R")

# source("secretary.R")
# source("forplumber.R")
source("modals.R")

# Adding initial data


# Define UI for application
ui <- navbarPage(
  theme = shinytheme("cosmo"),
  title = "LoanRisk",
  panel1
  
)

# Define server logic required to draw a histogram
server <- function(input, output) {
  # Data
  
  # Some reactive values
  collateral.dt<-reactiveVal()
  transaction.dt<-reactiveVal()
  
  # Adding initial data to reactive values
  collateral.dt(read.csv("./data/collateral.csv") )
  
  
  transaction.dt(read.csv("./data/transactions.csv") )
  
  # Transforming Data
  new.data <- reactive({
    ndt<- transaction.dt()
    withProgress(message = "Trying to read",
                 detail = "Hope the handwriting is legible!",
                 value = 0,
                 {
                   setProgress(value = 1, message = "Patience test 1 of 2")
                   ndt <-
                     ndt %>%
                     # Change format of some columns
                     mutate(
                       origination_date = ymd(origination_date),
                       maturity_date = ymd(maturity_date),
                       report_date = ymd(report_date)
                     ) %>%
                     # Add age of loan, loan tenure in months, which are compulsory parameters
                     mutate(age_of_asset_months = round(as.numeric(
                       report_date - origination_date
                     ) / 30)) %>%
                     mutate(loan_tenure_months = round(as.numeric(
                       maturity_date - origination_date
                     ) / 30)) %>%
                     group_by(id) %>%
                     # Arranging to avoid mistakes in lag
                     arrange(report_date) %>%
                     # Add lag of bureau score and total number of defaults. Lag added for delta creation
                     mutate(
                       cum_default = cumsum(default_flag),
                       bureau_score_lag = ifelse(
                         is.na(lag(bureau_score_orig, 1)),
                         bureau_score_orig,
                         lag(bureau_score_orig, 1)
                       )
                     ) %>%
                     # Adding delta of bureau score (ask me the reason if you do not know why)
                     mutate(bureau_score_delta = bureau_score_lag -
                              bureau_score_orig) %>%
                     # Adding quarter info for matching later with macroeconomic data
                     mutate(qtr = paste0(year(report_date), "-Q", quarter(report_date))) %>%
                     # Removing dummy
                     dplyr::select(-bureau_score_orig)
                   setProgress(value = 2, message = "Patience test 1 of 2")
                 })
    ndt
    
    
  })
  
  
  
  # Show uploaded data
  output$up_data <- renderDataTable({
    
    DT::datatable(
      new.data(),
      extensions = c("Buttons"),
      options = list(
        pageLength = 5,
        scrollX = TRUE,
        dom = 'Bfrtip',
        filter = list(position = 'top', clear = FALSE),
        buttons = list(
          list(extend = "csv", text = "Download Visible", filename = "page",
               exportOptions = list(
                 modifier = list(page = "current")
               )
          ),
          list(extend = "csv", text = "Download All", filename = "data",
               exportOptions = list(
                 modifier = list(page = "all")
               )
          )
        )
      )
    )
    
  })
  
  maxdate <- reactive(max(new.data()$report_date))
  # Show some stats
  
  basicstatServer("nofloans", dt = new.data())
  
  
  
  # IMF Data
  
  imf.data <- reactive({
    input$fetchimf
    req(!is.null(new.data()), input$fetchimf)
    # Defining database and other parameters for query of macroeconomic data
    databaseID <- "IFS"
    startdate = min(new.data()$report_date)
    enddate = max(new.data()$report_date)
    country = countries[countries$Country == input$country,]$Alpha.2.code
    withProgress(message = "Extracting data from IMF",
                 detail = "Hope their server is up!",
                 value = 0,
                 {
                   setProgress(value = 1, message = "Trying to reach..")
                   print(country)
                   imf.data <- tryCatch(
                     expr={
                       imf_data(
                         databaseID,
                         c("NGDP_NSA_XDC",
                           "PCPI_IX"),
                         country = stringr::str_trim(country),
                         start = startdate,
                         end = enddate,
                         freq = "Q",
                         return_raw = FALSE,
                         print_url = T,
                         times = 3
                       ) 
                     },
                     error = function(e){          # Specifying error message
                       showModal(
                         modalDialog(
                           "Error in IMF database. Sorry for the inconvenience. Can you please try again later?"
                         )
                       )
                       message("Error with IMF database. This is usually temporary. Sorry for the inconvenience. Please try again later.")
                     },
                     finally = {                   # Specifying final message
                       message("Error with IMF database. Please try again later.")
                     }
                   )
                   
                   
                   
                   
                   
                   setProgress(value = 2, message = "Done")
                 })
    
    imf.data
  })
  
  # Adding macroeconomic data. Currently GDP and prices data are extracted
  dataset_with_eco <- reactive({
    req(!is.null(imf.data()))
    
    withProgress(message = "Trying Hard",
                 detail = "Hang on",
                 value = 0,
                 {
                   setProgress(value = 1, message = "working..")
                   
                   # New dataset by joining GDP and prices data
                   dataset_with_eco <-
                     new.data() %>%
                     left_join(imf.data(), by = c("qtr" = "year_quarter")) %>%
                     rename(gdp = NGDP_NSA_XDC, prices = PCPI_IX) %>%
                     select(-iso2c) %>%
                     group_by(id) %>%
                     mutate(gdp_lag = lag(gdp, 1),
                            prices_lag = lag(prices, 1)) %>%
                     dplyr::select(-qtr) %>%
                     dplyr::select(-c(gdp, prices))
                   
                   # removing rows with no macroeconomic data
                   dataset_eco <-
                     dataset_with_eco[!is.na(dataset_with_eco$gdp_lag) &
                                        !is.na(dataset_with_eco$prices_lag), ]
                   setProgress(value = 2, message = "working..")
                 })
    
    return(dataset_eco)
  })
  
  # GDP Forecast
  
  gdp.forecast <- reactive({
    imf.data <- imf.data()
    imf.data$year_quarter <-
      zoo::as.yearqtr(imf.data$year_quarter, format = "%Y-Q%q")
    ga <- imf.data[, c(2, 3)]
    minqg <- min(imf.data$year_quarter)
    gats <- ts(ga$NGDP_NSA_XDC, # AIP_IX,
               start = minqg,
               frequency = 4)
    
    withProgress(message = "Forecasting GDP",
                 detail = "Hang on",
                 value = 0,
                 {
                   setProgress(value = 1, message = "working..")
                   fit.gdp <-
                     forecast::auto.arima(gats, seasonal = FALSE)
                   gdp.forecast <-
                     fit.gdp %>% forecast::forecast(h = 60)
                   setProgress(value = 2, message = "Done")
                 })
    gdp.forecast
  })
  
  # Price forecast
  pats.forecast <- reactive({
    imf.data <- imf.data()
    imf.data$year_quarter <-
      zoo::as.yearqtr(imf.data$year_quarter, format = "%Y-Q%q")
    gp <- imf.data[, c(2, 4)]
    minqg <- min(imf.data$year_quarter)
    pats <- ts(gp$PCPI_IX, start = minqg, frequency = 4)
    withProgress(message = "Forecasting price",
                 detail = "Hang on",
                 value = 0,
                 {
                   setProgress(value = 1, message = "working..")
                   fit.pats <-
                     forecast::auto.arima(pats, seasonal = FALSE)
                   pats.forecast <-
                     fit.pats %>% forecast::forecast(h = 60)
                   setProgress(value = 2, message = "Done")
                 })
    
    pats.forecast
  })
  
  # Show full data
  
  output$fulldata <- renderDataTable({
    validate(need(!is.null(dataset_with_eco()),
                  message = "Data Not ready yet"))
    a <- dataset_with_eco()
    
    DT::datatable(
      a,
      extensions = c("Buttons"),
      options = list(
        pageLength = 5,
        scrollX = TRUE,
        dom = 'Bfrtip',
        filter = list(position = 'top', clear = FALSE),
        buttons = list(
          list(extend = "csv", text = "Download Visible", filename = "page",
               exportOptions = list(
                 modifier = list(page = "current")
               )
          ),
          list(extend = "csv", text = "Download All", filename = "data",
               exportOptions = list(
                 modifier = list(page = "all")
               )
          )
        )
      )
    )
  })
  
  # Probability of default
  
  output$expcalcu <- renderUI({
    req(!is.null(dataset_with_eco()), input$fetchimf)
    fluidRow(
      h3("Probability of asset going bad"),
      p(
        "In this section you can see the distribution of probabilities of the assets going bad by year, upto 5 years from the last reporting date. Please click the button to proceed."
      ),
      actionButton(
        "start_model_selection",
        "Initiate Model Selection",
        class = "glow"
      ),
      withSpinner(
        plotOutput("probability_default"),
        type = 7,
        color = "black"
      )
    )
  })
  
  
  
  # Model Selection. Using functions. In production these are to be converted to APIs
  selected_model <- reactive({
    req(input$start_model_selection,!is.null(dataset_with_eco()))
    model_sel(dff=dataset_with_eco())
  })
  
  # Preparing tables with predictions. Using functions. In production these are to be converted to APIs
  predicted_table <- reactive({
    input$start_model_selection
    req(!is.null(selected_model()))
    
    z<-predic_t(dff=dataset_with_eco(),
                gdpfor=gdp.forecast(),
                prfor=pats.forecast(),
                maxdate=maxdate(), 
                final.model= selected_model())
    z
  })
  
  
  # Plot Probability of default
  output$probability_default <- renderPlot({
    input$start_model_selection
    req(predicted_table())
    withProgress(message = "Plotting",
                 detail = "Just a moment",
                 value = 0,
                 {
                   setProgress(value = 1, message = "Rushing")
                   pl <- predicted_table() %>%
                     group_by(name) %>%
                     #summarise(pd.me=median(value), pd.min=quantile(value,0.05), pd.max=quantile(value, 0.95))%>%
                     slice_head(n = 6) %>%
                     mutate(name = factor(
                       name,
                       levels = c(
                         "risk_current",
                         "risk_1yr",
                         "risk_2yr",
                         "risk_3yr",
                         "risk_4yr",
                         "risk_5yr"
                       )
                     )) %>%
                     filter(!is.na(name)) %>%
                     ggplot(aes(
                       x = name,
                       y = value,
                       fill = name
                     )) +
                     geom_violin() +
                     geom_boxplot(width = 0.1,
                                  color = "black",
                                  alpha = 0.2) +
                     labs(x = "",
                          y = "Probability",
                          title = "Probability of default") +
                     theme_bw() +
                     theme(legend.title = element_blank(),
                           legend.position = "bottom")
                   setProgress(value = 2, message = "Done")
                 })
    pl
  })
  
  # Show scenario creation options when exposures and risks are calculated
  
  output$scenario_opts <- renderUI({
    req(!is.null(predicted_table()),
        input$start_model_selection)
    fluidRow(
      # remove class when btn clicked
      br(),
      h3("Possible Scenarios"),
      p(
        "In this section, the parameters for simulation will be selected. Please click on start simulation button to proceed"
      ),
    )
    fluidRow(
      column(
        width = 3,
        h5("Select expected change in collateral value"),
        p("-ve means reduction in value"),
        sliderInput(
          "mode_dep",
          "Most Probable",
          min = -1,
          max = 1,
          value = -0.7,
          step = 0.1
        ),
        sliderInput(
          "min_dep",
          "Minimum",
          min = -1,
          max = 1,
          value = -1,
          step = 0.1
        ),
        sliderInput(
          "max_dep",
          "Max",
          min = -1,
          max = 1,
          value = 1,
          step = 0.1
        )
      ),
      column(
        width = 3,
        sliderInput(
          "discount_rate",
          "Select Discount rate / WACC",
          min = 0.00,
          max = 0.2,
          value = 0.02,
          step = 0.01
        ),
        actionButton("update", "Start Simulation", class = "glow"),
        h3("Please do not forget to click this button.")
      ),
      column(
        width = 6,
        p(
          "This section shows the possible exposure in case the asset goes bad."
        ),
        plotOutput("exposure_on_default")
      )
    )
    
  })
  
  # Exposure on Default
  
  output$exposure_on_default <- renderPlot({
    input$update
    req(nrow(predicted_table())>0)
    discount_rate_pa <- input$discount_rate
    
    withProgress(message = "Plotting",
                 detail = "Just a moment",
                 value = 0,
                 {
                   setProgress(value = 1, message = "Almost there")
                   pl <- predicted_table() %>%
                     mutate(pv.balance = balance / (1 + discount_rate_pa) ^
                              (r_n - 1)) %>%
                     mutate(exposure_on_default = pv.balance * value) %>%
                     group_by(name) %>%
                     summarise(
                       pv.balance = sum(pv.balance),
                       exposure_on_default = sum(exposure_on_default)
                     ) %>%
                     mutate(name = factor(
                       name,
                       levels = c(
                         "risk_current",
                         "risk_1yr",
                         "risk_2yr",
                         "risk_3yr",
                         "risk_4yr",
                         "risk_5yr"
                       )
                     )) %>%
                     pivot_longer(
                       cols = c("pv.balance", "exposure_on_default"),
                       names_to = "type",
                       values_to = "amount"
                     ) %>%
                     ggplot(aes(
                       x = name,
                       y = amount / 1000000,
                       fill = type,
                       label = paste0(round(amount / 1000000), " M")
                     )) +
                     geom_col(position = "dodge") +
                     geom_text(aes(y = (amount / 1000000) + 5), position = position_dodge(width = 1)) +
                     labs(
                       x = "",
                       y = "Amount",
                       title = "Exposure on default",
                       subtitle = "Amounts discounted"
                     ) +
                     theme_bw() +
                     theme(legend.title = element_blank(),
                           legend.position = "bottom")
                   setProgress(value = 2, message = "Done")
                 })
    
    pl
    
  })
  
  output$credit_loss <- renderUI({
    req(!is.null(predicted_table()), input$update)
    validate(
      need(input$mode_dep <= input$max_dep, message = "Most probable value should be less than or equal to Max"),
      need(input$min_dep <= input$mode_dep, message = "Min value should be less than or equal to Most probable value"),
      need(input$min_dep <= input$mode_dep, message = "Min value should be less than or equal to Max value")
    )
    fluidRow(
      # remove class when btn clicked
      h3("Simulated Credit Loss with probability"),
      p(
        "This section shows the distribution of the simlated possible losses. Please select the time to check the corresponding distribution."
      ),
      br(),
      selectInput(
        "riskperiod",
        "Select time",
        choices = c(
          "risk_current",
          "risk_1yr",
          "risk_2yr",
          "risk_3yr",
          "risk_4yr",
          "risk_5yr"
        ),
        selected = "risk_1yr"
      ),
      br(),
      column(
        width = 6,
        p(
          "Please click an drag to check the probabilities between ranges of possible loss."
        ),
        plotOutput("simres",
                   brush = brushOpts(
                     id = "sim_res_sel", direction = "x"
                   )),
        verbatimTextOutput("cumprob")
      ),
      column(
        width = 6,
        p(
          "This section shows the overall estimated credit loss by year from last reporting date, upto 5 years"
        ),
        plotOutput("exp_credit_loss")
      )
      
    )
  })
  
  # collateral<-reactive({
  #     collateral
  # })
  
  simdata <- reactive({
    req(nrow(predicted_table())>0)
    req(!is.null(input$discount_rate))
    input$update
    collateral<-collateral.dt()
    discount_rate_pa <- input$discount_rate
    withProgress(message = "Monte Carlo Simulation",
                 detail = "1000 simulations",
                 value = 0,
                 {
                   setProgress(value = 1, message = "preparing")
                   zz <-
                     predicted_table() %>%
                     ungroup() %>%
                     group_by(id) %>%
                     mutate(pv.balance = balance / (1 + discount_rate_pa) ^
                              (r_n - 1)) %>%
                     left_join(collateral, by = "id") %>%
                     ungroup() %>%
                     select(name, pv.balance, value, collateral)
                   setProgress(value = 2, message = "simulating")
                   
                   sim.t <-
                     data.frame(matrix(ncol = 1000, nrow = nrow(zz)))
                   colnames(sim.t) <-
                     paste0("sim", seq(1:1000))
                   sim.col.prob <-
                     (1 + rtriangle(
                       1000,
                       a = input$min_dep,
                       b = input$max_dep,
                       c = input$mode_dep
                     ))
                   
                   for (i in 1:1000) {
                     sim.t[, i] <-
                       round(zz$value * (zz$pv.balance - zz$collateral * sim.col.prob[i]),
                             2)
                   }
                   
                   sim.t[sim.t < 0] <- 0
                   sim.t <- cbind(zz, sim.t)
                   
                   setProgress(value = 3, message = "Done")
                   
                 })
    
    sim.t
  })
  
  simresdata <- reactive({
    dt <- simdata() %>%
      filter(name == input$riskperiod) %>%
      ungroup() %>%
      select(matches("sim?"))
    hist.pro <- density(colSums(dt))
    hist.pro
  })
  
  output$simres <- renderPlot({
    req(!is.null(simresdata()))
    hist.pro <- simresdata()
    pro_dens <- data.frame(hist.pro$x, hist.pro$y)
    
    ggplot(pro_dens,
           aes(x = hist.pro.x,
               y = hist.pro.y)) + geom_area(aes(y = hist.pro.y)) +
      labs(
        x = "Amount",
        y = "Chance",
        title = "Simulated expected credit loss",
        subtitle = "Amounts discounted"
      ) +
      theme_bw()
  })
  
  output$cumprob <- renderText({
    req(!is.null(input$sim_res_sel), !is.null(simresdata()))
    hist.pro <- simresdata()
    pro_dens <- data.frame(hist.pro$x, hist.pro$y)
    res <- brushedPoints(pro_dens, input$sim_res_sel)
    if (nrow(res) == 0) {
      return()
    }
    paste0(
      "Probability of the loss to be between ",
      min(res$hist.pro.x),
      " and ",
      max(res$hist.pro.x),
      " is " ,
      sum(res$hist.pro.y)
    )
  })
  
  finaldt <- reactive({
    sim_fin(simdata())
    
  })
  
  output$exp_credit_loss <- renderPlot({
    req(!is.null(finaldt()))
    finaldt() %>%
      mutate(amount = x * y) %>%
      group_by(r) %>%
      summarise(amount = sum(amount)) %>%
      ggplot(aes(
        x = r,
        y = round(amount / 1000),
        label = paste0(round(amount / 1000), " K")
      )) +
      geom_col() +
      geom_text(aes(y = round(amount / 1000) + 1)) +
      labs(
        title = "Estimated Credit Loss",
        subtitle = "Weighted sum of 1000 simulations",
        x = "",
        y = "Amount"
      ) +
      theme_bw()
  })
  
  output$dlmanager <- renderUI({
    input$update
    req(!is.null(finaldt()))
    fluidRow(
      textInput("author", "Enter Your Name"),
      downloadButton("report", "Generate report")
    )
  })
  
  output$report <- downloadHandler(
    # For PDF output, change this to "report.pdf"
    filename = "easyIFRSriskReport.pdf",
    content = function(file) {
      withProgress(message = "Monte Carlo Simulation",
                   detail = "1000 simulations",
                   value = 0,
                   
                   {
                     setProgress(value = 1, message = "Copying to temp directory")
                     
                     
                     tempReport <-
                       file.path(tempdir(), "easyIFRSriskReport.Rmd")
                     file.copy("easyIFRSriskReport.Rmd", tempReport, overwrite = TRUE)
                     setProgress(value = 2, message = "Passing parameters")
                     
                     ndata <- new.data()
                     edata <- dataset_with_eco()
                     predtable <- predicted_table()
                     simudata <- simdata()
                     fidata <- finaldt()
                     gfor <- gdp.forecast()
                     pfor <- pats.forecast()
                     
                     params <- list(
                       user = input$author,
                       newdata = ndata,
                       predtabledata = predtable,
                       disrate = input$discount_rate,
                       simdata = simudata,
                       finaldata = fidata,
                       gdpf = gfor,
                       pf = pfor
                     )
                     setProgress(value = 3, message = "Knitting. Hang on.")
                     
                     rmarkdown::render(
                       tempReport,
                       output_file = file,
                       params = params,
                       envir = new.env(parent = globalenv())
                     )
                     setProgress(value = 4, message = "Done.")
                   })
    }
  )
  
  # File uploading Module
  
  observeEvent(input$uploadnew, {
    showModal(
      modal1
    )
  })
  
  observeEvent(input$closemodal1,{
    removeModal()
  })
  observeEvent(input$closemodal2,{
    removeModal()
  })
  
  # Uploading temp file and collecting info about the columns
  observeEvent(input$uploadfiles,{
    
    df.tr <- reactive({
      inFile <- input$transaction
      if (is.null(inFile))
        return(NULL)
      df <- read.csv(inFile$datapath,
                     header = input$header,
                     sep = input$sep,
                     quote = input$quote)
      df
    })
    
    df.c <- reactive({
      inFile <- input$collaterals
      if (is.null(inFile))
        return(NULL)
      df <- read.csv(inFile$datapath,
                     header = input$header,
                     sep = input$sep,
                     quote = input$quote)
      df
    })
    updateVarSelectInput("collateralid","Select id column", df.c(),session = getDefaultReactiveDomain())
    updateVarSelectInput("collateralvalue","Select Collateral value column", df.c(),session = getDefaultReactiveDomain())
    updateVarSelectInput("reportdate","Select report date column", df.tr(),session = getDefaultReactiveDomain())
    updateVarSelectInput("origindate","Select origin date column", df.tr(),session = getDefaultReactiveDomain())
    updateVarSelectInput("maturitydate","Select maturity date column", df.tr(),session = getDefaultReactiveDomain())
    updateVarSelectInput("assettype","Select asset classifier column", df.tr(),session = getDefaultReactiveDomain())
    updateVarSelectInput("customertype","Select customer classifier column", df.tr(),session = getDefaultReactiveDomain())
    updateVarSelectInput("otherfact","Select any other classifier column", df.tr(),session = getDefaultReactiveDomain())
    updateVarSelectInput("bureauscore","Select bureau score column", df.tr(),session = getDefaultReactiveDomain())
    updateVarSelectInput("balance","Select asset balance column", df.tr(),session = getDefaultReactiveDomain())
    updateVarSelectInput("status","Select loan status", df.tr(),session = getDefaultReactiveDomain())
    updateVarSelectInput("defaultflag","Select default flag column", df.tr(),session = getDefaultReactiveDomain())
    updateVarSelectInput("transid","Select id column", df.tr(),session = getDefaultReactiveDomain())
    
  })
  
  # Modal 2
  
  observeEvent(input$uploadfiles, {
    showModal(
      modal2
    )
  })
  
  
  
  observeEvent(input$confirmupload,{
    transaction.dt(
      {
        df <- read.csv(input$transaction$datapath,
                       header = input$header,
                       sep = input$sep,
                       quote = input$quote)
        head(df)
        
        df<-
          df %>% 
          rename(report_date=input$reportdate,
                 origination_date=input$origindate,
                 maturity_date=input$maturitydate,
                 asset_type=input$assettype,
                 customer_type=input$customertype,
                 bureau_score_orig=input$bureauscore,
                 balance=input$balance,
                 loan_status=input$status,
                 default_flag=input$defaultflag,
                 id=input$transid)
        
        if(input$dateformat=="ymd"){
          df$report_date<-ymd(df$report_date)
          df$origination_date<-ymd(df$origination_date)
          df$maturity_date<-ymd(df$maturity_date)
        } else {
          df$report_date<-dmy(df$report_date)
          df$origination_date<-dmy(df$origination_date)
          df$maturity_date<-dmy(df$maturity_date)
        }
        
        df$loan_status<-as.integer(df$loan_status)
        
        # Covert to factors
        df$asset_type<-as.factor(df$asset_type)
        df$customer_type<-as.factor(df$customer_type)
        if(!is.null(input$otherfact)){
          for(i in 1:length(input$otherfact)){
            df$input$otherfact[i]<-as.factor(df$input$otherfact[i])
          } 
        }
        
        df
      }
    )
    
    collateral.dt(
      {
        df <- read.csv(input$collaterals$datapath,
                       header = input$header,
                       sep = input$sep,
                       quote = input$quote)
        df<-
          df %>% 
          rename(id=input$collateralid,
                 collateral=input$collateralvalue)
        df
      }
    )
    removeModal()
  })
  
  
}

# Run the application
shinyApp(ui = ui, server = server)