R - Random number generation in ggplot and Shiny - r

Question

I am building a plot of Net Present Value (NPV), using FinCal package, and its odds. For the NPV, the cash-flows are simulated using a triangular distribution for sales, normal distribution for costs and so on. So, here is a snippet of what I am doing:
npvCdf <- function(n) {
N <- sort(n)
P <- ecdf(N)
return(P)
}
makePlot <- function(C, m) {
N <- m$NPV / C$MILLION
P <- npvCdf(N)
#
# NPV distribution curve
n <- sort(N)
p <- P(n) * 100
df <- data.frame(npv = n, odds = p)
#
# Points of interest
o <- C$NPV_BREAK_EVEN_WORST_ODDS
q <- round((quantile(n, o)), C$DIGITS)
e <- C$NPV_BREAK_EVEN_VALUE
b <- P(e) * 100 # THIS IS THE ERROR I CANT FIGURE OUT
w <- o * 100
s <- getBreakEven(C, m)
#
# Labels
npvOdds <- paste("Odds of break-even : ", b, "%")
salesThresh <- paste("Sales threshold : ", s)
worstCase <-
paste("Worst case (# 5% odds) : ", q, "million")
#
# Make plot
#
g <- ggplot(df, aes(x = npv, y = odds)) +
geom_line(colour = "blue") +
labs(title = "NPV and Odds") +
labs(x = "NPV (million)") +
labs(y = "Percent (%)") +
geom_vline(xintercept = e,
colour = "red",
linetype = "longdash") +
geom_hline(yintercept = b,
colour = "green",
linetype = "longdash") +
geom_vline(xintercept = q,
colour = "green",
linetype = "dotdash") +
geom_hline(yintercept = w,
colour = "red",
linetype = "dotdash")
The C is a data frame of all the constants that are used for calculations of cash-flows, NPV calculations, etc. For example, C$MILLION=1000000 used to divide NPV for simpler representation. The m is a data-frame of sales, cash-flows and NPV per simulation. The simulations are used for cash-flows (triangular distribution), variable cost (normal distribution) and so on.
And, here is the Shiny code that uses the above snippet.
library(shiny)
source("../npd-c.R")
# Define server logic
shinyServer(function(input, output) {
output$npdPlot <- renderPlot({
C <- data.frame(2017,5000,1000000,3,100,500000,0.0,0.05,0.1,
input$salesRange[1],
input$salesRange[2],
input$salesMode,
input$demDeclMean,
input$demDeclSd,
input$varCostMean,
input$varCostSd,
input$fixedCostRange[1],
input$fixedCostRange[2]
)
names(C) <-
c(
"SEED",
"ITERATIONS",
"MILLION",
"DIGITS",
"PRICE",
"OUTLAY",
"NPV_BREAK_EVEN_VALUE",
"NPV_BREAK_EVEN_WORST_ODDS",
"HURDLE_RATE",
"SALES_TRIANG_MIN",
"SALES_TRIANG_MAX",
"SALES_TRIANG_MODE",
"DEM_DECL_FACTOR_MEAN",
"DEM_DECL_FACTOR_SD",
"VAR_COST_RATE_MEAN",
"VAR_COST_RATE_SD",
"FIX_COST_RATE_MIN",
"FIX_COST_RATE_MAX"
)
n <- npd(C,-1)
g <- makePlot(C,n)
g
})
})
The problem is as follows.
The same code when run in R, I get the plot right in terms of the NPV curve, horizontal and vertical lines. Whereas, when run as a Shiny application, the horizontal and vertical lines are hugely displaced. This is despite, hiving of the NPV and cash-flows code into a separate .R file and setting the same seed value for both the Shiny and non-shiny versions. For example, P(0)=40.07 without Shiny and P(0)=4.7 with Shiny application.
What am I missing?

Answer 1

First of all, let me say this is pretty useful code. It is a nice represenation of a monte-carlo simulation using NPV and I like the plots. It is a post I am pretty sure I will refer back to.
I think I see where the problem is though, it is basically more a matter of mis-interpretation and one small programming error.
The stated problem is that these plots are not showing the same results although they should be. The blue ecdf-NPV curves do look at first glance to be the same:
Shiny version:
Stand alone version:
However if you look carefully, you will see that in fact they are not the same, the expected NPV value (50 percent) in the first case is about 1.5 million, whereas it is only about 0.2 in the second case.
The curves look the same, but they are not. The other point is that there is an error in one of the calculations further confusing things. The "Odds of break-even" are incorrectly calculated and are actually the "Odds of losing money".
The correct calculation should be:
b <- (1-P(e)) * 100
And the correct odds of breaking even in the first case would be around 60%, and in the second case around 95%, which matches up with the expected NPV as well.