#install package igraph for network analysis
install.packages("igraph")
library(igraph)

#load data with the information on transactions between 50 different banks
data <- read.csv2('network.csv')
#understand the structure of the data: lender, borrower, loan amount, interest rate, date of transaction
str(data)
#load the characteristics of the banks (assets size)
size <- read.csv2('vertices.csv')

#create graph object using both datasets
network <- graph.data.frame(data, vertices = size)

#analyze network's basic properties
#Is network fully connected? Did all banks trade with each other?
is.connected(network)
#Are there multiple edges? Do banks lend/borrow from the same partners?
table(is.multiple(network))
#What is the percentage of banks lending/borrowing to each other?
reciprocity(network)
#Are there any loops in the data? Are the data collected accurately?
is.loop(network)
#exclude loops from the network
snetwork <- simplify(network, edge.attr.comb = "sum")

#plot the network
plot(snetwork, edge.arrow.size = 0.2)

#subset the data to analyze crisis period
monthlynetwork <- subset(data, (Year == 2008) & (Month == 9))
#calculate the changes in the amount of borrowings
mAmount <- with(data, aggregate(Amount, by = list(Month = Month, Year = Year), FUN = sum))
#plot the amounts of borrowings
plot(ts(mAmount$x, start = c(2007, 1), frequency = 12), ylab = 'Amount')
##calculate the network density over time (calculating the values for each month)
ds <- sapply(2007:2010, function(year) {
  sapply(1:12, function(month) {
    mdata <- subset(data, (Year == year) & (Month == month))
    graph.density(graph.data.frame(mdata))
  })
})
#plot the changes in density
plot(ts(as.vector(ds), start = c(2007, 1), frequency=12))
abline(v = 2008 + 259/366, col = 'red')

## calculate the systemic importance of banks based on different centrality measures
g <- graph.data.frame(data)
degree <- degree(g, normalized = TRUE)
between <- betweenness(g, normalized = TRUE)
closeness <- closeness(g, normalized = TRUE)
eigenv <- evcent(g, directed = TRUE)$vector
prank <- page.rank(g)
#plot the histograms for all centrality measures
hist(degree)
hist(between)
hist(closeness)
hist(eigenv)
hist(prank$vector)