How plausible are more than 1MM Refugees hitting the sea shores of Europe?

Rationale

As of 31 December 2015, how many refugees and migrants will UNHCR report as having arrived in Europe by sea? Europe is facing a refugee crisis […]. Question will be resolved using the data available on UNHCR’s Emergency Response Page for the Mediterranean as of 31 December 2015 (http://data.unhcr.org/mediterranean/regional.html ).

– from an Open Good Judgment Question

The code is in the public domain and I encourage you to play! (CC BY 3.0) Roland Kofler. Any Idea how to improve? Contact me via Gmail: my name separated by a dot.

Source on GitHub

Found the freaking raw data

Its all about Lesbos!

Facts

October=10 
September=9   #you get the idea
reportedAmount=723221 #1 November
aMillion=10^6
N=10000;
projectionMonth=30/26;
arrivals2014= c(3270, 4369, 7283, 17084, 16627, 26221, 28303, 33478, 33944, 23050, 13318, 9107)
arrivals2015= c( 5546, 7343, 10184, 29441,  40117,  53987,  75485,  130839, 172845, 220535, 128324*projectionMonth)
                 
lesbosArrivals2014= c(290, 413, 555, 620, 461, 824, 873, 1064, 1778, 2072, 959, 802)
lesbosArrivals2015= c(737, 1002, 3348, 4990, 7228, 14796, 23721, 56579, 95384, 135021, 81691*projectionMonth)

greeceArrivals2014= c(955, 1001, 1501,  1257,   1703,   3198,   3927,   6742, 7454, 7432,   3812,   2056)   
greeceArrivals2015= c(1690, 2945,   7634,   13133,  18444,  30717,  51951,  107843, 156457, 210824, 125311*projectionMonth)



syrianRefugees=4000000

Data Analysis

arrivalsFor2Years <- c(arrivals2014, arrivals2015)
arrivalTimeseries <- ts(arrivalsFor2Years, start=c(2014, 1), end=c(2015, 11), frequency=12)

lesbosArrivalsFor2Years <- c(lesbosArrivals2014, lesbosArrivals2015)
lesbosTimeseries <- ts(lesbosArrivalsFor2Years, start=c(2014, 1), end=c(2015, 11), frequency=12)

greeceArrivalsFor2Years <- c(greeceArrivals2014, greeceArrivals2015)
greeceTimeseries <- ts(greeceArrivalsFor2Years, start=c(2014, 1), end=c(2015, 11), frequency=12)

Seasonplots

The trend has reversed in November

suppressMessages(library(forecast, quietly = TRUE, warn.conflicts = FALSE, verbose=F))
#monthplot(arrivalTimeseries)
#library(forecast)
seasonplot(arrivalTimeseries) 

#seasonplot(lesbosTimeseries) 
seasonplot(greeceTimeseries)

Extreme situation in Lesbos

Lesbos is the main door to the EU, being the greatest Island of the EU only a few km ashore of Turkey it is reachable with a dingi boat

plot(lesbosArrivals2015/ lesbosArrivals2014[1:11], type = 'o', main="Lesbos 2015 vs Lesbos 2014")

plot(greeceArrivals2015/greeceArrivals2014[1:11], type = 'o', main="Greece 2015/2014")

changeRateLesbosArrivalsYoY <- lesbosArrivalsFor2Years/arrivalsFor2Years
plot(changeRateLesbosArrivalsYoY, type = 'o', main="How Lesbos became the door to Europe", xlab="month since Jan 2014", ylab="*100=% of Lesbos Refugees to Total")

If you are a Refugee Maximalista – someone who believes that more than a million will come – its worth to take a look at How Greek Island behaved in Winter 2014

Climate on the route

UNHCR has a useful Map of minimal temperatures

• Gaziantep, TR
• Adana, TR
• Konya, TR
• Lesbos, GR
• Skopje, MK
• Belgrade, SR
• Zagreb, CR

Snow cover in the Balkans http://data.unhcr.org/mediterranean/uploads/uploadsmediterranean/images/documents/dt_mediterranean_111.jpg

Monte Carlo Simulation

model parameters

if refugee inflow behave the same like last year in greece

changeMoM= greeceArrivals2015[11]/greeceArrivals2015[10] 

we get changeMoM=0.6858309

arrivalsImputedFuture= c(arrivalsFor2Years, arrivals2015[11] * changeMoM)
arrivalTimeseriesWithImputedFutureValues <- ts(arrivalsImputedFuture, start=c(2014, 1), end=c(2015, 12), frequency=12)

seasonplot(arrivalTimeseriesWithImputedFutureValues);

BIG OPEN Question: how to estimate our Standard Deviation? I’ve calibrated it according to what my subjective upper and lower boundary of uncertainty is. Theory: https://en.wikipedia.org/wiki/Calibrated_probability_assessment more than 1,050,000 would mean growth, and less than 900,000 is a dramatic shutdown, both unlikely but possible.

standardDeviation= 0.18
weight= rnorm(N, mean=changeMoM, sd=standardDeviation)

Calculations

arrivalsNovember = arrivals2015[11]
arrivalsDecember = arrivalsNovember * weight


futureArrivals= arrivalsNovember + arrivalsDecember
refugeesAmount = reportedAmount + futureArrivals
plot(refugeesAmount)
abline(h=aMillion, col="red");
abline(h=mean(refugeesAmount), col="green");

overAmillion = refugeesAmount > 10^6
ProbabilityOverMillion= sum(overAmillion)/N

Probability of more than a Million Refuges

print(ProbabilityOverMillion * 100)

## [1] 15.24