ARIMAtestForElephant.R

search <- read.csv("latest.search.csv", stringsAsFactors = FALSE)
openstack.month <- "2009-09"
search <- search[which(search$Month == openstack.month):nrow(search),]

search$austin <- 0
austin.month <- which(search$Month == "2016-04")
search[austin.month:nrow(search),]$austin <- 1

# test for stationary data
library(tseries)
adf.test(search$cloud.tco...Worldwide., alternative = "stationary")

library(ggplot2)
library(ggthemes)

library(forecast)
arimax <- auto.arima(search$cloud.tco...Worldwide.,xreg=data.frame(blog=search$blog), stepwise=FALSE, parallel = TRUE)
p<-(1-pnorm(abs(arimax$coef)/sqrt(diag(arimax$var.coef))))*2
print(p)  
months <- sum(search$austin)

# 12 months
(12*arimax$coef["blog"])/sum(search[0:(11)+austin.month,]$cloud.tco...Worldwide.)
# all
(months*arimax$coef["blog"])/sum(search[0:(months-1)+austin.month,]$cloud.tco...Worldwide.)

awith.df <- data.frame(Month = search$Month, prediction=arimax$coef["intercept"]+arimax$coef["blog"]*search$blog)

png("CloudTCOSearchVolumeMST.png", width=1600,height=750,res=150)
ggplot(search, aes(y=cloud.tco...Worldwide.,x=Month,group=1)) +
  geom_line(data=awith.df, aes(x=Month, y=prediction)) +
  geom_line(color="gray50") +
  geom_vline(xintercept=which(search$Month=="2016-01"), color="red") +
  theme_tufte(base_family="Overpass") + 
  labs(y='"cloud tco" search volume on Google',x="") +
  theme(axis.text.x = element_text(angle=90, hjust=1)) 
dev.off()

summary(lm(cloud.tco...Worldwide.~austin,data=search))