Daniel RG

11/4/2020

Series: United States Industrial Production (2012 = 100, NSA)

Source: FRED, Federal Reserve Bank of St. Louis

library(fpp2)
library(ggplot2)
library(ggthemes)
library(tidyverse)
library(kableExtra)

usip <- read.csv("F:/MSAE/2020fall_PredictiveAnalytics/Discussions/USIP.csv")
usip <- usip[order(usip$DATE),]


ip <- ts(usip[,2],start=c(1980,1),end=c(2020,9),frequency=12)

plot1 <- autoplot(ip) + geom_line() +
  xlab("Year") + ylab("") + ggtitle("US Industrial Production (2012 = 100)") + theme_classic()

plot1

Decompositions

decadd <- decompose(ip, type = "additive")
decmul <- decompose(ip, type = "multiplicative")

autoplot(decadd)

autoplot(decmul)

The multiplicative model has lower remainder values and is likely to work better for projections. This is likely due to increased seasonal variation towards the end of the series and increasing values.

Predictions

ip1 <- stl(ip, t.window=15, s.window="periodic", robust=TRUE)

ipf1 <- seasadj(ip1)
autoplot(naive(ipf1, h=24))

ipf2 <- forecast(ip1, method="naive", h=24)
autoplot(ipf2)

ipf3 <- stlf(ip, method="rwdrift", h=24, lambda=0)
autoplot(ipf3)

# table1 <- accuracy(ipf1)
# kable(table1, caption = "Naive") + kable_classic()

table2 <- accuracy(ipf2)
kable(round(table2, 2), caption = "Random Walk", full_width = F)
Random Walk
ME RMSE MAE MPE MAPE MASE ACF1
Training set 0.1 1.12 0.7 0.12 0.88 0.23 0.13 
table3 <- accuracy(ipf3)
kable(round(table3, 2), caption = "Random Walk with Drift", full_width = F)
Random Walk with Drift
ME RMSE MAE MPE MAPE MASE ACF1
Training set -0.01 0.88 0.5 0 0.61 0.16 0.23