#************************************************************ 
# Smoothing and Decompostion of Time Series R 
#------------------------------------------------------------


# Financial Market Forecasting model
rm(list=ls())

library(quantmod)
library("forecast")

myForc <- function(x,f1) {
  x_m <-  meanf(x, h=f1)
  # Naive 
  x_naive = naive(x,level=c(90,95),h=f1) 
  # Random Walk with Drift
  x_rd = rwf(x,drift=T,level=c(90,95),h=f1) 
  
  # Holt Exponential Smooth Method
  x_h <- holt(x,h=f1, exponential = FALSE)
  
  # Holt-Winters Smooth Method
  x_hwest <- HoltWinters(x,gamma=FALSE)
  x_hw <- forecast(x_hwest,h=f1)
  
  # Neural Network Method
  x_nnest <- nnetar(x)
  x_nn <- forecast(x_nnest,h=f1)
  
  out_test <- rbind(accuracy(x_naive),accuracy(x_rd),accuracy(x_h),accuracy(x_hw),accuracy(x_nn))
  row.names(out_test) <- c("Naive","Random Walk","Holt Exp","Holt_Winters","Neural Net")
  return(out_test)  
}

# Here is the unemployment rate without seasonal adjustment
getSymbols("UNRATENSA", src = "FRED")
unemp = UNRATENSA
chartSeries(unemp, theme="white")
unempts <- ts( unemp, start=1948-1-1, frequency=12)
unempcomp <- (decompose(unempts))
plot(unempcomp)
forc_unemp <- myForc(unemp,20)
forc_unemp

# Random Number Cases
x_random = rnorm(1000,1,1)
x_train <- ts(x_random,start=1,   end=990, frequency=1)
x_test <- ts(x_random,start=991, end=1000, frequency=1)
ts.plot(x_random)

forc_x_in <- myForc(x_train,10)
forc_x_in

f1 = 10
# Holt Exponential Smooth Method
x_h <- holt(x_train,h=f1, exponential = FALSE)
x_hffit <- predict(x_h,x_test)
e_hfe <- x_ts - x_hffit$mean[1:f1]
all_error <- cbind(x_test,x_hffit$mean[1:f1],e_hfe)
all_error
ts.plot(x_test,e_hfe, col=c("blue","red"))

# Holt-Winters Exponential Smooth Method
x_hw <-  HoltWinters(x_train,gamma=FALSE)
x_hwffit <- predict(x_hw,x_test)
e_hwfe <- x_test - x_hwffit$mean[1:f1]
ts.plot(x_test,e_hfe,e_hwfe, col=c("blue","yellow","red"))