IHSG Prediction using Time Series
Ricky Lie Jaya
2021-04-18
Intro
knitr::include_graphics("ihsg.jpg")
Rmd kali ini akan memprediksi harga penutupan ihsg menggunakan algoritma time series yaitu dengan model arima dengan menggunakan variable variable tambahan seperti harga emas, minyak dan lain lain
Import Library and Dataset
library(tidyverse)
library(tidyquant)
library(priceR)
library(Quandl)
library(zoo)
library(forecast)
library(lubridate)
library(padr)
library(tseries)
library(randomForest)gold <- Quandl("LBMA/GOLD", collapse = "daily")
currency <- historical_exchange_rates("USD", to = "IDR", start_date = "2020-01-01", end_date = "2021-02-28")## For full currency exchange rate API documentation visit:
## https://exchangerate.host/#/#docs
## (this message will only appear once per session)ihsg <- tq_get("^JKSE")
dowjones <- tq_get("^DJI")
oil <- Quandl("CHRIS/CME_CL1", collapse = "daily")
shanghai <- tq_get("000001.SS")Data Pre Processing
ihsg <- ihsg %>%
select(c(date, open, close)) %>%
mutate(date = ymd(date)) %>%
filter(date >= "2020-01-01",
date <= "2021-02-26")
currency <- currency %>%
rename(USDIDR = one_USD_equivalent_to_x_IDR)
gold <- gold %>%
select(c(Date, `USD (AM)`)) %>%
rename(Goldprices = `USD (AM)`,
date = Date) %>%
mutate(date = ymd(date)) %>%
filter(date >= "2020-01-01",
date <= "2021-02-28")
dowjones <- dowjones %>%
select(date, adjusted) %>%
rename(dowjones = adjusted) %>%
mutate(date = ymd(date)) %>%
filter(date >= "2020-01-01",
date <= "2021-02-28")
oil <- oil %>%
select(Date, Settle) %>%
rename(date = Date,
OilPrices = Settle) %>%
mutate(date = ymd(date)) %>%
filter(date >= "2020-01-01",
date <= "2021-02-28")
shanghai <- shanghai %>%
select(date, adjusted) %>%
rename(shanghai = adjusted) %>%
mutate(date = ymd(date)) %>%
filter(date >= "2020-01-01",
date <= "2021-02-28")a <- left_join(ihsg, currency, "date")
b <- left_join(a, gold, "date")
c <- left_join(b, dowjones, "date")
d <- left_join(c, shanghai, "date")
dataset <- left_join(d, oil, "date")Add variable Inflasi and BI Rate
dataset$Inflasi <- month(dataset$date)
dataset[1:22, ]$Inflasi <- 2.68
dataset[23:42, ]$Inflasi <- 2.98
dataset[43:63, ]$Inflasi <- 2.96
dataset[64:84, ]$Inflasi <- 2.67
dataset[85:100, ]$Inflasi <- 2.19
dataset[101:121, ]$Inflasi <- 1.96
dataset[122:143, ]$Inflasi <- 1.54
dataset[144:161, ]$Inflasi <- 1.32
dataset[162:183, ]$Inflasi <- 1.42
dataset[184:202, ]$Inflasi <- 1.44
dataset[203:223, ]$Inflasi <- 1.59
dataset[224:242, ]$Inflasi <- 1.68
dataset[243:262, ]$Inflasi <- 1.55
dataset[263:281, ]$Inflasi <- 1.38dataset[1:22, "BiRate"] <- 5
dataset[23:42, "BiRate"] <- 4.75
dataset[43:63, "BiRate"] <- 4.50
dataset[64:84, "BiRate"] <- 4.50
dataset[85:100, "BiRate"] <- 4.50
dataset[101:121, "BiRate"] <- 4.25
dataset[122:143, "BiRate"] <- 4
dataset[144:161, "BiRate"] <- 4
dataset[162:183, "BiRate"] <- 4
dataset[184:202, "BiRate"] <- 4
dataset[203:223, "BiRate"] <- 3.75
dataset[224:242, "BiRate"] <- 3.75
dataset[243:262, "BiRate"] <- 3.75
dataset[263:281, "BiRate"] <- 3.50Dataset Full
rmarkdown::paged_table(dataset)Check If There is Variable that didnt influence
model1 <- lm(close~USDIDR + Goldprices + dowjones + OilPrices + shanghai+Inflasi+BiRate, data = dataset)
summary(model1)##
## Call:
## lm(formula = close ~ USDIDR + Goldprices + dowjones + OilPrices +
## shanghai + Inflasi + BiRate, data = dataset)
##
## Residuals:
## Min 1Q Median 3Q Max
## -375.69 -95.49 9.89 88.33 406.13
##
## Coefficients:
## Estimate Std. Error t value Pr(>|t|)
## (Intercept) -2.546e+03 5.385e+02 -4.727 3.86e-06 ***
## USDIDR -3.765e-03 9.948e-03 -0.378 0.705
## Goldprices -8.230e-01 1.711e-01 -4.809 2.67e-06 ***
## dowjones 1.425e-01 7.228e-03 19.717 < 2e-16 ***
## OilPrices 8.453e+00 1.544e+00 5.475 1.09e-07 ***
## shanghai 1.271e+00 1.024e-01 12.408 < 2e-16 ***
## Inflasi 4.906e+02 4.532e+01 10.825 < 2e-16 ***
## BiRate 3.821e+01 5.170e+01 0.739 0.461
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Residual standard error: 138.1 on 242 degrees of freedom
## (31 observations deleted due to missingness)
## Multiple R-squared: 0.9501, Adjusted R-squared: 0.9486
## F-statistic: 657.9 on 7 and 242 DF, p-value: < 2.2e-16Convert to Time Series
ts_dataset <- dataset %>%
select(-date) %>%
na.approx() %>%
msts(seasonal.periods = c(7, 7*4))ts_dataset [ , "close"] %>%
autoplot()
Cross Validation
test <- tail(ts_dataset, 30)
train <- head(ts_dataset, length(ts_dataset) - 30)Modeling
modelarima2 <- stlm(train[ , "close"],
method = "arima",
xreg = train[ , c("open", "USDIDR", "Goldprices", "dowjones", "OilPrices", "BiRate", "Inflasi", "shanghai")])Forecasting and Evaluating
forecastarima2 <- forecast(modelarima2, h = 30, xreg = test[ , c("open", "USDIDR", "Goldprices", "dowjones", "OilPrices", "BiRate", "Inflasi", "shanghai")])accuracy(forecastarima2$mean, as.numeric(test[, "close"]))## ME RMSE MAE MPE MAPE
## Test set 15.06163 80.28048 66.05978 0.2383274 1.063973Plot Model arima
z <- as.data.frame(forecastarima2)
y <- as.data.frame(test)
compare <- cbind(forecast = z$`Point Forecast`, real = y$close)
compare <- as.data.frame(compare)
compare$date <- dataset$date %>%
tail(30)
plot <- compare %>%
ggplot(aes(x = date)) +
geom_line(aes(y = forecast), colour = "red", size = 2) +
geom_line(aes(y = real), colour = "green", size = 2) +
labs(x = "Date",
y = NULL,
title = "Comparison actual vs prediction") +
theme_dark()
plot
Green = Aktual