xgb.R

# Importing the Packages --------------------------------------------------

library(pacman)
p_load(dplyr, tidyr, readxl, quantmod, ModelMetrics, randomForest, rpart ,rpart.plot, 
       xts, lubridate, TTR, PerformanceAnalytics, ggplot2, dygraphs)

# Importing data ----------------------------------------------------------

safaricom <- read_excel("stock_prices.xlsx") %>% 
  mutate(Date = ymd(Date))

id <- safaricom$Date

safaricom2 <- safaricom %>% 
  select(-Date) %>% 
  xts(order.by = ymd(id))

head(safaricom)

# Descriptive ------------------------------------------------------------

data.frame(
  mean = format(sapply(safaricom[,2:ncol(safaricom)], mean), scientific = FALSE), 
  variance = format(sapply(safaricom[,2:ncol(safaricom)], var), scientific = TRUE),
  skewness = sapply(safaricom[,2:ncol(safaricom)], skewness),
  kurtosis = sapply(safaricom[,2:ncol(safaricom)], kurtosis)
)

# convert data to long format ---------------------------------------------

safaricom_long <- pivot_longer(
  safaricom[ ,1:5],
  cols = -Date,
  values_to = "price",
  names_to = "stock"
)


# Plotting ----------------------------------------------------------------

safaricom_long %>% 
  ggplot(aes(x  = Date)) +
  geom_line(aes(y = price, col = stock), linewidth = .7) +
  facet_wrap(~stock, ncol = 2) +
  theme_minimal() +
  theme(axis.line = element_line(colour = "black"),
        axis.ticks = element_line(colour = "black", size = 1),
        axis.text = element_text(colour = "black"),
        plot.title = element_text(colour = "black", hjust = .5)) +
  guides(col = "none") +
  labs(title = "SCOM Stock Prices")

# Technical Indicators ---------------------------------------------------------------

safaricom_returns <- safaricom2$Close

#MACD
MACD <- MACD(safaricom_returns)

# RSI

RSI <- RSI(safaricom_returns)

# ADX
ADX <- ADX( safaricom2[,1:4] )

# BBANDS
bband <- BBands(safaricom_returns)

# Stochastic Oscillator 
sc <- stoch(safaricom_returns)

# Average True Range
ATR <- ATR(safaricom2[,1:4])

# Commodity Channel Index
CCI <- CCI(safaricom_returns)

# Rate of Change
ROC <- ROC(safaricom_returns) 

# Money Flow Index (MFI)
MFI <- MFI(safaricom_returns, volume = safaricom2$Volume)

# On-Balance Volume (OBV)
OBV <- OBV(safaricom_returns, volume = safaricom2$Volume)

# Returns -----------------------------------------------------------------

safaricom_returns$Return <- CalculateReturns(safaricom_returns)
adf.test(na.omit(safaricom_returns$Return))

safaricom_returns <- safaricom_returns %>% 
  data.frame()

safaricom_returns$Direction <- case_when(as.numeric(safaricom_returns$Return) < 0 ~ "Down", 
                                         as.numeric(safaricom_returns$Return) >0 ~ "Up", 
                                         .default = "Down")
saf_var <- data.frame(MACD, RSI, ADX, bband, sc, ATR, CCI, ROC,
                      MFI, OBV,
                      row.names = row.names(safaricom_returns),
                      "Direction" = safaricom_returns$Direction)

returns_data <- safaricom_returns %>% 
  na.omit() %>% 
  mutate(Date = as.Date(row.names(na.omit(safaricom_returns))))


# Plotting the returns
return_plot <- returns_data %>% 
  ggplot((aes(x = Date))) +
  geom_line(aes(y = Return), col = "#8B008B") + #lightgoldenrod4, steelblue4, midnightblue, peru,coral4
  theme_minimal() +
  theme(axis.line = element_line(colour = "black"),
        axis.ticks = element_line(colour = "black", size = 1),
        axis.text = element_text(colour = "black"),
        plot.title = element_text(colour = "black", hjust = .5)) +
  guides(col = "none") +
  labs(title = "SCOM Stock Prices")
return_plot

# Plotting the direction using candlesticks

safaricom2 %>%
  as.data.frame() %>% 
  select(-c(Volume)) %>% 
  tail(n = 60) %>% 
  dygraph() %>%
  dyCandlestick()

# Feature Engineering -----------------------------------------------------

#saf_var$macd_indicator <- ifelse(saf_var$macd < saf_var$signal, "bearish", "bullish")
glimpse(saf_var)


# leading Directions -----------------------------------------------------

saf_var$Direction <- lead(saf_var$Direction)  
saf_var$Direction = as.factor(saf_var$Direction)
saf_var <- saf_var %>% 
  na.omit() %>% 
  data.frame() %>% 
  `row.names<-`(NULL)

# Splitting the data ------------------------------------------------------

size <- 0.8
sample_size <- floor(size * nrow(saf_var))
set.seed(123)
indicies <- sample(seq_len(nrow(saf_var)), size = sample_size)
train_data <- saf_var[indicies,]
test_data <- saf_var[-indicies,]

# 1. XGBOOST --------------------------------------------------------------

train_data$Direction <- ifelse(train_data$Direction == "Up", 1, 0)
test_data$Direction <- ifelse(test_data$Direction == "Up", 1, 0)
y_train = train_data$Direction


test_labels = 1:nrow(test_data)

xgb_train = xgb.DMatrix(as.matrix(train_data), label = y_train)
xgb_test = xgb.DMatrix(as.matrix(test_data))
xgb_params <- list(
  booster = "gbtree",
  eta = 0.01,
  max_depth = 8,
  gamma = 4,
  subsample = 0.75,
  colsample_bytree = 1,
  eval_metric = "auc",
  objective = "binary:logistic"
)

xgb_model <- xgb.train(
  params = xgb_params,
  data = xgb_train,
  nrounds = 5000,
  verbose = 1
)

test_data$Direction <- ifelse(test_data$Direction == 1, "Up", "Down") %>% 
  as.factor()

train_data$Direction <- ifelse(train_data$Direction == 1, "Up", "Down") %>% 
  as.factor()

xgb_preds_train <- predict(xgb_model, xgb_train, type = "response")
xgb_preds_train <- ifelse(xgb_preds_train <0.4, "Down", "Up") %>% 
  as.factor()
caret::confusionMatrix(train_data$Direction, xgb_preds_train)

xgb_preds <- predict(xgb_model, xgb_test, type = "response")
xgb_preds <- ifelse(xgb_preds <0.4, "Down", "Up") %>% 
  as.factor()
caret::confusionMatrix(test_data$Direction, xgb_preds)