Quantitative Finance: Compare Investment Strategies Using Backtesting

R Finance Series

R Codes and Results

#=====================================================
#   Reference: https://www.pfaffikus.de/books/jwex2/
#=====================================================

# Clear workspace: 
rm(list = ls())

# Import data: 

library(tidyverse)
read_csv("C:/Users/Admin/Documents/CafeF.HSX.Upto27.01.2021.csv") -> raw_data

# Select and rename some columns: 

raw_data %>% 
  select(1, 2, 6) -> raw_data_mini

names(raw_data_mini) <- c("symbol", "date_ymd", "close")

# Select some symbols: 

symbols_vn30 <- readxl::read_excel("C:/Users/Admin/Documents/data.xlsx", sheet = 1) %>% pull(2)

# Filter data belongs to VN30

library(lubridate)

my_date <- "2016-12-28"

raw_data_mini %>% 
  mutate(date_ymd = ymd(date_ymd)) %>%  
  filter(symbol %in% symbols_vn30, date_ymd >= ymd(my_date)) -> df_return_long

# Stocks that >= 1021 observations: 

df_return_long %>% 
  group_by(symbol) %>% 
  count(sort = TRUE) %>% 
  ungroup() %>% 
  filter(n >= 1021) %>% 
  pull(symbol) -> my_symbols

# Convert to wide form and claculate return: 

df_return_long %>% 
  filter(symbol %in% my_symbols) %>% 
  pivot_wider(names_from = symbol, values_from = close) %>% 
  slice(order(date_ymd)) %>% 
  mutate_if(is.numeric, function(x) {x / lag(x, n = 1L) - 1}) %>% 
  na.omit() -> data_wide_return

# Function calculates portfolio return: 

library(fPortfolio)

window_size <- 700

portfolio_return <- function(k_initial) {
  
  data_wide_return %>% slice((1 + k_initial):(window_size + k_initial)) -> train_data

  data_wide_return %>% slice(window_size + k_initial + 1) -> test_data

  # Prepare data for back-testing and portfolio optimization: 
  
  date_ymd <- train_data$date_ymd

  stocks_data <- train_data %>% select(-date_ymd)

  stocks_data %>% 
    timeSeries::timeSeries(charvec = as.character(date_ymd)) %>% 
    na.omit() -> stock_return 
  
  # Portfolio 1: 

  pspec <- portfolioSpec()

  gmv <-  pspec

  gmvpf <- minvariancePortfolio(data = stock_return, spec = pspec, constraints = "LongOnly")
  
  wGMV <- getWeights(gmvpf)

  test_data %>% 
    select(-date_ymd) %>% 
    as.vector() -> asset_returns

  p_return_minVar <- sum(asset_returns*wGMV) 
  
  # Portfolio 2: 
  
  cvar <- pspec

  setType(cvar) <- "CVaR"

  setAlpha(cvar) <- 0.1

  setSolver(cvar) <- "solveRglpk.CVAR"

  cvarpf <- minriskPortfolio(data = stock_return, spec = cvar, constraints = "LongOnly")
  
  wCVAR <- getWeights(cvarpf)
  
  p_return_minRisk <- sum(asset_returns*wCVAR) 

  test_data %>% 
    mutate(p_return_minVar = p_return_minVar, 
           p_return_minRisk = p_return_minRisk) -> port_return
  
  return(port_return)
  
}

# Use the function: 

n_days <- nrow(data_wide_return)

n_future_days <- n_days - window_size - 1 

lapply(0:n_future_days, portfolio_return) -> port_return_list

do.call("bind_rows", port_return_list) -> port_return_rolling

initial_investment <- 100

k <- nrow(port_return_rolling)

calculate_port_return_t <- function(t) {
  
  port_return_rolling$p_return_minVar[1:t] + 1 -> r_minVar
  
  port_return_rolling$p_return_minRisk[1:t] + 1 -> r_minRisk
  
  prod(r_minVar) -> r_minVar
  
  value_minVar <- r_minVar*initial_investment
  
  prod(r_minRisk) -> r_minRisk
  
  value_minRisk <- r_minRisk*initial_investment
  
  port_return_rolling %>% 
    slice(t) %>% 
    select(date_ymd, contains("return")) %>% 
    mutate(value_minRisk = value_minRisk, value_minVar = value_minVar) -> df_final
  
  return(df_final)
  
}

lapply(1:k, calculate_port_return_t) -> return_port_list

do.call("bind_rows", return_port_list) -> port_values


port_values %>% 
  select(date_ymd, contains("value")) %>% 
  pivot_longer(cols = c("value_minRisk", "value_minVar"), names_to = "type", values_to = "value") -> df_long


df_long %>% 
  ggplot(aes(date_ymd, value, color = type)) + 
  geom_line() + 
  theme(legend.title = element_blank()) + 
  labs(title = "Backtesting for comparing Daily-Value of Portfolio", x = NULL, y = "Investment Value") -> p

plotly::ggplotly(p)