#Leandro Guerra
#www.outspokenmarket.com
#21/07/2018

#Loading Libraries
library(quantmod)
library(ggplot2)
library(data.table)
library(gridExtra)
library(fTrading)
library(TTR)
library(dplyr)


#Loading Data
EURUSD <- fread("D:/Projetos/LBI/Outspoken Market/EURUSD_H1_Hour.csv", sep = ";",header=T)
EURUSD <- as.data.frame(EURUSD)

head(EURUSD)

#Manipulating date to get hours and years
names(EURUSD)[1] <- "Date"
EURUSD$Hours <- substr(as.character(EURUSD$Date), 12, 13)
EURUSD$Hours <- as.numeric(EURUSD$Hours)
EURUSD$Year <- substr(as.character(EURUSD$Date), 7, 10)
head(EURUSD,5)

#Creating more info
EURUSD$STD <- runSD(EURUSD$Close,15)                                          # I will explain it later on
EURUSD$RSI <- RSI(EURUSD$Close)                                              # I will explain it later on
EURUSD$CCI <- CCI(EURUSD[,c("High","Low","Close")])
EURUSD$CLV <- CLV(EURUSD[,c("High","Low","Close")])
EURUSD$VHF <- VHF(EURUSD[,"Close"])

EURUSD <- na.omit(EURUSD)
head(EURUSD,5)

#Evaluating the variation in the number os pips per hour, from 2015 to 20/07/2018 and creating the plot
AcumHour <- data.frame(Pips=0, Hours=seq(0,23),stringsAsFactors=FALSE)
for(i in 0:23){
  AcumHour$Pips[i+1] = 100*sum(EURUSD[EURUSD$Hours==(i),]$Pips)
}

hist <- ggplot(data=AcumHour, aes(x=Hours, y=Pips)) +
          geom_bar(stat="identity") + ggtitle("Outspoken Market - Test of Hours Strategy") +
          xlab("") + ylab("2015 to 2018")

#2015
EUR2015 <- EURUSD[EURUSD$Year==2015,]

AcumHour <- data.frame(Pips=0,Hours=seq(0,23),stringsAsFactors=FALSE)

for(i in 0:23){
  AcumHour$Pips[i+1] = 100*sum(EUR2015[EUR2015$Hours==(i),]$Pips)
}

hist2015 <- ggplot(data=AcumHour, aes(x=Hours, y=Pips)) +
              geom_bar(stat="identity") +
              xlab("") + ylab("2015")

#2016
EUR2016 <- EURUSD[EURUSD$Year==2016,]

AcumHour <- data.frame(Pips=0,Hours=seq(0,23),stringsAsFactors=FALSE)

for(i in 0:23){
  AcumHour$Pips[i+1] = 100*sum(EUR2016[EUR2016$Hours==(i),]$Pips)
}

hist2016 <- ggplot(data=AcumHour, aes(x=Hours, y=Pips)) +
            geom_bar(stat="identity") +
            xlab("") + ylab("2016") 

#2017
EUR2017 <- EURUSD[EURUSD$Year==2017,]

AcumHour <- data.frame(Pips=0,Hours=seq(0,23),stringsAsFactors=FALSE)

for(i in 0:23){
  AcumHour$Pips[i+1] = 100*sum(EUR2017[EUR2017$Hours==(i),]$Pips)
}

hist2017<-ggplot(data=AcumHour, aes(x=Hours, y=Pips)) +
          geom_bar(stat="identity") +
          xlab("") + ylab("2017")

#2018
EUR2018 <- EURUSD[EURUSD$Year==2018,]

AcumHour <- data.frame(Pips=0,Hours=seq(0,23),stringsAsFactors=FALSE)

for(i in 0:23){
  AcumHour$Pips[i+1] = 100*sum(EUR2018[EUR2018$Hours==(i),]$Pips)
}

hist2018 <- ggplot(data=AcumHour, aes(x=Hours, y=Pips)) +
          geom_bar(stat="identity") +
          xlab("Hour GMT +1") + ylab("2018")

#Ploting the results
grid.arrange(hist,hist2015, hist2016, hist2017, hist2018, nrow = 5, ncol=1)

#There is a relevant negative accumulation of pips in the hours 8 and 13 and a positive one in the hours 18 and 19
#Lets create a simple and ilustrative strategy on it where:
    # Go short at 8am and 13 (1pm) GMT + 1
    # Go long at 18 (6pm) and 19 (7pm) GMT + 1

#Creating the Hours Strategy
pips_hours <- subset(EURUSD,
                   (EURUSD$Hours == 23) |      #sell
                     (EURUSD$Hours == 13)|    #sell
                     (EURUSD$Hours == 18) |   #buy
                     (EURUSD$Hours == 19)    #buy
                     )

#Adjusting the pips signal when the hours are to selling
pips_hours$Pips <- ifelse(pips_hours$Hours == "23",-1*pips_hours$Pips ,pips_hours$Pips )
pips_hours$Pips <- ifelse(pips_hours$Hours == "13",-1*pips_hours$Pips ,pips_hours$Pips )
cumulative_buy_pips <- cumsum(pips_hours$Pips)


#Evaluating strategy KPIs

#KPIs
Bull_wins_PIPS <- sum(pips_hours$Pips > 0)
Bull_losses_PIPS <- sum(pips_hours$Pips <= 0)
Bull_totalTrades_PIPS <- Bull_wins_PIPS + Bull_losses_PIPS
Bull_AVG_Win_PIPS <- mean(pips_hours[10][which(pips_hours$Pips > 0),])
Bull_AVG_Loss_PIPS <- -mean(pips_hours[10][which(pips_hours$Pips < 0),])
Bull_WL_Ratio_PIPS <- Bull_wins_PIPS/Bull_totalTrades_PIPS
Bull_EV_PIPS <- Bull_AVG_Win_PIPS*Bull_WL_Ratio_PIPS - Bull_AVG_Loss_PIPS*(1-Bull_WL_Ratio_PIPS)
Bull_PF_PIPS <- -sum(pips_hours[10][which(pips_hours$Pips > 0),])/sum(pips_hours[10][which(pips_hours$Pips < 0),])

SHARPE <- mean(pips_hours$Pips, na.rm = T)/sd(pips_hours$Pips, na.rm = T)
print ("Sharpe Ratio")
SHARPE
print ("W/L Ration")
Bull_WL_Ratio_PIPS
print ("Profit Factor")
Bull_PF_PIPS

#DrawDown PIPS
Bull_maxDD_PIPS <- maxDrawDown(cumulative_buy_pips)$maxdrawdown[1]
Bull_maxDD_duration_PIPS <-maxDrawDown(cumulative_buy_pips)$to - maxDrawDown(cumulative_buy_pips)$from[1]
print ("Max DD in Pips")
Bull_maxDD_PIPS
print ("Max DD Duration")
Bull_maxDD_duration_PIPS

#Ploting Strategy Results - Only to illustrate the results. The strategy lacks of consistency verifications
df <- data.frame(seq.int(dim(pips_hours)[1]),cumulative_buy_pips)
g <- ggplot(df, aes(seq.int(dim(pips_hours)[1])))
g <- g + geom_line(aes(y=cumulative_buy_pips,linetype="Pips"), colour="blue")
g <- g + ggtitle("Outspoken Market - Test of Hours Strategy")+labs(x="Test timeframe: 2015 - 2018",
                                                                                      y="Total Pips") 
g <- g + annotate("text", x = dim(pips_hours)[1]-500, y = tail(cumulative_buy_pips,1)-45, label = paste(round(tail(cumulative_buy_pips,1),2)))
g <- g + theme(legend.position=c(.24, .95), legend.justification = c(1, 1),
               legend.background = element_rect(color = "black",fill = "grey90", size = 1, linetype = "solid")) + scale_shape_discrete(name="")
g <- g + labs(linetype="Hours Strategy")
g



#Don't forget: Strategies may look good until you add operational costs

#Simulating 0.5 and 1 pips of operational costs

costs1 <- 0.5
costs2 <- 1
pips_hours$Pips2 <- pips_hours$Pips-costs2
pips_hours$Pips <- pips_hours$Pips-costs1
cumulative_buy_pips <- cumsum(pips_hours$Pips)
cumulative_buy_pips2 <- cumsum(pips_hours$Pips2)

#Ploting Strategy Results not with operational costs - All is different now
df <- data.frame(seq.int(dim(pips_hours)[1]),cumulative_buy_pips)
g <- ggplot(df, aes(seq.int(dim(pips_hours)[1])))
g <- g + geom_line(aes(y=cumulative_buy_pips,linetype="0.5 Pips"), colour="blue")
g <- g + geom_line(aes(y=cumulative_buy_pips2,linetype="1.0 Pips"), colour="red")
g <- g + ggtitle("Outspoken Market - Test of Hours Strategy")+labs(x="Test timeframe: 2015 - 2018",
                                                                                      y="Total Pips") 
g <- g + annotate("text", x = dim(pips_hours)[1]-500, y = tail(cumulative_buy_pips,1)-45, label = paste(round(tail(cumulative_buy_pips,1),2)))
g <- g + annotate("text", x = dim(pips_hours)[1]-500, y = tail(cumulative_buy_pips2,1)-45, label = paste(round(tail(cumulative_buy_pips2,1),2)))
g <- g + theme(legend.position=c(.30, .95), legend.justification = c(1, 1),
               legend.background = element_rect(color = "black",fill = "grey90", size = 1, linetype = "solid")) + scale_shape_discrete(name="")
g <- g + labs(linetype="With operational Costs")
g



#How could we improve the strategy and make it profitable even with operational costs?
#Remember the elements Pip_Shift and STD that I mentioned in the begining?
#Let's use them as target and predictiors for a simple linear regression
set.seed(2016)
#Traning set - 20015 to 2017 data
pips_hours_traning <- subset(EURUSD,EURUSD$Year < 2018) 
#Test set - 2018
pips_hours_test <- subset(EURUSD,EURUSD$Year == 2018) 

linear1 = lm(Pips_Shift ~ STD + VHF + CCI + VHF*Hours + STD*RSI + Hours*STD + RSI*CCI*Hours , 
             data = pips_hours_traning)
summary(linear1)

#You can see that the linear regression is not good. However, could it still be profitable?
#Let's apply it to the data and find out
#In the end only the RSI variable has a decent significance level. In theory you can remove the other variables
#I will keep all for the simplicity of the example. Please try it removing and adding more variables :)

#Applying the linear regression to the data 

pips_hours_traning$linear1 = predict(linear1,pips_hours_traning)
pips_hours_test$linear1 = predict(linear1,pips_hours_test)

#Now let's not only run the regression, but discover the most profitable quantiles
q = quantile(pips_hours_traning$linear1, seq(0, 1, by = 0.05))
pips_hours_traning$qlinear <- cut(pips_hours_traning$linear1, q, include.lowest=TRUE)


Pips_qlinear <- aggregate(Pips_Shift ~ qlinear, data = pips_hours_traning, FUN = sum)
Pips_qlinear


#Creating the Linear Regression Strategy
pips_linear <- subset(pips_hours_test,
                   (pips_hours_test$linear1 <=  0) |      #sell
                     (pips_hours_test$linear1 > 0)    #buy
                     )

#Adjusting the pips signal when the hours are to selling
pips_linear$Pips_Shift <- ifelse(pips_linear$linear1 <=  0,-1*pips_linear$Pips_Shift ,pips_linear$Pips_Shift )
#Setting a Stop
pips_linear$Pips_Shift <- ifelse(pips_linear$Pips_Shift <=  -40,-41 ,pips_linear$Pips_Shift )
cumulative_buy_pips <- cumsum(pips_linear$Pips_Shift)
names(pips_linear)

#Evaluating the Linear Regressoin Strategy KPIs

#KPIs
Bull_wins_PIPS <- sum(pips_linear$Pips_Shift > 0)
Bull_losses_PIPS <- sum(pips_linear$Pips_Shift <= 0)
Bull_totalTrades_PIPS <- Bull_wins_PIPS + Bull_losses_PIPS
Bull_AVG_Win_PIPS <- mean(pips_linear[7][which(pips_linear$Pips_Shift > 0),])
Bull_AVG_Loss_PIPS <- -mean(pips_linear[7][which(pips_linear$Pips_Shift < 0),])
Bull_WL_Ratio_PIPS <- Bull_wins_PIPS/Bull_totalTrades_PIPS
Bull_EV_PIPS <- Bull_AVG_Win_PIPS*Bull_WL_Ratio_PIPS - Bull_AVG_Loss_PIPS*(1-Bull_WL_Ratio_PIPS)
Bull_PF_PIPS <- -sum(pips_linear[7][which(pips_linear$Pips_Shift > 0),])/sum(pips_linear[11][which(pips_linear$Pips_Shift < 0),])

SHARPE <- mean(pips_linear$Pips_Shift, na.rm = T)/sd(pips_linear$Pips_Shift, na.rm = T)
print ("Sharpe Ratio")
SHARPE
print ("W/L Ration")
Bull_WL_Ratio_PIPS
print ("Profit Factor")
Bull_PF_PIPS

#DrawDown PIPS
Bull_maxDD_PIPS <- maxDrawDown(cumulative_buy_pips)$maxdrawdown[1]
Bull_maxDD_duration_PIPS <-maxDrawDown(cumulative_buy_pips)$to - maxDrawDown(cumulative_buy_pips)$from[1]
print ("Max DD in Pips")
Bull_maxDD_PIPS
print ("Max DD Duration")
Bull_maxDD_duration_PIPS

#Ploting Linear Regression Strategy Results - Only to illustrate the results. The strategy lacks of consistency verifications
df <- data.frame(seq.int(dim(pips_linear)[1]),cumulative_buy_pips)
g <- ggplot(df, aes(seq.int(dim(pips_linear)[1])))
g <- g + geom_line(aes(y=cumulative_buy_pips,linetype="Pips"), colour="blue")
g <- g + ggtitle("Outspoken Market - Linear Regression in Hours Strategy")+labs(x="Test timeframe: 2018",
                                                                                      y="Total Pips") 
g <- g + annotate("text", x = dim(pips_linear)[1]-10, y = tail(cumulative_buy_pips,1)-45, label = paste(round(tail(cumulative_buy_pips,1),2)))
g <- g + theme(legend.position=c(.24, .95), legend.justification = c(1, 1),
               legend.background = element_rect(color = "black",fill = "grey90", size = 1, linetype = "solid")) + scale_shape_discrete(name="")
g <- g + labs(linetype="Hours Strategy")
g


#Don't forget: Strategies may look good untill you add operational costs

#Simulating 0.5 and 1 pips of operational costs

costs1 <- 0.5
costs2 <- 1
pips_linear$Pips_Shift2 <- pips_linear$Pips_Shift-costs2
pips_linear$Pips_Shift <- pips_linear$Pips_Shift-costs1
cumulative_buy_pips <- cumsum(pips_linear$Pips_Shift)
cumulative_buy_pips2 <- cumsum(pips_linear$Pips_Shift2)

#Ploting Strategy Results not with operational costs - All is different now
df <- data.frame(seq.int(dim(pips_linear)[1]),cumulative_buy_pips)
g <- ggplot(df, aes(seq.int(dim(pips_linear)[1])))
g <- g + geom_line(aes(y=cumulative_buy_pips,linetype="0.5 Pips"), colour="blue")
g <- g + geom_line(aes(y=cumulative_buy_pips2,linetype="1.0 Pips"), colour="red")
g <- g + ggtitle("Outspoken Market - Linear Regression in Hours Strategy")+labs(x="Test timeframe: 2018",
                                                                                      y="Total Pips") 
g <- g + annotate("text", x = dim(pips_linear)[1]-10, y = tail(cumulative_buy_pips,1)-45, label = paste(round(tail(cumulative_buy_pips,1),2)))
g <- g + annotate("text", x = dim(pips_linear)[1]-10, y = tail(cumulative_buy_pips2,1)-45, label = paste(round(tail(cumulative_buy_pips2,1),2)))
g <- g + theme(legend.position=c(.30, .95), legend.justification = c(1, 1),
               legend.background = element_rect(color = "black",fill = "grey90", size = 1, linetype = "solid")) + scale_shape_discrete(name="")
g <- g + labs(linetype="With operational Costs")
g


#Leandro Guerra - www.outspokenmarket.com