The definition of H shape candlestick chart
Meaning: The stock price is going to rise.
t day: Close > Open
t-1 day: The close price is close to the open price.
t-2 day: Close > Open
The reference of code: https://bap.cm.nsysu.edu.tw/?page_id=446
Start To Program
Load the packages and historical stock data
library(quantmod)
library(dplyr)
library(RMySQL)# Query price historical data from database
# Connect to the database
StocksDB <- dbConnect(MySQL(), dbname = "Stocks(TW)", username="root", password="")# Query data of which dates are after 2015
data <- dbGetQuery(StocksDB ,"select * from History_Prices where date > '2015-01-01'")
colnames(data) <-
c("Date","High","Close","Volume","Low","Open","AdjClose","Id")# Load the industries of stocks
stock_cat <- read.csv('/Users/chenlyn/Desktop/Quantative Analyst/Project/Stock Price Database/Stocks_Categories.csv')stock_cat$Id <- as.character(stock_cat$Id)
Find the Buying Points which Meet the H Shape Definition
# Create the buying function
H_shape_buy <- function(data=data, StockId){
stock <- data %>% filter(Id == StockId)
stock$Date <- as.Date(stock$Date) # Remove volume == 0
stock <- stock %>% filter(Volume != 0)# Calculate candle lengths, t-1 and t-2 close and open
stock <- stock %>% mutate(
# t-1 day close
lagClose1 = lag(Close, 1),
# t-2 day close
lagClose2 = lag(Close, 2),
# t-1 day open
lagOpen1 = lag(Open, 1),
# t-2 day open
lagOpen2 = lag(Open, 2),
# t day - candle length
candleValue = abs(Close / Open-1),
# t-1 day - candle length
lagcandleValue1 = lag(candleValue, 1),
# t-2 dat - candle length
lagcandleValue2 = lag(candleValue, 2),
)
# Find H shape candlestick plot
fitStrategy <- stock %>% filter(
# t day Close > t day Open
Close > Open,
# t-2 day Close > t-2 day Open
lagClose2 > lagOpen2,
# t day - candle length > 1%
candleValue > 0.01,
# t-1 day - candle length < 0.5%
lagcandleValue1 < 0.005,
# t-2 day - candle length > 1%
lagcandleValue2 > 0.01
) %>% select(BuyDate = Date,BuyPrice = Close,Id = Id) # Return the stock data and buying points
buyStock <- list(Stock = stock, Buy = fitStrategy)
return(buyStock)
}
Find the Selling Points
# Create the finding selling points function
# Function parameters including different moving average functions and different periods to average over
sellStock <- function(Stock, MAday = 20, MAFUN = SMA){
sellStock <- Stock %>%
mutate(
# Calculate t day MA
MA = MAFUN(Close, MAday),
# Calculate t-1 day MA
lagMA = lag(MA, 1)) %>% # When close is less than MA, sell the stock
filter(
# t day Close < t day MA
Close < MA,
# t-1 day Close > t-1 day MA
lagClose1 > MA) %>%
select(SellDate = Date, SellPrice = Close)
return(sellStock)
}
Create the Trading Table
# Combine the buying points and selling points to create trading tablesTradingRecord <- function(fitStrategy, SellPoint){
TradingTable <- NULL
for (i in 1:nrow(fitStrategy)){
# buy the stock
inDate <- fitStrategy$BuyDate[i]
# sell the stock
outdate <- which(SellPoint$SellDate > inDate)[1]
# Foolproof mechanism - if the buying point is at the end of data period, it can't find the selling point.
if(length(outdate)>0){
# Create trading table
TradingTable <- rbind(TradingTable, cbind(fitStrategy[i,], SellPoint[outdate,]))
}
TradingTable <- TradingTable[complete.cases(TradingTable),]
}
return(TradingTable)
}
Draw The Plot to Verify
# Take the stock Id 2330 as the example
Buying_2330 <- H_shape_buy(data, 2330)
Selling_2330 <- sellStock(Buying_2330$Stock)
TradingTable <- TradingRecord(Buying_2330$Buy,Selling_2330)# Create plot function
PlotGraph <- function(stock, TradingTable,plotSample = 5){
stock <- stock %>%
mutate(
MA5 = SMA(Close, 5),
MA20 = SMA(Close, 20)) # Find the buy point and sell point in the trading table
buyDate <- TradingTable$BuyDate[plotSample]
sellDate <- TradingTable$SellDate[plotSample] # start date of the plot is 35 trading days before the buy point
match <- which(stock$Date == buyDate)- 35
plotStartDate <- stock$Date[ifelse(match <1, 1, match)] # end date of the plot is 35 trading days after the sell point
match <- which(stock$Date == sellDate)+35
plotEndDate <- stock$Date[ifelse(match > nrow(stock), nrow(stock), match)] # Select the data meeting the holding stock date range
plotData <- stock[which((stock$Date >= plotStartDate)&(stock$Date <= plotEndDate)),]
plotData <- plotData %>% select(Date:Open, MA5:MA20) # Find the buying point
plotData$buy <- rep(NA, nrow(plotData))
plotData$buy[which(plotData$Date == buyDate)] <- plotData$Open[which(plotData$Date == buyDate)]*0.97 # Ｆind the selling point
plotData$sell <- rep(NA, nrow(plotData))
plotData$sell[which(plotData$Date == sellDate)] <- plotData$Close[which(plotData$Date == sellDate)]*1.03 # Convert the data format into xts
plotData <- xts(plotData[,-1], order.by = plotData$Date) # Set the color of candlestick chart
myTheme <- chart_theme()
myTheme$col$dn.col <- c('firebrick3')
myTheme$col$up.col <- c('chartreuse3') # Draw the candlestick chart
pic <- chart_Series(x=plotData[,1:6], name = paste('2330', 'Technical Analysis Plot',sep =' '), theme = myTheme) # Add Volume
pic <- add_Vo() # Add line plot of 5 day moving average of price
pic <- add_TA(x=plotData$MA5, on=1, type= 'l', col = 'blue', lwd = 1.5) # Add line plot of 20 day moving average of price
pic <- add_TA(x = plotData$MA20, on = 1, type = 'l', col = 'orange', lwd = 1.5) # Mark the buying point
pic <- add_TA(x = plotData$buy, on = 1, type = 'p', col='red', pch = 2, cex = 5, lwd = 1.5) # Mark the selling point
pic <- add_TA(x = plotData$sell, on = 1, type = 'p', col = 'green', pch = 6, cex =5, lwd =1.5)
return(pic)
}# Call the plot function
PlotGraph(Buying_2330$Stock, TradingTable,2)
Create the Performance Metrics and Apply to All Taiwan Stock
# Calculate the performanceTradingPerformance <- function(TradingTable){ # Fee and taxes
buyCost <- 0.001425
sellCost <- 0.004425 TradingTable <- TradingTable %>% mutate(
netProfit = SellPrice*(1-sellCost)*1000 - BuyPrice*(1+buyCost)*1000, # the rate of investment(ROI)
ret = ((SellPrice*(1-sellCost)) / (BuyPrice*(1+buyCost)) -1)*100,
holdDays = as.numeric(SellDate - BuyDate),
profitPerDay = ((SellPrice*(1-sellCost))- (BuyPrice*(1+buyCost)))*1000 / holdDays
)
return(TradingTable)
}# Integrate all function into one function
H_shape_performance <- function(StockID, MAday=20, MAFUN = SMA){
tryCatch({
BuyIn <- H_shape_buy(data, StockID)
SellOut <- sellStock(BuyIn$Stock, MAday, MAFUN)
Trade <- TradingRecord(BuyIn$Buy, SellOut)
performance <- TradingPerformance(Trade)
return(performance)},
error = function(e){return(NULL)}
)
}# Apply to all Taiwan Stock and compare the different periods to average overStockID <- unique(data$Id)
H_shape_SMA5 <- do.call(rbind, lapply(StockID,H_shape_performance, MAday =5))
H_shape_SMA5$MAdays <- 5H_shape_SMA10 <- do.call(rbind, lapply(StockID,H_shape_performance, MAday =10))
H_shape_SMA10$MAdays <- 10H_shape_SMA15 <- do.call(rbind, lapply(StockID,H_shape_performance, MAday =15))
H_shape_SMA15$MAdays <- 15H_shape_SMA20 <- do.call(rbind, lapply(StockID,H_shape_performance, MAday =20))
H_shape_SMA20$MAdays <- 20H_shape_SMA40 <- do.call(rbind, lapply(StockID,H_shape_performance, MAday =40))
H_shape_SMA40$MAdays <- 40H_shape_SMA60 <- do.call(rbind, lapply(StockID,H_shape_performance, MAday =60))
H_shape_SMA60$MAdays <- 60H_shape_SMA80 <- do.call(rbind, lapply(StockID,H_shape_performance, MAday =80))
H_shape_SMA80$MAdays <- 80H_shape_SMA100 <- do.call(rbind, lapply(StockID,H_shape_performance, MAday =100))
H_shape_SMA100$MAdays <- 100H_shape_SMA120 <- do.call(rbind, lapply(StockID,H_shape_performance, MAday =120))
H_shape_SMA120$MAdays <- 120SMA_Stock <- rbind(H_shape_SMA5, H_shape_SMA10, H_shape_SMA15, H_shape_SMA20,
H_shape_SMA40, H_shape_SMA60, H_shape_SMA80, H_shape_SMA100, H_shape_SMA120)# Left join with industries information
SMA_Stock <- left_join(SMA_Stock, stock_cat) SMA_Stock <- SMA_Stock %>% mutate(Year = ifelse(BuyDate > '2017-01-01', 'After 2017', 'After 2015'))
The Performance of H shape candlestick chart
# The overall winRatio, ROI, EV(stock price), EROI(expected ROI)Overall_Performance <- SMA_Stock %>%
summarise(WinRatio = sum(ifelse(netProfit >0, 1, 0))/n(), Mean_netProfit = mean(netProfit),
Mean_ROI = mean(ret)*100, EV = WinRatio*Mean_netProfit, EROI = WinRatio*Mean_ROI)
The win ratio is quite disappointing.
Even puting different moving average functions, the candle length of H shape or volume on selling points can’t make the signicant improvement on the win ratio.
# Compare the performances on different periods of moving average over and industries Performance <- SMA_Stock %>% inner_join(stock_cat)%>% group_by(MainCategory, SubCategory, MAdays, Year) %>%
summarise(WinRatio = sum(ifelse(netProfit >0, 1, 0))/n(), Mean_netProfit = mean(netProfit),
Mean_ROI = mean(ret)*100, EV = WinRatio*Mean_netProfit, EROI = WinRatio*Mean_ROI, number = n())anova <- aov(WinRatio ~ MAdays+MainCategory+Year+
MAdays:MainCategory+MainCategory:Year, data = Performance)summary(anova)
- Selling points with different periods of moving average can make win ratios have significant differences of means.
- Win ratios show significant differences between different industies.
- Year and Maincategory have interaction effects. The rising trend of industries are different in years.
# Do pairwise t-test between different MAdays groups
pairwise.t.test(Performance$WinRatio, Performance$MAdays)Performance %>% group_by(MAdays) %>% summarise(WinRatio = mean(WinRatio))%>%
ggplot(aes(x = MAdays, y = WinRatio))+
geom_bar(stat="identity")
The selling points depending on 40 trading days’s average prices started to show significantly more win ratio than less moving average periods. And it’s more suitable holding days for short-term traders.
# Select the better stocks
# Applying selling points on 40 days' MA prices strategy
# Find the stocks whose ROI is better than 75% and win Ratio is better than 50% of the H shape trading in 2015~2016 and 2017~now separately
# The stock needs to meet the H shape requirements twice in both periods.
# Filter the stocks which perform well both in 2015~2016 and 2017~nowAfter_2015_Stock <- SMA_Stock %>% group_by(Id)%>% mutate(WinRatio = sum(ifelse(netProfit >0, 1, 0))/n()) %>%
filter(ret > 0, ret > quantile(SMA_Stock$ret, 0.75), MAdays == 40, Year == 'After 2015', WinRatio > quantile(Performance$WinRatio, 0.50)) %>%
group_by(Id) %>% count() %>% filter(n > 1) %>% select(Id)%>% distinct()After_2017_Stock <- SMA_Stock %>% group_by(Id)%>% mutate(WinRatio = sum(ifelse(netProfit >0, 1, 0))/n()) %>%
filter(ret > 0, ret > quantile(SMA_Stock$ret, 0.75), MAdays == 40, Year == 'After 2017', WinRatio > quantile(Performance$WinRatio, 0.50)) %>%
group_by(Id) %>% count() %>% filter(n > 1) %>% select(Id)%>% distinct()Better_Stock <- rbind(After_2015_Stock, After_2017_Stock) %>% group_by(Id) %>% count() %>% filter(n == 2)
Set Today’s Signal
# Query the stock price in recent 3 days from the database
Recent_Prices <- dbGetQuery(StocksDB ,"select Date, high, close, open, volume, low, AdjClose, Id
FROM (SELECT *,
@cur := IF(id=@id, @cur+1, 1) AS ROW,
@id := id
FROM History_Prices CROSS JOIN (SELECT @id := (SELECT MAX(Id) FROM History_Prices), @cur := 0) AS init
WHERE date > '2020-05'
ORDER BY History_Prices.id, Date DESC) As t
WHERE ROW < 4")
dbDisconnect(StocksDB)
colnames(Recent_Prices) <-
c("Date","High","Close","Volume","Low","Open","AdjClose","Id")# Filter out the data of better stock in H shape
Recent_Prices <- Recent_Prices %>% filter(Id %in% Better_Stock$Id)
Better_Stock_Id <- unique(Recent_Prices)# Today's signal - which meet the H shape requirements
Hshape_Buy_Signal <- lapply(Better_Stock_Id$Id,H_shape_buy, data = Recent_Prices)
Hshape_Buy_Signal <- do.call(rbind,sapply(temp1, `[`,'Buy'))
Summary
- H shape has about 35% win ratio overall.
- The industries performing well are varied by years.
- The win ratio performed better for longer period which the moving average prices ove.
Reference: Build Stock Price Database With R Website Crawling & MYSQL
If there is any suggestion, please feel free to leave comments.
