Low Level of evolution for ITX stock at Madrid stockmarket

Introduction

This work allow us to keep the focus on the stockmarket operation and how to transform it properly:

# Load data
suppressPackageStartupMessages(library(googleVis))
suppressPackageStartupMessages(library(xtable))
suppressPackageStartupMessages(library(quantmod))
suppressPackageStartupMessages(library(TTR))
suppressPackageStartupMessages(library(segmented))
suppressPackageStartupMessages(library(caret))
suppressPackageStartupMessages(library(caTools))
suppressPackageStartupMessages(library(mlbench))
suppressPackageStartupMessages(library(xlsx))
suppressPackageStartupMessages(library(stringr))
suppressPackageStartupMessages(library(zoo))
suppressPackageStartupMessages(library(gbm))
suppressPackageStartupMessages(library(plyr))
suppressPackageStartupMessages(library(doMC))
suppressPackageStartupMessages(library(multicore))
suppressPackageStartupMessages(library(parallel))
suppressPackageStartupMessages(library(foreach))
suppressPackageStartupMessages(library(compiler))
#
consolide <- function(x, xts) {
    xts <- try.xts(xts, error = FALSE)
    dftmp <- xts[x, ]
    vol <- sum(dftmp$Stocks)
    if (vol == 0) {
        rs <- data.frame(Price = 0, Volume = 0, Open = 0, Close = 0, Max = 0, 
            Min = 0)
        return(rs)
    }
    pri <- sum(dftmp$Price * dftmp$Stocks)/vol
    opn <- dftmp$Price[1]
    cls <- dftmp$Price[nrow(dftmp)]
    mx <- max(dftmp$Price)
    mn <- min(dftmp$Price)
    rs <- data.frame(Price = pri, Volume = vol, Open = as.numeric(opn), Close = as.numeric(cls), 
        Max = mx, Min = mn)
    return(rs)
}
cconsolide <- cmpfun(consolide)
#
to.ohlc <- function(ts, rgt, nam = "TS") {
    zt <- zoo(ts[, -1], order.by = ts[, 1])
    zoot <- as.xts(zt, order.by = as.POSIXct(rownames(zt)))
    sf <- cut(index(zoot), breaks = rgt, labels = rgt[-1])
    blks <- split(index(zoot), sf)
    blk <- ldply(blks, cconsolide, zoot)
    zt <- zoo(blk[, -1], blk[, 1])
    zoot <- as.xts(zt, order.by = as.POSIXct(index(zt)))
    ohlc <- as.quantmod.OHLC(zoot, name = nam, col.names = c("Adjusted", "Volume", 
        "Open", "Close", "High", "Low"))
    return(ohlc)
}
cto.ohlc <- cmpfun(to.ohlc)
#
prepare <- function(itx) {
    itx_p <- tapply(itx[, 3], itx[, 2], mean)
    ts_itx <- cbind(names(itx_p), as.numeric(itx_p), tapply(itx[, 3], itx[, 
        2], length), tapply(itx[, 4], itx[, 2], sum))
    ts_itx <- as.data.frame(ts_itx)
    ts_itx[, -1] <- apply(ts_itx[, -1], 2, function(x) as.numeric(as.character(x)))
    colnames(ts_itx) <- c("Date", "Price", "Nex", "Stocks")
    return(ts_itx)
}
cprepare <- cmpfun(prepare)
#
setwd("~/git/HFT/")
if (file.exists("HFT_bd.RData")) {
    load("HFT_bd.RData")
} else {
    itx <- read.xlsx("ITX_SAN_TEL270412.xls", 1, )
    itx[, 2] <- as.POSIXct(strptime(itx[, 2], "%Y-%m-%d %H:%M:%OS"))
    # op <- options(digits.secs = 3) options(op)
    tef <- read.xlsx("ITX_SAN_TEL270412.xls", 3, )
    san <- read.xlsx("ITX_SAN_TEL270412.xls", 5, )
    san <- san[, 1:6]
    tef <- tef[, 1:6]
    san[, 2] <- as.POSIXct(strptime(san[, 2], "%Y-%m-%d %H:%M:%OS"))
    tef[, 2] <- as.POSIXct(strptime(tef[, 2], "%Y-%m-%d %H:%M:%OS"))
    #
    ts_itx <- cprepare(itx)
    ts_san <- cprepare(san)
    ts_tef <- cprepare(tef)
    #
    zoo_itx <- zoo(ts_itx[, -1], ts_itx[, 1])
    zoo_san <- zoo(ts_san[, -1], ts_san[, 1])
    zoo_tef <- zoo(ts_tef[, -1], ts_tef[, 1])

    pos_itx <- 2:(nrow(ts_itx) - 1)
    pos_san <- 2:(nrow(ts_san) - 1)
    pos_tef <- 2:(nrow(ts_tef) - 1)
    #
    RegTimes <- ISOdatetime(2012, 4, 27, 8, 59, 30) + seq(0:1020) * 30
    save.image("HFT_bd.RData")
}

Data is now ready as time is properly managed.

Series Overview

We can have a look at the ITX stock

ITX

Let’s make a short overview on the specific data. We will construct OHLC like time series.

#
ohlc_itx <- cto.ohlc(ts_itx[pos_itx, ], RegTimes, "ITX")
chartSeries(ohlc_itx)

plot of chunk plot1

reChart(subset = "first 10 minutes")

plot of chunk plot1

#

We will prepare the dataset now, and to do that we will consider the technical indexes provided by TTR package. In particular we will calculate:

#
chartSeries(ohlc_itx)

plot of chunk tidx1

reChart(subset = "first 10 minutes")

plot of chunk tidx1

#
cols <- paste("ITX", c("High", "Low", "Close"), sep = ".")
nmcl <- paste("ITX", "Adjusted", sep = ".")
nmvl <- paste("ITX", "Volume", sep = ".")
nmfl <- paste("ITX", "Close", sep = ".")
nc <- ncol(ohlc_itx)
d_itx <- as.data.frame(ohlc_itx)
# Adding indexes depending on HLC estructure
d_itx[, nc + 1] <- ADX(as.xts(ohlc_itx[, cols]))$ADX
colnames(d_itx)[nc + 1] <- "ADX"
d_itx[, nc + 2] <- ATR(as.xts(ohlc_itx[, cols]))$atr
colnames(d_itx)[nc + 2] <- "ATR"
d_itx[, nc + 3] <- CCI(as.xts(ohlc_itx[, cols]))$cci
colnames(d_itx)[nc + 3] <- "CCI"
d_itx[, nc + 4] <- CLV(as.xts(ohlc_itx[, cols]))$clv
colnames(d_itx)[nc + 4] <- "CLV"
d_itx[, nc + 5] <- SMI(as.xts(ohlc_itx[, cols]))$SMI
colnames(d_itx)[nc + 5] <- "SMI"
d_itx[, nc + 6] <- WPR(as.xts(ohlc_itx[, cols]))
colnames(d_itx)[nc + 6] <- "WPR"
# Adding indexes depending on Price
d_itx[, nc + 7] <- CMO(as.xts(ohlc_itx[, nmcl]))
colnames(d_itx)[nc + 7] <- "CMO"
d_itx[, nc + 8] <- SMA(as.xts(ohlc_itx[, nmcl]))
colnames(d_itx)[nc + 8] <- "SMA"
d_itx[, nc + 9] <- EMA(as.xts(ohlc_itx[, nmcl]))
colnames(d_itx)[nc + 9] <- "EMA"
d_itx[, nc + 10] <- MACD(as.xts(ohlc_itx[, nmcl]))$macd
colnames(d_itx)[nc + 10] <- "MACD"
d_itx[, nc + 11] <- DEMA(as.xts(ohlc_itx[, nmcl]))
colnames(d_itx)[nc + 11] <- "DEMA"
d_itx[, nc + 12] <- ZLEMA(as.xts(ohlc_itx[, nmcl]))
colnames(d_itx)[nc + 12] <- "ZLEMA"
d_itx[, nc + 13] <- WMA(as.xts(ohlc_itx[, nmcl]))
colnames(d_itx)[nc + 13] <- "WMA"
# Adding indexes depending on Price and Volume
d_itx[, nc + 14] <- CMF(as.xts(ohlc_itx[, cols]), as.xts(ohlc_itx[, nmvl]))
colnames(d_itx)[nc + 14] <- "CMF"
d_itx[, nc + 15] <- VWAP(as.xts(ohlc_itx[, nmcl]), as.xts(ohlc_itx[, nmvl]))
colnames(d_itx)[nc + 15] <- "VWAP"
d_itx[, nc + 15] <- VWAP(as.xts(ohlc_itx[, nmcl]), as.xts(ohlc_itx[, nmvl]))
colnames(d_itx)[nc + 15] <- "VWAP"
d_itx[, nc + 15] <- EVWMA(as.xts(ohlc_itx[, nmcl]), as.xts(ohlc_itx[, nmvl]))
colnames(d_itx)[nc + 15] <- "EVWMA"
d_itx[, nc + 16] <- ifelse(d_itx[, nmcl] > d_itx[, nmfl], 1, -1)
colnames(d_itx)[nc + 16] <- "STT"
# Defining the columns of interest
selco <- c(colnames(d_itx)[(nc + 1):ncol(d_itx)], nmvl, nmfl)
# Now, preparing the target data x min ahead to see if values grow up y
# cents
trainset <- function(d_itx, nm, min, cents) {
    nf <- nrow(d_itx)
    af <- 2 * min
    tmp <- cbind(d_itx[1:(nf - af), ], d_itx[-(1:af), nm])
    tmp[, ncol(tmp) + 1] <- ifelse((tmp[, ncol(tmp)] - tmp[, nm]) > cents/100, 
        1, -1)
    colnames(tmp)[ncol(tmp) - 1] <- "VF"
    colnames(tmp)[ncol(tmp)] <- "Target"
    return(tmp)
}
tset <- trainset(d_itx[, selco], "ITX.Close", 2, 10)
tset <- tset[-(1:37), ]
tset <- tset[tset[, "ITX.Volume"] > 0, ]
tset[, ncol(tset)] <- factor(tset[, ncol(tset)])

Let’s have a look at the training set:

tb1 <- xtable(tset)
print(tb1, "html")