Machine learning and commonly used trading indicators (part 1)

Portfolio trading comparison example - part 1

In the previous post, I reworked the tradingPortfolio() procedures to handle trading signals produced by both quantmod library based model and custom ones. As custom trading signal, I mean any trading signal whose syntax is a sequence of {-1,0,1} values respectively indicating {sell,hold,buy} action. That can be build by any user procedure following any criteria which makes sense.

In this post, first I build a quantmod based model which forecats next week close price to next week open price gain by taking into account last two weeks gains. The trading dates are chosen so that both trading strategies at the end determine a sell action. Signal thresholds are set to 0 to evaluate trading strategies signal issuing at their core. Also buy.all and sell.all flags are set to TRUE for the same reasons. The output portfolio will be referenced as machine learning based portfolio.

suppressPackageStartupMessages(library(quantmod))
suppressPackageStartupMessages(library(rpart))
suppressPackageStartupMessages(library(rpart.plot))
suppressPackageStartupMessages(library(knitr))

ticker <- "YHOO"
getSymbols(ticker, src='yahoo')

##     As of 0.4-0, 'getSymbols' uses env=parent.frame() and
##  auto.assign=TRUE by default.
## 
##  This  behavior  will be  phased out in 0.5-0  when the call  will
##  default to use auto.assign=FALSE. getOption("getSymbols.env") and 
##  getOptions("getSymbols.auto.assign") are now checked for alternate defaults
## 
##  This message is shown once per session and may be disabled by setting 
##  options("getSymbols.warning4.0"=FALSE). See ?getSymbols for more details.

## [1] "YHOO"

ticker = to.weekly(YHOO)

training.dates <- c('2012-01-01','2013-12-31')
trading.dates <- c('2014-01-01', '2015-02-20')

q.model <- specifyModel(Next(OpCl(ticker)) ~ (Lag(OpCl(ticker), 0:2)))

model.data <- modelData(q.model)
head(model.data)

##            Next.OpCl.ticker Lag.OpCl.ticker.0.2.Lag.0
## 2007-01-19      0.006822298              -0.074966535
## 2007-01-26      0.025668486               0.006822298
## 2007-02-02      0.037321242               0.025668486
## 2007-02-09      0.089450287               0.037321242
## 2007-02-16      0.009433931               0.089450287
## 2007-02-23     -0.072560947               0.009433931
##            Lag.OpCl.ticker.0.2.Lag.1 Lag.OpCl.ticker.0.2.Lag.2
## 2007-01-19               0.063176893               0.073114120
## 2007-01-26              -0.074966535               0.063176893
## 2007-02-02               0.006822298              -0.074966535
## 2007-02-09               0.025668486               0.006822298
## 2007-02-16               0.037321242               0.025668486
## 2007-02-23               0.089450287               0.037321242

model <- buildModel(q.model, method='rpart', training.per=training.dates,
                    control=rpart.control(minsplit=15))
rpart.plot(model@fitted.model)

signal.threshold <- c(0.00, 0.00)
control <- list("initial" = c("portfolio" = 0, 
                              "own.shares" = 0, 
                              "average.price"= 0, 
                              "buy.first" = TRUE, 
                              "sell.first" = FALSE),
                "available.funds" = 6000,
                "single.buy" = 0, 
                "buy.all" = TRUE, 
                "single.sell" = 0,
                "sell.all" = TRUE, 
                "signal.threshold" = signal.threshold)

res <- tradePortfolio(ticker, model, trading.dates, control)

portfolioTradingHistory <- res[[1]]

filt.col <- c("portfolio.Open", "portfolio.Close", "portfolio.Gain", "action", 
              "own.shares", "available.funds.Close")

portfolioTradingHistFilt <- subset(portfolioTradingHistory[,filt.col], action != "hold")
kable(portfolioTradingHistFilt, caption="Machine learning based portfolio trading history") 

Machine learning based portfolio trading history

	portfolio.Open	portfolio.Close	portfolio.Gain	action	own.shares	available.funds.Close
2014-01-10	0.000	6176.779	0.0000000	buy	149.8127	0.000
2014-01-31	5394.756	0.000	-10.0873935	sell	0.0000	5394.756
2014-02-07	0.000	5588.391	-10.0873935	buy	150.1045	0.000
2014-03-07	5809.045	0.000	-3.1825805	sell	0.0000	5809.045
2014-03-14	0.000	5654.157	-3.1825805	buy	150.3765	0.000
2014-03-28	5398.517	0.000	-10.0247097	sell	0.0000	5398.517
2014-04-11	0.000	5202.265	-10.0247097	buy	158.2679	0.000
2014-05-30	5483.982	0.000	-8.6002962	sell	0.0000	5483.982
2014-06-06	0.000	5678.427	-8.6002962	buy	158.0854	0.000
2014-06-13	5839.674	0.000	-2.6720938	sell	0.0000	5839.674
2014-06-20	0.000	5681.169	-2.6720938	buy	166.8478	0.000
2014-07-25	6026.544	0.000	0.4423964	sell	0.0000	6026.544
2014-08-01	0.000	5925.075	0.4423964	buy	166.3413	0.000
2014-08-15	6066.466	0.000	1.1077642	sell	0.0000	6066.466
2014-08-22	0.000	6271.046	1.1077642	buy	164.9841	0.000
2014-08-29	6353.538	0.000	5.8922980	sell	0.0000	6353.538
2014-09-05	0.000	6466.235	5.8922980	buy	163.3300	0.000
2014-10-24	7104.856	0.000	18.4142603	sell	0.0000	7104.856
2014-10-31	0.000	7554.343	18.4142603	buy	164.0465	0.000
2014-12-05	8364.733	0.000	39.4122196	sell	0.0000	8364.733
2014-12-12	0.000	8318.373	39.4122196	buy	165.5727	0.000
2014-12-26	8421.028	0.000	40.3504677	sell	0.0000	8421.028
2015-01-02	0.000	8337.931	40.3504677	buy	166.1936	0.000
2015-02-20	7330.799	0.000	22.1799786	sell	0.0000	7330.799

Afterwards, I build a portfolio based on the 10-weeks moving average indicator. The difference between the weekly closing price and its MA-10 moving average is used to trigger buy/sell actions. When such difference changes sign from negative to positive, a buy signal +1 is determined When on the contrary such difference changes sign from positive to negative, a sell signal -1 is determined. The output portfolio will be referenced as MA-10 based portfolio.

adj.trading.dates <- adjustTradingDate(ticker, trading.dates)
trading.window <- seq(from = adj.trading.dates[[1]], to = adj.trading.dates[[2]], by=1)

ticker.ma10 <- EMA(Cl(ticker), n = 10)
ticker.signal <- Cl(ticker) - ticker.ma10

ticker.trade <- ticker[trading.window]
len = nrow(ticker.trade)

signal <- rep(0, len)
trading.signal <- ticker.signal[trading.window]
for (i in 2:len) {
  if (trading.signal[i-1] <= 0 && trading.signal[i] > 0) {
    signal[i] <- 1
  } 
  if (trading.signal[i-1] >=0 && trading.signal[i] < 0) {
    signal[i] <- (-1)
  }
}

portfolioMaTradingHistory <- tradePortfolioWithSignal(ticker.trade, control, signal)

portfolioMaTradingHistFilt <- subset(portfolioMaTradingHistory[,filt.col], action != "hold")
kable(portfolioMaTradingHistFilt, caption="MA-10 based portfolio trading history") 

MA-10 based portfolio trading history

	portfolio.Open	portfolio.Close	portfolio.Gain	action	own.shares	available.funds.Close
2014-02-14	0.000	6036.316	0.0000000	buy	157.8947	0.000
2014-02-21	5887.895	0.000	-1.8684184	sell	0.0000	5887.895
2014-02-28	0.000	6115.629	-1.8684184	buy	158.1492	0.000
2014-03-14	5946.410	0.000	-0.8931703	sell	0.0000	5946.410
2014-04-17	0.000	6448.000	-0.8931703	buy	177.2402	0.000
2014-04-25	6111.243	0.000	1.8540564	sell	0.0000	6111.243
2014-05-02	0.000	6499.035	1.8540564	buy	176.2689	0.000
2014-05-09	5950.839	0.000	-0.8193557	sell	0.0000	5950.839
2014-06-06	0.000	6161.837	-0.8193557	buy	171.5434	0.000
2014-06-20	5841.051	0.000	-2.6491514	sell	0.0000	5841.051
2014-07-03	0.000	6043.389	-2.6491514	buy	167.2216	0.000
2014-07-18	5573.497	0.000	-7.1083889	sell	0.0000	5573.497
2014-07-25	0.000	6036.423	-7.1083889	buy	167.1213	0.000
2014-10-17	6425.816	0.000	7.0969342	sell	0.0000	6425.816
2014-10-24	0.000	7265.999	7.0969342	buy	167.0345	0.000
2015-01-16	7762.092	0.000	29.3681962	sell	0.0000	7762.092
2015-01-23	0.000	8120.419	29.3681962	buy	165.8921	0.000
2015-01-30	7297.594	0.000	21.6265674	sell	0.0000	7297.594

Saving both portfolio datasets for later analysis.

write.csv(portfolioTradingHistory, file = "portfolioTradingHistory.csv", row.names=TRUE)
write.csv(portfolioMaTradingHistory, file = "portfolioMaTradingHistory.csv", row.names=TRUE)

In my next post, I will comment on and perform graphic comparison of the results.