Blackbox trading Strategy using Rapidminer and R

January 23, 2011
By

(This article was first published on a Physicist in Wall Street, and kindly contributed to R-bloggers)

This my first post in 2011. this post has cost me a bit more than usual, but I hope it meets expectations. The aim of this tutorial is to generate an algorithm based on black box trading, with all the necessary elements for evaluation. That is a first post of several, in order to explore the problems, features of this strategy.
In recent posts we have explored the performance assessment of a simple strategy. To improve this kind of evaluation and to enhance the capabilities of the use of R and RapidMiner for trading we have included the blotter library possibilities into the model evaluation (that not was simple due to the limitations of R extension). Blotter is an R package that tracks the P&L of your trading systems (or simulations), even if your portfolio spans many security types and/or currencies. This post uses blotter to track the blackbox trading strategy and it will allows us to manage different profiles with Rapidminer.
The basic strategy was done for GSPC and it is described in the following diagram
clip_image002
The trading strategy is divided into the generation of models, prediction and filtering spurious signals. The first step is the model generation, I tested different regression models, neural networks, SVM, etc for the prediction model generation. For spurious signals littering we have used a binary classification (bad prediction or correct prediction). The binary classification (correct or false prediction) is done using different rule extraction algorithms using as input only technical indicators.
clip_image004

The second step is the Evaluation of the strategy using the models generated in the previous step
clip_image006
To implement this strategy we have used Rapidminer and R plugin, you can see the complexity of the algorithm in the following picture. The main requirement is Blotter, however it is not yet in the cran repository, so you need to install from R-forge
install.packages("blotter", repos="http://R-Forge.R-project.org")
you can see more information in: https://r-forge.r-project.org/R/?group_id=316
To use the last version of blotter you need to update R to the last one in that moment 2.12 version.
image
I will describe the most important R models, and in a future post I will describe, inside one video, all elements of the architecture.
The main elements are:
– Obtain Technical Data for Training model
– Evaluation Strategy

Obtain Technical Data for Training module

   1:  

   2: library(quantmod)

   3: library(TTR)

   4: library(PerformanceAnalytics)

   5: # Parameters

   6: SymbolName="^GSPC"

   7: initDate="2003-01-01"

   8: secondDate="2006-01-01::"

   9: # pull stock data from Yahoo Finance

  10: symbol<-getSymbols(SymbolName,from=initDate)

  11: stock<-xts(get(symbol))

  12: #remove stock name

  13: names(stock)[names(stock)==paste(symbol,'.Open',sep="")] <- 'Open' 

  14: names(stock)[names(stock)==paste(symbol,'.Close',sep="")] <- 'Close' 

  15: names(stock)[names(stock)==paste(symbol,'.Volume',sep="")] <- 'Volume' 

  16: names(stock)[names(stock)==paste(symbol,'.Adjusted',sep="")] <- 'Adjusted' 

  17: names(stock)[names(stock)==paste(symbol,'.High',sep="")] <- 'High' 

  18: names(stock)[names(stock)==paste(symbol,'.Low',sep="")] <- 'Low'

  19:  

  20: # Introduce RSI Indicator

  21: stock$RSI2 = RSI(Cl(stock), 2)

  22: #Introduce Eponential Moving Average indicator

  23: stock$EMA7=EMA(Cl(stock), n=7, wilder=FALSE, ratio=NULL)

  24: stock$EMA50=EMA(Cl(stock), n=50, wilder=FALSE, ratio=NULL)

  25: stock$EMA200=EMA(Cl(stock), n=200, wilder=FALSE, ratio=NULL)

  26: #Introduce MACD indicator

  27: stock$MACD26=MACD(Cl(stock), nFast=12, nSlow=26, nSig=9)

  28: #Introduce ADX indicator 

  29: stock$ADX14=ADX(stock, n=14)

  30: #Introduce AROON indicator

  31: stock$AROON = aroon(Cl(stock), n=20 )

  32: #Introduce ATR indicator  

  33: stock$ATR=ATR(stock, n=14)

  34: #Introduce BOLLINGER indicator

  35: stock$bbands = BBands(Cl(stock))

  36: #Introduce Commodity Channel Index indicator

  37: stock$Commodity =  CCI(Cl(stock))

  38: #Introduce Chaikin Accumulation  Distribution indicator

  39: #stock$chaikin =  chaikinAD(stock, stock$Volume)

  40: #Introduce chaikinVolatility Chaikin Volatilit indicator

  41: stock$chaikinVolatility = chaikinVolatility(stock)

  42: #Introduce Close Location Value indicator

  43: #stock$clv = CLV(stock) # infinites values

  44: #Introduce Chaikin Money Flow indicator

  45: #stock$cmf = CMF(stock, stock$Volume)

  46: #LAG

  47: stock$label=lag(stock$Adjusted,-1)

  48: #

  49: # remove 2003,2004,2005 in order to avoid NaN from EMA indicators

  50: # To maintain time it is necessary to conver in texts

  51: results <-data.frame(stock[secondDate],TIME=as.character(index(stock[secondDate])))

  52:  

Evaluation Strategy
This module define, using blotter, the strategy described before.

   1: # Load required libraries

   2:  

   3: library(quantmod)

   4:  

   5: library(TTR)

   6:  

   7: library(xts)

   8:  

   9: library(FinancialInstrument)

  10:  

  11: library(zoo)

  12:  

  13: library(blotter) 

  14:  

  15: library(PerformanceAnalytics)

  16:  

  17:  

  18: # Try to clean up in case the demo was run previously

  19:  

  20: try(rm("account.rapidminer","portfolio.rapidminer",pos=.blotter),silent=TRUE)

  21:  

  22: try(rm("ltaccount","ltportfolio","ClosePrice","CurrentDate","equity","Symbol","i","initDate","initEq","Posn","UnitSize","verbose"),silent=TRUE)

  23:  

  24: # Set initial values you can change to reduce the test period

  25:  

  26: initDate='2009-01-27'

  27:  

  28: initEq=100000

  29:  

  30: stopLoss=0.005 

  31:  

  32: # it is important to introduc this sentence to work using blotter

  33:  

  34: Sys.setenv(TZ="GMT")

  35:  

  36:  

  37: # Load data with quantmod

  38:  

  39: print("Loading data")

  40:  

  41: currency("USD")

  42:  

  43: stock("Symbol",currency="USD",multiplier=1)

  44:  

  45:  

  46: # Set up a portfolio object and an account object in blotter

  47:  

  48: print("Initializing portfolio and account structure")

  49:  

  50: ltportfolio='rapidminer'

  51:  

  52: ltaccount='rapidminer'

  53:  

  54: initPortf(ltportfolio,'Symbol', initDate=initDate)

  55:  

  56: initAcct(ltaccount,portfolios='rapidminer', initDate=initDate, initEq=initEq)

  57:  

  58: verbose=TRUE

  59:  

  60: datos <- data # for output data 

  61:  

  62:  

  63: dates <- as.Date(data$TIME)

  64:  

  65: data$TIME=NULL # remove string information

  66:  

  67: data$correct_prediction=NULL # remove information from data

  68:  

  69: Symbol <- xts(data,order.by=as.POSIXct(dates),index.class=c("POSIXt","POSIXct"))

  70:  

  71:  

  72: PREDICTION = data$prediction

  73:  

  74: close_ROC <- ROC(data$label)

  75:  

  76: prediction_ROC <-ROC(data$prediction)

  77:  

  78: close_ROC[1] <- 0

  79:  

  80: prediction_ROC[1] <- 0

  81:  

  82: #generate signals from prediction values

  83:  

  84: sigup <- ifelse(prediction_ROC > 0, 1, 0)

  85:  

  86: sigdn <- ifelse(prediction_ROC < 0, -1, 0)

  87:  

  88: # Replace missing signals with no position

  89:  

  90: # (generally just at beginning of series)

  91:  

  92: sigup[is.na(sigup)] <- 0

  93:  

  94: sigdn[is.na(sigdn)] <- 0

  95:  

  96: sig <- sigup + sigdn

  97:  

  98: # Create trades

  99:  

 100: for( i in 2:NROW(prediction_ROC) ) { 

 101:  

 102: CurrentDate=dates[i]

 103:  

 104: cat(".")

 105:  

 106: equity = getEndEq(ltaccount, CurrentDate)

 107:  

 108: ClosePrice = as.numeric(Symbol$Adjusted[i])

 109:  

 110: OpenPrice = as.numeric(Symbol$Open[i])

 111:  

 112: LowPrice = as.numeric(Symbol$Low[i])

 113:  

 114: HighPrice = as.numeric(Symbol$High[i])

 115:  

 116: filter<-data$filter[i-1]

 117:  

 118: # filter<-1 #if you want to remove filter un-comment this line

 119:  

 120: Posn = getPosQty(ltportfolio, Symbol='Symbol', Date=CurrentDate)

 121:  

 122: UnitSize = as.numeric(trunc(equity/ClosePrice))

 123:  

 124: #print(paste("UnitSize",UnitSize))

 125:  

 126: # Position Entry (assume fill at close)

 127:  

 128: if( Posn == 0 ) { 

 129:  

 130: # No position, so start Long position

 131:  

 132: if( prediction_ROC[i-1] >0 && filter==1 ) { 

 133:  

 134:  

 135: StopPrice = OpenPrice *(1-stopLoss)

 136:  

 137: # Store trade with blotter

 138:  

 139: addTxn(ltportfolio, Symbol='Symbol', TxnDate=CurrentDate, TxnPrice=OpenPrice, TxnQty = UnitSize , TxnFees=0)

 140:  

 141:  

 142:  

 143: if (LowPrice<StopPrice){

 144:  

 145: # Exit

 146:  

 147: print(paste("Exit",CurrentDate))

 148:  

 149: addTxn(ltportfolio, Symbol='Symbol', TxnDate=CurrentDate, TxnPrice=StopPrice, TxnQty = -UnitSize , TxnFees=0)

 150:  

 151:  

 152: }

 153:  

 154: }else{

 155:  

 156: if( prediction_ROC[i-1] <0 && filter==1) { 

 157:  

 158: #print("Prediction_ROC is SHELL")

 159:  

 160:  

 161: StopPrice = OpenPrice *(1+stopLoss)

 162:  

 163:  

 164: # Store trade with blotter

 165:  

 166: addTxn(ltportfolio, Symbol='Symbol', TxnDate=CurrentDate, TxnPrice=OpenPrice, TxnQty = -UnitSize , TxnFees=0)

 167:  

 168:  

 169: if (HighPrice>StopPrice){

 170:  

 171: # Exit

 172:  

 173: print(paste("Exit",CurrentDate))

 174:  

 175: addTxn(ltportfolio, Symbol='Symbol', TxnDate=CurrentDate, TxnPrice=StopPrice, TxnQty = UnitSize , TxnFees=0)

 176:  

 177: }

 178:  

 179:  

 180:  

 181: }else{

 182:  

 183:  

 184: #print("Prediction_ROC is 0")

 185:  

 186: }

 187:  

 188: }

 189:  

 190: }else{ 

 191:  

 192: if(Posn >0) {

 193:  

 194: # Have a position, so check exit

 195:  

 196: if( prediction_ROC[i-1] < 0 && filter==1) { 

 197:  

 198: print("Prediction_ROC is SHELL")

 199:  

 200:  

 201:  

 202: StopPrice = OpenPrice *(1+stopLoss)

 203:  

 204:  

 205: # Store trade with blotter

 206:  

 207: print("close position")

 208:  

 209: addTxn(ltportfolio, Symbol='Symbol', TxnDate=CurrentDate, TxnPrice=OpenPrice, TxnQty = -2*abs(Posn) , TxnFees=0)

 210:  

 211:  

 212: if (HighPrice>StopPrice){

 213:  

 214: # Exit

 215:  

 216: print(paste("Exit",CurrentDate))

 217:  

 218: addTxn(ltportfolio, Symbol='Symbol', TxnDate=CurrentDate, TxnPrice=StopPrice, TxnQty = 2*abs(Posn) , TxnFees=0)

 219:  

 220:  

 221: }

 222:  

 223: }else{

 224:  

 225:  

 226: if (LowPrice<StopPrice){

 227:  

 228: # Exit

 229:  

 230: print(paste("Exit",CurrentDate))

 231:  

 232: addTxn(ltportfolio, Symbol='Symbol', TxnDate=CurrentDate, TxnPrice=StopPrice, TxnQty = -abs(Posn) , TxnFees=0)

 233:  

 234:  

 235: }

 236:  

 237:  

 238: }

 239:  

 240: }else{

 241:  

 242:  

 243: if( prediction_ROC[i-1] > 0 && filter==1) { 

 244:  

 245:  

 246: StopPrice = OpenPrice *(1-stopLoss)

 247:  

 248:  

 249: # Store trade with blotter

 250:  

 251: #print("close position")

 252:  

 253: addTxn(ltportfolio, Symbol='Symbol', TxnDate=CurrentDate, TxnPrice=OpenPrice, TxnQty = 2*abs(Posn) , TxnFees=0)

 254:  

 255:  

 256: if (LowPrice<StopPrice){

 257:  

 258: # Exit

 259:  

 260: print(paste("Exit",CurrentDate))

 261:  

 262: addTxn(ltportfolio, Symbol='Symbol', TxnDate=CurrentDate, TxnPrice=StopPrice, TxnQty = -2*abs(Posn) , TxnFees=0)

 263:  

 264:  

 265: }

 266:  

 267: }else{

 268:  

 269: # we maintain short position except exit

 270:  

 271:  

 272: if (HighPrice>StopPrice){

 273:  

 274: # Exit

 275:  

 276: print(paste("Exit",CurrentDate))

 277:  

 278: addTxn(ltportfolio, Symbol='Symbol', TxnDate=CurrentDate, TxnPrice=StopPrice, TxnQty = abs(Posn) , TxnFees=0)

 279:  

 280:  

 281: }

 282:  

 283:  

 284: }

 285:  

 286:  

 287: }

 288:  

 289: } 

 290:  

 291: # Calculate P&L and resulting equity with blotter

 292:  

 293: updatePortf(ltportfolio, Dates = CurrentDate,Prices= Symbol$Adjusted)

 294:  

 295: updateAcct(ltaccount, Dates = CurrentDate)

 296:  

 297: updateEndEq(ltaccount, Dates = CurrentDate)

 298:  

 299: } # End dates loop

 300:  

 301: cat('\n')

 302:  

 303: transactions=getTxns(Portfolio="rapidminer", Symbol="Symbol")

 304:  

 305: # Copy the results into the local environment

 306:  

 307: print("Retrieving resulting portfolio")

 308:  

 309: ltportfolio = getPortfolio("rapidminer")

 310:  

 311: print("Retrieving resulting account")

 312:  

 313: ltaccount = getAccount("rapidminer")

 314:  

 315: results <-data.frame(ltaccount$summary)

 316:  

 317: # generate buy and hold

 318:  

 319: buyhold = ROC(Symbol$Adjusted)

 320:  

 321: # Plot Strategy Summary 

 322:  

 323: png(filename="20110109_blotter_strategy.png", 1000, 1000, units = "px", pointsize = 12, bg = "white", res = 200, restoreConsole = TRUE)

 324:  

 325: Eq<-ROC(getAccount('rapidminer')$summary$End.Eq)

 326:  

 327: #names(stock)[names(stock)==paste(symbol,'.Low',sep="")] <- 'Low'

 328:  

 329: buyhold$Blackbox <- Eq

 330:  

 331: charts.PerformanceSummary(buyhold,colorset=rich6equal, lwd=2, ylog=TRUE)

 332:  

 333: AnnualizedReturns <- table.AnnualizedReturns(buyhold)

 334:  

 335: Stats <- table.Stats(buyhold)

 336:  

 337: DownsideRisk <- table.DownsideRisk(buyhold)

 338:  

 339: dev.off()

 340:  

It was defined three time interval, one for training (800 days), one for evaluation genetic space search (300 days) and finally the rest of the days for testing ( >200 days)

Several tests have been done using Neural Networks and SVM, obtaining diverse results, however always positive results. You can see in the following image the results obtained

Using neural network we have improved the performance of the strategy in the training and evaluation data, during 7 hours, 100 generations we obtain the following improvement (figure). We use the same technique than in previous post. It is not possible to use for evaluation Blotter due to the time consumption of this package.

image

The results obtained using this strategy, without filter can be observed in the following graph, the comparison was done with buy&hold strategy. Although backtesting is no guarantee of future performance, it gives the trader confidence that the strategy has worked in the past. If the strategy is not over-optimized, data-mined, or based on random coincidences, it might have a good chance of working in the future.

image

The Sharpe Ratio (Rf=0%) of this strategy is 2.46.

On the other hand if we use the filter

image

Using the filter, we reduce the drawdown peak; however we obtained lower return and a Sharpe Ratio of 2.18. Currently I’m doing several test with different filters however the Sharpe Ratio of the prediction algorithm without filter is very high for this index.

Also, you can modify the type of filter, for example you can see the filter obtained using J48 algorithm. The idea is to auto-generate and understand  filters using technical indicators.

J48 pruned tree

——————

RSI2-0 <= 44.156798: false (409.0/168.0)

RSI2-0 > 44.156798

| DX-0 <= 5.179328

| | DX-0 <= 2.667342

| | | tr-1 <= 4.81: false (10.33/2.33)

| | | tr-1 > 4.81: true (20.67/3.0)

| | DX-0 > 2.667342: true (24.0/1.0)

| DX-0 > 5.179328: true (536.0/255.0)

Number of Leaves : 5

Size of the tree : 9

The current problems of the model are:

– The accuracy of the long and short position prediction, it is not good should be improved.

– My objective is to reduce the risk of the strategy, so the the filter should be improved in order to reduce false signals.

– Reduce the number of variable to adjust for the strategy…

In the following post, I will describe in a video the model with all elements of the architecture, and I will try to discuss about the problems discovered in the strategy.

clip_image012

We do not guarantee that those strategies will give you profits or have mistakes. It is important to remember that each trading situation is unique. You can never copy a strategy and think that it will always work. It is allot factors to take consideration too. For example, how the game looks like, how the odds are moving etc. This blog is a research activity to share knowledge related to datamining and trading algorithms. All code described in this blog is GPL v3, We try to put all information about the model and how to generate the results obtained and also if you have any doubt about the model you will receive the answer by email or comments in the blog. If you don¡t have patient and you want to obtain the files you can, paying one small donation to improve the website,

To leave a comment for the author, please follow the link and comment on his blog: a Physicist in Wall Street.

R-bloggers.com offers daily e-mail updates about R news and tutorials on topics such as: visualization (ggplot2, Boxplots, maps, animation), programming (RStudio, Sweave, LaTeX, SQL, Eclipse, git, hadoop, Web Scraping) statistics (regression, PCA, time series, trading) and more...



If you got this far, why not subscribe for updates from the site? Choose your flavor: e-mail, twitter, RSS, or facebook...

Comments are closed.