Systematic Investor's Blog    About    Contact

RSI and Random Forest

04 Nov 2014

To install Systematic Investor Toolbox (SIT) please visit About page.

Tad Slaff at InovanceTech published How to Trade the RSI: An analysis using a Support Vector Machine post that I like. Below I will explore it a bit more.

Load Historical Prices. Let’s start with AUD/USD 4 hour bar pricesdropbox.

#*****************************************************************
# Load historical data
#****************************************************************** 
library(SIT)
load.packages('quantmod')

data <- new.env()
data$AUDUSD = read.xts('data/AUDUSD.csv', format='%m/%d/%y %H:%M', index.class = c("POSIXlt", "POSIXt"))
#plota(data$AUDUSD, type='l')
bt.prep(data, align='remove.na')

Create Indicators and Train SVM:

# helper functions
make.predictor <- function(x) { iif(x > 0, 1, -1) }

#*****************************************************************
# Code Strategies
#****************************************************************** 
load.packages('e1071,ggplot2')

indicators = as.xts(list(
RSI3 = RSI(Cl(data$AUDUSD), 3),
DEMA10 = DEMA(Cl(data$AUDUSD), 10),
CCI2 = CCI(HLC(data$AUDUSD), 20)
))

SMA50 = SMA(Op(data$AUDUSD), 50)

DataSet = as.xts(list(
RSI3 = RSI(Op(data$AUDUSD), 3),
Trend = Op(data$AUDUSD) - SMA50,
Direction = make.predictor(Cl(data$AUDUSD) - Op(data$AUDUSD))
))

# remove NA's
DataSet = DataSet[-(1:49),]

#Separate the data into 60% training set to build our model, 20% test set to 
# test the patterns we found, and 20% validation set to run our strategy over new 

Training = DataSet[1:4528,]
Test = DataSet[4529:6038,]
Val = DataSet[6039:7548,]

#Build our support vector machine using a radial basis function as our kernel, the cost, or C, at 1, and the gamma function at , or 1 over the number of inputs we are using

SVM = svm(Direction ~ RSI3 + Trend, data=Training, kernel="radial",cost=1,gamma=1/2)

#Run the algorithm once more over the training set to visualize the patterns it found
TrainingPredictions = predict(SVM,Training,type="class")
TrainingPredictions = make.predictor(TrainingPredictions)

# Plot
TrainingData = data.frame(Training,TrainingPredictions)
ggplot(TrainingData, aes(x=Trend,y=RSI3)) + 
stat_density2d(geom="contour",aes(color=as.factor(TrainingPredictions)))+
labs(title="SVM RSI3 and Trend Predictions",
x="Open - SMA50",y="RSI3",color="Training Predictions")

plot of chunk plot-3

Now that we have found a basic set of rules that the SVM uncovered, lets test to see how well they hold up over new data, our test set.

# Now that we have found a basic set of rules that the SVM uncovered, 
# lets test to see how well they hold up over new data, our test set.

ShortRange1 = Test$RSI3 < 25 & Test$Trend > -.010 & Test$Trend < -.005
ShortRange2 = Test$RSI3 > 70 & Test$Trend < -.005
ShortRange3 = Test$RSI3 > 75 & Test$Trend > .015

LongRange1 = Test$RSI3 < 25 & Test$Trend < -.02
LongRange2 = Test$RSI3 > 50 & Test$RSI3 < 75 & Test$Trend > .005 & Test$Trend < .01


# Now lets see how well these patterns hold up over out test set:
ShortTrades = Test[ShortRange1 | ShortRange2 | ShortRange3,]
ShortCorrect = sum(ShortTrades$Direction == -1)/nrow(ShortTrades)*100
ShortCorrect

[1] 57.82313

LongTrades  = Test[LongRange1 | LongRange2, ]
LongCorrect = sum(LongTrades$Direction == 1)/nrow(LongTrades)*100
LongCorrect

[1] 57.14286

Wow! 58% (85 correct out of 147 trades) for our short trades and 57% (80 correct out of 140 trades) for our long trades.

Now lets turn this into an actual trading strategy with a stop loss and take profit as our exit conditions.

(this report was produced on: 2014-12-07)

Related Posts