Testing A Strategy For Robustness With Time Series Cross Validation

In two previous posts we attempted to increase our sample size for back testing using time series bootstrapping. One method used the maximum entropy bootstrap for time series using the meboot R package, the post can be found here. The second method involved using block bootstrap with sample with replacement (one may edit code for without), by any means the post can be found here.

After questioning the validity of the above procedures. We now focus on applying cross validation techniques on our time series data. Instead of optimizing over the full data set which can lead to over fitting. We split the data into many train (in-sample) and test (out-sample) periods. A good explanation can be found here.

In this example. We split each train set by a total number of 252 rows, which equates to 1 year of trading data. Each test period is split into a total number of 126 rows, which equates to 6 months of trading data. For this example we will use the VXV / VXMT sma ratio strategy. As a reminder the rules are as follows:

  1. long VXX when VXV/VXMT ratio is over 1 & the VXV/VXMT ratio is over an ‘n’ period sma of the VXV/VXMT ratio.
  2. long XIV when VXV/VXMT ratio is less than 1 & the VXV/VXMT ratio is below an ‘n’ period sma of the VXV/VXMT ratio.

The free parameters which can be optimized are the VXV/VXMT ratio value and also the VXV/VXMT simple moving average look back period.

We optimize the vxv/vxmt ratio sma look back period over a sma length of 2:170. Each sma ratio is optimized over each train set and ranked per their sharpe ratio. Those that produced the highest sharpe ratio were used in the next test set period.

An example of the procedure: after optimizing the sma lookback on train set 1, from rows 1:253, [1+252 = 253] we use the sma look back which produced the highest sharpe ratio in the next test set 1, rows 253:378 (+125 rows, inclusive start row). After this test set period. We move the whole train set window up by +126 rows, making train set 2 range from rows, 127 to 379 (+252 row width) the next test set 1 then begins on row 379 to 504 (+126 row width, inclusive start row). After this test set, we then move the train set 3 window up another 126 rows, so on and so forth until we reach the end of the series.

First before splitting the train and test sets, we wish to continuously calculate the VXV/VXMT ratio and all VXV/VXMT ratio SMA’s. This is so that we have the long run history and we do not try and calculate anything on each smaller train/test block.

Next we run the back test over all train set periods and record the highest sharpe ratio in each train period, the code that achieves this:


# Train / Test Set VXV / VXMT Strategy 
# Andrew Bannerman 

################################################
# Procedure 
# 1. Split data into train / test sets 
# 2. Select 1 year look back for each train set (252 days)
# 3. Select 6 months test set (126)
# 4. Re-calibrate on a moving window through entire time series 
#################################################

# Required Packages 
library(readxl)
require(xts)
require(data.table)
require(ggplot2)
require(RColorBrewer)
require(lubridate)
require(magrittr)
require(scales)
require(reshape2)
require(PerformanceAnalytics)
require(dplyr)
require(TTR)

#################################################
# 1. Load Data 
#################################################
# Load Syntehtic and join to alpha vantage adjusted prices 
# Load synthetic VXX and XIV data 
synth <- read_excel("D:/R Projects/Final Scripts/VIX_term_structure/vix-funds-models-no-formulas.xls", col_names = TRUE)
synth1 <- read_excel("D:/R Projects/Final Scripts/VIX_term_structure/vix-mt-funds-models-no-formulas.xls", col_names = TRUE)
synth <- as.data.frame(synth)
synth1 <- as.data.frame(synth1)
# Extract synthetic series 
vxx.synth <- data.frame(synth$Date, synth$'VXX calc')
xiv.synth <- data.frame(synth$Date, synth$'XIV calc')
ziv.synth <- data.frame(synth1$Date, synth1$'ZIV calc')
vxz.synth <- data.frame(synth1$Date, synth1$'VXZ calc')
colnames(vxx.synth)[1] <- "Date"
colnames(vxx.synth)[2] <- "vxx_close"
colnames(xiv.synth)[1] <- "Date"
colnames(xiv.synth)[2] <- "xiv_close"
colnames(ziv.synth)[1] <- "Date"
colnames(ziv.synth)[2] <- "ziv_close"
colnames(vxz.synth)[1] <- "Date"
colnames(vxz.synth)[2] <- "vxz_close"
vxx.synth$Date <- ymd(vxx.synth$Date)
xiv.synth$Date <- ymd(xiv.synth$Date)
ziv.synth$Date <- ymd(ziv.synth$Date)
vxz.synth$Date <- ymd(vxz.synth$Date)

# Download data from alphavantage
VXX <- fread("https://www.alphavantage.co/query?function=TIME_SERIES_DAILY_ADJUSTED&symbol=VXX&outputsize=full&apikey=your_api_key&datatype=csv") #fread() data.table for downloading directly to a data frame
VXX$timestamp <- ymd(VXX$timestamp)   #Lubridate to change character date to date format
VXX <- arrange(VXX,timestamp)   #dplyr to sort data frame by date ascending order
colnames(VXX)[1] <- "Date"
VXX$Date <- ymd(VXX$Date)
VXX <- as.data.frame(VXX)

XIV <- fread("https://www.alphavantage.co/query?function=TIME_SERIES_DAILY_ADJUSTED&symbol=XIV&outputsize=full&apikey=your_api_key&datatype=csv") #fread() data.table for downloading directly to a data frame
XIV$timestamp <- ymd(XIV$timestamp)   #Lubridate to change character date to date format
XIV <- arrange(XIV,timestamp)   #dplyr to sort data frame by date ascending order
colnames(XIV)[1] <- "Date"
XIV$Date <- ymd(XIV$Date)
XIV <- as.data.frame(XIV)

# Join sythentic data to alpha vantage 
vxx.synth <- subset(vxx.synth, Date <= as.POSIXct("2009-01-29"))
xiv.synth <- subset(xiv.synth, Date <= as.POSIXct("2010-11-29"))
# Subset only date and close from alpha vantage data 
VXX <- VXX[ -c(2:5, 7:9) ]  # subset adjusted close
XIV <- XIV[ -c(2:5, 7:9) ]  # subset adjusted close
colnames(VXX)[2] <- "vxx_close"
colnames(XIV)[2] <- "xiv_close"

# row bind 
VXX <- rbind(vxx.synth,VXX)
XIV <- rbind(xiv.synth,XIV)
df <- cbind(VXX,XIV)

# Download VXV and VXMT data from CBOE website
# Download VXV Data From CBOE website 
VXV_cboe <- fread("http://www.cboe.com/publish/scheduledtask/mktdata/datahouse/vix3mdailyprices.csv")
VXV_cboe <- as.data.frame(VXV_cboe)
VXV_cboe <- VXV_cboe[3:nrow(VXV_cboe), ]
colnames(VXV_cboe)[1] = "Date"
colnames(VXV_cboe)[2] = "vxv_cboe_open"
colnames(VXV_cboe)[3] = "vxv_cboe_high"
colnames(VXV_cboe)[4] = "vxv_cboe_low"
colnames(VXV_cboe)[5] = "vxv_cboe_close"
VXV_cboe$Date <- mdy(VXV_cboe$Date)
cols <-c(2:5)
VXV_cboe[,cols] %<>% lapply(function(x) as.numeric(as.character(x)))

# Download VXMT Data from CBOE website
VXMT_cboe <- fread("http://www.cboe.com/publish/scheduledtask/mktdata/datahouse/vxmtdailyprices.csv")
VXMT_cboe <- as.data.frame(VXMT_cboe)
VXMT_cboe <- VXMT_cboe[3:nrow(VXMT_cboe), ]
colnames(VXMT_cboe)[1] = "Date"
colnames(VXMT_cboe)[2] = "vxmt_cboe_open"
colnames(VXMT_cboe)[3] = "vxmt_cboe_high"
colnames(VXMT_cboe)[4] = "vxmt_cboe_low"
colnames(VXMT_cboe)[5] = "vxmt_cboe_close"
VXMT_cboe$Date <- mdy(VXMT_cboe$Date)
cols <-c(2:5)
VXMT_cboe[,cols] %<>% lapply(function(x) as.numeric(as.character(x)))

# Join VIX, VIX3m (VXV) and VXMT CBOE data to ETF df
cboe.df <- merge(VIX_cboe,VXV_cboe, by="Date")
cboe.df <- merge(cboe.df,VXMT_cboe, by="Date")
start_date <- cboe.df$Date[1]
# Subset Strategy To Start At VXMT Date 
df <- subset(df, Date >= as.POSIXct(start_date) )
df <- df[,c(-3)] # Drop unused dates
df <- full_join(df, cboe.df, by = c("Date" = "Date"))

####################################################################
# Create VXV VXMT Ratio and ratio SMA 
###################################################################
# VXV / VXMT Ratio 
df$vxv.vxmt.ratio <- df$vxv_cboe_close / df$vxmt_cboe_close
# Calculate Close-to-Close returns
df$vxx.close.ret <- ROC(df$vxx_close, type = c("discrete"))
df$xiv.close.ret <- ROC(df$xiv_close, type = c("discrete"))

# Calculate SMA of VXV , VXMT ratio 
sma.lookback <- 2:170
getSMA <- function(y) {
  function(x) {
    SMA(df[,"vxv.vxmt.ratio"], sma.lookback[i])    # Calls TTR package to create SMA
  }
}
# Create a matrix to put the SMAs in
sma.matrix <- matrix(nrow=nrow(df), ncol=0)
# Loop for filling it
for (i in 1:length(sma.lookback)) {
  sma.matrix <- cbind(sma.matrix, getSMA(i)(x))
}
# Rename columns
colnames(sma.matrix) <- sapply(sma.lookback, function(n)paste("ratio.sma.n", n, sep=""))
# Bind to existing dataframe
df <-  cbind(df, sma.matrix)

####################################################################
# Split Data Into Train and Test Sets
# 252 days train , 126 days test 
##################################################################
# Set row numbers for splitting to train and test sets
total.row <- nrow(df) # total rows
train.length <- 252
test.length <- 125
train.list <- list()
test.list <- list()
# Set start and ending rows for subsetting by row
train_num_set <- seq(1, nrow(df), by=test.length+1)
test_num_set <- seq(train.length+1, nrow(df), by=test.length+1)
train.list<- lapply(train_num_set, function(i) df[c(i:(i+train.length)),])
test.list<- lapply(test_num_set, function(i) df[c(i:(i+test.length)),])


####################################################################
# Run Optimizations over all training windows
# Select optimal sharpe ratio 
####################################################################
train.set.results <- data.frame()
i=1
train.loop.length[i]
train.loop.length <- rep(1:length(test.list),each=length(sma.lookback)) # run through the dfs the same length as the number of smas
sma.loop.length <- rep(2:length(sma.lookback),length(train.list)) # run each sma over every df
train.xts <- list()
# Loop for running all smas through all train sets, note need to wrap train.loop.length and sma.loop.lenght in [[i]]
for (i in 1:length(train.loop.length)) {
  tryCatch({
    df.num <- train.loop.length[[i]]
  # Enter buy / sell rules
  train.temp.df <- data.frame(Date=train.list[[df.num]]$Date)
  train.temp.df$vxx.signal <- ifelse(train.list[[df.num]]$vxv.vxmt.ratio > 1 & train.list[[df.num]]$vxv.vxmt.ratio > train.list[[df.num]][,paste0("ratio.sma.n", sma.loop.length[[i]])], 1,0)
  train.temp.df$xiv.signal <- ifelse(train.list[[df.num]]$vxv.vxmt.ratio < 1 & train.list[[df.num]]$vxv.vxmt.ratio < train.list[[df.num]][,paste0("ratio.sma.n", sma.loop.length[[i]])], 1,0)
  
  # lag signal by two forward days
  # CBOE data is available next day
  train.temp.df$vxx.signal <- lag(train.temp.df$vxx.signal,2) # Note k=1 implies a move *forward*
  train.temp.df$xiv.signal <- lag(train.temp.df$xiv.signal,2) # Note k=1 implies a move *forward*
  
  train.temp.df[is.na(train.temp.df)] <- as.Date(0)  # Set NA to 0
  
  # Calculate equity curves
  train.temp.df$vxx.signal.ret <-   train.temp.df$vxx.signal *  train.list[[df.num]]$vxx.close.ret
  train.temp.df$xiv.signal.ret <-   train.temp.df$xiv.signal *  train.list[[df.num]]$xiv.close.ret
  
  # Combine signals 
   train.temp.df$total.signal.ret <-  train.temp.df$vxx.signal.ret +  train.temp.df$xiv.signal.ret
  
  # Pull select columns from data frame to make XTS whilst retaining formats
  xts1 = xts(train.temp.df$vxx.signal.ret, order.by=as.Date(train.temp.df$Date, format="%y-%m-%d"))
  xts2 = xts(train.temp.df$xiv.signal.ret, order.by=as.Date(train.temp.df$Date, format="%y/%m/%d")) 
  xts3 = xts(train.temp.df$total.signal.ret, order.by=as.Date(train.temp.df$Date, format="%y/%m/%d")) 
  
  # Join XTS together
  compare <- cbind(xts1,xts2,xts3)
  
  # Use the PerformanceAnalytics package for trade statistics
  require(PerformanceAnalytics)
  colnames(compare) <- c("vxx","xiv","combined")
  #charts.PerformanceSummary(compare,main="Long when current month is higher than previous 12 month", wealth.index=TRUE, colorset=rainbow12equal)
  #performance.table <- rbind(table.AnnualizedReturns(compare),maxDrawdown(compare), CalmarRatio(compare),table.DownsideRisk(compare))
  #drawdown.table <- rbind(table.Drawdowns(xts3))
  #dev.off()  
  # logRets <- log(cumprod(1+compare))
  # chart.TimeSeries(logRets, legend.loc='topleft', colorset=rainbow12equal,main="Log Returns")
  
  #print(performance.table)
  #print(drawdown.table)
  cum.ret <- Return.cumulative(xts3, geometric = TRUE)
  annualized <- Return.annualized(xts3, scale = NA, geometric = TRUE)
  dd <- maxDrawdown(xts3)
  sharpe <- SharpeRatio.annualized(xts3, Rf = 0, scale = NA, geometric = TRUE)
  
  
  # Output data
  temp <- data.frame("Annualized Return" = annualized,"Annualized Sharpe" = sharpe,"Cumulative Return" = cum.ret,"Maximum Draw Down" =dd, ID=as.numeric(sma.loop.length[[i]]))
  rownames(temp) <- paste0("train.set.",train.loop.length[[i]],"_sma_",sma.loop.length[[i]])
  train.set.results <- rbind.data.frame(train.set.results,temp)
 compare <- data.frame(compare)
 compare <-  data.frame(date=index(compare), coredata(compare))
 train.xts[[i]] <- cbind.data.frame(compare)
  
  ptm0 <- proc.time()
  Sys.sleep(0.1)  
  ptm1=proc.time() - ptm0
  time=as.numeric(ptm1[3])
  cat('\n','Iteration Train Set',train.loop.length[i],'sma',sma.loop.length[i],'took', time, "seconds to complete")
}, error = function(e) { print(paste("i =", i, "failed:")) })
}


# Seperate train set results from data frame output
train.set.results <- setDT(train.set.results, keep.rownames = TRUE)[] # Set row names
train.set.results$ID <- as.numeric(train.set.results$ID)
# Subset test set results
l <- nrow(train.set.results)
train.result.list <- list()
output_num <- seq(1, l, by=length(sma.lookback))
train.result.list<- lapply(output_num, function(i) train.set.results[c(i:(i+length(sma.lookback)-1)),])

# Extract highest performance metric from each data frame within list
# Sharpe Ratio in this case
highest.sharpe <- list()
i=1
for (i in 1:length(train.result.list)) {  # for every data frame inside results list
  temp <- train.result.list[[i]][order(train.result.list[[i]]$Annualized.Sharpe),]
  highest.sharpe[[i]] <- as.data.frame(tail(temp[,6],1))
}

# Save performance metric
sharpe.df <- do.call(cbind,highest.sharpe)
sharpe.vec <- paste(sharpe.df)
sharpe.vec <- as.numeric(sharpe.vec)
cat("Optimal In-Sample Train Set Parameters:SMA of VXV/VXMT Ratio - Optimal SMAs =",sharpe.vec)

###### Sharpe Optimization Complete ##########

As we can see, there is a large degree of movement in the optimal sma look back period:

cat("Optimal In-Sample Train Set Parameters:SMA of VXV/VXMT Ratio - Optimal SMAs =",sharpe.vec)
Optimal In-Sample Train Set Parameters:SMA of VXV/VXMT Ratio - Optimal SMAs = 35 108 139 138 135 65 63 63 145 2 18 23 21 32 4 13 24 66 136 2

For train set 1, optimal sma of 35, train set 2, optimal sma of 108, train set 3, optimal sma of 139 etc..

These smas produced the highest sharpe ratios during in-sample training. We then use these sma parameters for each out of sample test set. Data that the model has never seen before. The code to run optimized parameters over each test test period:

#################################################################
# Run Optimal Parameters over all train sets 
#################################################################

test.set.results <- data.frame()
test.xts <- list()
test.loop.length <- rep(1:length(test.list),1) # run through the all df's 1x 
optimal.sma <- sharpe.vec # run each sma over every df
i=1
################### Manual Intervention ############################ 
# Remove nas on last test data frame inside list (Note this willl change adjust as needed)
#######################################################################
length.test.set <- length(test.list)
tail(test.list[[length.test.set]])
nrow.test.list <- NROW(test.list[[length.test.set]])
test.list[[length.test.set]] <- test.list[[length.test.set]][1:((nrow.test.list)-5),]
# Loop for running all smas through all train sets, note need to wrap train.loop.length and sma.loop.lenght in [[i]]
for (i in 1:length(test.loop.length )) {
  tryCatch({
    df.num <- test.loop.length[i]
    # Enter buy / sell rules
    test.temp.df <- data.frame(Date=test.list[[df.num]]$Date)
    test.temp.df$vxx.signal <- ifelse(test.list[[df.num]]$vxv.vxmt.ratio > 1 & test.list[[df.num]]$vxv.vxmt.ratio > test.list[[df.num]][,paste0("ratio.sma.n", optimal.sma[[i]])], 1,0)
    test.temp.df$xiv.signal <- ifelse(test.list[[df.num]]$vxv.vxmt.ratio < 1 & test.list[[df.num]]$vxv.vxmt.ratio < test.list[[df.num]][,paste0("ratio.sma.n", optimal.sma[[i]])], 1,0)
    
    # lag signal by two forward days
    # CBOE data is available next day
    test.temp.df$vxx.signal <- lag(test.temp.df$vxx.signal,2) # Note k=1 implies a move *forward*
    test.temp.df$xiv.signal <- lag(test.temp.df$xiv.signal,2) # Note k=1 implies a move *forward*
    
    test.temp.df[is.na(test.temp.df)] <- as.Date(0)  # Set NA to 0
    
    # Calculate equity curves
    test.temp.df$vxx.signal.ret <-   test.temp.df$vxx.signal *  test.list[[df.num]]$vxx.close.ret
    test.temp.df$xiv.signal.ret <-   test.temp.df$xiv.signal *  test.list[[df.num]]$xiv.close.ret
    
    # Combine signals 
    test.temp.df$total.signal.ret <-  test.temp.df$vxx.signal.ret +  test.temp.df$xiv.signal.ret
    
    # Pull select columns from data frame to make XTS whilst retaining formats
    xts1 = xts(test.temp.df$vxx.signal.ret, order.by=as.Date(test.temp.df$Date, format="%y-%m-%d"))
    xts2 = xts(test.temp.df$xiv.signal.ret, order.by=as.Date(test.temp.df$Date, format="%y/%m/%d")) 
    xts3 = xts(test.temp.df$total.signal.ret, order.by=as.Date(test.temp.df$Date, format="%y/%m/%d")) 
    
    # Join XTS together
    compare <- cbind(xts1,xts2,xts3)
    
    # Use the PerformanceAnalytics package for trade statistics
    require(PerformanceAnalytics)
    colnames(compare) <- c("vxx","xiv","combined")
    charts.PerformanceSummary(compare,main="Long when current month is higher than previous 12 month", wealth.index=TRUE, colorset=rainbow12equal)
    #performance.table <- rbind(table.AnnualizedReturns(compare),maxDrawdown(compare), CalmarRatio(compare),table.DownsideRisk(compare))
    #drawdown.table <- rbind(table.Drawdowns(xts3))
    #dev.off()  
    # logRets <- log(cumprod(1+compare))
    # chart.TimeSeries(logRets, legend.loc='topleft', colorset=rainbow12equal,main="Log Returns")
    
    #print(performance.table)
    #print(drawdown.table)
    cum.ret <- Return.cumulative(xts3, geometric = TRUE)
    annualized <- Return.annualized(xts3, scale = NA, geometric = TRUE)
    dd <- maxDrawdown(xts3)
    sharpe <- SharpeRatio.annualized(xts3, Rf = 0, scale = NA, geometric = TRUE)
    
    
    # Save test set results
    temp <- data.frame("Annualized Return" = annualized,"Annualized Sharpe" = sharpe,"Cumulative Return" = cum.ret,"Maximum Draw Down" =dd, ID=as.numeric(optimal.sma[[i]]))
    test.set.results <- rbind.data.frame(test.set.results)
    test.xts[[i]] <- cbind.data.frame(compare)
    
    ptm0 <- proc.time()
    Sys.sleep(0.1)  
    ptm1=proc.time() - ptm0
    time=as.numeric(ptm1[3])
    cat('\n','Iteration Test Set',test.loop.length[i],'optimal.sma',optimal.sma[i],'took', time, "seconds to complete")
  }, error = function(e) { print(paste("i =", i, "failed:")) })
}

# Extract xts3 cumulative returns 
cum.rets <- do.call(rbind,test.xts)
cum.rets <- setDT(cum.rets, keep.rownames = TRUE)[] # Set row names
colnames(cum.rets)[1] <- "Date"
cum.rets[is.na(cum.rets)] <- 0
cum.rets$Date <- ymd(cum.rets$Date)
cum.rets.xts <- xts(cum.rets$combined, order.by=as.Date(cum.rets$Date, format="%y-%m-%d")) 

# Summary Statistic on final walk forward curve 
charts.PerformanceSummary(cum.rets,main="Long when current month is higher than previous 12 month", wealth.index=TRUE, colorset=rainbow12equal)
table.AnnualizedReturns(cum.rets.xts)
maxDrawdown(cum.rets.xts)
CalmarRatio(cum.rets.xts)
table.DownsideRisk(cum.rets.xts)
drawdown.table <- rbind(table.Drawdowns(cum.rets.xts))
drawdown.table 

Here are some performance metrics for all test set periods:

 > table.AnnualizedReturns(cum.rets.xts)

Annualized Return         0.6797
Annualized Std Dev        0.4312
Annualized Sharpe (Rf=0%) 1.5765
> maxDrawdown(cum.rets.xts)
[1] 0.4312712
> CalmarRatio(cum.rets.xts)
                 [,1]
Calmar Ratio 1.576093
> table.DownsideRisk(cum.rets.xts)

Semi Deviation                 0.0192
Gain Deviation                 0.0235
Loss Deviation                 0.0266
Downside Deviation (MAR=210%)  0.0220
Downside Deviation (Rf=0%)     0.0183
Downside Deviation (0%)        0.0183
Maximum Drawdown               0.4313
Historical VaR (95%)          -0.0409
Historical ES (95%)           -0.0677
Modified VaR (95%)            -0.0385
Modified ES (95%)             -0.0621

The strategy reports an annualized return of 68% for all out of sample test periods with an annualized sharpe ratio of 1.57. Note that is data from VXMT inception, 2008 to present day.

Next we can validate our code by creating plots of the test set equity curve and plot the train and test set windows:

# Extract train and test dates for plotting 
# Train Dates
train.dateslist <- list()
train.dates.loop.length <- rep(1:length(train.list),each=1)
temp <- data.frame()
i=1
for (i in 1:length(train.dates.loop.length)) {
  temp <- data.frame("ID"=paste("train.id",train.dates.loop.length[[i]]))
  date <- train.list[[i]]$Date
  train.dateslist[[i]] <- cbind("Date" = date,ID=temp)

# Test Dates
test.dateslist <- list()
test.dates.loop.length <- rep(1:length(test.list),each=1)
i=1
for (i in 1:length(test.dates.loop.length)) {
  temp <- data.frame("ID"=paste("test.id",test.dates.loop.length[[i]]))
  date <- test.list[[i]]$Date
  test.dateslist[[i]] <- cbind("Date" = date,ID=temp)
  }

# Loop to extract start and end dates of each train / test segment 
train.dates.start <- list()
train.dates.end <- list()
for (i in 1:length(train.dateslist)) {
train.dates.start[[i]] <- train.dateslist[[i]]$Date[1]
train.dates.end[[i]] <- tail(train.dateslist[[i]]$Date,1)
}
test.dates.start <- list()
test.dates.end <- list()
for (i in 1:length(test.dateslist)) {
  test.dates.start[[i]] <- test.dateslist[[i]]$Date[1]
  test.dates.end[[i]] <- tail(test.dateslist[[i]]$Date,1)
}

# Plot train and test starting / ending periods
  ggplot()+
    theme_bw()+
    scale_y_continuous(expand = c(0, 0),breaks = round(seq(min(1), max(length(test.dateslist)), by = 1)))+ # set y axis to same amount as total train / test sample periosd
  theme(legend.position = "none")+
  theme(plot.title = element_text(hjust=0.5),plot.subtitle =element_text(hjust=0.5))+
  ggtitle("Train And Test Windows",subtitle=paste("Train Width =",train.length,"Days, Test Width =",test.length,"Days"))+
    geom_rect(aes(xmin=as.Date(train.dates.start[[1]]),xmax=as.Date(train.dates.end[[1]]),ymin=1, ymax=2),fill="#0072B2",alpha=.8)+
    geom_rect(aes(xmin=as.Date(train.dates.start[[2]]),xmax=as.Date(train.dates.end[[2]]),ymin=2, ymax=3),fill="#0072B2",alpha=.8)+
    geom_rect(aes(xmin=as.Date(train.dates.start[[3]]),xmax=as.Date(train.dates.end[[3]]),ymin=3, ymax=4),fill="#0072B2",alpha=.8)+
    geom_rect(aes(xmin=as.Date(train.dates.start[[4]]),xmax=as.Date(train.dates.end[[4]]),ymin=4, ymax=5),fill="#0072B2",alpha=.8)+    
    geom_rect(aes(xmin=as.Date(train.dates.start[[5]]),xmax=as.Date(train.dates.end[[5]]),ymin=5, ymax=6),fill="#0072B2",alpha=.8)+    
    geom_rect(aes(xmin=as.Date(train.dates.start[[6]]),xmax=as.Date(train.dates.end[[6]]),ymin=6, ymax=7),fill="#0072B2",alpha=.8)+    
    geom_rect(aes(xmin=as.Date(train.dates.start[[7]]),xmax=as.Date(train.dates.end[[7]]),ymin=7, ymax=8),fill="#0072B2",alpha=.8)+
    geom_rect(aes(xmin=as.Date(train.dates.start[[8]]),xmax=as.Date(train.dates.end[[8]]),ymin=8, ymax=9),fill="#0072B2",alpha=.8)+
    geom_rect(aes(xmin=as.Date(train.dates.start[[9]]),xmax=as.Date(train.dates.end[[9]]),ymin=9, ymax=10),fill="#0072B2",alpha=.8)+
    geom_rect(aes(xmin=as.Date(train.dates.start[[10]]),xmax=as.Date(train.dates.end[[10]]),ymin=10, ymax=11),fill="#0072B2",alpha=.8)+
    geom_rect(aes(xmin=as.Date(train.dates.start[[11]]),xmax=as.Date(train.dates.end[[11]]),ymin=11, ymax=12),fill="#0072B2",alpha=.8)+
    geom_rect(aes(xmin=as.Date(train.dates.start[[12]]),xmax=as.Date(train.dates.end[[12]]),ymin=12, ymax=13),fill="#0072B2",alpha=.8)+
    geom_rect(aes(xmin=as.Date(train.dates.start[[13]]),xmax=as.Date(train.dates.end[[13]]),ymin=13, ymax=14),fill="#0072B2",alpha=.8)+
    geom_rect(aes(xmin=as.Date(train.dates.start[[14]]),xmax=as.Date(train.dates.end[[14]]),ymin=14, ymax=15),fill="#0072B2",alpha=.8)+
    geom_rect(aes(xmin=as.Date(train.dates.start[[15]]),xmax=as.Date(train.dates.end[[15]]),ymin=15, ymax=16),fill="#0072B2",alpha=.8)+
    geom_rect(aes(xmin=as.Date(train.dates.start[[16]]),xmax=as.Date(train.dates.end[[16]]),ymin=16, ymax=17),fill="#0072B2",alpha=.8)+
    geom_rect(aes(xmin=as.Date(train.dates.start[[17]]),xmax=as.Date(train.dates.end[[17]]),ymin=17, ymax=18),fill="#0072B2",alpha=.8)+
    geom_rect(aes(xmin=as.Date(train.dates.start[[18]]),xmax=as.Date(train.dates.end[[18]]),ymin=18, ymax=19),fill="#0072B2",alpha=.8)+
    geom_rect(aes(xmin=as.Date(test.dates.start[[1]]),xmax=as.Date(test.dates.end[[1]]),ymin=1, ymax=2),fill="#D55E00",alpha=.8)+
    geom_rect(aes(xmin=as.Date(test.dates.start[[2]]),xmax=as.Date(test.dates.end[[2]]),ymin=2, ymax=3),fill="#D55E00",alpha=.8)+
    geom_rect(aes(xmin=as.Date(test.dates.start[[3]]),xmax=as.Date(test.dates.end[[3]]),ymin=3, ymax=4),fill="#D55E00",alpha=.8)+
    geom_rect(aes(xmin=as.Date(test.dates.start[[4]]),xmax=as.Date(test.dates.end[[4]]),ymin=4, ymax=5),fill="#D55E00",alpha=.8)+
    geom_rect(aes(xmin=as.Date(test.dates.start[[5]]),xmax=as.Date(test.dates.end[[5]]),ymin=5, ymax=6),fill="#D55E00",alpha=.8)+
    geom_rect(aes(xmin=as.Date(test.dates.start[[6]]),xmax=as.Date(test.dates.end[[6]]),ymin=6, ymax=7),fill="#D55E00",alpha=.8)+
    geom_rect(aes(xmin=as.Date(test.dates.start[[7]]),xmax=as.Date(test.dates.end[[7]]),ymin=7, ymax=8),fill="#D55E00",alpha=.8)+
    geom_rect(aes(xmin=as.Date(test.dates.start[[8]]),xmax=as.Date(test.dates.end[[8]]),ymin=8, ymax=9),fill="#D55E00",alpha=.8)+
    geom_rect(aes(xmin=as.Date(test.dates.start[[9]]),xmax=as.Date(test.dates.end[[9]]),ymin=9, ymax=10),fill="#D55E00",alpha=.8)+
    geom_rect(aes(xmin=as.Date(test.dates.start[[10]]),xmax=as.Date(test.dates.end[[10]]),ymin=10, ymax=11),fill="#D55E00",alpha=.8)+
    geom_rect(aes(xmin=as.Date(test.dates.start[[11]]),xmax=as.Date(test.dates.end[[11]]),ymin=11, ymax=12),fill="#D55E00",alpha=.8)+
    geom_rect(aes(xmin=as.Date(test.dates.start[[12]]),xmax=as.Date(test.dates.end[[12]]),ymin=12, ymax=13),fill="#D55E00",alpha=.8)+
    geom_rect(aes(xmin=as.Date(test.dates.start[[13]]),xmax=as.Date(test.dates.end[[13]]),ymin=13, ymax=14),fill="#D55E00",alpha=.8)+
    geom_rect(aes(xmin=as.Date(test.dates.start[[14]]),xmax=as.Date(test.dates.end[[14]]),ymin=14, ymax=15),fill="#D55E00",alpha=.8)+
    geom_rect(aes(xmin=as.Date(test.dates.start[[15]]),xmax=as.Date(test.dates.end[[15]]),ymin=15, ymax=16),fill="#D55E00",alpha=.8)+
    geom_rect(aes(xmin=as.Date(test.dates.start[[16]]),xmax=as.Date(test.dates.end[[16]]),ymin=16, ymax=17),fill="#D55E00",alpha=.8)+
    geom_rect(aes(xmin=as.Date(test.dates.start[[17]]),xmax=as.Date(test.dates.end[[17]]),ymin=17, ymax=18),fill="#D55E00",alpha=.8)+
    geom_rect(aes(xmin=as.Date(test.dates.start[[18]]),xmax=as.Date(test.dates.end[[18]]),ymin=18, ymax=19),fill="#D55E00",alpha=.8)+
    labs(x="Date",y="Train / Test Set No.")
    
    
# Extract final equity curve from test set 
  final.plot <- list()
  i=1
  for (i in 1:length(test.xts)) {
    to.sum <- data.frame("to.sum"=test.xts[[i]]$combined)
    dates <- test.list[[i]]$Date
      final.plot[[i]] <- data.frame("Date"=dates,to.sum)
  }

  # Make df of output 
  final.df <- do.call(rbind,final.plot)
  # Cumsum equity curve
final.df$equity <- cumprod(final.df$to.sum + 1) -1 
  final.df <- final.df[c(1,3) ]
  # Melt df 
  final.plot.df <- melt(data = final.df,id.vars = 'Date')   # melt df for plotting with ggplot2
  # Create ID
  final.plot.df$final_num <- as.numeric(rep(1:test.length+1, each=test.length+1, length.out=nrow(final.plot.df)))
  # Subset plots to plot each test period a different colour
  subset.plots <- list()
  test.width <- 125
  subset.plots<- lapply(row_nums, function(i) final.plot.df[c(i:(i+test.length)),])
 # Create plot 
  # Create colour scheme
  myPalette <- colorRampPalette(rev(brewer.pal(11, "Paired")))
  sc <- scale_colour_gradientn(colours = myPalette(100), limits=c(1, 18))
  # Plot out of sample test results equity curve                           
  ggplot(data = subset.plots[[1]], aes(x =Date, y = value,colour=final_num))+
    geom_line(data = subset.plots[[2]], aes(x =Date, y = value))+
    geom_line(data = subset.plots[[3]], aes(x =Date, y = value,colour=final_num))+
    geom_line(data = subset.plots[[4]], aes(x =Date, y = value,colour=final_num))+
    geom_line(data = subset.plots[[5]], aes(x =Date, y = value,colour=final_num))+
    geom_line(data = subset.plots[[6]], aes(x =Date, y = value,colour=final_num))+
    geom_line(data = subset.plots[[7]], aes(x =Date, y = value,colour=final_num))+
    geom_line(data = subset.plots[[8]], aes(x =Date, y = value,colour=final_num))+
    geom_line(data = subset.plots[[9]], aes(x =Date, y = value,colour=final_num))+
    geom_line(data = subset.plots[[10]], aes(x =Date, y = value,colour=final_num))+
    geom_line(data = subset.plots[[11]], aes(x =Date, y = value,colour=final_num))+
    geom_line(data = subset.plots[[12]], aes(x =Date, y = value,colour=final_num))+
    geom_line(data = subset.plots[[13]], aes(x =Date, y = value,colour=final_num))+
    geom_line(data = subset.plots[[14]], aes(x =Date, y = value,colour=final_num))+
    geom_line(data = subset.plots[[15]], aes(x =Date, y = value,colour=final_num))+
    geom_line(data = subset.plots[[16]], aes(x =Date, y = value,colour=final_num))+
    geom_line(data = subset.plots[[17]], aes(x =Date, y = value,colour=final_num))+
    geom_line(data = subset.plots[[18]], aes(x =Date, y = value,colour=final_num))+
    sc+ 
    theme_bw()+
    theme(legend.position = "none")+
    scale_y_continuous(breaks = seq(1, 110, by = 5))+
    ggtitle("Test Set Results", subtitle =paste("Test Set Lengths =",test.length," rows")) +
    labs(x="Date",y="Cumulative Return")+
    theme(plot.title = element_text(hjust=0.5),plot.subtitle =element_text(hjust=0.5))

  
Rplot204
Train and Test Sets – Train = 252 trading days. Test = 126 days.

The plot above was created using the same data created by subsetting the rows into train and test set periods and stitching them back together. The blue periods indicate in-sample test periods and the orange represent out of sample test periods. The final equity curve below is the result of all test set periods.

cumretrs
Cumulative Returns All Test Set Periods

Each color on the equity curve represents each test set period, 126 trading days long.

To conclude, the benefit of this type of cross validation is that it demonstrates robustness. Each in sample parameter optimization is subsequently used in the out of sample period. This is data that the model has never seen before. If an optimized parameter fell apart in out of sample testing, we know we have over fit in some way or another, or the edge itself has disappeared. Failing this test would be cause to abandon the strategy.

Thanks for reading!

Full code can be found on my github.

Robustness Testing – Volatility Strategy – Time Series Bootstrapping

In this post we will look at a simple volatility strategy. We will optimize the parameters over a back test and select a region of parameters which show the best performance. We will then subject this optimal performance band to further testing by resampling the original time series using the meboot package.

Traditionally bootstrapping is usually performed on data that is independent and identically distributed which of course it not the case with time series data due to the presence of auto correlation, if auto correlation is present then time series is dependent on its past. For lag 1 correlation at varying time periods x is the correlation between, (x4,x3),(x3,x2),(x2,x1),…). and for lag 2 correlation between time x, (x5,x3),(x4,x2),(x3,x1),…) so on and so forth for the entire sample at varying time lags.

acf xiv

This is the auto correlation of XIV. We may remove the auto correlation by differencing, Xt−Xt−1:

Rplot153One may notice there is no trend, the series is stationary. We notice the variance has increased over time.

mean(df$xiv.close.diff, na.rm = TRUE)
[1] 0.050235
> var(df$xiv.close.diff, na.rm = TRUE)
[1] 1.299783

We may stabilize the variance by taking the log of the close and differencing.

Rplot209

mean(df$log.xiv.close.diff, na.rm = TRUE)
[1] 0.001229805
> var(df$log.xiv.close.diff, na.rm = TRUE)
[1] 0.001382537

 

Rplot154

With the auto correlation removed, one may attempt to perform traditional methods of bootstrapping.

With stationary series, the mean is constant over time, there is no trend. An idea I was experimenting with was differencing as per above, and in an attempt to retain the correlation structure of the original series, block bootstrapping through resample and replace and stitch the series back together. As the resampled series is still stationary, cumulatively sum the output in order to create a trending series. This would then serve as a new series to run new back tests on.  See my attempt here.

I so far have opted to use the meboot (maximum entropy bootstrap) R package which does not require the series to be stationary, it does not require block lengths as a parameter and it keeps the correlation structure of the data intact.

We will need XIV, VXX, CBOE VXV and CBOE VXMT data, I also use sythnetic vxx and xiv data stored on my HD, feel free to edit these portions to include the locations of your own synthetic (if you choose to want more history).

Strategy rules as follows:

  1. long VXX when VXV/VXMT ratio is over 1 & the VXV/VXMT ratio is over an ‘n’ period sma of the VXV/VXMT ratio.
  2. long XIV when VXV/VXMT ratio is less than 1 & the VXV/VXMT ratio is below an ‘n’ period sma of the VXV/VXMT ratio.

We have two free parameters here. The ratio value currently set to 1 and the lookback period for the vxv/vxmt sma ratio.

We optimize the vxv/vxmt sma ratio from sma 2:300 and plot the annualized return for each sma period, the code to do this:

Note im using alphavantage API, you will need your api key…

# Meboot time series resampling
# Andrew Bannerman 12.29.2017

require(xts)
require(data.table)
require(ggplot2)
require(lubridate)
require(magrittr)
require(scales)
require(reshape2)
require(PerformanceAnalytics)
require(dplyr)
require(TTR)
require(meboot)
require(np)

######### Download ETF Data ###############
# Load Syntehtic and join to alpha vantage adjusted prices
# Load synthetic VXX and XIV data
library(readxl)
synth <- read_excel("D:/R Projects/Final Scripts/VIX_term_structure/vix-funds-models-no-formulas.xls", col_names = TRUE)
synth1 <- read_excel("D:/R Projects/Final Scripts/VIX_term_structure/vix-mt-funds-models-no-formulas.xls", col_names = TRUE)
synth <- as.data.frame(synth)
synth1 <- as.data.frame(synth1)
# Extract synthetic series
vxx.synth <- data.frame(synth$Date, synth$'VXX calc')
xiv.synth <- data.frame(synth$Date, synth$'XIV calc')
ziv.synth <- data.frame(synth1$Date, synth1$'ZIV calc')
vxz.synth <- data.frame(synth1$Date, synth1$'VXZ calc')
colnames(vxx.synth)[1] <- "Date"
colnames(vxx.synth)[2] <- "vxx_close"
colnames(xiv.synth)[1] <- "Date"
colnames(xiv.synth)[2] <- "xiv_close"
colnames(ziv.synth)[1] <- "Date"
colnames(ziv.synth)[2] <- "ziv_close"
colnames(vxz.synth)[1] <- "Date"
colnames(vxz.synth)[2] <- "vxz_close"
vxx.synth$Date <- ymd(vxx.synth$Date)
xiv.synth$Date <- ymd(xiv.synth$Date)
ziv.synth$Date <- ymd(ziv.synth$Date)
vxz.synth$Date <- ymd(vxz.synth$Date)

# Download SPY data
# Note you need tyo place your API key...your_key_here
SPY <- fread("https://www.alphavantage.co/query?function=TIME_SERIES_DAILY_ADJUSTED&symbol=SPY&outputsize=full&apikey=your_api_key&datatype=csv") #fread() data.table for downloading directly to a data frame
SPY$timestamp <- ymd(SPY$timestamp)   #Lubridate to change character date to date format
SPY <- arrange(SPY,timestamp)   #dplyr to sort data frame by date ascending order
colnames(SPY)[1] <- "Date"
SPY$Date <- ymd(SPY$Date)
SPY <- as.data.frame(SPY)
SPY <- subset(SPY, Date >= as.POSIXct("2004-03-26") ) # synthetic data start
head(SPY)

VXX <- fread("https://www.alphavantage.co/query?function=TIME_SERIES_DAILY_ADJUSTED&symbol=VXX&outputsize=full&apikey=your_api_key&datatype=csv") #fread() data.table for downloading directly to a data frame
VXX$timestamp <- ymd(VXX$timestamp)   #Lubridate to change character date to date format
VXX <- arrange(VXX,timestamp)   #dplyr to sort data frame by date ascending order
colnames(VXX)[1] <- "Date"
VXX$Date <- ymd(VXX$Date)
VXX <- as.data.frame(VXX)
head(VXX)

XIV <- fread("https://www.alphavantage.co/query?function=TIME_SERIES_DAILY_ADJUSTED&symbol=XIV&outputsize=full&apikey=your_api_key&datatype=csv") #fread() data.table for downloading directly to a data frame
XIV$timestamp <- ymd(XIV$timestamp)   #Lubridate to change character date to date format
XIV <- arrange(XIV,timestamp)   #dplyr to sort data frame by date ascending order
colnames(XIV)[1] <- "Date"
XIV$Date <- ymd(XIV$Date)
XIV <- as.data.frame(XIV)
head(XIV)
#XIV <- subset(XIV, Date >= as.POSIXct("2012-01-01"))

ZIV <- fread("https://www.alphavantage.co/query?function=TIME_SERIES_DAILY_ADJUSTED&symbol=ZIV&outputsize=full&apikey=your_api_key&datatype=csv") #fread() data.table for downloading directly to a data frame
ZIV$timestamp <- ymd(ZIV$timestamp)   #Lubridate to change character date to date format
ZIV <- arrange(ZIV,timestamp)   #dplyr to sort data frame by date ascending order
colnames(ZIV)[1] <- "Date"
ZIV$Date <- ymd(ZIV$Date)
ZIV <- as.data.frame(ZIV)
head(ZIV)

VXZ <- fread("https://www.alphavantage.co/query?function=TIME_SERIES_DAILY_ADJUSTED&symbol=VXZ&outputsize=full&apikey=your_api_key&datatype=csv") #fread() data.table for downloading directly to a data frame
VXZ$timestamp <- ymd(VXZ$timestamp)   #Lubridate to change character date to date format
VXZ <- arrange(VXZ,timestamp)   #dplyr to sort data frame by date ascending order
colnames(VXZ)[1] <- "Date"
VXZ$Date <- ymd(VXZ$Date)
VXZ <- as.data.frame(VXZ)
tail(VXZ)

# Join sythentic data to alpha vantage
vxx.synth <- subset(vxx.synth, Date <= as.POSIXct("2009-01-29"))
xiv.synth <- subset(xiv.synth, Date <= as.POSIXct("2010-11-29"))
ziv.synth <- subset(ziv.synth, Date <= as.POSIXct("2010-11-29"))
vxz.synth <- subset(vxz.synth, Date <= as.POSIXct("2009-02-19"))
# Subset only date and close from alpha vantage data
VXX <- VXX[ -c(2:5, 7:9) ]  # subset adjusted close
XIV <- XIV[ -c(2:5, 7:9) ]  # subset adjusted close
ZIV <- ZIV[ -c(2:5, 7:9) ]  # subset adjusted close
VXZ <- VXZ[ -c(2:5, 7:9) ]  # subset adjusted close
SPY <- SPY[ -c(3:5, 7:9) ]  # subset adjusted close
colnames(VXX)[2] <- "vxx_close"
colnames(XIV)[2] <- "xiv_close"
colnames(ZIV)[2] <- "ziv_close"
colnames(VXZ)[2] <- "vxz_close"
colnames(SPY)[2] <- "spy_open"
colnames(SPY)[3] <- "spy_close"

# row bind
VXX <- rbind(vxx.synth,VXX)
XIV <- rbind(xiv.synth,XIV)
ZIV <- rbind(ziv.synth,ZIV)
VXZ <- rbind(vxz.synth,VXZ)
df <- cbind(VXX,XIV,ZIV,VXZ,SPY)
tail(df)

# Download Spot VIX Price and VXV Price from CBOE website
VIX_cboe <- fread("http://www.cboe.com/publish/scheduledtask/mktdata/datahouse/vixcurrent.csv")
VIX_cboe <- as.data.frame(VIX_cboe)
VIX_cboe <- VIX_cboe[2:nrow(VIX_cboe), ]
colnames(VIX_cboe)[1] = "Date"
colnames(VIX_cboe)[2] = "vix_open"
colnames(VIX_cboe)[3] = "vix_high"
colnames(VIX_cboe)[4] = "vix_low"
colnames(VIX_cboe)[5] = "vix_close"
VIX_cboe$Date <- mdy(VIX_cboe$Date)
cols <-c(2:5)
VIX_cboe[,cols] %<>% lapply(function(x) as.numeric(as.character(x)))
VIX_cboe <- subset(VIX_cboe, Date >= as.POSIXct("2007-12-04") )
# Download VXV Data From CBOE website
VXV_cboe <- fread("http://www.cboe.com/publish/scheduledtask/mktdata/datahouse/vix3mdailyprices.csv")
VXV_cboe <- as.data.frame(VXV_cboe)
VXV_cboe <- VXV_cboe[3:nrow(VXV_cboe), ]
colnames(VXV_cboe)[1] = "Date"
colnames(VXV_cboe)[2] = "vxv_cboe_open"
colnames(VXV_cboe)[3] = "vxv_cboe_high"
colnames(VXV_cboe)[4] = "vxv_cboe_low"
colnames(VXV_cboe)[5] = "vxv_cboe_close"
VXV_cboe$Date <- mdy(VXV_cboe$Date)
cols <-c(2:5)
VXV_cboe[,cols] %<>% lapply(function(x) as.numeric(as.character(x)))

# Download VXMT Data from CBOE website
VXMT_cboe <- fread("http://www.cboe.com/publish/scheduledtask/mktdata/datahouse/vxmtdailyprices.csv")
VXMT_cboe <- as.data.frame(VXMT_cboe)
VXMT_cboe <- VXMT_cboe[3:nrow(VXMT_cboe), ]
colnames(VXMT_cboe)[1] = "Date"
colnames(VXMT_cboe)[2] = "vxmt_cboe_open"
colnames(VXMT_cboe)[3] = "vxmt_cboe_high"
colnames(VXMT_cboe)[4] = "vxmt_cboe_low"
colnames(VXMT_cboe)[5] = "vxmt_cboe_close"
VXMT_cboe$Date <- mdy(VXMT_cboe$Date)
cols <-c(2:5)
VXMT_cboe[,cols] %<>% lapply(function(x) as.numeric(as.character(x)))

# Join VIX, VIX3m (VXV) and VXMT CBOE data to ETF df
cboe.df <- merge(VIX_cboe,VXV_cboe, by="Date")
cboe.df <- merge(cboe.df,VXMT_cboe, by="Date")
tail(cboe.df)
df <- df[,c(-3,-5,-7,-9)] # Drop unused dates
df <- full_join(df, cboe.df, by = c("Date" = "Date"))

# Remove last rows
nrow <- NROW(df)-2
df <- head(df,nrow)
tail(df)

############################################
# Back Test VXV / VXMT Ratio
# Find best params without bootstrapping
############################################
# Calculate Close-to-Close returns
df$vxx.close.ret <- ROC(df$vxx_close, type = c("discrete"))
df$xiv.close.ret <- ROC(df$xiv_close, type = c("discrete"))

# VXV / VXMT Ratio
df$vxv.vxmt.ratio <- df$vxv_cboe_close / df$vxmt_cboe_close
df[is.na(df)] <- 0
head(df)

# Calculate SMA of ratio
numdays <- 2:500
getSMA <- function(numdays) {
  function(df) {
    SMA(df[,"vxv.vxmt.ratio"], numdays)    # Calls TTR package to create SMA
  }
}
# Create a matrix to put the SMAs in
sma.matrix <- matrix(nrow=nrow(df), ncol=0)

# Loop for filling it
for (i in numdays) {
  sma.matrix <- cbind(sma.matrix, getSMA(i)(df))
}

# Rename columns
colnames(sma.matrix) <- sapply(numdays, function(n)paste("ratio.sma.n", n, sep=""))

# Bind to existing dataframe
df <-  cbind(df, sma.matrix)
tail(df)

##############################################
# Optimize Strategy Params
##############################################
num.days <- 2:300
i=3
# Initialize data frame
data_output_df <- data.frame()
# Optimize #########
optIMIZE = function(x){
  #spx.sma <- df[,paste0("close.sma.n", sma[i])]
  names(df)
  # Enter buy / sell rules
  #df$vxx.signal <- ifelse(df$vxv.vxmt.ratio > 1 & df$vxv.vxmt.ratio > df$ratio.sma , 1,0)
  #df$xiv.signal <- ifelse(df$vxv.vxmt.ratio < 1 & df$vxv.vxmt.ratio < df$ratio.sma , 1,0)
  df$vxx.signal <- ifelse(df$vxv.vxmt.ratio > 1 & df$vxv.vxmt.ratio > df[,paste0("ratio.sma.n", num.days[i])], 1,0)
  df$xiv.signal <- ifelse(df$vxv.vxmt.ratio < 1 & df$vxv.vxmt.ratio < df[,paste0("ratio.sma.n", num.days[i])], 1,0)

  # lag signal by two forward days
  # CBOE data is available next day
  df$vxx.signal <- lag(df$vxx.signal,2) # Note k=1 implies a move *forward*
  df$xiv.signal <- lag(df$xiv.signal,2) # Note k=1 implies a move *forward*

  df[is.na(df)] <- 0  # Set NA to 0

  # Calculate equity curves
  df$vxx.signal.ret <-  df$vxx.signal * df$vxx.close.ret
  df$xiv.signal.ret <-  df$xiv.signal * df$xiv.close.ret

  # Combine signals
  df$total.signal.ret <- df$vxx.signal.ret + df$xiv.signal.ret

  # Pull select columns from data frame to make XTS whilst retaining formats
  xts1 = xts(df$vxx.signal.ret, order.by=as.Date(df$Date, format="%m/%d/%Y"))
  xts2 = xts(df$xiv.signal.ret, order.by=as.Date(df$Date, format="%m/%d/%Y"))
  xts3 = xts(df$total.signal.ret, order.by=as.Date(df$Date, format="%m/%d/%Y"))
  tail(xts3)

  # Join XTS together
  compare <- cbind(xts1,xts2,xts3)

  # Use the PerformanceAnalytics package for trade statistics

  require(PerformanceAnalytics)
  colnames(compare) <- c("vxx","xiv","combined")
  #charts.PerformanceSummary(compare,main="Long when current month is higher than previous 12 month", wealth.index=TRUE, colorset=rainbow12equal)
  # performance.table <- rbind(table.AnnualizedReturns(compare),maxDrawdown(compare), CalmarRatio(compare),table.DownsideRisk(compare))
  # drawdown.table <- rbind(table.Drawdowns(xts3))
  #dev.off()
  # logRets <- log(cumprod(1+compare))
  # chart.TimeSeries(logRets, legend.loc='topleft', colorset=rainbow12equal,main="Log Returns")

  #print(performance.table)
  #print(drawdown.table)
  cum.ret <- Return.cumulative(xts3, geometric = TRUE)
  annualized <- Return.annualized(xts3, scale = NA, geometric = TRUE)
  dd <- maxDrawdown(xts3)
  sharpe <- SharpeRatio.annualized(xts3, Rf = 0, scale = NA, geometric = TRUE)

  # Create data output of rep and close.diff columns rbind
  data_output_df <- data.frame("Annualized Return" = annualized,"Annualized Sharpe" = sharpe,"Cumulative Return" = cum.ret,"Maximum Draw Down" = dd)

}

for (i in 1:length(num.days)){    # Length of optimization
  tryCatch({
    temp <- optIMIZE(num.days[[i]])
    rownames(temp) <- paste0("",num.days[i])
    #cum_ret <- rbind.data.frame(cum_ret, temp)
    data_output_df <- rbind.data.frame(data_output_df,temp)
    ptm0 <- proc.time()
    Sys.sleep(0.1)
    ptm1=proc.time() - ptm0
    time=as.numeric(ptm1[3])
    cat('\n','Iteration',i,'took', time, "seconds to complete")
  }, error = function(e) { print(paste("i =", i, "failed:")) })
}

# Join SMA number to data frame
data_output_df <- data.frame(data_output_df,num.days)
show <- data_output_df[51:70,]

# Plot
colnames(data_output_df)
library(ggplot2)
ggplot(data=data_output_df, aes(x=num.days,Annualized.Return))+
  geom_bar(stat="identity")+
  theme_classic()+
  scale_x_continuous(breaks = seq(min(data_output_df$num.days), max(data_output_df$num.days)))+
  theme(axis.text.x = element_text(angle = 90, hjust = 0.5,size=15))+
  ggtitle("VXV/VXMT Volatility Strategy - Optimized sma vzx/vxmt ratio",subtitle="2008 to present")+
  theme(plot.title = element_text(hjust=0.5),plot.subtitle =element_text(hjust=0.5))+
  geom_rect(aes(xmin=51,xmax=70,ymin=0,ymax=Inf),alpha=0.01,fill="#CC6666")+
  geom_rect(aes(xmin=117,xmax=161,ymin=0,ymax=Inf),alpha=0.01,fill="#CC6666")+
  annotate("text", label = "sma 51 to 70", x = 60, y = .65, color = "red")+
  annotate("text", label = "sma 117 to 161", x = 135, y = .65, color = "red")
Rplot162
Annualized Returns

We see two parameter bands so to speak using sma look back periods of 51 to 70 and 117 to 161.  Those are sma’s which provided annualized returns, since 2008, of over 50%.

This is considered cross sectional stability, see Tony Copper comment here however, one issue is that these adjacent areas have correlated trades and perhaps would need to scrutinize the returns on a time series basis to rule out lucky periods or months.

Next, we select the optimzed sma band range of sma 51 to 70 and run this over the resampled time series. This is done with this code:

############################
# Meboot
# Generate Maximum Entropy Bootstrapped Time Series Ensemble
############################
# XIV
df[is.na(df)] <- 0
df <- head(df,NROW(df)-1) # Remove missng value...
tail(df$xiv_close,10)
xiv.boot <- meboot(df$xiv_close, reps=100, trim=list(trim=0.10, xmin=NULL, xmax=NULL), reachbnd=FALSE,
               expand.sd=TRUE, force.clt=FALSE, scl.adjustment = FALSE, sym = FALSE,
               elaps=TRUE, colsubj, coldata, coltimes)
# Place meboot results in data frame
xiv.ensemble.df <- data.frame(xiv.boot$ensemble)
xiv.ensemble.df <- data.frame(xiv.ensemble.df,"Date"=df$Date)
# Melt for plotting
xiv.plot.df <- melt(xiv.ensemble.df,id.vars = "Date")
# Plot ggplot2
ggplot(data = xiv.plot.df, aes(x=Date,y=value))+
  geom_line(aes(group = variable))+
  theme_classic()+
  theme(legend.position = "none")+
  geom_line(data=df,aes(x=Date,y=xiv_close,colour="red"))+
  theme(plot.title = element_text(hjust=0.5),plot.subtitle =element_text(hjust=0.5))+
  ggtitle("Resampled Time Series - XIV",subtitle="100 Iterations")

# VXX
vxx.boot <- meboot(df$vxx_close, reps=100, trim=list(trim=0.10, xmin=NULL, xmax=NULL), reachbnd=FALSE,
                   expand.sd=TRUE, force.clt=FALSE, scl.adjustment = FALSE, sym = FALSE,
                   elaps=TRUE, colsubj, coldata, coltimes)
# Place meboot results in data frame
vxx.ensemble.df <- data.frame(vxx.boot$ensemble)
vxx.ensemble.df <- data.frame(vxx.ensemble.df,"Date"=df$Date)
# Melt for plotting
vxx.plot.df <- melt(vxx.ensemble.df,id.vars = "Date")
# Plot ggplot2
ggplot(data = vxx.plot.df, aes(x=Date,y=value))+
  geom_line(aes(group = variable))+
  theme_classic()+
  theme(legend.position = "none")+
  geom_line(data=df,aes(x=Date,y=vxx_close,colour="red"))+
  theme(plot.title = element_text(hjust=0.5),plot.subtitle =element_text(hjust=0.5))+
  ggtitle("Resampled Time Series - VXX", subtitle="100 Iterations")

# VXV
vxv.boot <- meboot(df$vxv_cboe_close, reps=100, trim=list(trim=0.10, xmin=NULL, xmax=NULL), reachbnd=FALSE,
                   expand.sd=TRUE, force.clt=FALSE, scl.adjustment = FALSE, sym = FALSE,
                   elaps=TRUE, colsubj, coldata, coltimes)
# Place meboot results in data frame
vxv.ensemble.df <- data.frame(vxv.boot$ensemble)
vxv.ensemble.df <- data.frame(vxv.ensemble.df,"Date"=df$Date)
# Melt for plotting
vxv.plot.df <- melt(vxv.ensemble.df,id.vars = "Date")
# Plot ggplot2
ggplot(data = vxv.plot.df, aes(x=Date,y=value))+
  geom_line(aes(group = variable))+
  theme_classic()+
  theme(legend.position = "none")+
  geom_line(data=df,aes(x=Date,y=vxv_cboe_close,colour="red"))+
  theme(plot.title = element_text(hjust=0.5),plot.subtitle =element_text(hjust=0.5))+
  ggtitle("Resampled Time Series - VXV", subtitle="100 Iterations")

# VXMT
vxmt.boot <- meboot(df$vxmt_cboe_close, reps=100, trim=list(trim=0.10, xmin=NULL, xmax=NULL), reachbnd=FALSE,
                   expand.sd=TRUE, force.clt=FALSE, scl.adjustment = FALSE, sym = FALSE,
                   elaps=TRUE, colsubj, coldata, coltimes)
# Place meboot results in data frame
vxmt.ensemble.df <- data.frame(vxmt.boot$ensemble)
vxmt.ensemble.df <- data.frame(vxmt.ensemble.df,"Date"=df$Date)
# Melt for plotting
vxmt.plot.df <- melt(vxmt.ensemble.df,id.vars = "Date")
# Plot ggplot2
ggplot(data = vxmt.plot.df, aes(x=Date,y=value))+
  geom_line(aes(group = variable))+
  theme_classic()+
  theme(legend.position = "none")+
  geom_line(data=df,aes(x=Date,y=vxmt_cboe_close,colour="red"))+
  theme(plot.title = element_text(hjust=0.5),plot.subtitle =element_text(hjust=0.5))+
  ggtitle("Resampled Time Series - VXMT")

An example of a resample series using meboot:

Rplot163
XIV – Resampled 100 times

One thing that is noticeable is the the resampled series is tightly bound to the original series. In this example this is 100 iterations for the sake of illustration purposes, one would create upwards of a 1000 re-sampled series.

Next course of action is to back test the original strategy using the optimal parameter band of vxv/vxmt sma ratio 51 to 70. In this example the strategy will run from sma 51 over all 100 series to sma 70. We will record the results in a list and then plot the distribution of results over key metrics: Annualized returns, maximum draw down and annualized sharpe ratio. The code that achieves this:

#################################
# Back test optimal band found in initial sample over boot strapped series
##################################
df.boot <- data.frame("Date" = xiv.ensemble.df$Date)
# boot strap sample series results
# Initialize list
boot_output <- list()
#Drop dates (not used for plotting)
xiv.ensemble.df <- xiv.ensemble.df[,-101]
vxx.ensemble.df <- vxx.ensemble.df[,-101]
vxv.ensemble.df <- vxv.ensemble.df[,-101]
vxmt.ensemble.df <- vxmt.ensemble.df[,-101]

i=1
sma <- rep(51:70,each=length(xiv.ensemble.df))
length.dfs <- rep(1:100,11)
for (i in 1:length(length.dfs)) {
    tryCatch({
  ############################################
  # Back Test VXV / VXMT Ratio Over Boot Strapped Series
  ############################################
  # Calculate Close-to-Close returns
  df.boot$vxx.close.ret <- ROC(vxx.ensemble.df[,length.dfs[i]], type = c("discrete"))
  df.boot$xiv.close.ret <- ROC(xiv.ensemble.df[,length.dfs[i]], type = c("discrete"))

  # VXV / VXMT Ratio
  df.boot$vxv.vxmt.ratio <- vxv.ensemble.df[,length.dfs[i]]/ vxmt.ensemble.df[,length.dfs[i]]
  df.boot$vxv.vxmt.ratio[is.na(df.boot$vxv.vxmt.ratio)] <- 0
  df.boot$vxv.vxmt.ratio[is.nan(df.boot$vxv.vxmt.ratio)] <- 0
  df.boot$vxv.vxmt.ratio[is.infinite(df.boot$vxv.vxmt.ratio)] <- 0

  # Create sma
  df.boot$sma <- SMA(df.boot[,"vxv.vxmt.ratio"], sma[i])    # Calls TTR package to create SMA

    # Enter buy / sell rules
  sma.n <- sma[i]
    #df.boot$vxx.signal <- ifelse(df.boot$vxv.vxmt.ratio > 1 & df.boot$vxv.vxmt.ratio > df.boot[,paste0("ratio.sma.n", sma.n)], 1,0)
  df.boot$vxx.signal <- ifelse(df.boot$vxv.vxmt.ratio > 1 & df.boot$vxv.vxmt.ratio > df.boot$sma, 1,0)
  df.boot$xiv.signal <- ifelse(df.boot$vxv.vxmt.ratio < 1 & df.boot$vxv.vxmt.ratio < df.boot$sma, 1,0)

    # lag signal by two forward days
    # CBOE data is available next day
    df.boot$vxx.signal <- lag(df.boot$vxx.signal,2) # Note k=1 implies a move *forward*
    df.boot$xiv.signal <- lag(df.boot$xiv.signal,2) # Note k=1 implies a move *forward*

    # Calculate equity curves
    df.boot$vxx.signal.ret <-  df.boot$vxx.signal * df.boot$vxx.close.ret
    df.boot$xiv.signal.ret <-  df.boot$xiv.signal * df.boot$xiv.close.ret

    # Combine signals
    df.boot$total.signal.ret <- df.boot$vxx.signal.ret + df.boot$xiv.signal.ret

    # Pull select columns from data frame to make XTS whilst retaining formats
    xts1 = xts(df.boot$vxx.signal.ret, order.by=as.Date(df.boot$Date, format="%m/%d/%Y"))
    xts2 = xts(df.boot$xiv.signal.ret, order.by=as.Date(df.boot$Date, format="%m/%d/%Y"))
    xts3 = xts(df.boot$total.signal.ret, order.by=as.Date(df.boot$Date, format="%m/%d/%Y"))
    tail(xts3)

    # Join XTS together
    compare <- cbind(xts1,xts2,xts3)

    # Use the PerformanceAnalytics package for trade statistics

    require(PerformanceAnalytics)
    colnames(compare) <- c("vxx","xiv","combined")
    #charts.PerformanceSummary(compare,main="Long when current month is higher than previous 12 month", wealth.index=TRUE, colorset=rainbow12equal)
    # performance.table <- rbind(table.AnnualizedReturns(compare),maxDrawdown(compare), CalmarRatio(compare),table.DownsideRisk(compare))
    # drawdown.table <- rbind(table.Drawdowns(xts3))
    #dev.off()
    # logRets <- log(cumprod(1+compare))
    # chart.TimeSeries(logRets, legend.loc='topleft', colorset=rainbow12equal,main="Log Returns")

    #print(performance.table)
    #print(drawdown.table)
    cum.ret <- Return.cumulative(xts3, geometric = TRUE)
    annualized <- Return.annualized(xts3, scale = NA, geometric = TRUE)
    dd <- maxDrawdown(xts3)
    sharpe <- SharpeRatio.annualized(xts3, Rf = 0, scale = NA, geometric = TRUE)
    id <- paste0("col",length.dfs[i],"sma",sma[i])

    # Create data output of rep and close.diff columns rbind
   out <- data.frame("Annualized Return" = annualized,"Annualized Sharpe" = sharpe,"Cumulative Return" = cum.ret,"Maximum Draw Down" = dd, id = id)
   rownames(out) <- paste0("col",length.dfs[i],"sma",sma[i])
    boot_output[[i]] <- rbind(out)

     ptm0 <- proc.time()
    Sys.sleep(0.1)
    ptm1=proc.time() - ptm0
    time=as.numeric(ptm1[3])
    cat('\n','Iteration',i,'took', time, "seconds to complete")
  }, error = function(e) { print(paste("i =", i, "failed:")) })
}

  # Join  boot output
  master <- do.call(rbind, boot_output)
  names(master)

  # Find confidence intervals
  annualized.ret.mean <- mean(master$Annualized.Return)
  annualized.ret.sdev <- sd(master$Annualized.Return)
  annualized.ret.sample.size <- nrow(master)
  annualized.error <- qnorm(0.975)* annualized.ret.sdev/sqrt( annualized.ret.sample.size)
  annualized.left <- annualized.ret.mean - annualized.error
  annualized.right <- annualized.ret.mean + annualized.error

  #annualized.SE <- annualized.ret.sdev / sqrt(annualized.ret.sample.size)
  #t.test(master$Annualized.Sharpe,conf.level = 0.98)

  annualized.sharpe.mean <- mean(master$Annualized.Sharpe)
  annualized.sharpe.sdev <- sd(master$Annualized.Sharpe)
  annualized.sharpe.sample.size <- nrow(master)
  annualized.sharpe.error <- qnorm(0.975)* annualized.sharpe.sdev/sqrt(annualized.sharpe.sample.size)
  annualized.sharpe.left <- annualized.sharpe.mean - annualized.error
  annualized.sharpe.right <- annualized.sharpe.mean + annualized.sharpe.error

  Maximum.Draw.Down.mean <- mean(master$Maximum.Draw.Down)
  Maximum.Draw.Down.sdev <- sd(master$Maximum.Draw.Down)
  Maximum.Draw.Down.sample.size <- nrow(master)
  Maximum.Draw.Down.error <- qnorm(0.975)* Maximum.Draw.Down.sdev/sqrt( Maximum.Draw.Down.sample.size)
  Maximum.Draw.Down.left <- Maximum.Draw.Down.mean - annualized.error
  Maximum.Draw.Down.right <- Maximum.Draw.Down.mean + Maximum.Draw.Down.error

  #t.test(master$Annualized.Return)
 # jarque.bera.test(master$Maximum.Draw.Down)
  require(tseries)

  # Plot
  # Annualized Return
  library(ggplot2)
  p1 <- ggplot(data=master, aes(Annualized.Return,col=I("red")))+
    geom_histogram(binwidth = 0.02)+
    theme_classic()+
    ggtitle("Resampled Strategy Results - Sma 51 to 70 - Annualized Return",subtitle="Strategy run over 100 resampled time series")+
    labs(x="Annualized Return",y="Count")+
    theme(plot.title = element_text(hjust=0.5),plot.subtitle =element_text(hjust=0.5))+
   # geom_vline(xintercept = left, color="blue", linetype="dashed")+
    #geom_vline(xintercept = right, color="blue", linetype="dashed")+
    geom_vline(xintercept = annualized.ret.mean, color="blue", linetype="dashed")+
    annotate("text", label = "Mean = 0.425727", x = 0.35, y = 117, color = "blue")

 # Sharpe Ratio
 p3 <- ggplot(data=master, aes(Annualized.Sharpe,col=I("red")))+
    geom_histogram(binwidth = 0.001)+
    theme_classic()+
    ggtitle("Resampled Strategy Results - Sma 51 to 70 - Annualized Sharpe Ratio",subtitle="Strategy run over 100 resampled time series")+
    labs(x="Annualized.Sharpe",y="Count")+
    theme(plot.title = element_text(hjust=0.5),plot.subtitle =element_text(hjust=0.5))+
    geom_vline(xintercept = annualized.sharpe.mean, color="blue", linetype="dashed")+
   geom_vline(xintercept = annualized.sharpe.right, color="blue", linetype="dashed")+
   geom_vline(xintercept = annualized.sharpe.left, color="blue", linetype="dashed")+
    annotate("text", label = "Mean = 1.02887", x = 1.02887, y = 15, color = "blue")

 annualized.sharpe.right

  # Maximum DD
  # Sharpe Ratio
  p5 <- ggplot(data=master, aes(Maximum.Draw.Down,col=I("red")))+
    geom_histogram(binwidth = 0.01)+
    theme_classic()+
    ggtitle("Resampled Strategy Results - Sma 51 to 70 - Maximum Draw Down",subtitle="Strategy run over 100 resampled time series")+
    labs(x="Maximum.Draw.Down",y="Count")+
    theme(plot.title = element_text(hjust=0.5),plot.subtitle =element_text(hjust=0.5))+
    theme(plot.title = element_text(hjust=0.5),plot.subtitle =element_text(hjust=0.5))+
    geom_vline(xintercept = Maximum.Draw.Down.mean, color="blue", linetype="dashed")+
    annotate("text", label = "Mean = 0.5020338", x = 0.6, y = 65, color = "blue")

  ###################
  # Plot Original back test results
  ##################
  # Plot
  # Annualized Return
  library(ggplot2)
  # find original mean annualized
  mean.orig.annaulized <- mean(show$Annualized.Return)
  p2 <- ggplot(data=show, aes(Annualized.Return,col=I("red")))+
    geom_histogram(binwidth = 0.01)+
    theme_classic()+
    ggtitle("Original Strategy Results - Sma 51 to 70 - Annualized Return",subtitle="")+
    labs(x="Annualized Return",y="Count")+
    theme(plot.title = element_text(hjust=0.5),plot.subtitle =element_text(hjust=0.5))+
    # geom_vline(xintercept = left, color="blue", linetype="dashed")+
    #geom_vline(xintercept = right, color="blue", linetype="dashed")+
    geom_vline(xintercept = mean.orig.annaulized, color="blue", linetype="dashed")+
    annotate("text", label = "Mean = 0.6039258", x = 0.585, y = 15, color = "blue")
  require(gridExtra)
  gridExtra::grid.arrange(p1, p2, nrow = 1)

  # Sharpe Ratio
  # find original mean annualized
  mean.orig.sharpe <- mean(show$Annualized.Sharpe)
 p4 <- ggplot(data=show, aes(Annualized.Sharpe,col=I("red")))+
    geom_histogram(binwidth = 0.02)+
    theme_classic()+
    ggtitle("Original Strategy Results - Sma 51 to 70 - Annualized Sharpe Ratio",subtitle="")+
    labs(x="Annualized.Sharpe",y="Count")+
    theme(plot.title = element_text(hjust=0.5),plot.subtitle =element_text(hjust=0.5))+
    geom_vline(xintercept = mean.orig.sharpe, color="blue", linetype="dashed")+
    annotate("text", label = "Mean = 1.565738", x = 1.52, y = 6.5, color = "blue")
 gridExtra::grid.arrange(p3, p4, nrow = 1)

  # Maximum DD
  # Sharpe Ratio
 names(show)
 show$Maximum.Draw.Down
 mean.orig.maxdd <- mean(show$Maximum.Draw.Down)
  p6 <- ggplot(data=show, aes(Maximum.Draw.Down,col=I("red")))+
    geom_histogram(binwidth = 0.0001)+
    theme_classic()+
    ggtitle("Original Strategy Results - Sma 51 to 70 - Maximum Draw Down",subtitle="")+
    labs(x="Maximum.Draw.Down",y="Count")+
    theme(plot.title = element_text(hjust=0.5),plot.subtitle =element_text(hjust=0.5))+
    theme(plot.title = element_text(hjust=0.5),plot.subtitle =element_text(hjust=0.5))+
    geom_vline(xintercept = mean.orig.maxdd, color="blue", linetype="dashed")+
    annotate("text", label = "Mean = 0.3420499", x = 0.35, y = 7, color = "blue")
  gridExtra::grid.arrange(p5, p6, nrow = 1)

ann ret

Mean annualized return over the re-sampled time series is .43% vs original strategy back test of 0.6039258%.

sharpe

Mean annualized sharpe ratio over the re-sampled time series is 1.02887 vs original strategy back test of 1.565738.

max dd

Mean maximum draw down over the re-sampled time series is 0.5020338 % vs original strategy back test of 0.3420499 %.

Table original optimized parameters:
area in 50 to 60

Over all the monte carlo distributions, on average, show a reduction in performance on annualized return, sharpe ratio and maximum draw down.

It should be noted however that the expectancy is somewhat positive. The mean, variance of the resampled series is different to the original series, however, this may still not be the best way to proceed. The meboot method retains the structure of the original series and the resampled series bear a strong resemblance to the original series. However, the results do show that a change in mean / variance, although drop in performance, the strategy parameters themselves did not fall apart. I deem that useful information moving forward as a type of sensitivity test to changes in mean / variance.

The range 51 to 70 sma, its like throwing darts, your likely to land in a zone that would make money going forward.

This research on my end is very infant and I look for other ways to stress test strategies. Some which come to mind include:

1. Run monte carlo simulations on the distribution of returns. Changing the order many times may give an expectancy of losing streak and maximum draw down.
2. Find other ways to bootstrap time series data.
3. Randomly add / subtract an amount to original price series changing the data itself, run multiple iterations.

4. Walk forward analysis, multiple train and test periods

5. In out of sample testing, hold out method

More to come

Please feel free to drop me line on other methods of strategy stress test testing and any other general collaboration.

Thanks for reading!

Full code can be found on my github.

XIV | VXX – Testing Mean Reversion, Momentum, Random Walk – Rolling Hurst Exponent

In addition to the previous posts studying volatility strategies here and here we aim to study the nature of the XIV,VXX series. We subject the series to the Hurst exponent and we follow the same procedure as Ernie Chan (2013) where we take the lagged price differences and perform a regression on the log time lags vs log variance of the lagged differences, an example of this here.

For this post we will download VIX futures data from CBOE website and join synthetic XIV and XIV data to make the largest data set we possibly can per this previous post. After we have our XIV and VXX data we then proceed to create the hurst exponent of each series and the procedure for this is as follows:

1. Compute lagged differences on varying time lags. For example. lag 2 = todays close price – close price 2 days ago. lag 3 = todays close price – close price 3 days ago.
2. Next, Compute the variance of the lagged differences. Ernie chan recommends at least 100 days of data for this. Compute variance over a period of 100 days for each lagged difference.
3. Perform linear regression of the x,y,  log(time_lags) ~ log(variance_lagged_differences) and divide the slope by 2 to obtain the hurst exponent.

We compute the procedure above on a rolling basis and the look back chosen for the variance is 100 trading days. We also use the R package RcppEigen and use fastlm to perform a rolling linear regression. The code that achieves this:

###############################
# Hurst Exponent (varying lags)
###############################

require(magrittr)
require(zoo)
require(lattice)

## Set lags
lags <- 2:(252*6)

# Function for finding differences in lags. Todays Close - 'n' lag period
getLAG.DIFF <- function(lagdays) {
  function(term.structure.df) {
    c(rep(NA, lagdays), diff(term.structure.df$vxx_close, lag = lagdays, differences = 1, arithmetic = TRUE, na.pad = TRUE))
  }
}
# Create a matrix to put the lagged differences in
lag.diff.matrix <- matrix(nrow=nrow(term.structure.df), ncol=0)

# Loop for filling it
for (i in lags) {
  lag.diff.matrix <- cbind(lag.diff.matrix, getLAG.DIFF(i)(term.structure.df))
}

# Rename columns
colnames(lag.diff.matrix) <- sapply(lags, function(n)paste("lagged.diff.n", n, sep=""))

# Bind to existing dataframe
term.structure.df <-  cbind(term.structure.df, lag.diff.matrix)
head(term.structure.df,25)

############################################################
# Calculate rolling variances of 'n period' differences
# Set variance look back to 100 days
############################################################
# Convert NA to 0
term.structure.df[is.na(term.structure.df)] <- as.Date(0)

get.VAR <- function(varlag) {
  function(term.structure.df) {
    runVar(term.structure.df[,paste0("lagged.diff.n", lags[i])], y = NULL, n = 100, sample = TRUE, cumulative = FALSE)
  }
}
# Create a matrix to put the variances in
lag.var.matrix <- matrix(nrow=nrow(term.structure.df), ncol=0)

# Loop for filling it
for (i in 1:length(lags)) {
  lag.var.matrix <- cbind(lag.var.matrix, get.VAR(i)(term.structure.df))
}

# Rename columns
colnames(lag.var.matrix) <- sapply(lags, function(n)paste("roll.var.diff.", n, sep=""))

# Bind to existing dataframe
term.structure.df <-  cbind(term.structure.df, lag.var.matrix)

########################################
# Subset to remove all leading NA
########################################
#NonNAindex <- which(!is.na(term.structure.df))
#set_lag_threshold <- 50 # Set Variance
#na <- which(!is.na(term.structure.df[,paste0("roll.var.diff.", set_lag_threshold)]))
#firstNonNA <- min(na)
#term.structure.df<-term.structure.df[firstNonNA:nrow(term.structure.df),]

########################################
# Rolling linear regression to compute hurst exponent
########################################
variance <- list()
lag.vec <- c(2:30)  # Select short term lags
# Select column selection
for (i in 1:nrow(term.structure.df)) {
  variance[i] <- list(c(term.structure.df$roll.var.diff.2[i],
                        term.structure.df$roll.var.diff.3[i],
                        term.structure.df$roll.var.diff.4[i],
                        term.structure.df$roll.var.diff.5[i],
                        term.structure.df$roll.var.diff.6[i],
                        term.structure.df$roll.var.diff.7[i],
                        term.structure.df$roll.var.diff.8[i],
                        term.structure.df$roll.var.diff.9[i],
                        term.structure.df$roll.var.diff.10[i],
                        term.structure.df$roll.var.diff.11[i],
                        term.structure.df$roll.var.diff.12[i],
                        term.structure.df$roll.var.diff.13[i],
                        term.structure.df$roll.var.diff.14[i],
                        term.structure.df$roll.var.diff.15[i],
                        term.structure.df$roll.var.diff.16[i],
                        term.structure.df$roll.var.diff.17[i],
                        term.structure.df$roll.var.diff.18[i],
                        term.structure.df$roll.var.diff.19[i],
                        term.structure.df$roll.var.diff.20[i],
                        term.structure.df$roll.var.diff.21[i],
                        term.structure.df$roll.var.diff.22[i],
                        term.structure.df$roll.var.diff.23[i],
                        term.structure.df$roll.var.diff.24[i],
                        term.structure.df$roll.var.diff.25[i],
                        term.structure.df$roll.var.diff.26[i],
                        term.structure.df$roll.var.diff.27[i],
                        term.structure.df$roll.var.diff.28[i],
                        term.structure.df$roll.var.diff.29[i],
                        term.structure.df$roll.var.diff.30[i]))
}

#Initialize list, pre allocate memory
results<-vector("list", length(variance))
hurst<-vector("list", length(variance))
library(RcppEigen)
i=1
for(i in 1:length(variance)){
  results[[i]]<-fastLm( log(lag.vec) ~ log(variance[[i]]), data=variance)
  hurst[[i]]<- coef(results[[i]])[2]/2
  ptm0 <- proc.time()
  Sys.sleep(0.1)
  ptm1=proc.time() - ptm0
  time=as.numeric(ptm1[3])
  cat('\n','Iteration',i,'took', time, "seconds to complete")
}

# Join results to data frame
hurst <- do.call(rbind, hurst)
hurst.df <- as.data.frame(hurst)
hurst.df <- data.frame(hurst.df,Date=term.structure.df$Date)
colnames(hurst.df)[1] <- "Hurst"
hurst.df <- subset(hurst.df, Date >= as.POSIXct("2008-04-28") ) # subset data remove leading NA VXX only
# Plot Data
ggplot() +
  geom_line(data=hurst.df ,aes(x=Date,y=Hurst), colour="black") +
  theme_classic()+
  scale_y_continuous(breaks = round(seq(min(hurst.df$Hurst), max(hurst.df$Hurst), by = 0.2),2))+
  scale_x_date(breaks = date_breaks("years"), labels = date_format("%Y"))+
  ggtitle("VXX Hurst Exponent - Daily Bars - Lags 2:30", subtitle = "Regression log(variances) ~ log(time_lags) - Hurst = Coef/2") +
  labs(x="Date",y="Hurst")+
  theme(plot.title = element_text(hjust=0.5),plot.subtitle =element_text(hjust=0.5))+
  #geom_hline(yintercept = 0.5, color = "red", size=0.5,linetype="dashed")+
  geom_rect(aes(xmin=as.Date(head(hurst.df$Date,1)),xmax=as.Date(Inf),ymin=0.5,ymax=Inf),alpha=0.1,fill="green")+
  geom_rect(aes(xmin=as.Date(head(hurst.df$Date,1)),xmax=as.Date(Inf),ymin=-Inf,ymax=0.5),alpha=0.1,fill="orange")+
  geom_rect(aes(xmin=as.Date(head(hurst.df$Date,1)),xmax=as.Date(Inf),ymin=0.48,ymax=0.52),alpha=.7,fill="red")

The output:

Rplot140
Random Walk Band = Red Between .48 and .52. Momentum = Green > 0.52. Mean Reversion = Red < .48.
Rplot142
Random Walk Band = Red Between .48 and .52. Momentum = Green > 0.52. Mean Reversion = Red < .48.

This is for lagged differences of 2 to 30 days. A rolling variance of 100 days. We chose the smaller lagged differences as the strategies going forward will likely hold no longer than 30 days so it makes sense to see the nature of the series on this time period.

We can compute how often XIV and VXX is in a momentum, mean reversion and random walk regime. The code for this:

# Count how often in each regime
momo <- sum(hurst.df$Hurst > 0.52, na.rm=TRUE)
mean.rev <- sum(hurst.df$Hurst < 0.48 , na.rm=TRUE)
random <- sum(hurst.df$Hurst >= 0.48 & hurst.df$Hurst <=.52, na.rm=TRUE)
exact.random <- sum(hurst.df$Hurst >= 0.50 & hurst.df$Hurst <.51, na.rm=TRUE)
total.rows <- NROW(hurst.df)

# Percentage of time in momentum, mean reversion, random walk
momo.perc <- momo / total.rows
mean.rev.perc <- mean.rev / total.rows
random.perc <- random / total.rows
exact.random.perc <- exact.random / total.rows

VXX:

vxx.percs.df <- data.frame ("Momentum, Over 0.50" = momo.perc,"Mean Reversion, Less than 0.5" = mean.rev.perc, "Random Walk Band, 0.48 to 0.52" = random.perc, "Exact Random Walk, 0.50" = exact.random.perc) vxx.percs.df > vxx.percs.df
  Momentum..Over.0.50 Mean.Reversion..Less.than.0.5 Random.Walk.Band..0.48.to.0.52 Exact.Random.Walk..0.50
1           0.7471264                     0.1395731                      0.1133005              0.02791461<span 				data-mce-type="bookmark" 				id="mce_SELREST_start" 				data-mce-style="overflow:hidden;line-height:0" 				style="overflow:hidden;line-height:0" 			></span>

XIV:

 xiv.percs.df <- data.frame ("Momentum, Over 0.50" = momo.perc,"Mean Reversion, Less than 0.5" = mean.rev.perc, "Random Walk Band, 0.48 to 0.52" = random.perc, "Exact Random Walk, 0.50" = exact.random.perc) xiv.percs.df  > xiv.percs.df
  Momentum..Over.0.50 Mean.Reversion..Less.than.0.5 Random.Walk.Band..0.48.to.0.52 Exact.Random.Walk..0.50
1           0.7081281                     0.2085386                     0.08333333                0.022578

What we see is VXX is in momentum phase 74% of the time and XIV in momentum phase 70% of the time. That is the dominating theme where mean reversion 13% (VXX) and 20% (XIV) and times of random walk 11%(VXX) and 8% (XIV).

If fitting a model to the series itself without using the volatility risk premium / roll yield as entry signals. One may try fitting models based on the theme of momentum.

In subsequent posts we will be applying models based on extracting profits when the market is in contango/backwardation as well as applying models to the series itself. We will also look at the auto correlatation of the series and this will serve as a primer for testing a strategy for robustness.

Thanks for reading!