Manual

User Manual:

Open the PDF directly: View PDF PDF.
Page Count: 21

Package ‘fmlr’
March 5, 2019
Title Financial Machine Learning using R
Version 0.0.8.0
Author Larry Lei Hua [aut, cre],
Maintainer Larry Lei Hua <larry.lei.hua@gmail.com>
Description This is an R package for tools of machine learning for financial data.
Depends R (>= 3.5)
License file LICENSE
Encoding UTF-8
LazyData true
RoxygenNote 6.1.1
LinkingTo Rcpp
Imports stats,
lubridate,
pracma,
zoo,
Rcpp,
stringr,
readr
Rtopics documented:
acc_lucky .......................................... 2
bar_tick ........................................... 3
bar_tick_imbalance..................................... 3
bar_tick_runs ........................................ 4
bar_time........................................... 5
bar_unit ........................................... 6
bar_unit_runs ........................................ 6
bar_volume ......................................... 7
bar_volume_imbalance................................... 8
bar_volume_runs ...................................... 9
ema ............................................. 10
1
2acc_lucky
fracDiff ........................................... 10
imbalance_tick ....................................... 11
imbalance_volume ..................................... 11
istar_CUSUM........................................ 12
istar_CUSUM_R ...................................... 13
label_meta.......................................... 13
purged_k_CV........................................ 14
read_algoseek_futures_fullDepth . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
Tstar_tib........................................... 16
Tstar_trb_cpp........................................ 17
Tstar_vib .......................................... 18
Tstar_vrb_cpp........................................ 19
weights_fracDiff ...................................... 20
Index 21
acc_lucky A function to check whether a classification is better than a guess
Description
A function to check whether a classification is better than a guess
Usage
acc_lucky(train_class, test_class, my_acc, s = 1000)
Arguments
train_class a vector for the distribution of classes in the training set
test_class a vector for the distribution of classes in the test set
my_acc a number between 0 and 1 for the classification accuracy to be evaluated
ssample size of simulations used to check p-values
Author(s)
Larry Lei Hua
Examples
train_class <- c(1223,1322,1144)
test_class <- c(345,544,233)
my_acc <- 0.45
acc_lucky(train_class, test_class, my_acc)
bar_tick 3
bar_tick Construct tick bars
Description
Construct tick bars
Usage
bar_tick(dat, nTic)
Arguments
dat dat input with at least the following columns: Price, Size
nTic the number of ticks of each window
Value
time stamp at the end of each bar (if timestamp is provided), and H,L,O,C,V
Author(s)
Larry Lei Hua
bar_tick_imbalance Construct tick imbalance bars
Description
Construct tick imbalance bars
Usage
bar_tick_imbalance(dat, w0 = 10, bkw_T = 5, bkw_b = 5)
Arguments
dat dat input with at least the following column: Price, Size
w0 the time window length of the first bar
bkw_T backward window length when using pracma::movavg for exponentially weighted
average T
bkw_b backward window length when using pracma::movavg for exponentially weighted
average b_t
4bar_tick_runs
Value
a list of vectors for tStamp (if returned), and HLOCV of tick imbalance bars. Note that the remain-
ing data after the latest imbalance time point will be formed as a bar.
Author(s)
Larry Lei Hua
Examples
set.seed(1)
dat <- data.frame(Price = c(rep(0.5, 4), runif(50)), Size = rep(10,54))
bar_tick_imbalance(dat)
bar_tick_runs Construct tick runs bars
Description
Construct tick runs bars
Usage
bar_tick_runs(dat, w0 = 10, de = 1, bkw_T = 5, bkw_Pb1 = 5,
filter = FALSE)
Arguments
dat dat input with at least the following column: Price, Size
w0 the time window length of the first bar
de a positive value for adjusting the expected window size, that is, de*E0T; default:
1
bkw_T backward window length when using pracma::movavg for exponentially weighted
average T
bkw_Pb1 backward window length when using pracma::movavg for exponentially weighted
average P[b_t=1]
filter whether used as a filter; default FALSE. If TRUE, then only i_feabar, the ending
time index of feature bars, is returned
Value
If filter==FALSE, a list of vectors for tStamp (if returned), and HLOCV of tick runs bars. Note
that the remaining data after the latest ending time point detected will be formed as a bar. If fil-
ter==TRUE, i_feabar a vector of integers for the time index.
bar_time 5
Author(s)
Larry Lei Hua
Examples
set.seed(1)
dat <- data.frame(Price = c(rep(0.5, 4), runif(1000)), Size = rep(10,1004))
x1 <- bar_tick_runs(dat)
x2 <- bar_tick_runs(dat, filter=TRUE)
bar_time Construct time bars
Description
Construct time bars
Usage
bar_time(dat, tDur = 1)
Arguments
dat dat input with at least the following columns: tStamp, Price, Size, where tStamp
should be sorted already
tDur the time duration in seconds of each window
Value
tStamp, seconds since the starting time point, and H,L,O,C,V
Author(s)
Larry Lei Hua
6bar_unit_runs
bar_unit Construct unit bars
Description
Construct unit bars
Usage
bar_unit(dat, unit)
Arguments
dat dat input with at least the following columns: Price, Size. If timestamp is pro-
vided than output also contains timestamp of the unit bars
unit the total dollar (unit) traded of each window
Value
time stamp at the end of each bar (if timestamp is provided), and H,L,O,C,V
Author(s)
Larry Lei Hua
bar_unit_runs Construct unit runs bars
Description
Construct unit runs bars
Usage
bar_unit_runs(dat, u_0 = 2000, w0 = 10, de = 1, bkw_T = 5,
bkw_Pb1 = 5, bkw_U = 5, filter = FALSE)
Arguments
dat dat input with at least the following column: Price, Size
u_0 average unit (volume*price) for each trade, and it is used to create the first bar
w0 the time window length of the first bar
de a positive value for adjusting the expected window size, that is, de*E0T; default:
1
bar_volume 7
bkw_T backward window length when using pracma::movavg for exponentially weighted
average T
bkw_Pb1 backward window length when using pracma::movavg for exponentially weighted
average P[b_t=1]
bkw_U backward window length for exponentially weighted average volumes
filter whether used as a filter; default FALSE. If TRUE, then only i_feabar, the ending
time index of feature bars, is returned
Value
If filter==FALSE, a list of vectors for tStamp (if returned), and HLOCV of volume runs bars.
Note that the remaining data after the latest ending time point detected will be formed as a bar. If
filter==TRUE, i_feabar a vector of integers for the time index.
Author(s)
Larry Lei Hua
Examples
set.seed(1)
dat <- data.frame(Price = c(rep(0.5, 4), runif(50)), Size = floor(runif(54)*100))
bar_unit_runs(dat, u_0=mean(dat$Price)*mean(dat$Size))
bar_unit_runs(dat, u_0=mean(dat$Price)*mean(dat$Size), filter=TRUE)
bar_volume Construct volume bars
Description
Construct volume bars
Usage
bar_volume(dat, vol)
Arguments
dat dat input with at least the following columns: Price, Size
vol the volume of each window
Value
time stamp at the end of each bar (if timestamp is provided), and H,L,O,C,V
8bar_volume_imbalance
Author(s)
Larry Lei Hua
bar_volume_imbalance Construct volume imbalance bars
Description
Construct volume imbalance bars
Usage
bar_volume_imbalance(dat, w0 = 10, bkw_T = 5, bkw_b = 5)
Arguments
dat dat input with at least the following column: Price, Size
w0 the time window length of the first bar
bkw_T backward window length when using pracma::movavg for exponentially weighted
average T
bkw_b backward window length when using pracma::movavg for exponentially weighted
average b_tv_t
Value
a list of vectors for tStamp (if returned), and HLOCV of volume imbalance bars. Note that the
remaining data after the latest imbalance time point will be formed as a bar.
Author(s)
Larry Lei Hua
Examples
set.seed(1)
dat <- data.frame(Price = c(rep(0.5, 4), runif(50)), Size = rep(10,54))
bar_volume_imbalance(dat)
bar_volume_runs 9
bar_volume_runs Construct volume runs bars
Description
Construct volume runs bars
Usage
bar_volume_runs(dat, v_0 = 20, w0 = 10, de = 1, bkw_T = 5,
bkw_Pb1 = 5, bkw_V = 5, filter = FALSE)
Arguments
dat dat input with at least the following column: Price, Size
v_0 average volume for each trade, and it is used to create the first bar
w0 the time window length of the first bar
de a positive value for adjusting the expected window size, that is, de*E0T; default:
1
bkw_T backward window length when using pracma::movavg for exponentially weighted
average T
bkw_Pb1 backward window length when using pracma::movavg for exponentially weighted
average P[b_t=1]
bkw_V backward window length for exponentially weighted average volumes
filter whether used as a filter; default FALSE. If TRUE, then only i_feabar, the ending
time index of feature bars, is returned
Value
If filter==FALSE, a list of vectors for tStamp (if returned), and HLOCV of volume runs bars.
Note that the remaining data after the latest ending time point detected will be formed as a bar. If
filter==TRUE, i_feabar a vector of integers for the time index.
Author(s)
Larry Lei Hua
Examples
set.seed(1)
dat <- data.frame(Price = c(rep(0.5, 4), runif(50)), Size = floor(runif(54)*100))
bar_volume_runs(dat)
bar_volume_runs(dat, filter=TRUE)
10 fracDiff
ema exponentially weighted moving average; only return the last value
Description
exponentially weighted moving average; only return the last value
Usage
ema(x, n)
Arguments
xa numeric vector
nwindow size
Value
a numeric value
Author(s)
Larry Lei Hua
fracDiff convert a time series into a fractionally differentiated series
Description
convert a time series into a fractionally differentiated series
Usage
fracDiff(x, d = 0.5, nWei = 10, tau = NULL)
Arguments
xa vector of time series to be fractionally differentiated
dthe order for fractionally differentiated features
nWei number of weights for output
tau threshold where weights are cut off; default is NULL, if not NULL then use tau
and nWei is not used
Author(s)
Larry Lei Hua
imbalance_tick 11
imbalance_tick The auxiliary function b_t for constructing tick imbalance bars. The
first b_t is assigned the value 0 because no information is available
Description
The auxiliary function b_t for constructing tick imbalance bars. The first b_t is assigned the value
0 because no information is available
Usage
imbalance_tick(dat)
Arguments
dat dat input with at least the following columns: Price
Author(s)
Larry Lei Hua, ming08108(github)
Examples
set.seed(1)
dat <- data.frame(Price = c(rep(0.5, 4), runif(2), 0.5, 0.5, 0.4, runif(2) ))
b_t <- imbalance_tick(dat)
imbalance_volume The auxiliary function b_tv_t for constructing volume imbalance bars.
The first b_tv_t is assigned the value 0 because no information is avail-
able
Description
The auxiliary function b_tv_t for constructing volume imbalance bars. The first b_tv_t is assigned
the value 0 because no information is available
Usage
imbalance_volume(dat)
Arguments
dat dat input with at least the following columns: Price, Size
12 istar_CUSUM
Author(s)
Larry Lei Hua
Examples
set.seed(1)
dat <- data.frame(Price = c(rep(0.5, 4), runif(10)), Size = rep(10,14))
b_tv_t <- imbalance_volume(dat)
istar_CUSUM time index that triggers a symmetric CUSUM filter
Description
time index that triggers a symmetric CUSUM filter
Usage
istar_CUSUM(x, h)
Arguments
xa vector of time series to be filtered
ha vector of the thresholds
Author(s)
Larry Lei Hua
Examples
set.seed(1)
x <- runif(100, 1, 3)
h <- rep(1.5, 100)
i_CUSUM <- istar_CUSUM(x,h)
abline(v=i_CUSUM, lty = 2)
## Comparing C and R versions
# set.seed(1)
# x <- runif(1000000, 1, 3)
# h <- rep(1.5, 100)
# start_time <- Sys.time()
# i_CUSUM <- istar_CUSUM(x,h)
# end_time <- Sys.time()
# C_time <- end_time - start_time
istar_CUSUM_R 13
# start_time <- Sys.time()
# i_CUSUM_R <- istar_CUSUM_R(x,h)
# end_time <- Sys.time()
# R_time <- end_time - start_time
# cat("C and R time: ", C_time, R_time)
# all(i_CUSUM-i_CUSUM_R==0)
istar_CUSUM_R time index that triggers a symmetric CUSUM filter (R version for is-
tar_CUSUM(), for shorter x, the R version can be faster than the C
version)
Description
time index that triggers a symmetric CUSUM filter (R version for istar_CUSUM(), for shorter x,
the R version can be faster than the C version)
Usage
istar_CUSUM_R(x, h)
Arguments
xa vector of time series to be filtered
ha vector of the thresholds
Author(s)
Larry Lei Hua
label_meta Meta labeling, including three options: triple barriers, upper and ver-
tical barriers, and lower and vertical barriers.
Description
Meta labeling, including three options: triple barriers, upper and vertical barriers, and lower and
vertical barriers.
Usage
label_meta(x, events, ptSl, ex_vert = T, n_ex = 0)
14 purged_k_CV
Arguments
xa vector of prices series to be labeled.
events a dataframe that has the following columns:
• t0: event’s starting time index.
• t1: event’s ending time index; if t1==Inf then no vertical barrier, i.e., last
observation in x is the vertical barrier.
• trgt: the unit absolute return used to set up the upper and lower barriers.
• side: 0: both upper and lower barriers; 1: only upper; -1: only lower.
ptSl a vector of two multipliers for the upper and lower barriers, respectively.
ex_vert whether exclude the output when the vertical barrier is hit; default is T.
n_ex number of excluded observations at the begining of x; default is 0.
Value
data frame with the following columns:
• T_up: local time index when the upper barrier is hit; Inf means that upper is not hit.
• T_lo: local time index when the lower barrier is hit; Inf means that lower is not hit.
• t1: local time index when the vertical barrier is hit.
• ret: return associated with the event.
• label: low:-1, vertical:0, upper:1.
• t0Fea: begining time index of feature bars.
• t1Fea: ending time index of feature bars.
• tLabel: ending time index of events, i.e., when the labels are created. Both t1Fea and tLabel
will be useful for sequential bootstrap.
Author(s)
Larry Lei Hua
purged_k_CV Purged k-fold CV with embargo
Description
Purged k-fold CV with embargo
Usage
purged_k_CV(feaMat, k = 5, gam = 0.01)
read_algoseek_futures_fullDepth 15
Arguments
feaMat a data.frame for feature matrix with the first column being the label
knumber of folds for k-fold CV
gam gamma for embargo
Value
a list of k data.frame, each containing a test set and a training set
Author(s)
Larry Lei Hua
Examples
feaMat <- data.frame(Y = c(1,1,0,1,0),
V = c(2,4,2,4,1),
t1Fea = c(2,5,8,14,20),
tLabel = c(4,12,16,23,38))
purged_k_CV(feaMat, k=2, gam=0.1)
read_algoseek_futures_fullDepth
Load AlgoSeek Futures Full Depth data from zip files
Description
Load AlgoSeek Futures Full Depth data from zip files
Usage
read_algoseek_futures_fullDepth(zipdata, whichData = NULL)
Arguments
zipdata the original zip data provided by AlgoSeek
whichData the specific data to be loaded; by default load all data in the zip file
Author(s)
Larry Lei Hua
16 Tstar_tib
Examples
zipdata <- tempfile()
download.file("https://www.algoseek.com/static/files/sample_data/
futures_and_future_options/ESH5.Futures.FullDepth.20150128.csv.zip",zipdata)
dat <- read_algoseek_futures_fullDepth(zipdata)
# Do not run unless the file 20160104.zip is avaliable
# dat <- read_algoseek_futures_fullDepth("20160104.zip", whichData="ES/ESH6.csv")
Tstar_tib Tstar index for Tick Imbalance Bars (bar_tib)
Description
Tstar index for Tick Imbalance Bars (bar_tib)
Usage
Tstar_tib(dat, w0 = 10, bkw_T = 5, bkw_b = 5)
Arguments
dat dat input with at least the following columns: Price
w0 the time window length of the first bar
bkw_T backward window length when using pracma::movavg for exponentially weighted
average T
bkw_b backward window length when using pracma::movavg for exponentially weighted
average b_t
Value
a vector for the lengths of the tick imbalance bars. For example, if the return is c(10,26), then the 2
tick imbalance bars are (0,10] and (10, 36]
Author(s)
Larry Lei Hua
Tstar_trb_cpp 17
Examples
set.seed(1)
dat <- data.frame(Price = c(rep(0.5, 4), runif(50)))
T_tib <- Tstar_tib(dat)
b_t <- imbalance_tick(dat)
cumsum(b_t)[cumsum(T_tib)] # check the accumulated b_t's where the imbalances occur
Tstar_trb_cpp Tstar index for Tick Runs Bars (bar_trb)
Description
Tstar index for Tick Runs Bars (bar_trb)
Usage
Tstar_trb_cpp(b_t, w0, de, bkw_T, bkw_Pb1)
Arguments
b_t output of imbalance_tick(dat) with the dat has at least the following columns:
Price
w0 the time window length of the first bar
de a positive value for adjusting the expected window size, that is, de*E0
bkw_T backward window length for exponentially weighted average T
bkw_Pb1 backward window length for exponentially weighted average P[b_t=1]
Value
a list of the following two vectors: a vector for the lengths of the tick imbalance bars. For example,
if the return is c(10,26), then the 2 tick imbalance bars are (0,10] and (10, 36] a vector indicating
up runs or down runs
Examples
set.seed(1)
dat <- data.frame(Price = c(rep(0.5, 5), runif(100)))
b_t <- imbalance_tick(dat)
T_trb <- Tstar_trb_cpp(b_t, 10, 1.0, 10, 10)
col <- ifelse(T_trb$Type==1, "red", "blue")
T <- cumsum(T_trb$Tstar)
plot(dat$Price)
for(i in 1:length(T)) abline(v = T[i], col = col[i])
18 Tstar_vib
Tstar_vib Tstar index for Volume Imbalance Bars (bar_vib)
Description
Tstar index for Volume Imbalance Bars (bar_vib)
Usage
Tstar_vib(dat, w0 = 10, bkw_T = 5, bkw_b = 5)
Arguments
dat dat input with at least the following columns: Price, Size
w0 the time window length of the first bar
bkw_T backward window length when using pracma::movavg for exponentially weighted
average T
bkw_b backward window length when using pracma::movavg for exponentially weighted
average b_tv_t
Value
a vector for the lengths of the tick imbalance bars. For example, if the return is c(10,26), then the 2
tick imbalance bars are (0,10] and (10, 36]
Author(s)
Larry Lei Hua
Examples
set.seed(1)
dat <- data.frame(Price = c(rep(0.5, 4), runif(50)), Size = rep(10, 54))
T_vib <- Tstar_vib(dat)
b_tv_t <- imbalance_volume(dat)
cumsum(b_tv_t)[cumsum(T_vib)] # check the accumulated b_t's where the imbalances occur
Tstar_vrb_cpp 19
Tstar_vrb_cpp Tstar index for Volume Runs Bars (bar_vrb)
Description
Tstar index for Volume Runs Bars (bar_vrb)
Usage
Tstar_vrb_cpp(b_t, v_t, v_0, w0, de, bkw_T, bkw_Pb1, bkw_V)
Arguments
b_t output of imbalance_tick(dat) with the data ’dat’ has at least the following columns:
Price
v_t volume of the same data
v_0 average volume for each trade, and it is used to create the first bar
w0 the time window length of the first bar
de a positive value for adjusting the expected window size, that is, de*E0T; default:
1.
bkw_T backward window length for exponentially weighted average T
bkw_Pb1 backward window length for exponentially weighted average P[b_t=1]
bkw_V backward window length for exponentially weighted average volumes
Value
a list of the following two vectors: a vector for the lengths of the tick imbalance bars. For example,
if the return is c(10,26), then the 2 tick imbalance bars are (0,10] and (10, 36] a vector indicating
up runs or down runs
Examples
set.seed(1)
dat <- data.frame(Price = c(rep(0.5, 5), runif(100)), Size = runif(105, 10, 100))
b_t <- imbalance_tick(dat)
v_t <- dat$Size
T_vrb <- Tstar_vrb_cpp(b_t, v_t, 55, 10, 1.0, 10, 10, 10)
col <- ifelse(T_vrb$Type==1, "red", "blue")
T <- cumsum(T_vrb$Tstar)
plot(dat$Price)
for(i in 1:length(T)) abline(v = T[i], col = col[i])
20 weights_fracDiff
weights_fracDiff calculate the weights for deriving fractionally differentiated series
Description
calculate the weights for deriving fractionally differentiated series
Usage
weights_fracDiff(d = 0.5, nWei = 10, tau = NULL)
Arguments
dthe order for fractionally differentiated features
nWei number of weights for output
tau threshold where weights are cut off; default is NULL, if not NULL then use tau
and nWei is not used
Author(s)
Larry Lei Hua
Examples
weights_fracDiff(0.5,tau=1e-3)
Index
acc_lucky,2
bar_tick,3
bar_tick_imbalance,3
bar_tick_runs,4
bar_time,5
bar_unit,6
bar_unit_runs,6
bar_volume,7
bar_volume_imbalance,8
bar_volume_runs,9
ema,10
fracDiff,10
imbalance_tick,11
imbalance_volume,11
istar_CUSUM,12
istar_CUSUM_R,13
label_meta,13
purged_k_CV,14
read_algoseek_futures_fullDepth,15
Tstar_tib,16
Tstar_trb_cpp,17
Tstar_vib,18
Tstar_vrb_cpp,19
weights_fracDiff,20
21

Navigation menu