Deriv.com - Interday & Intraday High Frequency Trading Models Comparison Review (Part II)
®γσ, Lian Hu 
® 
2021-04-23
Theme Song
FRA : Elle me dit
CHN : 她对我说
ENG : She told me
JPN : 彼女は私に言う
GRE : Μου λέει
SPN : Ella me dijo
GER : Sie sagte mir
ITA : Lei mi dice
KRO : 그녀는 나에게 말한다
FRA : écris une chanson contente
CHN : 写一首欢快的歌
ENG : Write a happy song
JPN : 幸せな歌を書く
GRE : Γράψτε ένα χαρούμενο τραγούδι
SPN : Escribe una canción alegre
GER : Schreibe ein fröhliches Lied
ITA : Scrivi una canzone allegra
KRO : 쾌활한 노래 쓰기
FRA : Pas une chanson déprimante
CHN : 而不是悲伤的歌
ENG : Not a depressing song
JPN : 気のめいるような歌ではない
GRE : Όχι θλιβερό τραγούδι
SPN : No es una canción triste
GER : Kein trauriges Lied
ITA : Canzone non triste
KRO : 슬픈 노래가 아님
FRA : Une chanson que tout le monde aime
CHN : 一首让所有人都喜欢的歌
ENG : A song that everyone loves
JPN : みんなが大好きな曲
GRE : Ένα τραγούδι που αρέσει σε όλους
SPN : Una cancion que a todos les gusta
GER : Ein Lied, das jeder mag
ITA : Una canzone che piace a tutti
KRO : 모두가 좋아하는 노래
FRA : Elle me dit
CHN : 她对我说
ENG : She told me
JPN : 彼女は私に言う
GRE : Μου λέει
SPN : Ella me dijo
GER : Sie sagte mir
ITA : Lei mi dice
KRO : 그녀는 나에게 말한다
FRA : Tu deviendras milliardaire
CHN : 你将成为亿万富翁
ENG : You will become a millionaire
JPN : あなたは億万長者になります
GRE : Θα γίνετε δισεκατομμυριούχος
SPN : Te convertirás en multimillonario
GER : Sie werden Milliardär
ITA : Diventerai un miliardario
KRO : 당신은 억만 장자가 될 것입니다
FRA : Tu auras de quoi être fier
CHN : 你将为此感到骄傲
ENG : You will be proud
JPN : あなたは誇りに思うでしょう
GRE : Θα είσαι περήφανος για αυτό
SPN : Estarás orgulloso de ello
GER : Sie werden stolz darauf sein
ITA : Ne sarai orgoglioso/span>
KRO : 당신은 그것을 자랑스러워 할 것입니다
1 Introduction
Due to below issues from Deriv.com - Interday High Frequency Trading Models Comparison Blooper, here I review the research by using same dataset as we can know from below Part I.
- Deriv.com - Interday High Frequency Trading Models Comparison Review (Part I) (in RPubs.com)
- Deriv.com - Interday High Frequency Trading Models Comparison Review (Part I) (in RStudioConnect.com)
Here I use the βest model (ETS MNZ & MNZ) from above study but adjust the data size as we can know from next section.
The purpose of adjust the data size to be intraday :
- compare if improve the accuracy.
- easier the workload for other models comparison (use minimum data size to get optimal model).
Load Packages
if(!suppressPackageStartupMessages(require('BBmisc'))) {
install.packages('BBmisc', dependencies = TRUE, INSTALL_opts = '--no-lock')
}
suppressPackageStartupMessages(require('BBmisc'))
# suppressPackageStartupMessages(require('rmsfuns'))
pkgs <- c('devtools', 'knitr', 'kableExtra', 'tint', 'furrr', 'tidyr',
'devtools','readr', 'lubridate', 'data.table', 'reprex',
'feather', 'purrr', 'quantmod', 'tidyquant', 'tibbletime',
'timetk', 'plyr', 'dplyr', 'stringr', 'magrittr', 'tdplyr',
'tidyverse', 'memoise', 'htmltools', 'formattable', 'dtplyr',
'zoo', 'forecast', 'seasonal', 'seasonalview', 'rjson',
'rugarch', 'rmgarch', 'mfGARCH', 'sparklyr', 'jcolors',
'microbenchmark', 'dendextend', 'lhmetools', 'ggthemr',
'stringr', 'pacman', 'profmem', 'ggthemes', 'flyingfox',
'htmltools', 'echarts4r', 'viridis', 'hrbrthemes', 'gtools',
'fable', 'fabletools', 'Rfast', 'Metrics', 'MLmetrics')
suppressAll(lib(pkgs))
# load_pkg(pkgs)
.dtr <- 'C:/Users/User/Documents/GitHub/binary.com-interview-question-data/'
## Set the timezone but not change the datetime
Sys.setenv(TZ = 'Asia/Tokyo')
## options(knitr.table.format = 'html') will set all kableExtra tables to be 'html', otherwise need to set the parameter on every single table.
options(warn = -1, knitr.table.format = 'html')#, digits.secs = 6)
## https://stackoverflow.com/questions/39417003/long-vectors-not-supported-yet-abnor-in-rmd-but-not-in-r-script
knitr::opts_chunk$set(warning = FALSE, #cache = TRUE,
message = FALSE, cache.lazy = FALSE)
rm(pkgs)2 Data
2.1 Read Data
## check if data path set
if(!exists('dtr')) {
dtr <- 'C:/Users/User/Documents/GitHub/binary.com-interview-question-data/'}
## save files if not exists
if(!file.exists(paste0(dtr, 'data/fx/USDJPY/dsmp.rds')) & exists('dsmp')) {
saveRDS(dsmp, paste0(dtr, 'data/fx/USDJPY/dsmp.rds'))}
## read files if not exists
if(!exists('dsmp')) {
dsmp <- readRDS(paste0(dtr, 'data/fx/USDJPY/dsmp.rds'))}## plot sample data
dsmp[c(1:3, (nrow(dsmp)-3):nrow(dsmp)),] %>%
kbl(caption = '1 min Close Price Dataset', escape = FALSE) %>%
## https://www.w3schools.com/cssref/css_colors.asp
## https://public.tableau.com/en-us/gallery/100-color-palettes?gallery=votd
row_spec(0, background = 'DimGrey', color = 'gold', bold = TRUE) %>%
column_spec(1, background = 'CornflowerBlue') %>%
column_spec(2, background = 'Gray') %>%
column_spec(3, background = 'DarkGrey') %>%
column_spec(4, background = 'Gray') %>%
column_spec(5, background = 'DarkGrey') %>%
column_spec(6, background = '#4897D8') %>%
column_spec(7, background = '#556DAC') %>%
column_spec(8, background = '#92AAC7') %>%
column_spec(9, background = '#556DAC') %>%
column_spec(10, background = '#375E97') %>%
column_spec(11, background = 'CornflowerBlue') %>%
column_spec(12, background = 'LightGray', color = 'goldenrod') %>%
kable_styling(bootstrap_options = c('striped', 'hover', 'condensed', 'responsive')) %>%
kable_material(full_width = FALSE) %>% ##`full_width = FALSE` will auto adjust every single columns width to fit the table full width.
scroll_box(width = '100%', fixed_thead = TRUE, height = '400px')| index | year | quarter | month | week | wkdays | wk_1m | dy_1m | hr_1m | sq | date | close |
|---|---|---|---|---|---|---|---|---|---|---|---|
| 2015-01-05 00:01:00 | 2015 | 1 | 1 | 1 | Monday | 1 | 1 | 1 | 1 | 2015-01-05 | 120.5740 |
| 2015-01-05 00:02:00 | 2015 | 1 | 1 | 1 | Monday | 2 | 2 | 2 | 2 | 2015-01-05 | 120.5900 |
| 2015-01-05 00:03:00 | 2015 | 1 | 1 | 1 | Monday | 3 | 3 | 3 | 3 | 2015-01-05 | 120.6035 |
| 2018-07-06 23:57:00 | 2018 | 3 | 7 | 27 | Friday | 7197 | 1437 | 57 | 1324797 | 2018-07-06 | 110.4635 |
| 2018-07-06 23:58:00 | 2018 | 3 | 7 | 27 | Friday | 7198 | 1438 | 58 | 1324798 | 2018-07-06 | 110.4740 |
| 2018-07-06 23:59:00 | 2018 | 3 | 7 | 27 | Friday | 7199 | 1439 | 59 | 1324799 | 2018-07-06 | 110.4740 |
| 2018-07-07 00:00:00 | 2018 | 3 | 7 | 27 | Saturday | 7200 | 1440 | 60 | 1324800 | 2018-07-07 | 110.4740 |
source : 1324800 x 12
From above table, we can know the dataset gather from 2015-01-05 to 2018-07-07 which is used in Deriv.com - Interday High Frequency Trading Models Comparison Review (Part I).
3 ETS Modelling
3.1 Seasonal Data Modeling
Here I start my 1st forecast date from 1st trading datetime of 2016 (2016-01-04 which is 2nd year in dataset) which is used in Deriv.com - Interday High Frequency Trading Models Comparison Review (Part I).
The default data size 7200 to forecast 1440 as we can know that’s the optimal volume of observation data as we know from Deriv.com - Interday High Frequency Trading Models Comparison Review (Part I) to compare with below seasonality.
3.2 Daily Seasonal Data
3.2.1 Modelling
I set the length of data as Daily (1440 mins which is 1 trading days) to forecast 1440 minutes (1440 minutes is a trading day).
- 8.7 ARIMA modelling in R in Forecasting: Principles and Practice (2nd Edition)
- 8.9 Seasonal ARIMA models in Forecasting: Principles and Practice (2nd Edition)
- Seasonality in
auto.arima()fromforecastpackage
Refer to above function or load below r function.
3.2.2 Dy >> Dy
I set the length of data as Daily (1440 mins which is 1 trading days) to forecast 1440 mins (24 hrs * 60 mins = 1440 mins is a trading day).
tseas <- function(timeID, data = dsmp, data_len,
hrz1 = c(1440, 7200), hrz2 = 1440, .model) {
if(hrz1 == 1440) {
tmp <- llply(1:length(timeID), function(i) {
if(i == 1) {
cat('\n===========================================\n')
cat('train[', i, ']\n')
print(train <- dsmp[date < timeID[i]][(.N - (data_len - 1)):.N])
ctr <- train$sq[1]:(range(train$sq)[2] + hrz1)
cat('\n-------------------------------------------\n')
cat('train_test[', i, ']\n')
print(train_test <- dsmp[sq %in% ctr])
sets <- train[, .(index, close)] %>%
as_tibble %>%
tk_ts(frequency = hrz1) %>%
ets(model = .model) %>%
forecast(h = hrz1) %>%
tk_tbl %>%
dplyr::mutate(index = train_test[(.N - hrz1 + 1):.N,]$index,
mk.price = train_test[(.N - hrz1 + 1):.N,]$close) %>%
dplyr::rename(fc.price = `Point Forecast`) %>%
dplyr::select(index, mk.price, fc.price)
cat('\n-------------------------------------------\n')
cat('forecast[', i, ']\n')
print(sets %>% as.data.table)
fl_pth <- paste0(.dtr, 'data/fx/USDJPY/ts_ets_', data_len,
'_', hrz1, '.', as_date(sets$index[1]), '.rds')
saveRDS(sets, fl_pth)
cat('\n', i, '=', paste0('~/data/fx/USDJPY/ts_ets_', .model, '_',
data_len, '_', hrz1, '.',
as_date(sets$index[1]), '.rds saved!'))
cat('\n\n')
rm(sets)
} else if(i %in% seq(1, length(timeID), by = 6)[-1]) {
} else if(i == length(timeID)) {
} else {
lst_sq <- dsmp[date < timeID[i],][.N]$sq + 1
cat('\n===========================================\n')
cat('train[', i, ']\n')
print(train <- dsmp[(lst_sq - data_len + 1):lst_sq])
ctr <- train$sq[1]:(range(train$sq)[2] + hrz1)
cat('\n-------------------------------------------\n')
cat('train_test[', i, ']\n')
print(train_test <- dsmp[sq %in% ctr])
sets <- train[, .(index, close)] %>%
as_tibble %>%
tk_ts(frequency = hrz1) %>%
ets(model = .model) %>%
forecast(h = hrz1) %>%
tk_tbl %>%
dplyr::mutate(index = train_test[(.N - hrz1 + 1):.N,]$index,
mk.price = train_test[(.N - hrz1 + 1):.N,]$close) %>%
dplyr::rename(fc.price = `Point Forecast`) %>%
dplyr::select(index, mk.price, fc.price)
cat('\n-------------------------------------------\n')
cat('forecast[', i, ']\n')
print(sets %>% as.data.table)
fl_pth <- paste0(.dtr, 'data/fx/USDJPY/ts_ets_', data_len,
'_', hrz1, '.', as_date(sets$index[1]), '.rds')
saveRDS(sets, fl_pth)
cat('\n', i, '=', paste0('~/data/fx/USDJPY/ts_ets_', .model, '_',
data_len, '_', hrz1, '.',
as_date(sets$index[1]), '.rds saved!'))
cat('\n\n')
rm(sets)
}
})
} else if(hrz1 == 7200) {
tmp <- llply(1:length(timeID), function(i) {
if(i == 1) {
cat('\n===========================================\n')
cat('train[', i, ']\n')
print(train <- dsmp[date < timeID[i]][(.N - (data_len - 1)):.N])
ctr <- train$sq[1]:(range(train$sq)[2] + hrz1)
cat('\n-------------------------------------------\n')
cat('train_test[', i, ']\n')
print(train_test <- dsmp[sq %in% ctr])
sets <- train[, .(index, close)] %>%
as_tibble %>%
tk_ts(frequency = hrz1) %>%
ets(model = .model) %>%
forecast(h = hrz1) %>%
tk_tbl %>%
dplyr::mutate(index = train_test[(.N - hrz1 + 1):.N,]$index,
mk.price = train_test[(.N - hrz1 + 1):.N,]$close) %>%
dplyr::rename(fc.price = `Point Forecast`) %>%
dplyr::select(index, mk.price, fc.price)
cat('\n-------------------------------------------\n')
cat('forecast[', i, ']\n')
print(sets %>% as.data.table)
fl_pth <- paste0(.dtr, 'data/fx/USDJPY/ts_ets_', data_len,
'_', hrz1, '.', as_date(sets$index[1]), '.rds')
saveRDS(sets, fl_pth)
cat('\n', i, '=', paste0('~/data/fx/USDJPY/ts_ets_', .model, '_',
data_len, '_', hrz1, '.',
as_date(sets$index[1]), '.rds saved!'))
cat('\n\n')
rm(sets)
} else if(i > (length(timeID) - hrz1/hrz2) & i != length(timeID)) {
lst_sq <- dsmp[date < timeID[i],][.N]$sq + 1
## filter the length of forecasted data to fit with train_test data
## when the length of forecasted data more then length of test data.
#lst_date <- timeID[(length(timeID) - (hrz1/hrz2)):length(timeID)]
lst_date <- timeID[timeID >= timeID[i]]
lst_date_sq <- grep(
timeID[i], timeID[
(length(timeID) - (hrz1/hrz2 - 1)):length(timeID)])
cat('\n===========================================\n')
cat('train[', i, ']\n')
print(train <- dsmp[(lst_sq - data_len + 1):lst_sq])
ctr <- train$sq[1]:(range(train$sq)[2] + hrz1)
cat('\n-------------------------------------------\n')
cat('train_test[', i, ']\n')
print(train_test <- dsmp[sq %in% ctr])
sets <- train[, .(index, close)] %>%
as_tibble %>%
tk_ts(frequency = hrz1) %>%
ets(model = .model) %>%
forecast(h = hrz1) %>%
tk_tbl
sets <- sets[1:(hrz1 - (hrz2 * lst_date_sq)),] %>%
dplyr::mutate(index = train_test[
(.N - (hrz1 - (hrz2 * lst_date_sq)) + 1):.N, ]$index,
mk.price = train_test[
(.N - (hrz1 - (hrz2 * lst_date_sq)) + 1):.N, ]$close) %>%
dplyr::rename(fc.price = `Point Forecast`) %>%
dplyr::select(index, mk.price, fc.price)
cat('\n-------------------------------------------\n')
cat('forecast[', i, ']\n')
print(sets %>% as.data.table)
fl_pth <- paste0(.dtr, 'data/fx/USDJPY/ts_ets_', data_len,
'_', hrz1, '.', as_date(sets$index[1]), '.rds')
saveRDS(sets, fl_pth)
cat('\n', i, '=', paste0('~/data/fx/USDJPY/ts_ets_', .model, '_',
data_len, '_', hrz1, '.',
as_date(sets$index[1]), '.rds saved!'))
cat('\n\n')
rm(sets)
} else if(i %in% seq(1, length(timeID), by = 6)[-1]) {
} else if(i == length(timeID)) {
} else {
lst_sq <- dsmp[date < timeID[i],][.N]$sq + 1
cat('\n===========================================\n')
cat('train[', i, ']\n')
print(train <- dsmp[(lst_sq - data_len + 1):lst_sq])
ctr <- train$sq[1]:(range(train$sq)[2] + hrz1)
cat('\n-------------------------------------------\n')
cat('train_test[', i, ']\n')
print(train_test <- dsmp[sq %in% ctr])
sets <- train[, .(index, close)] %>%
as_tibble %>%
tk_ts(frequency = hrz1) %>%
ets(model = .model) %>%
forecast(h = hrz1) %>%
tk_tbl %>%
dplyr::mutate(index = train_test[(.N - hrz1 + 1):.N,]$index,
mk.price = train_test[(.N - hrz1 + 1):.N,]$close) %>%
dplyr::rename(fc.price = `Point Forecast`) %>%
dplyr::select(index, mk.price, fc.price)
cat('\n-------------------------------------------\n')
cat('forecast[', i, ']\n')
print(sets %>% as.data.table)
fl_pth <- paste0(.dtr, 'data/fx/USDJPY/ts_ets_', data_len,
'_', hrz1, '.', as_date(sets$index[1]), '.rds')
saveRDS(sets, fl_pth)
cat('\n', i, '=', paste0('~/data/fx/USDJPY/ts_ets_', .model, '_',
data_len, '_', hrz1, '.',
as_date(sets$index[1]), '.rds saved!'))
cat('\n\n')
rm(sets)
}
})
} else {
}
return(tmp)
}intra_1440 <- function(timeID, data = dsmp, data_len,
hrz1 = 60, .model, vb = TRUE) {
## data_len 1440; hrz1 = 60
intr <- data_len/hrz1
tmp <- llply(1:length(timeID), function(i) {
if(i == 1) {
tmp2 <-llply(1:intr, function(j) {
if(j == 1) {
train <- dsmp[date < timeID[i]][(.N - (data_len - 1)):.N]
} else {
lst_sq <- dsmp[date < timeID[i]][(.N - (hrz1 * (intr - j + 1) - 1))]$sq
train <- dsmp[lst_sq:(lst_sq + data_len - 1)]
}
ctr <- train$sq[1]:(range(train$sq)[2] + hrz1)
if(vb == TRUE) {
cat('\n===========================================\n')
cat('train[', i, '-', j, ']\n')
print(train)
}
train_test <- dsmp[sq %in% ctr]
if(vb == TRUE) {
cat('\n-------------------------------------------\n')
cat('train_test[', i, '-', j, ']\n')
print(train_test)
}
sets <- train[, .(index, close)] %>%
as_tibble %>%
tk_ts(frequency = hrz1) %>%
ets(model = .model) %>%
forecast(h = hrz1) %>%
tk_tbl %>%
dplyr::mutate(index = train_test[(.N - hrz1 + 1):.N,]$index,
mk.price = train_test[(.N - hrz1 + 1):.N,]$close) %>%
dplyr::rename(fc.price = `Point Forecast`) %>%
dplyr::select(index, mk.price, fc.price)
if(vb == TRUE) {
cat('\n-------------------------------------------\n')
cat('forecast[', i, '-', j, ']\n')
print(sets %>% as.data.table)
}
fl_pth <- paste0(
.dtr, 'data/fx/USDJPY/intraday/ts_ets_', .model, '_', data_len,
'_', hrz1, '.p', j, '.', as_date(sets$index[1]), '.rds')
saveRDS(sets, fl_pth)
cat('\n', i, '-', j, '=',
paste0('~/data/fx/USDJPY/intraday/ts_ets_', .model, '_',
data_len, '_', hrz1, '.p', j, '.',
as_date(sets$index[1]), '.rds saved!'))
cat('\n\n')
rm(sets)
})
return(tmp2)
} else if(i %in% seq(1, length(timeID), by = 6)[-1]) {
} else if(i == length(timeID)) {
} else {
tmp2 <-llply(1:intr, function(j) {
if(j == 1) {
lst_sq <- dsmp[date < timeID[i],][.N]$sq + 1
} else {
lst_sq <- dsmp[date < timeID[i],][.N]$sq + 1 + hrz1 * (j - 1)
}
train <- dsmp[(lst_sq - data_len + 1):lst_sq]
ctr <- train$sq[1]:(range(train$sq)[2] + hrz1)
if(vb == TRUE) {
cat('\n===========================================\n')
cat('train[', i, '-', j, ']\n')
print(train)
}
train_test <- dsmp[sq %in% ctr]
if(vb == TRUE) {
cat('\n-------------------------------------------\n')
cat('train_test[', i, '-', j, ']\n')
print(train_test)
}
sets <- train[, .(index, close)] %>%
as_tibble %>%
tk_ts(frequency = hrz1) %>%
ets(model = .model) %>%
forecast(h = hrz1) %>%
tk_tbl %>%
dplyr::mutate(index = train_test[(.N - hrz1 + 1):.N,]$index,
mk.price = train_test[(.N - hrz1 + 1):.N,]$close) %>%
dplyr::rename(fc.price = `Point Forecast`) %>%
dplyr::select(index, mk.price, fc.price)
if(vb == TRUE) {
cat('\n-------------------------------------------\n')
cat('forecast[', i, '-', j, ']\n')
print(sets %>% as.data.table)
}
fl_pth <- paste0(
.dtr, 'data/fx/USDJPY/intraday/ts_ets_', .model, '_', data_len,
'_', hrz1, '.p', j, '.', as_date(sets$index[1]), '.rds')
saveRDS(sets, fl_pth)
cat('\n', i, '-', j, '=',
paste0('~/data/fx/USDJPY/intraday/ts_ets_', .model, '_',
data_len, '_', hrz1, '.p', j, '.',
as_date(sets$index[1]), '.rds saved!'))
cat('\n\n')
rm(sets)
})
return(tmp2)
}
})
return(tmp)
}# --------- eval=FALSE ---------
ets.m <- 'MNN'
source('function/tseas.R')
source('function/intra_1440.R')
timeID <- unique(dsmp$date)
bse <- dsmp[year == 2016]$date[1] #"2016-01-04" #1st trading date in 2nd year
timeID %<>% .[. >= bse]
#timeID %<>% .[. >= as_date('2016-01-04')]
data_len <- 1440
hrz1 <- 1440
hrz2 <- 1440
llply(ets.m, function(md) {
tseas(timeID = timeID, dsmp,
data_len = data_len, hrz1 = hrz1,
hrz2 = hrz2, .model = md)
})fl <- list.files(paste0(.dtr, 'data/fx/USDJPY/'), pattern = '^ts_ets_MNN_1440_1440.[0-9]{4}')[1]
if(all(is.na(fl))){
smp <- NA
} else {
smp <- readRDS(paste0(.dtr, 'data/fx/USDJPY/', fl))
data.frame(smp)[
c(1:3, (nrow(smp)-2):nrow(smp)),] %>%
kbl(caption = 'Data Sample (ETS MNN 1440 forecast 1440)', escape = FALSE) %>%
row_spec(0, background = 'DimGrey', color = 'yellow') %>%
column_spec(1, background = 'CornflowerBlue') %>%
column_spec(2, background = '#556DAC') %>%
#column_spec(3, background = 'LightSlateGrey') %>%
#column_spec(3, background = '#556DAC') %>%
column_spec(3, background = 'Gainsboro', color = 'goldenrod') %>%
column_spec(4, background = 'LightGray', color = 'goldenrod') %>%
kable_styling(bootstrap_options = c('striped', 'hover', 'condensed', 'responsive')) %>%
kable_material(full_width = FALSE) %>%
scroll_box(width = '100%', fixed_thead = TRUE)
}| index | mk.price | fc.price | |
|---|---|---|---|
| 1 | 2016-01-04 00:01:00 | 120.2200 | 120.208 |
| 2 | 2016-01-04 00:02:00 | 120.2200 | 120.208 |
| 3 | 2016-01-04 00:03:00 | 120.2200 | 120.208 |
| 1438 | 2016-01-04 23:58:00 | 119.4535 | 120.208 |
| 1439 | 2016-01-04 23:59:00 | 119.4450 | 120.208 |
| 1440 | 2016-01-05 00:00:00 | 119.4530 | 120.208 |
source : 1440 x 3
3.2.3 Dy >> 12Hr
I set the length of data as Daily (1440 mins which is 1 trading days) to forecast 720 mins (12 hrs * 60 mins = 720 mins is half trading day).
# --------- eval=FALSE ---------
ets.m <- 'MNN'
source('function/intra_1440.R')
timeID <- unique(dsmp$date)
bse <- dsmp[year == 2016]$date[1] #"2016-01-04" #1st trading date in 2nd year
timeID %<>% .[. >= bse]
#timeID %<>% .[. >= as_date('2016-01-04')]
data_len <- 1440
hrz1 <- 720
llply(ets.m, function(md) {
intra_1440(timeID = timeID, dsmp,
data_len = data_len, hrz1 = hrz1,
.model = md)
})fl <- list.files(paste0(.dtr, 'data/fx/USDJPY/intraday/'), pattern = '^ts_ets_MNN_1440_720.p_[0-9]{0,}.[0-9]{4}')[1]
if(all(is.na(fl))){
smp <- NA
} else {
smp <- readRDS(paste0(.dtr, 'data/fx/USDJPY/intraday/', fl))
data.frame(smp)[
c(1:3, (nrow(smp)-2):nrow(smp)),] %>%
kbl(caption = 'Data Sample (ETS MNN 1440 forecast 720)', escape = FALSE) %>%
row_spec(0, background = 'DimGrey', color = 'yellow') %>%
column_spec(1, background = 'CornflowerBlue') %>%
column_spec(2, background = '#556DAC') %>%
#column_spec(3, background = 'LightSlateGrey') %>%
#column_spec(3, background = '#556DAC') %>%
column_spec(3, background = 'Gainsboro', color = 'goldenrod') %>%
column_spec(4, background = 'LightGray', color = 'goldenrod') %>%
kable_styling(bootstrap_options = c('striped', 'hover', 'condensed', 'responsive')) %>%
kable_material(full_width = FALSE) %>%
scroll_box(width = '100%', fixed_thead = TRUE)
}| index | mk.price | fc.price | |
|---|---|---|---|
| 1 | 2016-01-04 00:01:00 | 120.2200 | 120.208 |
| 2 | 2016-01-04 00:02:00 | 120.2200 | 120.208 |
| 3 | 2016-01-04 00:03:00 | 120.2200 | 120.208 |
| 718 | 2016-01-04 11:58:00 | 118.9050 | 120.208 |
| 719 | 2016-01-04 11:59:00 | 118.8765 | 120.208 |
| 720 | 2016-01-04 12:00:00 | 118.8800 | 120.208 |
source : 720 x 3
3.2.4 Dy >> 8Hr
I set the length of data as Daily (1440 mins which is 1 trading days) to forecast 480 mins (8 hrs * 60 mins = 480 mins is 8 hours in a trading day).
# --------- eval=FALSE ---------
ets.m <- 'MNN'
source('function/intra_1440.R')
timeID <- unique(dsmp$date)
bse <- dsmp[year == 2016]$date[1] #"2016-01-04" #1st trading date in 2nd year
timeID %<>% .[. >= bse]
#timeID %<>% .[. >= as_date('2016-01-04')]
data_len <- 1440
hrz1 <- 480
llply(ets.m, function(md) {
intra_1440(timeID = timeID, dsmp,
data_len = data_len, hrz1 = hrz1,
.model = md)
})fl <- list.files(paste0(.dtr, 'data/fx/USDJPY/intraday/'), pattern = '^ts_ets_MNN_1440_480.p_[0-9]{0,}.[0-9]{4}')[1]
if(all(is.na(fl))){
smp <- NA
} else {
smp <- readRDS(paste0(.dtr, 'data/fx/USDJPY/intraday/', fl))
data.frame(smp)[
c(1:3, (nrow(smp)-2):nrow(smp)),] %>%
kbl(caption = 'Data Sample (ETS MNN 1440 forecast 480)', escape = FALSE) %>%
row_spec(0, background = 'DimGrey', color = 'yellow') %>%
column_spec(1, background = 'CornflowerBlue') %>%
column_spec(2, background = '#556DAC') %>%
#column_spec(3, background = 'LightSlateGrey') %>%
#column_spec(3, background = '#556DAC') %>%
column_spec(3, background = 'Gainsboro', color = 'goldenrod') %>%
column_spec(4, background = 'LightGray', color = 'goldenrod') %>%
kable_styling(bootstrap_options = c('striped', 'hover', 'condensed', 'responsive')) %>%
kable_material(full_width = FALSE) %>%
scroll_box(width = '100%', fixed_thead = TRUE)
}| index | mk.price | fc.price | |
|---|---|---|---|
| 1 | 2016-01-04 00:01:00 | 120.2200 | 120.208 |
| 2 | 2016-01-04 00:02:00 | 120.2200 | 120.208 |
| 3 | 2016-01-04 00:03:00 | 120.2200 | 120.208 |
| 478 | 2016-01-04 07:58:00 | 119.3670 | 120.208 |
| 479 | 2016-01-04 07:59:00 | 119.3810 | 120.208 |
| 480 | 2016-01-04 08:00:00 | 119.3885 | 120.208 |
source : 480 x 3
3.2.5 Dy >> 6Hr
I set the length of data as Daily (1440 mins which is 1 trading days) to forecast 360 mins (6 hrs * 60 mins = 360 mins is 6 hours in a trading day).
# --------- eval=FALSE ---------
ets.m <- 'MNN'
source('function/intra_1440.R')
timeID <- unique(dsmp$date)
bse <- dsmp[year == 2016]$date[1] #"2016-01-04" #1st trading date in 2nd year
timeID %<>% .[. >= bse]
#timeID %<>% .[. >= as_date('2016-01-04')]
data_len <- 1440
hrz1 <- 360
llply(ets.m, function(md) {
intra_1440(timeID = timeID, dsmp,
data_len = data_len, hrz1 = hrz1,
.model = md)
})fl <- list.files(paste0(.dtr, 'data/fx/USDJPY/intraday/'), pattern = '^ts_ets_MNN_1440_360.p_[0-9]{0,}.[0-9]{4}')[1]
if(all(is.na(fl))){
smp <- NA
} else {
smp <- readRDS(paste0(.dtr, 'data/fx/USDJPY/intraday/', fl))
data.frame(smp)[
c(1:3, (nrow(smp)-2):nrow(smp)),] %>%
kbl(caption = 'Data Sample (ETS MNN 1440 forecast 360)', escape = FALSE) %>%
row_spec(0, background = 'DimGrey', color = 'yellow') %>%
column_spec(1, background = 'CornflowerBlue') %>%
column_spec(2, background = '#556DAC') %>%
#column_spec(3, background = 'LightSlateGrey') %>%
#column_spec(3, background = '#556DAC') %>%
column_spec(3, background = 'Gainsboro', color = 'goldenrod') %>%
column_spec(4, background = 'LightGray', color = 'goldenrod') %>%
kable_styling(bootstrap_options = c('striped', 'hover', 'condensed', 'responsive')) %>%
kable_material(full_width = FALSE) %>%
scroll_box(width = '100%', fixed_thead = TRUE)
}| index | mk.price | fc.price | |
|---|---|---|---|
| 1 | 2017-06-09 06:01:00 | 110.2445 | 110.237 |
| 2 | 2017-06-09 06:02:00 | 110.2605 | 110.237 |
| 3 | 2017-06-09 06:03:00 | 110.2620 | 110.237 |
| 358 | 2017-06-09 11:58:00 | 110.3850 | 110.237 |
| 359 | 2017-06-09 11:59:00 | 110.3775 | 110.237 |
| 360 | 2017-06-09 12:00:00 | 110.3790 | 110.237 |
source : 360 x 3
3.2.6 Dy >> 4Hr
I set the length of data as Daily (1440 mins which is 1 trading days) to forecast 240 mins (4 hrs * 60 mins = 240 mins is 4 hours in a trading day).
# --------- eval=FALSE ---------
ets.m <- 'MNN'
source('function/intra_1440.R')
timeID <- unique(dsmp$date)
bse <- dsmp[year == 2016]$date[1] #"2016-01-04" #1st trading date in 2nd year
timeID %<>% .[. >= bse]
#timeID %<>% .[. >= as_date('2016-01-04')]
data_len <- 1440
hrz1 <- 240
llply(ets.m, function(md) {
intra_1440(timeID = timeID, dsmp,
data_len = data_len, hrz1 = hrz1,
.model = md)
})fl <- list.files(paste0(.dtr, 'data/fx/USDJPY/intraday/'), pattern = '^ts_ets_MNN_1440_240.p_[0-9]{0,}.[0-9]{4}')[1]
if(all(is.na(fl))){
smp <- NA
} else {
smp <- readRDS(paste0(.dtr, 'data/fx/USDJPY/intraday/', fl))
data.frame(smp)[
c(1:3, (nrow(smp)-2):nrow(smp)),] %>%
kbl(caption = 'Data Sample (ETS MNN 1440 forecast 240)', escape = FALSE) %>%
row_spec(0, background = 'DimGrey', color = 'yellow') %>%
column_spec(1, background = 'CornflowerBlue') %>%
column_spec(2, background = '#556DAC') %>%
#column_spec(3, background = 'LightSlateGrey') %>%
#column_spec(3, background = '#556DAC') %>%
column_spec(3, background = 'Gainsboro', color = 'goldenrod') %>%
column_spec(4, background = 'LightGray', color = 'goldenrod') %>%
kable_styling(bootstrap_options = c('striped', 'hover', 'condensed', 'responsive')) %>%
kable_material(full_width = FALSE) %>%
scroll_box(width = '100%', fixed_thead = TRUE)
}| index | mk.price | fc.price | |
|---|---|---|---|
| 1 | 2016-01-04 00:01:00 | 120.2200 | 120.208 |
| 2 | 2016-01-04 00:02:00 | 120.2200 | 120.208 |
| 3 | 2016-01-04 00:03:00 | 120.2200 | 120.208 |
| 238 | 2016-01-04 03:58:00 | 120.0205 | 120.208 |
| 239 | 2016-01-04 03:59:00 | 120.0300 | 120.208 |
| 240 | 2016-01-04 04:00:00 | 119.9450 | 120.208 |
source : 240 x 3
3.2.7 Dy >> 3Hr
I set the length of data as Daily (1440 mins which is 1 trading days) to forecast 180 mins (3 hrs * 60 mins = 180 mins is 3 hours in a trading day).
# --------- eval=FALSE ---------
ets.m <- 'MNN'
source('function/intra_1440.R')
timeID <- unique(dsmp$date)
bse <- dsmp[year == 2016]$date[1] #"2016-01-04" #1st trading date in 2nd year
timeID %<>% .[. >= bse]
#timeID %<>% .[. >= as_date('2016-01-04')]
data_len <- 1440
hrz1 <- 180
llply(ets.m, function(md) {
intra_1440(timeID = timeID, dsmp,
data_len = data_len, hrz1 = hrz1,
.model = md)
})fl <- list.files(paste0(.dtr, 'data/fx/USDJPY/intraday/'), pattern = '^ts_ets_MNN_1440_180.p_[0-9]{0,}.[0-9]{4}')[1]
if(all(is.na(fl))){
smp <- NA
} else {
smp <- readRDS(paste0(.dtr, 'data/fx/USDJPY/intraday/', fl))
data.frame(smp)[
c(1:3, (nrow(smp)-2):nrow(smp)),] %>%
kbl(caption = 'Data Sample (ETS MNN 1440 forecast 180)', escape = FALSE) %>%
row_spec(0, background = 'DimGrey', color = 'yellow') %>%
column_spec(1, background = 'CornflowerBlue') %>%
column_spec(2, background = '#556DAC') %>%
#column_spec(3, background = 'LightSlateGrey') %>%
#column_spec(3, background = '#556DAC') %>%
column_spec(3, background = 'Gainsboro', color = 'goldenrod') %>%
column_spec(4, background = 'LightGray', color = 'goldenrod') %>%
kable_styling(bootstrap_options = c('striped', 'hover', 'condensed', 'responsive')) %>%
kable_material(full_width = FALSE) %>%
scroll_box(width = '100%', fixed_thead = TRUE)
}| index | mk.price | fc.price | |
|---|---|---|---|
| 1 | 2018-03-30 03:01:00 | 106.4765 | 106.4817 |
| 2 | 2018-03-30 03:02:00 | 106.4880 | 106.4817 |
| 3 | 2018-03-30 03:03:00 | 106.4820 | 106.4817 |
| 178 | 2018-03-30 05:58:00 | 106.2140 | 106.4817 |
| 179 | 2018-03-30 05:59:00 | 106.2130 | 106.4817 |
| 180 | 2018-03-30 06:00:00 | 106.2320 | 106.4817 |
source : 180 x 3
3.2.8 Dy >> 2Hr
I set the length of data as Daily (1440 mins which is 1 trading days) to forecast 120 mins (2 hrs * 60 mins = 120 mins is 2 hours in a trading day).
# --------- eval=FALSE ---------
ets.m <- 'MNN'
source('function/intra_1440.R')
timeID <- unique(dsmp$date)
bse <- dsmp[year == 2016]$date[1] #"2016-01-04" #1st trading date in 2nd year
timeID %<>% .[. >= bse]
#timeID %<>% .[. >= as_date('2016-01-04')]
data_len <- 1440
hrz1 <- 120
llply(ets.m, function(md) {
intra_1440(timeID = timeID, dsmp,
data_len = data_len, hrz1 = hrz1,
.model = md)
})fl <- list.files(paste0(.dtr, 'data/fx/USDJPY/intraday/'), pattern = '^ts_ets_MNN_1440_120.p_[0-9]{0,}.[0-9]{4}')[1]
if(all(is.na(fl))){
smp <- NA
} else {
smp <- readRDS(paste0(.dtr, 'data/fx/USDJPY/intraday/', fl))
data.frame(smp)[
c(1:3, (nrow(smp)-2):nrow(smp)),] %>%
kbl(caption = 'Data Sample (ETS MNN 1440 forecast 120)', escape = FALSE) %>%
row_spec(0, background = 'DimGrey', color = 'yellow') %>%
column_spec(1, background = 'CornflowerBlue') %>%
column_spec(2, background = '#556DAC') %>%
#column_spec(3, background = 'LightSlateGrey') %>%
#column_spec(3, background = '#556DAC') %>%
column_spec(3, background = 'Gainsboro', color = 'goldenrod') %>%
column_spec(4, background = 'LightGray', color = 'goldenrod') %>%
kable_styling(bootstrap_options = c('striped', 'hover', 'condensed', 'responsive')) %>%
kable_material(full_width = FALSE) %>%
scroll_box(width = '100%', fixed_thead = TRUE)
}| index | mk.price | fc.price | |
|---|---|---|---|
| 1 | 2016-08-29 20:01:00 | 102.1715 | 102.1637 |
| 2 | 2016-08-29 20:02:00 | 102.1675 | 102.1637 |
| 3 | 2016-08-29 20:03:00 | 102.1465 | 102.1637 |
| 118 | 2016-08-29 21:58:00 | 101.9625 | 102.1637 |
| 119 | 2016-08-29 21:59:00 | 101.9695 | 102.1637 |
| 120 | 2016-08-29 22:00:00 | 101.9745 | 102.1637 |
source : 120 x 3
3.2.9 Dy >> 1Hr
I set the length of data as Daily (1440 mins which is 1 trading days) to forecast 60 mins (1 hr * 60 mins = 60 mins is 1 hour in a trading day).
# --------- eval=FALSE ---------
ets.m <- 'MNN'
source('function/intra_1440.R')
timeID <- unique(dsmp$date)
bse <- dsmp[year == 2016]$date[1] #"2016-01-04" #1st trading date in 2nd year
timeID %<>% .[. >= bse]
#timeID %<>% .[. >= as_date('2016-01-04')]
data_len <- 1440
hrz1 <- 60
llply(ets.m, function(md) {
intra_1440(timeID = timeID, dsmp,
data_len = data_len, hrz1 = hrz1,
.model = md)
})fl <- list.files(paste0(.dtr, 'data/fx/USDJPY/intraday/'), pattern = '^ts_ets_MNN_1440_60.p_[0-9]{0,}.[0-9]{4}')[1]
if(all(is.na(fl))){
smp <- NA
} else {
smp <- readRDS(paste0(.dtr, 'data/fx/USDJPY/intraday/', fl))
data.frame(smp)[
c(1:3, (nrow(smp)-2):nrow(smp)),] %>%
kbl(caption = 'Data Sample (ETS MNN 1440 forecast 60)', escape = FALSE) %>%
row_spec(0, background = 'DimGrey', color = 'yellow') %>%
column_spec(1, background = 'CornflowerBlue') %>%
column_spec(2, background = '#556DAC') %>%
#column_spec(3, background = 'LightSlateGrey') %>%
#column_spec(3, background = '#556DAC') %>%
column_spec(3, background = 'Gainsboro', color = 'goldenrod') %>%
column_spec(4, background = 'LightGray', color = 'goldenrod') %>%
kable_styling(bootstrap_options = c('striped', 'hover', 'condensed', 'responsive')) %>%
kable_material(full_width = FALSE) %>%
scroll_box(width = '100%', fixed_thead = TRUE)
}| index | mk.price | fc.price | |
|---|---|---|---|
| 1 | 2018-02-07 00:01:00 | 109.5340 | 109.5435 |
| 2 | 2018-02-07 00:02:00 | 109.5650 | 109.5435 |
| 3 | 2018-02-07 00:03:00 | 109.5570 | 109.5435 |
| 58 | 2018-02-07 00:58:00 | 109.6780 | 109.5435 |
| 59 | 2018-02-07 00:59:00 | 109.7050 | 109.5435 |
| 60 | 2018-02-07 01:00:00 | 109.6945 | 109.5435 |
source : 60 x 3
3.2.10 Dy >> 30mins
I set the length of data as Daily (1440 mins which is 1 trading days) to forecast 30 mins in advance.
# --------- eval=FALSE ---------
ets.m <- 'MNN'
source('function/intra_1440.R')
timeID <- unique(dsmp$date)
bse <- dsmp[year == 2016]$date[1] #"2016-01-04" #1st trading date in 2nd year
timeID %<>% .[. >= bse]
#timeID %<>% .[. >= as_date('2016-01-04')]
data_len <- 1440
hrz1 <- 30
llply(ets.m, function(md) {
intra_1440(timeID = timeID, dsmp,
data_len = data_len, hrz1 = hrz1,
.model = md)
})fl <- list.files(paste0(.dtr, 'data/fx/USDJPY/intraday/'), pattern = '^ts_ets_MNN_1440_30.p_[0-9]{0,}.[0-9]{4}')[1]
if(all(is.na(fl))){
smp <- NA
} else {
smp <- readRDS(paste0(.dtr, 'data/fx/USDJPY/intraday/', fl))
data.frame(smp)[
c(1:3, (nrow(smp)-2):nrow(smp)),] %>%
kbl(caption = 'Data Sample (ETS MNN 1440 forecast 30)', escape = FALSE) %>%
row_spec(0, background = 'DimGrey', color = 'yellow') %>%
column_spec(1, background = 'CornflowerBlue') %>%
column_spec(2, background = '#556DAC') %>%
#column_spec(3, background = 'LightSlateGrey') %>%
#column_spec(3, background = '#556DAC') %>%
column_spec(3, background = 'Gainsboro', color = 'goldenrod') %>%
column_spec(4, background = 'LightGray', color = 'goldenrod') %>%
kable_styling(bootstrap_options = c('striped', 'hover', 'condensed', 'responsive')) %>%
kable_material(full_width = FALSE) %>%
scroll_box(width = '100%', fixed_thead = TRUE)
}| index | mk.price | fc.price | |
|---|---|---|---|
| 1 | 2017-01-26 00:01:00 | 113.2650 | 113.261 |
| 2 | 2017-01-26 00:02:00 | 113.2620 | 113.261 |
| 3 | 2017-01-26 00:03:00 | 113.2795 | 113.261 |
| 28 | 2017-01-26 00:28:00 | 113.3025 | 113.261 |
| 29 | 2017-01-26 00:29:00 | 113.2880 | 113.261 |
| 30 | 2017-01-26 00:30:00 | 113.2870 | 113.261 |
source : 30 x 3
3.2.11 Dy >> 20mins
I set the length of data as Daily (1440 mins which is 1 trading days) to forecast 20 mins in advance.
# --------- eval=FALSE ---------
ets.m <- 'MNN'
source('function/intra_1440.R')
timeID <- unique(dsmp$date)
bse <- dsmp[year == 2016]$date[1] #"2016-01-04" #1st trading date in 2nd year
timeID %<>% .[. >= bse]
#timeID %<>% .[. >= as_date('2016-01-04')]
data_len <- 1440
hrz1 <- 20
llply(ets.m, function(md) {
intra_1440(timeID = timeID, dsmp,
data_len = data_len, hrz1 = hrz1,
.model = md)
})fl <- list.files(paste0(.dtr, 'data/fx/USDJPY/intraday/'), pattern = '^ts_ets_MNN_1440_20.p_[0-9]{0,}.[0-9]{4}')[1]
if(all(is.na(fl))){
smp <- NA
} else {
smp <- readRDS(paste0(.dtr, 'data/fx/USDJPY/intraday/', fl))
data.frame(smp)[
c(1:3, (nrow(smp)-2):nrow(smp)),] %>%
kbl(caption = 'Data Sample (ETS MNN 1440 forecast 20)', escape = FALSE) %>%
row_spec(0, background = 'DimGrey', color = 'yellow') %>%
column_spec(1, background = 'CornflowerBlue') %>%
column_spec(2, background = '#556DAC') %>%
#column_spec(3, background = 'LightSlateGrey') %>%
#column_spec(3, background = '#556DAC') %>%
column_spec(3, background = 'Gainsboro', color = 'goldenrod') %>%
column_spec(4, background = 'LightGray', color = 'goldenrod') %>%
kable_styling(bootstrap_options = c('striped', 'hover', 'condensed', 'responsive')) %>%
kable_material(full_width = FALSE) %>%
scroll_box(width = '100%', fixed_thead = TRUE)
}| index | mk.price | fc.price | |
|---|---|---|---|
| 1 | 2018-04-12 03:01:00 | 106.8290 | 106.8198 |
| 2 | 2018-04-12 03:02:00 | 106.8085 | 106.8198 |
| 3 | 2018-04-12 03:03:00 | 106.8085 | 106.8198 |
| 18 | 2018-04-12 03:18:00 | 106.7110 | 106.8198 |
| 19 | 2018-04-12 03:19:00 | 106.7040 | 106.8198 |
| 20 | 2018-04-12 03:20:00 | 106.7260 | 106.8198 |
source : 20 x 3
3.2.12 Dy >> 15mins
I set the length of data as Daily (1440 mins which is 1 trading days) to forecast 15 mins in advance.
# --------- eval=FALSE ---------
ets.m <- 'MNN'
source('function/intra_1440.R')
timeID <- unique(dsmp$date)
bse <- dsmp[year == 2016]$date[1] #"2016-01-04" #1st trading date in 2nd year
timeID %<>% .[. >= bse]
#timeID %<>% .[. >= as_date('2016-01-04')]
data_len <- 1440
hrz1 <- 15
llply(ets.m, function(md) {
intra_1440(timeID = timeID, dsmp,
data_len = data_len, hrz1 = hrz1,
.model = md)
})fl <- list.files(paste0(.dtr, 'data/fx/USDJPY/intraday/'), pattern = '^ts_ets_MNN_1440_15.p_[0-9]{0,}.[0-9]{4}')[1]
if(all(is.na(fl))){
smp <- NA
} else {
smp <- readRDS(paste0(.dtr, 'data/fx/USDJPY/intraday/', fl))
data.frame(smp)[
c(1:3, (nrow(smp)-2):nrow(smp)),] %>%
kbl(caption = 'Data Sample (ETS MNN 1440 forecast 15)', escape = FALSE) %>%
row_spec(0, background = 'DimGrey', color = 'yellow') %>%
column_spec(1, background = 'CornflowerBlue') %>%
column_spec(2, background = '#556DAC') %>%
#column_spec(3, background = 'LightSlateGrey') %>%
#column_spec(3, background = '#556DAC') %>%
column_spec(3, background = 'Gainsboro', color = 'goldenrod') %>%
column_spec(4, background = 'LightGray', color = 'goldenrod') %>%
kable_styling(bootstrap_options = c('striped', 'hover', 'condensed', 'responsive')) %>%
kable_material(full_width = FALSE) %>%
scroll_box(width = '100%', fixed_thead = TRUE)
}| index | mk.price | fc.price | |
|---|---|---|---|
| 1 | 2016-06-15 02:16:00 | 106.0410 | 106.0411 |
| 2 | 2016-06-15 02:17:00 | 106.0240 | 106.0411 |
| 3 | 2016-06-15 02:18:00 | 106.0060 | 106.0411 |
| 13 | 2016-06-15 02:28:00 | 106.0210 | 106.0411 |
| 14 | 2016-06-15 02:29:00 | 106.0120 | 106.0411 |
| 15 | 2016-06-15 02:30:00 | 106.0205 | 106.0411 |
source : 15 x 3
3.2.13 Dy >> 10mins
I set the length of data as Daily (1440 mins which is 1 trading days) to forecast 10 mins in advance.
# --------- eval=FALSE ---------
ets.m <- 'MNN'
source('function/intra_1440.R')
timeID <- unique(dsmp$date)
bse <- dsmp[year == 2016]$date[1] #"2016-01-04" #1st trading date in 2nd year
timeID %<>% .[. >= bse]
#timeID %<>% .[. >= as_date('2016-01-04')]
data_len <- 1440
hrz1 <- 10
llply(ets.m, function(md) {
intra_1440(timeID = timeID, dsmp,
data_len = data_len, hrz1 = hrz1,
.model = md)
})fl <- list.files(paste0(.dtr, 'data/fx/USDJPY/intraday/'), pattern = '^ts_ets_MNN_1440_10.p_[0-9]{0,}.[0-9]{4}')[1]
if(all(is.na(fl))){
smp <- NA
} else {
smp <- readRDS(paste0(.dtr, 'data/fx/USDJPY/intraday/', fl))
data.frame(smp)[
c(1:3, (nrow(smp)-2):nrow(smp)),] %>%
kbl(caption = 'Data Sample (ETS MNN 1440 forecast 10)', escape = FALSE) %>%
row_spec(0, background = 'DimGrey', color = 'yellow') %>%
column_spec(1, background = 'CornflowerBlue') %>%
column_spec(2, background = '#556DAC') %>%
#column_spec(3, background = 'LightSlateGrey') %>%
#column_spec(3, background = '#556DAC') %>%
column_spec(3, background = 'Gainsboro', color = 'goldenrod') %>%
column_spec(4, background = 'LightGray', color = 'goldenrod') %>%
kable_styling(bootstrap_options = c('striped', 'hover', 'condensed', 'responsive')) %>%
kable_material(full_width = FALSE) %>%
scroll_box(width = '100%', fixed_thead = TRUE)
}| index | mk.price | fc.price | |
|---|---|---|---|
| 1 | 2018-04-18 03:11:00 | 107.0615 | 107.0543 |
| 2 | 2018-04-18 03:12:00 | 107.0755 | 107.0543 |
| 3 | 2018-04-18 03:13:00 | 107.0855 | 107.0543 |
| 8 | 2018-04-18 03:18:00 | 107.1230 | 107.0543 |
| 9 | 2018-04-18 03:19:00 | 107.1195 | 107.0543 |
| 10 | 2018-04-18 03:20:00 | 107.1275 | 107.0543 |
source : 10 x 3
3.2.14 Dy >> 8mins
I set the length of data as Daily (1440 mins which is 1 trading days) to forecast 8 mins in advance.
# --------- eval=FALSE ---------
ets.m <- 'MNN'
source('function/intra_1440.R')
timeID <- unique(dsmp$date)
bse <- dsmp[year == 2016]$date[1] #"2016-01-04" #1st trading date in 2nd year
timeID %<>% .[. >= bse]
#timeID %<>% .[. >= as_date('2016-01-04')]
data_len <- 1440
hrz1 <- 8
llply(ets.m, function(md) {
intra_1440(timeID = timeID, dsmp,
data_len = data_len, hrz1 = hrz1,
.model = md)
})fl <- list.files(paste0(.dtr, 'data/fx/USDJPY/intraday/'), pattern = '^ts_ets_MNN_1440_8.p_[0-9]{0,}.[0-9]{4}')[1]
if(all(is.na(fl))){
smp <- NA
} else {
smp <- readRDS(paste0(.dtr, 'data/fx/USDJPY/intraday/', fl))
data.frame(smp)[
c(1:3, (nrow(smp)-2):nrow(smp)),] %>%
kbl(caption = 'Data Sample (ETS MNN 1440 forecast 8)', escape = FALSE) %>%
row_spec(0, background = 'DimGrey', color = 'yellow') %>%
column_spec(1, background = 'CornflowerBlue') %>%
column_spec(2, background = '#556DAC') %>%
#column_spec(3, background = 'LightSlateGrey') %>%
#column_spec(3, background = '#556DAC') %>%
column_spec(3, background = 'Gainsboro', color = 'goldenrod') %>%
column_spec(4, background = 'LightGray', color = 'goldenrod') %>%
kable_styling(bootstrap_options = c('striped', 'hover', 'condensed', 'responsive')) %>%
kable_material(full_width = FALSE) %>%
scroll_box(width = '100%', fixed_thead = TRUE)
}| index | mk.price | fc.price | |
|---|---|---|---|
| 1 | 2016-05-04 13:13:00 | 106.8070 | 106.8169 |
| 2 | 2016-05-04 13:14:00 | 106.8020 | 106.8169 |
| 3 | 2016-05-04 13:15:00 | 106.7955 | 106.8169 |
| 6 | 2016-05-04 13:18:00 | 106.7970 | 106.8169 |
| 7 | 2016-05-04 13:19:00 | 106.8155 | 106.8169 |
| 8 | 2016-05-04 13:20:00 | 106.7935 | 106.8169 |
source : 8 x 3
3.2.15 Dy >> 6mins
I set the length of data as Daily (1440 mins which is 1 trading days) to forecast 6 mins in advance.
# --------- eval=FALSE ---------
ets.m <- 'MNN'
source('function/intra_1440.R')
timeID <- unique(dsmp$date)
bse <- dsmp[year == 2016]$date[1] #"2016-01-04" #1st trading date in 2nd year
timeID %<>% .[. >= bse]
#timeID %<>% .[. >= as_date('2016-01-04')]
data_len <- 1440
hrz1 <- 6
llply(ets.m, function(md) {
intra_1440(timeID = timeID, dsmp,
data_len = data_len, hrz1 = hrz1,
.model = md)
})fl <- list.files(paste0(.dtr, 'data/fx/USDJPY/intraday/'), pattern = '^ts_ets_MNN_1440_6.p_[0-9]{0,}.[0-9]{4}')[1]
if(all(is.na(fl))){
smp <- NA
} else {
smp <- readRDS(paste0(.dtr, 'data/fx/USDJPY/intraday/', fl))
smp %>%
kbl(caption = 'Data Sample (ETS MNN 1440 forecast 6)', escape = FALSE) %>%
row_spec(0, background = 'DimGrey', color = 'yellow') %>%
column_spec(1, background = '#556DAC') %>%
column_spec(2, background = 'Gainsboro', color = 'goldenrod') %>%
column_spec(3, background = 'LightGray', color = 'goldenrod') %>%
kable_styling(bootstrap_options = c('striped', 'hover', 'condensed', 'responsive')) %>%
kable_material(full_width = FALSE) %>%
scroll_box(width = '100%', fixed_thead = TRUE)
}| index | mk.price | fc.price |
|---|---|---|
| 2016-02-01 00:01:00 | 121.2675 | 121.129 |
| 2016-02-01 00:02:00 | 121.2360 | 121.129 |
| 2016-02-01 00:03:00 | 121.2585 | 121.129 |
| 2016-02-01 00:04:00 | 121.2880 | 121.129 |
| 2016-02-01 00:05:00 | 121.2500 | 121.129 |
| 2016-02-01 00:06:00 | 121.2455 | 121.129 |
source : 6 x 3
3.2.16 Dy >> 5mins
I set the length of data as Daily (1440 mins which is 1 trading days) to forecast 5 mins in advance.
# --------- eval=FALSE ---------
ets.m <- 'MNN'
source('function/intra_1440.R')
timeID <- unique(dsmp$date)
bse <- dsmp[year == 2016]$date[1] #"2016-01-04" #1st trading date in 2nd year
timeID %<>% .[. >= bse]
#timeID %<>% .[. >= as_date('2016-01-04')]
data_len <- 1440
hrz1 <- 5
llply(ets.m, function(md) {
intra_1440(timeID = timeID, dsmp,
data_len = data_len, hrz1 = hrz1,
.model = md)
})#```{r, warning = FALSE, message = FALSE, results = 'asis'}
fl <- list.files(paste0(.dtr, 'data/fx/USDJPY/intraday/'), pattern = '^ts_ets_MNN_1440_5.p_[0-9]{0,}.[0-9]{4}')[1]
if(all(is.na(fl))){
smp <- NA
} else {
smp <- readRDS(paste0(.dtr, 'data/fx/USDJPY/intraday/', fl))
smp %>%
kbl(caption = 'Data Sample (ETS MNN 1440 forecast 5)', escape = FALSE) %>%
row_spec(0, background = 'DimGrey', color = 'yellow') %>%
column_spec(1, background = '#556DAC') %>%
column_spec(2, background = 'Gainsboro', color = 'goldenrod') %>%
column_spec(3, background = 'LightGray', color = 'goldenrod') %>%
kable_styling(bootstrap_options = c('striped', 'hover', 'condensed', 'responsive')) %>%
kable_material(full_width = FALSE) %>%
scroll_box(width = '100%', fixed_thead = TRUE)
}3.2.17 Dy >> 4mins
I set the length of data as Daily (1440 mins which is 1 trading days) to forecast 4 mins in advance.
# --------- eval=FALSE ---------
ets.m <- 'MNN'
source('function/intra_1440.R')
timeID <- unique(dsmp$date)
bse <- dsmp[year == 2016]$date[1] #"2016-01-04" #1st trading date in 2nd year
timeID %<>% .[. >= bse]
#timeID %<>% .[. >= as_date('2016-01-04')]
data_len <- 1440
hrz1 <- 4
llply(ets.m, function(md) {
intra_1440(timeID = timeID, dsmp,
data_len = data_len, hrz1 = hrz1,
.model = md)
})#```{r, warning = FALSE, message = FALSE, results = 'asis'}
fl <- list.files(paste0(.dtr, 'data/fx/USDJPY/intraday/'), pattern = '^ts_ets_MNN_1440_4.p_[0-9]{0,}.[0-9]{4}')[1]
if(all(is.na(fl))){
smp <- NA
} else {
smp <- readRDS(paste0(.dtr, 'data/fx/USDJPY/intraday/', fl))
smp %>%
kbl(caption = 'Data Sample (ETS MNN 1440 forecast 4)', escape = FALSE) %>%
row_spec(0, background = 'DimGrey', color = 'yellow') %>%
column_spec(1, background = '#556DAC') %>%
column_spec(2, background = 'Gainsboro', color = 'goldenrod') %>%
column_spec(3, background = 'LightGray', color = 'goldenrod') %>%
kable_styling(bootstrap_options = c('striped', 'hover', 'condensed', 'responsive')) %>%
kable_material(full_width = FALSE) %>%
scroll_box(width = '100%', fixed_thead = TRUE)
}3.2.18 Dy >> 3mins
I set the length of data as Daily (1440 mins which is 1 trading days) to forecast 3 mins in advance.
# --------- eval=FALSE ---------
ets.m <- 'MNN'
source('function/intra_1440.R')
timeID <- unique(dsmp$date)
bse <- dsmp[year == 2016]$date[1] #"2016-01-04" #1st trading date in 2nd year
timeID %<>% .[. >= bse]
#timeID %<>% .[. >= as_date('2016-01-04')]
data_len <- 1440
hrz1 <- 3
llply(ets.m, function(md) {
intra_1440(timeID = timeID, dsmp,
data_len = data_len, hrz1 = hrz1,
.model = md)
})#```{r, warning = FALSE, message = FALSE, results = 'asis'}
fl <- list.files(paste0(.dtr, 'data/fx/USDJPY/intraday/'), pattern = '^ts_ets_MNN_1440_3.p_[0-9]{0,}.[0-9]{4}')[1]
if(all(is.na(fl))){
smp <- NA
} else {
smp <- readRDS(paste0(.dtr, 'data/fx/USDJPY/intraday/', fl))
smp %>%
kbl(caption = 'Data Sample (ETS MNN 1440 forecast 3)', escape = FALSE) %>%
row_spec(0, background = 'DimGrey', color = 'yellow') %>%
column_spec(1, background = '#556DAC') %>%
column_spec(2, background = 'Gainsboro', color = 'goldenrod') %>%
column_spec(3, background = 'LightGray', color = 'goldenrod') %>%
kable_styling(bootstrap_options = c('striped', 'hover', 'condensed', 'responsive')) %>%
kable_material(full_width = FALSE) %>%
scroll_box(width = '100%', fixed_thead = TRUE)
}3.2.19 Dy >> 2mins
I set the length of data as Daily (1440 mins which is 1 trading days) to forecast 2 mins in advance.
# --------- eval=FALSE ---------
ets.m <- 'MNN'
source('function/intra_1440.R')
timeID <- unique(dsmp$date)
bse <- dsmp[year == 2016]$date[1] #"2016-01-04" #1st trading date in 2nd year
timeID %<>% .[. >= bse]
#timeID %<>% .[. >= as_date('2016-01-04')]
data_len <- 1440
hrz1 <- 2
llply(ets.m, function(md) {
intra_1440(timeID = timeID, dsmp,
data_len = data_len, hrz1 = hrz1,
.model = md)
})#```{r, warning = FALSE, message = FALSE, results = 'asis'}
fl <- list.files(paste0(.dtr, 'data/fx/USDJPY/intraday/'), pattern = '^ts_ets_MNN_1440_2.p_[0-9]{0,}.[0-9]{4}')[1]
if(all(is.na(fl))){
smp <- NA
} else {
smp <- readRDS(paste0(.dtr, 'data/fx/USDJPY/intraday/', fl))
smp %>%
kbl(caption = 'Data Sample (ETS MNN 1440 forecast 2)', escape = FALSE) %>%
row_spec(0, background = 'DimGrey', color = 'yellow') %>%
column_spec(1, background = '#556DAC') %>%
column_spec(2, background = 'Gainsboro', color = 'goldenrod') %>%
column_spec(3, background = 'LightGray', color = 'goldenrod') %>%
kable_styling(bootstrap_options = c('striped', 'hover', 'condensed', 'responsive')) %>%
kable_material(full_width = FALSE) %>%
scroll_box(width = '100%', fixed_thead = TRUE)
}3.2.20 Dy >> 1min
I set the length of data as Daily (1440 mins which is 1 trading days) to forecast 1 min in advance.
# --------- eval=FALSE ---------
ets.m <- 'MNN'
source('function/intra_1440.R')
timeID <- unique(dsmp$date)
bse <- dsmp[year == 2016]$date[1] #"2016-01-04" #1st trading date in 2nd year
timeID %<>% .[. >= bse]
#timeID %<>% .[. >= as_date('2016-01-04')]
data_len <- 1440
hrz1 <- 1
llply(ets.m, function(md) {
intra_1440(timeID = timeID, dsmp,
data_len = data_len, hrz1 = hrz1,
.model = md)
})#```{r, warning = FALSE, message = FALSE, results = 'asis'}
fl <- list.files(paste0(.dtr, 'data/fx/USDJPY/intraday/'), pattern = '^ts_ets_MNN_1440_1.p_[0-9]{0,}.[0-9]{4}')[1:6]
#smp <- readRDS(paste0(.dtr, 'data/fx/USDJPY/intraday/', fl))
if(all(is.na(fl))){
smp <- NA
} else {
smp <- ldply(fl, function(x) {
readRDS(paste0(.dtr, 'data/fx/USDJPY/intraday/', x))
})
smp %>%
kbl(caption = 'Data Sample (ETS MNN 1440 forecast 1)', escape = FALSE) %>%
row_spec(0, background = 'DimGrey', color = 'yellow') %>%
column_spec(1, background = '#556DAC') %>%
column_spec(2, background = 'Gainsboro', color = 'goldenrod') %>%
column_spec(3, background = 'LightGray', color = 'goldenrod') %>%
kable_styling(bootstrap_options = c('striped', 'hover', 'condensed', 'responsive')) %>%
kable_material(full_width = FALSE) %>%
scroll_box(width = '100%', fixed_thead = TRUE)
}| index | mk.price | fc.price |
|---|---|---|
| 2016-01-04 00:01:00 | 120.220 | 120.2080 |
| 2016-01-05 00:01:00 | 119.449 | 119.4530 |
| 2016-01-06 00:01:00 | 119.061 | 119.0605 |
| 2016-01-07 00:01:00 | 118.477 | 118.4770 |
| 2016-01-08 00:01:00 | 117.665 | 117.6655 |
| 2016-01-11 00:01:00 | 117.182 | 117.2420 |
source : 6 x 3
3.3 Weekly Seasonal Data
3.3.1 Wk >> 12Hr
I set the length of data as Daily (5 days * 1440 mins = 7200 mins which is 5 trading days) to forecast 720 mins (12 hrs * 60 mins = 720 mins is half trading day).
# --------- eval=FALSE ---------
## set all models provided by ets function.
if(!exists('ets.m')) ets.m <- 'MNN'
if(!exists('intra_7200')) source('function/intra_7200.R')
timeID <- unique(dsmp$date)
bse <- dsmp[year == 2016]$date[1] #"2016-01-04" #1st trading date in 2nd year
timeID %<>% .[. >= bse]
#timeID %<>% .[. >= as_date('2016-01-04')]
data_len <- 7200
hrz1 <- 720
llply(ets.m, function(md) {
intra_7200(timeID = timeID, dsmp,
data_len = data_len, hrz1 = hrz1,
.model = md)
})fl <- list.files(paste0(.dtr, 'data/fx/USDJPY/intraday/'), pattern = '^ts_ets_MNN_7200_720.p_[0-9]{0,}.[0-9]{4}')[1]
if(all(is.na(fl))){
smp <- NA
} else {
smp <- readRDS(paste0(.dtr, 'data/fx/USDJPY/intraday/', fl))
data.frame(smp)[
c(1:3, (nrow(smp)-2):nrow(smp)),] %>%
kbl(caption = 'Data Sample (ETS MNN 7200 forecast 720)', escape = FALSE) %>%
row_spec(0, background = 'DimGrey', color = 'yellow') %>%
column_spec(1, background = 'CornflowerBlue') %>%
column_spec(2, background = '#556DAC') %>%
#column_spec(3, background = 'LightSlateGrey') %>%
#column_spec(3, background = '#556DAC') %>%
column_spec(3, background = 'Gainsboro', color = 'goldenrod') %>%
column_spec(4, background = 'LightGray', color = 'goldenrod') %>%
kable_styling(bootstrap_options = c('striped', 'hover', 'condensed', 'responsive')) %>%
kable_material(full_width = FALSE) %>%
scroll_box(width = '100%', fixed_thead = TRUE)
}| index | mk.price | fc.price | |
|---|---|---|---|
| 1 | 2016-08-26 00:01:00 | 100.5295 | 100.5295 |
| 2 | 2016-08-26 00:02:00 | 100.5295 | 100.5295 |
| 3 | 2016-08-26 00:03:00 | 100.5295 | 100.5295 |
| 718 | 2016-08-26 11:58:00 | 100.4645 | 100.5295 |
| 719 | 2016-08-26 11:59:00 | 100.4720 | 100.5295 |
| 720 | 2016-08-26 12:00:00 | 100.4685 | 100.5295 |
source : 720 x 3
3.3.2 Wk >> 6Hr
I set the length of data as Daily (5 days * 1440 mins = 7200 mins which is 5 trading days) to forecast 360 mins (6 hrs * 60 mins = 360 mins is half trading day).
# --------- eval=FALSE ---------
## set all models provided by ets function.
if(!exists('ets.m')) ets.m <- 'MNN'
if(!exists('intra_7200')) source('function/intra_7200.R')
timeID <- unique(dsmp$date)
bse <- dsmp[year == 2016]$date[1] #"2016-01-04" #1st trading date in 2nd year
timeID %<>% .[. >= bse]
#timeID %<>% .[. >= as_date('2016-01-04')]
data_len <- 7200
hrz1 <- 360
llply(ets.m, function(md) {
intra_7200(timeID = timeID, dsmp,
data_len = data_len, hrz1 = hrz1,
.model = md)
})fl <- list.files(paste0(.dtr, 'data/fx/USDJPY/intraday/'), pattern = '^ts_ets_MNN_7200_360.p_[0-9]{0,}.[0-9]{4}')[1]
if(all(is.na(fl))){
smp <- NA
} else {
smp <- readRDS(paste0(.dtr, 'data/fx/USDJPY/intraday/', fl))
data.frame(smp)[
c(1:3, (nrow(smp)-2):nrow(smp)),] %>%
kbl(caption = 'Data Sample (ETS MNN 7200 forecast 360)', escape = FALSE) %>%
row_spec(0, background = 'DimGrey', color = 'yellow') %>%
column_spec(1, background = 'CornflowerBlue') %>%
column_spec(2, background = '#556DAC') %>%
#column_spec(3, background = 'LightSlateGrey') %>%
#column_spec(3, background = '#556DAC') %>%
column_spec(3, background = 'Gainsboro', color = 'goldenrod') %>%
column_spec(4, background = 'LightGray', color = 'goldenrod') %>%
kable_styling(bootstrap_options = c('striped', 'hover', 'condensed', 'responsive')) %>%
kable_material(full_width = FALSE) %>%
scroll_box(width = '100%', fixed_thead = TRUE)
}3.3.3 Wk >> 3Hr
I set the length of data as Daily (5 days * 1440 mins = 7200 mins which is 5 trading days) to forecast 180 mins (3 hrs * 60 mins = 180 mins is half trading day).
# --------- eval=FALSE ---------
## set all models provided by ets function.
if(!exists('ets.m')) ets.m <- 'MNN'
if(!exists('intra_7200')) source('function/intra_7200.R')
timeID <- unique(dsmp$date)
bse <- dsmp[year == 2016]$date[1] #"2016-01-04" #1st trading date in 2nd year
timeID %<>% .[. >= bse]
#timeID %<>% .[. >= as_date('2016-01-04')]
data_len <- 7200
hrz1 <- 180
llply(ets.m, function(md) {
intra_7200(timeID = timeID, dsmp,
data_len = data_len, hrz1 = hrz1,
.model = md)
})fl <- list.files(paste0(.dtr, 'data/fx/USDJPY/intraday/'), pattern = '^ts_ets_MNN_7200_180.p_[0-9]{0,}.[0-9]{4}')[1]
if(all(is.na(fl))){
smp <- NA
} else {
smp <- readRDS(paste0(.dtr, 'data/fx/USDJPY/intraday/', fl))
data.frame(smp)[
c(1:3, (nrow(smp)-2):nrow(smp)),] %>%
kbl(caption = 'Data Sample (ETS MNN 7200 forecast 180)', escape = FALSE) %>%
row_spec(0, background = 'DimGrey', color = 'yellow') %>%
column_spec(1, background = 'CornflowerBlue') %>%
column_spec(2, background = '#556DAC') %>%
#column_spec(3, background = 'LightSlateGrey') %>%
#column_spec(3, background = '#556DAC') %>%
column_spec(3, background = 'Gainsboro', color = 'goldenrod') %>%
column_spec(4, background = 'LightGray', color = 'goldenrod') %>%
kable_styling(bootstrap_options = c('striped', 'hover', 'condensed', 'responsive')) %>%
kable_material(full_width = FALSE) %>%
scroll_box(width = '100%', fixed_thead = TRUE)
}3.3.4 Wk >> 1Hr
I set the length of data as Daily (5 days * 1440 mins = 7200 mins which is 5 trading days) to forecast 60 mins (3 hrs * 60 mins = 60 mins is half trading day).
# --------- eval=FALSE ---------
## set all models provided by ets function.
if(!exists('ets.m')) ets.m <- 'MNN'
if(!exists('intra_7200')) source('function/intra_7200.R')
timeID <- unique(dsmp$date)
bse <- dsmp[year == 2016]$date[1] #"2016-01-04" #1st trading date in 2nd year
timeID %<>% .[. >= bse]
#timeID %<>% .[. >= as_date('2016-01-04')]
data_len <- 7200
hrz1 <- 60
llply(ets.m, function(md) {
intra_7200(timeID = timeID, dsmp,
data_len = data_len, hrz1 = hrz1,
.model = md)
})fl <- list.files(paste0(.dtr, 'data/fx/USDJPY/intraday/'), pattern = '^ts_ets_MNN_7200_60.p_[0-9]{0,}.[0-9]{4}')[1]
if(all(is.na(fl))){
smp <- NA
} else {
smp <- readRDS(paste0(.dtr, 'data/fx/USDJPY/intraday/', fl))
data.frame(smp)[
c(1:3, (nrow(smp)-2):nrow(smp)),] %>%
kbl(caption = 'Data Sample (ETS MNN 7200 forecast 60)', escape = FALSE) %>%
row_spec(0, background = 'DimGrey', color = 'yellow') %>%
column_spec(1, background = 'CornflowerBlue') %>%
column_spec(2, background = '#556DAC') %>%
#column_spec(3, background = 'LightSlateGrey') %>%
#column_spec(3, background = '#556DAC') %>%
column_spec(3, background = 'Gainsboro', color = 'goldenrod') %>%
column_spec(4, background = 'LightGray', color = 'goldenrod') %>%
kable_styling(bootstrap_options = c('striped', 'hover', 'condensed', 'responsive')) %>%
kable_material(full_width = FALSE) %>%
scroll_box(width = '100%', fixed_thead = TRUE)
}3.4 Intraday Seasonal Data
3.4.1 20Hr >> 1min
I set the length of data as Daily (1200 mins which is 1 trading days) to forecast 1 min in advance.
# --------- eval=FALSE ---------
ets.m <- 'MNN'
source('function/intra_min.R')
timeID <- unique(dsmp$date)
bse <- dsmp[year == 2016]$date[1] #"2016-01-04" #1st trading date in 2nd year
timeID %<>% .[. >= bse]
#timeID %<>% .[. >= as_date('2016-01-04')]
data_len <- 1200
hrz1 <- 1
llply(ets.m, function(md) {
intra_min(timeID = timeID, dsmp,
data_len = data_len, hrz1 = hrz1,
.model = md)
})#```{r, warning = FALSE, message = FALSE, results = 'asis'}
fl <- list.files(paste0(.dtr, 'data/fx/USDJPY/intraday/'), pattern = '^ts_ets_MNN_1200_1.p_[0-9]{0,}.[0-9]{4}')[1:6]
#smp <- readRDS(paste0(.dtr, 'data/fx/USDJPY/intraday/', fl))
if(all(is.na(fl))){
smp <- NA
} else {
smp <- ldply(fl, function(x) {
readRDS(paste0(.dtr, 'data/fx/USDJPY/intraday/', x))
})
smp %>%
kbl(caption = 'Data Sample (ETS MNN 1200 forecast 1)', escape = FALSE) %>%
row_spec(0, background = 'DimGrey', color = 'yellow') %>%
column_spec(1, background = '#556DAC') %>%
column_spec(2, background = 'Gainsboro', color = 'goldenrod') %>%
column_spec(3, background = 'LightGray', color = 'goldenrod') %>%
kable_styling(bootstrap_options = c('striped', 'hover', 'condensed', 'responsive')) %>%
kable_material(full_width = FALSE) %>%
scroll_box(width = '100%', fixed_thead = TRUE)
}3.4.2 18Hr >> 1min
I set the length of data as Daily (1080 mins which is 1 trading days) to forecast 1 min in advance.
# --------- eval=FALSE ---------
ets.m <- 'MNN'
source('function/intra_min.R')
timeID <- unique(dsmp$date)
bse <- dsmp[year == 2016]$date[1] #"2016-01-04" #1st trading date in 2nd year
timeID %<>% .[. >= bse]
#timeID %<>% .[. >= as_date('2016-01-04')]
data_len <- 1080
hrz1 <- 1
llply(ets.m, function(md) {
intra_min(timeID = timeID, dsmp,
data_len = data_len, hrz1 = hrz1,
.model = md)
})#```{r, warning = FALSE, message = FALSE, results = 'asis'}
fl <- list.files(paste0(.dtr, 'data/fx/USDJPY/intraday/'), pattern = '^ts_ets_MNN_1080_1.p_[0-9]{0,}.[0-9]{4}')[1:6]
#smp <- readRDS(paste0(.dtr, 'data/fx/USDJPY/intraday/', fl))
if(all(is.na(fl))){
smp <- NA
} else {
smp <- ldply(fl, function(x) {
readRDS(paste0(.dtr, 'data/fx/USDJPY/intraday/', x))
})
smp %>%
kbl(caption = 'Data Sample (ETS MNN 1080 forecast 1)', escape = FALSE) %>%
row_spec(0, background = 'DimGrey', color = 'yellow') %>%
column_spec(1, background = '#556DAC') %>%
column_spec(2, background = 'Gainsboro', color = 'goldenrod') %>%
column_spec(3, background = 'LightGray', color = 'goldenrod') %>%
kable_styling(bootstrap_options = c('striped', 'hover', 'condensed', 'responsive')) %>%
kable_material(full_width = FALSE) %>%
scroll_box(width = '100%', fixed_thead = TRUE)
}3.4.3 16Hr >> 1min
I set the length of data as Daily (960 mins which is 1 trading days) to forecast 1 min in advance.
# --------- eval=FALSE ---------
ets.m <- 'MNN'
source('function/intra_min.R')
timeID <- unique(dsmp$date)
bse <- dsmp[year == 2016]$date[1] #"2016-01-04" #1st trading date in 2nd year
timeID %<>% .[. >= bse]
#timeID %<>% .[. >= as_date('2016-01-04')]
data_len <- 960
hrz1 <- 1
llply(ets.m, function(md) {
intra_min(timeID = timeID, dsmp,
data_len = data_len, hrz1 = hrz1,
.model = md)
})#```{r, warning = FALSE, message = FALSE, results = 'asis'}
fl <- list.files(paste0(.dtr, 'data/fx/USDJPY/intraday/'), pattern = '^ts_ets_MNN_960_1.p_[0-9]{0,}.[0-9]{4}')[1:6]
#smp <- readRDS(paste0(.dtr, 'data/fx/USDJPY/intraday/', fl))
if(all(is.na(fl))){
smp <- NA
} else {
smp <- ldply(fl, function(x) {
readRDS(paste0(.dtr, 'data/fx/USDJPY/intraday/', x))
})
smp %>%
kbl(caption = 'Data Sample (ETS MNN 960 forecast 1)', escape = FALSE) %>%
row_spec(0, background = 'DimGrey', color = 'yellow') %>%
column_spec(1, background = '#556DAC') %>%
column_spec(2, background = 'Gainsboro', color = 'goldenrod') %>%
column_spec(3, background = 'LightGray', color = 'goldenrod') %>%
kable_styling(bootstrap_options = c('striped', 'hover', 'condensed', 'responsive')) %>%
kable_material(full_width = FALSE) %>%
scroll_box(width = '100%', fixed_thead = TRUE)
}| index | mk.price | fc.price |
|---|---|---|
| 2016-01-04 00:02:00 | 120.220 | 120.2104 |
| 2016-01-05 00:01:00 | 119.449 | 119.4530 |
| 2016-01-06 00:01:00 | 119.061 | 119.0605 |
| 2016-01-07 00:01:00 | 118.477 | 118.4769 |
| 2016-01-08 00:01:00 | 117.665 | 117.6655 |
| 2016-01-11 00:02:00 | 117.183 | 117.1844 |
3.4.4 14Hr >> 1min
I set the length of data as Daily (840 mins which is 1 trading days) to forecast 1 min in advance.
# --------- eval=FALSE ---------
ets.m <- 'MNN'
source('function/intra_min.R')
timeID <- unique(dsmp$date)
bse <- dsmp[year == 2016]$date[1] #"2016-01-04" #1st trading date in 2nd year
timeID %<>% .[. >= bse]
#timeID %<>% .[. >= as_date('2016-01-04')]
data_len <- 840
hrz1 <- 1
llply(ets.m, function(md) {
intra_min(timeID = timeID, dsmp,
data_len = data_len, hrz1 = hrz1,
.model = md)
})#```{r, warning = FALSE, message = FALSE, results = 'asis'}
fl <- list.files(paste0(.dtr, 'data/fx/USDJPY/intraday/'), pattern = '^ts_ets_MNN_840_1.p_[0-9]{0,}.[0-9]{4}')[1:6]
#smp <- readRDS(paste0(.dtr, 'data/fx/USDJPY/intraday/', fl))
if(all(is.na(fl))){
smp <- NA
} else {
smp <- ldply(fl, function(x) {
readRDS(paste0(.dtr, 'data/fx/USDJPY/intraday/', x))
})
smp %>%
kbl(caption = 'Data Sample (ETS MNN 840 forecast 1)', escape = FALSE) %>%
row_spec(0, background = 'DimGrey', color = 'yellow') %>%
column_spec(1, background = '#556DAC') %>%
column_spec(2, background = 'Gainsboro', color = 'goldenrod') %>%
column_spec(3, background = 'LightGray', color = 'goldenrod') %>%
kable_styling(bootstrap_options = c('striped', 'hover', 'condensed', 'responsive')) %>%
kable_material(full_width = FALSE) %>%
scroll_box(width = '100%', fixed_thead = TRUE)
}| index | mk.price | fc.price |
|---|---|---|
| 2016-02-01 00:01:00 | 121.2675 | 121.1290 |
| 2016-02-02 00:01:00 | 121.0180 | 121.0090 |
| 2016-02-03 00:01:00 | 119.9710 | 119.9625 |
| 2016-02-04 00:01:00 | 117.8940 | 117.8950 |
| 2016-02-05 00:01:00 | 116.7770 | 116.7750 |
| 2016-02-08 00:02:00 | 116.8525 | 116.8475 |
3.4.5 12Hr >> 1min
I set the length of data as Daily (720 mins which is 1 trading days) to forecast 1 min in advance.
# --------- eval=FALSE ---------
ets.m <- 'MNN'
source('function/intra_min.R')
timeID <- unique(dsmp$date)
bse <- dsmp[year == 2016]$date[1] #"2016-01-04" #1st trading date in 2nd year
timeID %<>% .[. >= bse]
#timeID %<>% .[. >= as_date('2016-01-04')]
data_len <- 720
hrz1 <- 1
llply(ets.m, function(md) {
intra_min(timeID = timeID, dsmp,
data_len = data_len, hrz1 = hrz1,
.model = md)
})#```{r, warning = FALSE, message = FALSE, results = 'asis'}
fl <- list.files(paste0(.dtr, 'data/fx/USDJPY/intraday/'), pattern = '^ts_ets_MNN_720_1.p_[0-9]{0,}.[0-9]{4}')[1:6]
#smp <- readRDS(paste0(.dtr, 'data/fx/USDJPY/intraday/', fl))
if(all(is.na(fl))){
smp <- NA
} else {
smp <- ldply(fl, function(x) {
readRDS(paste0(.dtr, 'data/fx/USDJPY/intraday/', x))
})
smp %>%
kbl(caption = 'Data Sample (ETS MNN 720 forecast 1)', escape = FALSE) %>%
row_spec(0, background = 'DimGrey', color = 'yellow') %>%
column_spec(1, background = '#556DAC') %>%
column_spec(2, background = 'Gainsboro', color = 'goldenrod') %>%
column_spec(3, background = 'LightGray', color = 'goldenrod') %>%
kable_styling(bootstrap_options = c('striped', 'hover', 'condensed', 'responsive')) %>%
kable_material(full_width = FALSE) %>%
scroll_box(width = '100%', fixed_thead = TRUE)
}| index | mk.price | fc.price |
|---|---|---|
| 2016-01-04 00:02:00 | 120.220 | 120.2104 |
| 2016-01-05 00:01:00 | 119.449 | 119.4530 |
| 2016-01-06 00:01:00 | 119.061 | 119.0605 |
| 2016-01-07 00:01:00 | 118.477 | 118.4768 |
| 2016-01-08 00:01:00 | 117.665 | 117.6655 |
| 2016-01-11 00:02:00 | 117.183 | 117.1843 |
3.4.6 10Hr >> 1min
I set the length of data as Daily (600 mins which is 1 trading days) to forecast 1 min in advance.
# --------- eval=FALSE ---------
ets.m <- 'MNN'
source('function/intra_min.R')
timeID <- unique(dsmp$date)
bse <- dsmp[year == 2016]$date[1] #"2016-01-04" #1st trading date in 2nd year
timeID %<>% .[. >= bse]
#timeID %<>% .[. >= as_date('2016-01-04')]
data_len <- 600
hrz1 <- 1
llply(ets.m, function(md) {
intra_min(timeID = timeID, dsmp,
data_len = data_len, hrz1 = hrz1,
.model = md)
})#```{r, warning = FALSE, message = FALSE, results = 'asis'}
fl <- list.files(paste0(.dtr, 'data/fx/USDJPY/intraday/'), pattern = '^ts_ets_MNN_600_1.p_[0-9]{0,}.[0-9]{4}')[1:6]
#smp <- readRDS(paste0(.dtr, 'data/fx/USDJPY/intraday/', fl))
if(all(is.na(fl))){
smp <- NA
} else {
smp <- ldply(fl, function(x) {
readRDS(paste0(.dtr, 'data/fx/USDJPY/intraday/', x))
})
smp %>%
kbl(caption = 'Data Sample (ETS MNN 600 forecast 1)', escape = FALSE) %>%
row_spec(0, background = 'DimGrey', color = 'yellow') %>%
column_spec(1, background = '#556DAC') %>%
column_spec(2, background = 'Gainsboro', color = 'goldenrod') %>%
column_spec(3, background = 'LightGray', color = 'goldenrod') %>%
kable_styling(bootstrap_options = c('striped', 'hover', 'condensed', 'responsive')) %>%
kable_material(full_width = FALSE) %>%
scroll_box(width = '100%', fixed_thead = TRUE)
}| index | mk.price | fc.price |
|---|---|---|
| 2016-01-04 00:02:00 | 120.220 | 120.2104 |
| 2016-01-05 00:01:00 | 119.449 | 119.4530 |
| 2016-01-06 00:01:00 | 119.061 | 119.0605 |
| 2016-01-07 00:01:00 | 118.477 | 118.4770 |
| 2016-01-08 00:01:00 | 117.665 | 117.6655 |
| 2016-01-11 00:02:00 | 117.183 | 117.1841 |
3.4.7 8Hr >> 1min
I set the length of data as Daily (480 mins which is 1 trading days) to forecast 1 min in advance.
# --------- eval=FALSE ---------
ets.m <- 'MNN'
source('function/intra_min.R')
timeID <- unique(dsmp$date)
bse <- dsmp[year == 2016]$date[1] #"2016-01-04" #1st trading date in 2nd year
timeID %<>% .[. >= bse]
#timeID %<>% .[. >= as_date('2016-01-04')]
data_len <- 480
hrz1 <- 1
llply(ets.m, function(md) {
intra_min(timeID = timeID, dsmp,
data_len = data_len, hrz1 = hrz1,
.model = md)
})#```{r, warning = FALSE, message = FALSE, results = 'asis'}
fl <- list.files(paste0(.dtr, 'data/fx/USDJPY/intraday/'), pattern = '^ts_ets_MNN_480_1.p_[0-9]{0,}.[0-9]{4}')[1:6]
#smp <- readRDS(paste0(.dtr, 'data/fx/USDJPY/intraday/', fl))
smp <- ldply(fl, function(x) {
readRDS(paste0(.dtr, 'data/fx/USDJPY/intraday/', x))
})
smp %>%
kbl(caption = 'Data Sample (ETS MNN 480 forecast 1)', escape = FALSE) %>%
row_spec(0, background = 'DimGrey', color = 'yellow') %>%
column_spec(1, background = '#556DAC') %>%
column_spec(2, background = 'Gainsboro', color = 'goldenrod') %>%
column_spec(3, background = 'LightGray', color = 'goldenrod') %>%
kable_styling(bootstrap_options = c('striped', 'hover', 'condensed', 'responsive')) %>%
kable_material(full_width = FALSE) %>%
scroll_box(width = '100%', fixed_thead = TRUE)| index | mk.price | fc.price |
|---|---|---|
| 2016-01-04 00:02:00 | 120.220 | 120.2104 |
| 2016-01-05 00:01:00 | 119.449 | 119.4530 |
| 2016-01-06 00:01:00 | 119.061 | 119.0605 |
| 2016-01-07 00:01:00 | 118.477 | 118.4769 |
| 2016-01-08 00:01:00 | 117.665 | 117.6655 |
| 2016-01-11 00:02:00 | 117.183 | 117.1872 |
source : 6 x 3
3.4.8 6Hr >> 1min
I set the length of data as Daily (360 mins which is 1 trading days) to forecast 1 min in advance.
# --------- eval=FALSE ---------
ets.m <- 'MNN'
source('function/intra_min.R')
timeID <- unique(dsmp$date)
bse <- dsmp[year == 2016]$date[1] #"2016-01-04" #1st trading date in 2nd year
timeID %<>% .[. >= bse]
#timeID %<>% .[. >= as_date('2016-01-04')]
data_len <- 360
hrz1 <- 1
llply(ets.m, function(md) {
intra_min(timeID = timeID, dsmp,
data_len = data_len, hrz1 = hrz1,
.model = md)
})#```{r, warning = FALSE, message = FALSE, results = 'asis'}
fl <- list.files(paste0(.dtr, 'data/fx/USDJPY/intraday/'), pattern = '^ts_ets_MNN_360_1.p_[0-9]{0,}.[0-9]{4}')[1:6]
#smp <- readRDS(paste0(.dtr, 'data/fx/USDJPY/intraday/', fl))
smp <- ldply(fl, function(x) {
readRDS(paste0(.dtr, 'data/fx/USDJPY/intraday/', x))
})
smp %>%
kbl(caption = 'Data Sample (ETS MNN 360 forecast 1)', escape = FALSE) %>%
row_spec(0, background = 'DimGrey', color = 'yellow') %>%
column_spec(1, background = '#556DAC') %>%
column_spec(2, background = 'Gainsboro', color = 'goldenrod') %>%
column_spec(3, background = 'LightGray', color = 'goldenrod') %>%
kable_styling(bootstrap_options = c('striped', 'hover', 'condensed', 'responsive')) %>%
kable_material(full_width = FALSE) %>%
scroll_box(width = '100%', fixed_thead = TRUE)| index | mk.price | fc.price |
|---|---|---|
| 2016-01-04 00:01:00 | 120.220 | 120.2080 |
| 2016-01-05 00:01:00 | 119.449 | 119.4528 |
| 2016-01-06 00:01:00 | 119.061 | 119.0605 |
| 2016-01-07 00:01:00 | 118.477 | 118.4766 |
| 2016-01-08 00:01:00 | 117.665 | 117.6656 |
| 2016-01-11 00:01:00 | 117.182 | 117.2420 |
source : 6 x 3
3.4.9 4Hr >> 1min
I set the length of data as Daily (240 mins which is 1 trading days) to forecast 1 min in advance.
# --------- eval=FALSE ---------
ets.m <- 'MNN'
source('function/intra_min.R')
timeID <- unique(dsmp$date)
bse <- dsmp[year == 2016]$date[1] #"2016-01-04" #1st trading date in 2nd year
timeID %<>% .[. >= bse]
#timeID %<>% .[. >= as_date('2016-01-04')]
data_len <- 240
hrz1 <- 1
llply(ets.m, function(md) {
intra_min(timeID = timeID, dsmp,
data_len = data_len, hrz1 = hrz1,
.model = md)
})#```{r, warning = FALSE, message = FALSE, results = 'asis'}
fl <- list.files(paste0(.dtr, 'data/fx/USDJPY/intraday/'), pattern = '^ts_ets_MNN_240_1.p_[0-9]{0,}.[0-9]{4}')[1:6]
#smp <- readRDS(paste0(.dtr, 'data/fx/USDJPY/intraday/', fl))
if(all(is.na(fl))){
smp <- NA
} else {
smp <- ldply(fl, function(x) {
readRDS(paste0(.dtr, 'data/fx/USDJPY/intraday/', x))
})
smp %>%
kbl(caption = 'Data Sample (ETS MNN 240 forecast 1)', escape = FALSE) %>%
row_spec(0, background = 'DimGrey', color = 'yellow') %>%
column_spec(1, background = '#556DAC') %>%
column_spec(2, background = 'Gainsboro', color = 'goldenrod') %>%
column_spec(3, background = 'LightGray', color = 'goldenrod') %>%
kable_styling(bootstrap_options = c('striped', 'hover', 'condensed', 'responsive')) %>%
kable_material(full_width = FALSE) %>%
scroll_box(width = '100%', fixed_thead = TRUE)
}| index | mk.price | fc.price |
|---|---|---|
| 2016-01-04 00:02:00 | 120.220 | 120.2104 |
| 2016-01-05 00:01:00 | 119.449 | 119.4528 |
| 2016-01-06 00:01:00 | 119.061 | 119.0605 |
| 2016-01-07 00:01:00 | 118.477 | 118.4763 |
| 2016-01-08 00:01:00 | 117.665 | 117.6655 |
| 2016-01-11 00:02:00 | 117.183 | 117.1820 |
source : 6 x 3
3.4.10 3Hr >> 1min
I set the length of data as Daily (180 mins which is 1 trading days) to forecast 1 min in advance.
# --------- eval=FALSE ---------
ets.m <- 'MNN'
source('function/intra_min.R')
timeID <- unique(dsmp$date)
bse <- dsmp[year == 2016]$date[1] #"2016-01-04" #1st trading date in 2nd year
timeID %<>% .[. >= bse]
#timeID %<>% .[. >= as_date('2016-01-04')]
data_len <- 180
hrz1 <- 1
llply(ets.m, function(md) {
intra_min(timeID = timeID, dsmp,
data_len = data_len, hrz1 = hrz1,
.model = md)
})#```{r, warning = FALSE, message = FALSE, results = 'asis'}
fl <- list.files(paste0(.dtr, 'data/fx/USDJPY/intraday/'), pattern = '^ts_ets_MNN_180_1.p_[0-9]{0,}.[0-9]{4}')[1:6]
#smp <- readRDS(paste0(.dtr, 'data/fx/USDJPY/intraday/', fl))
if(all(is.na(fl))){
smp <- NA
} else {
smp <- ldply(fl, function(x) {
readRDS(paste0(.dtr, 'data/fx/USDJPY/intraday/', x))
})
smp %>%
kbl(caption = 'Data Sample (ETS MNN 180 forecast 1)', escape = FALSE) %>%
row_spec(0, background = 'DimGrey', color = 'yellow') %>%
column_spec(1, background = '#556DAC') %>%
column_spec(2, background = 'Gainsboro', color = 'goldenrod') %>%
column_spec(3, background = 'LightGray', color = 'goldenrod') %>%
kable_styling(bootstrap_options = c('striped', 'hover', 'condensed', 'responsive')) %>%
kable_material(full_width = FALSE) %>%
scroll_box(width = '100%', fixed_thead = TRUE)
}| index | mk.price | fc.price |
|---|---|---|
| 2016-01-04 00:02:00 | 120.220 | 120.2104 |
| 2016-01-05 00:01:00 | 119.449 | 119.4527 |
| 2016-01-06 00:01:00 | 119.061 | 119.0605 |
| 2016-01-07 00:01:00 | 118.477 | 118.4760 |
| 2016-01-08 00:01:00 | 117.665 | 117.6656 |
| 2016-01-11 00:02:00 | 117.183 | 117.1820 |
source : 6 x 3
3.4.11 2Hr >> 1min
I set the length of data as Daily (120 mins which is 1 trading days) to forecast 1 min in advance.
# --------- eval=FALSE ---------
ets.m <- 'MNN'
source('function/intra_min.R')
timeID <- unique(dsmp$date)
bse <- dsmp[year == 2016]$date[1] #"2016-01-04" #1st trading date in 2nd year
timeID %<>% .[. >= bse]
#timeID %<>% .[. >= as_date('2016-01-04')]
data_len <- 120
hrz1 <- 1
llply(ets.m, function(md) {
intra_min(timeID = timeID, dsmp,
data_len = data_len, hrz1 = hrz1,
.model = md)
})#```{r, warning = FALSE, message = FALSE, results = 'asis'}
fl <- list.files(paste0(.dtr, 'data/fx/USDJPY/intraday/'), pattern = '^ts_ets_MNN_120_1.p_[0-9]{0,}.[0-9]{4}')[1:6]
#smp <- readRDS(paste0(.dtr, 'data/fx/USDJPY/intraday/', fl))
if(all(is.na(fl))){
smp <- NA
} else {
smp <- ldply(fl, function(x) {
readRDS(paste0(.dtr, 'data/fx/USDJPY/intraday/', x))
})
smp %>%
kbl(caption = 'Data Sample (ETS MNN 120 forecast 1)', escape = FALSE) %>%
row_spec(0, background = 'DimGrey', color = 'yellow') %>%
column_spec(1, background = '#556DAC') %>%
column_spec(2, background = 'Gainsboro', color = 'goldenrod') %>%
column_spec(3, background = 'LightGray', color = 'goldenrod') %>%
kable_styling(bootstrap_options = c('striped', 'hover', 'condensed', 'responsive')) %>%
kable_material(full_width = FALSE) %>%
scroll_box(width = '100%', fixed_thead = TRUE)
}4 Comparison
4.1 1st Stage
4.1.1 Read Models
4.1.1.1 Grouped Models
Here I read the saved models. As some have mentioned a possible way is using git interactive staging. This is great when you have files with different extensions
$ git add -A
$ git statusSource : git add -A和 git add . git add -u的区别
$ git add -i
staged unstaged path
1: unchanged +0/-1 TODO
2: unchanged +1/-1 index.html
3: unchanged +5/-1 lib/simplegit.rb
*** Commands ***
1: status 2: update 3: revert 4: add untracked
5: patch 6: diff 7: quit 8: help
What now>If you press 2 then enter you will get a list of available files to be added:
What now> 2
staged unstaged path
1: unchanged +0/-1 TODO
2: unchanged +1/-1 index.html
3: unchanged +5/-1 lib/simplegit.rb
Update>>Now you just have to insert the number of the files you want to add, so if we wanted to add TODO and index.html we would type 1,2
Update>> 1,2
staged unstaged path
* 1: unchanged +0/-1 TODO
* 2: unchanged +1/-1 index.html
3: unchanged +5/-1 lib/simplegit.rb
Update>>You see the * before the number? that means that the file was added.
Now imagine that you have 7 files and you want to add them all except the 7th? Sure we could type 1,2,3,4,5,6 but imagine instead of 7 we have 16, that would be quite cumbersome, the good thing we don’t need to type them all because we can use ranges,by typing 1-6
Update>> 1-6
staged unstaged path
* 1: unchanged +0/-1 TODO
* 2: unchanged +1/-1 index.html
* 3: unchanged +5/-1 lib/simplegit.rb
* 4: unchanged +5/-1 file4.html
* 5: unchanged +5/-1 file5.html
* 6: unchanged +5/-1 file6.html
7: unchanged +5/-1 file7.html
Update>>We can even use multiple ranges, so if we want from 1 to 3 and from 5 to 7 we type 1-3, 5-7:
Update>> 1-3, 5-7
staged unstaged path
* 1: unchanged +0/-1 TODO
* 2: unchanged +1/-1 index.html
* 3: unchanged +5/-1 lib/simplegit.rb
4: unchanged +5/-1 file4.html
* 5: unchanged +5/-1 file5.html
* 6: unchanged +5/-1 file6.html
* 7: unchanged +5/-1 file7.html
Update>>We can also use this to unstage files, if we type -number, so if we wanted to unstage file number 1 we would type -1:
Update>> -1
staged unstaged path
1: unchanged +0/-1 TODO
* 2: unchanged +1/-1 index.html
* 3: unchanged +5/-1 lib/simplegit.rb
4: unchanged +5/-1 file4.html
* 5: unchanged +5/-1 file5.html
* 6: unchanged +5/-1 file6.html
* 7: unchanged +5/-1 file7.html
Update>>And as you can imagine we can also unstage a range of files, so if we type -range all the files on that range would be unstaged. If we wanted to unstage all the files from 5 to 7 we would type -5-7:
Update>> -5-7
staged unstaged path
1: unchanged +0/-1 TODO
* 2: unchanged +1/-1 index.html
* 3: unchanged +5/-1 lib/simplegit.rb
4: unchanged +5/-1 file4.html
5: unchanged +5/-1 file5.html
6: unchanged +5/-1 file6.html
7: unchanged +5/-1 file7.html
Update>>Source : How to add multiple files to Git at the same time.
You can commit all files by doing a `git add .` or just specifying the directory that they're in: `git add ./folder/` or to add all files that begin with "test": `git add test*`Source : How to add or commit specific files without specifying their full path?
## Get all files.
fls <- paste0(.dtr, 'data/fx/USDJPY/') %>%
list.files(., pattern = '^ts_ets_MNN_7200_1440.[0-9]{4}')
fls_pth <- paste0(.dtr, 'data/fx/USDJPY/', fls)
mds_ets_MNN_7200_1440 <- ldply(1:length(fls), function(i) {
nms <- fls[i] %>%
str_replace_all('.rds', '') %>%
str_split('_|\\.') %>%
.[[1]]
nms <- nms
names(nms) <- c('t_series', 'model', 'sub_model',
'data_min', 'fc_min', 'date')
datset2 <- t(nms) %>%
data.frame %>%
mutate(model =factor(model), sub_model = factor(sub_model),
data_min = as.numeric(data_min),
fc_min = as.numeric(fc_min),
part = 1,
date = as_date(date))
datset <- read_rds(fls_pth[i])
res <- tibble(datset2, datset) %>%
dplyr::select(index, t_series, model, sub_model, data_min,
fc_min, part, date, mk.price, fc.price)
res <- res[!is.na('mk.price'),]
cat(nms, '\n')
return(res)
}) %>%
as_tibble
## https://tysonbarrett.com/jekyll/update/2019/10/06/datatable_memory/
## http://brooksandrew.github.io/simpleblog/articles/advanced-data-table/
## https://atrebas.github.io/post/2019-03-03-datatable-dplyr/
saveRDS(mds_ets_MNN_7200_1440,
paste0(.dtr, 'data/fx/USDJPY/mds_ets_MNN_7200_1440.rds'))## Get all files.
fls <- paste0(.dtr, 'data/fx/USDJPY/intraday/') %>%
list.files(., pattern = '^ts_ets_MNN_')
fls_pth <- paste0(.dtr, 'data/fx/USDJPY/intraday/', fls)
mds_ets_MNN_intraday <- ldply(1:length(fls), function(i) {
nms <- fls[i] %>%
str_replace_all('.rds', '') %>%
str_split('_|\\.') %>%
.[[1]]
nms <- nms[-6]
names(nms) <- c('t_series', 'model', 'sub_model',
'data_min', 'fc_min', 'part', 'date')
datset2 <- t(nms) %>%
data.frame %>%
mutate(model =factor(model), sub_model = factor(sub_model),
data_min = as.numeric(data_min),
fc_min = as.numeric(fc_min),
part = as.numeric(part),
date = as_date(date))
datset <- read_rds(fls_pth[i])
res <- tibble(datset2, datset) %>%
dplyr::select(index, t_series, model, sub_model, data_min,
fc_min, part, date, mk.price, fc.price)
res <- res[!is.na('mk.price'),]
cat(nms, '\n')
return(res)
}) %>%
as_tibble
## https://tysonbarrett.com/jekyll/update/2019/10/06/datatable_memory/
## http://brooksandrew.github.io/simpleblog/articles/advanced-data-table/
## https://atrebas.github.io/post/2019-03-03-datatable-dplyr/
saveRDS(mds_ets_MNN_intraday, paste0(.dtr, 'data/fx/USDJPY/mds_ets_MNN_intraday.rds'))if(!exists('mds_ets_MNN_7200_1440')) {
mds_ets_MNN_7200_1440 <- read_rds(
paste0(.dtr, 'data/fx/USDJPY/mds_ets_MNN_7200_1440.rds')) }
if(!exists('mds_ets_MNN_intraday')) {
mds_ets_MNN_intraday <- read_rds(
paste0(.dtr, 'data/fx/USDJPY/mds_ets_MNN_intraday.rds')) }
mds_ets_MNN_intraday <- rbind(mds_ets_MNN_7200_1440, mds_ets_MNN_intraday) %>%
as.data.table
mds_ets_MNN_intraday <- mds_ets_MNN_intraday[order(index, part, data_min, fc_min)]
saveRDS(mds_ets_MNN_intraday, paste0(.dtr, 'data/fx/USDJPY/mds_ets_MNN_intraday.rds'))
rm(mds_ets_MNN_7200_1440)
if(file.exists(paste0(.dtr, 'data/fx/USDJPY/mds_ets_MNN_7200_1440.rds'))) {
file.remove(paste0(.dtr, 'data/fx/USDJPY/mds_ets_MNN_7200_1440.rds')) }4.1.1.2 Long Format
rm(list = ls())
if(!exists('mds_ets_MNN_intraday')) {
mds_ets_MNN_intraday <- read_rds(
paste0(.dtr, 'data/fx/USDJPY/mds_ets_MNN_intraday.rds')) }
mds_ets_MNN_intraday_cmp <- mds_ets_MNN_intraday %>%
#tidyr::unite(model, t_series:part) %>%
tidyr::unite(model, t_series:fc_min) %>%
as.data.table %>%
.[, .(index, model, mk.price, fc.price)] %>% as.data.table
mds_ets_MNN_intraday_cmp <- as.data.table(mds_ets_MNN_intraday_cmp)
## ----------------------------------------
## fromLast = FALSE
mds_ets_MNN_intraday_cmp_fLF <- unique(
mds_ets_MNN_intraday_cmp[, .(index, model, mk.price, fc.price)], fromLast = FALSE)
## save dataset in data.table format
saveRDS(mds_ets_MNN_intraday_cmp_fLF,
paste0(.dtr, 'data/fx/USDJPY/mds_ets_MNN_intraday_cmp_fLF.rds'))
## ----------------------------------------
## fromLast = TRUE
mds_ets_MNN_intraday_cmp_fLT <- unique(
mds_ets_MNN_intraday_cmp[, .(index, model, mk.price, fc.price)], fromLast = TRUE)
## save dataset in data.table format
saveRDS(mds_ets_MNN_intraday_cmp_fLT,
paste0(.dtr, 'data/fx/USDJPY/mds_ets_MNN_intraday_cmp_fLT.rds'))
## ----------------------------------------
## convert to long format for plot chart
## fromLast = FALSE
prc <- unique(mds_ets_MNN_intraday_cmp[, .(index, model, mk.price)], fromLast = FALSE)
prc <- prc[, model := 'actual'][]
setnames(prc, old = 'mk.price', new = 'price')
prc <- unique(prc, fromLast = FALSE)
mds_ets_MNN_intraday_grph <- unique(
mds_ets_MNN_intraday_cmp[, .(index, model, fc.price)], fromLast = FALSE)
setnames(mds_ets_MNN_intraday_grph, old = 'fc.price', new = 'price')
mds_ets_MNN_intraday_grph <- rbind(mds_ets_MNN_intraday_grph, prc)
mds_ets_MNN_intraday_grph_fLF <- data.table(mds_ets_MNN_intraday_grph)[order(index)]
rm(prc)
## save dataset in data.table format
saveRDS(mds_ets_MNN_intraday_grph_fLF,
paste0(.dtr, 'data/fx/USDJPY/mds_ets_MNN_intraday_grph_fLF.rds'))
## ----------------------------------------
## fromLast = TRUE
prc <- unique(mds_ets_MNN_intraday_cmp[, .(index, model, mk.price)], fromLast = TRUE)
prc <- prc[, model := 'actual'][]
setnames(prc, old = 'mk.price', new = 'price')
prc <- unique(prc, fromLast = TRUE)
mds_ets_MNN_intraday_grph <- unique(
mds_ets_MNN_intraday_cmp[, .(index, model, fc.price)], fromLast = TRUE)
setnames(mds_ets_MNN_intraday_grph, old = 'fc.price', new = 'price')
mds_ets_MNN_intraday_grph <- rbind(mds_ets_MNN_intraday_grph, prc)
mds_ets_MNN_intraday_grph_fLT <- data.table(mds_ets_MNN_intraday_grph)[order(index)]
rm(prc)
## save dataset in data.table format
saveRDS(mds_ets_MNN_intraday_grph_fLT,
paste0(.dtr, 'data/fx/USDJPY/mds_ets_MNN_intraday_grph_fLT.rds'))if(!exists('mds_ets_MNN_intraday')) {
mds_ets_MNN_intraday <- read_rds(
paste0(.dtr, 'data/fx/USDJPY/mds_ets_MNN_intraday.rds')) }
## plot sample data
mds_ets_MNN_intraday[c(1:3, (nrow(mds_ets_MNN_intraday)-3):nrow(mds_ets_MNN_intraday)),] %>%
kbl(caption = 'Data Sample (ETS MNN 7200/1440 from "2016-01-04" to "2018-07-06" forecast intraday)', escape = FALSE) %>%
## https://www.w3schools.com/cssref/css_colors.asp
## https://public.tableau.com/en-us/gallery/100-color-palettes?gallery=votd
row_spec(0, background = 'DimGrey', color = 'gold', bold = TRUE) %>%
column_spec(1, background = 'CornflowerBlue') %>%
column_spec(2, background = 'Gray') %>%
column_spec(3, background = 'DarkGrey') %>%
column_spec(4, background = 'Gray') %>%
column_spec(5, background = 'DarkGrey') %>%
column_spec(6, background = '#4897D8') %>%
column_spec(7, background = '#375E97') %>%
column_spec(8, background = 'CornflowerBlue') %>%
column_spec(9, background = 'Gainsboro', color = 'goldenrod') %>%
column_spec(10, background = 'LightGray', color = 'goldenrod') %>%
kable_styling(bootstrap_options = c('striped', 'hover', 'condensed', 'responsive')) %>%
kable_material(full_width = FALSE) %>% ##`full_width = FALSE` will auto adjust every single columns width to fit the table full width.
scroll_box(width = '100%', fixed_thead = TRUE, height = '400px')| index | t_series | model | sub_model | data_min | fc_min | part | date | mk.price | fc.price |
|---|---|---|---|---|---|---|---|---|---|
| 2016-01-04 00:01:00 | ts | ets | MNN | 1440 | 1 | 1 | 2016-01-04 | 120.220 | 120.2080 |
| 2016-01-04 00:01:00 | ts | ets | MNN | 1440 | 6 | 1 | 2016-01-04 | 120.220 | 120.2080 |
| 2016-01-04 00:01:00 | ts | ets | MNN | 1440 | 8 | 1 | 2016-01-04 | 120.220 | 120.2080 |
| 2018-07-07 00:00:00 | ts | ets | MNN | 1440 | 15 | 96 | 2018-07-06 | 110.474 | 110.4192 |
| 2018-07-07 00:00:00 | ts | ets | MNN | 1440 | 10 | 144 | 2018-07-06 | 110.474 | 110.4571 |
| 2018-07-07 00:00:00 | ts | ets | MNN | 1440 | 8 | 180 | 2018-07-06 | 110.474 | 110.4586 |
| 2018-07-07 00:00:00 | ts | ets | MNN | 1440 | 6 | 240 | 2018-07-06 | 110.474 | 110.4796 |
source : 18745921 x 10
There has a small mistake in previous paper Deriv.com - Interday High Frequency Trading Models Comparison Review (Part I) but doesn’t affect since I only want to compare the models and here I use the same MNN model and filter unique(fromLast = FALSE) and unique(fromLast = TRUE).
if(!exists('mds_ets_MNN_intraday_grph_fLF')) {
mds_ets_MNN_intraday_grph_fLF <- readRDS(
paste0(.dtr, 'data/fx/USDJPY/mds_ets_MNN_intraday_grph_fLF.rds')) }
if(!exists('mds_ets_MNN_intraday_grph_fLT')) {
mds_ets_MNN_intraday_grph_fLT <- readRDS(
paste0(.dtr, 'data/fx/USDJPY/mds_ets_MNN_intraday_grph_fLT.rds')) }data.frame(mds_ets_MNN_intraday_grph_fLF)[
c(1:5, (nrow(mds_ets_MNN_intraday_grph_fLF)-5):nrow(mds_ets_MNN_intraday_grph_fLF)),] %>%
kbl(caption = 'Data Sample', escape = FALSE) %>%
row_spec(0, background = 'DimGrey', color = 'yellow') %>%
column_spec(1, background = 'CornflowerBlue') %>%
column_spec(2, background = 'grey') %>%
column_spec(3, background = '#556DAC') %>%
column_spec(4, background = 'LightGray', color = 'goldenrod') %>%
kable_styling(bootstrap_options = c('striped', 'hover', 'condensed', 'responsive')) %>%
kable_material(full_width = FALSE) %>%
scroll_box(width = '100%', fixed_thead = TRUE)| index | model | price | |
|---|---|---|---|
| 1 | 2016-01-04 00:01:00 | ts_ets_MNN_7200_1440 | 120.2080 |
| 2 | 2016-01-04 00:01:00 | ts_ets_MNN_1440_1 | 120.2080 |
| 3 | 2016-01-04 00:01:00 | ts_ets_MNN_1440_10 | 120.2080 |
| 4 | 2016-01-04 00:01:00 | ts_ets_MNN_1440_120 | 120.2080 |
| 5 | 2016-01-04 00:01:00 | ts_ets_MNN_1440_15 | 120.2080 |
| 15913492 | 2018-07-07 00:00:00 | ts_ets_MNN_1440_60 | 110.4234 |
| 15913493 | 2018-07-07 00:00:00 | ts_ets_MNN_1440_720 | 110.6322 |
| 15913494 | 2018-07-07 00:00:00 | ts_ets_MNN_1440_8 | 110.4586 |
| 15913495 | 2018-07-07 00:00:00 | ts_ets_MNN_7200_720 | 110.6322 |
| 15913496 | 2018-07-07 00:00:00 | ts_ets_MNN_7200_720 | 110.4740 |
| 15913497 | 2018-07-07 00:00:00 | actual | 110.4740 |
source : 15913497 x 3
data.frame(mds_ets_MNN_intraday_grph_fLT)[
c(1:6, (nrow(mds_ets_MNN_intraday_grph_fLT)-5):nrow(mds_ets_MNN_intraday_grph_fLT)),] %>%
kbl(caption = 'Data Sample', escape = FALSE) %>%
row_spec(0, background = 'DimGrey', color = 'yellow') %>%
column_spec(1, background = 'CornflowerBlue') %>%
column_spec(2, background = 'grey') %>%
column_spec(3, background = '#556DAC') %>%
column_spec(4, background = 'LightGray', color = 'goldenrod') %>%
kable_styling(bootstrap_options = c('striped', 'hover', 'condensed', 'responsive')) %>%
kable_material(full_width = FALSE) %>%
scroll_box(width = '100%', fixed_thead = TRUE)| index | model | price | |
|---|---|---|---|
| 1 | 2016-01-04 00:01:00 | ts_ets_MNN_7200_1440 | 120.2080 |
| 2 | 2016-01-04 00:01:00 | ts_ets_MNN_1440_1 | 120.2080 |
| 3 | 2016-01-04 00:01:00 | ts_ets_MNN_1440_10 | 120.2080 |
| 4 | 2016-01-04 00:01:00 | ts_ets_MNN_1440_120 | 120.2080 |
| 5 | 2016-01-04 00:01:00 | ts_ets_MNN_1440_15 | 120.2080 |
| 6 | 2016-01-04 00:01:00 | ts_ets_MNN_1440_180 | 120.2080 |
| 15913492 | 2018-07-07 00:00:00 | ts_ets_MNN_1440_60 | 110.4234 |
| 15913493 | 2018-07-07 00:00:00 | ts_ets_MNN_1440_720 | 110.6322 |
| 15913494 | 2018-07-07 00:00:00 | ts_ets_MNN_1440_8 | 110.4586 |
| 15913495 | 2018-07-07 00:00:00 | ts_ets_MNN_7200_720 | 110.6322 |
| 15913496 | 2018-07-07 00:00:00 | ts_ets_MNN_7200_720 | 110.4740 |
| 15913497 | 2018-07-07 00:00:00 | actual | 110.4740 |
source : 15913497 x 3
4.1.2 Line Chart
4.1.2.1 unique(x, fromLast = FALSE) Dataset
## https://www.r-graph-gallery.com/line-chart-several-groups-ggplot2.html
plt <- mds_ets_MNN_intraday_grph_fLF %>%
ggplot(aes(x = index, y = price, group = model, color = model)) +
geom_line() +
labs(title = '1 min Open Price Forecasting',
subtitle = paste('From', range(unique(mds_ets_MNN_intraday_grph_fLF$index))[1L],
'to', range(unique(mds_ets_MNN_intraday_grph_fLF$index))[2L]),
caption = "Data source: fxcm") +
ylab('Exchange Rates USD/JPY') +
theme_economist() +
theme(legend.position = 'right')
#ggplotly(plt)
plt
Due to the high volume of data, here I skip the line chart.
4.1.2.2 unique(x, fromLast = TRUE) Dataset
## https://www.r-graph-gallery.com/line-chart-several-groups-ggplot2.html
plt <- mds_ets_MNN_intraday_grph_fLT %>%
ggplot(aes(x = index, y = price, group = model, color = model)) +
geom_line() +
labs(title = '1 min Open Price Forecasting',
subtitle = paste('From', range(unique(mds_ets_MNN_intraday_grph_fLT$index))[1L],
'to', range(unique(mds_ets_MNN_intraday_grph_fLT$index))[2L]),
caption = "Data source: fxcm") +
ylab('Exchange Rates USD/JPY') +
theme_economist() +
theme(legend.position = 'right')
#ggplotly(plt)
plt
Due to the high volume of data, here I skip the line chart.
4.1.3 Table
4.1.3.1 unique(x, fromLast = FALSE) Dataset
Below table compares the models.
rm(list = ls())
## report for mds_ets_MNN_intraday_cmp_fLF
if(!exists('mds_ets_MNN_intraday_cmp_fLF')) {
mds_ets_MNN_intraday_cmp_fLF <- read_rds(
paste0(.dtr, 'data/fx/USDJPY/mds_ets_MNN_intraday_cmp_fLF.rds')) }
if(!is.data.table('mds_ets_MNN_intraday_cmp_fLF')) {
mds_ets_MNN_intraday_cmp_fLF %<>% as.data.table }
mds_ets_MNN_intraday_cmp_fLF <- unique(
mds_ets_MNN_intraday_cmp_fLF, fromLast = FALSE, by = c('index', 'model')) %>%
na.omit
rp_fLF <- mds_ets_MNN_intraday_cmp_fLF[, {
mk.price = mk.price
fc.price = fc.price
.SD[, .(.N,
mae = MLmetrics::MAE(y_true = mk.price, y_pred = fc.price),
mape = MLmetrics::MAPE(y_true = mk.price, y_pred = fc.price),
rmse = MLmetrics::RMSE(y_true = mk.price, y_pred = fc.price),
smape = Metrics::smape(actual = mk.price, predicted = fc.price),
mse = MLmetrics::MSE(y_true = mk.price, y_pred = fc.price)),
by=.(model)]}][order(model), ]
## report
saveRDS(rp_fLF, paste0(.dtr, 'data/fx/USDJPY/rp_fLF.rds'))rp_fLF <- readRDS(paste0(.dtr, 'data/fx/USDJPY/rp_fLF.rds'))
rp_tbl <- rp_fLF %>%
dplyr::mutate(
mae = ifelse(
rank(mae) <= 3,
cell_spec(
paste0(round(mae, 7), ' (rank: ', sprintf('%1.f', rank(mae)), ')'),
color = 'darkgoldenrod', bold = TRUE),
cell_spec(
paste0(round(mae, 7), ' (rank: ', sprintf('%1.f', rank(mae)), ')'),
color = 'grey', italic = TRUE)),
mape = ifelse(
rank(mape) <= 3,
cell_spec(
paste0(round(mape, 7), ' (rank: ', sprintf('%1.f', rank(mape)), ')'),
color = 'darkgoldenrod', bold = TRUE),
cell_spec(
paste0(round(mape, 7), ' (rank: ', sprintf('%1.f', rank(mape)), ')'),
color = 'grey', italic = TRUE)),
rmse = ifelse(
rank(rmse) <= 3,
cell_spec(
paste0(round(rmse, 7), ' (rank: ', sprintf('%1.f', rank(rmse)), ')'),
color = 'darkgoldenrod', bold = TRUE),
cell_spec(
paste0(round(rmse, 7), ' (rank: ', sprintf('%1.f', rank(rmse)), ')'),
color = 'grey', italic = TRUE)),
smape = ifelse(
rank(smape) <= 3,
cell_spec(
paste0(round(smape, 7), ' (rank: ', sprintf('%1.f', rank(smape)), ')'),
color = 'darkgoldenrod', bold = TRUE),
cell_spec(
paste0(round(smape, 7), ' (rank: ', sprintf('%1.f', rank(smape)), ')'),
color = 'grey', italic = TRUE)),
mse = ifelse(
rank(mse) <= 3,
cell_spec(
paste0(round(mse, 7), ' (rank: ', sprintf('%1.f', rank(mse)), ')'),
color = 'darkgoldenrod', bold = TRUE),
cell_spec(
paste0(round(mse, 7), ' (rank: ', sprintf('%1.f', rank(mse)), ')'),
color = 'grey', italic = TRUE))) %>%
kbl(caption = 'Accurcy Report (Dataset from "2016-01-04" to "2018-07-06" Forecasted Up to Max 1440 mins)', escape = FALSE) %>%
## https://www.w3schools.com/cssref/css_colors.asp
row_spec(0, background = 'DimGrey', color = 'yellow') %>%
#column_spec(1, background = 'CornflowerBlue') %>%
column_spec(1, background = '#556DAC') %>%
column_spec(2, background = 'LightSlateGrey') %>%
column_spec(3, background = 'Gainsboro') %>%
column_spec(4, background = 'LightGray') %>%
column_spec(5, background = 'Gainsboro') %>%
column_spec(6, background = 'LightGray') %>%
column_spec(7, background = 'Gainsboro') %>%
kable_styling(bootstrap_options = c('striped', 'hover', 'condensed', 'responsive')) %>%
kable_material(full_width = FALSE) %>%
##`full_width = FALSE` will auto adjust every single columns width to fit the table full width.
scroll_box(width = '100%', fixed_thead = TRUE, height = '400px')
rp_tbl| model | N | mae | mape | rmse | smape | mse |
|---|---|---|---|---|---|---|
| ts_ets_MNN_1440_1 | 850925 | 0.0116135 (rank: 1) | 0.0001056 (rank: 1) | 0.0196293 (rank: 1) | 0.0001056 (rank: 1) | 0.0003853 (rank: 1) |
| ts_ets_MNN_1440_10 | 936012 | 0.0261097 (rank: 4) | 0.0002374 (rank: 4) | 0.0439667 (rank: 4) | 0.0002374 (rank: 4) | 0.0019331 (rank: 4) |
| ts_ets_MNN_1440_120 | 936008 | 0.0876931 (rank: 9) | 0.0007977 (rank: 9) | 0.142061 (rank: 9) | 0.0007976 (rank: 9) | 0.0201813 (rank: 9) |
| ts_ets_MNN_1440_15 | 936005 | 0.0315207 (rank: 5) | 0.0002866 (rank: 5) | 0.0527373 (rank: 5) | 0.0002866 (rank: 5) | 0.0027812 (rank: 5) |
| ts_ets_MNN_1440_180 | 936004 | 0.1090491 (rank: 10) | 0.0009924 (rank: 10) | 0.1758124 (rank: 10) | 0.0009922 (rank: 10) | 0.03091 (rank: 10) |
| ts_ets_MNN_1440_20 | 936008 | 0.036005 (rank: 6) | 0.0003275 (rank: 6) | 0.0602137 (rank: 6) | 0.0003275 (rank: 6) | 0.0036257 (rank: 6) |
| ts_ets_MNN_1440_240 | 936004 | 0.1239933 (rank: 11) | 0.0011291 (rank: 11) | 0.2000525 (rank: 11) | 0.0011289 (rank: 11) | 0.040021 (rank: 11) |
| ts_ets_MNN_1440_30 | 936008 | 0.0442656 (rank: 7) | 0.0004027 (rank: 7) | 0.0733471 (rank: 7) | 0.0004026 (rank: 7) | 0.0053798 (rank: 7) |
| ts_ets_MNN_1440_360 | 936004 | 0.1484447 (rank: 12) | 0.0013515 (rank: 12) | 0.2372316 (rank: 12) | 0.0013513 (rank: 12) | 0.0562789 (rank: 12) |
| ts_ets_MNN_1440_480 | 936004 | 0.1871978 (rank: 13) | 0.0017046 (rank: 13) | 0.2854502 (rank: 13) | 0.0017041 (rank: 13) | 0.0814818 (rank: 13) |
| ts_ets_MNN_1440_6 | 936012 | 0.0208291 (rank: 2) | 0.0001894 (rank: 2) | 0.0352063 (rank: 2) | 0.0001894 (rank: 2) | 0.0012395 (rank: 2) |
| ts_ets_MNN_1440_60 | 936008 | 0.0623901 (rank: 8) | 0.0005676 (rank: 8) | 0.1026414 (rank: 8) | 0.0005675 (rank: 8) | 0.0105353 (rank: 8) |
| ts_ets_MNN_1440_720 | 936004 | 0.2254901 (rank: 14) | 0.0020559 (rank: 14) | 0.3519291 (rank: 14) | 0.002055 (rank: 14) | 0.1238541 (rank: 14) |
| ts_ets_MNN_1440_8 | 936012 | 0.0236924 (rank: 3) | 0.0002155 (rank: 3) | 0.0405026 (rank: 3) | 0.0002155 (rank: 3) | 0.0016405 (rank: 3) |
| ts_ets_MNN_7200_1440 | 936004 | 0.3237977 (rank: 16) | 0.002952 (rank: 16) | 0.4913689 (rank: 16) | 0.0029499 (rank: 16) | 0.2414434 (rank: 16) |
| ts_ets_MNN_7200_720 | 936005 | 0.2258944 (rank: 15) | 0.0020595 (rank: 15) | 0.3523549 (rank: 15) | 0.0020585 (rank: 15) | 0.124154 (rank: 15) |
source :
from above models we know the βest model.
4.1.3.2 unique(x, fromLast = TRUE) Dataset
Below table compares the models.
rm(list = ls())
## report for mds_ets_MNN_intraday_cmp_fLT
if(!exists('mds_ets_MNN_intraday_cmp_fLT')) {
mds_ets_MNN_intraday_cmp_fLT <- read_rds(
paste0(.dtr, 'data/fx/USDJPY/mds_ets_MNN_intraday_cmp_fLT.rds')) }
if(!is.data.table('mds_ets_MNN_intraday_cmp_fLT')) {
mds_ets_MNN_intraday_cmp_fLT %<>% as.data.table }
mds_ets_MNN_intraday_cmp_fLT <- unique(
mds_ets_MNN_intraday_cmp_fLT, fromLast = TRUE, by = c('index', 'model')) %>%
na.omit
rp_fLT <- mds_ets_MNN_intraday_cmp_fLT[, {
mk.price = mk.price
fc.price = fc.price
.SD[, .(.N,
mae = MLmetrics::MAE(y_true = mk.price, y_pred = fc.price),
mape = MLmetrics::MAPE(y_true = mk.price, y_pred = fc.price),
rmse = MLmetrics::RMSE(y_true = mk.price, y_pred = fc.price),
smape = Metrics::smape(actual = mk.price, predicted = fc.price),
mse = MLmetrics::MSE(y_true = mk.price, y_pred = fc.price)),
by=.(model)]}][order(model), ]
## report
saveRDS(rp_fLT, paste0(.dtr, 'data/fx/USDJPY/rp_fLT.rds'))rp_fLT <- readRDS(paste0(.dtr, 'data/fx/USDJPY/rp_fLT.rds'))
rp_tbl <- rp_fLT %>%
dplyr::mutate(
mae = ifelse(
rank(mae) <= 3,
cell_spec(
paste0(round(mae, 7), ' (rank: ', sprintf('%1.f', rank(mae)), ')'),
color = 'darkgoldenrod', bold = TRUE),
cell_spec(
paste0(round(mae, 7), ' (rank: ', sprintf('%1.f', rank(mae)), ')'),
color = 'grey', italic = TRUE)),
mape = ifelse(
rank(mape) <= 3,
cell_spec(
paste0(round(mape, 7), ' (rank: ', sprintf('%1.f', rank(mape)), ')'),
color = 'darkgoldenrod', bold = TRUE),
cell_spec(
paste0(round(mape, 7), ' (rank: ', sprintf('%1.f', rank(mape)), ')'),
color = 'grey', italic = TRUE)),
rmse = ifelse(
rank(rmse) <= 3,
cell_spec(
paste0(round(rmse, 7), ' (rank: ', sprintf('%1.f', rank(rmse)), ')'),
color = 'darkgoldenrod', bold = TRUE),
cell_spec(
paste0(round(rmse, 7), ' (rank: ', sprintf('%1.f', rank(rmse)), ')'),
color = 'grey', italic = TRUE)),
smape = ifelse(
rank(smape) <= 3,
cell_spec(
paste0(round(smape, 7), ' (rank: ', sprintf('%1.f', rank(smape)), ')'),
color = 'darkgoldenrod', bold = TRUE),
cell_spec(
paste0(round(smape, 7), ' (rank: ', sprintf('%1.f', rank(smape)), ')'),
color = 'grey', italic = TRUE)),
mse = ifelse(
rank(mse) <= 3,
cell_spec(
paste0(round(mse, 7), ' (rank: ', sprintf('%1.f', rank(mse)), ')'),
color = 'darkgoldenrod', bold = TRUE),
cell_spec(
paste0(round(mse, 7), ' (rank: ', sprintf('%1.f', rank(mse)), ')'),
color = 'grey', italic = TRUE))) %>%
kbl(caption = 'Accurcy Report (Forecasted 1440 mins)', escape = FALSE) %>%
## https://www.w3schools.com/cssref/css_colors.asp
row_spec(0, background = 'DimGrey', color = 'yellow') %>%
#column_spec(1, background = 'CornflowerBlue') %>%
column_spec(1, background = '#556DAC') %>%
column_spec(2, background = 'LightSlateGrey') %>%
column_spec(3, background = 'Gainsboro') %>%
column_spec(4, background = 'LightGray') %>%
column_spec(5, background = 'Gainsboro') %>%
column_spec(6, background = 'LightGray') %>%
column_spec(7, background = 'Gainsboro') %>%
kable_styling(bootstrap_options = c('striped', 'hover', 'condensed', 'responsive')) %>%
kable_material(full_width = FALSE) %>%
##`full_width = FALSE` will auto adjust every single columns width to fit the table full width.
scroll_box(width = '100%', fixed_thead = TRUE, height = '400px')
rp_tbl| model | N | mae | mape | rmse | smape | mse |
|---|---|---|---|---|---|---|
| ts_ets_MNN_1440_1 | 853801 | 0.012137 (rank: 1) | 0.0001103 (rank: 1) | 0.023317 (rank: 1) | 0.0001103 (rank: 1) | 0.0005437 (rank: 1) |
| ts_ets_MNN_1440_10 | 939592 | 0.0266157 (rank: 4) | 0.0002419 (rank: 4) | 0.0456574 (rank: 4) | 0.0002419 (rank: 4) | 0.0020846 (rank: 4) |
| ts_ets_MNN_1440_120 | 939596 | 0.0879918 (rank: 9) | 0.0008003 (rank: 9) | 0.1423749 (rank: 9) | 0.0008002 (rank: 9) | 0.0202706 (rank: 9) |
| ts_ets_MNN_1440_15 | 939599 | 0.0320082 (rank: 5) | 0.0002909 (rank: 5) | 0.0541282 (rank: 5) | 0.0002909 (rank: 5) | 0.0029299 (rank: 5) |
| ts_ets_MNN_1440_180 | 939600 | 0.1092829 (rank: 10) | 0.0009943 (rank: 10) | 0.1759537 (rank: 10) | 0.0009942 (rank: 10) | 0.0309597 (rank: 10) |
| ts_ets_MNN_1440_20 | 939596 | 0.0364745 (rank: 6) | 0.0003316 (rank: 6) | 0.0614091 (rank: 6) | 0.0003316 (rank: 6) | 0.0037711 (rank: 6) |
| ts_ets_MNN_1440_240 | 939600 | 0.1241567 (rank: 11) | 0.0011304 (rank: 11) | 0.2000916 (rank: 11) | 0.0011302 (rank: 11) | 0.0400366 (rank: 11) |
| ts_ets_MNN_1440_30 | 939596 | 0.0447015 (rank: 7) | 0.0004065 (rank: 7) | 0.0742782 (rank: 7) | 0.0004065 (rank: 7) | 0.0055172 (rank: 7) |
| ts_ets_MNN_1440_360 | 939600 | 0.1485413 (rank: 12) | 0.0013522 (rank: 12) | 0.2371252 (rank: 12) | 0.001352 (rank: 12) | 0.0562284 (rank: 12) |
| ts_ets_MNN_1440_480 | 939600 | 0.1871554 (rank: 13) | 0.001704 (rank: 13) | 0.2852255 (rank: 13) | 0.0017036 (rank: 13) | 0.0813536 (rank: 13) |
| ts_ets_MNN_1440_6 | 939592 | 0.0213556 (rank: 2) | 0.0001941 (rank: 2) | 0.0373323 (rank: 2) | 0.0001941 (rank: 2) | 0.0013937 (rank: 2) |
| ts_ets_MNN_1440_60 | 939596 | 0.0627605 (rank: 8) | 0.0005708 (rank: 8) | 0.1032182 (rank: 8) | 0.0005707 (rank: 8) | 0.010654 (rank: 8) |
| ts_ets_MNN_1440_720 | 939600 | 0.22533 (rank: 14) | 0.0020542 (rank: 14) | 0.3514996 (rank: 14) | 0.0020533 (rank: 14) | 0.123552 (rank: 14) |
| ts_ets_MNN_1440_8 | 939592 | 0.0242083 (rank: 3) | 0.00022 (rank: 3) | 0.0423498 (rank: 3) | 0.00022 (rank: 3) | 0.0017935 (rank: 3) |
| ts_ets_MNN_7200_1440 | 939600 | 0.3240886 (rank: 16) | 0.0029543 (rank: 16) | 0.4914093 (rank: 16) | 0.0029522 (rank: 16) | 0.2414831 (rank: 16) |
| ts_ets_MNN_7200_720 | 943198 | 0.2273692 (rank: 15) | 0.0020721 (rank: 15) | 0.3538666 (rank: 15) | 0.0020712 (rank: 15) | 0.1252216 (rank: 15) |
source :
from above models we know the βest model.
4.1.3.3 Summary
rm(list = ls())
## report for rp_fLF and
if(!exists('rp_fLF')) {
rp_fLF <- read_rds(
paste0(.dtr, 'data/fx/USDJPY/rp_fLF.rds')) %>%
as.data.table }
## report for rp_fLT and
if(!exists('rp_fLT')) {
rp_fLT <- read_rds(
paste0(.dtr, 'data/fx/USDJPY/rp_fLT.rds')) %>%
as.data.table }
rp_fLF <- rp_fLF[, arrange := 'from_first'][]
rp_fLT <- rp_fLT[, arrange := 'from_last'][]
rp <- rbindlist(list(rp_fLF, rp_fLT))
rp <- rp[, .(model, arrange, N, mae, mape, rmse, smape, mse)]
rp %<>% unique
#sum <- rp[mae == min(mae) | mape == min(mape) | rmse == min(rmse) | smape == min(smape) | mse == min(mse)]
rp %>%
dplyr::mutate(
mae = ifelse(
rank(mae) <= 3,
cell_spec(
paste0(round(mae, 7), ' (rank: ', sprintf('%1.f', rank(mae)), ')'),
color = 'darkgoldenrod', bold = TRUE),
cell_spec(
paste0(round(mae, 7), ' (rank: ', sprintf('%1.f', rank(mae)), ')'),
color = 'grey', italic = TRUE)),
mape = ifelse(
rank(mape) <= 3,
cell_spec(
paste0(round(mape, 7), ' (rank: ', sprintf('%1.f', rank(mape)), ')'),
color = 'darkgoldenrod', bold = TRUE),
cell_spec(
paste0(round(mape, 7), ' (rank: ', sprintf('%1.f', rank(mape)), ')'),
color = 'grey', italic = TRUE)),
rmse = ifelse(
rank(rmse) <= 3,
cell_spec(
paste0(round(rmse, 7), ' (rank: ', sprintf('%1.f', rank(rmse)), ')'),
color = 'darkgoldenrod', bold = TRUE),
cell_spec(
paste0(round(rmse, 7), ' (rank: ', sprintf('%1.f', rank(rmse)), ')'),
color = 'grey', italic = TRUE)),
smape = ifelse(
rank(smape) <= 3,
cell_spec(
paste0(round(smape, 7), ' (rank: ', sprintf('%1.f', rank(smape)), ')'),
color = 'darkgoldenrod', bold = TRUE),
cell_spec(
paste0(round(smape, 7), ' (rank: ', sprintf('%1.f', rank(smape)), ')'),
color = 'grey', italic = TRUE)),
mse = ifelse(
rank(mse) <= 3,
cell_spec(
paste0(round(mse, 7), ' (rank: ', sprintf('%1.f', rank(mse)), ')'),
color = 'darkgoldenrod', bold = TRUE),
cell_spec(
paste0(round(mse, 7), ' (rank: ', sprintf('%1.f', rank(mse)), ')'),
color = 'grey', italic = TRUE))) %>%
kbl(caption = 'Accurcy Report (Initial Stage Comparison Dataset from "2016-01-04" to "2018-07-06")', escape = FALSE) %>%
## https://www.w3schools.com/cssref/css_colors.asp
row_spec(0, background = 'DimGrey', color = 'yellow') %>%
#column_spec(1, background = 'CornflowerBlue') %>%
column_spec(1, background = '#556DAC') %>%
column_spec(2, background = 'grey') %>%
column_spec(3, background = 'LightSlateGrey') %>%
column_spec(4, background = 'Gainsboro') %>%
column_spec(5, background = 'LightGray') %>%
column_spec(6, background = 'Gainsboro') %>%
column_spec(7, background = 'LightGray') %>%
column_spec(8, background = 'Gainsboro') %>%
kable_styling(bootstrap_options = c('striped', 'hover', 'condensed', 'responsive')) %>%
kable_material(full_width = FALSE) %>%
scroll_box(width = '100%', fixed_thead = TRUE)| model | arrange | N | mae | mape | rmse | smape | mse |
|---|---|---|---|---|---|---|---|
| ts_ets_MNN_1440_1 | from_first | 850925 | 0.0116135 (rank: 1) | 0.0001056 (rank: 1) | 0.0196293 (rank: 1) | 0.0001056 (rank: 1) | 0.0003853 (rank: 1) |
| ts_ets_MNN_1440_10 | from_first | 936012 | 0.0261097 (rank: 7) | 0.0002374 (rank: 7) | 0.0439667 (rank: 7) | 0.0002374 (rank: 7) | 0.0019331 (rank: 7) |
| ts_ets_MNN_1440_120 | from_first | 936008 | 0.0876931 (rank: 17) | 0.0007977 (rank: 17) | 0.142061 (rank: 17) | 0.0007976 (rank: 17) | 0.0201813 (rank: 17) |
| ts_ets_MNN_1440_15 | from_first | 936005 | 0.0315207 (rank: 9) | 0.0002866 (rank: 9) | 0.0527373 (rank: 9) | 0.0002866 (rank: 9) | 0.0027812 (rank: 9) |
| ts_ets_MNN_1440_180 | from_first | 936004 | 0.1090491 (rank: 19) | 0.0009924 (rank: 19) | 0.1758124 (rank: 19) | 0.0009922 (rank: 19) | 0.03091 (rank: 19) |
| ts_ets_MNN_1440_20 | from_first | 936008 | 0.036005 (rank: 11) | 0.0003275 (rank: 11) | 0.0602137 (rank: 11) | 0.0003275 (rank: 11) | 0.0036257 (rank: 11) |
| ts_ets_MNN_1440_240 | from_first | 936004 | 0.1239933 (rank: 21) | 0.0011291 (rank: 21) | 0.2000525 (rank: 21) | 0.0011289 (rank: 21) | 0.040021 (rank: 21) |
| ts_ets_MNN_1440_30 | from_first | 936008 | 0.0442656 (rank: 13) | 0.0004027 (rank: 13) | 0.0733471 (rank: 13) | 0.0004026 (rank: 13) | 0.0053798 (rank: 13) |
| ts_ets_MNN_1440_360 | from_first | 936004 | 0.1484447 (rank: 23) | 0.0013515 (rank: 23) | 0.2372316 (rank: 24) | 0.0013513 (rank: 23) | 0.0562789 (rank: 24) |
| ts_ets_MNN_1440_480 | from_first | 936004 | 0.1871978 (rank: 26) | 0.0017046 (rank: 26) | 0.2854502 (rank: 26) | 0.0017041 (rank: 26) | 0.0814818 (rank: 26) |
| ts_ets_MNN_1440_6 | from_first | 936012 | 0.0208291 (rank: 3) | 0.0001894 (rank: 3) | 0.0352063 (rank: 3) | 0.0001894 (rank: 3) | 0.0012395 (rank: 3) |
| ts_ets_MNN_1440_60 | from_first | 936008 | 0.0623901 (rank: 15) | 0.0005676 (rank: 15) | 0.1026414 (rank: 15) | 0.0005675 (rank: 15) | 0.0105353 (rank: 15) |
| ts_ets_MNN_1440_720 | from_first | 936004 | 0.2254901 (rank: 28) | 0.0020559 (rank: 28) | 0.3519291 (rank: 28) | 0.002055 (rank: 28) | 0.1238541 (rank: 28) |
| ts_ets_MNN_1440_8 | from_first | 936012 | 0.0236924 (rank: 5) | 0.0002155 (rank: 5) | 0.0405026 (rank: 5) | 0.0002155 (rank: 5) | 0.0016405 (rank: 5) |
| ts_ets_MNN_7200_1440 | from_first | 936004 | 0.3237977 (rank: 31) | 0.002952 (rank: 31) | 0.4913689 (rank: 31) | 0.0029499 (rank: 31) | 0.2414434 (rank: 31) |
| ts_ets_MNN_7200_720 | from_first | 936005 | 0.2258944 (rank: 29) | 0.0020595 (rank: 29) | 0.3523549 (rank: 29) | 0.0020585 (rank: 29) | 0.124154 (rank: 29) |
| ts_ets_MNN_1440_1 | from_last | 853801 | 0.012137 (rank: 2) | 0.0001103 (rank: 2) | 0.023317 (rank: 2) | 0.0001103 (rank: 2) | 0.0005437 (rank: 2) |
| ts_ets_MNN_1440_10 | from_last | 939592 | 0.0266157 (rank: 8) | 0.0002419 (rank: 8) | 0.0456574 (rank: 8) | 0.0002419 (rank: 8) | 0.0020846 (rank: 8) |
| ts_ets_MNN_1440_120 | from_last | 939596 | 0.0879918 (rank: 18) | 0.0008003 (rank: 18) | 0.1423749 (rank: 18) | 0.0008002 (rank: 18) | 0.0202706 (rank: 18) |
| ts_ets_MNN_1440_15 | from_last | 939599 | 0.0320082 (rank: 10) | 0.0002909 (rank: 10) | 0.0541282 (rank: 10) | 0.0002909 (rank: 10) | 0.0029299 (rank: 10) |
| ts_ets_MNN_1440_180 | from_last | 939600 | 0.1092829 (rank: 20) | 0.0009943 (rank: 20) | 0.1759537 (rank: 20) | 0.0009942 (rank: 20) | 0.0309597 (rank: 20) |
| ts_ets_MNN_1440_20 | from_last | 939596 | 0.0364745 (rank: 12) | 0.0003316 (rank: 12) | 0.0614091 (rank: 12) | 0.0003316 (rank: 12) | 0.0037711 (rank: 12) |
| ts_ets_MNN_1440_240 | from_last | 939600 | 0.1241567 (rank: 22) | 0.0011304 (rank: 22) | 0.2000916 (rank: 22) | 0.0011302 (rank: 22) | 0.0400366 (rank: 22) |
| ts_ets_MNN_1440_30 | from_last | 939596 | 0.0447015 (rank: 14) | 0.0004065 (rank: 14) | 0.0742782 (rank: 14) | 0.0004065 (rank: 14) | 0.0055172 (rank: 14) |
| ts_ets_MNN_1440_360 | from_last | 939600 | 0.1485413 (rank: 24) | 0.0013522 (rank: 24) | 0.2371252 (rank: 23) | 0.001352 (rank: 24) | 0.0562284 (rank: 23) |
| ts_ets_MNN_1440_480 | from_last | 939600 | 0.1871554 (rank: 25) | 0.001704 (rank: 25) | 0.2852255 (rank: 25) | 0.0017036 (rank: 25) | 0.0813536 (rank: 25) |
| ts_ets_MNN_1440_6 | from_last | 939592 | 0.0213556 (rank: 4) | 0.0001941 (rank: 4) | 0.0373323 (rank: 4) | 0.0001941 (rank: 4) | 0.0013937 (rank: 4) |
| ts_ets_MNN_1440_60 | from_last | 939596 | 0.0627605 (rank: 16) | 0.0005708 (rank: 16) | 0.1032182 (rank: 16) | 0.0005707 (rank: 16) | 0.010654 (rank: 16) |
| ts_ets_MNN_1440_720 | from_last | 939600 | 0.22533 (rank: 27) | 0.0020542 (rank: 27) | 0.3514996 (rank: 27) | 0.0020533 (rank: 27) | 0.123552 (rank: 27) |
| ts_ets_MNN_1440_8 | from_last | 939592 | 0.0242083 (rank: 6) | 0.00022 (rank: 6) | 0.0423498 (rank: 6) | 0.00022 (rank: 6) | 0.0017935 (rank: 6) |
| ts_ets_MNN_7200_1440 | from_last | 939600 | 0.3240886 (rank: 32) | 0.0029543 (rank: 32) | 0.4914093 (rank: 32) | 0.0029522 (rank: 32) | 0.2414831 (rank: 32) |
| ts_ets_MNN_7200_720 | from_last | 943198 | 0.2273692 (rank: 30) | 0.0020721 (rank: 30) | 0.3538666 (rank: 30) | 0.0020712 (rank: 30) | 0.1252216 (rank: 30) |
source : 32 x 8
4.2 2nd Stage (1 Year Dataset)
In order to save the time, here I try to compare if the accuracy of the model by using 1 year dataset good enough.
4.2.1 Read Models
if(!exists('mds_ets_MNN_intraday')) {
mds_ets_MNN_intraday <- read_rds(
paste0(.dtr, 'data/fx/USDJPY/mds_ets_MNN_intraday.rds')) }
mds_ets_MNN_intraday_1y <- mds_ets_MNN_intraday[date <= as_date('2016-12-31')]
saveRDS(mds_ets_MNN_intraday_1y, paste0(.dtr, 'data/fx/USDJPY/mds_ets_MNN_intraday_1y.rds'))
mds_ets_MNN_intraday_1y_cmp <- mds_ets_MNN_intraday_1y %>%
#tidyr::unite(model, t_series:part) %>%
tidyr::unite(model, t_series:fc_min) %>%
as.data.table %>%
.[, .(index, model, mk.price, fc.price)] %>% as.data.table
mds_ets_MNN_intraday_1y_cmp <- as.data.table(mds_ets_MNN_intraday_1y_cmp)
## ----------------------------------------
## fromLast = FALSE
mds_ets_MNN_intraday_1y_cmp_fLF <- unique(
mds_ets_MNN_intraday_1y_cmp[, .(index, model, mk.price, fc.price)], fromLast = FALSE)
## save dataset in data.table format
saveRDS(mds_ets_MNN_intraday_1y_cmp_fLF,
paste0(.dtr, 'data/fx/USDJPY/mds_ets_MNN_intraday_1y_cmp_fLF.rds'))
## ----------------------------------------
## fromLast = TRUE
mds_ets_MNN_intraday_1y_cmp_fLT <- unique(
mds_ets_MNN_intraday_1y_cmp[, .(index, model, mk.price, fc.price)], fromLast = TRUE)
## save dataset in data.table format
saveRDS(mds_ets_MNN_intraday_1y_cmp_fLT,
paste0(.dtr, 'data/fx/USDJPY/mds_ets_MNN_intraday_1y_cmp_fLT.rds'))
## ----------------------------------------
## convert to long format for plot chart
## fromLast = FALSE
prc <- unique(mds_ets_MNN_intraday_1y_cmp[, .(index, model, mk.price)], fromLast = FALSE)
prc <- prc[, model := 'actual'][]
setnames(prc, old = 'mk.price', new = 'price')
prc <- unique(prc, fromLast = FALSE)
mds_ets_MNN_intraday_1y_grph <- unique(
mds_ets_MNN_intraday_1y_cmp[, .(index, model, fc.price)], fromLast = FALSE)
setnames(mds_ets_MNN_intraday_1y_grph, old = 'fc.price', new = 'price')
mds_ets_MNN_intraday_1y_grph <- rbind(mds_ets_MNN_intraday_1y_grph, prc)
mds_ets_MNN_intraday_1y_grph_fLF <- data.table(mds_ets_MNN_intraday_1y_grph)[order(index)]
rm(prc)
## save dataset in data.table format
saveRDS(mds_ets_MNN_intraday_1y_grph_fLF,
paste0(.dtr, 'data/fx/USDJPY/mds_ets_MNN_intraday_1y_grph_fLF.rds'))
## ----------------------------------------
## fromLast = TRUE
prc <- unique(mds_ets_MNN_intraday_1y_cmp[, .(index, model, mk.price)], fromLast = TRUE)
prc <- prc[, model := 'actual'][]
setnames(prc, old = 'mk.price', new = 'price')
prc <- unique(prc, fromLast = TRUE)
mds_ets_MNN_intraday_1y_grph <- unique(
mds_ets_MNN_intraday_1y_cmp[, .(index, model, fc.price)], fromLast = TRUE)
setnames(mds_ets_MNN_intraday_1y_grph, old = 'fc.price', new = 'price')
mds_ets_MNN_intraday_1y_grph <- rbind(mds_ets_MNN_intraday_1y_grph, prc)
mds_ets_MNN_intraday_1y_grph_fLT <- data.table(mds_ets_MNN_intraday_1y_grph)[order(index)]
rm(prc)
## save dataset in data.table format
saveRDS(mds_ets_MNN_intraday_1y_grph_fLT,
paste0(.dtr, 'data/fx/USDJPY/mds_ets_MNN_intraday_1y_grph_fLT.rds'))4.2.2 Table
4.2.2.1 unique(x, fromLast = FALSE) Dataset
Below table compares the models.
rm(list = ls())
## report for mds_ets_MNN_intraday_1y_cmp_fLF
if(!exists('mds_ets_MNN_intraday_1y_cmp_fLF')) {
mds_ets_MNN_intraday_1y_cmp_fLF <- read_rds(
paste0(.dtr, 'data/fx/USDJPY/mds_ets_MNN_intraday_1y_cmp_fLF.rds')) }
if(!is.data.table('mds_ets_MNN_intraday_1y_cmp_fLF')) {
mds_ets_MNN_intraday_1y_cmp_fLF %<>% as.data.table }
mds_ets_MNN_intraday_1y_cmp_fLF <- unique(
mds_ets_MNN_intraday_1y_cmp_fLF, fromLast = FALSE, by = c('index', 'model')) %>%
na.omit
rp_1y_fLF <- mds_ets_MNN_intraday_1y_cmp_fLF[, {
mk.price = mk.price
fc.price = fc.price
.SD[, .(.N,
mae = MLmetrics::MAE(y_true = mk.price, y_pred = fc.price),
mape = MLmetrics::MAPE(y_true = mk.price, y_pred = fc.price),
rmse = MLmetrics::RMSE(y_true = mk.price, y_pred = fc.price),
smape = Metrics::smape(actual = mk.price, predicted = fc.price),
mse = MLmetrics::MSE(y_true = mk.price, y_pred = fc.price)),
by=.(model)]}][order(model), ]
## report
saveRDS(rp_1y_fLF, paste0(.dtr, 'data/fx/USDJPY/rp_1y_fLF.rds'))4.2.2.2 unique(x, fromLast = TRUE) Dataset
Below table compares the models.
rm(list = ls())
## report for mds_ets_MNN_intraday_cmp_fLT
if(!exists('mds_ets_MNN_intraday_1y_cmp_fLT')) {
mds_ets_MNN_intraday_1y_cmp_fLT <- read_rds(
paste0(.dtr, 'data/fx/USDJPY/mds_ets_MNN_intraday_1y_cmp_fLT.rds')) }
if(!is.data.table('mds_ets_MNN_intraday_1y_cmp_fLT')) {
mds_ets_MNN_intraday_1y_cmp_fLT %<>% as.data.table }
mds_ets_MNN_intraday_1y_cmp_fLT <- unique(
mds_ets_MNN_intraday_1y_cmp_fLT, fromLast = TRUE, by = c('index', 'model')) %>%
na.omit
rp_1y_fLT <- mds_ets_MNN_intraday_1y_cmp_fLT[, {
mk.price = mk.price
fc.price = fc.price
.SD[, .(.N,
mae = MLmetrics::MAE(y_true = mk.price, y_pred = fc.price),
mape = MLmetrics::MAPE(y_true = mk.price, y_pred = fc.price),
rmse = MLmetrics::RMSE(y_true = mk.price, y_pred = fc.price),
smape = Metrics::smape(actual = mk.price, predicted = fc.price),
mse = MLmetrics::MSE(y_true = mk.price, y_pred = fc.price)),
by=.(model)]}][order(model), ]
## report
saveRDS(rp_1y_fLT, paste0(.dtr, 'data/fx/USDJPY/rp_1y_fLT.rds'))4.2.3 Summary
4.2.3.1 1 Year
rm(list = ls())
## report for rp_fLF and
if(!exists('rp_fLF')) {
rp_fLF <- read_rds(
paste0(.dtr, 'data/fx/USDJPY/rp_fLF.rds')) %>%
as.data.table }
## report for rp_fLT and
if(!exists('rp_fLT')) {
rp_fLT <- read_rds(
paste0(.dtr, 'data/fx/USDJPY/rp_fLT.rds')) %>%
as.data.table }
rp_fLF <- rp_fLF[, arrange := 'from_first'][]
rp_fLT <- rp_fLT[, arrange := 'from_last'][]
rp <- rbindlist(list(rp_fLF, rp_fLT))
rp <- rp[, .(model, arrange, N, mae, mape, rmse, smape, mse)]
rp %<>% unique
## report for rp_1y_fLF and
if(!exists('rp_1y_fLF')) {
rp_1y_fLF <- read_rds(
paste0(.dtr, 'data/fx/USDJPY/rp_1y_fLF.rds')) %>%
as.data.table }
## report for rp_1y_fLT and
if(!exists('rp_1y_fLT')) {
rp_1y_fLT <- read_rds(
paste0(.dtr, 'data/fx/USDJPY/rp_1y_fLT.rds')) %>%
as.data.table }
rp_1y_fLF <- rp_1y_fLF[, arrange := 'from_first'][]
rp_1y_fLT <- rp_1y_fLT[, arrange := 'from_last'][]
rp_1y <- rbindlist(list(rp_1y_fLF, rp_1y_fLT))
rp_1y <- rp_1y[, .(model, arrange, N, mae, mape, rmse, smape, mse)]
rp_1y %<>% unique
#sum <- rp[mae == min(mae) | mape == min(mape) | rmse == min(rmse) | smape == min(smape) | mse == min(mse)]
rp_1y %>%
dplyr::mutate(
mae = ifelse(
rank(mae) <= 3,
cell_spec(
paste0(round(mae, 7), ' (rank: ', sprintf('%1.f', rank(mae)), ')'),
color = 'darkgoldenrod', bold = TRUE),
cell_spec(
paste0(round(mae, 7), ' (rank: ', sprintf('%1.f', rank(mae)), ')'),
color = 'grey', italic = TRUE)),
mape = ifelse(
rank(mape) <= 3,
cell_spec(
paste0(round(mape, 7), ' (rank: ', sprintf('%1.f', rank(mape)), ')'),
color = 'darkgoldenrod', bold = TRUE),
cell_spec(
paste0(round(mape, 7), ' (rank: ', sprintf('%1.f', rank(mape)), ')'),
color = 'grey', italic = TRUE)),
rmse = ifelse(
rank(rmse) <= 3,
cell_spec(
paste0(round(rmse, 7), ' (rank: ', sprintf('%1.f', rank(rmse)), ')'),
color = 'darkgoldenrod', bold = TRUE),
cell_spec(
paste0(round(rmse, 7), ' (rank: ', sprintf('%1.f', rank(rmse)), ')'),
color = 'grey', italic = TRUE)),
smape = ifelse(
rank(smape) <= 3,
cell_spec(
paste0(round(smape, 7), ' (rank: ', sprintf('%1.f', rank(smape)), ')'),
color = 'darkgoldenrod', bold = TRUE),
cell_spec(
paste0(round(smape, 7), ' (rank: ', sprintf('%1.f', rank(smape)), ')'),
color = 'grey', italic = TRUE)),
mse = ifelse(
rank(mse) <= 3,
cell_spec(
paste0(round(mse, 7), ' (rank: ', sprintf('%1.f', rank(mse)), ')'),
color = 'darkgoldenrod', bold = TRUE),
cell_spec(
paste0(round(mse, 7), ' (rank: ', sprintf('%1.f', rank(mse)), ')'),
color = 'grey', italic = TRUE))) %>%
kbl(caption = 'Accurcy Report (1 Year Dataset from "2016-01-04" to "2016-12-31")', escape = FALSE) %>%
## https://www.w3schools.com/cssref/css_colors.asp
row_spec(0, background = 'DimGrey', color = 'yellow') %>%
#column_spec(1, background = 'CornflowerBlue') %>%
column_spec(1, background = '#556DAC') %>%
column_spec(2, background = 'grey') %>%
column_spec(3, background = 'LightSlateGrey') %>%
column_spec(4, background = 'Gainsboro') %>%
column_spec(5, background = 'LightGray') %>%
column_spec(6, background = 'Gainsboro') %>%
column_spec(7, background = 'LightGray') %>%
column_spec(8, background = 'Gainsboro') %>%
kable_styling(bootstrap_options = c('striped', 'hover', 'condensed', 'responsive')) %>%
kable_material(full_width = FALSE) %>%
scroll_box(width = '100%', fixed_thead = TRUE)| model | arrange | N | mae | mape | rmse | smape | mse |
|---|---|---|---|---|---|---|---|
| ts_ets_MNN_1440_1 | from_first | 374398 | 0.0135726 (rank: 2) | 0.0001248 (rank: 2) | 0.0235634 (rank: 2) | 0.0001248 (rank: 2) | 0.0005552 (rank: 2) |
| ts_ets_MNN_1440_10 | from_first | 374400 | 0.0310264 (rank: 8) | 0.0002852 (rank: 8) | 0.0535552 (rank: 8) | 0.0002852 (rank: 8) | 0.0028682 (rank: 8) |
| ts_ets_MNN_1440_120 | from_first | 374400 | 0.1046746 (rank: 18) | 0.0009631 (rank: 18) | 0.1716097 (rank: 18) | 0.000963 (rank: 18) | 0.0294499 (rank: 18) |
| ts_ets_MNN_1440_15 | from_first | 374400 | 0.0373874 (rank: 10) | 0.0003437 (rank: 10) | 0.064235 (rank: 10) | 0.0003437 (rank: 10) | 0.0041261 (rank: 10) |
| ts_ets_MNN_1440_180 | from_first | 374400 | 0.129417 (rank: 20) | 0.0011914 (rank: 20) | 0.2116718 (rank: 20) | 0.0011911 (rank: 20) | 0.044805 (rank: 20) |
| ts_ets_MNN_1440_20 | from_first | 374400 | 0.0427459 (rank: 12) | 0.0003931 (rank: 12) | 0.0730338 (rank: 12) | 0.0003931 (rank: 12) | 0.0053339 (rank: 12) |
| ts_ets_MNN_1440_240 | from_first | 374400 | 0.148679 (rank: 22) | 0.0013704 (rank: 22) | 0.245548 (rank: 22) | 0.00137 (rank: 22) | 0.0602938 (rank: 22) |
| ts_ets_MNN_1440_30 | from_first | 374400 | 0.0524834 (rank: 14) | 0.0004828 (rank: 14) | 0.0890827 (rank: 14) | 0.0004827 (rank: 14) | 0.0079357 (rank: 14) |
| ts_ets_MNN_1440_360 | from_first | 374400 | 0.1784428 (rank: 24) | 0.0016443 (rank: 24) | 0.2897381 (rank: 24) | 0.0016439 (rank: 24) | 0.0839482 (rank: 24) |
| ts_ets_MNN_1440_480 | from_first | 374400 | 0.2263737 (rank: 26) | 0.0020871 (rank: 26) | 0.3521562 (rank: 26) | 0.0020863 (rank: 26) | 0.124014 (rank: 26) |
| ts_ets_MNN_1440_6 | from_first | 374400 | 0.0247197 (rank: 4) | 0.0002272 (rank: 4) | 0.0428744 (rank: 4) | 0.0002272 (rank: 4) | 0.0018382 (rank: 4) |
| ts_ets_MNN_1440_60 | from_first | 374400 | 0.0738649 (rank: 16) | 0.0006797 (rank: 16) | 0.1238921 (rank: 16) | 0.0006796 (rank: 16) | 0.0153493 (rank: 16) |
| ts_ets_MNN_1440_720 | from_first | 374400 | 0.2755691 (rank: 28) | 0.0025449 (rank: 28) | 0.4395358 (rank: 30) | 0.0025432 (rank: 28) | 0.1931917 (rank: 30) |
| ts_ets_MNN_1440_8 | from_first | 374400 | 0.0281814 (rank: 6) | 0.0002591 (rank: 6) | 0.0494359 (rank: 6) | 0.0002591 (rank: 6) | 0.0024439 (rank: 6) |
| ts_ets_MNN_7200_1440 | from_first | 374400 | 0.4072728 (rank: 32) | 0.0037581 (rank: 32) | 0.6228677 (rank: 32) | 0.0037537 (rank: 32) | 0.3879641 (rank: 32) |
| ts_ets_MNN_7200_720 | from_first | 374400 | 0.275585 (rank: 30) | 0.0025451 (rank: 30) | 0.4395006 (rank: 28) | 0.0025434 (rank: 30) | 0.1931608 (rank: 28) |
| ts_ets_MNN_1440_1 | from_last | 374398 | 0.0135726 (rank: 2) | 0.0001248 (rank: 2) | 0.0235634 (rank: 2) | 0.0001248 (rank: 2) | 0.0005552 (rank: 2) |
| ts_ets_MNN_1440_10 | from_last | 374400 | 0.0310264 (rank: 8) | 0.0002852 (rank: 8) | 0.0535552 (rank: 8) | 0.0002852 (rank: 8) | 0.0028682 (rank: 8) |
| ts_ets_MNN_1440_120 | from_last | 374400 | 0.1046746 (rank: 18) | 0.0009631 (rank: 18) | 0.1716097 (rank: 18) | 0.000963 (rank: 18) | 0.0294499 (rank: 18) |
| ts_ets_MNN_1440_15 | from_last | 374400 | 0.0373874 (rank: 10) | 0.0003437 (rank: 10) | 0.064235 (rank: 10) | 0.0003437 (rank: 10) | 0.0041261 (rank: 10) |
| ts_ets_MNN_1440_180 | from_last | 374400 | 0.129417 (rank: 20) | 0.0011914 (rank: 20) | 0.2116718 (rank: 20) | 0.0011911 (rank: 20) | 0.044805 (rank: 20) |
| ts_ets_MNN_1440_20 | from_last | 374400 | 0.0427459 (rank: 12) | 0.0003931 (rank: 12) | 0.0730338 (rank: 12) | 0.0003931 (rank: 12) | 0.0053339 (rank: 12) |
| ts_ets_MNN_1440_240 | from_last | 374400 | 0.148679 (rank: 22) | 0.0013704 (rank: 22) | 0.245548 (rank: 22) | 0.00137 (rank: 22) | 0.0602938 (rank: 22) |
| ts_ets_MNN_1440_30 | from_last | 374400 | 0.0524834 (rank: 14) | 0.0004828 (rank: 14) | 0.0890827 (rank: 14) | 0.0004827 (rank: 14) | 0.0079357 (rank: 14) |
| ts_ets_MNN_1440_360 | from_last | 374400 | 0.1784428 (rank: 24) | 0.0016443 (rank: 24) | 0.2897381 (rank: 24) | 0.0016439 (rank: 24) | 0.0839482 (rank: 24) |
| ts_ets_MNN_1440_480 | from_last | 374400 | 0.2263737 (rank: 26) | 0.0020871 (rank: 26) | 0.3521562 (rank: 26) | 0.0020863 (rank: 26) | 0.124014 (rank: 26) |
| ts_ets_MNN_1440_6 | from_last | 374400 | 0.0247197 (rank: 4) | 0.0002272 (rank: 4) | 0.0428744 (rank: 4) | 0.0002272 (rank: 4) | 0.0018382 (rank: 4) |
| ts_ets_MNN_1440_60 | from_last | 374400 | 0.0738649 (rank: 16) | 0.0006797 (rank: 16) | 0.1238921 (rank: 16) | 0.0006796 (rank: 16) | 0.0153493 (rank: 16) |
| ts_ets_MNN_1440_720 | from_last | 374400 | 0.2755691 (rank: 28) | 0.0025449 (rank: 28) | 0.4395358 (rank: 30) | 0.0025432 (rank: 28) | 0.1931917 (rank: 30) |
| ts_ets_MNN_1440_8 | from_last | 374400 | 0.0281814 (rank: 6) | 0.0002591 (rank: 6) | 0.0494359 (rank: 6) | 0.0002591 (rank: 6) | 0.0024439 (rank: 6) |
| ts_ets_MNN_7200_1440 | from_last | 374400 | 0.4072728 (rank: 32) | 0.0037581 (rank: 32) | 0.6228677 (rank: 32) | 0.0037537 (rank: 32) | 0.3879641 (rank: 32) |
| ts_ets_MNN_7200_720 | from_last | 374400 | 0.275585 (rank: 30) | 0.0025451 (rank: 30) | 0.4395006 (rank: 28) | 0.0025434 (rank: 30) | 0.1931608 (rank: 28) |
source : 32 x 8
4.2.3.2 Comparison between 1 Year & 3.5 Year Dataset
rp_1y_3.5y <- rbind(rp[order(mae, mape, rmse, smape, mse)][1:4], rp_1y[order(mae, mape, rmse, smape, mse)][1:4])
rp_1y_3.5y %>%
dplyr::mutate(
mae = ifelse(
rank(mae) <= 3,
cell_spec(
paste0(round(mae, 7), ' (rank: ', sprintf('%1.f', rank(mae)), ')'),
color = 'darkgoldenrod', bold = TRUE),
cell_spec(
paste0(round(mae, 7), ' (rank: ', sprintf('%1.f', rank(mae)), ')'),
color = 'grey', italic = TRUE)),
mape = ifelse(
rank(mape) <= 3,
cell_spec(
paste0(round(mape, 7), ' (rank: ', sprintf('%1.f', rank(mape)), ')'),
color = 'darkgoldenrod', bold = TRUE),
cell_spec(
paste0(round(mape, 7), ' (rank: ', sprintf('%1.f', rank(mape)), ')'),
color = 'grey', italic = TRUE)),
rmse = ifelse(
rank(rmse) <= 3,
cell_spec(
paste0(round(rmse, 7), ' (rank: ', sprintf('%1.f', rank(rmse)), ')'),
color = 'darkgoldenrod', bold = TRUE),
cell_spec(
paste0(round(rmse, 7), ' (rank: ', sprintf('%1.f', rank(rmse)), ')'),
color = 'grey', italic = TRUE)),
smape = ifelse(
rank(smape) <= 3,
cell_spec(
paste0(round(smape, 7), ' (rank: ', sprintf('%1.f', rank(smape)), ')'),
color = 'darkgoldenrod', bold = TRUE),
cell_spec(
paste0(round(smape, 7), ' (rank: ', sprintf('%1.f', rank(smape)), ')'),
color = 'grey', italic = TRUE)),
mse = ifelse(
rank(mse) <= 3,
cell_spec(
paste0(round(mse, 7), ' (rank: ', sprintf('%1.f', rank(mse)), ')'),
color = 'darkgoldenrod', bold = TRUE),
cell_spec(
paste0(round(mse, 7), ' (rank: ', sprintf('%1.f', rank(mse)), ')'),
color = 'grey', italic = TRUE))) %>%
kbl(caption = 'Accurcy Report (Comparison between 1 Year & 3.5 Year Dataset)', escape = FALSE) %>%
## https://www.w3schools.com/cssref/css_colors.asp
row_spec(0, background = 'DimGrey', color = 'yellow') %>%
#column_spec(1, background = 'CornflowerBlue') %>%
column_spec(1, background = '#556DAC') %>%
column_spec(2, background = 'grey') %>%
column_spec(3, background = 'LightSlateGrey') %>%
column_spec(4, background = 'Gainsboro') %>%
column_spec(5, background = 'LightGray') %>%
column_spec(6, background = 'Gainsboro') %>%
column_spec(7, background = 'LightGray') %>%
column_spec(8, background = 'Gainsboro') %>%
kable_styling(bootstrap_options = c('striped', 'hover', 'condensed', 'responsive')) %>%
kable_material(full_width = FALSE) %>%
scroll_box(width = '100%', fixed_thead = TRUE)| model | arrange | N | mae | mape | rmse | smape | mse |
|---|---|---|---|---|---|---|---|
| ts_ets_MNN_1440_1 | from_first | 850925 | 0.0116135 (rank: 1) | 0.0001056 (rank: 1) | 0.0196293 (rank: 1) | 0.0001056 (rank: 1) | 0.0003853 (rank: 1) |
| ts_ets_MNN_1440_1 | from_last | 853801 | 0.012137 (rank: 2) | 0.0001103 (rank: 2) | 0.023317 (rank: 2) | 0.0001103 (rank: 2) | 0.0005437 (rank: 2) |
| ts_ets_MNN_1440_6 | from_first | 936012 | 0.0208291 (rank: 5) | 0.0001894 (rank: 5) | 0.0352063 (rank: 5) | 0.0001894 (rank: 5) | 0.0012395 (rank: 5) |
| ts_ets_MNN_1440_6 | from_last | 939592 | 0.0213556 (rank: 6) | 0.0001941 (rank: 6) | 0.0373323 (rank: 6) | 0.0001941 (rank: 6) | 0.0013937 (rank: 6) |
| ts_ets_MNN_1440_1 | from_first | 374398 | 0.0135726 (rank: 4) | 0.0001248 (rank: 4) | 0.0235634 (rank: 4) | 0.0001248 (rank: 4) | 0.0005552 (rank: 4) |
| ts_ets_MNN_1440_1 | from_last | 374398 | 0.0135726 (rank: 4) | 0.0001248 (rank: 4) | 0.0235634 (rank: 4) | 0.0001248 (rank: 4) | 0.0005552 (rank: 4) |
| ts_ets_MNN_1440_6 | from_first | 374400 | 0.0247197 (rank: 8) | 0.0002272 (rank: 8) | 0.0428744 (rank: 8) | 0.0002272 (rank: 8) | 0.0018382 (rank: 8) |
| ts_ets_MNN_1440_6 | from_last | 374400 | 0.0247197 (rank: 8) | 0.0002272 (rank: 8) | 0.0428744 (rank: 8) | 0.0002272 (rank: 8) | 0.0018382 (rank: 8) |
From the table above, N is the sample data size, 3.5 years dataset is more accurate than 1 year dataset. However due to save the resource and time:
- we can know the βest model based on
1 yeardataset. - we can know the βest model based on the unique
from_firstdataset.
4.3 3rd Stage (Intraday 1 min)
4.3.1 Read Models
4.3.1.1 Grouped Models
## Get all intraday files.
fls <- list.files(paste0(.dtr, 'data/fx/USDJPY/intraday/'), pattern = '^ts_ets_MNN_1080_1.p_[0-9]{0,}.[0-9]{4}|^ts_ets_MNN_1200_1.p_[0-9]{0,}.[0-9]{4}|^ts_ets_MNN_480_1.p_[0-9]{0,}.[0-9]{4}|^ts_ets_MNN_360_1.p_[0-9]{0,}.[0-9]{4}|^ts_ets_MNN_240_1.p_[0-9]{0,}.[0-9]{4}|^ts_ets_MNN_180_1.p_[0-9]{0,}.[0-9]{4}') %>%
sort
fls_pth <- paste0(.dtr, 'data/fx/USDJPY/intraday/', fls)
mds_ets_MNN_intraday_1min <- ldply(1:length(fls), function(i) {
nms <- fls[i] %>%
str_replace_all('.rds', '') %>%
str_split('_|\\.') %>%
.[[1]]
nms <- nms[-6]
names(nms) <- c('t_series', 'model', 'sub_model',
'data_min', 'fc_min', 'part', 'date')
datset2 <- t(nms) %>%
data.frame %>%
mutate(model =factor(model), sub_model = factor(sub_model),
data_min = as.numeric(data_min),
fc_min = as.numeric(fc_min),
part = as.numeric(part),
date = as_date(date))
datset <- read_rds(fls_pth[i])
res <- tibble(datset2, datset) %>%
dplyr::select(index, t_series, model, sub_model, data_min,
fc_min, part, date, mk.price, fc.price)
res <- res[!is.na('mk.price'),]
cat(i, '/', length(fls), '=', nms, '\n')
return(res)
}) %>%
as_tibble
## https://tysonbarrett.com/jekyll/update/2019/10/06/datatable_memory/
## http://brooksandrew.github.io/simpleblog/articles/advanced-data-table/
## https://atrebas.github.io/post/2019-03-03-datatable-dplyr/
saveRDS(mds_ets_MNN_intraday_1min, paste0(.dtr, 'data/fx/USDJPY/mds_ets_MNN_intraday_1min.rds'))
mds_ets_MNN_intraday_1min_1y <- mds_ets_MNN_intraday_1min[date <= as_date('2016-12-31')]
saveRDS(mds_ets_MNN_intraday_1min_1y, paste0(.dtr, 'data/fx/USDJPY/mds_ets_MNN_intraday_1min_1y.rds'))## Get all intraday files.
fls <- list.files(paste0(.dtr, 'data/fx/USDJPY/intraday/'), pattern = '^ts_ets_MNN_960_1.p_[0-9]{0,}.[0-9]{4}|^ts_ets_MNN_840_1.p_[0-9]{0,}.[0-9]{4}|^ts_ets_MNN_720_1.p_[0-9]{0,}.[0-9]{4}|^ts_ets_MNN_600_1.p_[0-9]{0,}.[0-9]{4}') %>%
sort
#'2016-02-14'
fls_pth <- paste0(.dtr, 'data/fx/USDJPY/intraday/', fls)
mds_ets_MNN_intraday2_1min <- ldply(1:length(fls), function(i) {
nms <- fls[i] %>%
str_replace_all('.rds', '') %>%
str_split('_|\\.') %>%
.[[1]]
nms <- nms[-6]
names(nms) <- c('t_series', 'model', 'sub_model',
'data_min', 'fc_min', 'part', 'date')
datset2 <- t(nms) %>%
data.frame %>%
mutate(model =factor(model), sub_model = factor(sub_model),
data_min = as.numeric(data_min),
fc_min = as.numeric(fc_min),
part = as.numeric(part),
date = as_date(date))
datset <- read_rds(fls_pth[i])
res <- tibble(datset2, datset) %>%
dplyr::select(index, t_series, model, sub_model, data_min,
fc_min, part, date, mk.price, fc.price)
res <- res[!is.na('mk.price'),]
cat(i, '/', length(fls), '=', nms, '\n')
return(res)
}) %>%
as_tibble
## https://tysonbarrett.com/jekyll/update/2019/10/06/datatable_memory/
## http://brooksandrew.github.io/simpleblog/articles/advanced-data-table/
## https://atrebas.github.io/post/2019-03-03-datatable-dplyr/
mds_ets_MNN_intraday2_1min %<>% arrange(index, data_min)
saveRDS(mds_ets_MNN_intraday2_1min, paste0(.dtr, 'data/fx/USDJPY/mds_ets_MNN_intraday2_1min.rds'))
#ts_ets_MNN_960_1.p_277.2016-06-14
#ts_ets_MNN_840_1.p_1316.2016-06-14
#ts_ets_MNN_720_1.p_103.2016-06-13
#ts_ets_MNN_600_1.p_839.2016-05-06
mds_ets_MNN_intraday2_1min_1y <- mds_ets_MNN_intraday2_1min[date <= as_date('2016-12-31')]
saveRDS(mds_ets_MNN_intraday2_1min_1y, paste0(.dtr, 'data/fx/USDJPY/mds_ets_MNN_intraday2_1min_1y.rds'))4.3.1.2 Long Format
rm(list = ls())
if(!exists('mds_ets_MNN_intraday_1min_1y')) {
mds_ets_MNN_intraday_1min_1y <- read_rds(
paste0(.dtr, 'data/fx/USDJPY/mds_ets_MNN_intraday_1min_1y.rds')) }
if(!exists('mds_ets_MNN_intraday2_1min_1y')) {
mds_ets_MNN_intraday2_1min_1y <- read_rds(
paste0(.dtr, 'data/fx/USDJPY/mds_ets_MNN_intraday2_1min_1y.rds')) }
mds_ets_MNN_intraday3_1min_1y <- rbind(mds_ets_MNN_intraday_1min_1y, mds_ets_MNN_intraday2_1min_1y)
saveRDS(mds_ets_MNN_intraday3_1min_1y, paste0(.dtr, 'data/fx/USDJPY/mds_ets_MNN_intraday3_1min_1y.rds'))
mds_ets_MNN_intraday3_1min_1y <- readRDS(paste0(.dtr, 'data/fx/USDJPY/mds_ets_MNN_intraday3_1min_1y.rds'))
mds_ets_MNN_intraday3_1min_1y_cmp <- mds_ets_MNN_intraday3_1min_1y %>%
#tidyr::unite(model, t_series:part) %>%
tidyr::unite(model, t_series:fc_min) %>%
as.data.table %>%
.[, .(index, model, mk.price, fc.price)] %>% as.data.table
mds_ets_MNN_intraday3_1min_1y_cmp <- as.data.table(mds_ets_MNN_intraday3_1min_1y_cmp)
## ----------------------------------------
## fromLast = FALSE
mds_ets_MNN_intraday3_1min_1y_cmp_fLF <- unique(
mds_ets_MNN_intraday3_1min_1y_cmp[, .(index, model, mk.price, fc.price)], fromLast = FALSE)
## save dataset in data.table format
saveRDS(mds_ets_MNN_intraday3_1min_1y_cmp_fLF,
paste0(.dtr, 'data/fx/USDJPY/mds_ets_MNN_intraday3_1min_1y_cmp_fLF.rds'))
## ----------------------------------------
## fromLast = TRUE
mds_ets_MNN_intraday3_1min_1y_cmp_fLT <- unique(
mds_ets_MNN_intraday3_1min_1y_cmp[, .(index, model, mk.price, fc.price)], fromLast = TRUE)
## save dataset in data.table format
saveRDS(mds_ets_MNN_intraday3_1min_1y_cmp_fLT,
paste0(.dtr, 'data/fx/USDJPY/mds_ets_MNN_intraday3_1min_1y_cmp_fLT.rds'))
## ----------------------------------------
## convert to long format for plot chart
## fromLast = FALSE
prc <- unique(mds_ets_MNN_intraday3_1min_1y_cmp[, .(index, model, mk.price)], fromLast = FALSE)
prc <- prc[, model := 'actual'][]
setnames(prc, old = 'mk.price', new = 'price')
prc <- unique(prc, fromLast = FALSE)
mds_ets_MNN_intraday3_1min_1y_grph <- unique(
mds_ets_MNN_intraday3_1min_1y_cmp[, .(index, model, fc.price)], fromLast = FALSE)
setnames(mds_ets_MNN_intraday3_1min_1y_grph, old = 'fc.price', new = 'price')
mds_ets_MNN_intraday3_1min_1y_grph <- rbind(mds_ets_MNN_intraday3_1min_1y_grph, prc)
mds_ets_MNN_intraday3_1min_1y_grph_fLF <- data.table(mds_ets_MNN_intraday3_1min_1y_grph)[order(index)]
rm(prc)
## save dataset in data.table format
saveRDS(mds_ets_MNN_intraday3_1min_1y_grph_fLF,
paste0(.dtr, 'data/fx/USDJPY/mds_ets_MNN_intraday3_1min_1y_grph_fLF.rds'))
## ----------------------------------------
## fromLast = TRUE
prc <- unique(mds_ets_MNN_intraday3_1min_1y_cmp[, .(index, model, mk.price)], fromLast = TRUE)
prc <- prc[, model := 'actual'][]
setnames(prc, old = 'mk.price', new = 'price')
prc <- unique(prc, fromLast = TRUE)
mds_ets_MNN_intraday3_1min_1y_grph <- unique(
mds_ets_MNN_intraday3_1min_1y_cmp[, .(index, model, fc.price)], fromLast = TRUE)
setnames(mds_ets_MNN_intraday3_1min_1y_grph, old = 'fc.price', new = 'price')
mds_ets_MNN_intraday3_1min_1y_grph <- rbind(mds_ets_MNN_intraday3_1min_1y_grph, prc)
mds_ets_MNN_intraday3_1min_1y_grph_fLT <- data.table(mds_ets_MNN_intraday3_1min_1y_grph)[order(index)]
rm(prc)
## save dataset in data.table format
saveRDS(mds_ets_MNN_intraday3_1min_1y_grph_fLT,
paste0(.dtr, 'data/fx/USDJPY/mds_ets_MNN_intraday3_1min_1y_grph_fLT.rds'))if(!exists('mds_ets_MNN_intraday3_1min_1y')) {
mds_ets_MNN_intraday3_1min_1y <- read_rds(
paste0(.dtr, 'data/fx/USDJPY/mds_ets_MNN_intraday3_1min_1y.rds')) }
## plot sample data
mds_ets_MNN_intraday3_1min_1y[c(1:3, (nrow(mds_ets_MNN_intraday3_1min_1y)-3):nrow(mds_ets_MNN_intraday3_1min_1y)),] %>%
kbl(caption = '1 Year Data Sample (ETS MNN intraday forecast 1min)', escape = FALSE) %>%
## https://www.w3schools.com/cssref/css_colors.asp
## https://public.tableau.com/en-us/gallery/100-color-palettes?gallery=votd
row_spec(0, background = 'DimGrey', color = 'gold', bold = TRUE) %>%
column_spec(1, background = 'CornflowerBlue') %>%
column_spec(2, background = 'Gray') %>%
column_spec(3, background = 'DarkGrey') %>%
column_spec(4, background = 'Gray') %>%
column_spec(5, background = 'DarkGrey') %>%
column_spec(6, background = '#4897D8') %>%
column_spec(7, background = '#375E97') %>%
column_spec(8, background = 'CornflowerBlue') %>%
column_spec(9, background = 'Gainsboro', color = 'goldenrod') %>%
column_spec(10, background = 'LightGray', color = 'goldenrod') %>%
kable_styling(bootstrap_options = c('striped', 'hover', 'condensed', 'responsive')) %>%
kable_material(full_width = FALSE) %>% ##`full_width = FALSE` will auto adjust every single columns width to fit the table full width.
scroll_box(width = '100%', fixed_thead = TRUE, height = '400px')| .id | index | t_series | model | sub_model | data_min | fc_min | part | date | mk.price | fc.price |
|---|---|---|---|---|---|---|---|---|---|---|
| 180 | 2016-01-04 00:02:00 | ts | ets | MNN | 180 | 1 | 1 | 2016-01-04 | 120.2200 | 120.2104 |
| 180 | 2016-01-04 00:03:00 | ts | ets | MNN | 180 | 1 | 2 | 2016-01-04 | 120.2200 | 120.2200 |
| 180 | 2016-01-04 00:04:00 | ts | ets | MNN | 180 | 1 | 3 | 2016-01-04 | 120.2185 | 120.2200 |
| 1200 | 2016-12-31 23:57:00 | ts | ets | MNN | 1200 | 1 | 1437 | 2016-12-31 | 117.0270 | 117.0270 |
| 1200 | 2016-12-31 23:58:00 | ts | ets | MNN | 1200 | 1 | 1438 | 2016-12-31 | 117.0270 | 117.0270 |
| 1200 | 2016-12-31 23:59:00 | ts | ets | MNN | 1200 | 1 | 1439 | 2016-12-31 | 117.0270 | 117.0270 |
| 1200 | 2017-01-01 00:00:00 | ts | ets | MNN | 1200 | 1 | 1440 | 2017-01-01 | 117.0270 | 117.0270 |
source : 3742909 x 11
There has a small mistake in previous paper Deriv.com - Interday High Frequency Trading Models Comparison Review (Part I) but doesn’t affect since I only want to compare the models and here I use the same MNN model and filter unique(fromLast = FALSE) and unique(fromLast = TRUE).
if(!exists('mds_ets_MNN_intraday3_1min_1y_grph_fLF')) {
mds_ets_MNN_intraday3_1min_1y_grph_fLF <- readRDS(
paste0(.dtr, 'data/fx/USDJPY/mds_ets_MNN_intraday3_1min_1y_grph_fLF.rds')) }
if(!exists('mds_ets_MNN_intraday3_1min_1y_grph_fLT')) {
mds_ets_MNN_intraday3_1min_1y_grph_fLT <- readRDS(
paste0(.dtr, 'data/fx/USDJPY/mds_ets_MNN_intraday3_1min_1y_grph_fLT.rds')) }data.frame(mds_ets_MNN_intraday3_1min_1y_grph_fLF)[
c(1:5, (nrow(mds_ets_MNN_intraday3_1min_1y_grph_fLF)-5):nrow(mds_ets_MNN_intraday3_1min_1y_grph_fLF)),] %>%
kbl(caption = '1 Year Data Sample', escape = FALSE) %>%
row_spec(0, background = 'DimGrey', color = 'yellow') %>%
column_spec(1, background = 'CornflowerBlue') %>%
column_spec(2, background = 'grey') %>%
column_spec(3, background = '#556DAC') %>%
column_spec(4, background = 'LightGray', color = 'goldenrod') %>%
kable_styling(bootstrap_options = c('striped', 'hover', 'condensed', 'responsive')) %>%
kable_material(full_width = FALSE) %>%
scroll_box(width = '100%', fixed_thead = TRUE)| index | model | price | |
|---|---|---|---|
| 1 | 2016-01-04 00:01:00 | ts_ets_MNN_360_1 | 120.2080 |
| 2 | 2016-01-04 00:01:00 | actual | 120.2200 |
| 3 | 2016-01-04 00:02:00 | ts_ets_MNN_180_1 | 120.2104 |
| 4 | 2016-01-04 00:02:00 | ts_ets_MNN_240_1 | 120.2104 |
| 5 | 2016-01-04 00:02:00 | ts_ets_MNN_360_1 | 120.2104 |
| 4117208 | 2016-12-31 23:59:00 | ts_ets_MNN_1080_1 | 117.0270 |
| 4117209 | 2016-12-31 23:59:00 | ts_ets_MNN_1200_1 | 117.0270 |
| 4117210 | 2016-12-31 23:59:00 | actual | 117.0270 |
| 4117211 | 2017-01-01 00:00:00 | ts_ets_MNN_1080_1 | 117.0270 |
| 4117212 | 2017-01-01 00:00:00 | ts_ets_MNN_1200_1 | 117.0270 |
| 4117213 | 2017-01-01 00:00:00 | actual | 117.0270 |
source : 4117213 x 3
data.frame(mds_ets_MNN_intraday3_1min_1y_grph_fLT)[
c(1:6, (nrow(mds_ets_MNN_intraday3_1min_1y_grph_fLT)-5):nrow(mds_ets_MNN_intraday3_1min_1y_grph_fLT)),] %>%
kbl(caption = '1 Year Data Sample', escape = FALSE) %>%
row_spec(0, background = 'DimGrey', color = 'yellow') %>%
column_spec(1, background = 'CornflowerBlue') %>%
column_spec(2, background = 'grey') %>%
column_spec(3, background = '#556DAC') %>%
column_spec(4, background = 'LightGray', color = 'goldenrod') %>%
kable_styling(bootstrap_options = c('striped', 'hover', 'condensed', 'responsive')) %>%
kable_material(full_width = FALSE) %>%
scroll_box(width = '100%', fixed_thead = TRUE)| index | model | price | |
|---|---|---|---|
| 1 | 2016-01-04 00:01:00 | ts_ets_MNN_360_1 | 120.2080 |
| 2 | 2016-01-04 00:01:00 | actual | 120.2200 |
| 3 | 2016-01-04 00:02:00 | ts_ets_MNN_180_1 | 120.2104 |
| 4 | 2016-01-04 00:02:00 | ts_ets_MNN_240_1 | 120.2104 |
| 5 | 2016-01-04 00:02:00 | ts_ets_MNN_360_1 | 120.2104 |
| 6 | 2016-01-04 00:02:00 | ts_ets_MNN_480_1 | 120.2104 |
| 4117208 | 2016-12-31 23:59:00 | ts_ets_MNN_1080_1 | 117.0270 |
| 4117209 | 2016-12-31 23:59:00 | ts_ets_MNN_1200_1 | 117.0270 |
| 4117210 | 2016-12-31 23:59:00 | actual | 117.0270 |
| 4117211 | 2017-01-01 00:00:00 | ts_ets_MNN_1080_1 | 117.0270 |
| 4117212 | 2017-01-01 00:00:00 | ts_ets_MNN_1200_1 | 117.0270 |
| 4117213 | 2017-01-01 00:00:00 | actual | 117.0270 |
source : 4117213 x 3
4.3.2 Line Chart
4.3.2.1 unique(x, fromLast = FALSE) Dataset
## https://www.r-graph-gallery.com/line-chart-several-groups-ggplot2.html
plt <- mds_ets_MNN_intraday3_1min_1y_grph_fLF %>%
ggplot(aes(x = index, y = price, group = model, color = model)) +
geom_line() +
labs(title = '1 min Open Price Forecasting',
subtitle = paste('From', range(unique(mds_ets_MNN_intraday3_1min_1y_grph_fLF$index))[1L],
'to', range(unique(mds_ets_MNN_intraday3_1min_1y_grph_fLF$index))[2L]),
caption = "Data source: fxcm") +
ylab('Exchange Rates USD/JPY') +
theme_economist() +
theme(legend.position = 'right')
#ggplotly(plt)
plt
Due to the high volume of data, here I skip the line chart.
4.3.2.2 unique(x, fromLast = TRUE) Dataset
## https://www.r-graph-gallery.com/line-chart-several-groups-ggplot2.html
plt <- mds_ets_MNN_intraday3_1min_1y_grph_fLT %>%
ggplot(aes(x = index, y = price, group = model, color = model)) +
geom_line() +
labs(title = '1 min Open Price Forecasting',
subtitle = paste('From', range(unique(mds_ets_MNN_intraday3_1min_1y_grph_fLT$index))[1L],
'to', range(unique(mds_ets_MNN_intraday3_1min_1y_grph_fLT$index))[2L]),
caption = "Data source: fxcm") +
ylab('Exchange Rates USD/JPY') +
theme_economist() +
theme(legend.position = 'right')
#ggplotly(plt)
plt
Due to the high volume of data, here I skip the line chart.
4.3.3 Table
4.3.3.1 unique(x, fromLast = FALSE) Dataset
Below table compares the models.
rm(list = ls())
## report for mds_ets_MNN_intraday3_1min_1y_cmp_fLF
if(!exists('mds_ets_MNN_intraday3_1min_1y_cmp_fLF')) {
mds_ets_MNN_intraday3_1min_1y_cmp_fLF <- read_rds(
paste0(.dtr, 'data/fx/USDJPY/mds_ets_MNN_intraday3_1min_1y_cmp_fLF.rds')) }
if(!is.data.table('mds_ets_MNN_intraday3_1min_1y_cmp_fLF')) {
mds_ets_MNN_intraday3_1min_1y_cmp_fLF %<>% as.data.table }
mds_ets_MNN_intraday3_1min_1y_cmp_fLF <- unique(
mds_ets_MNN_intraday3_1min_1y_cmp_fLF, fromLast = FALSE, by = c('index', 'model')) %>%
na.omit
rp3_1min_1y_fLF <- mds_ets_MNN_intraday3_1min_1y_cmp_fLF[, {
mk.price = mk.price
fc.price = fc.price
.SD[, .(.N,
mae = MLmetrics::MAE(y_true = mk.price, y_pred = fc.price),
mape = MLmetrics::MAPE(y_true = mk.price, y_pred = fc.price),
rmse = MLmetrics::RMSE(y_true = mk.price, y_pred = fc.price),
smape = Metrics::smape(actual = mk.price, predicted = fc.price),
mse = MLmetrics::MSE(y_true = mk.price, y_pred = fc.price)),
by=.(model)]}][order(model), ]
## report
saveRDS(rp3_1min_1y_fLF, paste0(.dtr, 'data/fx/USDJPY/rp3_1min_1y_fLF.rds'))rp3_1min_1y_fLF <- readRDS(paste0(.dtr, 'data/fx/USDJPY/rp3_1min_1y_fLF.rds'))
rp3_1min_1y_tbl <- rp3_1min_1y_fLF %>%
dplyr::mutate(
mae = ifelse(
rank(mae) <= 3,
cell_spec(
paste0(round(mae, 7), ' (rank: ', sprintf('%1.f', rank(mae)), ')'),
color = 'darkgoldenrod', bold = TRUE),
cell_spec(
paste0(round(mae, 7), ' (rank: ', sprintf('%1.f', rank(mae)), ')'),
color = 'grey', italic = TRUE)),
mape = ifelse(
rank(mape) <= 3,
cell_spec(
paste0(round(mape, 7), ' (rank: ', sprintf('%1.f', rank(mape)), ')'),
color = 'darkgoldenrod', bold = TRUE),
cell_spec(
paste0(round(mape, 7), ' (rank: ', sprintf('%1.f', rank(mape)), ')'),
color = 'grey', italic = TRUE)),
rmse = ifelse(
rank(rmse) <= 3,
cell_spec(
paste0(round(rmse, 7), ' (rank: ', sprintf('%1.f', rank(rmse)), ')'),
color = 'darkgoldenrod', bold = TRUE),
cell_spec(
paste0(round(rmse, 7), ' (rank: ', sprintf('%1.f', rank(rmse)), ')'),
color = 'grey', italic = TRUE)),
smape = ifelse(
rank(smape) <= 3,
cell_spec(
paste0(round(smape, 7), ' (rank: ', sprintf('%1.f', rank(smape)), ')'),
color = 'darkgoldenrod', bold = TRUE),
cell_spec(
paste0(round(smape, 7), ' (rank: ', sprintf('%1.f', rank(smape)), ')'),
color = 'grey', italic = TRUE)),
mse = ifelse(
rank(mse) <= 3,
cell_spec(
paste0(round(mse, 7), ' (rank: ', sprintf('%1.f', rank(mse)), ')'),
color = 'darkgoldenrod', bold = TRUE),
cell_spec(
paste0(round(mse, 7), ' (rank: ', sprintf('%1.f', rank(mse)), ')'),
color = 'grey', italic = TRUE))) %>%
kbl(caption = 'Accurcy Report (1 Year Dataset Intraday Forecasted 1 min)', escape = FALSE) %>%
## https://www.w3schools.com/cssref/css_colors.asp
row_spec(0, background = 'DimGrey', color = 'yellow') %>%
#column_spec(1, background = 'CornflowerBlue') %>%
column_spec(1, background = '#556DAC') %>%
column_spec(2, background = 'LightSlateGrey') %>%
column_spec(3, background = 'Gainsboro') %>%
column_spec(4, background = 'LightGray') %>%
column_spec(5, background = 'Gainsboro') %>%
column_spec(6, background = 'LightGray') %>%
column_spec(7, background = 'Gainsboro') %>%
kable_styling(bootstrap_options = c('striped', 'hover', 'condensed', 'responsive')) %>%
kable_material(full_width = FALSE) %>%
##`full_width = FALSE` will auto adjust every single columns width to fit the table full width.
scroll_box(width = '100%', fixed_thead = TRUE, height = '400px')
rp3_1min_1y_tbl| model | N | mae | mape | rmse | smape | mse |
|---|---|---|---|---|---|---|
| ts_ets_MNN_1080_1 | 374296 | 0.0135525 (rank: 2) | 0.0001246 (rank: 2) | 0.0233866 (rank: 2) | 0.0001246 (rank: 2) | 0.0005469 (rank: 2) |
| ts_ets_MNN_1200_1 | 374296 | 0.0135521 (rank: 1) | 0.0001246 (rank: 1) | 0.0233844 (rank: 1) | 0.0001246 (rank: 1) | 0.0005468 (rank: 1) |
| ts_ets_MNN_180_1 | 374295 | 0.0135795 (rank: 10) | 0.0001248 (rank: 10) | 0.0234633 (rank: 10) | 0.0001248 (rank: 10) | 0.0005505 (rank: 10) |
| ts_ets_MNN_240_1 | 374295 | 0.0135735 (rank: 9) | 0.0001248 (rank: 9) | 0.0234467 (rank: 9) | 0.0001248 (rank: 9) | 0.0005497 (rank: 9) |
| ts_ets_MNN_360_1 | 374267 | 0.0135653 (rank: 8) | 0.0001247 (rank: 8) | 0.0234235 (rank: 8) | 0.0001247 (rank: 8) | 0.0005487 (rank: 8) |
| ts_ets_MNN_480_1 | 374275 | 0.0135592 (rank: 7) | 0.0001247 (rank: 7) | 0.0234082 (rank: 7) | 0.0001247 (rank: 7) | 0.0005479 (rank: 7) |
| ts_ets_MNN_600_1 | 374296 | 0.0135569 (rank: 6) | 0.0001246 (rank: 6) | 0.0234 (rank: 6) | 0.0001246 (rank: 6) | 0.0005476 (rank: 6) |
| ts_ets_MNN_720_1 | 374297 | 0.013556 (rank: 5) | 0.0001246 (rank: 5) | 0.0233952 (rank: 5) | 0.0001246 (rank: 5) | 0.0005473 (rank: 5) |
| ts_ets_MNN_840_1 | 374297 | 0.0135551 (rank: 4) | 0.0001246 (rank: 4) | 0.0233913 (rank: 4) | 0.0001246 (rank: 4) | 0.0005472 (rank: 4) |
| ts_ets_MNN_960_1 | 374295 | 0.0135531 (rank: 3) | 0.0001246 (rank: 3) | 0.0233885 (rank: 3) | 0.0001246 (rank: 3) | 0.000547 (rank: 3) |
source :
from above models we know the βest model.
4.3.3.2 unique(x, fromLast = TRUE) Dataset
Below table compares the models.
rm(list = ls())
## report for mds_ets_MNN_intraday_1min_1y_cmp_fLT
if(!exists('mds_ets_MNN_intraday3_1min_1y_cmp_fLT')) {
mds_ets_MNN_intraday3_1min_1y_cmp_fLT <- read_rds(
paste0(.dtr, 'data/fx/USDJPY/mds_ets_MNN_intraday3_1min_1y_cmp_fLT.rds')) }
if(!is.data.table('mds_ets_MNN_intraday3_1min_1y_cmp_fLT')) {
mds_ets_MNN_intraday3_1min_1y_cmp_fLT %<>% as.data.table }
mds_ets_MNN_intraday3_1min_1y_cmp_fLT <- unique(
mds_ets_MNN_intraday3_1min_1y_cmp_fLT, fromLast = TRUE, by = c('index', 'model')) %>%
na.omit
rp3_1min_1y_fLT <- mds_ets_MNN_intraday3_1min_1y_cmp_fLT[, {
mk.price = mk.price
fc.price = fc.price
.SD[, .(.N,
mae = MLmetrics::MAE(y_true = mk.price, y_pred = fc.price),
mape = MLmetrics::MAPE(y_true = mk.price, y_pred = fc.price),
rmse = MLmetrics::RMSE(y_true = mk.price, y_pred = fc.price),
smape = Metrics::smape(actual = mk.price, predicted = fc.price),
mse = MLmetrics::MSE(y_true = mk.price, y_pred = fc.price)),
by=.(model)]}][order(model), ]
## report
saveRDS(rp3_1min_1y_fLT, paste0(.dtr, 'data/fx/USDJPY/rp3_1min_1y_fLT.rds'))rp3_1min_1y_fLT <- readRDS(paste0(.dtr, 'data/fx/USDJPY/rp3_1min_1y_fLT.rds'))
rp3_1min_1y_tbl <- rp3_1min_1y_fLT %>%
dplyr::mutate(
mae = ifelse(
rank(mae) <= 3,
cell_spec(
paste0(round(mae, 7), ' (rank: ', sprintf('%1.f', rank(mae)), ')'),
color = 'darkgoldenrod', bold = TRUE),
cell_spec(
paste0(round(mae, 7), ' (rank: ', sprintf('%1.f', rank(mae)), ')'),
color = 'grey', italic = TRUE)),
mape = ifelse(
rank(mape) <= 3,
cell_spec(
paste0(round(mape, 7), ' (rank: ', sprintf('%1.f', rank(mape)), ')'),
color = 'darkgoldenrod', bold = TRUE),
cell_spec(
paste0(round(mape, 7), ' (rank: ', sprintf('%1.f', rank(mape)), ')'),
color = 'grey', italic = TRUE)),
rmse = ifelse(
rank(rmse) <= 3,
cell_spec(
paste0(round(rmse, 7), ' (rank: ', sprintf('%1.f', rank(rmse)), ')'),
color = 'darkgoldenrod', bold = TRUE),
cell_spec(
paste0(round(rmse, 7), ' (rank: ', sprintf('%1.f', rank(rmse)), ')'),
color = 'grey', italic = TRUE)),
smape = ifelse(
rank(smape) <= 3,
cell_spec(
paste0(round(smape, 7), ' (rank: ', sprintf('%1.f', rank(smape)), ')'),
color = 'darkgoldenrod', bold = TRUE),
cell_spec(
paste0(round(smape, 7), ' (rank: ', sprintf('%1.f', rank(smape)), ')'),
color = 'grey', italic = TRUE)),
mse = ifelse(
rank(mse) <= 3,
cell_spec(
paste0(round(mse, 7), ' (rank: ', sprintf('%1.f', rank(mse)), ')'),
color = 'darkgoldenrod', bold = TRUE),
cell_spec(
paste0(round(mse, 7), ' (rank: ', sprintf('%1.f', rank(mse)), ')'),
color = 'grey', italic = TRUE))) %>%
kbl(caption = 'Accurcy Report (1 Year Dataset Intraday Forecasted 1 mins)', escape = FALSE) %>%
## https://www.w3schools.com/cssref/css_colors.asp
row_spec(0, background = 'DimGrey', color = 'yellow') %>%
#column_spec(1, background = 'CornflowerBlue') %>%
column_spec(1, background = '#556DAC') %>%
column_spec(2, background = 'LightSlateGrey') %>%
column_spec(3, background = 'Gainsboro') %>%
column_spec(4, background = 'LightGray') %>%
column_spec(5, background = 'Gainsboro') %>%
column_spec(6, background = 'LightGray') %>%
column_spec(7, background = 'Gainsboro') %>%
kable_styling(bootstrap_options = c('striped', 'hover', 'condensed', 'responsive')) %>%
kable_material(full_width = FALSE) %>%
##`full_width = FALSE` will auto adjust every single columns width to fit the table full width.
scroll_box(width = '100%', fixed_thead = TRUE, height = '400px')
rp3_1min_1y_tbl| model | N | mae | mape | rmse | smape | mse |
|---|---|---|---|---|---|---|
| ts_ets_MNN_1080_1 | 374296 | 0.0135525 (rank: 2) | 0.0001246 (rank: 2) | 0.0233866 (rank: 2) | 0.0001246 (rank: 2) | 0.0005469 (rank: 2) |
| ts_ets_MNN_1200_1 | 374296 | 0.0135521 (rank: 1) | 0.0001246 (rank: 1) | 0.0233844 (rank: 1) | 0.0001246 (rank: 1) | 0.0005468 (rank: 1) |
| ts_ets_MNN_180_1 | 374295 | 0.0135795 (rank: 10) | 0.0001248 (rank: 10) | 0.0234633 (rank: 10) | 0.0001248 (rank: 10) | 0.0005505 (rank: 10) |
| ts_ets_MNN_240_1 | 374295 | 0.0135735 (rank: 9) | 0.0001248 (rank: 9) | 0.0234467 (rank: 9) | 0.0001248 (rank: 9) | 0.0005497 (rank: 9) |
| ts_ets_MNN_360_1 | 374267 | 0.0135653 (rank: 8) | 0.0001247 (rank: 8) | 0.0234235 (rank: 8) | 0.0001247 (rank: 8) | 0.0005487 (rank: 8) |
| ts_ets_MNN_480_1 | 374275 | 0.0135592 (rank: 7) | 0.0001247 (rank: 7) | 0.0234082 (rank: 7) | 0.0001247 (rank: 7) | 0.0005479 (rank: 7) |
| ts_ets_MNN_600_1 | 374296 | 0.0135569 (rank: 6) | 0.0001246 (rank: 6) | 0.0234 (rank: 6) | 0.0001246 (rank: 6) | 0.0005476 (rank: 6) |
| ts_ets_MNN_720_1 | 374297 | 0.013556 (rank: 5) | 0.0001246 (rank: 5) | 0.0233952 (rank: 5) | 0.0001246 (rank: 5) | 0.0005473 (rank: 5) |
| ts_ets_MNN_840_1 | 374297 | 0.0135551 (rank: 4) | 0.0001246 (rank: 4) | 0.0233913 (rank: 4) | 0.0001246 (rank: 4) | 0.0005472 (rank: 4) |
| ts_ets_MNN_960_1 | 374295 | 0.0135531 (rank: 3) | 0.0001246 (rank: 3) | 0.0233885 (rank: 3) | 0.0001246 (rank: 3) | 0.000547 (rank: 3) |
source :
from above models we know the βest model.
4.3.3.3 Summary
rm(list = ls())
## report for rp_1y_fLF and
if(!exists('rp_1y_fLF')) {
rp_1y_fLF <- read_rds(
paste0(.dtr, 'data/fx/USDJPY/rp_1y_fLF.rds')) %>%
as.data.table }
## report for rp_1y_fLT and
if(!exists('rp_1y_fLT')) {
rp_1y_fLT <- read_rds(
paste0(.dtr, 'data/fx/USDJPY/rp_1y_fLT.rds')) %>%
as.data.table }
## report for rp_1min_1y_fLF and
if(!exists('rp3_1min_1y_fLF')) {
rp3_1min_1y_fLF <- read_rds(
paste0(.dtr, 'data/fx/USDJPY/rp3_1min_1y_fLF.rds')) %>%
as.data.table }
## report for rp_1min_1y_fLT and
if(!exists('rp3_1min_1y_fLT')) {
rp3_1min_1y_fLT <- read_rds(
paste0(.dtr, 'data/fx/USDJPY/rp3_1min_1y_fLT.rds')) %>%
as.data.table }
rp_1y_fLF <- rp_1y_fLF[, arrange := 'from_first'][]
rp_1y_fLT <- rp_1y_fLT[, arrange := 'from_last'][]
rp3_1min_1y_fLF <- rp3_1min_1y_fLF[, arrange := 'from_first'][]
rp3_1min_1y_fLT <- rp3_1min_1y_fLT[, arrange := 'from_last'][]
rp_1min_1y <- rbindlist(list(rp_1y_fLF, rp_1y_fLT, rp3_1min_1y_fLF, rp3_1min_1y_fLT))
rp_1min_1y <- rp_1min_1y[, .(model, arrange, N, mae, mape, rmse, smape, mse)]
rp_1min_1y %<>% unique
#sum <- rp_1min_1y[mae == min(mae) | mape == min(mape) | rmse == min(rmse) | smape == min(smape) | mse == min(mse)]
rp_1min_1y %>%
dplyr::mutate(
mae = ifelse(
rank(mae) <= 3,
cell_spec(
paste0(round(mae, 7), ' (rank: ', sprintf('%1.f', rank(mae)), ')'),
color = 'darkgoldenrod', bold = TRUE),
cell_spec(
paste0(round(mae, 7), ' (rank: ', sprintf('%1.f', rank(mae)), ')'),
color = 'grey', italic = TRUE)),
mape = ifelse(
rank(mape) <= 3,
cell_spec(
paste0(round(mape, 7), ' (rank: ', sprintf('%1.f', rank(mape)), ')'),
color = 'darkgoldenrod', bold = TRUE),
cell_spec(
paste0(round(mape, 7), ' (rank: ', sprintf('%1.f', rank(mape)), ')'),
color = 'grey', italic = TRUE)),
rmse = ifelse(
rank(rmse) <= 3,
cell_spec(
paste0(round(rmse, 7), ' (rank: ', sprintf('%1.f', rank(rmse)), ')'),
color = 'darkgoldenrod', bold = TRUE),
cell_spec(
paste0(round(rmse, 7), ' (rank: ', sprintf('%1.f', rank(rmse)), ')'),
color = 'grey', italic = TRUE)),
smape = ifelse(
rank(smape) <= 3,
cell_spec(
paste0(round(smape, 7), ' (rank: ', sprintf('%1.f', rank(smape)), ')'),
color = 'darkgoldenrod', bold = TRUE),
cell_spec(
paste0(round(smape, 7), ' (rank: ', sprintf('%1.f', rank(smape)), ')'),
color = 'grey', italic = TRUE)),
mse = ifelse(
rank(mse) <= 3,
cell_spec(
paste0(round(mse, 7), ' (rank: ', sprintf('%1.f', rank(mse)), ')'),
color = 'darkgoldenrod', bold = TRUE),
cell_spec(
paste0(round(mse, 7), ' (rank: ', sprintf('%1.f', rank(mse)), ')'),
color = 'grey', italic = TRUE))) %>%
kbl(caption = 'Accurcy Report (1 Year Dataset Initial Stage Comparison)', escape = FALSE) %>%
## https://www.w3schools.com/cssref/css_colors.asp
row_spec(0, background = 'DimGrey', color = 'yellow') %>%
#column_spec(1, background = 'CornflowerBlue') %>%
column_spec(1, background = '#556DAC') %>%
column_spec(2, background = 'grey') %>%
column_spec(3, background = 'LightSlateGrey') %>%
column_spec(4, background = 'Gainsboro') %>%
column_spec(5, background = 'LightGray') %>%
column_spec(6, background = 'Gainsboro') %>%
column_spec(7, background = 'LightGray') %>%
column_spec(8, background = 'Gainsboro') %>%
kable_styling(bootstrap_options = c('striped', 'hover', 'condensed', 'responsive')) %>%
kable_material(full_width = FALSE) %>%
scroll_box(width = '100%', fixed_thead = TRUE)| model | arrange | N | mae | mape | rmse | smape | mse |
|---|---|---|---|---|---|---|---|
| ts_ets_MNN_1440_1 | from_first | 374398 | 0.0135726 (rank: 18) | 0.0001248 (rank: 18) | 0.0235634 (rank: 22) | 0.0001248 (rank: 18) | 0.0005552 (rank: 22) |
| ts_ets_MNN_1440_10 | from_first | 374400 | 0.0310264 (rank: 28) | 0.0002852 (rank: 28) | 0.0535552 (rank: 28) | 0.0002852 (rank: 28) | 0.0028682 (rank: 28) |
| ts_ets_MNN_1440_120 | from_first | 374400 | 0.1046746 (rank: 38) | 0.0009631 (rank: 38) | 0.1716097 (rank: 38) | 0.000963 (rank: 38) | 0.0294499 (rank: 38) |
| ts_ets_MNN_1440_15 | from_first | 374400 | 0.0373874 (rank: 30) | 0.0003437 (rank: 30) | 0.064235 (rank: 30) | 0.0003437 (rank: 30) | 0.0041261 (rank: 30) |
| ts_ets_MNN_1440_180 | from_first | 374400 | 0.129417 (rank: 40) | 0.0011914 (rank: 40) | 0.2116718 (rank: 40) | 0.0011911 (rank: 40) | 0.044805 (rank: 40) |
| ts_ets_MNN_1440_20 | from_first | 374400 | 0.0427459 (rank: 32) | 0.0003931 (rank: 32) | 0.0730338 (rank: 32) | 0.0003931 (rank: 32) | 0.0053339 (rank: 32) |
| ts_ets_MNN_1440_240 | from_first | 374400 | 0.148679 (rank: 42) | 0.0013704 (rank: 42) | 0.245548 (rank: 42) | 0.00137 (rank: 42) | 0.0602938 (rank: 42) |
| ts_ets_MNN_1440_30 | from_first | 374400 | 0.0524834 (rank: 34) | 0.0004828 (rank: 34) | 0.0890827 (rank: 34) | 0.0004827 (rank: 34) | 0.0079357 (rank: 34) |
| ts_ets_MNN_1440_360 | from_first | 374400 | 0.1784428 (rank: 44) | 0.0016443 (rank: 44) | 0.2897381 (rank: 44) | 0.0016439 (rank: 44) | 0.0839482 (rank: 44) |
| ts_ets_MNN_1440_480 | from_first | 374400 | 0.2263737 (rank: 46) | 0.0020871 (rank: 46) | 0.3521562 (rank: 46) | 0.0020863 (rank: 46) | 0.124014 (rank: 46) |
| ts_ets_MNN_1440_6 | from_first | 374400 | 0.0247197 (rank: 24) | 0.0002272 (rank: 24) | 0.0428744 (rank: 24) | 0.0002272 (rank: 24) | 0.0018382 (rank: 24) |
| ts_ets_MNN_1440_60 | from_first | 374400 | 0.0738649 (rank: 36) | 0.0006797 (rank: 36) | 0.1238921 (rank: 36) | 0.0006796 (rank: 36) | 0.0153493 (rank: 36) |
| ts_ets_MNN_1440_720 | from_first | 374400 | 0.2755691 (rank: 48) | 0.0025449 (rank: 48) | 0.4395358 (rank: 50) | 0.0025432 (rank: 48) | 0.1931917 (rank: 50) |
| ts_ets_MNN_1440_8 | from_first | 374400 | 0.0281814 (rank: 26) | 0.0002591 (rank: 26) | 0.0494359 (rank: 26) | 0.0002591 (rank: 26) | 0.0024439 (rank: 26) |
| ts_ets_MNN_7200_1440 | from_first | 374400 | 0.4072728 (rank: 52) | 0.0037581 (rank: 52) | 0.6228677 (rank: 52) | 0.0037537 (rank: 52) | 0.3879641 (rank: 52) |
| ts_ets_MNN_7200_720 | from_first | 374400 | 0.275585 (rank: 50) | 0.0025451 (rank: 50) | 0.4395006 (rank: 48) | 0.0025434 (rank: 50) | 0.1931608 (rank: 48) |
| ts_ets_MNN_1440_1 | from_last | 374398 | 0.0135726 (rank: 18) | 0.0001248 (rank: 18) | 0.0235634 (rank: 22) | 0.0001248 (rank: 18) | 0.0005552 (rank: 22) |
| ts_ets_MNN_1440_10 | from_last | 374400 | 0.0310264 (rank: 28) | 0.0002852 (rank: 28) | 0.0535552 (rank: 28) | 0.0002852 (rank: 28) | 0.0028682 (rank: 28) |
| ts_ets_MNN_1440_120 | from_last | 374400 | 0.1046746 (rank: 38) | 0.0009631 (rank: 38) | 0.1716097 (rank: 38) | 0.000963 (rank: 38) | 0.0294499 (rank: 38) |
| ts_ets_MNN_1440_15 | from_last | 374400 | 0.0373874 (rank: 30) | 0.0003437 (rank: 30) | 0.064235 (rank: 30) | 0.0003437 (rank: 30) | 0.0041261 (rank: 30) |
| ts_ets_MNN_1440_180 | from_last | 374400 | 0.129417 (rank: 40) | 0.0011914 (rank: 40) | 0.2116718 (rank: 40) | 0.0011911 (rank: 40) | 0.044805 (rank: 40) |
| ts_ets_MNN_1440_20 | from_last | 374400 | 0.0427459 (rank: 32) | 0.0003931 (rank: 32) | 0.0730338 (rank: 32) | 0.0003931 (rank: 32) | 0.0053339 (rank: 32) |
| ts_ets_MNN_1440_240 | from_last | 374400 | 0.148679 (rank: 42) | 0.0013704 (rank: 42) | 0.245548 (rank: 42) | 0.00137 (rank: 42) | 0.0602938 (rank: 42) |
| ts_ets_MNN_1440_30 | from_last | 374400 | 0.0524834 (rank: 34) | 0.0004828 (rank: 34) | 0.0890827 (rank: 34) | 0.0004827 (rank: 34) | 0.0079357 (rank: 34) |
| ts_ets_MNN_1440_360 | from_last | 374400 | 0.1784428 (rank: 44) | 0.0016443 (rank: 44) | 0.2897381 (rank: 44) | 0.0016439 (rank: 44) | 0.0839482 (rank: 44) |
| ts_ets_MNN_1440_480 | from_last | 374400 | 0.2263737 (rank: 46) | 0.0020871 (rank: 46) | 0.3521562 (rank: 46) | 0.0020863 (rank: 46) | 0.124014 (rank: 46) |
| ts_ets_MNN_1440_6 | from_last | 374400 | 0.0247197 (rank: 24) | 0.0002272 (rank: 24) | 0.0428744 (rank: 24) | 0.0002272 (rank: 24) | 0.0018382 (rank: 24) |
| ts_ets_MNN_1440_60 | from_last | 374400 | 0.0738649 (rank: 36) | 0.0006797 (rank: 36) | 0.1238921 (rank: 36) | 0.0006796 (rank: 36) | 0.0153493 (rank: 36) |
| ts_ets_MNN_1440_720 | from_last | 374400 | 0.2755691 (rank: 48) | 0.0025449 (rank: 48) | 0.4395358 (rank: 50) | 0.0025432 (rank: 48) | 0.1931917 (rank: 50) |
| ts_ets_MNN_1440_8 | from_last | 374400 | 0.0281814 (rank: 26) | 0.0002591 (rank: 26) | 0.0494359 (rank: 26) | 0.0002591 (rank: 26) | 0.0024439 (rank: 26) |
| ts_ets_MNN_7200_1440 | from_last | 374400 | 0.4072728 (rank: 52) | 0.0037581 (rank: 52) | 0.6228677 (rank: 52) | 0.0037537 (rank: 52) | 0.3879641 (rank: 52) |
| ts_ets_MNN_7200_720 | from_last | 374400 | 0.275585 (rank: 50) | 0.0025451 (rank: 50) | 0.4395006 (rank: 48) | 0.0025434 (rank: 50) | 0.1931608 (rank: 48) |
| ts_ets_MNN_1080_1 | from_first | 374296 | 0.0135525 (rank: 4) | 0.0001246 (rank: 4) | 0.0233866 (rank: 4) | 0.0001246 (rank: 4) | 0.0005469 (rank: 4) |
| ts_ets_MNN_1200_1 | from_first | 374296 | 0.0135521 (rank: 2) | 0.0001246 (rank: 2) | 0.0233844 (rank: 2) | 0.0001246 (rank: 2) | 0.0005468 (rank: 2) |
| ts_ets_MNN_180_1 | from_first | 374295 | 0.0135795 (rank: 22) | 0.0001248 (rank: 22) | 0.0234633 (rank: 20) | 0.0001248 (rank: 22) | 0.0005505 (rank: 20) |
| ts_ets_MNN_240_1 | from_first | 374295 | 0.0135735 (rank: 20) | 0.0001248 (rank: 20) | 0.0234467 (rank: 18) | 0.0001248 (rank: 20) | 0.0005497 (rank: 18) |
| ts_ets_MNN_360_1 | from_first | 374267 | 0.0135653 (rank: 16) | 0.0001247 (rank: 16) | 0.0234235 (rank: 16) | 0.0001247 (rank: 16) | 0.0005487 (rank: 16) |
| ts_ets_MNN_480_1 | from_first | 374275 | 0.0135592 (rank: 14) | 0.0001247 (rank: 14) | 0.0234082 (rank: 14) | 0.0001247 (rank: 14) | 0.0005479 (rank: 14) |
| ts_ets_MNN_600_1 | from_first | 374296 | 0.0135569 (rank: 12) | 0.0001246 (rank: 12) | 0.0234 (rank: 12) | 0.0001246 (rank: 12) | 0.0005476 (rank: 12) |
| ts_ets_MNN_720_1 | from_first | 374297 | 0.013556 (rank: 10) | 0.0001246 (rank: 10) | 0.0233952 (rank: 10) | 0.0001246 (rank: 10) | 0.0005473 (rank: 10) |
| ts_ets_MNN_840_1 | from_first | 374297 | 0.0135551 (rank: 8) | 0.0001246 (rank: 8) | 0.0233913 (rank: 8) | 0.0001246 (rank: 8) | 0.0005472 (rank: 8) |
| ts_ets_MNN_960_1 | from_first | 374295 | 0.0135531 (rank: 6) | 0.0001246 (rank: 6) | 0.0233885 (rank: 6) | 0.0001246 (rank: 6) | 0.000547 (rank: 6) |
| ts_ets_MNN_1080_1 | from_last | 374296 | 0.0135525 (rank: 4) | 0.0001246 (rank: 4) | 0.0233866 (rank: 4) | 0.0001246 (rank: 4) | 0.0005469 (rank: 4) |
| ts_ets_MNN_1200_1 | from_last | 374296 | 0.0135521 (rank: 2) | 0.0001246 (rank: 2) | 0.0233844 (rank: 2) | 0.0001246 (rank: 2) | 0.0005468 (rank: 2) |
| ts_ets_MNN_180_1 | from_last | 374295 | 0.0135795 (rank: 22) | 0.0001248 (rank: 22) | 0.0234633 (rank: 20) | 0.0001248 (rank: 22) | 0.0005505 (rank: 20) |
| ts_ets_MNN_240_1 | from_last | 374295 | 0.0135735 (rank: 20) | 0.0001248 (rank: 20) | 0.0234467 (rank: 18) | 0.0001248 (rank: 20) | 0.0005497 (rank: 18) |
| ts_ets_MNN_360_1 | from_last | 374267 | 0.0135653 (rank: 16) | 0.0001247 (rank: 16) | 0.0234235 (rank: 16) | 0.0001247 (rank: 16) | 0.0005487 (rank: 16) |
| ts_ets_MNN_480_1 | from_last | 374275 | 0.0135592 (rank: 14) | 0.0001247 (rank: 14) | 0.0234082 (rank: 14) | 0.0001247 (rank: 14) | 0.0005479 (rank: 14) |
| ts_ets_MNN_600_1 | from_last | 374296 | 0.0135569 (rank: 12) | 0.0001246 (rank: 12) | 0.0234 (rank: 12) | 0.0001246 (rank: 12) | 0.0005476 (rank: 12) |
| ts_ets_MNN_720_1 | from_last | 374297 | 0.013556 (rank: 10) | 0.0001246 (rank: 10) | 0.0233952 (rank: 10) | 0.0001246 (rank: 10) | 0.0005473 (rank: 10) |
| ts_ets_MNN_840_1 | from_last | 374297 | 0.0135551 (rank: 8) | 0.0001246 (rank: 8) | 0.0233913 (rank: 8) | 0.0001246 (rank: 8) | 0.0005472 (rank: 8) |
| ts_ets_MNN_960_1 | from_last | 374295 | 0.0135531 (rank: 6) | 0.0001246 (rank: 6) | 0.0233885 (rank: 6) | 0.0001246 (rank: 6) | 0.000547 (rank: 6) |
source : 52 x 8
5 Conclusion
5.1 Final Conclude
sum <- rp_1min_1y[mae == min(mae) | mape == min(mape) | rmse == min(rmse) | smape == min(smape) | mse == min(mse)]
sum %>%
dplyr::mutate(
mae = ifelse(
rank(mae) <= 3,
cell_spec(
paste0(round(mae, 7), ' (rank: ', sprintf('%1.f', rank(mae)), ')'),
color = 'darkgoldenrod', bold = TRUE),
cell_spec(
paste0(round(mae, 7), ' (rank: ', sprintf('%1.f', rank(mae)), ')'),
color = 'grey', italic = TRUE)),
mape = ifelse(
rank(mape) <= 3,
cell_spec(
paste0(round(mape, 7), ' (rank: ', sprintf('%1.f', rank(mape)), ')'),
color = 'darkgoldenrod', bold = TRUE),
cell_spec(
paste0(round(mape, 7), ' (rank: ', sprintf('%1.f', rank(mape)), ')'),
color = 'grey', italic = TRUE)),
rmse = ifelse(
rank(rmse) <= 3,
cell_spec(
paste0(round(rmse, 7), ' (rank: ', sprintf('%1.f', rank(rmse)), ')'),
color = 'darkgoldenrod', bold = TRUE),
cell_spec(
paste0(round(rmse, 7), ' (rank: ', sprintf('%1.f', rank(rmse)), ')'),
color = 'grey', italic = TRUE)),
smape = ifelse(
rank(smape) <= 3,
cell_spec(
paste0(round(smape, 7), ' (rank: ', sprintf('%1.f', rank(smape)), ')'),
color = 'darkgoldenrod', bold = TRUE),
cell_spec(
paste0(round(smape, 7), ' (rank: ', sprintf('%1.f', rank(smape)), ')'),
color = 'grey', italic = TRUE)),
mse = ifelse(
rank(mse) <= 3,
cell_spec(
paste0(round(mse, 7), ' (rank: ', sprintf('%1.f', rank(mse)), ')'),
color = 'darkgoldenrod', bold = TRUE),
cell_spec(
paste0(round(mse, 7), ' (rank: ', sprintf('%1.f', rank(mse)), ')'),
color = 'grey', italic = TRUE))) %>%
kbl(caption = 'Best Model', escape = FALSE) %>%
## https://www.w3schools.com/cssref/css_colors.asp
row_spec(0, background = 'DimGrey', color = 'yellow') %>%
column_spec(1, background = 'CornflowerBlue') %>%
column_spec(2, background = '#556DAC') %>%
column_spec(3, background = 'LightSlateGrey') %>%
column_spec(4, background = 'LightGray') %>%
column_spec(5, background = 'Gainsboro') %>%
column_spec(6, background = 'LightGray') %>%
column_spec(7, background = 'Gainsboro') %>%
column_spec(8, background = 'LightGray') %>%
kable_styling(bootstrap_options = c('striped', 'hover', 'condensed', 'responsive')) %>%
kable_material(full_width = FALSE) %>%
scroll_box(width = '100%', fixed_thead = TRUE)| model | arrange | N | mae | mape | rmse | smape | mse |
|---|---|---|---|---|---|---|---|
| ts_ets_MNN_1200_1 | from_first | 374296 | 0.0135521 (rank: 2) | 0.0001246 (rank: 2) | 0.0233844 (rank: 2) | 0.0001246 (rank: 2) | 0.0005468 (rank: 2) |
| ts_ets_MNN_1200_1 | from_last | 374296 | 0.0135521 (rank: 2) | 0.0001246 (rank: 2) | 0.0233844 (rank: 2) | 0.0001246 (rank: 2) | 0.0005468 (rank: 2) |
source : 2 x 8
From 4.2.3.2 Comparison between 1 Year & 3.5 Year Dataset, we can know the βest model based on the unique from_first dataset.
From table above, we know that ts_ets_MNN_1200_1 from_first is the βest model.
5.2 Future Studies
Next papers will compare tbats, midas, sarimax etc.
Deriv.com - Interday High Frequency Trading Models Comparison Review (Part III)
Beyond the future, will be publish Chinese Language thesis, there as a lot of online website language translator.
6 Appendix
6.1 Blooper
There has a small mistake in previous paper Deriv.com - Interday High Frequency Trading Models Comparison Review (Part I) but doesn’t affect since I only want to compare the models and here I use the same MNN model and filter unique(fromLast = FALSE) and unique(fromLast = TRUE).
6.2 Documenting File Creation
It’s useful to record some information about how your file was created.
- File creation date: 2021-02-03
- File latest updated date: 2021-04-23
- R version 4.0.5 (2021-03-31)
- R version (short form): 4.0.5
- rmarkdown package version: 2.7
- File version: 1.0.0
- Author Profile: ®γσ, Eng Lian Hu
- GitHub: Source Code
- Additional session information:
suppressMessages(require('dplyr', quietly = TRUE))
suppressMessages(require('magrittr', quietly = TRUE))
suppressMessages(require('formattable', quietly = TRUE))
suppressMessages(require('knitr', quietly = TRUE))
suppressMessages(require('kableExtra', quietly = TRUE))
sys1 <- devtools::session_info()$platform %>%
unlist %>% data.frame(Category = names(.), session_info = .)
rownames(sys1) <- NULL
sys2 <- data.frame(Sys.info()) %>%
dplyr::mutate(Category = rownames(.)) %>% .[2:1]
names(sys2)[2] <- c('Sys.info')
rownames(sys2) <- NULL
if (nrow(sys1) == 9 & nrow(sys2) == 8) {
sys2 %<>% rbind(., data.frame(
Category = 'Current time',
Sys.info = paste(as.character(lubridate::now('Asia/Tokyo')), 'JST🗾')))
} else {
sys1 %<>% rbind(., data.frame(
Category = 'Current time',
session_info = paste(as.character(lubridate::now('Asia/Tokyo')), 'JST🗾')))
}
sys <- cbind(sys1, sys2) %>%
kbl(caption = 'Additional session information:') %>%
kable_styling(bootstrap_options = c('striped', 'hover', 'condensed', 'responsive')) %>%
row_spec(0, background = 'DimGrey', color = 'yellow') %>%
column_spec(1, background = 'CornflowerBlue', color = 'red') %>%
column_spec(2, background = 'grey', color = 'black') %>%
column_spec(3, background = 'CornflowerBlue', color = 'blue') %>%
column_spec(4, background = 'grey', color = 'white') %>%
row_spec(9, bold = T, color = 'yellow', background = '#D7261E')
rm(sys1, sys2)
sys| Category | session_info | Category | Sys.info |
|---|---|---|---|
| version | R version 4.0.5 (2021-03-31) | sysname | Windows |
| os | Windows 10 x64 | release | 10 x64 |
| system | x86_64, mingw32 | version | build 19042 |
| ui | RTerm | nodename | SCIBROKES-TRADI |
| language | (EN) | machine | x86-64 |
| collate | English_World.1252 | login | Owner |
| ctype | English_World.1252 | user | Owner |
| tz | Asia/Tokyo | effective_user | Owner |
| date | 2021-04-23 | Current time | 2021-04-23 21:21:36 JST<U+0001F5FE> |
6.3 Reference
- Deriv.com - Interday High Frequency Trading Models Comparison (Blooper)
- Deriv.com - Interday High Frequency Trading Models Comparison Review (Part I) (in RPubs.com)
- Deriv.com - Interday High Frequency Trading Models Comparison Review (Part I) (in RStudioConnect.com)
- binary.com 面试试题 I - GARCH模型中的
ARCH in Mean - binary.com 面试试题 I - GARCH模型中的
ARIMA(p,d,q)参数最优化 - Forecasting: Principles and Practice
- Tableau 100 Color Palettes
- How to add multiple files to Git at the same time
Powered by - Copyright® Intellectual Property Rights of 
Sςιβrοκεrs Trαdιηg ®️ 経営企業