Nothing
R/Simulation.R
In strand: A Framework for Investment Strategy Simulation
#' Simulation class
#'
#' @description Class for running a simulation and getting results.
#'
#' @details The \code{Simulation} class is used to set up and run a daily
#' simulation over a particular period. Portfolio construction parameters and
#' other simulator settings can be configured in a yaml file that is passed to
#' the object's constructor. See \code{vignette("strand")} for information on
#' configuration file setup.
#'
#' @export
Simulation <- R6Class(
"Simulation",
public = list(
#' @description Create a new \code{Simulation} object.
#' @param config An object of class \code{list} or \code{character}, or
#' \code{NULL}. If the value passed is a character vector, it should be of
#' length 1 and specify the path to a yaml configuration file that
#' contains the object's configuration info. If the value passed is of
#' class list(), the list should contain the object's configuration info
#' in list form (e.g, the return value of calling \code{yaml.load_file} on
#' the configuration file). If the value passed is \code{NULL}, then there
#' will be no configuration information associated with the simulation and
#' it will not possible to call the \code{run} method. Setting
#' \code{config = NULL} is useful when creating simulation objects into
#' which results will be loaded with \code{readFeather}.
#' @param raw_input_data A data frame that contains all of the input data
#' (for all periods) for the simulation. The data frame must have a
#' \code{date} column. Data supplied using this parameter will be
#' used if the configuration option \code{simulator/input_data/type} is
#' set to \code{object}. Defaults to \code{NULL}.
#' @param input_dates Vector of class \code{Date} that specifies when input
#' data should be updated. If data is being supplied using the
#' \code{raw_input_data} parameter, then \code{input_dates} defaults to
#' set of dates present in this data.
#' @param raw_pricing_data A data frame that contains all of the input data
#' (for all periods) for the simulation. The data frame must have a
#' \code{date} column. Data supplied using this parameter will only be
#' used if the configuration option \code{simulator/pricing_data/type} is
#' set to \code{object}. Defaults to \code{NULL}.
#' @param security_reference_data A data frame that contains reference data
#' on the securities in the simulation, including any categories that are
#' used in portfolio construction constraints. Note that the simulator
#' will throw an error if there are input data records for which there is
#' no entry in the security reference. Data supplied using this parameter
#' will only be used if the configuration option
#' \code{simulator/secref_data/type} is set to \code{object}. Defaults to
#' \code{NULL}.
#' @param delisting_data A data frame that contains delisting dates and
#' associated returns. It must contain three columns: id (character),
#' delisting_date (Date), and delisting_return (numeric). The date in the
#' delisting_date column means the day on which a stock will be removed
#' from the simulation portfolio. It is typically the day after the last
#' day of trading. The delisting_return column reflects what, if any, P&L
#' should be recorded on the delisting date. A delisting_return of -1
#' means that the shares were deemed worthless. The delisting return is
#' multiplied by the starting net market value of the position to
#' determine P&L for the delisted position on the delisting date. Note
#' that the portfolio optimization does not include stocks that are being
#' removed due to delisting. Data supplied using this parameter will only
#' be used if the configuration option
#' \code{simulator/delisting_data/type} is set to \code{object}. Defaults
#' to \code{NULL}.
#' @return A new \code{Simulation} object.
initialize = function(config = NULL,
raw_input_data = NULL,
input_dates = NULL,
raw_pricing_data = NULL,
security_reference_data = NULL,
delisting_data = NULL) {
if (is.character(config)) {
if (!file.exists(config)) {
stop(paste0("Config file not found: ", config))
}
config <- yaml.load_file(config)
private$config <- StrategyConfig$new(config)
} else if (is.list(config)) {
private$config <- StrategyConfig$new(config)
} else if (is.null(config)) {
return(self)
} else {
stop("config must be of class list or character, or NULL")
}
# Set security_reference field
#
# TODO Improve the mechanism by which we set up the object using
# constructor parameters or file / database resources.
secref_config <- private$config$getConfig("simulator")$secref_data
stopifnot(!is.null(secref_config$type),
length(secref_config$type) %in% 1,
is.character(secref_config$type))
if (secref_config$type %in% "file") {
private$security_reference <- read_feather(secref_config$filename)
} else if (secref_config$type %in% "object") {
private$security_reference <- security_reference_data
} else {
stop(paste0("Invalid config value for secref_config/type: ", secref_config$type))
}
# Set delisting_data field. Delisting data is not required. If no
# delisting_data parameter is specified in the config, set the
# delisting_data field to an empty data frame with the appropriate
# columns.
#
# TODO Pull setup logic for secref and delisting data into helper
# methods.
delistings_config <- private$config$getConfig("simulator")$delisting_data
if (is.null(delistings_config)) {
private$delisting_data <- data.frame(id = character(0),
delisting_date = structure(numeric(0), class = "Date"),
delisting_return = numeric(0))
} else {
stopifnot(!is.null(delistings_config$type),
length(delistings_config$type) %in% 1,
is.character(delistings_config$type))
if (delistings_config$type %in% "file") {
private$delisting_data <- read_feather(delistings_config$filename)
} else if (delistings_config$type %in% "object") {
private$delisting_data <- delisting_data
} else {
stop(paste0("Invalid config value for delistings_config/type: ", delistings_config$type))
}
}
# Set raw data from constuctor parameters.
if (!is.null(raw_input_data)) {
if (private$config$getConfig("simulator")$input_data$type %in% "file") {
stop("Passing data via raw_input_data but configuration specifies file-based inputs")
}
stopifnot("date" %in% names(raw_input_data))
private$raw_input_data <- raw_input_data
} else {
if (private$config$getConfig("simulator")$input_data$type %in% "object") {
stop("raw_input_data is NULL but configuration specifies object-based inputs")
}
}
if (!is.null(raw_pricing_data)) {
if (private$config$getConfig("simulator")$pricing_data$type %in% "file") {
stop("Passing data via raw_pricing_data but configuration specifies file-based inputs")
}
stopifnot("date" %in% names(raw_pricing_data))
private$raw_pricing_data <- raw_pricing_data
} else {
if (private$config$getConfig("simulator")$pricing_data$type %in% "object") {
stop("raw_pricing_data is NULL but configuration specifies object-based inputs")
}
}
if (isTRUE(private$config$getConfig("simulator")$verbose)) {
private$verbose <- TRUE
}
# Set dates
if (is.null(input_dates) & !is.null(raw_input_data)) {
private$input_dates <- unique(sort(raw_input_data$date))
} else {
private$input_dates <- input_dates
}
invisible(self)
},
#' @description Set the verbose flag to control info output.
#' @param verbose Logical flag indicating whether to be verbose or not.
#' @return No return value, called for side effects.
setVerbose = function(verbose) {
stopifnot(is.logical(verbose),
length(verbose) %in% 1)
private$verbose <- verbose
invisible(self)
},
#' @description Set the callback function for updating progress when running
#' a simulation in shiny.
#' @param callback A function suitable for updating a shiny Progress object.
#' It must have two parameters: \code{value}, indicating the progress
#' amount, and detail, and \code{detail}, a text string for display on the
#' progress bar.
#' @return No return value, called for side effects.
setShinyCallback = function(callback) {
if (!is.function(callback)) {
stop("callback must be a function")
}
private$shiny_callback <- callback
invisible(self)
},
#' @description Get security reference information.
#' @return An object of class \code{data.frame} that contains the security
#' reference data for the simulation.
getSecurityReference = function() {
invisible(private$security_reference)
},
#' @description Run the simulation.
#' @return No return value, called for side effects.
run = function() {
if (is.null(private$config)) {
stop("Can not run simulation: no configuration defined.")
}
# Grab simulator section of config
simulator_config <- private$config$getConfig("simulator")
# Input data object
stopifnot(!is.null(simulator_config$input_data$type),
length(simulator_config$input_data$type) %in% 1,
is.character(simulator_config$input_data$type),
simulator_config$input_data$type %in% c("object", "file"))
if (simulator_config$input_data$type %in% "object") {
input_data_obj <- CrossSection$new(TRUE)
input_data_obj$setRaw(private$raw_input_data)
} else if (simulator_config$input_data$type %in% "file") {
input_data_obj <- CrossSectionFromFile$new(TRUE,
simulator_config$input_data$directory,
simulator_config$input_data$prefix)
} else {
stop("Unsupported value for input_data/type")
}
if (!is.null(simulator_config$input_data$na_replace)) {
input_data_obj$setNAReplaceValue(simulator_config$input_data$na_replace)
}
# Pricing data object
stopifnot(!is.null(simulator_config$pricing_data$type),
length(simulator_config$pricing_data$type) %in% 1,
is.character(simulator_config$pricing_data$type),
simulator_config$pricing_data$type %in% c("object", "file"))
if (simulator_config$pricing_data$type %in% "object") {
pricing_data_obj <- CrossSection$new(TRUE)
pricing_data_obj$setRaw(private$raw_pricing_data)
} else if (simulator_config$pricing_data$type %in% "file") {
pricing_data_obj <- CrossSectionFromFile$new(TRUE,
simulator_config$pricing_data$directory,
simulator_config$pricing_data$prefix)
} else {
stop("Unsupported value for pricing_data/type")
}
pricing_data_obj$setColumnMap(simulator_config$pricing_data$columns)
pricing_data_obj$setCarryForwardValue(list(
volume = 0,
adjustment_ratio = 1,
dividend = as.numeric(NA),
distribution = as.numeric(NA)
))
all_strategies <- private$config$getStrategyNames()
portfolio <- Portfolio$new(all_strategies)
all_dates <- self$getSimDates()
for (current_date in as.list(all_dates)) {
if (isTRUE(private$verbose)) {
cat("[", private$config$getConfig("name"), "] Working on ", format(current_date), "\n", sep = "")
}
if (is.function(private$shiny_callback)) {
private$shiny_callback(which(all_dates %in% current_date) / length(all_dates),
paste0("Working on ", format(current_date)))
}
# Retrieve data for the current period.
#
# TODO Impose restrictions on allowable input data column names. For
# example, columns of the form shares_{strategy name} will collide with
# the simulator's work columns. There are other ways around this issue
# but imposing column restrictions is the easiest.
if (!is.null(private$input_dates) && !current_date %in% private$input_dates) {
input_data <- input_data_obj$getCurrent()
} else {
input_data <- input_data_obj$update(current_date)
# Collect metadata as specified in the config (such as correlation of
# values from one period to the next).
if (!is.null(simulator_config$input_data$track_metadata)) {
input_stats <- input_data_obj$periodStats(simulator_config$input_data$track_metadata)
private$saveInputStats(current_date, input_stats)
}
}
if (nrow(input_data) %in% 0) {
stop("Cannot formulate optimization: no input data found")
}
pricing_data <- pricing_data_obj$update(current_date)
# Properties we enforce on input data:
#
# 1. Each security with an entry in input_data must have an entry in
# pricing_data *unless* the simulator is configured to omit inputs
# records that have not yet been priced.
#
# 2. Each security in which there is a non-zero position has an entry in
# pricing_data and input_data.
#
# 3. All securities must be present in the security reference.
stopifnot(
all(input_data$id %in% private$security_reference$id),
all(portfolio$getPositions()$id %in% private$security_reference$id),
all(portfolio$getPositions()$id %in% pricing_data$id),
all(portfolio$getPositions()$id %in% input_data$id)
)
if (!all(input_data$id %in% pricing_data$id)) {
if (!isTRUE(simulator_config$inputs_without_pricing %in% "omit")) {
stop(paste0("Input records (",
sum(!input_data$id %in% pricing_data$id),
") found without pricing data. Consider setting simulator/inputs_without_pricing=omit"))
} else {
if (isTRUE(private$verbose)) {
cat("Omitting ",
sum(!input_data$id %in% pricing_data$id),
" input records without pricing data: ",
paste0(input_data$id[!input_data$id %in% pricing_data$id], collapse = ", "),
"\n")
}
input_data <- filter(input_data, .data$id %in% pricing_data$id)
}
}
input_data <- input_data %>%
rename(inputs_carry_forward = carry_forward)
pricing_data <- pricing_data %>%
select("id",
"close_price", "prior_close_price",
"adjustment_ratio", "volume",
"dividend", "distribution","carry_forward") %>%
rename(pricing_carry_forward = carry_forward)
stopifnot(all(pricing_data$close_price > 0),
all(pricing_data$prior_close_price > 0))
# A further adjustment is required for pricing data that is
# carried-forward: the prior close price must be set to the same value
# as the close price.
#
# TODO Stop requiring that the user passes in the prior close price.
# Once we remove this requirement we can keep track of prior values as
# we iterate over period-by-period cross sections (and can remove logic
# like the below).
pricing_data$prior_close_price <- ifelse(pricing_data$pricing_carry_forward,
pricing_data$close_price,
pricing_data$prior_close_price)
# Create start and end price columns to keep things clear. The start
# price must be adjusted by the adjustment ratio. In the case of a 2:1
# split, the adjustment ratio is 0.5 (number of old shares over number
# of new shares). So, we multiply the previous day's unadjusted price by
# the adjustment ratio to bring yesterday's price in line with today's.
#
# Note that data sanitization (replacing NAs with default values) should
# be moved to CrossSection.
pricing_data <- pricing_data %>%
mutate(start_price = .data$prior_close_price * .data$adjustment_ratio,
end_price = .data$close_price,
volume = replace_na(.data$volume, 0),
adjustment_ratio = replace_na(.data$adjustment_ratio, 1),
dividend = replace_na(.data$dividend, 0),
distribution = replace_na(.data$distribution, 0)
)
portfolio <- portfolio$applyAdjustmentRatio(
select(pricing_data, "id", "adjustment_ratio"))
# Process delistings
# Delistings are handled in the following way:
#
# 1. All securities that have a delisting date that is equal to or
# earlier than current_date are recorded in the id_delisted vector.
# These stocks are marked not investable.
#
# 2. Positions in delisted stocks are recorded in the pos_delisted data
# frame. They are omitted from the day's portfolio optimization.
#
# 3. "Delisting trades" are added as part of the EOD bookkeeping
# sequence so that positions in delisted stocks are flattened. In the
# day's detail data, the logical column 'delisting' indicates that the
# trade was due to a delisting.
#
# TODO For delisted stocks we can go back to our data interfaces and
# ensure data for them is not carried forward after the delisting date.
# This will decrease the size of our inputs by omitting securities we
# know we will never trade again.
# We could save time and use equality on current_date here, but we run
# the risk of having dead securities in the portfolio.
id_delisted <- private$delisting_data$id[private$delisting_data$delisting_date <= current_date]
pos_delisted <- portfolio$getPositions() %>% filter(.data$id %in% id_delisted &
(.data$int_shares != 0 | .data$ext_shares != 0))
pos_delisted <- pos_delisted %>%
left_join(select(pricing_data, "id", "start_price"), by = "id") %>%
left_join(private$delisting_data, by = "id")
if (nrow(pos_delisted) > 0) {
# Record delisting info
private$saveDelistings(current_date, pos_delisted)
}
# Merge together consolidated, wide format positions, pricing data,
# security reference and inputs data for passing to the optimization.
# This merge will create rows with NA volues for stocks that have not
# previously appeared in the data feed. May want to use a special column
# prefix for shares columns (as opposed to "shares_") to avoid name
# conflict with user columns.
#
# Note that the reference price for the optimization is the start_price.
# We could save some cycles by setting price_var = "start_price" in the
# config.
#
# Note the importance of the assertion above that each position appear
# in the input data feed. If the assertion were to be false, then we
# would lose position records in first left join.
#
# TODO slim down security reference so that it only contains columns
# used in category constraints.
input_data <-
left_join(input_data,
portfolio$getConsolidatedPositions(),
by = "id") %>%
left_join(pricing_data,
by = "id") %>%
left_join(private$security_reference,
by = "id") %>%
mutate(ref_price = .data$start_price,
investable = TRUE,
delisting = .data$id %in% pos_delisted$id) %>%
mutate_at(.vars = vars(portfolio$getShareColumns()),
.funs = ~ replace_na(., 0))
# Set investable flag based on
#
# 1. Delisting status
# 2. Universe configuration
#
# TODO Move universe logic into PortOpt class. It's OK to sort out
# delistings in the simulator, but each strategy should be able to have
# its own universe.
input_data$investable <- !input_data$id %in% id_delisted
# By default, presence in the latest set of input data affects
# investability: stocks that are not in the latest update are not
# investable. This makes it easy to define the investable universe
# simply as the stocks present in the input cross-section.
#
# This behavior may not always be desirable, and can be controlled by
# setting the simulator/inputs_define_universe configuration option to
# FALSE. For example, a user may want to pass in updated factor data for
# stocks that are no longer in the universe but may be in the portfolio.
#
# Note that simulator/inputs_define_universe = TRUE is the same as
# having the expression `!inputs_carry_forward` as part of the
# simulator/universe configuration parameter.
#
# Note also that any expression in simulator/universe will be applied
# regardless of the setting of simulator/inputs_define_universe.
# TODO We need to validate config entries and set defaults in the
# StrategyConfig class.
inputs_define_universe <- simulator_config$inputs_define_universe
stopifnot(is.null(inputs_define_universe) || is.logical(inputs_define_universe))
if (is.null(inputs_define_universe) || isTRUE(inputs_define_universe)) {
input_data$investable <- input_data$investable & !input_data$inputs_carry_forward
}
if (!is.null(simulator_config$universe) && length(simulator_config$universe) > 0) {
univ <- eval(rlang::parse_expr(simulator_config$universe), envir = input_data)
if (any(is.na(univ))) {
stop("Found NA universe values")
}
input_data$investable <- input_data$investable & univ
}
# Normalization of input variables and risk factors.
# Perform normalization for variables listed in
# simulator/normalize_in_vars.
#
# The procedure is as follows:
#
# 1. Set variable values for non-investable securities to NA.
# 2. Normalize variable to N(0, 1).
# 3. Replace NAs created in step 1 with 0.
#
# Store raw (pre-normalized) values for column foo in the detail dataset
# column foo_raw.
if (!is.null(simulator_config$normalize_in_vars)) {
for (normalize_var in simulator_config$normalize_in_vars) {
raw_normalize_var <- paste0(normalize_var, "_raw")
input_data <- input_data %>%
mutate(
!! raw_normalize_var := get(normalize_var),
!! normalize_var := replace_na(normalize(ifelse(investable, get(normalize_var), NA)), 0))
}
}
# Perform normalization for variables listed in
# simulator/normalize_factor_vars.
#
# factor_vars (variables used in constraint calculations) are normalized
# as follows:
#
# 1. Set variable values for non-investable securities *in which there
# is no position* to NA.
# 2. Normalize variable to N(0, 1).
# 3. Replace NAs created in step 1 with 0.
#
# The idea is that in_var values should be 0 for stocks that are
# non-investable to encourage exiting, while factor_vars for positions
# in such stocks should be preserved so that accurate exposures can be
# calculated.
if (!is.null(simulator_config$normalize_factor_vars)) {
for (normalize_var in simulator_config$normalize_factor_vars) {
raw_normalize_var <- paste0(normalize_var, "_raw")
# Preserve factor_var value if there is a position in any strategy.
non_zero_pos <- rowSums(abs(input_data[portfolio$getShareColumns()])) != 0
input_data <- input_data %>%
mutate(
!! raw_normalize_var := get(normalize_var),
!! normalize_var := replace_na(normalize(ifelse(investable | non_zero_pos, get(normalize_var), NA)), 0))
}
}
stopifnot(!any(is.na(input_data)))
# Make a copy of input_data to pass to the PortOpt class. We do this to
# accomodate delistings, which will be removed from the portfolio later
# in the process, and which should not contribute to the current day's
# constraint calculations.
opt_input <- input_data
if(nrow(pos_delisted) > 0) {
# Omit positions in stocks that will be exited due to delisting from the
# optimization. Do this by setting their current share levels to 0 in
# the optimization's input data.
opt_input <- opt_input %>%
mutate_at(.vars = portfolio$getShareColumns(),
.funs = ~ replace(., delisting, 0))
}
# Create problem and solve
portOpt <- PortOpt$new(private$config, opt_input)
portOpt$solve()
# Record information on loosened constraints
if (length(portOpt$getLoosenedConstraints()) > 0) {
loosened_df <- data.frame(date = current_date,
constraint_name = names(portOpt$getLoosenedConstraints()),
pct_loosened = 100 * (1 - as.vector(unlist(portOpt$getLoosenedConstraints()))))
private$saveLooseningInfo(current_date, loosened_df)
}
orders <- portOpt$getResultData() %>%
select("id", contains("shares"))
# Handle positions that have grown too large.
#
# After the optimization has generated orders, if the force_trim_factor
# is set, add orders to trim positions that are too large.
#
# A force_trim_factor value of X means "if a position grows larger than
# X times its max position size, force a trade to trim the position back
# to X times its max position size."
#
# So if a security has a max long position size of 10,000, its current
# size is 15,000, and force_trim_factor = 1.2, we add a trade to sell
# 3,000 so that the post-trade position size is 1.2 times max position
# size = 12,000.
#
# TODO Trim positions in PortOpt as part of the optimization process
# (when setting variable limits). We will still want the ability to force
# trim positions, however, since these limits will also be subject to
# loosening when no solution can be found.
if (!is.null(simulator_config$force_trim_factor)) {
force_trim_factor <- simulator_config$force_trim_factor
stopifnot(is.numeric(force_trim_factor),
force_trim_factor >= 1)
vol_var <- private$config$getConfig("vol_var")
too_big <- input_data %>%
filter(!delisting) %>%
select("id", !!vol_var, "start_price",
!!portfolio$getShareColumns()) %>%
pivot_longer(cols = portfolio$getShareColumns(),
names_to = "strategy",
names_prefix = "shares_",
values_to = "shares") %>%
left_join(portOpt$getMaxPosition(), by = c("id", "strategy")) %>%
mutate(pos_nmv = shares * start_price,
max_pos_trim = ifelse(shares > 0, max_pos_lmv, max_pos_smv) *
force_trim_factor) %>%
# Grab those positions that are above the market value trim
# threshold.
filter(abs(shares * start_price) > abs(max_pos_trim)) %>%
# Compute how much can be trimmed
left_join(portOpt$getMaxOrder(), by = c("id", "strategy")) %>%
mutate(
# trim_gmv is the lesser of the amount required to trade the
# position down to max_pos_trim and the maximum order size.
trim_gmv = pmin(abs(pos_nmv) - abs(max_pos_trim), max_order_gmv),
force_order_shares = -1 * sign(shares) * floor(trim_gmv / start_price)) %>%
# Filter out cases where we are less than 1 share away from the
# max
filter(force_order_shares != 0) %>%
select("id", "strategy", "force_order_shares")
if (nrow(too_big) > 0) {
# The orders records (in wide format) for the securities in the
# too_big data frame need to be reconstructed. This is more involved
# because there can be N strategies and M != N strategies that have
# positions over the limit.
#
# It may be more straightforward to do the entire force-trim process
# in a loop where we work on the columns for the strategies' orders
# one at a time, then recalculate the joint-level order.
revised_orders <- orders %>%
filter(id %in% too_big$id) %>%
pivot_longer(cols = paste0("order_", c("shares_joint", portfolio$getShareColumns())),
names_to = "strategy",
names_prefix = "order_shares_",
values_to = "order_shares") %>%
filter(strategy != "joint") %>%
# Left join with data frame that contains force orders
left_join(too_big, by = c("id", "strategy")) %>%
# Revise order_shares if it is smaller than force_order_shares.
#
# Also record whether the signs of order_shares and
# force_order_shares are different. If they are, then the original
# order would represent an increase in the size of the position.
mutate(sign_mismatch = sign(force_order_shares) * sign(order_shares) < 0,
order_shares = ifelse(!is.na(force_order_shares) &
abs(order_shares) < abs(force_order_shares),
force_order_shares,
order_shares))
stopifnot(!any(revised_orders$sign_mismatch, na.rm = TRUE))
revised_orders <- revised_orders %>% select(id, strategy, order_shares)
# Calculate joint level shares.
joint_level <- revised_orders %>%
group_by(id) %>%
summarise(
strategy = "joint",
order_shares = sum(order_shares))
revised_orders <- rbind(revised_orders, joint_level)
# Now pivot back to wide format
revised_orders <- revised_orders %>%
pivot_wider(names_from = "strategy",
values_from = "order_shares",
names_prefix = "order_shares_")
# Adjust column order to match data frame 'orders'
revised_orders <- revised_orders[names(orders)]
# Finally: remove orders generated by the optimization (if any) and
# add the force-exit orders.
orders <- filter(orders, !.data$id %in% revised_orders$id)
orders <- rbind(orders, revised_orders)
}
}
# Handle non-investable securities.
#
# Often due to a changing universe the portfolio will have positions in
# stocks that are not part of the investable universe. Setting the
# simulator/force_exit_non_investable configuration parameter controls
# how these positions are handled. If the
# simulator/force_exit_non_investable parameter is not set, such
# positions are allowed in the portfolio (but increasing their size is
# prohibited in the optimization).
#
# TODO Exit non-investable positions in PortOpt as part of the
# optimization process (when setting variable limits). Like position
# trimming, we will still want the ability to force exit these
# positions, since these limits will also be subject to loosening when
# no solution can be found.
#
# TODO Refactor force-trimming above and force-exiting below to remove
# duplicated code.
if (isTRUE(simulator_config$force_exit_non_investable)) {
vol_var <- private$config$getConfig("vol_var")
non_investable <- input_data %>%
filter(!investable & ! delisting) %>%
select("id", !!vol_var, start_price, !!portfolio$getShareColumns()) %>%
pivot_longer(cols = portfolio$getShareColumns(),
names_to = "strategy",
names_prefix = "shares_",
values_to = "shares") %>%
filter(shares != 0)
if (nrow(non_investable) > 0) {
# Call floor on abs share value to round toward zero (to avoid
# over-sizing in corner cases).
non_investable <- non_investable %>%
left_join(portOpt$getMaxOrder(), by = c("id", "strategy")) %>%
mutate(
pos_nmv = shares * start_price,
exit_gmv = pmin(abs(pos_nmv), max_order_gmv),
order_shares = -1 * sign(shares) * floor(exit_gmv / start_price)) %>%
select("id", "strategy", "order_shares")
# Calculate joint level shares.
joint_level <- non_investable %>%
group_by(id) %>%
summarise(
strategy = "joint",
order_shares = sum(order_shares))
non_investable <- rbind(non_investable, joint_level)
# Now pivot back to wide format
non_investable <- non_investable %>%
pivot_wider(names_from = "strategy",
values_from = "order_shares",
names_prefix = "order_shares_")
# Adjust column order to match data frame 'orders'
non_investable <- non_investable[names(orders)]
# Finally: remove orders generated by the optimization (if any) and
# add the force-exit orders.
orders <- filter(orders, !.data$id %in% non_investable$id)
orders <- rbind(orders, non_investable)
}
}
# Handle delistings
#
# Here we enter delisting trades so that the entire positions in stocks
# that are delisted are removed by EOD. Note that below, fill rate for
# delisting trades will be set to 1 regardless of volume.
if (any(input_data$delisting)) {
delistings <- input_data %>%
filter(delisting) %>%
select("id", !!portfolio$getShareColumns()) %>%
pivot_longer(cols = portfolio$getShareColumns(),
names_to = "strategy",
names_prefix = "shares_",
values_to = "order_shares") %>%
# Completely exit the position
mutate(order_shares = -1 * order_shares) %>%
# Pivot back
pivot_wider(names_from = "strategy",
values_from = "order_shares",
names_prefix = "order_shares_")
# Recalculate order_shares_joint
delistings$order_shares_joint <- rowSums(select(delistings, -"id"))
# Adjust column order to match data frame 'orders'
delistings <- delistings[names(orders)]
orders <- filter(orders, !.data$id %in% delistings$id)
orders <- rbind(orders, delistings)
}
# Now that the order generation process is complete, join orders back to the
# original cross-section.
stopifnot(setequal(orders$id, input_data$id))
input_data <-
inner_join(input_data, orders, by = "id")
# Now for each strategy, separate orders that need to be worked in the
# market from orders that net down with orders from other strategies.
# 0. Pre-compute the maximum number of shares that are available for
# fills, computed as the product of the configured fill rate and the
# day's dollar trading volume.
stopifnot(!is.null(simulator_config$fill_rate_pct_vol),
simulator_config$fill_rate_pct_vol > 0)
input_data$fill_shares_max <-
round(input_data$volume * simulator_config$fill_rate_pct_vol / 100)
# Compute the fill rate ahead of time for each stock. The fill rate is 1
# if the maximum number of shares available (fill_shares_max) is greater
# than the size of the total order across all strategies. Computing this
# value makes fill calculations easier.
input_data$fill_rate <-
ifelse(input_data$order_shares_joint %in% 0 |
abs(input_data$order_shares_joint) <= input_data$fill_shares_max |
input_data$delisting,
1,
input_data$fill_shares_max / abs(input_data$order_shares_joint))
stopifnot(!any(is.na(input_data$fill_shares_max)))
# 1. Compute the total number of shares bought and sold across all
# strategies.
input_data$buy_shares_joint <- input_data$sell_shares_joint <- 0
for (strategy in private$config$getStrategyNames()) {
this_order_shares <- input_data[[paste0("order_shares_", strategy)]]
input_data$buy_shares_joint <- input_data$buy_shares_joint +
ifelse(this_order_shares > 0, this_order_shares, 0)
input_data$sell_shares_joint <- input_data$sell_shares_joint +
ifelse(this_order_shares < 0, this_order_shares, 0)
}
strategy_name_regexp <- paste0(private$config$getStrategyNames(), collapse = "|")
# 2. Using the totals computed in (1), pivot the data to long format to
# calculate fills.
res <- input_data %>%
select("id", "fill_rate",
contains("shares")) %>%
# Is there a cleaner way to select all strategy columns for the pivot?
# Note that the below names_pattern requires that strategy names only
# consist of alpha numeric characters.
pivot_longer(cols = matches(!!strategy_name_regexp),
names_to = c(".value", "strategy"),
names_pattern = paste0("(.*)_(", strategy_name_regexp, ")$")) %>%
left_join(portfolio$getPositions(), by = c("id", "strategy")) %>%
mutate(
# Orders
#
# Calculate shares that must be traded in the market
market_order_shares = as.integer(round(
ifelse(.data$order_shares > 0 & .data$order_shares_joint > 0,
.data$order_shares / .data$buy_shares_joint * .data$order_shares_joint,
ifelse(.data$order_shares < 0 & .data$order_shares_joint < 0,
abs(.data$order_shares / .data$sell_shares_joint) * .data$order_shares_joint,
0)))),
# Shares transferred to other strategies
transfer_order_shares = as.integer(.data$order_shares - .data$market_order_shares),
# Fills
#
# Market fills are subject to market liquidity
market_fill_shares = as.integer(floor(round(.data$market_order_shares * .data$fill_rate))),
# Transfers are filled completely
transfer_fill_shares = as.integer(.data$transfer_order_shares),
int_shares = replace_na(.data$int_shares, 0L),
ext_shares = replace_na(.data$ext_shares, 0L),
# External and internal end positions are computed from market and
# transfer fills
end_int_shares = as.integer(.data$int_shares + .data$transfer_fill_shares),
end_ext_shares = as.integer(.data$ext_shares + .data$market_fill_shares),
# Total fill = market + transfer
fill_shares = .data$transfer_fill_shares + .data$market_fill_shares,
# End number of shares = start + fill
end_shares = .data$shares + .data$fill_shares)
res <- res %>%
select("id", "strategy", "shares",
ends_with("_shares"))
# Roll up strategy data to a "joint" top level with a sum.
joint_data <- res %>%
select(-"strategy") %>%
group_by(.data$id) %>%
summarise_all(sum) %>%
mutate(strategy = "joint")
res <- rbind(res, joint_data)
# Compute market values, costs, and P&L.
stopifnot(
!is.null(simulator_config$transaction_cost_pct),
simulator_config$transaction_cost_pct >= 0,
!is.null(simulator_config$financing_cost_pct),
simulator_config$financing_cost_pct >= 0
)
res <- res %>%
inner_join(select(input_data, "id", "start_price", "end_price", "dividend", "distribution",
"investable", "delisting",
!!simulator_config$add_detail_columns),
by = "id") %>%
mutate(
# Position P&L is computed by comparing the value of the starting
# position to the value of the ending position (including adjustments
# for dividends and distributions).
position_pnl = .data$shares *
(.data$end_price + .data$dividend + .data$distribution - .data$start_price),
# Trading P&L is computed by comparing the end price to the benchmark
# price.
#
# Temporarily: assuming trading at the close, so trading P&L is
# zero.
trading_pnl = 0,
# Apply trade costs
#
# Temporarily: trade costs are a fixed percentage of the notional of the market
# trade (valued at the close). There is no cost to transfer to other
# strategies.
trade_costs = abs(.data$market_fill_shares * .data$end_price) *
simulator_config$transaction_cost_pct / 100,
# Apply financing costs
#
# Temporarily: financing costs are a fixed percentage of the notional of the
# starting value of the portfolio's external positions. External
# positions are positions held on the street and are recorded in the
# ext_shares column.
#
# Use a standard 360-day-count methodology, charging for three days
# on Monday.
financing_costs = abs(.data$ext_shares * .data$start_price) *
simulator_config$financing_cost_pct / 100 / 360 *
ifelse(weekdays(current_date, FALSE) %in% "Monday", 3, 1),
gross_pnl = .data$position_pnl + .data$trading_pnl,
net_pnl = .data$gross_pnl - .data$trade_costs - .data$financing_costs,
# TODO use benchmark price in trade valuations
market_order_gmv = abs(.data$market_order_shares * .data$end_price),
market_fill_nmv = .data$market_fill_shares * .data$end_price,
market_fill_gmv = abs(.data$market_fill_nmv),
transfer_fill_nmv = .data$transfer_fill_shares * .data$end_price,
transfer_fill_gmv = abs(.data$transfer_fill_nmv),
start_nmv = (.data$int_shares + .data$ext_shares) * .data$start_price,
end_nmv = .data$end_shares * .data$end_price,
end_gmv = abs(.data$end_nmv))
# Finish processing delistings
if (any(res$delisting)) {
# Work on rows corresponding to delistings
res_delisting <- filter(res, delisting)
# First make sure the ending position is flat.
if (any(res_delisting$end_int_shares != 0) ||
any(res_delisting$end_ext_shares != 0) ||
any(res_delisting$end_shares != 0)) {
stop("Found non-zero ending position in delisted security")
}
# Bring in delisting_return
res_delisting <- res_delisting %>% left_join(select(pos_delisted, id, delisting_return), by = "id")
if (any(is.na(res_delisting$delisting_return))) {
stop("Found NA delisting return")
}
if (!is.numeric(res_delisting$delisting_return)) {
stop("Delisting return must be numeric")
}
if (any(res_delisting$delisting_return < -1)) {
stop("Delisting return can not be less than -1")
}
# Ensure that all P&L and costs for the delisting are zero. Then set
# gross and net P&L to the delisting return times starting net
# market value.
res_delisting <- res_delisting %>%
mutate(
position_pnl = 0,
trading_pnl = 0,
trade_costs = 0,
financing_costs = 0,
gross_pnl = start_nmv * delisting_return,
net_pnl = gross_pnl
) %>%
select(-delisting_return)
# Replace the original rows with the rows we worked on
res <- filter(res, !delisting)
res <- rbind(res, res_delisting)
}
# Bring in max position information
res <- res %>%
left_join(portOpt$getMaxPosition(), by = c("strategy", "id"))
# if (nrow(filter(res, strategy %in% "joint" & transfer_fill_gmv != 0))) browser()
# Simple sums
summary_sum_data <- res %>%
select("strategy",
ends_with("_nmv"),
ends_with("_gmv"),
ends_with("_pnl"),
ends_with("_costs")) %>%
group_by(.data$strategy) %>%
summarise_all(sum) %>%
ungroup() %>%
mutate(fill_rate_pct = 100 * .data$market_fill_gmv / .data$market_order_gmv)
# More complex aggregation on nmv
summary_addl_data <- res %>%
select("strategy", "end_nmv", "start_nmv", "investable") %>%
group_by(.data$strategy) %>%
summarise(
end_lmv = sum(.data$end_nmv[.data$end_nmv > 0]),
end_smv = sum(.data$end_nmv[.data$end_nmv < 0]),
start_lmv = sum(.data$start_nmv[.data$start_nmv > 0]),
start_smv = sum(.data$start_nmv[.data$start_nmv < 0]),
end_num = sum(.data$end_nmv != 0),
end_num_long = sum(.data$end_nmv > 0),
end_num_short = sum(.data$end_nmv < 0),
num_investable = sum(.data$investable)
)
summary_data <-
inner_join(summary_sum_data,
summary_addl_data, by = "strategy")
# Update positions
portfolio$setPositions(
filter(res, !.data$strategy %in% "joint") %>%
select("id",
"strategy",
"int_shares" = "end_int_shares",
"ext_shares" = "end_ext_shares"))
# Calculate EOD exposures.
#
# First prepare data. Columns for category grouping are static and live
# in this object's security_reference member. Factor data is different
# each day and can be found in 'input_data'.
#
# TODO automatically calculate exposures for all categories and factors
# used in constraints.
category_vars <- simulator_config$calculate_exposures$category_vars
factor_vars <- simulator_config$calculate_exposures$factor_vars
if (!is.null(category_vars) || !is.null(factor_vars)) {
exposures_input <-
res %>%
select("strategy", "id", "end_nmv", "end_gmv") %>%
left_join(select(input_data, "id", one_of(!!factor_vars), one_of(!!category_vars)),
by = "id") %>%
mutate(end_lmv = ifelse(end_nmv > 0, end_nmv, 0),
end_smv = ifelse(end_nmv < 0, end_nmv, 0))
# Save net exposures
exposures <- calculate_exposures(detail_df = exposures_input,
in_var = "end_nmv",
weight_divisor = private$getStrategyCapital(),
category_vars = category_vars,
factor_vars = factor_vars)
private$saveExposures(current_date, exposures, type = "net")
# Save long exposures
exposures <- calculate_exposures(detail_df = exposures_input,
in_var = "end_lmv",
weight_divisor = private$getStrategyCapital(),
category_vars = category_vars,
factor_vars = factor_vars)
private$saveExposures(current_date, exposures, type = "long")
# Save short exposures
exposures <- calculate_exposures(detail_df = exposures_input,
in_var = "end_smv",
weight_divisor = private$getStrategyCapital(),
category_vars = category_vars,
factor_vars = factor_vars)
private$saveExposures(current_date, exposures, type = "short")
# Save gross exposures
exposures <- calculate_exposures(detail_df = exposures_input,
in_var = "end_gmv",
weight_divisor = private$getStrategyCapital(),
category_vars = category_vars,
factor_vars = factor_vars)
private$saveExposures(current_date, exposures, type = "gross")
}
# Save sim summary, sim detail, and optimization data.
private$saveSimSummary(current_date, summary_data)
# TODO Add flags that control saving of other data. Right now we can
# only turn saving detail off and on.
if (is.null(simulator_config$skip_saving) ||
!"detail" %in% simulator_config$skip_saving) {
# The full detail dataset is large. To save specific columns, use the
# simulator/keep_detail_columns parameter.
if (!is.null(simulator_config$keep_detail_columns)) {
keep_detail_columns <-
unique(c("id", "strategy",
simulator_config$keep_detail_columns))
res <- res %>% select(!!keep_detail_columns)
}
# The 'keep_detail_all_rows' flag controls whether to save all detail
# rows, or only rows where there is a holding or trade for the stock.
# It defaults to FALSE.
keep_detail_all_rows <- FALSE
if (!is.null(simulator_config$keep_detail_all_rows) &&
is.logical(simulator_config$keep_detail_all_rows)) {
keep_detail_all_rows <- simulator_config$keep_detail_all_rows
}
if (!isTRUE(keep_detail_all_rows)) {
res <- res %>% filter(start_nmv != 0 | end_nmv != 0)
}
private$saveSimDetail(current_date, res)
}
private$saveOptimizationSummary(current_date, portOpt$summaryDf())
}
invisible(self)
},
#' @description Get a list of all date for the simulation.
#' @return A vector of class \code{Date} over which the simulation currently iterates: all
#' weekdays between the 'from' and 'to' dates in the simulation's config.
getSimDates = function() {
from <- as.Date(private$config$getConfig("from"))
to <- as.Date(private$config$getConfig("to"))
if (to < from) {
stop("to date must not be earlier than from date")
}
# The simulator currently iterates over each weekday between the dates
# 'from' and 'to'.
all_dates <- seq(from = from, to = to, by = "1 day")
all_dates <- all_dates[!weekdays(all_dates, abbreviate = FALSE) %in% c("Saturday", "Sunday")]
# Dates can be omitted using the omit_dates simulator config option.
omit_dates <- private$config$getConfig("simulator")$omit_dates
if (length(omit_dates) > 0) {
omit_dates <- as.Date(omit_dates)
all_dates <- all_dates[!all_dates %in% omit_dates]
}
all_dates
},
#' @description Get summary information.
#' @param strategy_name Character vector of length 1 that specifies the
#' strategy for which to get detail data. If \code{NULL} data for all
#' strategies is returned. Defaults to \code{NULL}.
#' @return An object of class \code{data.frame} that contains summary data
#' for the simulation, by period, at the joint and strategy level. The data
#' frame contains the following columns:
#' \describe{
#' \item{strategy}{Strategy name, or 'joint' for the aggregate strategy.}
#' \item{sim_date}{Date of the summary data.}
#' \item{market_fill_nmv}{Total net market value of fills that do not
#' net down across strategies.}
#' \item{transfer_fill_nmv}{Total net market value of fills that
#' represent "internal transfers", i.e., fills in one strategy that net
#' down with fills in another. Note that at the joint level this column
#' by definition is 0.}
#' \item{market_order_gmv}{Total gross market value of orders that do not
#' net down across strategies.}
#' \item{market_fill_gmv}{Total gross market value of fills that do not
#' net down across strategies.}
#' \item{transfer_fill_gmv}{Total gross market value of fills that
#' represent "internal transfers", i.e., fills in one strategy that net
#' down with fills in another.}
#' \item{start_nmv}{Total net market value of all positions at the start
#' of the period.}
#' \item{start_lmv}{Total net market value of all long positions at the
#' start of the period.}
#' \item{start_smv}{Total net market value of all short positions at the
#' start of the period.}
#' \item{end_nmv}{Total net market value of all positions at the end of
#' the period.}
#' \item{end_gmv}{Total gross market value of all positions at the end
#' of the period.}
#' \item{end_lmv}{Total net market value of all long positions at the
#' end of the period.}
#' \item{end_smv}{Total net market value of all short positions at the
#' end of the period.}
#' \item{end_num}{Total number of positions at the end of the period.}
#' \item{end_num_long}{Total number of long positions at the end of the
#' period.}
#' \item{end_num_short}{Total number of short positions at the end of
#' the period.}
#' \item{position_pnl}{The total difference between the end and start
#' market value of positions.}
#' \item{trading_pnl}{The total difference between the market value of
#' trades at the benchmark price and at the end price. Note: currently
#' assuming benchmark price is the closing price, so trading P&L is
#' zero.}
#' \item{gross_pnl}{Total P&L gross of costs, calculated as position_pnl
#' + trading_pnl.}
#' \item{trade_costs}{Total trade costs (slippage).}
#' \item{financing_costs}{Total financing/borrow costs.}
#' \item{net_pnl}{Total P&L net of costs, calculated as gross_pnl -
#' trade_costs - financing_costs.}
#' \item{fill_rate_pct}{Total fill rate across all market orders,
#' calculated as 100 * market_fill_gmv / market_order_gmv.}
#' \item{num_investable}{Number of investable securities (size of universe).}
#'
#' }
#'
getSimSummary = function(strategy_name = NULL) {
res <- bind_rows(private$sim_summary_list)
if (!is.null(strategy_name)) {
res <- filter(res, .data$strategy %in% !!strategy_name)
}
invisible(res)
},
#' @description Get detail information.
#' @param sim_date Vector of length 1 of class Date or character that
#' specifies the period for which to get detail information. If
#' \code{NULL} then data from all periods is returned. Defaults
#' to \code{NULL}.
#' @param strategy_name Character vector of length 1 that specifies the
#' strategy for which to get detail data. If \code{NULL} data for all
#' strategies is returned. Defaults to \code{NULL}.
#' @param security_id Character vector of length 1 that specifies the
#' security for which to get detail data. If \code{NULL} data for all
#' securities is returned. Defaults to \code{NULL}.
#' @param columns Vector of class character specifying the columns to
#' return. This parameter can be useful when dealing with very large
#' detail datasets.
#' @return An object of class \code{data.frame} that contains security-level
#' detail data for the simulation for the desired strategies, securities,
#' dates, and columns. Available columns include:
#' \describe{
#' \item{id}{Security identifier.}
#' \item{strategy}{Strategy name, or 'joint' for the aggregate strategy.}
#' \item{sim_date}{Date to which the data pertains.}
#' \item{shares}{Shares at the start of the period.}
#' \item{int_shares}{Shares at the start of the period that net down
#' with positions in other strategies.}
#' \item{ext_shares}{Shares at the start of the period that do not net
#' down with positions in other strategies.}
#' \item{order_shares}{Order, in shares.}
#' \item{market_order_shares}{Order that does not net down with orders
#' in other strategies, in shares.}
#' \item{transfer_order_shares}{Order that nets down with orders in
#' other strategies, in shares.}
#' \item{fill_shares}{Fill, in shares.}
#' \item{market_fill_shares}{Fill that does not net down with fills in
#' other strategies, in shares.}
#' \item{transfer_fill_shares}{Fill that nets down with fills in other
#' strategies, in shares.}
#' \item{end_shares}{Shares at the end of the period.}
#' \item{end_int_shares}{Shares at the end of the period that net down
#' with positions in other strategies.}
#' \item{end_ext_shares}{Shares at the end of the period that do not net
#' down with positions in other strategies.}
#' \item{start_price}{Price for the security at the beginning of the
#' period.}
#' \item{end_price}{Price for the security at the end of the period.}
#' \item{dividend}{Dividend for the security, if any, for the
#' period.}
#' \item{distribution}{Distribution (e.g., spin-off) for the security, if
#' any, for the period.}
#' \item{investable}{Logical indicating whether the security is part of
#' the investable universe. The value of the flag is set to TRUE if the
#' security has not been delisted and satisfies the universe criterion
#' provided (if any) in the \code{simulator/universe} configuration
#' option.}
#' \item{delisting}{Logical indicating whether a position in the
#' security was removed due to delisting. If delisting is set to TRUE,
#' the gross_pnl and net_pnl columns will contain the P&L
#' due to delisting, if any. P&L due to delisting is calculated as the
#' delisting return times the \code{start_nmv} of the position.}
#' \item{position_pnl}{Position P&L, calculated as shares * (end_price +
#' dividend + distribution - start_price)}
#' \item{trading_pnl}{The difference between the market value of
#' trades at the benchmark price and at the end price. Note: currently
#' assuming benchmark price is the closing price, so trading P&L is
#' zero.}
#' \item{trade_costs}{Trade costs, calculated as a fixed percentage (set
#' in the simulation configuration) of the notional of the market trade
#' (valued at the close).}
#' \item{financing_costs}{Financing cost for the position, calculated as
#' a fixed percentage (set in the simulation configuration) of the
#' notional of the starting value of the portfolio's external positions.
#' External positions are positions held on the street and are recorded
#' in the ext_shares column.}
#' \item{gross_pnl}{Gross P&L, calculated as position_pnl + trading_pnl.}
#' \item{net_pnl}{Net P&L, calculated as gross_pnl - trade_costs -
#' financing_costs.}
#' \item{market_order_nmv}{Net market value of the order that does not
#' net down with orders in other strategies.}
#' \item{market_fill_gmv}{Gross market value of the order that does not
#' net down with orders in other strategies.}
#' \item{market_fill_nmv}{Net market value of the fill that does not net
#' down with orders in other strategies.}
#' \item{market_fill_gmv}{Gross market value of the fill that does not
#' net down with orders in other strategies.}
#' \item{transfer_fill_nmv}{Net market value of the fill that nets down
#' with fills in other strategies.}
#' \item{transfer_fill_gmv}{Gross market value of the fill that nets down
#' with fills in other strategies.}
#' \item{start_nmv}{Net market value of the position at the start of the
#' period.}
#' \item{end_nmv}{Net market value of the position at the end of the
#' period.}
#' \item{end_gmv}{Gross market value of the position at the end of the
#' period.}
#'
#' }
getSimDetail = function(sim_date = NULL,
strategy_name = NULL,
security_id = NULL,
columns = NULL) {
if (!is.null(sim_date)) {
detail_data <- private$sim_detail_list[[sim_date]]
} else {
detail_data <- bind_rows(private$sim_detail_list)
}
if (!is.null(strategy_name)) {
detail_data <- detail_data %>% filter(.data$strategy %in% !!strategy_name)
}
if (!is.null(security_id)) {
detail_data <- detail_data %>% filter(.data$id %in% !!security_id)
}
if (!is.null(columns)) {
detail_data <- detail_data %>% select(!!columns)
}
invisible(detail_data)
},
#' @description Get summary information by security. This method can be
#' used, for example, to calculate the biggest winners and losers over the
#' course of the simulation.
#' @param strategy_name Character vector of length 1 that specifies the
#' strategy for which to get detail data. If \code{NULL} data for all
#' strategies is returned. Defaults to \code{NULL}.
#' @return An object of class \code{data.frame} that contains summary
#' information aggregated by security. The data frame contains the
#' following columns:
#' \describe{
#' \item{id}{Security identifier.}
#' \item{strategy}{Strategy name, or 'joint' for the aggregate
#' strategy.}
#' \item{gross_pnl}{Gross P&L for the position over the entire
#' simulation.}
#' \item{gross_pnl}{Net P&L for the position over the entire
#' simulation.}
#' \item{average_market_value}{Average net market value of the
#' position over days in the simulation where the position was not
#' flat.}
#' \item{total_trading}{Total gross market value of trades for the
#' security.}
#' \item{trade_costs}{Total cost of trades for the security over the
#' entire simulation.}
#' \item{trade_costs}{Total cost of financing for the position over the
#' entire simulation.}
#' \item{days_in_portfolio}{Total number of days there was a position in
#' the security in the portfolio over the entire simulation.}
#' }
getPositionSummary = function(strategy_name = NULL) {
detail_data <- bind_rows(private$sim_detail_list)
if (!is.null(strategy_name)) {
detail_data <- detail_data %>% filter(.data$strategy %in% !!strategy_name)
}
detail_data %>%
group_by(.data$id, .data$strategy) %>%
summarise(gross_pnl = round(sum(gross_pnl)),
net_pnl = round(sum(net_pnl)),
average_market_value = round(mean(end_nmv[end_shares != 0])),
total_trading = round(sum(market_fill_gmv)),
trade_costs = round(sum(trade_costs)),
financing_costs = round(sum(financing_costs)),
days_in_portfolio = sum(end_shares != 0)) %>%
arrange(-.data$gross_pnl)
},
#' @description Get input statistics.
#' @return An object of class \code{data.frame} that contains statistics on
#' select columns of input data. Statistics are tracked for the columns
#' listed in the configuration variable
#' \code{simulator/input_data/track_metadata}. The data frame contains the
#' following columns:
#' \describe{
#' \item{period}{Period to which statistics pertain.}
#' \item{input_rows}{Total number of rows of input data, including
#' rows carried forward from the previous period.}
#' \item{cf_rows}{Total number of rows carried forward from the previous
#' period.}
#' \item{num_na_\emph{column}}{Number of NA values in \emph{column}. This
#' measure appears for each element of \code{track_metadata}.}
#' \item{cor_\emph{column}}{Period-over-period correlation for \emph{column}.
#' This measure appears for each element of \code{track_metadata}.}
#' }
getInputStats = function() {
invisible(bind_rows(private$input_stats_list))
},
#' @description Get loosening information.
#' @return An object of class \code{data.frame} that contains, for each
#' period, which constraints were loosened in order to solve the portfolio
#' optimization problem, if any. The data frame contains the
#' following columns:
#' \describe{
#' \item{date}{Date for which the constraint was loosened.}
#' \item{constraint_name}{Name of the constraint that was loosened.}
#' \item{pct_loosened}{Percentage by which the constraint was loosened,
#' where 100 means loosened fully (i.e., the constraint is effectively
#' removed).}
#' }
getLooseningInfo = function() {
invisible(bind_rows(private$loosening_info_list))
},
#' @description Get optimization summary information.
#' @return An object of class \code{data.frame} that contains optimization
#' summary information, such as starting and ending factor constraint
#' values, at the strategy and joint level. The data frame contains the
#' following columns:
#' \describe{
#' \item{strategy}{Strategy name, or 'joint' for the aggregate strategy.}
#' \item{sim_date}{Date to which the data pertains.}
#' \item{order_gmv}{Total gross market value of orders generated by the
#' optimization.}
#' \item{start_smv}{Total net market value of short positions at the
#' start of the optimization.}
#' \item{start_lmv}{Total net market value of long positions at the
#' start of the optimization.}
#' \item{end_smv}{Total net market value of short positions at the end
#' of the optimization.}
#' \item{end_lmv}{Total net market value of long positions at the end of
#' the optimization.}
#' \item{start_\emph{factor}}{Total net exposure to \emph{factor} at the
#' start of the optimization, for each factor constraint.}
#' \item{end_\emph{factor}}{Total net exposure to \emph{factor} at the
#' start of the optimization, for each factor constraint.}
#' }
getOptimizationSummary = function() {
invisible(bind_rows(private$optimization_summary_list))
},
#' @description Get end-of-period exposure information.
#' @param type Vector of length 1 that may be one of \code{"net"},
#' \code{"long"}, \code{"short"}, and \code{"gross"}.
#' @return An object of class \code{data.frame} that contains end-of-period
#' exposure information for the simulation portfolio. The units of the
#' exposures are portfolio weight relative to strategy_captial (i.e., net
#' market value of exposure divided by strategy capital). The data frame
#' contains the following columns:
#' \describe{
#' \item{strategy}{Strategy name, or 'joint' for the aggregate strategy.}
#' \item{sim_date}{Date of the exposure data.}
#' \item{\emph{category}_\emph{level}}{Exposure to \emph{level}
#' within \emph{category}, for all levels of all category constraints, at the end
#' of the period.}
#' \item{\emph{factor}}{Exposure to \emph{factor}, for all factor
#' constraints, at the end of the period.}
#' }
getExposures = function(type = "net") {
stopifnot(type %in% c("net", "long", "short", "gross"))
invisible(bind_rows(private$exposures_list[[type]]))
},
#' @description Get information on positions removed due to delisting.
#' @return An object of class \code{data.frame} that contains a row for each
#' position that is removed from the simulation portfolio due to a
#' delisting. Each row contains the size of the position on the day on
#' which it was removed from the portfolio.
getDelistings = function() {
invisible(bind_rows(private$delistings_list))
},
#' @description Get summary information for a single strategy suitable for
#' plotting input.
#' @param strategy_name Strategy for which to return summary data.
#' @param include_zero_row Logical flag indicatiing whether to prepend a row
#' to the summary data with starting values at zero. Defaults to \code{TRUE}.
#' @return A data frame that contains summary information for the desired
#' strategy, as well as columns for cumulative net and gross total return,
#' calculated as pnl divided by ending gross market value.
getSingleStrategySummaryDf = function(strategy_name = "joint", include_zero_row = TRUE) {
res <- filter(self$getSimSummary(), .data$strategy %in% !!strategy_name)
if (isTRUE(include_zero_row)) {
# Create a zero-value starting row with date lagged by one day.
res <- res[c(1, 1:nrow(res)),]
res$sim_date[1] <- res$sim_date[1] - 1
res[1,] <- mutate_if(res[1,], is.numeric, ~ 0)
}
# Compute returns as a percentage GMV, by day and cumulative.
#
# These calculations should probably be done up front in the simulation
# loop and saved in the summary dataset.
mutate(res,
net_ret = ifelse(end_gmv %in% 0, 0, .data$net_pnl / .data$end_gmv),
net_cum_ret = cumsum(net_ret),
gross_ret = ifelse(end_gmv %in% 0, 0, .data$gross_pnl / .data$end_gmv),
gross_cum_ret = cumsum(gross_ret))
},
#' @description Draw a plot of cumulative gross and net return by date.
#' @param strategy_name Character vector of length 1 specifying the strategy
#' for the plot. Defaults to \code{"joint"}.
plotPerformance = function(strategy_name = "joint") {
self$getSingleStrategySummaryDf(strategy_name) %>%
select("sim_date", "gross_cum_ret", "net_cum_ret") %>%
rename(Gross = "gross_cum_ret", Net = "net_cum_ret") %>%
pivot_longer(
-"sim_date",
names_to = "type",
values_to = "cum_ret"
) %>%
ggplot(aes(x = sim_date, y = 100 * cum_ret, color = type, group = type)) + geom_line() +
xlab("Date") + ylab("Return (%)") +
ggtitle("Cumulative Return (% GMV)") +
theme_light() +
theme(
plot.background = element_rect(fill = NA, colour = NA),
plot.title = element_text(size = 18),
axis.text = element_text(size = 10),
axis.text.x = element_text(angle = 0),
legend.title = element_blank())
},
#' @description Draw a plot of contribution to net return on GMV for levels
#' of a specified category.
#' @param category_var Plot performance contribution for the levels of
#' \code{category_var}. \code{category_var} must be present in the
#' simulation's security reference, and detail data must be present in the
#' object's result data.
#' @param strategy_name Character vector of length 1 specifying the strategy
#' for the plot. Defaults to \code{"joint"}.
plotContribution = function(category_var, strategy_name = "joint") {
stopifnot(length(strategy_name) %in% 1)
summary_data <- self$getSimSummary(strategy_name) %>%
select(sim_date, end_gmv)
# Get the category var values from the security reference if possible.
# Otherwise look for them in the detail result data.
if (category_var %in% names(self$getSecurityReference())) {
contrib_data <- self$getSimDetail(strategy_name = strategy_name,
columns = c("id", "sim_date", "net_pnl")) %>%
left_join(select(self$getSecurityReference(), "id", !!category_var), by = "id")
} else {
contrib_data <- self$getSimDetail(strategy_name = strategy_name,
columns = c("id", "sim_date", "net_pnl", category_var))
}
contrib_data <- contrib_data %>%
group_by_at(c("sim_date", category_var)) %>%
summarise(net_pnl = sum(net_pnl)) %>%
ungroup() %>%
left_join(summary_data, by = c("sim_date")) %>%
group_by_at(category_var) %>%
mutate(net_ret = net_pnl / end_gmv,
cum_net_ret = cumsum(net_ret)) %>%
ungroup()
# Adding zero-row for each group
zero_rows <- contrib_data %>% filter(!duplicated(get(category_var))) %>%
mutate(sim_date = sim_date - 1) %>%
mutate_if(is.numeric, ~ 0)
contrib_data <- rbind(zero_rows, contrib_data)
contrib_data %>%
ggplot(aes(x = sim_date, y = 100 * cum_net_ret, color = get(category_var), group = get(category_var))) +
geom_line() +
xlab("Date") + ylab("Contribution (%)") +
ggtitle(paste0(category_var, " Contribution to Net Return (% GMV)")) +
theme_light() +
theme(
plot.background = element_rect(fill = NA, colour = NA),
plot.title = element_text(size = 18),
axis.text = element_text(size = 10),
axis.text.x = element_text(angle = 0),
legend.title = element_blank())
},
#' @description Draw a plot of total gross, long, short, and net market
#' value by date.
#' @param strategy_name Character vector of length 1 specifying the strategy
#' for the plot. Defaults to \code{"joint"}.
plotMarketValue = function(strategy_name = "joint") {
if (!strategy_name %in% c("joint", private$config$getStrategyNames())) {
stop(paste0("Invalid strategy name: ", strategy_name))
}
mv_plot_df <- select(self$getSingleStrategySummaryDf(strategy_name),
sim_date, GMV = end_gmv, LMV = end_lmv, SMV = end_smv, NMV = end_nmv) %>%
gather(type, value, GMV:NMV) %>%
filter(!is.na(value))
mv_plot_df$type <- factor(mv_plot_df$type, levels = c("GMV", "LMV", "NMV", "SMV"))
ggplot(mv_plot_df, aes(sim_date, value/1e6, color = type, group = type)) + geom_line() +
scale_color_manual(values = c("LMV" = "darkgreen", "SMV" = "darkred",
"GMV" = "dodgerblue1", "NMV" = "black")) +
xlab("Date") +
ylab("Market Value ($mm)") +
ggtitle("Market Values") +
theme_light() +
theme(
plot.background = element_rect(fill = NA, colour = NA),
plot.title = element_text(size = 18),
axis.text = element_text(size = 10),
axis.text.x = element_text(angle = 0),
legend.title = element_blank())
},
#' @description Draw a plot of exposure to all levels in a category by date.
#' @param in_var Category for which exposures are plotted. In order to plot
#' exposures for category \code{in_var}, we must have run the simulation
#' with \code{in_var} in the config setting
#' \code{simulator/calculate_exposures/category_vars}.
#' @param strategy_name Character vector of length 1 specifying the strategy
#' for the plot. Defaults to \code{"joint"}.
plotCategoryExposure = function(in_var, strategy_name = "joint") {
exposures <- self$getExposures() %>% filter(strategy %in% strategy_name) %>%
select("sim_date", starts_with(in_var))
# Make sure that there is at least one level for in_var present in the
# object's exposure data. If only sim_date is present, then most likely
# exposure to in_var was not calculated by the simulation.
if (ncol(exposures) %in% 1) {
stop(paste0("No exposure data found for in_var: ", in_var))
}
exposures %>%
pivot_longer(-"sim_date",
names_to = in_var,
names_prefix = paste0(in_var, "_"),
values_to = "exposure") %>%
ggplot(aes(sim_date, exposure * 100, color = get(in_var), group = get(in_var))) + geom_line() +
xlab("Date") +
ylab("Exposure (%)") +
ggtitle(paste0(in_var, " Exposure (% Capital)")) +
theme_light() +
theme(
plot.background = element_rect(fill = NA, colour = NA),
plot.title = element_text(size = 18),
axis.text = element_text(size = 10),
axis.text.x = element_text(angle = 0),
legend.title = element_blank())
},
#' @description Draw a plot of exposure to factors by date.
#' @param in_var Factors for which exposures are plotted.
#' @param strategy_name Character vector of length 1 specifying the strategy
#' for the plot. Defaults to \code{"joint"}.
plotFactorExposure = function(in_var, strategy_name = "joint") {
exposures <- self$getExposures() %>% filter(strategy %in% strategy_name)
exposures %>%
select("sim_date", one_of(!!in_var)) %>%
pivot_longer(-"sim_date",
names_to = "factor_name",
values_to = "exposure") %>%
ggplot(aes(sim_date, exposure * 100, color = factor_name, group = factor_name)) + geom_line() +
xlab("Date") +
ylab("Exposure (%)") +
ggtitle("Factor Exposure (% Capital)") +
theme_light() +
theme(
plot.background = element_rect(fill = NA, colour = NA),
plot.title = element_text(size = 18),
axis.text = element_text(size = 10),
axis.text.x = element_text(angle = 0),
legend.title = element_blank())
},
#' @description Draw a plot of number of long and short positions by date.
#' @param strategy_name Character vector of length 1 specifying the strategy
#' for the plot. Defaults to \code{"joint"}.
plotNumPositions = function(strategy_name = "joint") {
self$getSingleStrategySummaryDf(strategy_name) %>%
select("sim_date", "end_num_long", "end_num_short") %>%
rename(Long = "end_num_long", Short = "end_num_short") %>%
pivot_longer(
-"sim_date",
names_to = "side",
values_to = "count"
) %>%
ggplot(aes(x = sim_date, y = count, color = side, group = side)) + geom_line() +
xlab("Date") + ylab("Number of Positions") +
ggtitle("Number of Positions by Side") +
theme_light() +
theme(
plot.background = element_rect(fill = NA, colour = NA),
plot.title = element_text(size = 18),
axis.text = element_text(size = 10),
axis.text.x = element_text(angle = 0),
legend.title = element_blank())
},
#' @description Draw a plot of number of long and short positions by date.
#' @param strategy_name Character vector of length 1 specifying the strategy
#' for the plot. Defaults to \code{"joint"}.
plotTurnover = function(strategy_name = "joint") {
self$getSimSummary(strategy_name) %>%
select("sim_date", "market_fill_gmv") %>%
ggplot(aes(x = sim_date, y = market_fill_gmv)) + geom_bar(stat = "identity") +
xlab("Date") + ylab("Traded GMV ($)") +
ggtitle("Turnover") +
theme_light() +
theme(
plot.background = element_rect(fill = NA, colour = NA),
plot.title = element_text(size = 18),
axis.text = element_text(size = 10),
axis.text.x = element_text(angle = 0),
legend.title = element_blank())
},
#' @description Draw a plot of the universe size, or number of investable
#' stocks, over time.
#' @param strategy_name Character vector of length 1 specifying the strategy
#' for the plot. Defaults to \code{joint}.
plotUniverseSize = function(strategy_name = "joint") {
investable_data <- self$getSimSummary(strategy_name = strategy_name)
investable_data %>%
ggplot(aes(x = sim_date, y = num_investable)) + geom_line() +
xlab("Date") + ylab("Number of securities") +
ggtitle("Universe size") +
theme_light() +
theme(
plot.background = element_rect(fill = NA, colour = NA),
plot.title = element_text(size = 18),
axis.text = element_text(size = 10),
axis.text.x = element_text(angle = 0),
legend.title = element_blank())
},
#' @description Draw a plot of the percentage of portfolio GMV held in
#' non-investable stocks (e.g., stocks that do not satisfy universe criteria)
#' for a given strategy. Note that this plot requires detail data.
#' @param strategy_name Character vector of length 1 specifying the strategy
#' for the plot. Defaults to \code{"joint"}.
plotNonInvestablePct = function(strategy_name = "joint") {
investable_data <- self$getSimDetail(strategy_name = strategy_name,
columns = c("sim_date", "id","end_gmv", "investable"))
investable_data %>%
group_by(sim_date) %>%
summarise(pct_non_investable = 100 * (1 - sum(end_gmv[investable]) / sum(end_gmv))) %>%
ggplot(aes(x = sim_date, y = pct_non_investable)) + geom_line() +
xlab("Date") + ylab("Percentage GMV") +
ggtitle("Percentage of GMV in\nnon-investable securites") +
theme_light() +
theme(
plot.background = element_rect(fill = NA, colour = NA),
plot.title = element_text(size = 18),
axis.text = element_text(size = 10),
axis.text.x = element_text(angle = 0),
legend.title = element_blank())
},
#' @description Calculate overall simulation summary statistics, such as
#' total P&L, Sharpe, average market values and counts, etc.
#' @return A data frame that contains summary statistics, suitable for
#' reporting.
overallStatsDf = function() {
res <- self$getSingleStrategySummaryDf("joint", include_zero_row = FALSE)
data.frame(
Item = c(
"Total P&L",
"Total Return on GMV (%)",
"Annualized Return on GMV (%)",
"Annualized Vol (%)",
"Annualized Sharpe",
"Max Drawdown (%)",
"Avg GMV",
"Avg NMV",
"Avg Count",
"Avg Daily Turnover",
"Holding Period (months)"
),
Gross = c(
formatC(sum(res$gross_pnl), big.mark = ",", digit = 0, format = "f"),
sprintf("%0.1f", sum(res$gross_pnl / res$end_gmv) * 100),
sprintf("%0.1f", mean(res$gross_pnl / res$end_gmv) * 100 * 252),
sprintf("%0.1f", sd(res$gross_pnl / res$end_gmv) * 100 * sqrt(252)),
sprintf("%0.2f", mean(res$gross_pnl / res$end_gmv) / sd(res$gross_pnl / res$end_gmv) * sqrt(252)),
sprintf("%0.1f", drawdown(res$gross_pnl / res$end_gmv) * 100),
rep("", 5)
),
Net = c(
formatC(sum(res$net_pnl), big.mark = ",", digit = 0, format = "f"),
sprintf("%0.1f", sum(res$net_pnl / res$end_gmv) * 100),
sprintf("%0.1f", mean(res$net_pnl / res$end_gmv) * 100 * 252),
sprintf("%0.1f", sd(res$net_pnl / res$end_gmv) * 100 * sqrt(252)),
sprintf("%0.2f", mean(res$net_pnl / res$end_gmv) / sd(res$net_pnl / res$end_gmv) * sqrt(252)),
sprintf("%0.1f", drawdown(res$net_pnl / res$end_gmv) * 100),
formatC(mean(res$end_gmv), big.mark = ",", digit = 0, format = "f"),
formatC(mean(res$end_nmv), big.mark = ",", digit = 0, format = "f"),
formatC(mean(res$end_num), big.mark = ",", digit = 0, format = "f"),
formatC(mean(res$market_fill_gmv), big.mark = ",", digit = 0, format = "f"),
sprintf("%0.1f", 12 / (mean(res$market_fill_gmv) / mean(res$end_gmv) * 252 / 2))
), stringsAsFactors = FALSE)
},
#' @description Calculate return for each month and summary statistics for
#' each year, such as total return and annualized Sharpe. Return in data
#' frame format suitable for reporting.
#' @return The data frame contains one row for each calendar year in the
#' simulation, and up to seventeen columns: one column for year, one
#' column for each calendar month, and columns for the year's total
#' return, annualized return, annualized volatility, and annualized
#' Sharpe. Total return is the sum of daily net returns. Annualized return
#' is the mean net return times 252. Annualized volatility is the standard
#' deviation of net return times the square root of 252. Annualized Sharpe
#' is the ratio of annualized return to annualized volatility. All returns
#' are in percent.
overallReturnsByMonthDf = function() {
res <- self$getSingleStrategySummaryDf("joint", include_zero_row = FALSE)
# Group by month and calculate return.
month_ret_long <- group_by(res, date = as.Date(paste0(format(sim_date, "%Y-%m"), "-01"))) %>%
summarise(ret = sum(net_ret) * 100) %>%
ungroup() %>%
transmute(year = lubridate::year(date),
month = lubridate::month(date),
ret)
# Organize monthly returns into rows, one for each year.
month_ret_yearly <- month_ret_long %>%
tidyr::spread(month, ret)
# Compute summary statistics for each year.
stats_yearly <- group_by(res, year = lubridate::year(sim_date)) %>%
summarise(
total_ret = 100 *sum(net_ret),
ann_ret = 100 * mean(net_ret) * 252,
ann_vol = 100 * sd(net_ret) * sqrt(252)) %>%
mutate(ann_sr = ann_ret / ann_vol)
# Add stats to monthly returns.
month_ret_yearly <- month_ret_yearly %>%
left_join(stats_yearly, by = "year")
month_cols <- names(month_ret_yearly)[names(month_ret_yearly) %in% as.character(1:12)]
# Replace numerical month strings with month abbreviation. Do it this way
# to handle cases where there are fewer than 12 months.
month_ret_yearly %>%
rename_at(month_cols, ~ month.abb[as.numeric(.x)])
},
#' @description Print overall simulation statistics.
print = function() {
if (is.null(private$sim_summary_list)) {
print("Simulation object with no result data.")
} else {
print(self$overallStatsDf())
}
invisible()
},
#' @description Write the data in the object to feather files.
#' @param out_loc Directory in which output files should be created.
#' @return No return value, called for side effects.
writeFeather = function(out_loc) {
# TODO Add getter and setter for raw config data in the Config class.
yaml::write_yaml(private$config$config, paste0(out_loc, "/config.yaml"))
write_feather(self$getSecurityReference(), paste0(out_loc, "/security_reference.feather"))
write_feather(self$getSimSummary(), paste0(out_loc, "/sim_summary.feather"))
write_feather(self$getSimDetail(), paste0(out_loc, "/sim_detail.feather"))
write_feather(self$getInputStats(), paste0(out_loc, "/input_stats.feather"))
write_feather(self$getLooseningInfo(), paste0(out_loc, "/loosening_info.feather"))
write_feather(self$getOptimizationSummary(), paste0(out_loc, "/optimization_summary.feather"))
write_feather(self$getExposures(), paste0(out_loc, "/exposures.feather"))
write_feather(self$getDelistings(), paste0(out_loc, "/delistings.feather"))
invisible(self)
},
#' @description Load files created with \code{writeFeather} into the object.
#' Note that because detail data is not re-split by period, it will not be
#' possible to use the \code{sim_date} parameter when calling
#' \code{getSimDetail} on the populated object.
#' @param in_loc Directory that contains files to be loaded.
#' @return No return value, called for side effects.
readFeather = function(in_loc) {
# TODO Check to see if this object is empty before loading up data.
private$config <- StrategyConfig$new(yaml::yaml.load_file(paste0(in_loc, "/config.yaml")))
private$security_reference <- read_feather(paste0(in_loc, "/security_reference.feather"))
private$sim_summary_list <- list(read_feather(paste0(in_loc, "/sim_summary.feather")))
private$sim_detail_list <- list(read_feather(paste0(in_loc, "/sim_detail.feather")))
private$input_stats_list <- list(read_feather(paste0(in_loc, "/input_stats.feather")))
private$loosening_info_list <- list(read_feather(paste0(in_loc, "/loosening_info.feather")))
private$optimization_summary_list <- list(read_feather(paste0(in_loc, "/optimization_summary.feather")))
private$exposures_list$net <- list(read_feather(paste0(in_loc, "/exposures.feather")))
private$delistings_list <- list(read_feather(paste0(in_loc, "/delistings.feather")))
warning("It will not be possible to use the sim_date parameter of getSimDetail on this object to filter detail records by period")
invisible(self)
},
#' @description Get the object's configuration information.
#' @return Object of class \code{list} that contains the simulation's
#' configuration information.
getConfig = function() {
invisible(private$config)
},
#' @description Write an html document of simulation results.
#' @param res The object of class 'Simulation' which we want to write the
#' report about.
#' @param out_dir Directory in which output files should be created
#' @param out_file File name for output
#' @param out_fmt Format in which output files should be created. The
#' default is html and that is currently the only option.
#' @param contrib_vars Security reference variables for which to plot return
#' contribution.
writeReport = function(out_dir, out_file, out_fmt = "html", contrib_vars = NULL) {
rmarkdown::render(input = system.file("reports/simReport.Rmd",
package = "strand"),
output_format = paste0(out_fmt, "_document"),
output_file = out_file,
output_dir = out_dir,
params = list(res = self, contrib_vars = contrib_vars),
quiet = TRUE)
}
),
private = list(
# Configuration and input data
config = NULL,
raw_input_data = NULL,
input_dates = NULL,
raw_pricing_data = NULL,
security_reference = NULL,
delisting_data = NULL,
shiny_callback = NULL,
verbose = FALSE,
# Results
# _list objects are lists whose elements are result data for single
# periods.
# High-level summary data, including period return, number of positions,
# etc.
sim_summary_list = NULL,
# Position-level simulation data
sim_detail_list = NULL,
# Input data statistics, such as number of NAs and period-over-period
# correlations.
input_stats_list = NULL,
# Information on any loosened constraints.
loosening_info_list = NULL,
# Optimization summary data.
optimization_summary_list = NULL,
# Exposures data.
exposures_list = NULL,
# Data for stocks removed due to to delisting.
delistings_list = NULL,
# @description Get the strategy capital levels for the strategy, based on
# the simulation's config.
# @return A list where the names are strategy names (or 'joint') and the
# values are the capital levels for each strategy,
getStrategyCapital = function() {
capital_list <- list(joint = 0)
for (strategy_name in private$config$getStrategyNames()) {
this_capital <- private$config$getStrategyConfig(strategy_name, "strategy_capital")
capital_list[[strategy_name]] <- this_capital
capital_list[["joint"]] <- capital_list[["joint"]] + this_capital
}
capital_list
},
# @description Save summary information.
# @param period Period to which the data pertains.
# @param data_obj Data frame to save.
saveSimSummary = function(period, data_obj) {
private$sim_summary_list[[as.character(period)]] <-
mutate(data_obj, sim_date = period)
invisible(self)
},
# @description Save detail (position-level) information.
# @param period Period to which the data pertains.
# @param data_obj Data frame to save.
saveSimDetail = function(period, data_obj) {
private$sim_detail_list[[as.character(period)]] <-
mutate(data_obj, sim_date = period)
invisible(self)
},
# @description Save input statistics.
# @param period Period to which the data pertains.
# @param data_obj Data frame to save.
saveInputStats = function(period, data_obj) {
private$input_stats_list[[as.character(period)]] <-
mutate(data_obj, sim_date = period)
invisible(self)
},
# @description Save loosening information.
# @param period Period to which the data pertains.
# @param data_obj Data frame to save.
saveLooseningInfo = function(period, data_obj) {
private$loosening_info_list[[as.character(period)]] <-
mutate(data_obj, sim_date = period)
invisible(self)
},
# @description Save optimization summary information.
# @param period Period to which the data pertains.
# @param data_obj Data frame to save.
saveOptimizationSummary = function(period, data_obj) {
private$optimization_summary_list[[as.character(period)]] <-
mutate(data_obj, sim_date = period)
invisible(self)
},
# @description Save exposure information.
# @param period Period to which the data pertains.
# @param data_obj Data frame to save.
saveExposures = function(period, data_obj, type = "net") {
stopifnot(type %in% c("net", "long", "short", "gross"))
private$exposures_list[[type]][[as.character(period)]] <-
mutate(data_obj, sim_date = period)
invisible(self)
},
# @description Save information on positions removed due to delisting.
# @param period Period to which the data pertains.
# @param data_obj Data frame to save.
saveDelistings = function(period, data_obj) {
private$delistings_list[[as.character(period)]] <-
mutate(data_obj, sim_date = period)
invisible(self)
}
)
)
Try the strand package in your browser
Any scripts or data that you put into this service are public.
strand documentation built on Nov. 20, 2020, 1:08 a.m.
R Package Documentation
Browse R Packages
We want your feedback!
Note that we can't provide technical support on individual packages. You should contact the package authors for that.
#' Simulation class
#'
#' @description Class for running a simulation and getting results.
#'
#' @details The \code{Simulation} class is used to set up and run a daily
#' simulation over a particular period. Portfolio construction parameters and
#' other simulator settings can be configured in a yaml file that is passed to
#' the object's constructor. See \code{vignette("strand")} for information on
#' configuration file setup.
#'
#' @export
Simulation <- R6Class(
"Simulation",
public = list(
#' @description Create a new \code{Simulation} object.
#' @param config An object of class \code{list} or \code{character}, or
#' \code{NULL}. If the value passed is a character vector, it should be of
#' length 1 and specify the path to a yaml configuration file that
#' contains the object's configuration info. If the value passed is of
#' class list(), the list should contain the object's configuration info
#' in list form (e.g, the return value of calling \code{yaml.load_file} on
#' the configuration file). If the value passed is \code{NULL}, then there
#' will be no configuration information associated with the simulation and
#' it will not possible to call the \code{run} method. Setting
#' \code{config = NULL} is useful when creating simulation objects into
#' which results will be loaded with \code{readFeather}.
#' @param raw_input_data A data frame that contains all of the input data
#' (for all periods) for the simulation. The data frame must have a
#' \code{date} column. Data supplied using this parameter will be
#' used if the configuration option \code{simulator/input_data/type} is
#' set to \code{object}. Defaults to \code{NULL}.
#' @param input_dates Vector of class \code{Date} that specifies when input
#' data should be updated. If data is being supplied using the
#' \code{raw_input_data} parameter, then \code{input_dates} defaults to
#' set of dates present in this data.
#' @param raw_pricing_data A data frame that contains all of the input data
#' (for all periods) for the simulation. The data frame must have a
#' \code{date} column. Data supplied using this parameter will only be
#' used if the configuration option \code{simulator/pricing_data/type} is
#' set to \code{object}. Defaults to \code{NULL}.
#' @param security_reference_data A data frame that contains reference data
#' on the securities in the simulation, including any categories that are
#' used in portfolio construction constraints. Note that the simulator
#' will throw an error if there are input data records for which there is
#' no entry in the security reference. Data supplied using this parameter
#' will only be used if the configuration option
#' \code{simulator/secref_data/type} is set to \code{object}. Defaults to
#' \code{NULL}.
#' @param delisting_data A data frame that contains delisting dates and
#' associated returns. It must contain three columns: id (character),
#' delisting_date (Date), and delisting_return (numeric). The date in the
#' delisting_date column means the day on which a stock will be removed
#' from the simulation portfolio. It is typically the day after the last
#' day of trading. The delisting_return column reflects what, if any, P&L
#' should be recorded on the delisting date. A delisting_return of -1
#' means that the shares were deemed worthless. The delisting return is
#' multiplied by the starting net market value of the position to
#' determine P&L for the delisted position on the delisting date. Note
#' that the portfolio optimization does not include stocks that are being
#' removed due to delisting. Data supplied using this parameter will only
#' be used if the configuration option
#' \code{simulator/delisting_data/type} is set to \code{object}. Defaults
#' to \code{NULL}.
#' @return A new \code{Simulation} object.
initialize = function(config = NULL,
raw_input_data = NULL,
input_dates = NULL,
raw_pricing_data = NULL,
security_reference_data = NULL,
delisting_data = NULL) {
if (is.character(config)) {
if (!file.exists(config)) {
stop(paste0("Config file not found: ", config))
}
config <- yaml.load_file(config)
private$config <- StrategyConfig$new(config)
} else if (is.list(config)) {
private$config <- StrategyConfig$new(config)
} else if (is.null(config)) {
return(self)
} else {
stop("config must be of class list or character, or NULL")
}
# Set security_reference field
#
# TODO Improve the mechanism by which we set up the object using
# constructor parameters or file / database resources.
secref_config <- private$config$getConfig("simulator")$secref_data
stopifnot(!is.null(secref_config$type),
length(secref_config$type) %in% 1,
is.character(secref_config$type))
if (secref_config$type %in% "file") {
private$security_reference <- read_feather(secref_config$filename)
} else if (secref_config$type %in% "object") {
private$security_reference <- security_reference_data
} else {
stop(paste0("Invalid config value for secref_config/type: ", secref_config$type))
}
# Set delisting_data field. Delisting data is not required. If no
# delisting_data parameter is specified in the config, set the
# delisting_data field to an empty data frame with the appropriate
# columns.
#
# TODO Pull setup logic for secref and delisting data into helper
# methods.
delistings_config <- private$config$getConfig("simulator")$delisting_data
if (is.null(delistings_config)) {
private$delisting_data <- data.frame(id = character(0),
delisting_date = structure(numeric(0), class = "Date"),
delisting_return = numeric(0))
} else {
stopifnot(!is.null(delistings_config$type),
length(delistings_config$type) %in% 1,
is.character(delistings_config$type))
if (delistings_config$type %in% "file") {
private$delisting_data <- read_feather(delistings_config$filename)
} else if (delistings_config$type %in% "object") {
private$delisting_data <- delisting_data
} else {
stop(paste0("Invalid config value for delistings_config/type: ", delistings_config$type))
}
}
# Set raw data from constuctor parameters.
if (!is.null(raw_input_data)) {
if (private$config$getConfig("simulator")$input_data$type %in% "file") {
stop("Passing data via raw_input_data but configuration specifies file-based inputs")
}
stopifnot("date" %in% names(raw_input_data))
private$raw_input_data <- raw_input_data
} else {
if (private$config$getConfig("simulator")$input_data$type %in% "object") {
stop("raw_input_data is NULL but configuration specifies object-based inputs")
}
}
if (!is.null(raw_pricing_data)) {
if (private$config$getConfig("simulator")$pricing_data$type %in% "file") {
stop("Passing data via raw_pricing_data but configuration specifies file-based inputs")
}
stopifnot("date" %in% names(raw_pricing_data))
private$raw_pricing_data <- raw_pricing_data
} else {
if (private$config$getConfig("simulator")$pricing_data$type %in% "object") {
stop("raw_pricing_data is NULL but configuration specifies object-based inputs")
}
}
if (isTRUE(private$config$getConfig("simulator")$verbose)) {
private$verbose <- TRUE
}
# Set dates
if (is.null(input_dates) & !is.null(raw_input_data)) {
private$input_dates <- unique(sort(raw_input_data$date))
} else {
private$input_dates <- input_dates
}
invisible(self)
},
#' @description Set the verbose flag to control info output.
#' @param verbose Logical flag indicating whether to be verbose or not.
#' @return No return value, called for side effects.
setVerbose = function(verbose) {
stopifnot(is.logical(verbose),
length(verbose) %in% 1)
private$verbose <- verbose
invisible(self)
},
#' @description Set the callback function for updating progress when running
#' a simulation in shiny.
#' @param callback A function suitable for updating a shiny Progress object.
#' It must have two parameters: \code{value}, indicating the progress
#' amount, and detail, and \code{detail}, a text string for display on the
#' progress bar.
#' @return No return value, called for side effects.
setShinyCallback = function(callback) {
if (!is.function(callback)) {
stop("callback must be a function")
}
private$shiny_callback <- callback
invisible(self)
},
#' @description Get security reference information.
#' @return An object of class \code{data.frame} that contains the security
#' reference data for the simulation.
getSecurityReference = function() {
invisible(private$security_reference)
},
#' @description Run the simulation.
#' @return No return value, called for side effects.
run = function() {
if (is.null(private$config)) {
stop("Can not run simulation: no configuration defined.")
}
# Grab simulator section of config
simulator_config <- private$config$getConfig("simulator")
# Input data object
stopifnot(!is.null(simulator_config$input_data$type),
length(simulator_config$input_data$type) %in% 1,
is.character(simulator_config$input_data$type),
simulator_config$input_data$type %in% c("object", "file"))
if (simulator_config$input_data$type %in% "object") {
input_data_obj <- CrossSection$new(TRUE)
input_data_obj$setRaw(private$raw_input_data)
} else if (simulator_config$input_data$type %in% "file") {
input_data_obj <- CrossSectionFromFile$new(TRUE,
simulator_config$input_data$directory,
simulator_config$input_data$prefix)
} else {
stop("Unsupported value for input_data/type")
}
if (!is.null(simulator_config$input_data$na_replace)) {
input_data_obj$setNAReplaceValue(simulator_config$input_data$na_replace)
}
# Pricing data object
stopifnot(!is.null(simulator_config$pricing_data$type),
length(simulator_config$pricing_data$type) %in% 1,
is.character(simulator_config$pricing_data$type),
simulator_config$pricing_data$type %in% c("object", "file"))
if (simulator_config$pricing_data$type %in% "object") {
pricing_data_obj <- CrossSection$new(TRUE)
pricing_data_obj$setRaw(private$raw_pricing_data)
} else if (simulator_config$pricing_data$type %in% "file") {
pricing_data_obj <- CrossSectionFromFile$new(TRUE,
simulator_config$pricing_data$directory,
simulator_config$pricing_data$prefix)
} else {
stop("Unsupported value for pricing_data/type")
}
pricing_data_obj$setColumnMap(simulator_config$pricing_data$columns)
pricing_data_obj$setCarryForwardValue(list(
volume = 0,
adjustment_ratio = 1,
dividend = as.numeric(NA),
distribution = as.numeric(NA)
))
all_strategies <- private$config$getStrategyNames()
portfolio <- Portfolio$new(all_strategies)
all_dates <- self$getSimDates()
for (current_date in as.list(all_dates)) {
if (isTRUE(private$verbose)) {
cat("[", private$config$getConfig("name"), "] Working on ", format(current_date), "\n", sep = "")
}
if (is.function(private$shiny_callback)) {
private$shiny_callback(which(all_dates %in% current_date) / length(all_dates),
paste0("Working on ", format(current_date)))
}
# Retrieve data for the current period.
#
# TODO Impose restrictions on allowable input data column names. For
# example, columns of the form shares_{strategy name} will collide with
# the simulator's work columns. There are other ways around this issue
# but imposing column restrictions is the easiest.
if (!is.null(private$input_dates) && !current_date %in% private$input_dates) {
input_data <- input_data_obj$getCurrent()
} else {
input_data <- input_data_obj$update(current_date)
# Collect metadata as specified in the config (such as correlation of
# values from one period to the next).
if (!is.null(simulator_config$input_data$track_metadata)) {
input_stats <- input_data_obj$periodStats(simulator_config$input_data$track_metadata)
private$saveInputStats(current_date, input_stats)
}
}
if (nrow(input_data) %in% 0) {
stop("Cannot formulate optimization: no input data found")
}
pricing_data <- pricing_data_obj$update(current_date)
# Properties we enforce on input data:
#
# 1. Each security with an entry in input_data must have an entry in
# pricing_data *unless* the simulator is configured to omit inputs
# records that have not yet been priced.
#
# 2. Each security in which there is a non-zero position has an entry in
# pricing_data and input_data.
#
# 3. All securities must be present in the security reference.
stopifnot(
all(input_data$id %in% private$security_reference$id),
all(portfolio$getPositions()$id %in% private$security_reference$id),
all(portfolio$getPositions()$id %in% pricing_data$id),
all(portfolio$getPositions()$id %in% input_data$id)
)
if (!all(input_data$id %in% pricing_data$id)) {
if (!isTRUE(simulator_config$inputs_without_pricing %in% "omit")) {
stop(paste0("Input records (",
sum(!input_data$id %in% pricing_data$id),
") found without pricing data. Consider setting simulator/inputs_without_pricing=omit"))
} else {
if (isTRUE(private$verbose)) {
cat("Omitting ",
sum(!input_data$id %in% pricing_data$id),
" input records without pricing data: ",
paste0(input_data$id[!input_data$id %in% pricing_data$id], collapse = ", "),
"\n")
}
input_data <- filter(input_data, .data$id %in% pricing_data$id)
}
}
input_data <- input_data %>%
rename(inputs_carry_forward = carry_forward)
pricing_data <- pricing_data %>%
select("id",
"close_price", "prior_close_price",
"adjustment_ratio", "volume",
"dividend", "distribution","carry_forward") %>%
rename(pricing_carry_forward = carry_forward)
stopifnot(all(pricing_data$close_price > 0),
all(pricing_data$prior_close_price > 0))
# A further adjustment is required for pricing data that is
# carried-forward: the prior close price must be set to the same value
# as the close price.
#
# TODO Stop requiring that the user passes in the prior close price.
# Once we remove this requirement we can keep track of prior values as
# we iterate over period-by-period cross sections (and can remove logic
# like the below).
pricing_data$prior_close_price <- ifelse(pricing_data$pricing_carry_forward,
pricing_data$close_price,
pricing_data$prior_close_price)
# Create start and end price columns to keep things clear. The start
# price must be adjusted by the adjustment ratio. In the case of a 2:1
# split, the adjustment ratio is 0.5 (number of old shares over number
# of new shares). So, we multiply the previous day's unadjusted price by
# the adjustment ratio to bring yesterday's price in line with today's.
#
# Note that data sanitization (replacing NAs with default values) should
# be moved to CrossSection.
pricing_data <- pricing_data %>%
mutate(start_price = .data$prior_close_price * .data$adjustment_ratio,
end_price = .data$close_price,
volume = replace_na(.data$volume, 0),
adjustment_ratio = replace_na(.data$adjustment_ratio, 1),
dividend = replace_na(.data$dividend, 0),
distribution = replace_na(.data$distribution, 0)
)
portfolio <- portfolio$applyAdjustmentRatio(
select(pricing_data, "id", "adjustment_ratio"))
# Process delistings
# Delistings are handled in the following way:
#
# 1. All securities that have a delisting date that is equal to or
# earlier than current_date are recorded in the id_delisted vector.
# These stocks are marked not investable.
#
# 2. Positions in delisted stocks are recorded in the pos_delisted data
# frame. They are omitted from the day's portfolio optimization.
#
# 3. "Delisting trades" are added as part of the EOD bookkeeping
# sequence so that positions in delisted stocks are flattened. In the
# day's detail data, the logical column 'delisting' indicates that the
# trade was due to a delisting.
#
# TODO For delisted stocks we can go back to our data interfaces and
# ensure data for them is not carried forward after the delisting date.
# This will decrease the size of our inputs by omitting securities we
# know we will never trade again.
# We could save time and use equality on current_date here, but we run
# the risk of having dead securities in the portfolio.
id_delisted <- private$delisting_data$id[private$delisting_data$delisting_date <= current_date]
pos_delisted <- portfolio$getPositions() %>% filter(.data$id %in% id_delisted &
(.data$int_shares != 0 | .data$ext_shares != 0))
pos_delisted <- pos_delisted %>%
left_join(select(pricing_data, "id", "start_price"), by = "id") %>%
left_join(private$delisting_data, by = "id")
if (nrow(pos_delisted) > 0) {
# Record delisting info
private$saveDelistings(current_date, pos_delisted)
}
# Merge together consolidated, wide format positions, pricing data,
# security reference and inputs data for passing to the optimization.
# This merge will create rows with NA volues for stocks that have not
# previously appeared in the data feed. May want to use a special column
# prefix for shares columns (as opposed to "shares_") to avoid name
# conflict with user columns.
#
# Note that the reference price for the optimization is the start_price.
# We could save some cycles by setting price_var = "start_price" in the
# config.
#
# Note the importance of the assertion above that each position appear
# in the input data feed. If the assertion were to be false, then we
# would lose position records in first left join.
#
# TODO slim down security reference so that it only contains columns
# used in category constraints.
input_data <-
left_join(input_data,
portfolio$getConsolidatedPositions(),
by = "id") %>%
left_join(pricing_data,
by = "id") %>%
left_join(private$security_reference,
by = "id") %>%
mutate(ref_price = .data$start_price,
investable = TRUE,
delisting = .data$id %in% pos_delisted$id) %>%
mutate_at(.vars = vars(portfolio$getShareColumns()),
.funs = ~ replace_na(., 0))
# Set investable flag based on
#
# 1. Delisting status
# 2. Universe configuration
#
# TODO Move universe logic into PortOpt class. It's OK to sort out
# delistings in the simulator, but each strategy should be able to have
# its own universe.
input_data$investable <- !input_data$id %in% id_delisted
# By default, presence in the latest set of input data affects
# investability: stocks that are not in the latest update are not
# investable. This makes it easy to define the investable universe
# simply as the stocks present in the input cross-section.
#
# This behavior may not always be desirable, and can be controlled by
# setting the simulator/inputs_define_universe configuration option to
# FALSE. For example, a user may want to pass in updated factor data for
# stocks that are no longer in the universe but may be in the portfolio.
#
# Note that simulator/inputs_define_universe = TRUE is the same as
# having the expression `!inputs_carry_forward` as part of the
# simulator/universe configuration parameter.
#
# Note also that any expression in simulator/universe will be applied
# regardless of the setting of simulator/inputs_define_universe.
# TODO We need to validate config entries and set defaults in the
# StrategyConfig class.
inputs_define_universe <- simulator_config$inputs_define_universe
stopifnot(is.null(inputs_define_universe) || is.logical(inputs_define_universe))
if (is.null(inputs_define_universe) || isTRUE(inputs_define_universe)) {
input_data$investable <- input_data$investable & !input_data$inputs_carry_forward
}
if (!is.null(simulator_config$universe) && length(simulator_config$universe) > 0) {
univ <- eval(rlang::parse_expr(simulator_config$universe), envir = input_data)
if (any(is.na(univ))) {
stop("Found NA universe values")
}
input_data$investable <- input_data$investable & univ
}
# Normalization of input variables and risk factors.
# Perform normalization for variables listed in
# simulator/normalize_in_vars.
#
# The procedure is as follows:
#
# 1. Set variable values for non-investable securities to NA.
# 2. Normalize variable to N(0, 1).
# 3. Replace NAs created in step 1 with 0.
#
# Store raw (pre-normalized) values for column foo in the detail dataset
# column foo_raw.
if (!is.null(simulator_config$normalize_in_vars)) {
for (normalize_var in simulator_config$normalize_in_vars) {
raw_normalize_var <- paste0(normalize_var, "_raw")
input_data <- input_data %>%
mutate(
!! raw_normalize_var := get(normalize_var),
!! normalize_var := replace_na(normalize(ifelse(investable, get(normalize_var), NA)), 0))
}
}
# Perform normalization for variables listed in
# simulator/normalize_factor_vars.
#
# factor_vars (variables used in constraint calculations) are normalized
# as follows:
#
# 1. Set variable values for non-investable securities *in which there
# is no position* to NA.
# 2. Normalize variable to N(0, 1).
# 3. Replace NAs created in step 1 with 0.
#
# The idea is that in_var values should be 0 for stocks that are
# non-investable to encourage exiting, while factor_vars for positions
# in such stocks should be preserved so that accurate exposures can be
# calculated.
if (!is.null(simulator_config$normalize_factor_vars)) {
for (normalize_var in simulator_config$normalize_factor_vars) {
raw_normalize_var <- paste0(normalize_var, "_raw")
# Preserve factor_var value if there is a position in any strategy.
non_zero_pos <- rowSums(abs(input_data[portfolio$getShareColumns()])) != 0
input_data <- input_data %>%
mutate(
!! raw_normalize_var := get(normalize_var),
!! normalize_var := replace_na(normalize(ifelse(investable | non_zero_pos, get(normalize_var), NA)), 0))
}
}
stopifnot(!any(is.na(input_data)))
# Make a copy of input_data to pass to the PortOpt class. We do this to
# accomodate delistings, which will be removed from the portfolio later
# in the process, and which should not contribute to the current day's
# constraint calculations.
opt_input <- input_data
if(nrow(pos_delisted) > 0) {
# Omit positions in stocks that will be exited due to delisting from the
# optimization. Do this by setting their current share levels to 0 in
# the optimization's input data.
opt_input <- opt_input %>%
mutate_at(.vars = portfolio$getShareColumns(),
.funs = ~ replace(., delisting, 0))
}
# Create problem and solve
portOpt <- PortOpt$new(private$config, opt_input)
portOpt$solve()
# Record information on loosened constraints
if (length(portOpt$getLoosenedConstraints()) > 0) {
loosened_df <- data.frame(date = current_date,
constraint_name = names(portOpt$getLoosenedConstraints()),
pct_loosened = 100 * (1 - as.vector(unlist(portOpt$getLoosenedConstraints()))))
private$saveLooseningInfo(current_date, loosened_df)
}
orders <- portOpt$getResultData() %>%
select("id", contains("shares"))
# Handle positions that have grown too large.
#
# After the optimization has generated orders, if the force_trim_factor
# is set, add orders to trim positions that are too large.
#
# A force_trim_factor value of X means "if a position grows larger than
# X times its max position size, force a trade to trim the position back
# to X times its max position size."
#
# So if a security has a max long position size of 10,000, its current
# size is 15,000, and force_trim_factor = 1.2, we add a trade to sell
# 3,000 so that the post-trade position size is 1.2 times max position
# size = 12,000.
#
# TODO Trim positions in PortOpt as part of the optimization process
# (when setting variable limits). We will still want the ability to force
# trim positions, however, since these limits will also be subject to
# loosening when no solution can be found.
if (!is.null(simulator_config$force_trim_factor)) {
force_trim_factor <- simulator_config$force_trim_factor
stopifnot(is.numeric(force_trim_factor),
force_trim_factor >= 1)
vol_var <- private$config$getConfig("vol_var")
too_big <- input_data %>%
filter(!delisting) %>%
select("id", !!vol_var, "start_price",
!!portfolio$getShareColumns()) %>%
pivot_longer(cols = portfolio$getShareColumns(),
names_to = "strategy",
names_prefix = "shares_",
values_to = "shares") %>%
left_join(portOpt$getMaxPosition(), by = c("id", "strategy")) %>%
mutate(pos_nmv = shares * start_price,
max_pos_trim = ifelse(shares > 0, max_pos_lmv, max_pos_smv) *
force_trim_factor) %>%
# Grab those positions that are above the market value trim
# threshold.
filter(abs(shares * start_price) > abs(max_pos_trim)) %>%
# Compute how much can be trimmed
left_join(portOpt$getMaxOrder(), by = c("id", "strategy")) %>%
mutate(
# trim_gmv is the lesser of the amount required to trade the
# position down to max_pos_trim and the maximum order size.
trim_gmv = pmin(abs(pos_nmv) - abs(max_pos_trim), max_order_gmv),
force_order_shares = -1 * sign(shares) * floor(trim_gmv / start_price)) %>%
# Filter out cases where we are less than 1 share away from the
# max
filter(force_order_shares != 0) %>%
select("id", "strategy", "force_order_shares")
if (nrow(too_big) > 0) {
# The orders records (in wide format) for the securities in the
# too_big data frame need to be reconstructed. This is more involved
# because there can be N strategies and M != N strategies that have
# positions over the limit.
#
# It may be more straightforward to do the entire force-trim process
# in a loop where we work on the columns for the strategies' orders
# one at a time, then recalculate the joint-level order.
revised_orders <- orders %>%
filter(id %in% too_big$id) %>%
pivot_longer(cols = paste0("order_", c("shares_joint", portfolio$getShareColumns())),
names_to = "strategy",
names_prefix = "order_shares_",
values_to = "order_shares") %>%
filter(strategy != "joint") %>%
# Left join with data frame that contains force orders
left_join(too_big, by = c("id", "strategy")) %>%
# Revise order_shares if it is smaller than force_order_shares.
#
# Also record whether the signs of order_shares and
# force_order_shares are different. If they are, then the original
# order would represent an increase in the size of the position.
mutate(sign_mismatch = sign(force_order_shares) * sign(order_shares) < 0,
order_shares = ifelse(!is.na(force_order_shares) &
abs(order_shares) < abs(force_order_shares),
force_order_shares,
order_shares))
stopifnot(!any(revised_orders$sign_mismatch, na.rm = TRUE))
revised_orders <- revised_orders %>% select(id, strategy, order_shares)
# Calculate joint level shares.
joint_level <- revised_orders %>%
group_by(id) %>%
summarise(
strategy = "joint",
order_shares = sum(order_shares))
revised_orders <- rbind(revised_orders, joint_level)
# Now pivot back to wide format
revised_orders <- revised_orders %>%
pivot_wider(names_from = "strategy",
values_from = "order_shares",
names_prefix = "order_shares_")
# Adjust column order to match data frame 'orders'
revised_orders <- revised_orders[names(orders)]
# Finally: remove orders generated by the optimization (if any) and
# add the force-exit orders.
orders <- filter(orders, !.data$id %in% revised_orders$id)
orders <- rbind(orders, revised_orders)
}
}
# Handle non-investable securities.
#
# Often due to a changing universe the portfolio will have positions in
# stocks that are not part of the investable universe. Setting the
# simulator/force_exit_non_investable configuration parameter controls
# how these positions are handled. If the
# simulator/force_exit_non_investable parameter is not set, such
# positions are allowed in the portfolio (but increasing their size is
# prohibited in the optimization).
#
# TODO Exit non-investable positions in PortOpt as part of the
# optimization process (when setting variable limits). Like position
# trimming, we will still want the ability to force exit these
# positions, since these limits will also be subject to loosening when
# no solution can be found.
#
# TODO Refactor force-trimming above and force-exiting below to remove
# duplicated code.
if (isTRUE(simulator_config$force_exit_non_investable)) {
vol_var <- private$config$getConfig("vol_var")
non_investable <- input_data %>%
filter(!investable & ! delisting) %>%
select("id", !!vol_var, start_price, !!portfolio$getShareColumns()) %>%
pivot_longer(cols = portfolio$getShareColumns(),
names_to = "strategy",
names_prefix = "shares_",
values_to = "shares") %>%
filter(shares != 0)
if (nrow(non_investable) > 0) {
# Call floor on abs share value to round toward zero (to avoid
# over-sizing in corner cases).
non_investable <- non_investable %>%
left_join(portOpt$getMaxOrder(), by = c("id", "strategy")) %>%
mutate(
pos_nmv = shares * start_price,
exit_gmv = pmin(abs(pos_nmv), max_order_gmv),
order_shares = -1 * sign(shares) * floor(exit_gmv / start_price)) %>%
select("id", "strategy", "order_shares")
# Calculate joint level shares.
joint_level <- non_investable %>%
group_by(id) %>%
summarise(
strategy = "joint",
order_shares = sum(order_shares))
non_investable <- rbind(non_investable, joint_level)
# Now pivot back to wide format
non_investable <- non_investable %>%
pivot_wider(names_from = "strategy",
values_from = "order_shares",
names_prefix = "order_shares_")
# Adjust column order to match data frame 'orders'
non_investable <- non_investable[names(orders)]
# Finally: remove orders generated by the optimization (if any) and
# add the force-exit orders.
orders <- filter(orders, !.data$id %in% non_investable$id)
orders <- rbind(orders, non_investable)
}
}
# Handle delistings
#
# Here we enter delisting trades so that the entire positions in stocks
# that are delisted are removed by EOD. Note that below, fill rate for
# delisting trades will be set to 1 regardless of volume.
if (any(input_data$delisting)) {
delistings <- input_data %>%
filter(delisting) %>%
select("id", !!portfolio$getShareColumns()) %>%
pivot_longer(cols = portfolio$getShareColumns(),
names_to = "strategy",
names_prefix = "shares_",
values_to = "order_shares") %>%
# Completely exit the position
mutate(order_shares = -1 * order_shares) %>%
# Pivot back
pivot_wider(names_from = "strategy",
values_from = "order_shares",
names_prefix = "order_shares_")
# Recalculate order_shares_joint
delistings$order_shares_joint <- rowSums(select(delistings, -"id"))
# Adjust column order to match data frame 'orders'
delistings <- delistings[names(orders)]
orders <- filter(orders, !.data$id %in% delistings$id)
orders <- rbind(orders, delistings)
}
# Now that the order generation process is complete, join orders back to the
# original cross-section.
stopifnot(setequal(orders$id, input_data$id))
input_data <-
inner_join(input_data, orders, by = "id")
# Now for each strategy, separate orders that need to be worked in the
# market from orders that net down with orders from other strategies.
# 0. Pre-compute the maximum number of shares that are available for
# fills, computed as the product of the configured fill rate and the
# day's dollar trading volume.
stopifnot(!is.null(simulator_config$fill_rate_pct_vol),
simulator_config$fill_rate_pct_vol > 0)
input_data$fill_shares_max <-
round(input_data$volume * simulator_config$fill_rate_pct_vol / 100)
# Compute the fill rate ahead of time for each stock. The fill rate is 1
# if the maximum number of shares available (fill_shares_max) is greater
# than the size of the total order across all strategies. Computing this
# value makes fill calculations easier.
input_data$fill_rate <-
ifelse(input_data$order_shares_joint %in% 0 |
abs(input_data$order_shares_joint) <= input_data$fill_shares_max |
input_data$delisting,
1,
input_data$fill_shares_max / abs(input_data$order_shares_joint))
stopifnot(!any(is.na(input_data$fill_shares_max)))
# 1. Compute the total number of shares bought and sold across all
# strategies.
input_data$buy_shares_joint <- input_data$sell_shares_joint <- 0
for (strategy in private$config$getStrategyNames()) {
this_order_shares <- input_data[[paste0("order_shares_", strategy)]]
input_data$buy_shares_joint <- input_data$buy_shares_joint +
ifelse(this_order_shares > 0, this_order_shares, 0)
input_data$sell_shares_joint <- input_data$sell_shares_joint +
ifelse(this_order_shares < 0, this_order_shares, 0)
}
strategy_name_regexp <- paste0(private$config$getStrategyNames(), collapse = "|")
# 2. Using the totals computed in (1), pivot the data to long format to
# calculate fills.
res <- input_data %>%
select("id", "fill_rate",
contains("shares")) %>%
# Is there a cleaner way to select all strategy columns for the pivot?
# Note that the below names_pattern requires that strategy names only
# consist of alpha numeric characters.
pivot_longer(cols = matches(!!strategy_name_regexp),
names_to = c(".value", "strategy"),
names_pattern = paste0("(.*)_(", strategy_name_regexp, ")$")) %>%
left_join(portfolio$getPositions(), by = c("id", "strategy")) %>%
mutate(
# Orders
#
# Calculate shares that must be traded in the market
market_order_shares = as.integer(round(
ifelse(.data$order_shares > 0 & .data$order_shares_joint > 0,
.data$order_shares / .data$buy_shares_joint * .data$order_shares_joint,
ifelse(.data$order_shares < 0 & .data$order_shares_joint < 0,
abs(.data$order_shares / .data$sell_shares_joint) * .data$order_shares_joint,
0)))),
# Shares transferred to other strategies
transfer_order_shares = as.integer(.data$order_shares - .data$market_order_shares),
# Fills
#
# Market fills are subject to market liquidity
market_fill_shares = as.integer(floor(round(.data$market_order_shares * .data$fill_rate))),
# Transfers are filled completely
transfer_fill_shares = as.integer(.data$transfer_order_shares),
int_shares = replace_na(.data$int_shares, 0L),
ext_shares = replace_na(.data$ext_shares, 0L),
# External and internal end positions are computed from market and
# transfer fills
end_int_shares = as.integer(.data$int_shares + .data$transfer_fill_shares),
end_ext_shares = as.integer(.data$ext_shares + .data$market_fill_shares),
# Total fill = market + transfer
fill_shares = .data$transfer_fill_shares + .data$market_fill_shares,
# End number of shares = start + fill
end_shares = .data$shares + .data$fill_shares)
res <- res %>%
select("id", "strategy", "shares",
ends_with("_shares"))
# Roll up strategy data to a "joint" top level with a sum.
joint_data <- res %>%
select(-"strategy") %>%
group_by(.data$id) %>%
summarise_all(sum) %>%
mutate(strategy = "joint")
res <- rbind(res, joint_data)
# Compute market values, costs, and P&L.
stopifnot(
!is.null(simulator_config$transaction_cost_pct),
simulator_config$transaction_cost_pct >= 0,
!is.null(simulator_config$financing_cost_pct),
simulator_config$financing_cost_pct >= 0
)
res <- res %>%
inner_join(select(input_data, "id", "start_price", "end_price", "dividend", "distribution",
"investable", "delisting",
!!simulator_config$add_detail_columns),
by = "id") %>%
mutate(
# Position P&L is computed by comparing the value of the starting
# position to the value of the ending position (including adjustments
# for dividends and distributions).
position_pnl = .data$shares *
(.data$end_price + .data$dividend + .data$distribution - .data$start_price),
# Trading P&L is computed by comparing the end price to the benchmark
# price.
#
# Temporarily: assuming trading at the close, so trading P&L is
# zero.
trading_pnl = 0,
# Apply trade costs
#
# Temporarily: trade costs are a fixed percentage of the notional of the market
# trade (valued at the close). There is no cost to transfer to other
# strategies.
trade_costs = abs(.data$market_fill_shares * .data$end_price) *
simulator_config$transaction_cost_pct / 100,
# Apply financing costs
#
# Temporarily: financing costs are a fixed percentage of the notional of the
# starting value of the portfolio's external positions. External
# positions are positions held on the street and are recorded in the
# ext_shares column.
#
# Use a standard 360-day-count methodology, charging for three days
# on Monday.
financing_costs = abs(.data$ext_shares * .data$start_price) *
simulator_config$financing_cost_pct / 100 / 360 *
ifelse(weekdays(current_date, FALSE) %in% "Monday", 3, 1),
gross_pnl = .data$position_pnl + .data$trading_pnl,
net_pnl = .data$gross_pnl - .data$trade_costs - .data$financing_costs,
# TODO use benchmark price in trade valuations
market_order_gmv = abs(.data$market_order_shares * .data$end_price),
market_fill_nmv = .data$market_fill_shares * .data$end_price,
market_fill_gmv = abs(.data$market_fill_nmv),
transfer_fill_nmv = .data$transfer_fill_shares * .data$end_price,
transfer_fill_gmv = abs(.data$transfer_fill_nmv),
start_nmv = (.data$int_shares + .data$ext_shares) * .data$start_price,
end_nmv = .data$end_shares * .data$end_price,
end_gmv = abs(.data$end_nmv))
# Finish processing delistings
if (any(res$delisting)) {
# Work on rows corresponding to delistings
res_delisting <- filter(res, delisting)
# First make sure the ending position is flat.
if (any(res_delisting$end_int_shares != 0) ||
any(res_delisting$end_ext_shares != 0) ||
any(res_delisting$end_shares != 0)) {
stop("Found non-zero ending position in delisted security")
}
# Bring in delisting_return
res_delisting <- res_delisting %>% left_join(select(pos_delisted, id, delisting_return), by = "id")
if (any(is.na(res_delisting$delisting_return))) {
stop("Found NA delisting return")
}
if (!is.numeric(res_delisting$delisting_return)) {
stop("Delisting return must be numeric")
}
if (any(res_delisting$delisting_return < -1)) {
stop("Delisting return can not be less than -1")
}
# Ensure that all P&L and costs for the delisting are zero. Then set
# gross and net P&L to the delisting return times starting net
# market value.
res_delisting <- res_delisting %>%
mutate(
position_pnl = 0,
trading_pnl = 0,
trade_costs = 0,
financing_costs = 0,
gross_pnl = start_nmv * delisting_return,
net_pnl = gross_pnl
) %>%
select(-delisting_return)
# Replace the original rows with the rows we worked on
res <- filter(res, !delisting)
res <- rbind(res, res_delisting)
}
# Bring in max position information
res <- res %>%
left_join(portOpt$getMaxPosition(), by = c("strategy", "id"))
# if (nrow(filter(res, strategy %in% "joint" & transfer_fill_gmv != 0))) browser()
# Simple sums
summary_sum_data <- res %>%
select("strategy",
ends_with("_nmv"),
ends_with("_gmv"),
ends_with("_pnl"),
ends_with("_costs")) %>%
group_by(.data$strategy) %>%
summarise_all(sum) %>%
ungroup() %>%
mutate(fill_rate_pct = 100 * .data$market_fill_gmv / .data$market_order_gmv)
# More complex aggregation on nmv
summary_addl_data <- res %>%
select("strategy", "end_nmv", "start_nmv", "investable") %>%
group_by(.data$strategy) %>%
summarise(
end_lmv = sum(.data$end_nmv[.data$end_nmv > 0]),
end_smv = sum(.data$end_nmv[.data$end_nmv < 0]),
start_lmv = sum(.data$start_nmv[.data$start_nmv > 0]),
start_smv = sum(.data$start_nmv[.data$start_nmv < 0]),
end_num = sum(.data$end_nmv != 0),
end_num_long = sum(.data$end_nmv > 0),
end_num_short = sum(.data$end_nmv < 0),
num_investable = sum(.data$investable)
)
summary_data <-
inner_join(summary_sum_data,
summary_addl_data, by = "strategy")
# Update positions
portfolio$setPositions(
filter(res, !.data$strategy %in% "joint") %>%
select("id",
"strategy",
"int_shares" = "end_int_shares",
"ext_shares" = "end_ext_shares"))
# Calculate EOD exposures.
#
# First prepare data. Columns for category grouping are static and live
# in this object's security_reference member. Factor data is different
# each day and can be found in 'input_data'.
#
# TODO automatically calculate exposures for all categories and factors
# used in constraints.
category_vars <- simulator_config$calculate_exposures$category_vars
factor_vars <- simulator_config$calculate_exposures$factor_vars
if (!is.null(category_vars) || !is.null(factor_vars)) {
exposures_input <-
res %>%
select("strategy", "id", "end_nmv", "end_gmv") %>%
left_join(select(input_data, "id", one_of(!!factor_vars), one_of(!!category_vars)),
by = "id") %>%
mutate(end_lmv = ifelse(end_nmv > 0, end_nmv, 0),
end_smv = ifelse(end_nmv < 0, end_nmv, 0))
# Save net exposures
exposures <- calculate_exposures(detail_df = exposures_input,
in_var = "end_nmv",
weight_divisor = private$getStrategyCapital(),
category_vars = category_vars,
factor_vars = factor_vars)
private$saveExposures(current_date, exposures, type = "net")
# Save long exposures
exposures <- calculate_exposures(detail_df = exposures_input,
in_var = "end_lmv",
weight_divisor = private$getStrategyCapital(),
category_vars = category_vars,
factor_vars = factor_vars)
private$saveExposures(current_date, exposures, type = "long")
# Save short exposures
exposures <- calculate_exposures(detail_df = exposures_input,
in_var = "end_smv",
weight_divisor = private$getStrategyCapital(),
category_vars = category_vars,
factor_vars = factor_vars)
private$saveExposures(current_date, exposures, type = "short")
# Save gross exposures
exposures <- calculate_exposures(detail_df = exposures_input,
in_var = "end_gmv",
weight_divisor = private$getStrategyCapital(),
category_vars = category_vars,
factor_vars = factor_vars)
private$saveExposures(current_date, exposures, type = "gross")
}
# Save sim summary, sim detail, and optimization data.
private$saveSimSummary(current_date, summary_data)
# TODO Add flags that control saving of other data. Right now we can
# only turn saving detail off and on.
if (is.null(simulator_config$skip_saving) ||
!"detail" %in% simulator_config$skip_saving) {
# The full detail dataset is large. To save specific columns, use the
# simulator/keep_detail_columns parameter.
if (!is.null(simulator_config$keep_detail_columns)) {
keep_detail_columns <-
unique(c("id", "strategy",
simulator_config$keep_detail_columns))
res <- res %>% select(!!keep_detail_columns)
}
# The 'keep_detail_all_rows' flag controls whether to save all detail
# rows, or only rows where there is a holding or trade for the stock.
# It defaults to FALSE.
keep_detail_all_rows <- FALSE
if (!is.null(simulator_config$keep_detail_all_rows) &&
is.logical(simulator_config$keep_detail_all_rows)) {
keep_detail_all_rows <- simulator_config$keep_detail_all_rows
}
if (!isTRUE(keep_detail_all_rows)) {
res <- res %>% filter(start_nmv != 0 | end_nmv != 0)
}
private$saveSimDetail(current_date, res)
}
private$saveOptimizationSummary(current_date, portOpt$summaryDf())
}
invisible(self)
},
#' @description Get a list of all date for the simulation.
#' @return A vector of class \code{Date} over which the simulation currently iterates: all
#' weekdays between the 'from' and 'to' dates in the simulation's config.
getSimDates = function() {
from <- as.Date(private$config$getConfig("from"))
to <- as.Date(private$config$getConfig("to"))
if (to < from) {
stop("to date must not be earlier than from date")
}
# The simulator currently iterates over each weekday between the dates
# 'from' and 'to'.
all_dates <- seq(from = from, to = to, by = "1 day")
all_dates <- all_dates[!weekdays(all_dates, abbreviate = FALSE) %in% c("Saturday", "Sunday")]
# Dates can be omitted using the omit_dates simulator config option.
omit_dates <- private$config$getConfig("simulator")$omit_dates
if (length(omit_dates) > 0) {
omit_dates <- as.Date(omit_dates)
all_dates <- all_dates[!all_dates %in% omit_dates]
}
all_dates
},
#' @description Get summary information.
#' @param strategy_name Character vector of length 1 that specifies the
#' strategy for which to get detail data. If \code{NULL} data for all
#' strategies is returned. Defaults to \code{NULL}.
#' @return An object of class \code{data.frame} that contains summary data
#' for the simulation, by period, at the joint and strategy level. The data
#' frame contains the following columns:
#' \describe{
#' \item{strategy}{Strategy name, or 'joint' for the aggregate strategy.}
#' \item{sim_date}{Date of the summary data.}
#' \item{market_fill_nmv}{Total net market value of fills that do not
#' net down across strategies.}
#' \item{transfer_fill_nmv}{Total net market value of fills that
#' represent "internal transfers", i.e., fills in one strategy that net
#' down with fills in another. Note that at the joint level this column
#' by definition is 0.}
#' \item{market_order_gmv}{Total gross market value of orders that do not
#' net down across strategies.}
#' \item{market_fill_gmv}{Total gross market value of fills that do not
#' net down across strategies.}
#' \item{transfer_fill_gmv}{Total gross market value of fills that
#' represent "internal transfers", i.e., fills in one strategy that net
#' down with fills in another.}
#' \item{start_nmv}{Total net market value of all positions at the start
#' of the period.}
#' \item{start_lmv}{Total net market value of all long positions at the
#' start of the period.}
#' \item{start_smv}{Total net market value of all short positions at the
#' start of the period.}
#' \item{end_nmv}{Total net market value of all positions at the end of
#' the period.}
#' \item{end_gmv}{Total gross market value of all positions at the end
#' of the period.}
#' \item{end_lmv}{Total net market value of all long positions at the
#' end of the period.}
#' \item{end_smv}{Total net market value of all short positions at the
#' end of the period.}
#' \item{end_num}{Total number of positions at the end of the period.}
#' \item{end_num_long}{Total number of long positions at the end of the
#' period.}
#' \item{end_num_short}{Total number of short positions at the end of
#' the period.}
#' \item{position_pnl}{The total difference between the end and start
#' market value of positions.}
#' \item{trading_pnl}{The total difference between the market value of
#' trades at the benchmark price and at the end price. Note: currently
#' assuming benchmark price is the closing price, so trading P&L is
#' zero.}
#' \item{gross_pnl}{Total P&L gross of costs, calculated as position_pnl
#' + trading_pnl.}
#' \item{trade_costs}{Total trade costs (slippage).}
#' \item{financing_costs}{Total financing/borrow costs.}
#' \item{net_pnl}{Total P&L net of costs, calculated as gross_pnl -
#' trade_costs - financing_costs.}
#' \item{fill_rate_pct}{Total fill rate across all market orders,
#' calculated as 100 * market_fill_gmv / market_order_gmv.}
#' \item{num_investable}{Number of investable securities (size of universe).}
#'
#' }
#'
getSimSummary = function(strategy_name = NULL) {
res <- bind_rows(private$sim_summary_list)
if (!is.null(strategy_name)) {
res <- filter(res, .data$strategy %in% !!strategy_name)
}
invisible(res)
},
#' @description Get detail information.
#' @param sim_date Vector of length 1 of class Date or character that
#' specifies the period for which to get detail information. If
#' \code{NULL} then data from all periods is returned. Defaults
#' to \code{NULL}.
#' @param strategy_name Character vector of length 1 that specifies the
#' strategy for which to get detail data. If \code{NULL} data for all
#' strategies is returned. Defaults to \code{NULL}.
#' @param security_id Character vector of length 1 that specifies the
#' security for which to get detail data. If \code{NULL} data for all
#' securities is returned. Defaults to \code{NULL}.
#' @param columns Vector of class character specifying the columns to
#' return. This parameter can be useful when dealing with very large
#' detail datasets.
#' @return An object of class \code{data.frame} that contains security-level
#' detail data for the simulation for the desired strategies, securities,
#' dates, and columns. Available columns include:
#' \describe{
#' \item{id}{Security identifier.}
#' \item{strategy}{Strategy name, or 'joint' for the aggregate strategy.}
#' \item{sim_date}{Date to which the data pertains.}
#' \item{shares}{Shares at the start of the period.}
#' \item{int_shares}{Shares at the start of the period that net down
#' with positions in other strategies.}
#' \item{ext_shares}{Shares at the start of the period that do not net
#' down with positions in other strategies.}
#' \item{order_shares}{Order, in shares.}
#' \item{market_order_shares}{Order that does not net down with orders
#' in other strategies, in shares.}
#' \item{transfer_order_shares}{Order that nets down with orders in
#' other strategies, in shares.}
#' \item{fill_shares}{Fill, in shares.}
#' \item{market_fill_shares}{Fill that does not net down with fills in
#' other strategies, in shares.}
#' \item{transfer_fill_shares}{Fill that nets down with fills in other
#' strategies, in shares.}
#' \item{end_shares}{Shares at the end of the period.}
#' \item{end_int_shares}{Shares at the end of the period that net down
#' with positions in other strategies.}
#' \item{end_ext_shares}{Shares at the end of the period that do not net
#' down with positions in other strategies.}
#' \item{start_price}{Price for the security at the beginning of the
#' period.}
#' \item{end_price}{Price for the security at the end of the period.}
#' \item{dividend}{Dividend for the security, if any, for the
#' period.}
#' \item{distribution}{Distribution (e.g., spin-off) for the security, if
#' any, for the period.}
#' \item{investable}{Logical indicating whether the security is part of
#' the investable universe. The value of the flag is set to TRUE if the
#' security has not been delisted and satisfies the universe criterion
#' provided (if any) in the \code{simulator/universe} configuration
#' option.}
#' \item{delisting}{Logical indicating whether a position in the
#' security was removed due to delisting. If delisting is set to TRUE,
#' the gross_pnl and net_pnl columns will contain the P&L
#' due to delisting, if any. P&L due to delisting is calculated as the
#' delisting return times the \code{start_nmv} of the position.}
#' \item{position_pnl}{Position P&L, calculated as shares * (end_price +
#' dividend + distribution - start_price)}
#' \item{trading_pnl}{The difference between the market value of
#' trades at the benchmark price and at the end price. Note: currently
#' assuming benchmark price is the closing price, so trading P&L is
#' zero.}
#' \item{trade_costs}{Trade costs, calculated as a fixed percentage (set
#' in the simulation configuration) of the notional of the market trade
#' (valued at the close).}
#' \item{financing_costs}{Financing cost for the position, calculated as
#' a fixed percentage (set in the simulation configuration) of the
#' notional of the starting value of the portfolio's external positions.
#' External positions are positions held on the street and are recorded
#' in the ext_shares column.}
#' \item{gross_pnl}{Gross P&L, calculated as position_pnl + trading_pnl.}
#' \item{net_pnl}{Net P&L, calculated as gross_pnl - trade_costs -
#' financing_costs.}
#' \item{market_order_nmv}{Net market value of the order that does not
#' net down with orders in other strategies.}
#' \item{market_fill_gmv}{Gross market value of the order that does not
#' net down with orders in other strategies.}
#' \item{market_fill_nmv}{Net market value of the fill that does not net
#' down with orders in other strategies.}
#' \item{market_fill_gmv}{Gross market value of the fill that does not
#' net down with orders in other strategies.}
#' \item{transfer_fill_nmv}{Net market value of the fill that nets down
#' with fills in other strategies.}
#' \item{transfer_fill_gmv}{Gross market value of the fill that nets down
#' with fills in other strategies.}
#' \item{start_nmv}{Net market value of the position at the start of the
#' period.}
#' \item{end_nmv}{Net market value of the position at the end of the
#' period.}
#' \item{end_gmv}{Gross market value of the position at the end of the
#' period.}
#'
#' }
getSimDetail = function(sim_date = NULL,
strategy_name = NULL,
security_id = NULL,
columns = NULL) {
if (!is.null(sim_date)) {
detail_data <- private$sim_detail_list[[sim_date]]
} else {
detail_data <- bind_rows(private$sim_detail_list)
}
if (!is.null(strategy_name)) {
detail_data <- detail_data %>% filter(.data$strategy %in% !!strategy_name)
}
if (!is.null(security_id)) {
detail_data <- detail_data %>% filter(.data$id %in% !!security_id)
}
if (!is.null(columns)) {
detail_data <- detail_data %>% select(!!columns)
}
invisible(detail_data)
},
#' @description Get summary information by security. This method can be
#' used, for example, to calculate the biggest winners and losers over the
#' course of the simulation.
#' @param strategy_name Character vector of length 1 that specifies the
#' strategy for which to get detail data. If \code{NULL} data for all
#' strategies is returned. Defaults to \code{NULL}.
#' @return An object of class \code{data.frame} that contains summary
#' information aggregated by security. The data frame contains the
#' following columns:
#' \describe{
#' \item{id}{Security identifier.}
#' \item{strategy}{Strategy name, or 'joint' for the aggregate
#' strategy.}
#' \item{gross_pnl}{Gross P&L for the position over the entire
#' simulation.}
#' \item{gross_pnl}{Net P&L for the position over the entire
#' simulation.}
#' \item{average_market_value}{Average net market value of the
#' position over days in the simulation where the position was not
#' flat.}
#' \item{total_trading}{Total gross market value of trades for the
#' security.}
#' \item{trade_costs}{Total cost of trades for the security over the
#' entire simulation.}
#' \item{trade_costs}{Total cost of financing for the position over the
#' entire simulation.}
#' \item{days_in_portfolio}{Total number of days there was a position in
#' the security in the portfolio over the entire simulation.}
#' }
getPositionSummary = function(strategy_name = NULL) {
detail_data <- bind_rows(private$sim_detail_list)
if (!is.null(strategy_name)) {
detail_data <- detail_data %>% filter(.data$strategy %in% !!strategy_name)
}
detail_data %>%
group_by(.data$id, .data$strategy) %>%
summarise(gross_pnl = round(sum(gross_pnl)),
net_pnl = round(sum(net_pnl)),
average_market_value = round(mean(end_nmv[end_shares != 0])),
total_trading = round(sum(market_fill_gmv)),
trade_costs = round(sum(trade_costs)),
financing_costs = round(sum(financing_costs)),
days_in_portfolio = sum(end_shares != 0)) %>%
arrange(-.data$gross_pnl)
},
#' @description Get input statistics.
#' @return An object of class \code{data.frame} that contains statistics on
#' select columns of input data. Statistics are tracked for the columns
#' listed in the configuration variable
#' \code{simulator/input_data/track_metadata}. The data frame contains the
#' following columns:
#' \describe{
#' \item{period}{Period to which statistics pertain.}
#' \item{input_rows}{Total number of rows of input data, including
#' rows carried forward from the previous period.}
#' \item{cf_rows}{Total number of rows carried forward from the previous
#' period.}
#' \item{num_na_\emph{column}}{Number of NA values in \emph{column}. This
#' measure appears for each element of \code{track_metadata}.}
#' \item{cor_\emph{column}}{Period-over-period correlation for \emph{column}.
#' This measure appears for each element of \code{track_metadata}.}
#' }
getInputStats = function() {
invisible(bind_rows(private$input_stats_list))
},
#' @description Get loosening information.
#' @return An object of class \code{data.frame} that contains, for each
#' period, which constraints were loosened in order to solve the portfolio
#' optimization problem, if any. The data frame contains the
#' following columns:
#' \describe{
#' \item{date}{Date for which the constraint was loosened.}
#' \item{constraint_name}{Name of the constraint that was loosened.}
#' \item{pct_loosened}{Percentage by which the constraint was loosened,
#' where 100 means loosened fully (i.e., the constraint is effectively
#' removed).}
#' }
getLooseningInfo = function() {
invisible(bind_rows(private$loosening_info_list))
},
#' @description Get optimization summary information.
#' @return An object of class \code{data.frame} that contains optimization
#' summary information, such as starting and ending factor constraint
#' values, at the strategy and joint level. The data frame contains the
#' following columns:
#' \describe{
#' \item{strategy}{Strategy name, or 'joint' for the aggregate strategy.}
#' \item{sim_date}{Date to which the data pertains.}
#' \item{order_gmv}{Total gross market value of orders generated by the
#' optimization.}
#' \item{start_smv}{Total net market value of short positions at the
#' start of the optimization.}
#' \item{start_lmv}{Total net market value of long positions at the
#' start of the optimization.}
#' \item{end_smv}{Total net market value of short positions at the end
#' of the optimization.}
#' \item{end_lmv}{Total net market value of long positions at the end of
#' the optimization.}
#' \item{start_\emph{factor}}{Total net exposure to \emph{factor} at the
#' start of the optimization, for each factor constraint.}
#' \item{end_\emph{factor}}{Total net exposure to \emph{factor} at the
#' start of the optimization, for each factor constraint.}
#' }
getOptimizationSummary = function() {
invisible(bind_rows(private$optimization_summary_list))
},
#' @description Get end-of-period exposure information.
#' @param type Vector of length 1 that may be one of \code{"net"},
#' \code{"long"}, \code{"short"}, and \code{"gross"}.
#' @return An object of class \code{data.frame} that contains end-of-period
#' exposure information for the simulation portfolio. The units of the
#' exposures are portfolio weight relative to strategy_captial (i.e., net
#' market value of exposure divided by strategy capital). The data frame
#' contains the following columns:
#' \describe{
#' \item{strategy}{Strategy name, or 'joint' for the aggregate strategy.}
#' \item{sim_date}{Date of the exposure data.}
#' \item{\emph{category}_\emph{level}}{Exposure to \emph{level}
#' within \emph{category}, for all levels of all category constraints, at the end
#' of the period.}
#' \item{\emph{factor}}{Exposure to \emph{factor}, for all factor
#' constraints, at the end of the period.}
#' }
getExposures = function(type = "net") {
stopifnot(type %in% c("net", "long", "short", "gross"))
invisible(bind_rows(private$exposures_list[[type]]))
},
#' @description Get information on positions removed due to delisting.
#' @return An object of class \code{data.frame} that contains a row for each
#' position that is removed from the simulation portfolio due to a
#' delisting. Each row contains the size of the position on the day on
#' which it was removed from the portfolio.
getDelistings = function() {
invisible(bind_rows(private$delistings_list))
},
#' @description Get summary information for a single strategy suitable for
#' plotting input.
#' @param strategy_name Strategy for which to return summary data.
#' @param include_zero_row Logical flag indicatiing whether to prepend a row
#' to the summary data with starting values at zero. Defaults to \code{TRUE}.
#' @return A data frame that contains summary information for the desired
#' strategy, as well as columns for cumulative net and gross total return,
#' calculated as pnl divided by ending gross market value.
getSingleStrategySummaryDf = function(strategy_name = "joint", include_zero_row = TRUE) {
res <- filter(self$getSimSummary(), .data$strategy %in% !!strategy_name)
if (isTRUE(include_zero_row)) {
# Create a zero-value starting row with date lagged by one day.
res <- res[c(1, 1:nrow(res)),]
res$sim_date[1] <- res$sim_date[1] - 1
res[1,] <- mutate_if(res[1,], is.numeric, ~ 0)
}
# Compute returns as a percentage GMV, by day and cumulative.
#
# These calculations should probably be done up front in the simulation
# loop and saved in the summary dataset.
mutate(res,
net_ret = ifelse(end_gmv %in% 0, 0, .data$net_pnl / .data$end_gmv),
net_cum_ret = cumsum(net_ret),
gross_ret = ifelse(end_gmv %in% 0, 0, .data$gross_pnl / .data$end_gmv),
gross_cum_ret = cumsum(gross_ret))
},
#' @description Draw a plot of cumulative gross and net return by date.
#' @param strategy_name Character vector of length 1 specifying the strategy
#' for the plot. Defaults to \code{"joint"}.
plotPerformance = function(strategy_name = "joint") {
self$getSingleStrategySummaryDf(strategy_name) %>%
select("sim_date", "gross_cum_ret", "net_cum_ret") %>%
rename(Gross = "gross_cum_ret", Net = "net_cum_ret") %>%
pivot_longer(
-"sim_date",
names_to = "type",
values_to = "cum_ret"
) %>%
ggplot(aes(x = sim_date, y = 100 * cum_ret, color = type, group = type)) + geom_line() +
xlab("Date") + ylab("Return (%)") +
ggtitle("Cumulative Return (% GMV)") +
theme_light() +
theme(
plot.background = element_rect(fill = NA, colour = NA),
plot.title = element_text(size = 18),
axis.text = element_text(size = 10),
axis.text.x = element_text(angle = 0),
legend.title = element_blank())
},
#' @description Draw a plot of contribution to net return on GMV for levels
#' of a specified category.
#' @param category_var Plot performance contribution for the levels of
#' \code{category_var}. \code{category_var} must be present in the
#' simulation's security reference, and detail data must be present in the
#' object's result data.
#' @param strategy_name Character vector of length 1 specifying the strategy
#' for the plot. Defaults to \code{"joint"}.
plotContribution = function(category_var, strategy_name = "joint") {
stopifnot(length(strategy_name) %in% 1)
summary_data <- self$getSimSummary(strategy_name) %>%
select(sim_date, end_gmv)
# Get the category var values from the security reference if possible.
# Otherwise look for them in the detail result data.
if (category_var %in% names(self$getSecurityReference())) {
contrib_data <- self$getSimDetail(strategy_name = strategy_name,
columns = c("id", "sim_date", "net_pnl")) %>%
left_join(select(self$getSecurityReference(), "id", !!category_var), by = "id")
} else {
contrib_data <- self$getSimDetail(strategy_name = strategy_name,
columns = c("id", "sim_date", "net_pnl", category_var))
}
contrib_data <- contrib_data %>%
group_by_at(c("sim_date", category_var)) %>%
summarise(net_pnl = sum(net_pnl)) %>%
ungroup() %>%
left_join(summary_data, by = c("sim_date")) %>%
group_by_at(category_var) %>%
mutate(net_ret = net_pnl / end_gmv,
cum_net_ret = cumsum(net_ret)) %>%
ungroup()
# Adding zero-row for each group
zero_rows <- contrib_data %>% filter(!duplicated(get(category_var))) %>%
mutate(sim_date = sim_date - 1) %>%
mutate_if(is.numeric, ~ 0)
contrib_data <- rbind(zero_rows, contrib_data)
contrib_data %>%
ggplot(aes(x = sim_date, y = 100 * cum_net_ret, color = get(category_var), group = get(category_var))) +
geom_line() +
xlab("Date") + ylab("Contribution (%)") +
ggtitle(paste0(category_var, " Contribution to Net Return (% GMV)")) +
theme_light() +
theme(
plot.background = element_rect(fill = NA, colour = NA),
plot.title = element_text(size = 18),
axis.text = element_text(size = 10),
axis.text.x = element_text(angle = 0),
legend.title = element_blank())
},
#' @description Draw a plot of total gross, long, short, and net market
#' value by date.
#' @param strategy_name Character vector of length 1 specifying the strategy
#' for the plot. Defaults to \code{"joint"}.
plotMarketValue = function(strategy_name = "joint") {
if (!strategy_name %in% c("joint", private$config$getStrategyNames())) {
stop(paste0("Invalid strategy name: ", strategy_name))
}
mv_plot_df <- select(self$getSingleStrategySummaryDf(strategy_name),
sim_date, GMV = end_gmv, LMV = end_lmv, SMV = end_smv, NMV = end_nmv) %>%
gather(type, value, GMV:NMV) %>%
filter(!is.na(value))
mv_plot_df$type <- factor(mv_plot_df$type, levels = c("GMV", "LMV", "NMV", "SMV"))
ggplot(mv_plot_df, aes(sim_date, value/1e6, color = type, group = type)) + geom_line() +
scale_color_manual(values = c("LMV" = "darkgreen", "SMV" = "darkred",
"GMV" = "dodgerblue1", "NMV" = "black")) +
xlab("Date") +
ylab("Market Value ($mm)") +
ggtitle("Market Values") +
theme_light() +
theme(
plot.background = element_rect(fill = NA, colour = NA),
plot.title = element_text(size = 18),
axis.text = element_text(size = 10),
axis.text.x = element_text(angle = 0),
legend.title = element_blank())
},
#' @description Draw a plot of exposure to all levels in a category by date.
#' @param in_var Category for which exposures are plotted. In order to plot
#' exposures for category \code{in_var}, we must have run the simulation
#' with \code{in_var} in the config setting
#' \code{simulator/calculate_exposures/category_vars}.
#' @param strategy_name Character vector of length 1 specifying the strategy
#' for the plot. Defaults to \code{"joint"}.
plotCategoryExposure = function(in_var, strategy_name = "joint") {
exposures <- self$getExposures() %>% filter(strategy %in% strategy_name) %>%
select("sim_date", starts_with(in_var))
# Make sure that there is at least one level for in_var present in the
# object's exposure data. If only sim_date is present, then most likely
# exposure to in_var was not calculated by the simulation.
if (ncol(exposures) %in% 1) {
stop(paste0("No exposure data found for in_var: ", in_var))
}
exposures %>%
pivot_longer(-"sim_date",
names_to = in_var,
names_prefix = paste0(in_var, "_"),
values_to = "exposure") %>%
ggplot(aes(sim_date, exposure * 100, color = get(in_var), group = get(in_var))) + geom_line() +
xlab("Date") +
ylab("Exposure (%)") +
ggtitle(paste0(in_var, " Exposure (% Capital)")) +
theme_light() +
theme(
plot.background = element_rect(fill = NA, colour = NA),
plot.title = element_text(size = 18),
axis.text = element_text(size = 10),
axis.text.x = element_text(angle = 0),
legend.title = element_blank())
},
#' @description Draw a plot of exposure to factors by date.
#' @param in_var Factors for which exposures are plotted.
#' @param strategy_name Character vector of length 1 specifying the strategy
#' for the plot. Defaults to \code{"joint"}.
plotFactorExposure = function(in_var, strategy_name = "joint") {
exposures <- self$getExposures() %>% filter(strategy %in% strategy_name)
exposures %>%
select("sim_date", one_of(!!in_var)) %>%
pivot_longer(-"sim_date",
names_to = "factor_name",
values_to = "exposure") %>%
ggplot(aes(sim_date, exposure * 100, color = factor_name, group = factor_name)) + geom_line() +
xlab("Date") +
ylab("Exposure (%)") +
ggtitle("Factor Exposure (% Capital)") +
theme_light() +
theme(
plot.background = element_rect(fill = NA, colour = NA),
plot.title = element_text(size = 18),
axis.text = element_text(size = 10),
axis.text.x = element_text(angle = 0),
legend.title = element_blank())
},
#' @description Draw a plot of number of long and short positions by date.
#' @param strategy_name Character vector of length 1 specifying the strategy
#' for the plot. Defaults to \code{"joint"}.
plotNumPositions = function(strategy_name = "joint") {
self$getSingleStrategySummaryDf(strategy_name) %>%
select("sim_date", "end_num_long", "end_num_short") %>%
rename(Long = "end_num_long", Short = "end_num_short") %>%
pivot_longer(
-"sim_date",
names_to = "side",
values_to = "count"
) %>%
ggplot(aes(x = sim_date, y = count, color = side, group = side)) + geom_line() +
xlab("Date") + ylab("Number of Positions") +
ggtitle("Number of Positions by Side") +
theme_light() +
theme(
plot.background = element_rect(fill = NA, colour = NA),
plot.title = element_text(size = 18),
axis.text = element_text(size = 10),
axis.text.x = element_text(angle = 0),
legend.title = element_blank())
},
#' @description Draw a plot of number of long and short positions by date.
#' @param strategy_name Character vector of length 1 specifying the strategy
#' for the plot. Defaults to \code{"joint"}.
plotTurnover = function(strategy_name = "joint") {
self$getSimSummary(strategy_name) %>%
select("sim_date", "market_fill_gmv") %>%
ggplot(aes(x = sim_date, y = market_fill_gmv)) + geom_bar(stat = "identity") +
xlab("Date") + ylab("Traded GMV ($)") +
ggtitle("Turnover") +
theme_light() +
theme(
plot.background = element_rect(fill = NA, colour = NA),
plot.title = element_text(size = 18),
axis.text = element_text(size = 10),
axis.text.x = element_text(angle = 0),
legend.title = element_blank())
},
#' @description Draw a plot of the universe size, or number of investable
#' stocks, over time.
#' @param strategy_name Character vector of length 1 specifying the strategy
#' for the plot. Defaults to \code{joint}.
plotUniverseSize = function(strategy_name = "joint") {
investable_data <- self$getSimSummary(strategy_name = strategy_name)
investable_data %>%
ggplot(aes(x = sim_date, y = num_investable)) + geom_line() +
xlab("Date") + ylab("Number of securities") +
ggtitle("Universe size") +
theme_light() +
theme(
plot.background = element_rect(fill = NA, colour = NA),
plot.title = element_text(size = 18),
axis.text = element_text(size = 10),
axis.text.x = element_text(angle = 0),
legend.title = element_blank())
},
#' @description Draw a plot of the percentage of portfolio GMV held in
#' non-investable stocks (e.g., stocks that do not satisfy universe criteria)
#' for a given strategy. Note that this plot requires detail data.
#' @param strategy_name Character vector of length 1 specifying the strategy
#' for the plot. Defaults to \code{"joint"}.
plotNonInvestablePct = function(strategy_name = "joint") {
investable_data <- self$getSimDetail(strategy_name = strategy_name,
columns = c("sim_date", "id","end_gmv", "investable"))
investable_data %>%
group_by(sim_date) %>%
summarise(pct_non_investable = 100 * (1 - sum(end_gmv[investable]) / sum(end_gmv))) %>%
ggplot(aes(x = sim_date, y = pct_non_investable)) + geom_line() +
xlab("Date") + ylab("Percentage GMV") +
ggtitle("Percentage of GMV in\nnon-investable securites") +
theme_light() +
theme(
plot.background = element_rect(fill = NA, colour = NA),
plot.title = element_text(size = 18),
axis.text = element_text(size = 10),
axis.text.x = element_text(angle = 0),
legend.title = element_blank())
},
#' @description Calculate overall simulation summary statistics, such as
#' total P&L, Sharpe, average market values and counts, etc.
#' @return A data frame that contains summary statistics, suitable for
#' reporting.
overallStatsDf = function() {
res <- self$getSingleStrategySummaryDf("joint", include_zero_row = FALSE)
data.frame(
Item = c(
"Total P&L",
"Total Return on GMV (%)",
"Annualized Return on GMV (%)",
"Annualized Vol (%)",
"Annualized Sharpe",
"Max Drawdown (%)",
"Avg GMV",
"Avg NMV",
"Avg Count",
"Avg Daily Turnover",
"Holding Period (months)"
),
Gross = c(
formatC(sum(res$gross_pnl), big.mark = ",", digit = 0, format = "f"),
sprintf("%0.1f", sum(res$gross_pnl / res$end_gmv) * 100),
sprintf("%0.1f", mean(res$gross_pnl / res$end_gmv) * 100 * 252),
sprintf("%0.1f", sd(res$gross_pnl / res$end_gmv) * 100 * sqrt(252)),
sprintf("%0.2f", mean(res$gross_pnl / res$end_gmv) / sd(res$gross_pnl / res$end_gmv) * sqrt(252)),
sprintf("%0.1f", drawdown(res$gross_pnl / res$end_gmv) * 100),
rep("", 5)
),
Net = c(
formatC(sum(res$net_pnl), big.mark = ",", digit = 0, format = "f"),
sprintf("%0.1f", sum(res$net_pnl / res$end_gmv) * 100),
sprintf("%0.1f", mean(res$net_pnl / res$end_gmv) * 100 * 252),
sprintf("%0.1f", sd(res$net_pnl / res$end_gmv) * 100 * sqrt(252)),
sprintf("%0.2f", mean(res$net_pnl / res$end_gmv) / sd(res$net_pnl / res$end_gmv) * sqrt(252)),
sprintf("%0.1f", drawdown(res$net_pnl / res$end_gmv) * 100),
formatC(mean(res$end_gmv), big.mark = ",", digit = 0, format = "f"),
formatC(mean(res$end_nmv), big.mark = ",", digit = 0, format = "f"),
formatC(mean(res$end_num), big.mark = ",", digit = 0, format = "f"),
formatC(mean(res$market_fill_gmv), big.mark = ",", digit = 0, format = "f"),
sprintf("%0.1f", 12 / (mean(res$market_fill_gmv) / mean(res$end_gmv) * 252 / 2))
), stringsAsFactors = FALSE)
},
#' @description Calculate return for each month and summary statistics for
#' each year, such as total return and annualized Sharpe. Return in data
#' frame format suitable for reporting.
#' @return The data frame contains one row for each calendar year in the
#' simulation, and up to seventeen columns: one column for year, one
#' column for each calendar month, and columns for the year's total
#' return, annualized return, annualized volatility, and annualized
#' Sharpe. Total return is the sum of daily net returns. Annualized return
#' is the mean net return times 252. Annualized volatility is the standard
#' deviation of net return times the square root of 252. Annualized Sharpe
#' is the ratio of annualized return to annualized volatility. All returns
#' are in percent.
overallReturnsByMonthDf = function() {
res <- self$getSingleStrategySummaryDf("joint", include_zero_row = FALSE)
# Group by month and calculate return.
month_ret_long <- group_by(res, date = as.Date(paste0(format(sim_date, "%Y-%m"), "-01"))) %>%
summarise(ret = sum(net_ret) * 100) %>%
ungroup() %>%
transmute(year = lubridate::year(date),
month = lubridate::month(date),
ret)
# Organize monthly returns into rows, one for each year.
month_ret_yearly <- month_ret_long %>%
tidyr::spread(month, ret)
# Compute summary statistics for each year.
stats_yearly <- group_by(res, year = lubridate::year(sim_date)) %>%
summarise(
total_ret = 100 *sum(net_ret),
ann_ret = 100 * mean(net_ret) * 252,
ann_vol = 100 * sd(net_ret) * sqrt(252)) %>%
mutate(ann_sr = ann_ret / ann_vol)
# Add stats to monthly returns.
month_ret_yearly <- month_ret_yearly %>%
left_join(stats_yearly, by = "year")
month_cols <- names(month_ret_yearly)[names(month_ret_yearly) %in% as.character(1:12)]
# Replace numerical month strings with month abbreviation. Do it this way
# to handle cases where there are fewer than 12 months.
month_ret_yearly %>%
rename_at(month_cols, ~ month.abb[as.numeric(.x)])
},
#' @description Print overall simulation statistics.
print = function() {
if (is.null(private$sim_summary_list)) {
print("Simulation object with no result data.")
} else {
print(self$overallStatsDf())
}
invisible()
},
#' @description Write the data in the object to feather files.
#' @param out_loc Directory in which output files should be created.
#' @return No return value, called for side effects.
writeFeather = function(out_loc) {
# TODO Add getter and setter for raw config data in the Config class.
yaml::write_yaml(private$config$config, paste0(out_loc, "/config.yaml"))
write_feather(self$getSecurityReference(), paste0(out_loc, "/security_reference.feather"))
write_feather(self$getSimSummary(), paste0(out_loc, "/sim_summary.feather"))
write_feather(self$getSimDetail(), paste0(out_loc, "/sim_detail.feather"))
write_feather(self$getInputStats(), paste0(out_loc, "/input_stats.feather"))
write_feather(self$getLooseningInfo(), paste0(out_loc, "/loosening_info.feather"))
write_feather(self$getOptimizationSummary(), paste0(out_loc, "/optimization_summary.feather"))
write_feather(self$getExposures(), paste0(out_loc, "/exposures.feather"))
write_feather(self$getDelistings(), paste0(out_loc, "/delistings.feather"))
invisible(self)
},
#' @description Load files created with \code{writeFeather} into the object.
#' Note that because detail data is not re-split by period, it will not be
#' possible to use the \code{sim_date} parameter when calling
#' \code{getSimDetail} on the populated object.
#' @param in_loc Directory that contains files to be loaded.
#' @return No return value, called for side effects.
readFeather = function(in_loc) {
# TODO Check to see if this object is empty before loading up data.
private$config <- StrategyConfig$new(yaml::yaml.load_file(paste0(in_loc, "/config.yaml")))
private$security_reference <- read_feather(paste0(in_loc, "/security_reference.feather"))
private$sim_summary_list <- list(read_feather(paste0(in_loc, "/sim_summary.feather")))
private$sim_detail_list <- list(read_feather(paste0(in_loc, "/sim_detail.feather")))
private$input_stats_list <- list(read_feather(paste0(in_loc, "/input_stats.feather")))
private$loosening_info_list <- list(read_feather(paste0(in_loc, "/loosening_info.feather")))
private$optimization_summary_list <- list(read_feather(paste0(in_loc, "/optimization_summary.feather")))
private$exposures_list$net <- list(read_feather(paste0(in_loc, "/exposures.feather")))
private$delistings_list <- list(read_feather(paste0(in_loc, "/delistings.feather")))
warning("It will not be possible to use the sim_date parameter of getSimDetail on this object to filter detail records by period")
invisible(self)
},
#' @description Get the object's configuration information.
#' @return Object of class \code{list} that contains the simulation's
#' configuration information.
getConfig = function() {
invisible(private$config)
},
#' @description Write an html document of simulation results.
#' @param res The object of class 'Simulation' which we want to write the
#' report about.
#' @param out_dir Directory in which output files should be created
#' @param out_file File name for output
#' @param out_fmt Format in which output files should be created. The
#' default is html and that is currently the only option.
#' @param contrib_vars Security reference variables for which to plot return
#' contribution.
writeReport = function(out_dir, out_file, out_fmt = "html", contrib_vars = NULL) {
rmarkdown::render(input = system.file("reports/simReport.Rmd",
package = "strand"),
output_format = paste0(out_fmt, "_document"),
output_file = out_file,
output_dir = out_dir,
params = list(res = self, contrib_vars = contrib_vars),
quiet = TRUE)
}
),
private = list(
# Configuration and input data
config = NULL,
raw_input_data = NULL,
input_dates = NULL,
raw_pricing_data = NULL,
security_reference = NULL,
delisting_data = NULL,
shiny_callback = NULL,
verbose = FALSE,
# Results
# _list objects are lists whose elements are result data for single
# periods.
# High-level summary data, including period return, number of positions,
# etc.
sim_summary_list = NULL,
# Position-level simulation data
sim_detail_list = NULL,
# Input data statistics, such as number of NAs and period-over-period
# correlations.
input_stats_list = NULL,
# Information on any loosened constraints.
loosening_info_list = NULL,
# Optimization summary data.
optimization_summary_list = NULL,
# Exposures data.
exposures_list = NULL,
# Data for stocks removed due to to delisting.
delistings_list = NULL,
# @description Get the strategy capital levels for the strategy, based on
# the simulation's config.
# @return A list where the names are strategy names (or 'joint') and the
# values are the capital levels for each strategy,
getStrategyCapital = function() {
capital_list <- list(joint = 0)
for (strategy_name in private$config$getStrategyNames()) {
this_capital <- private$config$getStrategyConfig(strategy_name, "strategy_capital")
capital_list[[strategy_name]] <- this_capital
capital_list[["joint"]] <- capital_list[["joint"]] + this_capital
}
capital_list
},
# @description Save summary information.
# @param period Period to which the data pertains.
# @param data_obj Data frame to save.
saveSimSummary = function(period, data_obj) {
private$sim_summary_list[[as.character(period)]] <-
mutate(data_obj, sim_date = period)
invisible(self)
},
# @description Save detail (position-level) information.
# @param period Period to which the data pertains.
# @param data_obj Data frame to save.
saveSimDetail = function(period, data_obj) {
private$sim_detail_list[[as.character(period)]] <-
mutate(data_obj, sim_date = period)
invisible(self)
},
# @description Save input statistics.
# @param period Period to which the data pertains.
# @param data_obj Data frame to save.
saveInputStats = function(period, data_obj) {
private$input_stats_list[[as.character(period)]] <-
mutate(data_obj, sim_date = period)
invisible(self)
},
# @description Save loosening information.
# @param period Period to which the data pertains.
# @param data_obj Data frame to save.
saveLooseningInfo = function(period, data_obj) {
private$loosening_info_list[[as.character(period)]] <-
mutate(data_obj, sim_date = period)
invisible(self)
},
# @description Save optimization summary information.
# @param period Period to which the data pertains.
# @param data_obj Data frame to save.
saveOptimizationSummary = function(period, data_obj) {
private$optimization_summary_list[[as.character(period)]] <-
mutate(data_obj, sim_date = period)
invisible(self)
},
# @description Save exposure information.
# @param period Period to which the data pertains.
# @param data_obj Data frame to save.
saveExposures = function(period, data_obj, type = "net") {
stopifnot(type %in% c("net", "long", "short", "gross"))
private$exposures_list[[type]][[as.character(period)]] <-
mutate(data_obj, sim_date = period)
invisible(self)
},
# @description Save information on positions removed due to delisting.
# @param period Period to which the data pertains.
# @param data_obj Data frame to save.
saveDelistings = function(period, data_obj) {
private$delistings_list[[as.character(period)]] <-
mutate(data_obj, sim_date = period)
invisible(self)
}
)
)
Try the strand package in your browser
Any scripts or data that you put into this service are public.
R Package Documentation
Browse R Packages
We want your feedback!
Note that we can't provide technical support on individual packages. You should contact the package authors for that.
Embedding an R snippet on your website
Add the following code to your website.
For more information on customizing the embed code, read Embedding Snippets.