Some tips on backtesting in quantstrat

You might run into odd errors when backtesting strategies on tickers Ford (F) and ATT (T) in quantstrat. Since F is also reserved word False and T is shorthand for TRUE these ticker symbols will likely result in errors. Assigning F <- TRUE and T <- FALSE before backtesting will be a workaround as long as TRUE and FALSE is used elsewhere in the backtests.

Stoploss creation is a bit tricky. One has to use order-sets to get the exit orders cancelled once either target or stop is filled.

If you want to use limit orders you should use the "prefer" parameter to specify a column which contains the price for the limit order, "prefer" can be used for all columns in mktdata.

Canceling Limit orders in quantstrat

In quantstrat you might want to use Limit orders to avoid crossing the market. Since all orders in quantstrat are modeled as GTC orders the unfilled ones have to be cancelled once the conditions for placing them are not true anymore.
For that I made this function to cancel the orders https://r-forge.r-project.org/scm/viewvc.php/pkg/quantstrat/R/ruleCancel.R?view=markup&root=blotter&sortby=log&pathrev=1233

Speed performance considerations have to be taken into account because when using limit orders in quantstrat the signals are probably more often TRUE which causes more processing.


The code of ruleCancel.R:

#' rule to cancel an unfilled limit order on a signal
#' 
#' As described elsewhere in the documentation, quantstrat models 
#' \emph{orders}.  All orders in quantstrat are GTC orders, which means that
#' unfilled limit orders have to be cancelled manually or replaced by other orders.
#' 
#' This function is used for canceling the orders based on a signal.
#' 
#' @param data an xts object containing market data.  depending on rules, may need to be in OHLCV or BBO formats, and may include indicator and signal information
#' @param timestamp timestamp coercible to POSIXct that will be the time the order will be inserted on 
#' @param sigcol column name to check for signal
#' @param sigval signal value to match against
#' @param orderside one of either "long" or "short", default NULL, see details 
#' @param orderset tag to identify an orderset
#' @param portfolio text name of the portfolio to place orders in
#' @param symbol identifier of the instrument to cancel orders for
#' @param ruletype must be 'risk' for ruleCancel, see \code{\link{add.rule}}
#' @author Niklas Kolster
#' @seealso \code{\link{osNoOp}} , \code{\link{add.rule}}
#' @export
ruleCancel <- function(data=mktdata, timestamp, sigcol, sigval, orderside=NULL, orderset=NULL, portfolio, symbol, ruletype)
{
 

  if (ruletype!='risk') {
    stop(paste('Ruletype can only be risk'))
  }
  
  if(is.null(orderset)) orderset=NA
    
  updateOrder(portfolio=portfolio, 
              symbol=symbol, 
              timespan=timespan,
              orderset=orderset, 
              newstatus='canceled'
             )
  
}

###############################################################################
# R (http://r-project.org/) Quantitative Strategy Model Framework
#
# Copyright (c) 2012
# Niklas Kolster
#
# This library is distributed under the terms of the GNU Public License (GPL)
# for full details see the file COPYING
#
# $Id$
#
###############################################################################

How to calculate higher timeframe indicators on xts data

Sometimes you have higher frequency data and you need to get a slower timeframe analysis, it can be accomplished with this undocumented but real powerful feature of by.period:

getSymbols("SPY", src="yahoo")

SPY$wklyRSI <- xts:::by.period(SPY, RSI, period="weeks", n=2)