I've read that IB provides sampled data and it certainly appears to, going by the (non)performance of indicators that depend on tick data. I would have to compare it e.g. to IQ to quantify that, however.
On the topic of IQ, I subscribed to the IQFeed trial last week and am still studying the cost/benefit. One of the first things I noticed, besides the fact the data appears to be complete as advertised, and hence am able to use Gom tools, unfortunately the (NT default) EURCAD.FXCM spreads are at least 2x IB spreads. Still have to look at IQ's Tenfore & Barclays Forex products.
Speaking from a bot operator's perspective (and sometime manual trader) and trying to shorten the time frame of a fairly reliable currency trading strategy from a 4 hour bar to something less, I'm finding Forex becomes more ill behaved the closer you look. Murrey Math & Fib continue to work as well as ever but the rules / indicator / money management combo complexity increases exponentially the shorter the time frame, as much to avoid whipsaw and to cut losses as to generate entries, quantities and exits. With bots at least any dependence on lagging indicators like ATR seems to generate losses more often than not; they appear to profess something about the future that is demonstrably not true, except if true then only briefly and likely coincidentally. Interestingly doing the exact opposite of what the indicators say turns out also to be an effective way to generate losses.
That said, I *can* (and do) trade short term Forex profitably by hand but any indicators on the chart are there pretty much only for comic relief--entries and exits are based almost exclusively on the instantaneous price pattern and tend to go all in and all out. Still a way to go if any set of hard & fast rules (i.e., a bot) I come up with is able to duplicate that.
More to the point, part of the reason I auto-trade virtually 100% of the time is by the time we've optimized, back- and forward-tested what appears to be a profitable set of rules on a machine, the machine ought to be able to manage it live while I do something more constructive (e.g., nap).
Sometimes the simplest realization can dramatically improve profitability. In my case (as a bottist trying to capitalize on shorter time frame moves) for example, this meant skipping the 2 hour bar, which showed only marginal improvement over the 4 hour bar in terms of profitability, and focusing on the 1 hour bar. It turns out what was happening is 24/5 calculate-on-bar-close bots using bars that are multiples of 2 hours synch to midnight by default, which means I was missing New York coming online at 9 AM (in my time zone--AST), crucial for Euro dynamics at the moment (i.e., while London is more or less bullish and New York is more or less bearish--significant action at the open). The bot was repeatedly throwing away what amounts to a decent day's wage in the 2 hours between 8 AM and 10 AM, which appears only as a blip when analyzing longer time frame trading results.
Edited to add: for anyone interested in time frame versus instrument analysis, IMO a 1 hour bar still under-samples the Euro in terms of potential bang for the buck (payback versus time spent worrying about it), but that remains to be proven.
Last edited by bnichols; March 2nd, 2011 at 11:03 AM.