Yes, it works quite well. Jeff Ryan is the man. www.quantmod.com. Using his other package, IBrokers, or by embedding the REngine in a Java or .NET thread and using Interactive Brokers' APIs, it is quite feasible to start using R for mid-frequency automated trading : market data (direction:in) and order management (direction:out) are done in Java/.NET via the broker's APIs, and you marshal the market data off to R on it's thread and run it's analytics and spit out a trading decision.
The problem is that most Technical Analysis is somewhere in between voodoo and immature quantitative analysis. Thus, the real practical use of the package (for me) is in visualizing non-TA stuff.
"The problem is that most Technical Analysis is somewhere in between voodoo and immature quantitative analysis."
If you accept that past prices have no bearing on future prices, would you care to explain the difference between technical analysis and quantitative analysis? My own opinion is that QA is a subset of TA that places emphasis on statistical analysis rather than trying to see chart patterns (the voodoo you refer to). Still, aren't both things, regardless of the sophistication of the approach, striving to predict future prices from historical prices?
in quantitative analysis, you try to figure out how to make decisions that are statistically likely to be profitable.
generally, those that practice technical analysis, look at charts and apply rules, however you don't have any statistical evidence that your ideas will be profitable.
one could properly backtest technical analysis based strategies, and if one did so properly, while controlling for overfitting, it would be as valid as most quantitative analysis.
That was sort of my point. If you're backtesting, then you are assuming that past prices do in fact have some sort of bearing on future prices. Besides, aren't there all sorts of problems with backtesting, one of which is survivor's bias?
Sure. My complaint is independent of predictability, it's about the disconnect of TA from established principles of science (objectivity and reproducible research) and the abuse of established terminology.
One example is when the description is vague or cannot be the same for everyone. An example that comes to mind is 45-degree angles on charts ("Gann angles"? Not sure what they are called), which are extremely sensitive to changes in the scale of the Y-axis!
Another is the use of the word "stochastics" for a TA indicator that has nothing to do with what the word "stochastic" means in established research. (Also found someone using "Gaussian" recently to lend an air of sophistication to his idea, which has nothing to do with Gaussian distributions as understood by anyone else!). But heck, at least stochastics are expressed using an objective formula.
