I have been having many problems with my main backtesting software, AmiBroker, and the data it uses. This past week has found me every evening uninstalling, installing, defragging drives, doing disk checks, so on and so forth… I thought I had the problem fixed yesterday, and then I find out that now my database is only going back to 1993 (should be 1985). So when I try to fix that issue, it seems that all the initial problems I had fixed, are back.
To be clear, I do not think this is an AmiBroker problem as I hear it is very very stable software. I also do not think it is my data provider’s problem. I think that it is probably the interface between the backtester and the data and my computer.
I have lots of things I want to test, and I can’t wait to get started on my Indicators and Edges series, but I have to get this problem straightened out first, so that I can trust the results. Hopefully by the end of this weekend everything will be up and running and ready to go.
Wood,
Just as a note to you and everyone else as posters to IBC, it is great to see the effort and honesty you all bring to the table. This is the difference between IBC and any of the billions and billions of stock blogs out there.
You, CA, RC, JG, Fly, Gio and the others care about the product, as informative and humorous as it might be, it still is a consumer product for the people who “tune in” regularly.
You guys are not just chest thumping screamers putting out the “look at me”.bullshit so common in this arena.
You may all be just humping screamers for all I know, but you all do good with this service and product.
I believe it is called intergrity.
Have a great day and thanks again
I bet Vincenzo spilled some gravy on your HDD. Fly should fire that guy!
Thanks Mustard. I am honest, to a fault. Glad you appreciate it.
I know, Wood is one of my absolute favorites. He was very helpful to point out to me and give me an article to read about moving averages, and how I was doing everything stupidly backwards when buying and selling.
I don’t comment much on this blog, but I read it every single day. And your power dip buying system helps me to know how the market might look that day and gives me new stock ideas to check out.
Thanks for how much you have and continue to help me, Wood!
Have you played around with neural networks yet? I’ve ordered the TradersStudio add-in ($200) and have also looked at Neural Trader ($1500).
The idea is that there are two types of systems. Rules-based (where you know what indicators make sense and can create a sensible strategy) and NN-based (where you select a group of inputs and let the NN determine how much weight to place on each indicator).
As for optimizing, you can brute force many rules-based systems, but genetic algorithms are more efficient if you have more than two variables, or if you have multiple inputs as found in many NNs.
Does AmiBroker have a GA optimizer?
A couple of thoughts on neural networks. The biggest problem I’ve seen with NN is the faith that is required – most NN systems don’t deal with stops. This makes position sizing an issue, and it also means you may watch the system crash and just have to hold on to see if it comes back. It’s very difficult to know, in my opinion, if the system has died.
The lack of a stop is a bigger issue for me – I use stops to size positions, but I also like to know there is a hard out if the system stops working. I’ve not seen a NN system that deals with this issue. The most interesting system I’ve seen is by Biocomp – called Dakota – which uses “intelligent” agents to make predictions. But again, no stops – check out the journal on Elitetrader by someone using the product to get a sense of what I’m talking about.
As for a GA optimizer, Amibroker offers 3 different optimizers – one, I believe is GA-based, but it is important to recognize that an optimizer doesn’t have to be GA-based to be fast, and Amibroker (both as an optimizer and in general backtesting) is very, very fast.
All just imho….
As far as hard stops and NNs, why couldn’t you just program in an ATR-based stop on your open positions that override the NN exit signal?
I’ll look into the Dakota system on ET.
BTW, isn’t the PPT essentially an NN? From what I gather, it takes a set of proprietary inputs (fundamental and perhaps technical) and spits out an output as to a buy/sell bias.
As to GA optimizers. I was running backtesting with the brute force method, which is pretty fast on TradersStudio I guess, but when I introduced three variables (MA1,MA2, BB1), the brute force method’s time requirement becomes more of an issue.
You could program hard stops but that defeats the point of a NN where the goal is to re-evaluate according to what the net is saying. While what you’re saying makes intuitive sense, I’ve not seen any ability to set a hard stop in any of the major NN software providers.
I’m not sure what you mean by PPT – I’m not familiar with the abbreviation.
Tradersstudio crawls as soon as you add a lot variables – in contrast, AB does much, much better.
I have yet to use an NN, but my expectation is that it provides an output (enter long, enter short) based on various inputs I provide. I’m not sure why you can’t just leave it there. Just like an MA crossover provides a signal. You can exit an MA crossover with its inverse, or after x bars, or after n ATR, no?
I’ve got some research to do.
PPT? Look to your right. A premier service provided by the good folks of ibankcoin.
Ah – I thought you were referring to an application vs. PPT – got it. No idea if it is NN or not.
The difference is that if you have an MA system, you can still set up hard stops (or variable stops based on, say, ATR). In the NN world, I think that would have to be done as an overlay because, as you mentioned, it just generates buy and sell signals. So you could create a NN and then, say, export the signal to Amibroker or something like it and then add anything you like as an overlay – I’m just saying that NN people would generally say that since it isn’t part of the system, it shouldn’t be used. As for whether you could have a stop as an input – I’ve never seen it as part of any NN application – they’re usually just buy/sells based on the current data.
The other issue I’ve found with NN is that they really blow apart at junctures in the market.
I’ll let you know if I can use an NN signal in TradersStudio after I get it up and running. I’ve seen this technique demonstrated on the Neural Shell Trader site.
I’m viewing NN as a tool to organize a group of data inputs, to keep track of them, and to seek out the most useful inputs.
I took a quick look at the BioComp Dakota system and it seems their NN has proprietary data inputs, but I’m not sure …
I think Biocomp is very different from the other NN vendors – I’m not even confident it would be called NN (as opposed to, say, Neuroshell). Biocomp uses the concept of agents and is all forward testing. In contrast, Neuroshell and other vendors seem to take inputs and then do a backtest to optimize against that data. Pretty different approach imho – and I think Biocomp is a better approach because it is all forward testing.