Hey all, I've been devoting some time to looking at Neural Net/Genetic algorithm software over the past week and wanted to share my findings and see if people had comments or wanted to share experiences. Software I've been looking at: Neuroshell, Tradecisions, TradingSolutions. I've only been able to actually try out TradingSolutions because they offer an evaluation version of their software - the other two vendors make you buy the software first. TradingSolutions seems to be pretty decent. They have the ability to combine both neurally generated systems with hard trading rules. It also has good data management and integration with Interactive Brokers (all the vendors highlighted provide this). It provide a large number of indicators to use as inputs to any neural network. What it lacks is portfolio-level controls. These may not be important to everyone, but in my research I've found that trading signals are less important than position sizing/money management. In fact, I've found that you can do, essentially, random entries on a stock or group of stocks and get decent results by just resizing the positions based on, say, a distance from a stop. One very unique feature that I liked was the ability to attach different time series data and perform a correlation analysis - this should make it a bit easier to identify good inputs for a neural net. So TradingSolutions looks like it would be great for generating a system assuming you had a good system to begin with. It clearly isn't (this is true with all the vendors) going to spin your random thoughts into gold. Neuroshell seems to have the best/most interesting neural analysis techniques and I've heard from many people that it is the standard in the field. Tough, however, for me to analyze. It too is lacking the money management tools. In one example they sent me, they showed a relative strength ranking system for creating a hedged portfolio - pretty interesting stuff. If you're looking for a well supported product, Neuroshell seems to do a great job. Tradecision seems to have some really interesting features. I particularly like the way it handles external data (such as fundamental data). Say you wanted to pull in data on the COT report - this looked very easy in Tradecision. It may well be easy in the other packages. Finally, Tradecision has some money management and position sizing tools. It allows you to define a position size using typical sizing methodologies such as Williams, Kelly's or Optimal F. I'm still not clear if it is really working a whole portfolio, and it looks to me like it can't do pyramiding. Lastly, you seem to be able to attach any script to manage stops. One issue with neural-based systems is drawdown - so I'm looking for a system that could define, say, an ATR-based stop on top of the system to improve drawdown. Of the three, I think only Tradecision has the ability to do this. On the downside, I tried to call Tradecision directly to get some more info but have received no call back - so support may well be an issue. Tradecision also includes an interesting moving average (Ingenious Moving Average) that seems to have some of the properties of the Jurik indicators). Incidentally, all the systems here can use Jurik. There is also a quote on the Tradecision website from Jurik saying that they recommend Tradecisions. Neuroshell says that they are adding portfolio tools - I'll have to wait and see what they provide. All three provide sample systems based on TASC systems highlighted in the magazine. All three provide backtesting and reporting, with in-sample and out-of-sample testing. That's what I've got at the moment. I would love to hear from users of these products as to what they think of them. Feedback welcome.
I am using Tradingsolutions for a large portion of my trades, approx. since one year. I love the software. It - like all software - has it's little tiny awkward problems but all in all I have to say it is a beautifully engineered piece of software. Whoever designed the user interface knew what they were doing. All the functionalities that are built in work beautifully. I have a portfolio of hundreds of models that I run daily and generate trading signals. I too find the lack of position/money management tools a bit dissapointing. But hey, when I finally started understanding Tradingsolutions and slowly got the first profitable models out of it you can imagine I was happy enough to find something that works for me. At this point, I do all my position/money mangagement in a huge Excel spreadsheet that I developed. Of course, if TS offered such functionatlity it would be a huge timesaver. Here are some things that I find are missing: - The portfolio display doesn't offer any portfolio level analysis. Example: You have a portfolio with 100 models. I would love to display the equity curve of the entire portfolio (which of course must share the same data periodicity etc.) if e.g. each model is attributed 1% of $10m US. Thus it would display aggregate trading results in different market timeframes which helps to feel comfident with your trading portfolio should the market tank/rally strongly - you don't want to be caught 90% long/short at the "wrong" time. - Taking the above portfolio analysis to a money management level would of course be a dream come true. Example: Optimize the % amount of funds for each model, ranging from say 0.5% - 2.5%, in the training time frame. Then display the total portfolio equity curve using these MM rules in the cross validation and testing time frame. This would not only save me a lot of headache... it would likely improve my trading results as well. - While you can analyze many properties of your signals, some essentials are lacking IMO. E.g. in the portfolio display I am unable to display Sharpe Ratio or Max Drawdown %, while it displays these metrics inside of the analysis table of each model (takes 2 mouseclicks). I had to program these in TS to display them in my portfolio display which was a bit cumbersome. As a final note I wanted to talk about the learning curve. TS offers great tech support as well as excellent documentation and instructional videos. However, anybody who is not willing to spend 2-3 months learning the software is going to be dissapointed. It is not a silver bullet. It doesn't "print money" just by itself. You have to invest huge amounts of blood, sweat and tears and be creative and curious enough to try out new things all the time. You will be rewarded in the long run. P.S. Regarding your ATR SL question: I have implemented my own ATR Stop Loss solution in TradingSolutions. Basically you take a trading signal and "feed it into" a mechanical (ATR SL) trading system and then use the result of this as the final trading signal. Took me a day or two to figure out, was one of the first entry/exit systems I tried to build
Thanks for the great comments - nice to hear from someone actually using one of these software packages. Comments below. Agreed - I've been admiring the user interface quite a bit. In general, this idea of generating an optimal signal is significantly different from the approach of other software packages. I do find it a bit too dialog heavy - you are in one dialog, that leads to another that leads to another, etc. I found myself confused at several points, but that may have more to do with getting comfortable with the TradingSolutions concepts than anything else. I'm curious about your approach here - do you tend to create individual models for each instrument, or use one model across them all? I'm guessing it is the former rather than the latter. If it is the former, then are you re-optimizing the models daily? Or just using the models to generate signals? Feel free to, of course, not answer any of these questions if you consider it proprietary - I'm just curious how you are using the software. Yes, the software is complex enough without adding portfolio management and optimization, but I want to make sure it can do it in the future. Tradecision seems to be heading this way for their January release. Yes, that is exactly what I find lacking. I'd like to be able to attach money management rules like "never let a position get above 3%" or "equally divide total capital into a number of sectors, keep the sectors at fixed sizes but vary the allocation in the sectors." That is a very interesting concept and one that I have not seen from any vendor. I'm not sure how it would work - given that most of the time the optimal signal, to begin with, is generated off of concepts such as minimum gain, etc. So, let's say that you're optimizing a portfolio - presumably you're optimizing position size relative to gain (probably applying more capital to a position that has performed better historically), then is there much to optimize? I suppose you could take something like a Williams portfolio allocation and then optimize around the % per position, but I'm not sure how it would work with the individual positions. Totally agree. Needs to be there. Yes these are not software packages for those afraid of hard work! It strikes me that, to begin with, these packages are actually best for optimizing strategies that are already profitable. I'd be interested, if you're willing to share, as to what your approach has been to creating strategies. Do you start with a successful strategy? Are you generating completely new strategies? Good to know - thanks for the feedback!