Bud uses both NeuroShell 2 and the NeuroShell Trader software for his networks and systems, but is shifting to the Trader because it is so much easier to use and does so much automatically for traders.

As to inputs, he says his farming experience helps decide what things might be important to try. But mostly he uses available data such as price, volume, and open interest figures. To represent the data, he uses common indicators, ratios, and rate-of-change calculations. There are over 800 indicators for him to use in the NeuroShell Trader. Persistence pays, he says. Last year he spent a good share of the winter trying to build a net for a favorite market with NeuroShell 2, but never got one to generalize well on out-of-sample data. But this winter he thinks that he has a good model - finally. He will start with many potential inputs - sometimes twenty to thirty - and narrow them down to eight to twelve for the final nets. To do that, he uses the contribution figures for each variable that his nets give him. He uses rather large training sets of between six and ten years, and then evaluates the nets on one to two years data to see how good they are performing on out-of-sample. That job is now automatically done by the NeuroShell Trader. Sometimes he trades with these nets and sometimes he retrains them with data up to the present time.

Bud also revealed that he uses multiple nets, a technique he learned from attending one of the Ward Systems Group classes. Multiple nets are a snap with the NeuroShell Trader, but much more difficult with traditional neural net products.   Bud has a wealth of philosophy about the grain markets. “Don’t expect to get great results on your out-of-sample data. If the market is too predictable using historical data, those patterns will not be there now. Learn to live with poor R squared values when predicting price change - in fact, if they are too high, be suspicious of them. There are no killer inputs,” he states flatly. “But you don’t have to predict a market every day to make money - you just need to know when the odds are on your side - and they don’t have to be on your side all that much to make a good return.”

Bud has observed that there are times when the nets are not as reliable as other times. In the grains, summer is an uncertain time, and the nets will show extreme variability during weather markets. Similar input patterns historically have probably resulted in big moves both up and down - no one knows on a Friday afternoon whether it will rain in the corn belt by Monday morning. He also observes that markets are gaining volatility year by year. Effective inputs two years ago have had to be replaced because they moved completely out of the range that they were in when the nets were trained. “More and more money is looking for anything that will trend up or down, and I think this will increase even more in the coming years.” He thinks that it will be more and more important to be able to retrain nets on the most current data in the future.

Update - 1/6/2006

I am writing to update what is on your website about me. I thought that I should replace the original post, as it is several years old. But upon reflection, I decided that I would write this update, and write it as honestly as I can. Hopefully my experiences will be of help to some of the newer traders that read this. It is important to be able to deal with both the peaks and the troughs that seem to be inevitable in trading markets, so I will write what I would have liked to have been able to read when I first began trading many years ago.

I made a spreadsheet of my profit & loss by month, and then I charted it. It was no surprise to me that just as the original piece was written about me here on this board, I had hit a peak. Over the next couple of years I did give back a portion of the profits that I had made. It was a combination of a couple of things. One was certainly that I had serious distractions from the trading that I was doing. The distractions were serious enough that I should have exited the market. And the other was what Victor Niederhoffer devoted a chapter to in the second book that he wrote – hubris. All through this time, however, I used the NeuroShell Trader very successfully with long term price and profitability analysis in my business of farming and cattle feeding. But even though successful at long term analysis, my trading suffered due to my being seriously distracted and overconfident.

The difficult thing for most people about trading the markets is that we have to do the “hard thing” rather than what we feel is the right thing to do. One thing that I really do like about the markets is that no committee decides whether I am right or wrong. Keeping score is simple. The report card is the balance in my trading account at the end of the year. There seems to be a very small percentage of people who are born with their brains hard-wired to be good traders. Most of us are born hard-wired with severe limitations when it comes to trading. To complicate matters, it also seems to me that good traders do not generally make good analysts, and the traits that make a person a good analyst usually make for a lousy trader. People who buy the NeuroShell Trader are probably better analysts than traders. This, IMHO, is important, as without a trading system that we develop ourselves, trading for us will most likely be frustrating. This is where software like the Trader can be so very important. Besides being fun to build and test systems, it can save a great deal of frustration and probably thousands of dollars testing ideas in the actual markets. And, without a system, when it quits working, what do you fix? Add to this the tremendous amount of learning that comes with putting in the effort to build a working system with software that allows a person to easily test any idea.

As for suggestions on building a system, I would first make a long term back-adjusted contract of the commodity and visually look for patterns. Does it trend much of the time, or might a swing trading system work? Draw lines from where you would like to buy it to where it might be sold. Ward System’s “Turning Points” add-on is very helpful for this. The points are computer generated, so this takes much of the subjectivity out of the exercise. If this is done with multiple markets, it will be seen, I think, that each market is individual. Another of Ward’s indicators that is helpful here is the Hodrick-Prescott filter. Using either the Turning Points indicator or the HP filter and then hiding the actual price sometimes makes patterns jump out better. The systems that I use came as a result of this and were a surprise to me. I never expected to see stats on out-of-sample data that good.

Using a system that has tested well on out-of-sample data is interesting. A swing system will get a person in long when it is least comfortable, it seems. And then, just as the long position becomes comfortable, the system says to get out of the long position and get short. And on it goes – just as it gets somewhat comfortable to be short, it says to go long. If you think about trading a swing system, this is how it should be. And this is why, for most of us, trading without a system is so frustrating, as we need to be doing exactly what we least want to do. In a trend following system it is similar. It seems like when the time comes to take a position, the market has everyone in a state of mind such that it is impossible to act. This is where having a system built in the Trader is so valuable. It is the “kick-in-the-pants” needed to take a position. The same applies to exiting a trade.

The current systems that I have built in the Trader greatly reduce stress. Entries and exits are clear, and even when I do not follow the signals exactly, I know at those times that what I am doing is contrary to the percentages. Trading is much more enjoyable now than it has ever been for me. Another thing I have done is to take money out of the account regularly, so that I keep a comfortable amount there at any given time. Quality of life is much better, even though I am not maximizing the potential to make money. I trade the markets mostly because I enjoy trading, and I enjoy the analysis and the challenge. In recent years, each year I have about doubled the balance at the beginning of the year. I have no desire to do what is necessary to get rich trading, as the cost is too high for me. But the Trader has been giving me a very nice supplemental income along with a hobby that is fascinating.

I would like to emphasize that this is not an impossible thing to accomplish. If a poor dirt farmer who is very busy Spring and Fall with farming, and who likes to spend much of the summer in the far North fishing, can build and use a working system, it should be an attainable goal for someone willing to put in the necessary time and work. I am sending this unsolicited, as I am a very satisfied long term customer of Ward Systems. Simply put, it is enjoyable working with the Trader. The people at Ward Systems can’t be beat. I am very impressed with how easy the Trader makes it to look at relationships and work with the data in addition to having the neural net capabilities. I have been in the futures markets now for over twenty-five years, and I have looked at a lot of software. Most of it is high priced and of questionable value. Much of it is junk. If I could only keep one market analysis program, it would without doubt be NeuroShell Trader. And for what it is capable of doing, it is a bargain.

Update - 2/9/2007

"Please sign me up for the upcoming seminar in March. I feel that the last one several years ago was extremely helpful in getting me going with the Trader, and looking at the list of topics you plan to cover this year makes me think that this one will be equally helpful.

I am forwarding results of one market that I have been working on. The first model is one that I had made a couple of years ago and have been observing and trading in real time since then. I am hoping to pick up some ideas at your seminar about how to build stop loss orders into the system to further smooth the equity curve. The annual return on account for six years at nearly 112% per year is good considering that the data is strictly out-of-sample. If I could smooth the equity curve more, I could trade this system with even less stress than I have for the last two years."

Bud

Update - September 2007

"Just a note to say that Judy and I had another great summer fishing in British Columbia.

I'm enjoying using ideas I got when attending your seminar in Florida. You presented some very good ideas, and even though I have used your products for many years, you provided a great kick-start for me to look at my models from new angles."

Bud