In this issue:
I. Microsoft Windows Vista
II. NeuroShell Trader tips
III. The John Ehlers sale
IV. Editorial Comment by Steve Ward, CEO
V. Possible email delays or problems next week
VI. One way to stop this newsletter
*******************************************************
I. Microsoft Windows Vista
Microsoft is working hard to get their new operating system Vista out the door, and we’re working hard to make a version of NeuroShell Trader that runs on it. Current versions DO NOT, so if you are tempted to install one of Microsoft’s Vista betas, keep that in mind.
We’re starting to test the other software now too, but it is too soon to tell how it will work on Vista. We know there are at least problems with help files, and maybe more.
*******************************************************
II. NeuroShell Trader tips
In an effort to help you understand why Genetic Algorithms (GAs) act differently than other optimizers to which you may be accustomed, we offer the following facts about our GAs in NeuroShell Trader. Much of this information is applicable to genetic algorithms in our other products too.
GAs emulate the theory of evolution. What they do is much like selective breeding. Multiple pairs of solutions are “mated” during many “generations” to hopefully produce more “fit” offspring.
GAs do not try every combination. Selective breeding causes exploration all around in the search space based on random starting numbers. Algorithms that try every combination are called “exhaustive search” algorithms, and when dealing with parameters that are not integers, exhaustive search optimizers must search in increments.
GAs do not require a search increment. That is because they do not try every combination, so they can deal with real numbers (with decimals) without increments.
GAs don’t know when they’re done. Since they don’t search in orderly increments, they never know when they have found the optimal solution. They use arbitrary stopping criteria in NeuroShell – they just stop when some number of generations pass with no better solutions being found, or when some amount of time has passed. Since when the GA stops is arbitrary, you are allowed to either stop it earlier or tell it to go a long period of time.
You can never assume you should get the same results with a GA unless you are doing EXACTLY the same thing. Anything different can cause differences because GAs start at random starting points, do not search every combination, and they have arbitrary stopping criteria.
The computation speed of the GA doesn’t depend on the GA or the number of parameters you are optimizing. It depends on how long it takes to backtest the model. Models that have indicators that take a long time to compute are going to take longer, and it has nothing to do with the GA. The computation in the GA is a minor part of the total calculations during optimization. However, when the GA actually ends can depend on the number of parameters, because more parameters may make it more likely that better solutions get found, thus continuing the search. Real parameters as opposed to integers can prolong the searching in the same way.
The GA will only try integers for any parameter that is designated as integer. If the parameter is not integer, the GA will try real numbers (with decimals). All that is decided when the programmer creates the indicator.
GAs are usually much faster than “exhaustive search” algorithms. That is again because they are not trying every combination. They can actually be slower, however, if only a couple of variables are being searched, or the exhaustive search algorithm has been given large increments in which to search.
GAs cannot be guaranteed to always find the optimal solution. That is again because they are not trying every combination. Exhaustive search algorithms ARE guaranteed to find the optimal solution, assuming it is exactly on the search increment of each variable, and assuming you have the time it could take with many variables. Trading usually does not require the optimal solution anyway, because it will most likely over-fit.
GAs do not require a continuous or differentiable search space. “Newtonian” or “hill climbing” type optimizers often do impose such conditions. In fact, exhaustive search algorithms can only work in continuous spaces if they are instructed to search in increments. Continuous and differentiable are mathematical terms you may not understand, but the bottom line is that many optimizers will stop or fail if the search space is not a nice smooth set of curves with no gaps. GAs don’t care what the search space looks like mathematically, but see the next item.
GAs can fail if the parameters they are given are too restrictive. Optimization may end early if the genetic algorithm cannot reproduce, i.e. if there is not enough diversity in the optimization parameter search space. This happens when too many parameter combinations yield the same result. It is like “inbreeding” in the selective breeding process. One or more parameter ranges may be too wide or too narrow or otherwise not appropriate for the underlying indicators. You may need to choose parameter ranges that do not result in the same profit result for a majority of the search space.
GAs are less likely to get stuck in local minima or maxima. That is because they are searching many points in the search space simultaneously, and are also not “hill climbing”. High dimensional search spaces frequently contain local maxima and minima that GAs are better suited to overcoming.
*******************************************************
III. The John Ehlers sale
Not long ago we had a sale of our own add-ons. Now we’re going to give equal time to the add-ons developed by John Ehlers – the Cybernetic Analysis add-on and the MESA8 add-on. This sale is only for existing users, and you have to buy both add-ons before the end of October to get the 20% discount. Our VIPs (those who have in the past spent $4000 on Ward Systems Group products) will get 30%.
If you are already the proud owner of one Ehlers add-on, we won’t leave you out in the cold. If that’s the case we’ll substitute two of our add-ons for one of Ehlers’ add-ons. So you buy one of Cybernetic or MESA8, plus two of ours (for a total of three) to receive the 20% or 30% discount.
*******************************************************
IV. Editorial Comment by Steve Ward, CEO
1. $0 Trades at Bank of America
If you live in Connecticut, Delaware, Massachusetts, Maine, New Hampshire, New Jersey, New York, Pennsylvania, Rhode Island or Vermont you can already make online self-directed equity trades with Bank of America. You do need to have $25,000 in accounts with the bank, but that includes CDs, savings, money market, and checking accounts. Other states are supposed to be coming soon. You can make up to 30 trades per month, including ETFs, and I think you can manage your IRAs under the plan.
Sounds like a great deal to me. Not that the $7 to $9 per trade other brokerages charge is a big deal. If you trade 30 times a month, though, you’ll save up to $270/month, which is $3240 a year. Now it does sound like a big deal. I called and asked about short trades. Of course they fall under margin fees, but if you have enough money in your accounts they don’t cost either, I’m told. Don’t take my word for the details, however, check it out yourself:
https://www.baisidirect.com/Static/SelfDirected/index.html
2. Some of you are struggling with your one copy of NeuroShell Trader Professional, or DayTrader Professional, trying to let several people use it under our liberal multi-computer authorization system. However, I’ll make you a deal – I’ll sell you a second copy for a 50% discount so you don’t have to worry about who’s on when. Call this month.
*******************************************************
V. Possible email delays or problems next week
Our Internet Service provider is updating our email service next week to make our email service faster and better. The update is supposed to be fast and transparent. We’ll believe that if we see it. Until then we are bracing for interruptions and delays, possibly all week. Please bear with us next week with respect to processing of your emails! If you don’t hear back from us in the usual 24 hours, send your email again, and we apologize in advance for any disruptions.
*******************************************************
VI. One way to stop this newsletter from coming
It is really easy. Just change your email address and don’t tell us.
*******************************************************