NeuroShell Trader

5 Pages·2003·0.61 MB·English

by  Peterson D.D.

Reprinted from Technical Analysis of STOCKS & COMMODITIES magazine. © 2003 Technical Analysis Inc., [phone], http://www.traders.com
PRODUCT REVIEW
NeuroShell Trader
Version 4.2
WARD SYSTEMS GROUP, INC. Systems’ NeuroShell Trader.
Executive Park West What makes this an outstanding prod-
5 Hillcrest Drive uct? As you get started, you find that the
Frederick, MD 21703 CD-ROM that contains the program code
Phone: [phone] also has a number of video clips. Each
Fax: [phone] one is short and to the point so you don’t
Email: [email protected]; have to listen to a long lecture. You can
[email protected] listen to just what you need. That’s a very
System requirements: 800 MHZ nice touch I seldom see, and it was my
Pentium or equivalent, 128 MB first clue as to the quality of the product.
RAM, Windows ME/2000/XP. NeuroShell Trader enables the user to develop
(Much of this review was done using DATA SOURCES AND CHARTING and test trading systems in a charting interface;
a 450 MHZ PC with Windows ME.) The first thing you want to do is access choose from over 800 indicators, neural networks
for predictions, and trading rules.
Price: Prices start at $795 for your data. NeuroShell Trader supports
NeuroShell Trader, and go to $2,295 a variety of data formats and allows you
for NeuroShell DayTrader Profes- to distinguish between data on your
sional. Numerous plug-ins available hard drive versus datafeeds you might namic Data Exchange, and Quote.com.
at $299 each. have. You can mix and match sources In Figure 2, I used the eSignal data
on a single chart. The menu item “Tools” server for the top chart of daily QQQ
by Dennis D. Peterson lets you select “Data sources,” which data, and for the bottom chart of Nasdaq
has three tabs (Figure 1) allowing the new highs and new lows I used
W hen you start review- identification of servers and datafiles. MetaStock data from my hard drive. To
ing an excellent prod- For example, under the “Server” tab put this data on the chart, you follow a
uct you can generally you can see that six different servers are pattern of user interface techniques of
tell immediately, and recognized: eSignal, TradeStation, dropdowns and dialog selections used
so it was with Ward Omega Research Global, Prophet, Dy- throughout the software. What you even-
FIGURE 1: DATA SOURCES INTERFACE. Establishing a link to data is easy, as the FIGURE 2: DAILY QQQ AND VOLUME, WITH NASDAQ NEW HIGHS AND LOWS. Legends
choices have been made for you through dialog screens and dropdown lists. All you have accurately describe the data shown. The top chart is daily QQQ data using an eSignal
to do is make a choice. The most popular servers are supported, while the Directories/ server, with three moving averages; note the legend at the top with volume in the middle
Files Tab allows you to follow a folder/file tree of your hard drive to identify MetaStock, CSI chart, while the bottom chart is the 10-day simple moving average of Nasdaq new highs/
Inc., and csv (comma separated variables) files. (new highs + new lows) using MetaStock files on the hard drive. The height of each chart
can be sized using cursor drag and click between charts, while time covered is governed
by zooming in or out over a scrolled area.
Reprinted from Technical Analysis of STOCKS & COMMODITIES magazine. © 2003 Technical Analysis Inc., [phone], http://www.traders.com
FIGURE 3: INDICATOR INTERFACE FOR TRADING STRATEGIES. In keeping with FIGURE 4: INDICATOR RELATIONS AVAILABLE. Rather than have the user type
a syntax-free interface, the user is given a list of categories to choose from for in what A and B might be, along with <, >, or =, you select one of the relations and
building the trading strategy rules. Choices include day and time flags to allow you the software will allow you to select a line that reads “A=” and fill in the blank using
to optimize on days of the week or hours of the day, and relational rules such as A>B, the indicator wizard where the numerous indicators are available.
where you can choose from any of the 800-plus indicators to be either A or B.
tually find out is that the user interface vides neural network solutions as well, indicator might work best? In bull mar-
doesn’t require you to become an expert but the optimization of trading state- kets, it seems at times the rule is to sell
in NeuroShell syntax, since the inter- ments or indicator parameters is incred- before lunch and buy in the last five
faces have all been set up for you. ibly quick, brought about by the use of minutes of the trading day. Even in bull
The charting capability provides crisp a genetic algorithm. markets, with a substantial gain the day
clean pictures (Figure 2) and allows you before, profit taking seems likely at the
to add charts of indicators and other GENETIC ALGORITHM beginning. Mind you, this exercise
data. How much you see of what you Unless you go out of your way to find a doesn’t even touch using neural net-
have created is controlled by interfaces description, you won’t see how works, but we’ll get to that.
that allow you to hide selected pieces of seamlessly this approach has been inte- From the main menu line, select “In-
data, zoom in or out on what is shown, grated into the software. What the algo- sert,” then “Trading Strategy.” You are
or simply change the frequency by us- rithm does is find the optimal solution presented with tabs that require trading
ing a dropdown list at the top right of the to either the parameters in custom indi- rules for long and short entry/exit. When
top chart. There are more than 800 indi- cators or the true/false statements in a you select “Add Conditions,” you get the
cators to choose from and you can select trading strategy. Indicator Wizard (Figure 3). The lists of
multiple periods for an indicator by The algorithm is only available in the indicators are strategy-specific, so if you
simply typing in the periods separated professional versions of NeuroShell try to insert an indicator not connected
by commas. Charts can be saved either Trader. The performance is amazing in with trading rules, it won’t work.
as a chart with your specific quote data that what would easily take hours with Now, select one of the categories. If
or as templates to retain indicators or other processing models, going through you choose relational, then you get a set
trading strategies you have built. every combination of parameters or state- of choices (Figure 4). Then select the
ments, gets you there in a relation that makes sense, such as A>B,
matter of minutes using and when the relation is shown in the tab
The software provides neural the genetic algorithm. for, say, long entry, you will be given
Now, of course, you might the chance to state what is going to be A
network solutions as well, but the
worry about curve-fitting, and B by first clicking on A then B.
optimization of indicator parameters
but NeuroShell Trader is one When you do, the indicator wizard ap-
is incredibly quick, brought about by step ahead of you on that ac- pears again.
the use of a genetic algorithm. count as well. As you create the statements for each
Let’s take a problem of the tabs, you’ll notice a small “+”
that usually consumes beside the rule, and clicking on it causes
You’ll note the absence of drawing many traders’ hours of either set-up or an expansion that shows the range of
tools such as trendlines, pitchforks, or computational time to see if there is a variables that are going to be tested
Fibonacci retracement lines, but the real winning trading strategy. There’s a prob- (Figure 5), which you can change. At
power of this software is in the genetic lem, though: Are there favorable times the top of dialog space, a dropdown lets
algorithm. Obviously, the software pro- to trade a stock during the day, and what you select from market, stop, limit, stop
Reprinted from Technical Analysis of STOCKS & COMMODITIES magazine. © 2003 Technical Analysis Inc., [phone], http://www.traders.com
FIGURE 5: TRADING STRATEGY INTERFACE. Separate tabs allow separate entry of rules/ FIGURE 6: TRADING DATES OPTIMIZATION. NeuroShell Trader Pro-
conditions for long and short entry/exit, and clicking on the “+” to the left of the condition fessional allows you to pick a span of dates to optimize on, and then paper
statement expands the statement to show the parameter range chosen by the system. Each trade another period using the optimized results. The other five tabs allow
line can be individually selected to allow the parameter range to be changed. Note that in the you to select all of the usual parameters for trading such as commission
middle condition, the relation is easily seen as A>B, where A is the MACD signal and B is a size, trading size, as well as choose from 18 different optimization
parameter with a separate range. There’s no syntax; just click and fill in the blank. techniques, such as maximize on net profit, minimize drawdown, and the
maximum number of winners.
limit, and market close for order types, range of values. It took a
and looking farther to the right, you see computer with a 2.6 GHZ
a box that says “If all of the following processor with Windows
are true.” The “all” can be changed to XP approximately five
one, two, three, or whatever number of minutes to arrive at the
statements you want to be true to place answer.
the order of your choice. The answer is presented
You could set all this up in 10 min- in a variety of forms, one
utes. Now, recall that NeuroShell Trader of which is the trading rules
was one step ahead of us in curve- (Figure 7). In case you are
fitting. You can select one period of curious about trading SPY,
time to optimize and then, using the the answer is: Buy long at
parameters or statements found in that 2:30 and sell at 4:00, given
FIGURE 7: OPTIMIZED TRADING RULES. The results of your trading
optimization period, use them to paper that the MACD signal is strategy are presented with the usual statistics, trade-by-trade rules as
trade. An interface is provided (Figure greater than -6.5, using a well as the rules that finally were chosen.
5) that allows you to select the first dates MACD of seven, eight,
for the optimization period and the pa- and 12 periods and a
per-trading period. The other tabs allow two-period RSI > 28;
you to set commissions, margins, buy a and buy short at the
fixed number of shares or buy against a first interval after the
fixed number of dollars, or use the whole market open if the
account, optimize against profit or mini- MACD of two, six, and
mum drawdown, and so forth. 19 periods is less than
88 and a nine-period
TRADING STRATEGY RSI is less than 57.
OPTIMIZATION This gives you 14 wins
Now, you let the software work on the and 10 losses in paper
problem. I used the genetic algorithm trading from Septem-
on five-minute interval data of SPY ber 20 to October 14,
(AMEX: Standard & Poor’s 500) start- 2003.
ing on July 21, 2003, and optimized The visual picture
through September 20, 2003, on a range of equity and trades
of both entry and exit times (see Figure (Figure 8) shows that FIGURE 8: TRADING STRATEGY PAPER TRADING RESULTS. Top chart is five-minute
6). I also used two indicators, the mov- while the optimization SPY price bars, and the bottom the equity curve. Long entry is shown with a blue up
arrow, and short by a down red arrow. Exits are shown with down blue triangles and red
ing average convergence/divergence was done on periods
up triangles. This is the paper-trading phase that uses the parameters set by the
(MACD) and relative strength index (RSI) of principally side- optimization period. Equity performance is one of a few charts available to examine
for long and short entry, each having a ways motion, there are performance.
Reprinted from Technical Analysis of STOCKS & COMMODITIES magazine. © 2003 Technical Analysis Inc., [phone], http://www.traders.com
FIGURE 9: NEURAL NETWORK INPUT INTERFACE. Shown are the six indicators that will
be used to create a neural network solution for trading the underlying security, which in this FIGURE 10: FIVE MINUTE SPY PRICE (TOP) AND EQUITY (BOTTOM). The gray price
case will be SPY at five-minute intervals. Any of the 800-plus indicators can be used, or bars are the ones used for neural network training, while the green bars are the ones used
other price or market data. One of the indicators, %R, has been expanded to show the to do the evaluation or paper trading. Entry and exit for longs and shorts are shown the same
parameters used in building the indicator, which in this case uses a range of periods from way as for trading strategies, with blue up arrows for long entry, blue down triangles for long
1 to 10. exit, red down arrows for short entry and red up triangles for short exit. Red and blue Xs
show the actual entry/exit fill price for longs and shorts. The equity curve is the result of
optimizing using the criteria of maximum percentage gain. The inputs are in Figure 9, while
the percentage choice the genetic algorithm made in optimizing is shown in Figure 11.
few serious drawdowns, in part due to input “Other Instrument Data,” mean- apparent that the genetic algorithm is a
limiting the number of shares traded. Ex- ing you can input other price or market powerful feature since it is always look-
amining the long and short entries/exits data. The plus signs to the left of each of ing for an optimal solution. Again, you
shows that the technique was successful. the inputs is the same as for trading get to define what is meant by optimal —
rules: associated with each indicator is for example, maximum percentage gain
NEURAL NET PREDICTIONS a range of parameters that will be opti- or minimum drawdown — but in the
As the name infers, NeuroShell Trader mized. Again, you can change the range case of neural network solutions, you get
provides neural net solutions. The user of the parameters. to choose from an additional five that are
interface is similar to that of inserting a When you are finished with your in- neural network–unique, giving you a
trading strategy. You start by inserting puts, your price chart (Figure 10) is choice of 23 different target solutions.
a prediction into a chart. Instead of annotated with the buy and sell entries An interesting twist that NeuroShell
filling in the six tabs associated with a and exits, but the price bars are also Trader provides is that rather than
trading strategy, you have only one (Fig- color-coded to indicate which were used show the Sharpe ratio†, one of your
ure 9), and going through “Add Inputs” for training and which for evaluation optimal solution choices is maximum
gets you to the same indicator wizard (paper trading). Sharpe ratio.
you have used elsewhere. You can also The results of the neural network train- With six inputs, you probably want to
ing include optimiza- know what was finally found by the ge-
tion. Associated with netic algorithm as the optimal solution.
each of the indicators Because the genetic algorithm is optimiz-
is a range of param- ing using the input parameter ranges as
eters, and as part of choices, some indicators end up being
the neural network so- used more than others. In this case, the
lution, the genetic al- optimized solution to five-minute SPY
gorithm (professional bars shows (Figure 11) that stochastic
version only) will be %D, volume price trend, and money flow
used to find the opti- index (MFI) are noncontributors to the
mal neural network optimal solution.
inputs and thresholds. Does it get better? Yes. You can select
Neural network solu- different walkforward periods, how many
tions are calculated as you want to use, and how many you want
the genetic algorithm to optimize. NeuroShell Trader defines
searches for the opti- walkforwards as containing both the train-
mal inputs and thresh- ing period and evaluation period. The
FIGURE 11: NEURAL NETWORK PREDICTION RESULTS. Six indicators, each with
olds. After using this dates you input to the neural network
a range of parameters such as the number of periods used in Williams’ %R, were used
to find the optimal solution, which in this case was maximum percentage gain. product, it becomes allow you to stagger training and evalua-
Reprinted from Technical Analysis of STOCKS & COMMODITIES magazine. © 2003 Technical Analysis Inc., [phone], http://www.traders.com
tion. It would be as if you ran a training The interface to TradeStation allows support people are, but it didn’t work in
many months ago and then decided you the import of both data and indicators. I this case, because NeuroShell calls you
wanted to train again but with more recent used MetaStock end-of-day data and after the first week to see if you need any
data. You can also access other neural eSignal. Every screen can be moved help. It is easy to understand why this
network models using the plug-ins. and sized. I turned on all of the options. software is the readers’ first choice.
For new users this might be a bit daunt-
SUMMARY ing. I exhausted the video clips, but they Dennis D. Peterson is a Staff Writer for
What becomes apparent as a strong were a great start. I liked the brevity. STOCKS & COMMODITIES.
feature with NeuroShell Trader is that it Finally, I liked the syntax-free way of
allows the recursive use of inputs. A constructing indicators and strategies.
neural network can be the input to a This is one of the most user-friendly
trading strategy or another neural net. interfaces I have used. I installed plug-
An indicator can be the input to another ins with ease. I use a combination of
indicator. Trading strategy signals can visual clues and indicators in my own
be input to neural nets, other trading trading, and I would recommend some
strategies, or indicators. Coupled with drawing tools be added.
the genetic algorithm optimization of I tried to keep my identity as a re-
the professional version, this is a pow- viewer a secret during the review pro-
erful piece of software. And fast! cess to get an idea of how helpful the S&C