NeuroForecaster/GENETICA version 5
Copyright 1995-1996, NIBS Inc. 
62 Fowlie Road, Republic of Singapore 428501.
Email: midaz@singnet.com.sg
Tel (+65)3442357, Fax (+65) 3442130.

Installation:

1. For Zipped File: eg. nfga.ZIP

Drag and drop it to unzip.exe or winzip.exe to unzip it, then 
run the setup.exe.

unzip nfga.zip
pkunzip -d nfga.zip

2. For Floppy Disk with Setup.exe:

Insert the NFGA program diskette into the floppy disk drive, 
select the Program Manager menu function "File\Run...", type
"a:\setup" or "b:\setup" depending on whether Drive a or 
Drive b is used.

3. For Self-extracing File: eg. nfgazip.exe

Double click on it to explode it, then run the setup.exe.

nfgazip.exe -d


Note - Manual Setup: If the above setup method does not 
work or the programs you receive does not have the setup.exe 
program, create a new directory (eg. c:\NFGA) in your hard 
disk and copy all the files from the diskette to the new 
directory.  Use the Windows Program Manager's "File\New..." 
menu to create the program group and items, and click on the 
"Browse" and "Change Icon" buttons, and set the correct 
directory path.  Check to make sure that you have all the 
above files.


1. What is NeuroForecaster?  

NeuroForecaster is an advanced windows-based, user-friendly, 
intelligent, neural network forecasting tool.  It is packed 
with the latest technologies including neural network, fuzzy 
computing and non-linear dynamics.  It excels at:

Time-Series Forecasting
eg.  Stock & Currency Market Forecasts, GDP Forecast
	 
Classification
eg.  Stock Selection,  Bond Rating, Credit Assignment, 
Property Valuation
	  
Indicator Analysis
Identification of useful input indicators

2. What is GENETICA Net Builder?  

GENETICA Net Builder is a search and optimization technique 
based on  genetic algorithms, designed to search for the 
best combination of input data, the most predictable 
forecaster horizon (i.e., how many steps ahead to forecast) 
for a given problem, and automatically creates and 
optimizes forecaster structures and control parameters by 
evolution and genetic search.  

3. Combining the Power of Neural Networks and GA
   
NeuroForecaster/GA combines the forecasting power of 
NeuroForecaster and the advanced search and optimization 
capabilities of GENETICA within an integrated user interface.  
With its revolutionary technology, major performance gains and 
functional improvements, NFGA retains the ease of use for which 
NeuroForecaster is renowned.  From time-series forecasting to 
classification problems, it provides the ideal tool for 
predictive analysis, charting and presentations.  Complete 
with a dozen of fully-featured applications and a host of 
useful functions, NFGA is the most exciting value-packed 
intelligent forecasting software available!   In the automatic 
mode, networks are trained and cross-validated for their 
generalization performance via out-sample tests, and the best 
performing nets are saved with incremented file extensions 
until the expected performance is reached and the training 
process stops.  With the C source code, DLL and example 
program, trained forecasters can be embedded and run from 
within your DOS or Windows applications.

4. Features of NeuroForecaster/GA

Best Net Search
Best Input Data Combination Search 
Best Forecast Horizon Search
Auto Test
Auto Save
Auto Stop
		
12 neural network models including Backprop, Fastprop, 
Radial Basis Function, NeuroFuzzy	

Multitasking & background training mode 		

Unlimited network capacity, unlimited number of chromosomes	

Rescaled range analysis & Hurst exponent to unveil hidden 
market cycles & check for predictability	

Correlation Analysis to compute correlation factors to 
analyze the significance of input indicators 		

Weight Histogram to monitor the progress of 
learning

Accumulated Error analysis to analyze the strength of 
Input Indicators	

Ease of Use:

Easy to install and ready to run
All functions are menu and button-driven
Color graphic user interface
Comprehensive User's Manual & Technical Reference with 
tips for neural network training 

5. Financial, Business & Engineering Applications 
Included in the Package:

Stock price prediction
Stock market 6 monthly returns forecast
Stock selection based on company ratios
US$ to Deutschmark exchange rate forecast
US$ to SGD exchange rate forecast
US$ to Yen exchange rate forecast
Credit rating of bank loan applications
Property price valuation
Construction demand forecast
Soccer Match Prediction
XOR - A classical problem
Chaos - Prediction of Mackey-Glass chaotic time series
SineWave - To demonstrate Rescaled Range analysis

6. Other Applications:

Stock market index forecast 
Interest rate forecast 
Commodity trading 
Bond rating 
National GDP forecast 
Sales forecast
Fraud detection and Fault diagnosis 
Business site selection 
Rain-fall & catchment run-off prediction 
Air passenger arrival forecast 
Recruitment screening , Selection of insurance policies 
Electricity Demand Forecast
And many others

7. More on GENETICA Net Builder Option

Genetic Algorithm (GA) is very effective at finding 
optimal or near optimal solutions to a wide variety of 
problems including neural network design, because it does 
not impose many of the limitations required by traditional 
methods.  Based on Darwinian evolution theory, GENETICA 
generates many possible networks which are evaluated, 
purged and recombined to produce better offspring to inherit 
the knowledge acquired by their parents.  Occasionally 
mutation will take place to ensure that networks will not 
be trapped in local minima during the learning process.

The integration of GENETICA with NeuroForecaster is 
seamless - All you need to do is to specify the GA model 
when creating new networks, and GENETICA will be activated.  
Genetic algorithm is used to search for the best combination 
of input data and the number of steps ahead to forecast from 
the input and output data window, and construct the best 
forecaster network structures.

Initialization of chromosome population, training, testing, 
purging, crossover, mutation, setting of learning rate, 
tolerance level are carried out automatically.  

The neural network training and evolution process can be 
stopped by preset iteration count, training error or 
testing error level.

Unlike most of other neural network simulator programs, 
NFGA allocates memory dynamically, so there is no preset 
limit on the data file size and number of chromosomes 
allowed -- be it 10 years of daily data, 1000 chromosomes 
or more.  The learning rate, momentum rate, tolerance 
level and evolution cycle can be set manually to auto mode 
and determined by the program.

The program will automatically determine the appropriate 
time to test the generalization capability of the network 
population, save the best nets and stop the learning process 
when the expected performance is reached.

One genetically evolved forecaster was trained with 9 
financial ratios to forecast the average price per share of 
a NYSE listed stock.  The downward turn was detected, 
signifying a sell signal.  The entire learning and evolution 
process took only 15 minutes on a 486 PC.  This pre-trained 
forecaster and the financial ratios used are included in the 
package.

7. Review on NeuroForecaster

Please see the May 1995 issue of the Technical Analysis of
Stocks and Commodities, page 85-89.  The study showed that
a neural network built using NeuroForecaster far outperformed
other competing approaches, and produced the best forecasting
results.  Send in your mailing address if you wish to receive
a reprint of that review paper.

8. NFGA Outperforms others with stunning accuracy

All the pre-trained forecasters illustrated below and others 
are included in the NeuroForecaster/GA package.

Title:   US$ to DM Daily Forecast
Type:  Time Series
Training Time:  25 mins (with GENETICA)
The genetically evolved forecaster was trained with 24 
technical indicators exported from a Computrac TradePlan.  
The network generated very accurate forecasts with correct 
turning points.  The data and network files are US$_DM2.txt 
and US$_DM2.ga respectively.

Title:   S&P500 Monthly Forecasts
Type:  Time Series
Training Time:  20 mins
The upward surge at the end of training was detected with a 
stunning accuracy.  Both the speed and accuracy in direction 
and magnitude outperform those of NeuroShell and others!
Other Similar Applications:  DJIA index forecast, Individual 
stock price forecasts

Title:   Stock Selection
Type:  Classification
Training Time:  15 mins
Trained with 13 company ratios for rating of other stocks 
based on their expected returns.  The actual returns confirmed 
the forecast.
Other Similar Applications:  Credit scoring, Real Estate 
Appraisal, bond rating, Fault diagnosis, Fraud detection

Title:   USD/YEN Daily Closing Forecast
Type:  Time Series
Training Time:  30 mins
Based on Gold price, DJIA, NYSE and cross rates including Yen, 
DM, GBP which are readily available.  The results and speed 
outperform BrainMaker Professional.    
Other Similar Applications:  USD/DM rates, Gold price forecast

Title:   USD/SGD 6-Month Ahead Forecast
Type:  Time Series
Training Time:  30 mins
The same set of data was trained on BrainMaker Professional 
and other neural net tools but none converge even for 1-month 
ahead forecast.
Other Similar Applications:  Mid to long-term GDP forecast, 
Sales, Market direction, Commodity Trading

Title:   Stock Price Prediction
Type:  Time Series
Training Time:  15 mins (with GENETICA)
The genetically evolved forecaster was trained with 9 
financial ratios to forecast the average price per share of a 
NYSE listed stock.  The downward turn was detected as the 
forecast started, signifying a sell signal.  The net was 
automatically created and optimized by GENETICA, which also 
searched for the best input data combination, number of steps 
ahead and control parameters.

9. Users' Feedback  

Since the launch of NeuroForecaster, we have received 
overwhelming response.  Our clients include United Nations, 
Bank of America, Smith Barney Shearson, Harris Upham Securities, 
BARRA, Merill Lynch, Mitsui, Public Financial Services of 
Germany, LBS Capital Management USA, US Central Supply, 
Prudential, Fidelity Investment, 3M, TeleSensory, Schonfeld 
Securities, Fininvest, Intelligent Financial Services, Chevron, 
Kaman Sciences Corp, Uniplan Software S.R.L, Morning Side, 
Ecole des Hautes Etudes Commeciales, NeuralWare, NeuroDynamx, 
Banca Comm Italiana, Small System Software, Chroous Corp, 
PRM Austria, OTE S.p.A. Italy, F.M.R. Computers & Software 
Ltd Isreal, Sprout Sdn. Bhd, InterConsult S.R.L., 
Petty-Bierut, Yoshiaki Kato, Centro Auxologico Italiano, 
Barclay Bank of England, InterPool USA, ING Bank Netherland, 
Ngee Ann Polytechnic Neural Network Center, Information 
Technology Institute of the National Computer Board, and the 
School of Building & Estate Management, Faculty of Engineering, 
Schools of Business Administration of many well-known 
universities and private investors.  Here are some of the 
feedbacks we received:

"We use it for daily USD/DM forecast, the results have been 
very successful."  R. Magenreuter, Public Financial Consulting, 
Germany, Aug 93.

"I run NeuroForecaster on stock data and I can get 40 correct 
trades out of every 50 on the average.  The graphical interface 
and overall design is much superior than others."  D. Lee, 
Smith Barney, Singapore, Jun 93.

"Congratulations.  We are pleased and delighted to have a good 
run of NeuroForecaster.  It is very interesting indeed.  
Certainly, we felt that such a package is well suited for 
the end-users."   Dr. S. Tan, Curtin University of Technology, 
Australia, Jun 93.

"An excellent product as it is user friendly for forecasting.  
I will include it in my courses in commoditiy futures markets 
to traders and hedgers at the commodity exchange."   Dr. M. 
S. Boyd, Manitoba, Canada, Jun 93.

"The results of your product has struck me as quite congruent 
with our experiment using a genetic algorithms shell combined 
with a neural network tool;  But they do not have the 
built-in application layer for forecasting like yours." K. 
Itanis, ITANIS International Inc., Aug 93.

"We have the highest praises for your program, especially the 
GENETICA.  I had conveyed these positive and superb features to 
Dr. DeBoeck of the World Bank, Dean Barr of LBS and Prof. 
Mark Fishman of Eckerd College, software reviewer." P. K. 
Holtz, Pretoria, 16 Jan 94.

"I'm very impressed with your product.  It beats BrainMaker 
Pro & its Genetic Training Option easily."  R. Gunn, 
Australia, May 94.

"Your NeuroForecaster/GA is very impressive." Zeno Lampe, 
Lampe Automatisering, Holland, Jan 94.

"We have used NeuroShell, BrainMaker, HNet and NeuralWare.  
None of them can come close to its performance."  H. 
Iskandar, Greece, Jun 94.

"I am particularly impressed with the genetically produced 
forecasters that I have generated so far." Dr. Pete Snow, 
Kaman Sciences USA, Jan 94.

"I am pleased with it and impressed with its features." Prof. 
J. Roy, Ecole des Hautes Etudes Commeciales, Canada, Jul 93.

"I am impressed by its capabilities and features.  The 
materials I've received looks interesting." F. Salvatori, 
Canada, Apr 93.

"I am very impressed." K. Rowe, Buchanan Partners, UK, Sep 93.

"I've got very good results from NeuroForecaster for US 
stock market forecast."  J. Krosp, Oxford Intelligent System, 
Inc., USA, Sep 93.

"After using your NeuroForecaster, I found that it is very 
much superior and user friendly than BrainMaker Pro.  A 
lot of people bought BrainMaker have given up after trying 
so hard to use it.  Your program is user friendly and the 
display is first class."  S. Ooi, May 94, Multi-purpose 
Commodity Trading, Malaysia.


10) VisuaData 2.1 for Windows

VisuaData for Windows (ViDA) is a user-friendly 
data management program which reads MetaStock, CSI, 
Computrac & ASCII data files directly, generates 
over 108 commonly used and new technical indicators 
including Chande's VIDYA & CMO, Percent Price Change, 
RWI, CCI, ADX, VHF, SAR, TSF, OBV, ULTIMATE, etc., and
mathematical operations.  It outputs ASCII data files 
ready for use by popular neural networks and other 
programs such as BrainMaker and NeuroForecaster/
Genetica.  Multiple data files can be opened and 
processed simultaneously, in spreadsheet-like data 
grids for easy data manipulation.  Other user-friendly 
features include in-cell graphics which supports 
drag-and-drop plots on standard price-volume charts, 
candle-stick chart, line charts, bar charts,  
context-sensitive help, smart-icon tool bars, tutorial 
sessions and examples. Unlimited free upgrades.

11) List of Technical Indicators and Maths Operations
above - true if first signal above second
accel - acceleration
adv - accumulation distribution volume 
advosc - adv oscillator
adx - average directional movement index
adxr - adx rating (adx + adx Period ago)/2
ama - moving average adaptive
atr - average true range
bbandb - bollinger band bottom signal
bbands - bollinger bands
bbandt - bollinger band top signal
bbandx - bollinger band close crossing
below - true if first signal below second 
cci - commodity channel index
chaikin - chaikin oscillator
change - change in value over the interval
changep - change in percent
cmo - chande momentum oscillator
cmoc - chande momentum count
cmot - chande momentum turning points
cross - true if first signal crosses above second 
di - directional movement index (wilder's) 
dic - crossing of pdi and mdi
dm_adx - average directional movement index
dm_adxr - adx rating
dm_di - directional movement index (wilder's) 
dm_dic - crossing of pdi and mdi
dm_dx - net directional movement
dm_mdi - minus directional movement downtrend index
dm_pdi - plus directional movement downtrend index
dx - net directional movement
ema - moving average exponential
filter - filter out insignificant moves
gapdown - true if current values gapped below last
gapup - true if current values gapped above last
hlc - high low close prices combination
inside - true if current values are inside previous ones 
log - natural log 
lsf - linear regression using least-squares-fit line 
macd - moving average convergence/divergence
macd fast - fast macd
macdbsc - macd crossing buy and sell counts
macdlx - last cross of macd and signal
macdosc - macd oscillator
macds - macd difference line slope
macdt - macd trigger signal
macdt fast - fast macd trigger signal
macdts - macd trigger signal slope
macdwt - macd with trigger
macdwt fast - fast macd with trigger
macdx - macd crossing signal
macdxc - macd crossing count
mdi - minus directional movement downtrend index
mfi - money flow index
mlr - moving linear regression
mo - momentum
mov3e - triple exponential moving average
mov3ewt - triple exponential moving average
mova - moving average adaptive
move - moving average exponential
movs - moving average simple
movt - moving average triangular
movv - moving average variable
movw - moving average weight
mp - median price 
nvi - negative volume index 
nviosc - negative volume index oscillator
obv - on-balance volume 
obvosc - on-balance volume oscillator
outside - true if current values are outside previous ones 
pdi - plus directional movement uptrend index
pvi - positive volume index 
pviosc - positive volume index oscillator
pvt - price volumn trend 
pvtosc - price volumn trend oscillator
qstick - quantitative candlestick
regress - difference between data and moving least-squares-fit
roc - rate of change
roc volume - rate of change of volume
rocp - rate of change in percent
rocs - regression oscillator
rsi - relative strength index (wilder's) 
rsic - relative strength index count
rsit - relative strength index turning count
rwih short - random walk index based on high price (short term) 
rwih medium - random walk index based on high price (medium term) 
rwih long - random walk index based on high price (long term) 
rwil short - random walk index based on low price (short term) 
rwil medium - random walk index based on low price (medium term) 
rwil long - random walk index based on low price (long term) 
sar - parabolic stop and reversal
stddev - standard deviation
stddevc - standard deviation directional change
stoch fast - stochastic %K and %D
stoch slow - slower stochastic %K and %D
stoch%d fast - stochastic %D
stoch%d slow - slower stochastic %D
stoch%k fast - stochastic %K
stoch%k slow - slower stochastic %K
stochobos - stochastic overbought oversold
stochosc - stochastic oscillator
tema - triple exponential moving average
temawt - triple exponential moving average with trigger
tma - moving average triangular
tr - true range 
trix - triple exponential moving average
trixwt - triple exponential moving average with trigger
tsf - time series forecast indicator
ultimate - william's ultimate oscillators
vhf - vertical horizontal filter
vidya - moving average variable
volatility - volatility
volatilityc - volatility (chaikin) using high low prices
w%r - william's %r
wad - william's accumulation/distribution indicator 
waddrf - wad daily raw figure based on buy and sell pressure
wma - moving average weight
zigzag - zigzag modifier
Add 2 columns
Substract 2 columns
Multiply 2 columns
Divide 2 columns
Add value to column
Subtract value from column
Multiply column by value
Divide column by value
Square root of column
Absolute values of column
Statistics of column..
Detrend by difference
Detrend by percentage difference
Detrend by log difference
Detrend by least square
Outliers, Identifying

For more info, please contact:
NIBS Pte Ltd, 62 Fowlie Road, 
Republic of Singapore 428501. 
Fax (+65) 344 2130 Tel (+65) 344 2357    
Email  midaz@singnet.com.sg
