Neural Network Prediction Output function - seems strange

Neural Network Prediction Output function - seems strange

Kiran
2017-03-08 10:12:48

1) I have a 3-day prediction model (attached image) - since the prediction is for 3 days later, shouldn't the Output function be perf(close,-3) ? Why is it perf(close,3)?

2) Also, what does WS and Lag columns mean? How do i use them?

3) If i have multiple symbols in the model, is the model created to best fit each of the symbols? Would i get a more accurate model if i had only 1 symbol?

thanks
Kiran

QuantShare
2017-03-09 12:29:45

1/ You can't attach images in the forum. Please send an email to support.

2/ If you have "close" as Input.
Putting WS = 2 means that QuantShare should generate two inputs
close
ref(close, 1)

Putting Lag = 1 means that should use
ref(close, 1)
instead of
close

Please check our documentation for more info.

3/ For the more accurate model that depends. If you have multiple symbols then only one network is used to train all these symbols.

Kiran
2017-03-10 01:19:40

Thanks, pl also answer this question - I have a 3-day prediction model. since the prediction is for 3 days later, shouldn't the Output function be perf(close,-3)? Why do the examples show Output functions of perf(close,1) for 1-day predictions, instead of perf(close,-1) ?

QuantShare
2017-03-10 03:32:01

The example is for predicting tomorrow rate of change.
If bar 1 to 10 are used for the training then the prediction automatically occurs at bar 11.
perf(close, 1) allows you to predict the rate of return using the close at bar 10 and 11.

If you use perf(close, -1) then the rate of return will be based on the close of bar 11 and 12.

Of course, for a 3-day prediction model, you should not use perf(close, 3).
Instead of adding a formula, you can also select "Symbol Return" under the "Type column" and select the number of bars.

Kiran
2017-03-11 14:00:11

Thanks,
In the AI Training Results, what's the difference between Directional Accuracy and Prediction Sign Accuracy? Shouldn't they both indicate the % of times the prediction direction was correct? How are they calculated?

QuantShare
2017-03-13 05:54:48

The "Directional Accuracy" is the percentage of time the upward or downward movement of the data was correct.
The "Prediction Sign Accuracy" is the percentage of time the prediction sign (positive/negative) was correct.

Kiran
2017-03-17 00:18:02

Sorry, still confused - isnt that the same thing? Prediction sign is predicting direction .. so, when Direction accuracy is correct, Prediction is correct.
What am i missing? Why are they so different - i'm getting 55% direction accuracy and 85% prediction sign accuracy

thx
kiran

QuantShare
2017-03-17 04:00:05

Sign here is not direction. It is whether a number is above 0 or not.
Depending on what you are trying to predict, prediction sign accuracy could be interesting or not.

Kiran
2017-03-27 18:04:08

Hi Azouz,

"Sign here is not direction. It is whether a number is above 0 or not."

Sorry this still doesn't make sense, so let's walk through an example. My prediction output is Symbol Return 3 bars later.
If the actual market return was bullish (+ve) 60% of the predicted days and berish (-ve) 40% of the days, but
the prediction was bullish (+ve) only 50% of days and bearish 50% of days.

Prediction sign accuracy is % of time predicted correctly = 50 bull days + 40 bear days = 90%
Direction accuracy is also the same: it's accurate 50 bull days and 40 bear days = 90%

What's the difference? they both are the same. Why am i getting dramatically different values for the two?

thanks
Kiran

QuantShare
2017-03-28 11:18:30

Here is an example:

Predict stock one-bar rate of return.

For day 0:
Predicted value: 2%
Next bar value: 1%
Prediction sign: 1 (true)
Direction accuracy: 1

For day 1:
Predicted value: 0.1%
Next bar value: -1%
Prediction sign: 0 (false)
Direction accuracy: 1 (because the prediction value moved down from 2 to 0.1 and the actual value also moved from 1 to -1

Kiran
2017-04-23 14:20:51

So, Direction Sign is the change in Predicted Value from previous bar to next, correct?
Why is Direction Sign useful and how does having a high Direction Sign Accuracy help?

I can see why Predicted Value is useful to determine Long vs Short direction, if Prediction accuracy is high.
However, my training model is giving high Direction Sign accuracy (85%) but not Prediction accuracy (55%)
- how can is use this model to to construct a profitable trading system?

Get more messages

No more messages

Back


Trading Items Choose function Stock market prediction using traderbots NORMDIST - Normal Probability Density Function Day of week and Week number function Average Output Per Year	How-to Lessons How to create and trade a Neural Network model How to optimize a neural network using a genetic algorithm How to calculate the average of a time series using the money man... How to debug a trading system using the money management tool How to create a hatched area in a chart
Related Forum Threads neural network Neural Network Strategy error Interpret Prediction results and making it useful? Save Rules Analyzer Output Composite Function	Blog Posts Stock Market Prediction Stock Market Prediction with QuantShare Forecast volatility using trading rules and neural networks The 'inside period' function How to create market indicators using the composite function - Pa...