Model Buy/Sell, How do I model Buy/Sell Options

[Scott Watson]

Hello to the Group

Currently I have my output {1=Sell, -1=Buy 0=Do Nothing}

My problem is that Synapse is looking at that and and building a model that matches the zeros because the data has 90% zero's.

So my question is how do I model the model so that it only tries to match the 1's and -1's?

Thank you for your Time
Scott Watson

[Scott Watson]

Hello to the Group

Currently I have my output {1=Sell, -1=Buy 0=Do Nothing}

My problem is that Synapse is looking at that and and building a model that matches the zeros because the data has 90% zero's.

So my question is how do I model the model so that it only tries to match the 1's and -1's?

Thank you for your Time
Scott Watson

I think if people (like me) have these type of basic questions they should go and DO the tutorials and not just read them. Once you actually do them, you will gain a much greater understanding of the software.

Thank you
Scott Watson

[kesk]

Hello,

Just started trying out Synapse. So a lot of questions come to my mind. But what are the inputs you have given and how did you get the output?

Your reply is not answering your own question. Would appreciate a bit more brief.

thanks

[Andre P.]

I read posts like this often, and work in trading, so I thought I'd add my two cents.

1. Synapse is not trading software. It is a data analysis, and a model building/testing package for machine learning algos. It does not have any built-in functionality for trading or trade simulation.
2. To use it for trading, you must use it with some other software, or roll your own. We use it with a high-end package that costs us 25k a year per client, but you can use it with something like Wealth-lab which is about $650. WL uses C# as the programming language and allows you to use external dlls such as the models output by Synapse.
3. Modelling approaches such as the one outlined in the original post will fail. The poster has already figured out why.
4. Broadly speaking, you can use two approaches. Try predicting futures percent returns of the series, or try calculating fair value and trade appropriately when it is out of line. Do not use raw price in, raw price out. Your model's scaling will be all wrong. Use pct changes i.e. use differentiation.
5. Another approach which has not worked for me but may work for you, is to set all your labels to either 1 for trending up, or 0 for trending down. You can further refine this to be 1 for up, 0 for sideways, and -1 for down. Then try using classification models such as SVM. Note that Synapse SVM component is a binary classifier so it will not work with 1, 0, -1 models, only with 1s & 0s.