How to define a list of shares (20 us shares for example) for predicting? Rather than the entire nasdaq100 or sp500?

[orchardc]

How to define a list of shares (20 us shares for example) for predicting? Rather than the entire nasdaq100 or sp500?

Is there a sample?

Many thanks!

[you-n-g]

There are two approaches I come up with now.

1. You can follow the example here ~/.qlib/qlib_data/cn_data/instruments to define your own universe. And create a dataset based on this universe.
2. If you want to train the model on a larger universe and test data on a smaller universe, you can filter your data when model inference. You can define some processors which only apply to inference data https://github.com/microsoft/qlib/blob/main/qlib/data/dataset/handler.py#L442

[orchardc]

Thank you-n-g, for your #1 suggestion.

For quick testing:
I create a file called: nasdaq2.txt that includes 2 records copied from the sample file, then put in the instruments folder.

Then change the market = "NASDAQ2" in the .ipynb file.

Is that everything or do I need to change something else? The purpose is only to use these 2 stocks for training and testing.

In the prediction result, I got:

[orchardc]

Any idea? @you-n-g

[you-n-g]

@orchardc
It looks like your model learned nothing (the initial model gets the lowest loss and predicts the same value for all samples).

[you-n-g]

I think more data will be helpful

[orchardc]

Thanks @you-n-g , I think you are right! With more stocks, I start seeing some data...although I'd like the model to only use selective stocks for training/testing.

[orchardc]

The NASDAQ prediction result isn't good, but I guess the sample workflow was designed for CSI300, not for the US market.