Order Book Dataset

[bukosabino]

Goal

Get Order Book as another dataset. The dataset can give us X bids/asks sampled every Y minutes/days. Example: 20 bids/asks sampled every minute.

We can get these data using:

context.exchanges[context.EXCHANGE].get_orderbook(symbol('etc_btc'), limit=20)

Inspiration

• https://medium.com/@alexrachnog/ai-for-algorithmic-trading-7-mistakes-that-could-make-me-broke-a41f94048b8c
• https://github.com/produvia/kryptos/issues/8
• https://www.amazon.com/Advances-Financial-Machine-Learning-Marcos/dp/1119482089
• https://github.com/enigmampc/catalyst/issues/381

Problem

We cannot get the historical order book, you only get the actual order book situation. Any ideas how it could be useful for us?

[slavakurilyak]

@bukosabino We either create our own dataset or buy one. Kaiko comes to mind, which collects:

• 1% Order Books: orders placed at +/- 1% of the mid price
• 10% Order Books: orders placed at +/- 10% of the mid price
• Full Order Books: available upon request

If we are to create our own dataset, we can do so by starting with daily order-book data. This new dataset will then be used for future machine learning predictions. There are a few public Github repositories which collect orderbook data.