rank_futures_spreads.py

import databento as db
import pandas as pd


client = db.Historical()


def rank_by_volume(top=500):
    """
    Returns instrument IDs of instruments that traded most, in descending rank
    """
    data = client.timeseries.get_range(
        dataset='GLBX.MDP3',
        symbols='ALL_SYMBOLS',
        schema='ohlcv-1d',
        start='2023-08-15',
        end='2023-08-15'
    )
    df = data.to_df()
    return df.sort_values(by='volume', ascending=False).instrument_id.tolist()[:top]


def get_symbol_mappings(instrument_id_list):
    """
    Fetch raw symbols and tick sizes of specified instrument IDs
    """
    data = client.timeseries.get_range(
        dataset='GLBX.MDP3',
        stype_in='instrument_id',
        symbols=instrument_id_list,
        schema='definition',
        start='2023-08-15',
        end='2023-08-15'
    )
    return data.to_df()[['instrument_id', 'raw_symbol', 'min_price_increment']]


def get_median_spread(instrument_id_list, metadata):
    """
    Get median spread of specified instrument IDs, in descending rank
    """
    data = client.timeseries.get_range(
        dataset='GLBX.MDP3',
        stype_in='instrument_id',
        symbols=instrument_id_list,
        schema='tbbo',
        start='2023-08-15T15:00',
        end='2023-08-15T16:00'
    )
    df = data.to_df().merge(metadata, on='instrument_id', how='inner')
    df.dropna(subset=['ask_px_00', 'bid_px_00', 'ask_sz_00', 'bid_sz_00', 'min_price_increment'], inplace=True)

    # Compute median spread at times of execution
    spread_ranking = df.groupby(by='raw_symbol').apply(lambda r: ((r.ask_px_00 -
                                                            r.bid_px_00)/r.min_price_increment).median()).sort_values().apply(round)
    touch_ranking = df.groupby(by='raw_symbol').apply(lambda r: ((r.ask_sz_00 +
                                                            r.bid_sz_00)/2).median()).apply(lambda x: round(x,1))
    ranking = pd.concat([spread_ranking, touch_ranking], axis=1)
    ranking.columns = ['median_spread', 'median_touch']
    return ranking


volume_ranking = rank_by_volume(500)
metadata = get_symbol_mappings(volume_ranking)
spread_ranking = get_median_spread(volume_ranking,
                                   metadata)
spread_ranking.to_csv('ranking.csv')