Order Book Analytics#

onetick-py offers functions for analyzing tick-by-tick order book. There are three representations of an order book. We’ll show top 3 levels only for the ease of exposition.

A book can be displayed with a tick per level per side. We refer to a level in the book as a ‘price level’ or ‘prl’.

import onetick.py as otp

s = otp.dt(2024, 2, 1, 10)

prl = otp.ObSnapshot(db='CME_SAMPLE', tick_type='PRL_FULL', max_levels=3)
# we can use the same timestamp for the start an the end times when we just need a snapshot
otp.run(prl, symbols=r'NQ\H24', start=s, end=s)
Time PRICE SIZE LEVEL UPDATE_TIME BUY_SELL_FLAG
0 2024-02-01 10:00:00 17303.50 3 1 2024-02-01 09:59:59.737771351 1
1 2024-02-01 10:00:00 17303.75 1 2 2024-02-01 09:59:59.968007113 1
2 2024-02-01 10:00:00 17304.00 3 3 2024-02-01 09:59:59.823591575 1
3 2024-02-01 10:00:00 17300.25 4 1 2024-02-01 09:59:59.682656319 0
4 2024-02-01 10:00:00 17299.50 1 2 2024-02-01 09:59:59.798148709 0
5 2024-02-01 10:00:00 17299.25 1 3 2024-02-01 09:59:59.881654499 0

Alternatively, a book can show a tick per level with both ask and bid price/size info.

prl = otp.ObSnapshotWide(db='CME_SAMPLE', tick_type='PRL_FULL', max_levels=3)
otp.run(prl, symbols=r'NQ\H24', start=s, end=s)
Time BID_PRICE BID_SIZE BID_UPDATE_TIME ASK_PRICE ASK_SIZE ASK_UPDATE_TIME LEVEL
0 2024-02-01 10:00:00 17300.25 4 2024-02-01 09:59:59.682656319 17303.50 3 2024-02-01 09:59:59.737771351 1
1 2024-02-01 10:00:00 17299.50 1 2024-02-01 09:59:59.798148709 17303.75 1 2024-02-01 09:59:59.968007113 2
2 2024-02-01 10:00:00 17299.25 1 2024-02-01 09:59:59.881654499 17304.00 3 2024-02-01 09:59:59.823591575 3

Finally, all levels can be displayed in one tick.

prl = otp.ObSnapshotFlat(db='CME_SAMPLE', tick_type='PRL_FULL', max_levels=3)
otp.run(prl, symbols=r'NQ\H24', start=s, end=s)
Time BID_PRICE1 BID_SIZE1 BID_UPDATE_TIME1 ASK_PRICE1 ASK_SIZE1 ASK_UPDATE_TIME1 BID_PRICE2 BID_SIZE2 BID_UPDATE_TIME2 ASK_PRICE2 ASK_SIZE2 ASK_UPDATE_TIME2 BID_PRICE3 BID_SIZE3 BID_UPDATE_TIME3 ASK_PRICE3 ASK_SIZE3 ASK_UPDATE_TIME3
0 2024-02-01 10:00:00 17300.25 4 2024-02-01 09:59:59.682656319 17303.5 3 2024-02-01 09:59:59.737771351 17299.5 1 2024-02-01 09:59:59.798148709 17303.75 1 2024-02-01 09:59:59.968007113 17299.25 1 2024-02-01 09:59:59.881654499 17304.0 3 2024-02-01 09:59:59.823591575

We can output the book (in any of the three representation) on every change to price/size at any of the levels.

prl = otp.ObSnapshotFlat(db='CME_SAMPLE', tick_type='PRL_FULL', max_levels=3, running=True)
prl = prl.drop(r".+TIME\d")
otp.run(prl, symbols=r'NQ\H24', start=s, end=s + otp.Milli(100))
Time BID_PRICE1 BID_SIZE1 ASK_PRICE1 ASK_SIZE1 BID_PRICE2 BID_SIZE2 ASK_PRICE2 ASK_SIZE2 BID_PRICE3 BID_SIZE3 ASK_PRICE3 ASK_SIZE3
0 2024-02-01 10:00:00.000000000 17300.25 4 17303.50 3 17299.50 1 17303.75 1 17299.25 1 17304.00 3
1 2024-02-01 10:00:00.000005635 17281.75 1 17284.75 1 17281.50 2 17292.75 30 17281.25 1 17303.50 3
2 2024-02-01 10:00:00.000012009 17281.75 1 17284.75 1 17281.50 2 17303.50 3 17281.25 1 17303.75 1
3 2024-02-01 10:00:00.000035337 17276.00 4 17264.50 2 17275.75 1 17264.75 1 17275.50 7 17265.00 1
4 2024-02-01 10:00:00.002605599 17296.75 2 17264.50 2 17276.00 4 17264.75 1 17275.75 1 17265.00 1
... ... ... ... ... ... ... ... ... ... ... ... ... ...
10 2024-02-01 10:00:00.032787693 17300.25 2 17264.50 2 17299.25 15 17264.75 1 17297.00 1 17265.00 1
11 2024-02-01 10:00:00.043854201 17300.25 2 17264.50 2 17297.00 1 17264.75 1 17296.75 2 17265.00 1
12 2024-02-01 10:00:00.054669411 17300.25 1 17264.50 2 17297.00 1 17264.75 1 17296.75 2 17265.00 1
13 2024-02-01 10:00:00.074736715 17300.25 1 17254.00 9 17297.00 1 17264.50 2 17296.75 2 17264.75 1
14 2024-02-01 10:00:00.083903067 17300.25 1 17253.00 3 17297.00 1 17254.00 9 17296.75 2 17264.50 2

15 rows × 13 columns

The otp.ObSnapshot method doesn’t require specifying max_levels. The entire book is returned when the parameter is not specified.

prl = otp.ObSnapshot(db='CME_SAMPLE', tick_type='PRL_FULL')
otp.run(prl, symbols=r'NQ\H24', start=s, end=s)
Time PRICE SIZE LEVEL UPDATE_TIME BUY_SELL_FLAG
0 2024-02-01 10:00:00 17303.50 3 1 2024-02-01 09:59:59.737771351 1
1 2024-02-01 10:00:00 17303.75 1 2 2024-02-01 09:59:59.968007113 1
2 2024-02-01 10:00:00 17304.00 3 3 2024-02-01 09:59:59.823591575 1
3 2024-02-01 10:00:00 17304.25 3 4 2024-02-01 09:59:59.668640001 1
4 2024-02-01 10:00:00 17304.50 4 5 2024-02-01 09:59:59.767992495 1
... ... ... ... ... ... ...
1570 2024-02-01 10:00:00 11111.00 1 782 2024-01-31 17:59:59.998000000 0
1571 2024-02-01 10:00:00 10000.00 1 783 2024-01-31 17:59:59.998000000 0
1572 2024-02-01 10:00:00 9600.00 1 784 2024-01-31 17:59:59.998000000 0
1573 2024-02-01 10:00:00 622.00 1 785 2024-01-31 17:59:59.998000000 0
1574 2024-02-01 10:00:00 200.00 1 786 2024-01-31 17:59:59.998000000 0

1575 rows × 6 columns

Book Imbalance#

Let’s find the time weighted book imbalance. The imbalance at a given time is defined as the sum of the bid sizes at the top x levels minus the sum of the ask sizes at the top x levels divided by the sum of these two terms: the values close to 1 mean the book is much heavier on the bid side, close to -1 – on the ask side, equal to zero means the sizes are the same.

We display top 3 levels of the book first on every update at any of these levels. There are three ticks (one per level) to represent the book after each update.

x = 3
prl = otp.ObSnapshotWide(db='CME_SAMPLE', tick_type='PRL_FULL', max_levels=x, running=True)
otp.run(prl, symbols=r'NQ\H24', start=s, end=s + otp.Milli(100))
Time BID_PRICE BID_SIZE BID_UPDATE_TIME ASK_PRICE ASK_SIZE ASK_UPDATE_TIME LEVEL
0 2024-02-01 10:00:00.000000000 17300.25 4 2024-02-01 09:59:59.682656319 17303.50 3 2024-02-01 09:59:59.737771351 1
1 2024-02-01 10:00:00.000000000 17299.50 1 2024-02-01 09:59:59.798148709 17303.75 1 2024-02-01 09:59:59.968007113 2
2 2024-02-01 10:00:00.000000000 17299.25 1 2024-02-01 09:59:59.881654499 17304.00 3 2024-02-01 09:59:59.823591575 3
3 2024-02-01 10:00:00.000005635 17281.75 1 2024-02-01 09:59:50.090523363 17284.75 1 2024-02-01 10:00:00.000005635 1
4 2024-02-01 10:00:00.000005635 17281.50 2 2024-02-01 09:59:57.426381413 17292.75 30 2024-02-01 10:00:00.000005635 2
... ... ... ... ... ... ... ... ...
40 2024-02-01 10:00:00.074736715 17297.00 1 2024-02-01 10:00:00.006940703 17264.50 2 2024-02-01 10:00:00.000035337 2
41 2024-02-01 10:00:00.074736715 17296.75 2 2024-02-01 10:00:00.002605599 17264.75 1 2024-02-01 10:00:00.000035337 3
42 2024-02-01 10:00:00.083903067 17300.25 1 2024-02-01 10:00:00.054669411 17253.00 3 2024-02-01 10:00:00.083903067 1
43 2024-02-01 10:00:00.083903067 17297.00 1 2024-02-01 10:00:00.006940703 17254.00 9 2024-02-01 10:00:00.074736715 2
44 2024-02-01 10:00:00.083903067 17296.75 2 2024-02-01 10:00:00.002605599 17264.50 2 2024-02-01 10:00:00.000035337 3

45 rows × 8 columns

Let’s compute the total ask and bid volumes and the corresponding imbalance.

prl = otp.ObSnapshotWide(db='CME_SAMPLE', tick_type='PRL_FULL', max_levels=x, running=True)
prl = prl.agg({'ask_vol': otp.agg.sum('ASK_SIZE'), 'bid_vol': otp.agg.sum('BID_SIZE')}, bucket_units='ticks', bucket_interval=x)
prl['imb'] = (prl['bid_vol'] - prl['ask_vol']) / (prl['bid_vol'] + prl['ask_vol'])
otp.run(prl, symbols=r'NQ\H24', start=s, end=s + otp.Milli(100))
Time ask_vol bid_vol imb
0 2024-02-01 10:00:00.000000000 7 6 -0.076923
1 2024-02-01 10:00:00.000005635 34 4 -0.789474
2 2024-02-01 10:00:00.000012009 5 4 -0.111111
3 2024-02-01 10:00:00.000035337 4 12 0.500000
4 2024-02-01 10:00:00.002605599 4 7 0.272727
... ... ... ... ...
10 2024-02-01 10:00:00.032787693 4 18 0.636364
11 2024-02-01 10:00:00.043854201 4 5 0.111111
12 2024-02-01 10:00:00.054669411 4 4 0.000000
13 2024-02-01 10:00:00.074736715 12 4 -0.500000
14 2024-02-01 10:00:00.083903067 14 4 -0.555556

15 rows × 4 columns

We can also compute that stats for the imbalance over time.

imb_stats = prl.agg({
    'tw_imb': otp.agg.tw_average('imb'),
    'mean':   otp.agg.average('imb'),
    'stdev':  otp.agg.stddev('imb'),
})
otp.run(imb_stats, symbols=r'NQ\H24', start=s, end=s + otp.Milli(100))
Time tw_imb mean stdev
0 2024-02-01 10:00:00.100 0.024032 0.025261 0.399156

Book sweep#

There are two versions of book sweep: by price and by quantity. Book sweep by price, takes a price as an input and returns the total quantity available at that price or better. Book sweep by quantity, takes a quantity as an input and returns the VWAP if the quantity were executed immediately.

prl = otp.ObSnapshot(db='CME_SAMPLE', tick_type='PRL_FULL', max_levels=10)
otp.run(prl, symbols=r'NQ\H24', start=s, end=s)
Time PRICE SIZE LEVEL UPDATE_TIME BUY_SELL_FLAG
0 2024-02-01 10:00:00 17303.50 3 1 2024-02-01 09:59:59.737771351 1
1 2024-02-01 10:00:00 17303.75 1 2 2024-02-01 09:59:59.968007113 1
2 2024-02-01 10:00:00 17304.00 3 3 2024-02-01 09:59:59.823591575 1
3 2024-02-01 10:00:00 17304.25 3 4 2024-02-01 09:59:59.668640001 1
4 2024-02-01 10:00:00 17304.50 4 5 2024-02-01 09:59:59.767992495 1
... ... ... ... ... ... ...
15 2024-02-01 10:00:00 17298.25 8 6 2024-02-01 09:59:59.681536321 0
16 2024-02-01 10:00:00 17298.00 3 7 2024-02-01 09:59:59.500380959 0
17 2024-02-01 10:00:00 17297.75 1 8 2024-02-01 09:59:59.251562445 0
18 2024-02-01 10:00:00 17297.50 2 9 2024-02-01 09:59:59.111633741 0
19 2024-02-01 10:00:00 17297.25 2 10 2024-02-01 09:59:59.883993699 0

20 rows × 6 columns

def side_to_direction(side):
    return 1 if side == 'ASK' else -1

def sweep_by_price(side, price):
    prl = otp.ObSnapshot(db='CME_SAMPLE', tick_type='PRL_FULL', side=side)
    direction = side_to_direction(side)
    prl = prl.where(direction * prl['PRICE'] <= direction * price)
    prl = prl.agg({'total_qty': otp.agg.sum('SIZE')})
    return otp.run(prl, symbols=r'NQ\H24', start=s, end=s)

print(sweep_by_price('BID', 11896))
print(sweep_by_price('ASK', 11898))
                 Time  total_qty
0 2024-02-01 10:00:00       2981
                 Time  total_qty
0 2024-02-01 10:00:00          0
def sweep_by_qty(side, qty):
    prl = otp.ObSnapshot(db='CME_SAMPLE', tick_type='PRL_FULL', side=side)
    prl = prl.agg({'total_qty': otp.agg.sum('SIZE')}, running=True, all_fields=True)
    prl = prl.where(prl['total_qty'] - prl['SIZE'] < qty)
    # update the SIZE in the last tick only so that total_qty is exactly qty
    prl['SIZE'] = prl.apply(lambda row: row['SIZE'] - (row['total_qty'] - qty) if row['total_qty'] > qty else row['SIZE'])
    prl = prl.agg({'VWAP': otp.agg.vwap('PRICE', 'SIZE')})
    return otp.run(prl, symbols=r'NQ\H24', start=s, end=s)
print(sweep_by_qty('BID', 10))
print(sweep_by_qty('ASK', 10))
                 Time     VWAP
0 2024-02-01 10:00:00  17299.4
                 Time     VWAP
0 2024-02-01 10:00:00  17303.9

Market By Order#

Order Book data may be annotated with ‘key’ field that lets us break down the book by each value of the ‘key’ field. For example, a book could by keyed by market participant ID, allowing us to see the book with the orders of a given market participant only. Some exchanges provide ‘market-by-order’ data where the book is keyed by order id. Set show_full_detail to True to see the book broken down to the most granular level. The example below is a market-by-order book.

prl = otp.ObSnapshot('CME_SAMPLE', tick_type='PRL_FULL', side='BID', show_full_detail=True)
orders = otp.run(prl, symbols=r'NQ\H24', start=s, end=s)
orders = orders[['ORDER_ID', 'PRICE', 'LEVEL', 'TIME_PRIORITY', 'SIZE', 'BUY_SELL_FLAG', 'ORDER_TYPE']]
orders.head()
ORDER_ID PRICE LEVEL TIME_PRIORITY SIZE BUY_SELL_FLAG ORDER_TYPE
0 6849720601921 17300.25 1 67070105795 1 0 L
1 6849720601880 17300.25 1 67070105719 1 0 L
2 6849720601879 17300.25 1 67070105718 2 0 L
3 6849720600537 17299.50 2 67070105850 1 0 L
4 6849719227337 17299.25 3 67070105870 1 0 L

Market-by-order data can be used to analyze/validate the priority mechanism used by the exchange.

prl = otp.ObSnapshot('CME_SAMPLE', tick_type='PRL_FULL', side='BID', show_full_detail=True)

"""
ORDER_TYPE:
L = Limit order
I = Implied order

Implied liquidity doesn't have priority as it's always last to execute at any price level.
It also doesn't have an order ID, so the IDs that we see in the db are synthetic
(consisting of 1 or 2 for the 1st/2nd implied level, and E/F for the buy/sell side respectively).

In order to rank the orders within a given price point by priority, we need to sort first by ORDER_TYPE (“L” comes before “I”),
then by TIME_PRIORITY (lowest value comes first).
"""
prl = prl.sort(['LEVEL', 'ORDER_TYPE', 'TIME_PRIORITY'], ascending=[True, False, True])
orders = otp.run(prl, symbols=r'NQ\H24', start=s, end=s)
orders = orders[['ORDER_ID', 'PRICE', 'LEVEL', 'TIME_PRIORITY', 'SIZE', 'BUY_SELL_FLAG', 'ORDER_TYPE']]
orders.head()
ORDER_ID PRICE LEVEL TIME_PRIORITY SIZE BUY_SELL_FLAG ORDER_TYPE
0 6849720601879 17300.25 1 67070105718 2 0 L
1 6849720601880 17300.25 1 67070105719 1 0 L
2 6849720601921 17300.25 1 67070105795 1 0 L
3 6849720600537 17299.50 2 67070105850 1 0 L
4 6849719227337 17299.25 3 67070105870 1 0 L