mirror of
https://github.com/manifoldbt/manifoldbt.git
synced 2026-11-02 07:56:11 +00:00
Jimmy7892
7e81ffe0d6
New directory benchmarks/vs_hft. The rules are written once (spec.py) and each engine implements them in its own idiom: a two-sided maker on 25 book levels and on best bid/ask, a flow taker on the trade clock, a maker that joins the best on an order-by-order feed, and an open-loop probe whose quotes depend on the market only, so that each one gets a verdict that does not depend on what the engine did before. The data is public and pinned. One Bybit spot BTCUSDT day (book and trades) and the Nasdaq TotalView-ITCH half session of 28 November 2025 (one symbol), each archive checked against a SHA-256 before anything runs. Parity comes before timing, and it is not read from fill counts: an engine that serves an order in pieces shows more fills for the same quantity. It is read as actions (which quotes were sent), verdicts (served or not, for how much, how late) and totals, with manifoldbt as the reference and the same exact / documented / failed statuses as the vectorbt benchmark. On the order-by-order day every engine is also judged against an exact replay of the raw messages (fifo_truth.py), so none of them has to be trusted as the reference. The README has a section, "Why the engines differ", that names each convention behind a difference in the results and gives its figure for each engine. Parameter sweeps are included: manifoldbt through run_sweep and run_sweep_lite, hftbacktest and NautilusTrader as the loop and the pool of processes a user would write. The result of a pool is checked against the loop, and a sample of sweep combinations against the same combination run alone. journal_of keeps the instant of a first fill as an integer. A nanosecond epoch in a float64 is only exact to 256 ns, and the exact-replay judgment compares those instants with the order of the raw messages, so the column never goes through a float. test_journal.py checks it to the nanosecond. The lock file pins the engines and their dependencies; every pinned version has a manylinux wheel for CPython 3.12, and manifoldbt resolves together with it.
147 lines
5.7 KiB
Python
147 lines
5.7 KiB
Python
"""The single source of truth for what every engine is asked to do.
|
|
|
|
Nothing in this file imports an engine. Every adapter (engine_mbt.py,
|
|
engine_hft.py, engine_nt.py) reads its numbers from here, so a change of a
|
|
latency or of a clip is made once and cannot drift between engines.
|
|
|
|
Venue: Bybit spot BTCUSDT (tick 0.1 USDT, lot 1e-6 BTC), public archives.
|
|
"""
|
|
from __future__ import annotations
|
|
|
|
import datetime as dt
|
|
|
|
SYMBOL = "BTCUSDT"
|
|
SYMBOL_ID = 1
|
|
TICK = 0.1
|
|
TICK_INV = 10 # 1 / TICK, exact: prices are handled as integer ticks
|
|
LOT = 1e-6
|
|
|
|
MS = 1_000_000 # nanoseconds in a millisecond
|
|
SEC = 1_000_000_000
|
|
|
|
# --------------------------------------------------------------------------- #
|
|
# Data
|
|
# --------------------------------------------------------------------------- #
|
|
# Days are pinned. A day that disappears upstream fails the run loudly (the
|
|
# fingerprint below is checked when it is set), it is never replaced silently.
|
|
DAYS = {
|
|
# day: (sha256 of the book archive, sha256 of the trades archive)
|
|
"2025-08-13": (
|
|
"63e1f91c9b3e28fd0189a0bf1773eadf0e02cc0d63c1608d4403013da848bc38",
|
|
"5a9277a3c73f93af1811917a29cf723fc34b5d4f1a17868dbd90d5fa26dd671c",
|
|
),
|
|
"2025-08-23": (
|
|
"989b9e3c69cdba8d7a1157e9a7659f93279f342cf55f1c0eff88ab62eebc20b0",
|
|
"aa5fc8f415e5259e7391d49c9a5225c41507ccdbb5d0a95761f6ebda4ca3f101",
|
|
),
|
|
}
|
|
# Nasdaq TotalView-ITCH 5.0 day files (4 to 18 GB). 2025-11-28 is the half session after Thanksgiving.
|
|
NASDAQ_DAYS = {
|
|
"2025-11-28": {"file": "S112825-v50.txt.gz", "bytes": 4_735_308_661,
|
|
"sha256": "e965ea6f9779c1b5d7b74de4549836ffe29a5f6ff64bb7894186a6f6404baaa4"},
|
|
}
|
|
BOOK_URL ="https://quote-saver.bycsi.com/orderbook/spot/{sym}/{day}_{sym}_ob200.data.zip"
|
|
TRADES_URL = "https://public.bybit.com/spot/{sym}/{sym}_{day}.csv.gz"
|
|
|
|
# Data tiers. "depth" is the number of price levels each side that an engine is
|
|
# allowed to see; every engine reads the same prepared stream.
|
|
TIERS = {
|
|
"bbo": 1, # best bid and best ask, plus the trade tape
|
|
"l2": 25, # 25 levels a side, plus the trade tape
|
|
}
|
|
|
|
# --------------------------------------------------------------------------- #
|
|
# Venue model, identical for all engines wherever the engine can express it
|
|
# --------------------------------------------------------------------------- #
|
|
LAT_ORDER_NS = 5 * MS # a post reaches the venue
|
|
LAT_CANCEL_NS = 5 * MS # a cancellation does
|
|
LAT_RESPONSE_NS = 5 * MS # an acknowledgement or a fill comes back
|
|
LAT_FEED_NS = 2 * MS # the book and the prints reach the strategy
|
|
MAKER_FEE_BPS = 0.0
|
|
TAKER_FEE_BPS = 4.0
|
|
|
|
# --------------------------------------------------------------------------- #
|
|
# Strategies
|
|
# --------------------------------------------------------------------------- #
|
|
WAKE_NS = 100 * MS # wake-up period of the quoting strategies
|
|
|
|
|
|
class MMLadder:
|
|
"""Example 31, parameterised: two post-only quotes, one level back per clip
|
|
of inventory held. The parameters are the sweep axes."""
|
|
name = "mm_ladder"
|
|
tier = "l2"
|
|
CLIP = 0.01 # BTC per quote
|
|
MAX_CLIPS = 5 # inventory bound, in clips
|
|
OFF_BID = 0 # extra levels back on the bid
|
|
OFF_ASK = 0 # extra levels back on the ask
|
|
|
|
|
|
class MMTouch:
|
|
"""Touch maker for a BBO feed: join the best price on both sides, withdraw a
|
|
side when the inventory bound is hit. The only data it needs is the best
|
|
bid and ask."""
|
|
name = "mm_touch"
|
|
tier = "bbo"
|
|
CLIP = 0.01
|
|
MAX_CLIPS = 5
|
|
|
|
|
|
class FlowTaker:
|
|
"""Example 33, on a fixed grid instead of the trade clock: aggressor
|
|
imbalance over a trailing window arms a side, a market (IOC) order follows
|
|
the armed side, and the position flips on the opposite crossing."""
|
|
name = "flow_taker"
|
|
tier = "bbo"
|
|
WINDOW_NS = 34 * SEC
|
|
THRESHOLD = 0.9
|
|
QTY = 0.01 # BTC per order
|
|
WAKE_NS = 1 * SEC # decision grid
|
|
|
|
|
|
# --------------------------------------------------------------------------- #
|
|
# Sweep grids (nested: each grid contains the smaller ones)
|
|
# --------------------------------------------------------------------------- #
|
|
GRID_AXES = {
|
|
4: dict(clip=[0.01, 0.02], max_clips=[3, 5], off_bid=[0], off_ask=[0]),
|
|
16: dict(clip=[0.005, 0.01, 0.02, 0.03], max_clips=[3, 5], off_bid=[0, 1], off_ask=[0]),
|
|
64: dict(clip=[0.005, 0.01, 0.02, 0.03], max_clips=[2, 3, 5, 8], off_bid=[0, 1], off_ask=[0, 1]),
|
|
256: dict(clip=[0.005, 0.01, 0.02, 0.03], max_clips=[2, 3, 5, 8], off_bid=[0, 1, 2, 3],
|
|
off_ask=[0, 1, 2, 3]),
|
|
}
|
|
|
|
|
|
def grid_combos(n):
|
|
"""The combinations, in the order an engine that sorts parameters by name
|
|
and varies the last fastest would enumerate them."""
|
|
import itertools
|
|
|
|
axes = GRID_AXES[n]
|
|
names = sorted(axes)
|
|
return names, [dict(zip(names, v)) for v in itertools.product(*[axes[k] for k in names])]
|
|
|
|
|
|
def day_bounds_ns(day: str):
|
|
d0 = dt.datetime.fromisoformat(day).replace(tzinfo=dt.timezone.utc)
|
|
t0 = int(d0.timestamp()) * SEC
|
|
return t0, t0 + 86_400 * SEC
|
|
|
|
|
|
class MBOJoin:
|
|
"""Order-by-order market (Nasdaq ITCH): join the best bid and ask with one
|
|
round lot, keep the order while it stays at the best, stand a side aside when
|
|
the displayed size at that best price is above CROWDED or the inventory
|
|
bound is reached."""
|
|
name = "mbo_join"
|
|
tier = "mbo"
|
|
LOT = 100.0 # shares per quote
|
|
MAX_LOTS = 5 # inventory bound, in lots
|
|
CROWDED = 5_000.0 # displayed shares at the best price above which the side stands aside
|
|
TICK = 0.01
|
|
WAKE_NS = 100 * MS
|
|
# regular session of the pinned day (New York): 09:30 to 13:00 (half day)
|
|
DAY = "2025-11-28"
|
|
OPEN_S = 9 * 3600 + 30 * 60
|
|
CLOSE_S = 13 * 3600
|
|
SYMBOL = "AAPL"
|