Skip to content

Configuration API

Set benchmark index, trading costs, slippage models, and other backtest parameters.


set_benchmark

Set the backtest benchmark.

set_benchmark('000300.XSHG')   # CSI 300

set_order_cost

Set trading costs.

set_order_cost(OrderCost(open_tax=0, close_tax=0.0005,
                         open_commission=0.00025, close_commission=0.00025,
                         min_commission=5))

OrderCost

Parameter Default Description
open_tax 0 Buy-side stamp duty
close_tax None Sell-side stamp duty; None applies the statutory rate (0.05% on/after 2023-08-28, 0.1% before); an explicit value takes precedence
open_commission 0.00025 Buy-side commission
close_commission 0.00025 Sell-side commission
close_today_commission 0 Intraday sell commission
min_commission 5 Minimum commission (¥)

set_option

Set strategy options.

set_option('use_real_price', True)

Available options

Option Default Description
check_price_limit True Price-limit fill enforcement (see "Price-limit rules" below). Flipped from False to True in 2026-10
avoid_future_data False Strict PIT mode: calling a snapshot API that is not point-in-time safe (e.g. get_fundamentals) inside a backtest raises ValueError instead of emitting a FutureWarning
st_status_map None Point-in-time ST status mapping: {code: bool \| (start, end) \| [(start, end), ...] \| callable(bare_code, date)}. No network detection in backtests; without it stocks are treated as non-ST (main board ±10%)
delist_gap_days 30 A held position with no bar for more than N trading days is treated as delisted and force-liquidated at the last close (<= 0 disables)
delisted_position_handling "liquidate" Delisting handling; "keep" keeps the position valued at its last price
delisted_liquidation_haircut 0.0 Haircut fraction (0–1) applied to delisting liquidation, e.g. 1.0 models a total loss

Price-limit rules

With check_price_limit=True (default), buy orders opening at the limit-up price and sell orders opening at the limit-down price are cancelled with Order.cancel_reason ("limit-up" / "limit-down"). Limits are board- and date-aware:

Board Limit
Main board (non-ST) ±10%
Main board ST / *ST ±5% (requires st_status_map for point-in-time status)
ChiNext (300xxx) ±10% before 2020-08-24, ±20% after (registration reform)
STAR Market (688xxx) ±20% (since inception)
BSE (codes 83/87/43/92) ±30%

Limits are computed off the security's last genuinely traded close (not updated during suspensions) and rounded to the fen. New-stock limit-free windows are not modelled. Orders on suspension days are cancelled the same day with a recorded reason ("no price data" for a missing bar, "suspended" for zero volume), never deferred. Positions in delisted stocks are force-liquidated after delist_gap_days without bars, recorded as DELIST_LIQUIDATE in the trade log.

Migration note (breaking): previous versions defaulted to no price-limit check (check_price_limit=False), so old momentum/limit-up-chasing backtest results were systematically optimistic. To reproduce the old behavior, set set_option('check_price_limit', False) explicitly.

set_slippage

Set the slippage model.

Class Description
SlippageModel Base class, no price adjustment
FixedSlippage(pct=0.001) Fixed percentage slippage
VolumeSlippage(impact=0.05, max_slippage_pct=0.05) Volume-proportional market impact

FixedSlippage: shifts every trade by a fixed fraction — buys at price * (1 + pct), sells at price * (1 - pct); pct defaults to 0.001 (0.1%).

VolumeSlippage: impact scales with the order's share of daily volume — impact_pct = impact * (order_shares / daily_volume), capped at max_slippage_pct; no slippage when daily_volume is 0 or unknown. impact defaults to 0.05 (an order at 10% of daily volume incurs ~0.5% slippage); max_slippage_pct defaults to 0.05 (5%).

BacktestSession and get_session

BacktestSession encapsulates the mutable state of a single backtest run. Regular strategies do not need to use it; for multi-threaded parallel backtesting, you can create an independent session per thread.