backtestingfx 0.2.0__tar.gz → 0.3.0__tar.gz
This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
- backtestingfx-0.3.0/.github/ISSUE_TEMPLATE/bug_report.md +39 -0
- backtestingfx-0.3.0/.github/ISSUE_TEMPLATE/feature_request.md +31 -0
- {backtestingfx-0.2.0 → backtestingfx-0.3.0}/.github/workflows/ci.yml +1 -1
- {backtestingfx-0.2.0 → backtestingfx-0.3.0}/.github/workflows/release.yml +11 -0
- {backtestingfx-0.2.0 → backtestingfx-0.3.0}/.gitignore +1 -0
- backtestingfx-0.3.0/CHANGELOG.md +182 -0
- backtestingfx-0.3.0/COMPATIBILITY.md +79 -0
- backtestingfx-0.3.0/CONTRIBUTING.md +97 -0
- {backtestingfx-0.2.0 → backtestingfx-0.3.0}/Cargo.lock +70 -36
- {backtestingfx-0.2.0 → backtestingfx-0.3.0}/Cargo.toml +2 -1
- backtestingfx-0.3.0/MARGINS.md +253 -0
- backtestingfx-0.3.0/MARKET_DATA.md +215 -0
- backtestingfx-0.3.0/PKG-INFO +541 -0
- backtestingfx-0.3.0/README.md +524 -0
- backtestingfx-0.3.0/REFERENCE_INPUT.md +146 -0
- backtestingfx-0.3.0/ROADMAP.md +63 -0
- backtestingfx-0.3.0/SWAPS.md +172 -0
- backtestingfx-0.3.0/backtestingfx/__init__.py +3 -0
- backtestingfx-0.3.0/backtestingfx/_input.py +157 -0
- backtestingfx-0.3.0/backtestingfx/_timestamps.py +43 -0
- backtestingfx-0.3.0/backtestingfx/backtest.py +519 -0
- {backtestingfx-0.2.0 → backtestingfx-0.3.0}/backtestingfx/plotting.py +134 -27
- backtestingfx-0.3.0/examples/benchmark.md +147 -0
- backtestingfx-0.3.0/examples/benchmark.py +81 -0
- backtestingfx-0.3.0/examples/benchmark_callbacks.md +176 -0
- backtestingfx-0.3.0/examples/benchmark_callbacks.py +264 -0
- backtestingfx-0.3.0/examples/benchmark_ingestion.md +158 -0
- backtestingfx-0.3.0/examples/benchmark_ingestion.py +175 -0
- backtestingfx-0.3.0/examples/benchmark_ingestion_results.json +723 -0
- backtestingfx-0.3.0/examples/benchmark_sl_tp.md +29 -0
- backtestingfx-0.3.0/examples/benchmark_sl_tp.rs +37 -0
- backtestingfx-0.3.0/examples/compare_bt.py +171 -0
- backtestingfx-0.3.0/examples/margin.py +27 -0
- {backtestingfx-0.2.0 → backtestingfx-0.3.0}/examples/simple_strategy.rs +6 -4
- backtestingfx-0.3.0/examples/swaps.py +30 -0
- {backtestingfx-0.2.0 → backtestingfx-0.3.0}/pyproject.toml +2 -1
- backtestingfx-0.3.0/src/broker.rs +1543 -0
- backtestingfx-0.3.0/src/data.rs +596 -0
- backtestingfx-0.3.0/src/engine.rs +711 -0
- backtestingfx-0.3.0/src/execution.rs +54 -0
- {backtestingfx-0.2.0 → backtestingfx-0.3.0}/src/lib.rs +12 -0
- backtestingfx-0.3.0/src/margin.rs +243 -0
- backtestingfx-0.3.0/src/optimise.rs +301 -0
- {backtestingfx-0.2.0 → backtestingfx-0.3.0}/src/stats.rs +57 -4
- backtestingfx-0.3.0/src/strategy.rs +17 -0
- backtestingfx-0.3.0/src/swap.rs +544 -0
- {backtestingfx-0.2.0 → backtestingfx-0.3.0}/src/types.rs +39 -2
- backtestingfx-0.3.0/src/validation.rs +59 -0
- backtestingfx-0.3.0/tests/data/timestamps.csv +7 -0
- {backtestingfx-0.2.0 → backtestingfx-0.3.0}/tests/test_backtest.py +57 -0
- backtestingfx-0.3.0/tests/test_bulk_ingestion.py +296 -0
- backtestingfx-0.3.0/tests/test_callback_contract.py +161 -0
- backtestingfx-0.3.0/tests/test_column_conversion.py +63 -0
- backtestingfx-0.3.0/tests/test_execution_timing.py +155 -0
- backtestingfx-0.3.0/tests/test_intrabar_execution.py +109 -0
- backtestingfx-0.3.0/tests/test_margin.py +234 -0
- backtestingfx-0.3.0/tests/test_market_data_validation.py +131 -0
- backtestingfx-0.3.0/tests/test_native_engine_reuse.py +34 -0
- backtestingfx-0.3.0/tests/test_numeric_validation.py +150 -0
- backtestingfx-0.3.0/tests/test_reference_comparison.py +69 -0
- backtestingfx-0.3.0/tests/test_reference_input.py +338 -0
- backtestingfx-0.3.0/tests/test_strategy_data.py +125 -0
- backtestingfx-0.3.0/tests/test_swaps.py +294 -0
- backtestingfx-0.3.0/tests/test_timestamps.py +123 -0
- backtestingfx-0.2.0/CHANGELOG.md +0 -64
- backtestingfx-0.2.0/PKG-INFO +0 -265
- backtestingfx-0.2.0/README.md +0 -250
- backtestingfx-0.2.0/backtestingfx/__init__.py +0 -2
- backtestingfx-0.2.0/backtestingfx/backtest.py +0 -240
- backtestingfx-0.2.0/examples/benchmark.py +0 -146
- backtestingfx-0.2.0/examples/compare_bt.py +0 -90
- backtestingfx-0.2.0/src/broker.rs +0 -454
- backtestingfx-0.2.0/src/data.rs +0 -28
- backtestingfx-0.2.0/src/engine.rs +0 -163
- backtestingfx-0.2.0/src/optimise.rs +0 -166
- backtestingfx-0.2.0/src/strategy.rs +0 -7
- {backtestingfx-0.2.0 → backtestingfx-0.3.0}/LICENSE +0 -0
- {backtestingfx-0.2.0 → backtestingfx-0.3.0}/data/EURUSD_1H.csv +0 -0
- {backtestingfx-0.2.0 → backtestingfx-0.3.0}/examples/html_report.py +0 -0
- {backtestingfx-0.2.0 → backtestingfx-0.3.0}/examples/optimize.py +0 -0
- {backtestingfx-0.2.0 → backtestingfx-0.3.0}/examples/sma_cross.py +0 -0
|
@@ -0,0 +1,39 @@
|
|
|
1
|
+
---
|
|
2
|
+
name: Bug report
|
|
3
|
+
about: Report a reproducible FX simulation, installation, or reporting bug
|
|
4
|
+
title: ""
|
|
5
|
+
labels: ""
|
|
6
|
+
assignees: ""
|
|
7
|
+
---
|
|
8
|
+
|
|
9
|
+
## Problem
|
|
10
|
+
Describe the bug and the expected versus actual behavior. Include exact errors
|
|
11
|
+
or a traceback and hand-calculated expected results when relevant.
|
|
12
|
+
|
|
13
|
+
## Environment
|
|
14
|
+
- backtestingfx version (or commit):
|
|
15
|
+
- Python version and interpreter:
|
|
16
|
+
- OS and architecture:
|
|
17
|
+
- pandas / numpy versions:
|
|
18
|
+
- Plotly version (if reporting):
|
|
19
|
+
- Installation method (wheel, source, development); Rust version if source-built:
|
|
20
|
+
|
|
21
|
+
## Data and simulation settings
|
|
22
|
+
- Instrument, account currency, and quote-to-account conversion:
|
|
23
|
+
- Row count, column names/dtypes, and index type:
|
|
24
|
+
- Timeframe, timestamp format/unit, and timezone:
|
|
25
|
+
- OHLC price convention (midpoint, bid, or ask):
|
|
26
|
+
- Cash, lot sizes, contract size, spread, commission, and SL/TP settings:
|
|
27
|
+
|
|
28
|
+
## Minimal reproduction
|
|
29
|
+
Provide a runnable code block that constructs a small synthetic DataFrame,
|
|
30
|
+
defines the strategy or signal function, and runs the failing call. Include
|
|
31
|
+
sample rows and output; avoid private data and external file dependencies.
|
|
32
|
+
|
|
33
|
+
```python
|
|
34
|
+
# Imports, synthetic input, strategy/signals, settings, and failing call
|
|
35
|
+
```
|
|
36
|
+
|
|
37
|
+
## Additional context
|
|
38
|
+
Does this happen on the latest release or main? If it is a regression, include
|
|
39
|
+
the last working version and relevant logs.
|
|
@@ -0,0 +1,31 @@
|
|
|
1
|
+
---
|
|
2
|
+
name: Feature request
|
|
3
|
+
about: Propose an FX research capability or API improvement
|
|
4
|
+
title: ""
|
|
5
|
+
labels: ""
|
|
6
|
+
assignees: ""
|
|
7
|
+
---
|
|
8
|
+
|
|
9
|
+
## FX use case and problem
|
|
10
|
+
Describe the research workflow and what currently prevents it.
|
|
11
|
+
|
|
12
|
+
## Proposed behavior
|
|
13
|
+
Give an API/code example and explain expected execution or result semantics.
|
|
14
|
+
|
|
15
|
+
## Environment and representative data
|
|
16
|
+
- backtestingfx and Python versions; OS/architecture:
|
|
17
|
+
- Instrument, timeframe, row count, column/index types, and timestamp timezone/unit:
|
|
18
|
+
- Relevant sizing, costs, conversion, or order assumptions:
|
|
19
|
+
|
|
20
|
+
## Current workaround
|
|
21
|
+
Include a small example of the current workflow if possible, and alternatives
|
|
22
|
+
you considered.
|
|
23
|
+
|
|
24
|
+
## Acceptance criteria
|
|
25
|
+
List observable outcomes and edge cases that would make the feature complete.
|
|
26
|
+
Explain how the behavior could be tested with synthetic data.
|
|
27
|
+
|
|
28
|
+
## Compatibility and scope
|
|
29
|
+
Could existing strategies or results change? Note any related
|
|
30
|
+
[roadmap](https://github.com/KhizarImran/backtestingfx/blob/main/ROADMAP.md)
|
|
31
|
+
items or issues.
|
|
@@ -19,6 +19,6 @@ jobs:
|
|
|
19
19
|
# clippy does everything cargo check does, so it replaces it rather than adding to it
|
|
20
20
|
- run: cargo clippy --all-targets -- -D warnings
|
|
21
21
|
- run: cargo test
|
|
22
|
-
- run: python -m pip install ".[report]"
|
|
22
|
+
- run: python -m pip install ".[report,test]"
|
|
23
23
|
- run: python -m unittest discover -s "$GITHUB_WORKSPACE/tests"
|
|
24
24
|
working-directory: /tmp
|
|
@@ -78,6 +78,17 @@ jobs:
|
|
|
78
78
|
target: x64
|
|
79
79
|
args: --release --out dist --find-interpreter
|
|
80
80
|
sccache: ${{ !startsWith(github.ref, 'refs/tags/') }}
|
|
81
|
+
# Preserve the 3.10 wheel when the runner image no longer includes it.
|
|
82
|
+
- uses: actions/setup-python@v6
|
|
83
|
+
with:
|
|
84
|
+
python-version: "3.10"
|
|
85
|
+
architecture: x64
|
|
86
|
+
- name: Build Python 3.10 wheel
|
|
87
|
+
uses: PyO3/maturin-action@v1
|
|
88
|
+
with:
|
|
89
|
+
target: x64
|
|
90
|
+
args: --release --out dist --interpreter python
|
|
91
|
+
sccache: ${{ !startsWith(github.ref, 'refs/tags/') }}
|
|
81
92
|
- uses: actions/upload-artifact@v6
|
|
82
93
|
with:
|
|
83
94
|
name: wheels-windows-x64
|
|
@@ -0,0 +1,182 @@
|
|
|
1
|
+
# Changelog
|
|
2
|
+
|
|
3
|
+
## [Unreleased]
|
|
4
|
+
|
|
5
|
+
## [0.3.0] - 2026-11-19
|
|
6
|
+
|
|
7
|
+
### Added
|
|
8
|
+
- Reference strategy data exposes case-sensitive OHLCV NumPy views, pandas
|
|
9
|
+
accessors, and pip size. Each callback sees only its visible prefix.
|
|
10
|
+
- Opt-in queued market execution through `trade_on_close` and `finalize_trades`.
|
|
11
|
+
Python strategies, native strategies, and signal runs share next-open or
|
|
12
|
+
previous-close fills and final-bar processing. Results expose open positions
|
|
13
|
+
and marked final equity. The queued timeline follows backtesting.py 0.6.6
|
|
14
|
+
with FX sizing and costs. See [queued market execution](README.md#queued-market-execution).
|
|
15
|
+
- Optional `data_mode="reference"` for backtesting.py 0.6.6 DataFrame input:
|
|
16
|
+
periods, duplicate labels, exact timestamps, aligned features, indexed equity
|
|
17
|
+
and event views, and reports that retain source labels. The default FX input
|
|
18
|
+
policy and order API remain unchanged. Reference mode rejects calendar swaps.
|
|
19
|
+
See [REFERENCE_INPUT.md](REFERENCE_INPUT.md) for scope and remaining API work.
|
|
20
|
+
- Required CI input-conformance fixtures against `backtesting==0.6.6` through
|
|
21
|
+
the `test` extra.
|
|
22
|
+
- Optional immutable `MarginConfig`, current-price gross used/free margin,
|
|
23
|
+
cost-aware entry checks, returned `OrderResult` values, durable rejection
|
|
24
|
+
records, and full-book liquidation with terminal account status. Python,
|
|
25
|
+
native, and signal execution share the policy; margin defaults to disabled.
|
|
26
|
+
- Margin configuration and stopout evidence in HTML reports, plus a
|
|
27
|
+
[margin guide](MARGINS.md) and hand-calculated examples.
|
|
28
|
+
- Explicit Python `timestamp_unit` and native CSV `TimestampUnit` options for
|
|
29
|
+
numeric epoch input. Naive datetimes mean UTC; offsets normalize to UTC.
|
|
30
|
+
- Optional `SwapConfig` for signed long/short overnight financing in account
|
|
31
|
+
currency per lot. Configure local rollover time, IANA timezone, weekday weights,
|
|
32
|
+
and holiday date overrides. Swaps are disabled by default.
|
|
33
|
+
- Separate position/trade financing, total swap cash flow, dated posting records,
|
|
34
|
+
and HTML report financing details. Native, Python, and signal runs share the
|
|
35
|
+
rollover policy and reconcile partial-close financing without charging twice.
|
|
36
|
+
- Constant-time Python `Strategy` and `Broker` properties `has_positions` and
|
|
37
|
+
`position_count`, avoiding position-list copies for presence and count checks.
|
|
38
|
+
- A callback benchmark with separate conversion, engine construction,
|
|
39
|
+
initialization, and simulation timings, native and Python no-op baselines,
|
|
40
|
+
indicator and position-access workloads, correctness checks, and peak RSS.
|
|
41
|
+
- Documented and regression-tested the stop-first intrabar SL/TP policy,
|
|
42
|
+
inclusive touches and next-bar checks for new entries.
|
|
43
|
+
- Contributor setup and validation guidance, bug/feature issue templates, an
|
|
44
|
+
API/platform compatibility policy, and a prioritized FX roadmap.
|
|
45
|
+
|
|
46
|
+
### Fixed
|
|
47
|
+
- Nonpositive-equity histories return zero Sharpe instead of dividing by zero
|
|
48
|
+
or computing percentage returns on debt.
|
|
49
|
+
- Timestamp conversion floors to exact whole seconds, including before 1970,
|
|
50
|
+
without floating-point rounding or row promotion of integer timestamps.
|
|
51
|
+
Reports use the same saved seconds as the engine and financing ledger.
|
|
52
|
+
- Default FX market-data boundaries reject nonfinite/nonpositive OHLC values, inconsistent
|
|
53
|
+
candle ranges, negative/nonfinite volume, and duplicate or backwards timestamps
|
|
54
|
+
after conversion to whole seconds. Python errors identify the bar and field;
|
|
55
|
+
empty inputs remain valid. Optimization checks data before calling signals.
|
|
56
|
+
- Native CSV loading resolves columns by case-insensitive name, defaults missing
|
|
57
|
+
volume to zero, and returns typed errors instead of panicking. Checked native
|
|
58
|
+
`Bar::try_new`, `Engine::try_new`, and `data::validate_bars` are also available.
|
|
59
|
+
- Pandas input uses bulk typed buffers to construct native bars, removing
|
|
60
|
+
per-row Series and temporary Python Bar allocation from numeric DatetimeIndex
|
|
61
|
+
ingestion. Initialization history avoids an intermediate native copy. Preserve
|
|
62
|
+
independent runs, report snapshots, and Bar identity. Timestamp policy is
|
|
63
|
+
explicit as described above.
|
|
64
|
+
See the ingestion benchmark for measured setup and total-runtime improvements.
|
|
65
|
+
- Python strategy dispatch reuses the Python method-name string for `next`,
|
|
66
|
+
while preserving per-bar method lookup, distinct Bar objects, and exceptions.
|
|
67
|
+
- The existing cross-library comparison and benchmark now use matched fixed-unit
|
|
68
|
+
next-open strategies, verify trades and full equity curves, and report setup,
|
|
69
|
+
simulation, and total timings separately. Spread is zero; commission is matched.
|
|
70
|
+
- Python constructors reject nonfinite or out-of-range broker settings. Order,
|
|
71
|
+
partial-close, stop-update, closing-price, and equity-price inputs are checked
|
|
72
|
+
before account mutation. Finite nonpositive partial sizes remain no-ops.
|
|
73
|
+
- Optimization rejects malformed or nonfinite signals, including direct
|
|
74
|
+
`run_grid()` calls. Native SMA calls reject invalid sizes and windows before
|
|
75
|
+
simulation rather than risking a Rust panic.
|
|
76
|
+
- SL/TP processing checks the midpoint open first: crossed stops fill at the
|
|
77
|
+
adverse open, while crossed targets receive the favorable open, before any
|
|
78
|
+
later range touches. Exit spread, commission, and conversion still apply.
|
|
79
|
+
Bars that open inside both levels retain stop-first intrabar behavior.
|
|
80
|
+
- Batch stop-loss/take-profit processing uses a single linear pass, preserving
|
|
81
|
+
survivor and trade order, fills, and accounting when many positions exit together.
|
|
82
|
+
- Native engine runs reset cash, positions, trade history, and position IDs before
|
|
83
|
+
each independent simulation, including repeated `run_native_sma()` calls.
|
|
84
|
+
- `spread` now means the full bid–ask width around midpoint OHLC prices. Each
|
|
85
|
+
fill applies half the width, so a one-pip EURUSD spread costs $10 per standard
|
|
86
|
+
lot on a flat round trip, before commission, instead of $20.
|
|
87
|
+
- Open equity uses the bid/ask price available to close each position and includes
|
|
88
|
+
the exit-side spread (future exit commission is still charged only at closing).
|
|
89
|
+
|
|
90
|
+
### Migration
|
|
91
|
+
- Strategy and Broker `buy`/`sell` return `OrderResult` instead of `None`.
|
|
92
|
+
Native `Engine::run` and `run_with_swap` return
|
|
93
|
+
`Result<Stats, SimulationError>`, and native broker close methods and
|
|
94
|
+
`check_sl_tp` return `Result<(), MarginError>`. Handle results with `?` or a
|
|
95
|
+
match. Margin checks
|
|
96
|
+
require an explicit `MarginConfig`. A stopout halts further callbacks, retains
|
|
97
|
+
real negative cash, and preserves the full equity timeline. See [MARGINS.md](MARGINS.md).
|
|
98
|
+
- Numeric timestamp columns now require `timestamp_unit="s"`, `"ms"`, `"us"`, or
|
|
99
|
+
`"ns"`. Use `"ns"` for the previous implicit pandas units. Fractional dates
|
|
100
|
+
floor instead of rounding and truncating; rerun affected backtests and resolve
|
|
101
|
+
collisions at whole-second precision. Native numeric CSV uses signed integers
|
|
102
|
+
with `load_csv_with_timestamp_unit`. See [timestamp rules](MARKET_DATA.md#timestamp-sources-and-ordering).
|
|
103
|
+
- Malformed candles and duplicate/reversed whole-second timestamps now raise
|
|
104
|
+
errors, including repeated timestamps previously accepted with swaps enabled.
|
|
105
|
+
Resolve subsecond collisions explicitly before simulation. Native `load_csv`
|
|
106
|
+
now returns `Result<Vec<Bar>, DataError>`; handle it with `?` or an explicit
|
|
107
|
+
match. See [MARKET_DATA.md](MARKET_DATA.md) for the complete input policy.
|
|
108
|
+
- Python cash, contract size, conversion rate, opening sizes, prices, and supplied
|
|
109
|
+
SL/TP levels must be positive and finite; spread and commission must be finite
|
|
110
|
+
and nonnegative. Invalid numeric inputs now raise `ValueError`. Signal arrays
|
|
111
|
+
must be one-dimensional, finite, and contain one target per bar. See README
|
|
112
|
+
numeric input validation for partial-close and native SMA rules.
|
|
113
|
+
- Gap-bar SL/TP fills now use the opening price instead of the configured level.
|
|
114
|
+
An opening target also takes precedence over a later stop touch. Rerun affected
|
|
115
|
+
backtests: PnL, equity, and strategy decisions can change; there is no legacy
|
|
116
|
+
gap-fill option. See README execution policy for long/short examples.
|
|
117
|
+
- Double the previous `spread` input to preserve pre-change fills and completed
|
|
118
|
+
trade PnL. Equity curves can still differ because marking now includes exit
|
|
119
|
+
spread. SL/TP levels remain midpoint levels. See README price conventions.
|
|
120
|
+
|
|
121
|
+
## [0.2.0] - 2026-08-30
|
|
122
|
+
|
|
123
|
+
First release to reach PyPI since 0.1.1 — the `v0.1.2` tag predated the release
|
|
124
|
+
workflow, so 0.1.2 was tagged but never published. Everything listed under 0.1.2
|
|
125
|
+
below ships here too.
|
|
126
|
+
|
|
127
|
+
### Added
|
|
128
|
+
- `Strategy.update_sl(id, stop_loss)` — move the stop on an open position in place,
|
|
129
|
+
so a trailing stop doesn't pay spread and commission to close and reopen
|
|
130
|
+
- `Strategy.close_partial(id, lot_size)` — scale out of a position, leaving the
|
|
131
|
+
remainder open under the same id
|
|
132
|
+
|
|
133
|
+
### Changed
|
|
134
|
+
- `Broker` closes positions through one internal `settle` helper instead of four
|
|
135
|
+
copies of the same PnL and commission arithmetic
|
|
136
|
+
|
|
137
|
+
## [0.1.2] - 2026-08-13
|
|
138
|
+
|
|
139
|
+
### Added
|
|
140
|
+
- `Backtest.optimize()` — parallel grid search over a vectorised signal function.
|
|
141
|
+
Simulations run on native threads with the GIL released (`src/optimise.rs`);
|
|
142
|
+
157x faster than looping `run()` over the same grid.
|
|
143
|
+
- `strategy_class` is now optional, so `Backtest(df, cash=...)` works for optimization
|
|
144
|
+
- Standalone HTML reports (`Backtest.plot()`) and `examples/html_report.py`
|
|
145
|
+
- Three-way speed benchmark against backtesting.py (`examples/benchmark.py`)
|
|
146
|
+
- Sample EURUSD hourly data is now in the repo, so the examples run on a fresh clone
|
|
147
|
+
|
|
148
|
+
### Fixed
|
|
149
|
+
- Data access inside `next()` is O(1) per bar instead of O(n) — ~2.6x faster
|
|
150
|
+
|
|
151
|
+
### Changed
|
|
152
|
+
- `numpy` is now a direct dependency (it was already pulled in by pandas)
|
|
153
|
+
- CI runs `cargo clippy -- -D warnings` in place of `cargo check`
|
|
154
|
+
|
|
155
|
+
## [0.1.1] - 2026-07-05
|
|
156
|
+
|
|
157
|
+
### Added
|
|
158
|
+
- `self.data`, `self.index`, `self.cash`, `self.equity` properties on Strategy
|
|
159
|
+
- Sharpe ratio in Stats output (unannualized)
|
|
160
|
+
- `__repr__` on Position — readable output when printing positions
|
|
161
|
+
- DataFrame column validation with clear error message
|
|
162
|
+
|
|
163
|
+
### Fixed
|
|
164
|
+
- Trade PnL in history now stores net PnL (after exit commission) — per-trade stats were slightly optimistic
|
|
165
|
+
- `Position` pyclass uses `from_py_object` to fix PyO3 deprecation warning
|
|
166
|
+
- Removed dead `AttributeError` swallow in engine.rs
|
|
167
|
+
|
|
168
|
+
### Examples
|
|
169
|
+
- Added `examples/sma_cross.py` — SMA 10/50 crossover on EURUSD hourly data
|
|
170
|
+
- Added `examples/compare_bt.py` — side-by-side comparison against backtesting.py
|
|
171
|
+
|
|
172
|
+
## [0.1.0] - 2026-06-21
|
|
173
|
+
|
|
174
|
+
### Added
|
|
175
|
+
- Event-driven backtesting engine on OHLCV bar data
|
|
176
|
+
- Simulated broker with buy, sell, close_all, close_position
|
|
177
|
+
- Per-position stop loss and take profit
|
|
178
|
+
- Realistic FX lot sizing (0.01 / 0.10 / 1.00) with contract_size and quote_to_account conversion
|
|
179
|
+
- Full trade history with PnL per trade
|
|
180
|
+
- Stats: return, win rate, avg PnL, best/worst trade, profit factor, max drawdown
|
|
181
|
+
- Python API — inherit Strategy, run Backtest
|
|
182
|
+
- PyO3 Rust extension with Python wrapper
|
|
@@ -0,0 +1,79 @@
|
|
|
1
|
+
# Compatibility policy
|
|
2
|
+
|
|
3
|
+
## Python and platforms
|
|
4
|
+
|
|
5
|
+
Package metadata declares Python >=3.9. CPython is the intended interpreter;
|
|
6
|
+
PyPy support is not currently verified. pandas and numpy are required.
|
|
7
|
+
HTML reporting additionally requires Plotly >=5 through the `report` extra.
|
|
8
|
+
|
|
9
|
+
| Area | Current coverage |
|
|
10
|
+
| --- | --- |
|
|
11
|
+
| Functional CI | Ubuntu, CPython 3.12; Rust checks and installed-package Python tests |
|
|
12
|
+
| Linux release wheel targets | x86_64 and aarch64, with manylinux selected by maturin |
|
|
13
|
+
| macOS release wheel targets | Intel x86_64 and Apple Silicon aarch64 |
|
|
14
|
+
| Windows release wheel target | x64 |
|
|
15
|
+
| Source distributions | Built by the release workflow; require Rust and a native compiler to install |
|
|
16
|
+
|
|
17
|
+
These wheel targets describe the configured release workflow, not a guarantee
|
|
18
|
+
that every Python version has a published wheel or has passed functional tests.
|
|
19
|
+
The extension does not currently use abi3, so wheel compatibility depends on
|
|
20
|
+
the interpreter and platform tags. The wheel jobs discover available
|
|
21
|
+
interpreters. Windows also builds Python 3.10 explicitly. Check the artifacts
|
|
22
|
+
for your interpreter before relying on wheel availability.
|
|
23
|
+
|
|
24
|
+
If pip falls back to a source build, install a current stable Rust toolchain
|
|
25
|
+
and your platform's compiler/linker, then retry installation. Contributor
|
|
26
|
+
setup is in [CONTRIBUTING.md](CONTRIBUTING.md). Broader wheel verification and
|
|
27
|
+
functional coverage are tracked in issues
|
|
28
|
+
[#35](https://github.com/KhizarImran/backtestingfx/issues/35) and
|
|
29
|
+
[#36](https://github.com/KhizarImran/backtestingfx/issues/36).
|
|
30
|
+
|
|
31
|
+
## Public API and stability
|
|
32
|
+
|
|
33
|
+
Input compatibility targets backtesting.py **0.6.6**, commit
|
|
34
|
+
`9c2b8e1c99f3a860883222ba8d6d5b5cac3a932d`, through the explicit
|
|
35
|
+
`data_mode="reference"` option. See [reference input](REFERENCE_INPUT.md) for
|
|
36
|
+
index, validation, strategy, and result behavior. CI requires the pinned
|
|
37
|
+
reference package through the `test` extra. Full strategy, order, sizing, cost,
|
|
38
|
+
and results compatibility remains tracked in
|
|
39
|
+
[#61](https://github.com/KhizarImran/backtestingfx/issues/61).
|
|
40
|
+
|
|
41
|
+
The project is pre-1.0. Public Python imports from `backtestingfx`, documented
|
|
42
|
+
strategy methods/properties, result fields, and documented execution assumptions
|
|
43
|
+
are the compatibility surface. Underscore-prefixed members and internal Rust
|
|
44
|
+
implementation details may change; existing examples using private strategy
|
|
45
|
+
members do not make those members stable.
|
|
46
|
+
|
|
47
|
+
Before 1.0, intentional incompatible API changes belong in a minor release
|
|
48
|
+
with a changelog entry and migration instructions. Patch releases should
|
|
49
|
+
preserve API signatures, but correctness fixes can change simulated fills,
|
|
50
|
+
PnL, equity, and statistics. Such changes must identify the previous and new
|
|
51
|
+
behavior and any migration required. After 1.0, intentional incompatible public
|
|
52
|
+
API changes require a major release.
|
|
53
|
+
|
|
54
|
+
Where practical, deprecated Python APIs emit `DeprecationWarning` and remain
|
|
55
|
+
available for at least one minor release before removal. Document the
|
|
56
|
+
replacement and planned removal version. A correctness fix that cannot retain
|
|
57
|
+
the old behavior safely may be made without a deprecation period, with an
|
|
58
|
+
explicit explanation and migration notes.
|
|
59
|
+
|
|
60
|
+
Changes to the minimum Python version or supported platform targets must be
|
|
61
|
+
announced in the changelog and this policy. Before 1.0 they belong in a minor
|
|
62
|
+
release; after 1.0 they require a major release.
|
|
63
|
+
|
|
64
|
+
## Reproducibility
|
|
65
|
+
|
|
66
|
+
Pin the library and dependency versions for reproducible research. Record data
|
|
67
|
+
source, instrument, timeframe, timezone, sizing, spread, commission, and
|
|
68
|
+
conversion assumptions. Read [CHANGELOG.md](CHANGELOG.md) before upgrading:
|
|
69
|
+
unchanged strategy code can produce different results after a modeling fix.
|
|
70
|
+
The full bid–ask spread correction in Unreleased is one such example.
|
|
71
|
+
|
|
72
|
+
With optional swaps enabled, also record signed long/short rates, the rollover
|
|
73
|
+
timezone and time, weekday weights, and date overrides. Timezone rules come
|
|
74
|
+
from the `chrono-tz` database embedded in the build. Pin the package and native
|
|
75
|
+
dependency versions when reproducing historical rollover results. See
|
|
76
|
+
[SWAPS.md](SWAPS.md) for calendar behavior.
|
|
77
|
+
|
|
78
|
+
This policy describes project expectations; the current tests and published
|
|
79
|
+
artifacts establish the coverage actually verified for each release.
|
|
@@ -0,0 +1,97 @@
|
|
|
1
|
+
# Contributing
|
|
2
|
+
|
|
3
|
+
backtestingfx focuses on FX research with OHLCV bars, lot sizing, spread,
|
|
4
|
+
commission, optional overnight financing, and quote-to-account conversion. See the [roadmap](ROADMAP.md)
|
|
5
|
+
for priorities and the [compatibility policy](COMPATIBILITY.md) before changing
|
|
6
|
+
public behavior. Search existing issues before opening a bug or feature request.
|
|
7
|
+
|
|
8
|
+
## Local setup
|
|
9
|
+
|
|
10
|
+
Install Git, CPython 3.9 or newer, and a current stable Rust toolchain with
|
|
11
|
+
Cargo and Clippy. The crate uses Rust edition 2024 (Rust 1.85 or newer);
|
|
12
|
+
dependency requirements may require a newer compiler. Python 3.12 on Linux
|
|
13
|
+
matches the current CI configuration. Building the extension also requires
|
|
14
|
+
your platform's native compiler/linker (for example, Xcode Command Line Tools
|
|
15
|
+
on macOS or MSVC Build Tools on Windows).
|
|
16
|
+
|
|
17
|
+
From your clone's root:
|
|
18
|
+
|
|
19
|
+
```bash
|
|
20
|
+
git clone https://github.com/KhizarImran/backtestingfx.git
|
|
21
|
+
cd backtestingfx
|
|
22
|
+
python -m venv .venv
|
|
23
|
+
source .venv/bin/activate
|
|
24
|
+
python -m pip install --upgrade pip
|
|
25
|
+
python -m pip install "maturin>=1.0,<2.0" pandas numpy "plotly>=5" "backtesting==0.6.6"
|
|
26
|
+
maturin develop
|
|
27
|
+
```
|
|
28
|
+
|
|
29
|
+
On Windows, activate with `.venv\Scripts\Activate.ps1` in PowerShell instead.
|
|
30
|
+
Use your virtual environment's Python for all commands. Run `maturin develop`
|
|
31
|
+
again after Rust changes. Python source changes are visible through the
|
|
32
|
+
development installation.
|
|
33
|
+
|
|
34
|
+
## Checks
|
|
35
|
+
|
|
36
|
+
Run the Rust checks from the repository root:
|
|
37
|
+
|
|
38
|
+
```bash
|
|
39
|
+
cargo clippy --all-targets -- -D warnings
|
|
40
|
+
cargo test
|
|
41
|
+
```
|
|
42
|
+
|
|
43
|
+
Check the installed Python package outside the source directory, as CI does.
|
|
44
|
+
This avoids accidentally importing the source package without its built extension.
|
|
45
|
+
|
|
46
|
+
```bash
|
|
47
|
+
python -m pip install ".[report,test]"
|
|
48
|
+
repo_dir="$PWD"
|
|
49
|
+
cd /tmp
|
|
50
|
+
python -m unittest discover -s "$repo_dir/tests"
|
|
51
|
+
cd "$repo_dir"
|
|
52
|
+
```
|
|
53
|
+
|
|
54
|
+
On Windows, save the repository path, change to a temporary directory, and pass
|
|
55
|
+
the absolute path to `tests` to `python -m unittest discover -s`. Reinstall
|
|
56
|
+
with `python -m pip install --force-reinstall ".[report,test]"` after changes when
|
|
57
|
+
using this regular installation rather than `maturin develop`.
|
|
58
|
+
|
|
59
|
+
## Examples
|
|
60
|
+
|
|
61
|
+
Run these from the repository root after building/installing the package:
|
|
62
|
+
|
|
63
|
+
```bash
|
|
64
|
+
python examples/sma_cross.py
|
|
65
|
+
python examples/optimize.py
|
|
66
|
+
python examples/swaps.py
|
|
67
|
+
python examples/margin.py
|
|
68
|
+
python examples/html_report.py
|
|
69
|
+
cargo run --example simple_strategy
|
|
70
|
+
```
|
|
71
|
+
|
|
72
|
+
SMA and optimization examples use the included `data/EURUSD_1H.csv`.
|
|
73
|
+
The HTML example generates synthetic data, writes
|
|
74
|
+
`backtestingfx-report.html`, and attempts to open a browser. It requires the
|
|
75
|
+
`report` extra. Comparison and benchmark scripts additionally require
|
|
76
|
+
`python -m pip install backtesting==0.6.6`; report their versions and execution
|
|
77
|
+
assumptions when sharing results. See [the matched comparison](examples/benchmark.md)
|
|
78
|
+
for correctness checks and separate setup/run timings. The `test` extra pins the
|
|
79
|
+
reference package. Input conformance tests require it, and CI installs it.
|
|
80
|
+
|
|
81
|
+
The swap example uses a small flat-price fixture and checks financing and
|
|
82
|
+
partial-close reconciliation. See [SWAPS.md](SWAPS.md) for rate units, rollover
|
|
83
|
+
calendars, and native execution. For profiling and total-runtime comparisons,
|
|
84
|
+
see [the ingestion benchmark](examples/benchmark_ingestion.md).
|
|
85
|
+
|
|
86
|
+
## Pull requests
|
|
87
|
+
|
|
88
|
+
Create a focused branch and describe the problem, resulting behavior, and
|
|
89
|
+
validation performed. Link the relevant issue using `Closes #number` when
|
|
90
|
+
the PR completes it. For behavior changes, add regression tests with small,
|
|
91
|
+
deterministic inputs and update documentation and the Unreleased changelog.
|
|
92
|
+
Explain intentional changes to fills, costs, timestamps, or statistics and
|
|
93
|
+
include migration guidance when existing results can change.
|
|
94
|
+
|
|
95
|
+
Use the bug template to report versions, input shape and timestamps, broker
|
|
96
|
+
settings, expected versus actual results, and a minimal runnable reproduction.
|
|
97
|
+
Use synthetic or shareable data rather than private trading records.
|