fix: unify backtest execution and add forward validation candidates

Share point-in-time signals and a cash-funded order ledger across backtests and trading. Correct eligibility, missing-price fills, quality factors, regime inputs, warmup windows, and cash-injection accounting. Add reproducible registered-strategy comparisons and paired forward candidates with a 20% resize band.

Validation: 125 Python tests, 23 web tests, TypeScript/Vite build, 87 strategy causality checks, and real-data target/order/equity parity for 10 strategies.
This commit is contained in:
2026-10-01 02:24:01 -07:00
parent 761725d2c8
commit 722cade974
45 changed files with 1561 additions and 1021 deletions

View File

@@ -102,13 +102,15 @@ uv run python -m unittest discover tests
**Data manager** (`data_manager.py`): Persistent price storage in `data/{market}.csv` (close) and `data/{market}_{open,high,low,volume}.csv`. First run downloads 10 years of history; every run re-downloads the last 30 days. New tickers from index rebalances are backfilled automatically; tickers missing from a partial batch are retried alone before being marked unavailable (retried weekly). `update(market, tickers, with_open=True)` returns (close, open); `update(..., fields=[...])` returns `{field: panel}`.
**Backtest engine** (`main.py`): Orchestrates data loading, strategy execution, and visualization. `backtest()` has two modes. Close mode (no `open_data`, used by `main.py` and old research scripts) is vectorized and trades at the close the signal saw. Live mode (`open_data` given, used by the web UI through `backtest_runner.py`) keeps the book like `trader.py`: same signal timing (`strategy_positions`), holdings that drift, the $50/0.1% order floor, `trader.RESIZE_BAND`, fees on the orders placed (`simulate_live_book`).
**Backtest engine** (`main.py`, `backtest_runner.py`): all replay paths use `signals.session_targets` and the cash-funded order ledger in `execution.py`. Price strategies may use the session open and fill at its close; close-only inputs and BarStrategy inputs use prior completed bars. Historical membership restricts ranks and selection separately from price history. Missing quotes and zero volume cannot fill; the book checks available cash after fees, handles whole shares/lots, and applies conservative CN closing-limit restrictions and sell-side stamp tax. Registry replays retain at least 315 prior sessions and score all compared strategies on a common actual window. `--offline` on main/precompute reads a frozen local snapshot. Data coverage and adjusted-share/corporate-action limitations are exposed in API metadata.
`precompute_backtest.py` reruns the ledger for cash-injection scenarios; metrics use unit NAV after actual fees. The API and `web/src/shared/strategyTypes.ts` share this cash-flow contract. `ENGINE_VERSION` in `execution.py` identifies the execution protocol; API cache keys also include its version.
**Daily trader** (`trader.py`): Live/forward-testing system with persistent portfolio state in `data/trader_{market}_{strategy}.json`. The `auto` subcommand runs both signal generation and execution in a single invocation — designed for cron. The `simulate` subcommand replays a date range day-by-day with realistic portfolio tracking (fractional shares, cash, commissions). Orders: whole A-share lots are allocated nearest-lot within cash (`_allocate_lots`); a held name is resized only when ≥ `RESIZE_BAND` (10%) off target. Available strategies: `recovery_mom_top10`, `recovery_mom_top20`, `momentum`, `momentum_quality`, `dual_momentum`, `inverse_vol`, `trend_following`, `buy_and_hold`, 32 factor-combo strategies (`fc_{signal}_{freq}` — see `strategies/factor_combo.py`), the robust-momentum bar strategies `overnight_recovery_top10`, `lowrisk_momentum_top15`, `lowbeta_sharpe_mom_top20`, `lowrisk_top20`, the US sleeve portfolios `us_three_sleeve` and `us_momentum_qqq_trend`, and the agent-reviewed `overnight_recovery_agent_top10` (US only, forward only).
**Live books** (`trader.LIVE_BOOKS`): a book is a cohort of strategies, each with its own account, started on one date with one capital and a flat fee per order. The original book (every live strategy, `MONITOR_CAPITAL`, `MARKET_FEES`) keeps state in `data/trader_*.json`; a named book keeps it in `data/books/<book>/`. The monitor runs every book; `auto`/`status`/`log`/`compare`/`replay` and `bridge.py live` take `--book`, and `bridge.py books` lists them for the UI's book selector. `selected_2026_09` (30 strategies, US $10,000 / CN ¥100,000, $2 / ¥10 per order) starts on 2026-09-25 (US) and 2026-09-28 (CN); `overnight_recovery_agent_top10` joined it on 2026-09-30 and starts with the book.
**Strategy protocol** (`strategies/base.py`): All strategies inherit from `Strategy` ABC and implement `generate_signals(data) → DataFrame` where the returned DataFrame contains portfolio weights (rows = dates, columns = assets, values sum to ~1.0 per row). Each strategy is responsible for applying its own 1-day lag via `.shift(1)` to avoid lookahead bias — the backtest engine does not shift. A `BarStrategy` declares `fields` (close/open/high/low/volume) and receives `{field: DataFrame}`; live it uses completed bars only (row d from bars through d-1, traded at the close of d), while close-price strategies read the session's open as their latest price.
**Strategy protocol** (`strategies/base.py`, `signals.py`): All strategies inherit from `Strategy` ABC and implement `generate_signals(data) → DataFrame` where the returned DataFrame contains portfolio weights (rows = dates, columns = assets, values sum to ~1.0 per row). Each strategy returns lagged weights (row t uses bars through t-1). `signals.session_targets` converts this to a feasible execution-day target; the ledger trades at that day’s close. `StockStrategy` accepts a separate `eligible` mask so pre-admission observations remain usable but nonmembers cannot be selected. A `BarStrategy` declares `fields` (close/open/high/low/volume) and receives `{field: DataFrame}`; live it uses completed bars only (row d from bars through d-1, traded at the close of d), while close-price strategies read the session's open as their latest price.
**Strategies**:
- `buy_and_hold.py` — Equal-weight static allocation
@@ -137,6 +139,16 @@ uv run python -m unittest discover tests
- Python 3.12+, managed with `uv`.
- Do not show matplotlib figures when running backtests; use `--no-plot`.
The `*_rb20` strategies are execution-only candidates with a 20% resize band; baseline strategies keep 10%. They run alongside their baselines only in `validation_2026_10`, starting with the first market session on/after 2026-10-02, with matching capital and fees. Existing forward histories are preserved and future daily snapshots record the engine version. These candidates were selected from previously examined historical data and have no independent forward performance yet.
Registered-strategy comparisons use the same ledger:
```bash
uv run python -m research.lab.evaluate --market us --registered overnight_recovery_top10 us_three_sleeve --start 2018-01-01 --end 2026-09-28 --output data/research/results/new_comparison
```
The comparison writes its fixed budget and acceptance rule before running and refuses to overwrite an existing study. Trial producers append versioned JSONL; legacy CSV trial counts remain a lower bound. `stepwise --objective maximin` selects only inside the design period. Previously inspected validation periods must not be described as independent OOS data.
## Server Deployment
Three `systemd --user` services, defined in `deploy/systemd/`, run from `/home/gahow/projects/quant-trading`: