Pybroker Rotational Trading

PyBroker Rotational Trading

Overview

Build rotational PyBroker strategies that hold the top-ranked symbols in a universe and rotate out names that fall from favor. Execution functions score symbols with ctx.long_score and ctx.short_score, cap positions with Strategy.set_max_long_positions and set_max_short_positions, and Strategy.enable_rotation(worst_rank_held=...) liquidates and refills slots each bar from the top-ranked candidates, optionally sized with a custom sizer over RotationContext. Also covers the simpler ranked-cap mode and dynamic universes via SymbolSelector.

Workflow

1. Extract the rotation spec: the candidate universe (fixed list or a SymbolSelector screen), the ranking signal for each side, long and/or short legs, position slots per side, hold band (worst_rank_held) versus ranked-cap prioritization only, sizing (default equal weight or a custom sizer), stops and fill prices, backtest versus walkforward, and the desired deliverable file.
2. Ask only for missing blockers. If details are absent but noncritical, make conservative assumptions and state them in the final answer or code comments.
3. Read references/wiki-index.md to choose the smallest relevant wiki page. For nontrivial rotation work, also read references/rotational-patterns.md.
4. Build a complete runnable rotation surface:
   • start scripts with pybroker.disable_progress_bar() and pybroker.enable_data_source_cache("<name>")
   • compute the ranking indicator as NumPy over BarData arrays and set ctx.long_score/ctx.short_score in the execution function
   • cap slots with strategy.set_max_long_positions(n)/set_max_short_positions(n) (never the deprecated StrategyConfig fields), then either stop there for ranked-cap mode or call strategy.enable_rotation(worst_rank_held=..., sizer=...) for hold-band rotation
   • under rotation, let the execution function set only scores, stops, and fill prices — orders it places are ignored; in ranked-cap mode, keep placing orders normally with at most one order side per symbol per bar
   • run backtest/walkforward with warmup= covering the ranking indicator's lookback and inspect result.orders to confirm rotation entries and hold-band exits
5. Validate the produced code as far as the environment allows. At minimum, run syntax checks for created Python files. When practical, run against a small local DataFrame and confirm result.orders shows entries capped at the position limits and exits for symbols that fall out of the hold band.

Implementation Rules

• Treat PyBroker as a backtesting framework, not a source of financial advice. State assumptions explicitly (universe, ranking signal, hold band, costs) and make no performance claims unsupported by the produced backtest.
• Use completed historical bar data only. Indicator logic must be lookahead-free: never index a full-length array with a negative index (it silently wraps to the end of the series, the future) and never shift future values backward; a value at bar i may depend only on inputs at index i and earlier. Self-test novel indicator logic with the bump-last-bar check: change only the final input bar and assert every earlier output is unchanged.
• Two ranking modes exist. Ranked-cap mode (set_max_*_positions plus scores, no enable_rotation) keeps execution functions in charge of orders and uses scores only to prioritize signals when a cap binds; symbols that set no score sort as 0.0 and unrankable scores sort last. Rotation mode (enable_rotation) drives all trading from scores. Choose ranked-cap for prioritizing entry signals, rotation for hold-the-top-N portfolios.
• Rank with ctx.long_score (buy and cover signals) and ctx.short_score (sell signals). Scores rank the whole portfolio across all executions, descending, with the symbol name as a deterministic tiebreak.
• strategy.set_max_long_positions(n)/set_max_short_positions(n) accept an int greater than 0, a searchable Hyperparam, or None for unlimited. The StrategyConfig fields of the same names are deprecated and the setters take precedence.
• Rotation mechanics: each bar, held positions ranked worse than worst_rank_held — or holding an unrankable score, even when another execution opened them — are liquidated, and the top-ranked candidates fill the remaining free slots at equal weight 1 / (long slots + short slots). Candidates ranked outside the hold band are never entered. enable_rotation(None) disables rotation and clears the sizer.
• Rotation is exclusive: orders placed by execution functions are discarded, but fill prices and stops (including hold_bars) set during execution are kept and applied to the orders rotation places. Under rotation, the execution function's job is scores, stops, and fill prices only.
• A None or NaN score excludes the symbol from the rank map, which liquidates a held position. NaN indicator warmup is harmless before positions exist, but an indicator that goes NaN mid-series forces an exit — confirm that is intended.
• A symbol picked by both the long and short leg goes to the side where it ranks better; ties go long. A symbol with no bar on the current date keeps its position slot, and in-flight pending orders hold their slots too.
• A rotation sizer is a Callable[[RotationContext], None] invoked after rotation decides what to trade; long_ranks/short_ranks are 1-based with 1 the best. Override entry sizes with ctx.buy_shares = ctx.calc_target_shares(weight) (or ctx.sell_shares for short entries) guarded by if ctx.buy_shares is not None:, and never override the sell or cover signals rotation set. A sizer without rotation enabled raises ValueError.
• worst_rank_held requires at least one position cap and must be greater than or equal to every cap that is set. On any rotation ValueError, match the message against the Common Errors table in references/rotational-patterns.md before changing code.
• To rotate within a screened universe, pass a SymbolSelector callable as the add_execution symbols: it runs once per walkforward window on the window's training data, requires a DataFrame data source and a training window (backtest and train_size=0 raise ValueError), and positions in symbols a later window drops are closed at that window's first bar.
• set_max_long_positions, set_max_short_positions, and enable_rotation(worst_rank_held=...) all accept a pybroker.hyperparam(...), so slots and the hold band are searchable with Strategy.optimize.
• Never use pandas to implement indicator or execution logic: write indicators as vectorized NumPy over BarData arrays (Numba @njit for explicit loops) and read ctx.* NumPy arrays in execution functions — no pd.Series/pd.DataFrame construction and no .rolling/.ewm/.shift/.apply in either. A SymbolSelector is a sanctioned pandas boundary: it receives the DataFrame PyBroker hands it.
• An indicator returns one full-length one-dimensional array with one value per input bar, warmup left-padded with NaN — never a shortened array.
• Keep feature data out-of-band: never widen or mutate the user's input DataFrame.
• Enable caching while iterating: pybroker.enable_data_source_cache(name) to skip refetching data, or pybroker.enable_caches(name) to also cache indicators. Call pybroker.disable_progress_bar() in agent-run scripts, and add pybroker.disable_logging() when running many backtests.
• Rotation's orders fill at PriceType.MIDDLE — the midpoint of the low and high of the execution bar, which under the default buy_delay/sell_delay of 1 is the bar after the score, so PriceType.CLOSE means the next bar's close. Fill prices set in the execution function survive rotation: assign ctx.buy_fill_price / ctx.sell_fill_price a PriceType (OPEN, HIGH, LOW, CLOSE, MIDDLE, AVERAGE), a number, or a (symbol, bar_data) callable, and note they read back as None rather than MIDDLE until set. Because a whole universe rotates on one bar, this choice moves every leg at once.
• StrategyConfig.exit_on_last_bar defaults to False. A rotational strategy is usually fully invested when the data ends, so leaving it off strands one open position per held slot: none of them become Trades, so trade_count, win_rate, total_pnl and every other trade-level metric silently exclude them while their P&L sits in unrealized_pnl. Set exit_on_last_bar=True whenever trade statistics are reported; exits fill at exit_sell_fill_price / exit_cover_fill_price, both PriceType.MIDDLE. calc_bootstrap is a backtest/walkforward parameter defaulting to False, not a StrategyConfig field; pass calc_bootstrap=True to populate result.bootstrap with BCa confidence intervals for profit factor and Sharpe plus percentile bounds on max drawdown.
• Report result.metrics_df as the human-readable summary, and inspect result.orders to confirm rotation behavior. When structured output is needed (agent parsing, saved report files, downstream tools), use result.to_json() / result.to_json_str(): the default payload serializes metrics, trades, orders, and bootstrap capped at max_rows=100 rows per table, symbols= filters to specific tickers, and include= opts into portfolio/positions/metrics_df/signals/stops. Do not replace the metrics_df print outright: the default JSON payload (trades plus orders) is usually larger than the metrics table.
• On a Numba compilation or typing error in an @njit indicator, re-run once with the environment variable NUMBA_DISABLE_JIT=1 to get a readable Python traceback, fix the error, then re-run with JIT enabled. Never leave JIT disabled in the final script.
• Guard lookbacks with ctx.bars or warmup=. In ranked-cap mode set at most one order side per symbol per bar; in rotation mode a symbol scored on both legs is resolved by rotation's overlap rule, never by placing both orders.
• If exact API names, constructor parameters, or methods matter, read references/api-public-surface.md.
• For exact type signatures — set_max_long_positions, set_max_short_positions, and enable_rotation in references/pybroker_strategy.pyi, RotationContext and the long_score/short_score attributes in references/pybroker_context.pyi, SymbolSelector in references/pybroker_types.pyi — read the matching references/pybroker_*.pyi stub.
• If the user wants a standalone file, copy and adapt assets/rotation_template.py.

Common Deliverables

• Standalone .py rotational backtest that ranks a universe and holds the top-N inside a hold band.
• Ranked-cap prioritization (long_score/short_score plus position caps) added to an existing multi-symbol strategy.
• Custom rotation sizer implementing rank-weighted or otherwise non-equal entry allocation.
• Long/short rotation with both legs, overlap handling, and stops carried into rotation orders.
• Migration of deprecated StrategyConfig.max_long_positions/max_short_positions code to the current API.
• Debugging notes for rotation ValueErrors, unrankable-score liquidations, and ignored execution-function orders.

Resources

• references/wiki-index.md: start here for topic routing across the bundled references.
• references/wiki-10-rotational-trading.md: hold-band rotation with enable_rotation and custom position sizing with a sizer.
• references/wiki-04-ranking-long-and-short-signals.md: ranking long and short signals with scores and position caps.
• references/wiki-18-dynamic-symbol-selection.md: screening a candidate universe with a SymbolSelector.
• references/rotational-patterns.md: load when writing nontrivial rotation code; the mode decision, rotation mechanics, sizer recipes, the rotation error table, and the validation checklist.
• references/api-public-surface.md: generated public API signatures and first docstring sentences from local source.
• references/pybroker_strategy.pyi: generated type stubs for Strategy, StrategyConfig, TestResult, and the optimization types.
• references/pybroker_context.pyi: generated type stubs for ExecContext (including its writable order/stop attributes), IntervalContext, RotationContext, ExecResult, and the slippage models.
• references/pybroker_model.pyi: generated type stubs for model(), indicator(), vector helpers, data sources, and top-level module functions.
• references/pybroker_types.pyi: generated type stubs for enums, BarData, Portfolio, order/trade/position records, and evaluation result types.
• assets/rotation_template.py: copy and adapt when creating a new standalone rotational trading script.