wehub-resource-sync
178 lines
7.6 KiB
YAML
178 lines
7.6 KiB
YAML
strategy_id: us_firm_characteristics
|
|
setup_version: v1
|
|
|
|
universe:
|
|
min_price_usd: 5
|
|
min_adv_usd: 1000000
|
|
exclusions: [ADRs, REITs, financials]
|
|
identifiers: anonymized_permno_based
|
|
note: Filters applied by the data provider before anonymization; cannot be independently verified.
|
|
|
|
decision:
|
|
cadence: monthly_month_end
|
|
snapshot: month_end_close
|
|
execution_delay: next_bar_open
|
|
characteristic_lag: 6_months
|
|
|
|
# Engine-level execution defaults. Single source of truth — Ch16-19 notebooks
|
|
# read these via get_backtest_config(); never declare a local INITIAL_CASH
|
|
# or share_type constant. Changing values here invalidates every existing
|
|
# backtest_hash for this case study (cash + share_type are spec-hash inputs).
|
|
#
|
|
# Moment-based allocators (inverse_vol, risk_parity, hrp, mvo_ledoit_wolf)
|
|
# are EXCLUDED for this case study. US firm characteristics is a returns-only
|
|
# cross-sectional panel keyed by integer firm IDs (no continuous per-symbol
|
|
# price series), so a per-symbol trailing volatility or covariance is not
|
|
# well-defined: firms enter and exit, and 12 monthly observations cannot
|
|
# support a covariance over a top-50 cross-section. Under that data the
|
|
# moment allocators either degenerate to equal weight (mvo/hrp hit their
|
|
# observation-count guards and fall back) or produce numbers from a
|
|
# covariance estimate the data cannot justify (inverse_vol, risk_parity).
|
|
# Only lookback-free allocators (equal_weight, score_weighted,
|
|
# conformal_weighted) are swept and eligible for carrier selection.
|
|
# ``allocator_lookback`` below is retained only because get_backtest_config()
|
|
# requires the key; it is unused while no moment allocator is configured.
|
|
#
|
|
# Cash defaults to 1_000_000 rather than 100_000: monthly-cadence
|
|
# long-short portfolios at top_k=50 carry ~$10K per leg per name at the
|
|
# 100k tier — below realistic round-trip granularity for institutional
|
|
# US equity. 1M keeps the spec-implied position sizes (notional per name
|
|
# / fixed-cost ratio) in a regime the backtest engine resolves cleanly.
|
|
execution:
|
|
initial_cash: 1_000_000 # Monthly long-short top_k=50 needs $1M to size cleanly
|
|
share_type: integer # US equities trade in whole shares
|
|
allocator_lookback: 12 # 1 year of monthly bars
|
|
|
|
mapping:
|
|
class: long_short_decile_rebalance
|
|
position_state_space: long_short
|
|
entry_logic: decile_sort_long_top_short_bottom
|
|
sizing: equal_weight_within_decile
|
|
|
|
costs:
|
|
class: material
|
|
components: [spread, commission, market_impact, borrow_cost]
|
|
per_leg_cost_bps_range: [5, 20]
|
|
borrow_cost_note: Long-short requires borrow for the short leg.
|
|
era_note: Pre-2001 spreads 15-30 bps; post-2001 (decimalization) 5-15 bps.
|
|
|
|
evaluation:
|
|
n_splits: 10
|
|
train_size: 10Y
|
|
val_size: 1Y
|
|
holdout_start: '2016-01-01'
|
|
holdout_end: '2016-12-31'
|
|
calendar: null # Monthly returns; calendar-aware splitting needs daily frequency.
|
|
periods_per_year: 12
|
|
|
|
labels:
|
|
primary: fwd_ret_1m
|
|
buffer: 1M
|
|
variants:
|
|
- fwd_ret_1m_win
|
|
- fwd_class_1m
|
|
variant_buffers:
|
|
fwd_ret_1m_win: 1M
|
|
fwd_class_1m: 1M
|
|
# Vectorized-backtest thinning step per label: number of schedule slots
|
|
# to advance per trade so holding periods don't overlap. Authored from
|
|
# (schedule cadence, label horizon); add an entry here for any new label.
|
|
rebalance_step:
|
|
fwd_ret_1m: 1
|
|
fwd_ret_1m_win: 1
|
|
fwd_class_1m: 1
|
|
# Continuous return that each classification label is derived from.
|
|
# IC for classification predictions is computed against this column;
|
|
# AUC/accuracy/log_loss are computed against the binary label itself.
|
|
classification_eval_label:
|
|
fwd_class_1m: fwd_ret_1m
|
|
|
|
backtest:
|
|
rebalance:
|
|
# Per-asset rebalance thresholds. A rebalance is skipped when the
|
|
# per-asset weight change is below min_weight_change AND the resulting
|
|
# trade notional is below min_trade_value. At top_k=50 the equal-weight
|
|
# per-asset weight is 1/50 = 2%, comfortably above the 0.5% threshold.
|
|
default:
|
|
min_weight_change: 0.005
|
|
min_trade_value: 100.0
|
|
benchmark:
|
|
min_weight_change: 0.0
|
|
min_trade_value: 0.0
|
|
sweep:
|
|
# Iteration controls per stage. ``signal: 0`` means "all predictions";
|
|
# downstream stages take the top-N from the upstream stage's rank-1.
|
|
# Notebooks read these via get_top_n_predictions(case_study, stage).
|
|
top_n_predictions:
|
|
signal: 0 # all signal predictions (eq-weight baseline)
|
|
allocation_cheap: 0 # score_weighted — all signal preds
|
|
allocation_expensive: 10 # conformal_weighted — top-10 by signal Sharpe
|
|
cost_sensitivity: 1 # top-1 of {signal+allocation} per label
|
|
risk_overlay: 1 # top-1 of {signal+allocation} per label
|
|
# Skip MVO/HRP when allocator runtime is the bottleneck (intraday CSes).
|
|
expensive_allocators_skip: false
|
|
# Allocators routed through ``allocation_expensive`` (top-10 by signal Sharpe).
|
|
# All others (equal_weight, score_weighted) take the cheap path.
|
|
expensive_allocators: [conformal_weighted]
|
|
# Ch16 backtest is equal-weight sized within the selected set. The
|
|
# selection rule is one of: top-k (rank), percentile band (long-only
|
|
# cutoff), or quantile buckets (e.g. quintile long-short — combines
|
|
# with the long-short mapping). Only top_k_grid is active here;
|
|
# uncomment percentile_grid / quantile_grid to add other selection axes
|
|
# for any label.
|
|
top_k_grid:
|
|
fwd_ret_1m: [5, 10, 20, 50]
|
|
fwd_ret_1m_win: [5, 10, 20, 50]
|
|
fwd_class_1m: [5, 10, 20, 50]
|
|
# percentile_grid:
|
|
# fwd_ret_1m: [80, 90, 95]
|
|
# quantile_grid:
|
|
# fwd_ret_1m: [5, 10]
|
|
# Ch17 portfolio: reuses top_k_grid above and sweeps over allocators.
|
|
# Only lookback-free allocators are listed — moment-based allocators
|
|
# (inverse_vol, risk_parity, hrp, mvo_ledoit_wolf) are excluded because
|
|
# this returns-only firm-characteristics panel has no per-symbol price
|
|
# series to estimate volatility or covariance from (see the execution
|
|
# block above). No max-weight cap.
|
|
allocators:
|
|
- {name: equal_weight, method: equal_weight}
|
|
- {name: score_weighted, method: score_weighted}
|
|
- {name: conformal_weighted, method: conformal_weighted}
|
|
# Ch18 cost sensitivity (bps regime).
|
|
cost_grid_bps: [0, 1, 2, 3, 5, 7, 10, 15, 20, 30, 50]
|
|
# Ch19 risk overlays.
|
|
risk_controls:
|
|
position:
|
|
- {name: stop_loss_3pct, type: stop_loss, threshold: 0.03}
|
|
- {name: stop_loss_5pct, type: stop_loss, threshold: 0.05}
|
|
- {name: stop_loss_10pct, type: stop_loss, threshold: 0.10}
|
|
- {name: stop_loss_15pct, type: stop_loss, threshold: 0.15}
|
|
- {name: trailing_1pct, type: trailing_stop, threshold: 0.01}
|
|
- {name: trailing_2pct, type: trailing_stop, threshold: 0.02}
|
|
- {name: trailing_3pct, type: trailing_stop, threshold: 0.03}
|
|
- {name: trailing_5pct, type: trailing_stop, threshold: 0.05}
|
|
- {name: trailing_10pct, type: trailing_stop, threshold: 0.10}
|
|
- {name: trailing_15pct, type: trailing_stop, threshold: 0.15}
|
|
- {name: trailing_20pct, type: trailing_stop, threshold: 0.20}
|
|
- {name: time_exit_10, type: time_exit, bars: 10}
|
|
- {name: time_exit_20, type: time_exit, bars: 20}
|
|
- {name: time_exit_40, type: time_exit, bars: 40}
|
|
|
|
modeling:
|
|
gbm:
|
|
libraries: [lightgbm]
|
|
preset: default
|
|
device: gpu
|
|
latent_factors:
|
|
persistent_entities: false
|
|
model_kwargs:
|
|
sdf:
|
|
checkpoint_epochs: [256, 512, 768, 1024, 1280]
|
|
beta_checkpoint_epochs: [256]
|
|
beta_default_checkpoint: 256
|
|
|
|
causal:
|
|
treatment: r12_2
|
|
confounders: [Beta, IdioVol, LME, Variance]
|
|
method: walk_forward_dml
|