Replay refreshes each price panel once for the union of tickers and then
replays offline; target weights for all sessions come from one causal
signal pass instead of re-running the strategy per day (verified identical
daily equity for v7_vt36, recovery_mom_top10 and fc monthly).
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
Tickers count as alive for 10 rows after their last bar, so a newest row
printed by a few early tickers is dropped until the rest arrive. The
benchmark-print session rule now applies only to the mixed US/HK ETF panel:
Yahoo's 510300.SS lags CN stocks by days and would stall the CN book.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
- Monthly/yearly returns chain from the previous period's close (the first
trading day of every period was dropped; up to ~9pp/yr error).
- With cash injections, metrics/drawdowns/periodic returns are time-weighted
on the no-injection curve; capitalFlows reports invested/profit separately.
- Benchmarks cover exactly the strategies' dates.
- CAGR uses calendar time, so gaps in live records do not inflate it.
- Backtests skip the per-trade fee for rebalances below the live minimum
order and keep a ruined portfolio at zero.
- live: exact strategy match, ineligible strategies hidden, currency,
startDate, same-window benchmark, stale-price fallback, equity anchored at
the initial capital.
- PIT membership falls back to the cached history when the Wikipedia change
log cannot be parsed (the table is no longer on the page).
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
- Process every completed session since the last recorded one, so monitor
downtime and late data are caught up; today's bar only counts after the
exchange closes. The monitor's evening phase always goes through auto and
catches up at startup.
- Tickers without a real closing print are valued at the last price but not
traded; held tickers missing from the panel keep their last recorded price
instead of being valued at zero.
- Use registry parameters (live momentum strategies ran top_n=len//10=49
instead of the backtested 20).
- A-share orders use 100-share board lots; per-market default capital
(US $10k, CN ¥70k); HK tickers are converted to USD (HKD=X, peg fallback)
and exempt from the US per-share fee.
- ETF strategies only run in the US book; buy_and_hold is excluded from live
because its ~$20 slices never reach the minimum order size.
- CN benchmark switches to 510300.SS (Yahoo's 000300.SS is truncated).
- New 'replay' command rebuilds live state files from each start date.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
The append-only updater only fetched when last_date+1 < local date, never
revisited appended rows and forward-filled before saving. First-seen values
(partial intraday bars, late or missing prints) were frozen permanently: vs
Yahoo, 22% (US), 58% (CN) and 27% (ETF) of recent days had >0.5% daily-return
errors, CN never received same-day data, and HK tickers created US-holiday rows.
Panels are now stored raw (NaN = no print) and forward-filled on read; every
update re-downloads the last 30 days for the requested tickers, rescales older
history when Yahoo's split/dividend adjustment moves, drops rows printed by a
minority of tickers, and writes atomically under a per-market file lock.
Tickers Yahoo has no data for are retried weekly instead of every run.
rebuild() repairs panels written by the old updater.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
Four research directions beyond V7+VT36:
1. single_stock_swing: 20 famous stocks (Mag 7 + others), per-stock
optimized swing trading. High-vol growth stocks (AMD Sharpe 1.55,
TSLA 1.54) work best, but overfitting risk is extreme — universal
params only TSLA is viable. Not competitive with V7.
2. v7_literature_alpha: 9 academic directions (VIX overlay, Kelly
sizing, multi-MA, cross-asset, momentum acceleration, VIX mean-
reversion, vol-adaptive PT, combined). V3's regime engine already
implicitly captures most literature signals. MA130 marginally
better than MA150 (+0.02 Sharpe, within noise).
3. new_frameworks_eval: volatility trading (SVXY risk-off) and
calendar effects (turn-of-month). SVXY and V7 regime structurally
conflict — SVXY crashes exactly when V7 goes risk-off.
Turn-of-month has decent Sharpe (1.30) but only 28% annual.
Nothing beats V7.
4. smart_dca_eval: fixed/VIX-scaled/MA-deviation/value-averaging/RSI
DCA into SPY/QQQ/TQQQ/UPRO + V7 hybrids. Smart DCA barely beats
fixed DCA. Any DCA hybrid dilutes V7's alpha. DCA only useful for
new monthly contributions that can't lump-sum into V7.
Conclusion: V7+VT36 remains SOTA across all tested frameworks.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
- Register trend_rider_v7_vt36 (target_vol=0.36, min_lev=0.75) in
strategy registry, ETF universe map, and bridge metadata.
10y backtest: Ann 60.5%, Sharpe 1.87, MaxDD -29.2%.
- Add hot-reload to monitor: each phase re-imports trader module to
pick up newly registered strategies without restart. New strategies
are logged on detection.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
V3's output only includes {SPY, TQQQ, UPRO, GLD, DBC}. When PT
triggered, park_col resolved to "" (cash at 0%) instead of SHY.
Now injects SHY column before the PT loop if present in data.
Impact: ~0 in 2016-2026 (rising rates made SHY slightly negative),
but fixes ~0.6%/yr drag in normal rate environments (SHY ~4%/yr,
14.3% of days in PT-park).
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
- DEFAULT_MONITOR_STRATEGIES now includes ALL strategies (stock + ETF)
instead of excluding ETF strategies. The cmd_morning/evening/auto
already route ETF strategies to the correct data pipeline via
strategy_universe() and strategy_data_market().
- Register trend_rider_v7, v7_vt24, v7_vt32 in bridge.py STRATEGY_META
so they appear in the stock-agent frontend via /api/strategies.
- Monitor now runs as a background daemon with logs written to
logs/monitor.log (PYTHONUNBUFFERED=1, no tmux dependency).
PID saved to logs/monitor.pid.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Three-layer strategy for leveraged ETF portfolios:
Layer 1: V3 regime engine (MA150) — SPY technicals for risk-on/off
Layer 2: Vol-target overlay (28%, clip 0.6-1.0) — scale by realized vol
Layer 3: Profit-take with hysteresis (+30% → clear to SHY, restore <20%)
The profit-take exploits a structural property of 3x leveraged ETFs:
after large gains, volatility drag on the inflated base erodes compound
returns. Clearing the position locks in geometric gains before the drag
takes effect — this is rebalancing alpha, not prediction alpha.
10y backtest (2016-2026, 10bps one-way cost):
Ann 54.7%, Sharpe(rf=5%) 1.72, MaxDD -25.7%, Sortino 2.23
Also registers trend_rider_v7, trend_rider_v7_vt24, trend_rider_v7_vt32
in the trader strategy registry and ETF_STRATEGY_UNIVERSES.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Add .qoder/ (local tool settings) and SEC fundamental data
(sec_frames/, sec_company_tickers.json) to prevent large
downloaded datasets from being tracked.
The existing framework fetches today's S&P 500 constituents from Wikipedia
and applies that list to the entire 10-year price history — classic
survivorship bias. Stocks that went bankrupt or were removed for poor
performance are absent, while today's winners (which may have been minor
names 10 years ago) are implicitly selected. This materially inflates
reported strategy returns.
New pipeline:
- universe_history.py reconstructs per-ticker membership intervals by
walking Wikipedia's "Selected changes" table backward from today.
- research/fetch_historical.py downloads prices for all 848 tickers
that were ever members (Yahoo returns ~675 of them; ~170 fully
delisted names are unavailable — remaining partial bias).
- research/pit_backtest.py masks prices to NaN outside membership
windows so strategies naturally cannot select non-members.
- research/strategies_plus.py adds RecoveryMomentumPlus (generalized
Recovery+Momentum with configurable weighting / blend / regime hook)
and an EnsembleStrategy.
- research/optimize.py runs five experiments: bias drift, hyperparameter
sweep (2016-2022 train / 2023-2026 test), SPY MA regime filter,
weighting schemes, and an uncorrelated-config ensemble.
Headline finding: the biased backtest reports 40.9% CAGR for
recovery_mom_top10 over 2016-2026; the point-in-time version reports
22.4% (vs 14.0% SPY buy-and-hold). True edge is ~8pp CAGR, not ~27pp.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
Add MARKET_FEES {us: 2, cn: 5} so the monitor and cron (auto) paths
automatically apply the correct local-currency fixed commission without
needing a per-strategy override. CLI --fixed-fee still wins when set
explicitly for auto; monitor now always resolves from the table so its
banner and each strategy sub-call agree.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
cmd_evening (used by the monitor path) only updated the simple daily_equity
dict, so daily_log had gaps on every monitor-driven day. Mirror cmd_auto's
pattern and call record_daily_snapshot so each strategy's NAV is recorded
every trading day, even when no trades execute.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
New FactorComboStrategy class (strategies/factor_combo.py) implements
8 champion factor signals (4 US, 4 CN) discovered through iterative
factor research, each at 4 rebalancing frequencies (daily/weekly/
biweekly/monthly). Registered in trader.py as fc_{signal}_{freq}.
Existing strategies and state files are untouched — safe to git pull
and restart monitor on server.
Also includes factor research scripts (factor_loop.py, factor_research.py,
etc.) used to discover and validate these factors.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
