merged_cs_ra_steve_live_v1 review package
Generated 2026-06-04T21:28:34.230049-04:00 Eastern Time / 2026-06-05T01:28:34.230049+00:00 Coordinated Universal Time (UTC). Advisory/review artifact only; no orders submitted for this task.
Financial glossary. Gross Sharpe Ratio is the annualized return/risk ratio before estimated Transaction Costs (TC). Net Sharpe Ratio is the same ratio after estimated Transaction Costs (TC). Compound Annual Growth Rate (CAGR) is annualized compounded return. Maximum Drawdown (MaxDD) is the worst peak-to-trough loss. Out-of-Sample (OOS) means the evaluation period not used for fitting or tuning. Algorithm Trading (AT) is the offline algorithm/backtest path, Paper Trading (PT) is broker paper-account execution, and Real Trading (RT) is live-money execution.
Review verdictneeds fix before live expand
Net Sharpe Ratio0.720
Out-of-Sample (OOS) Sharpe Ratio1.369
Live open orders0
Team-review summary. Code is ready for review. The sleeve is already live at small cap, but Strategy Eval still flags the P0 bias/lookahead gate failure and high component correlation, so the status is needs_fix_before_live_expand, not approved for larger live allocation.
Strategy Algorithm
1
Data inputs + cutoff
yfinance adjusted daily OHLC for 8 ETFs. Exclude current America/New_York date to avoid partial daily bars. Alpaca live_steve is read-only for live status.
2
Feature generation
Cross-sectional score averages 63/126/252-day returns. Regime sleeve computes SPY 20-day volatility, 252-day volatility quantiles, 126-day momentum, and 5-day short-term return.
3
Ranking + weighting
Select top 3 positive cross-sectional momentum names. Regime selects momentum names in low-vol, mean-reversion names in high-vol, or defensive inverse-vol allocation in mid-vol.
4
Merged target construction
Normalize each component, blend 50% cross-sectional momentum and 50% regime-adaptive sleeve, then normalize merged weights to 95% gross exposure.
5
Sizing + risk controls
Apply $5,000 sleeve budget, $100 per-symbol drift threshold, no shorts, marketable day limit policy, 0.3% limit buffer, extended_hours=false.
6
Schedule + duplicate guard
Daemon checks every 5 minutes; eligible after 09:35 Eastern Time on weekdays. Local state last_submit_date prevents more than one submit per date. Discretionary override requires Steve/CEO approval outside this automated path.
7
Order intent + broker path
For drift above threshold, create ExecutionGateway order intents, then submit through AlpacaBroker. No direct TradingClient.submit_order in strategy runner.
8
Monitoring feedback loop
Record signals, targets, and metrics; read Alpaca positions/open orders; publish review status and public signal/report artifacts. Expired/fill status feeds risk review.
Exact code links are embedded in each step. The discretionary override path is intentionally outside the automated runner: Steve/CEO must explicitly approve any override before this strategy changes risk outside the signal and threshold rules.
Algorithm Trading (AT) / Strategy Eval
| Window | 2007-01-03 to 2026-06-02 (4884 days) |
| Gross Sharpe Ratio / Compound Annual Growth Rate (CAGR) / Maximum Drawdown (MaxDD) | 0.803 / 10.99% / -33.55% |
| Net Sharpe Ratio / Compound Annual Growth Rate (CAGR) / Maximum Drawdown (MaxDD) | 0.720 / 9.68% / -34.13% |
| Net total return | 499.17% |
| Out-of-Sample (OOS) split | 2023-01-01; in-sample Sharpe Ratio 0.603, Out-of-Sample (OOS) Sharpe Ratio 1.369 |
| Transaction Costs (TC) | Annual Transaction Costs (TC) drag estimate 1.1876% of Net Asset Value (NAV) per year in the eval summary |
| Factor / correlation | Factor R-squared 0.761; SPY correlation 0.758; maximum strategy correlation 0.890 vs cross_sectional_component |
| Bias gate | FAIL; shift test passed=False |
Rolling Sharpe Ratio Chart
126 trading days: slow/full-context Rolling Sharpe Ratio30 trading days: fast/recent Rolling Sharpe Ratio
Both lines use net-of-Transaction Costs (TC) visualization: gross daily returns minus annual TC drag / 252. Latest 126-day value 2.351; latest 30-day value 2.438. Display is clipped to [-8, 8] for readability after a minimum-volatility filter.
Real Trading vs Paper Trading Cash Results
Comparison verdict: not apples-to-apples. Real Trading (RT) uses Steve's
live_steve Alpaca account and has strategy-tagged orders. Paper Trading (PT) uses the shared paper Alpaca account, but no orders with the av_mergcsra strategy prefix were found in the latest 500 queried paper orders. PT figures below are therefore closest comparable paper-account evidence, not a true mirror of this live strategy.| Field | Real Trading (RT) | Paper Trading (PT) / closest evidence | Difference |
|---|---|---|---|
| Account / environment | live_steve, ACTIVE | paper, ACTIVE | Different accounts |
| Comparison window | Strategy-tagged RT orders from 2026-05-14 13:39:55.222656+00:00 to 2026-06-01 13:39:57.800932+00:00; snapshot 2026-06-05T03:28:29.638204+00:00 | No strategy-tagged PT order window found; snapshot 2026-06-05T03:28:30.534138+00:00 | Not matched |
| Ending account cash | $76,767.35 | $969,051.06 | $-892,283.71 (-92.08% vs PT account cash) |
| Strategy/universe open market value | $4,767.79 | $8,911.93 | $-4,144.14 |
| Realized profit and loss | Not separately available from this read-only snapshot; RT account has pre-existing/account-level cash flows. | Not attributable; no PT mirror orders found. | Unavailable |
| Unrealized profit and loss | $-9.90 | $151.35 | $-161.25 (-106.54% vs PT unrealized P&L) |
| Strategy order evidence | {'EXPIRED': 3, 'FILLED': 29} ; open orders 0 | {} ; open orders 0 | PT has no tagged mirror orders |
Known causes of divergence. Live and paper accounts have different cash bases, account constraints, and holdings. PT is a shared account with unrelated ETF positions and no detected strategy-tagged order history. RT had live fills, 3 expired orders, marketable limit execution, timing differences, and possible manual risk/cap overlays. Because PT is not a true mirror, account-level cash differences should not be interpreted as strategy alpha.
Artifacts used. RT/PT snapshot:
reports/live_review_20260605/rt_pt_cash_compare.json generated by read-only Alpaca live_steve and paper queries. Live code path: scripts/run_merged_cs_ra_alpaca_live.py. Paper code path: scripts/run_merged_cs_ra_alpaca_paper.py.Real Trading (RT) / Current Live Status
Account profilelive_steve
Portfolio value$90,945.20
Cash$76,767.35
Rebalance neededFalse
| Signal date | 2026-06-03 completed close; latest yfinance daily bar 2026-06-04; excludes Eastern Time current-date partial bar=True |
| Regime | low_vol_momentum; SPY 20-day volatility 9.90%; low cut 10.18%; high cut 12.92% |
| Order status counts | {'EXPIRED': 3, 'FILLED': 29}; current strategy open orders 0; other open orders 0 |
| Review-task orders submitted | 0 |
Latest Signal Lines
| Symbol | CS rank | CS score | RA 126-day momentum | RA 5-day return | Target weight | Current | Target | Drift |
|---|---|---|---|---|---|---|---|---|
| EEM | 1 | 35.08% | 30.66% | 2.24% | 31.67% | $1,580.17 | $1,583.33 | $-3.16 |
| DBC | 2 | 32.72% | 35.42% | 2.71% | 31.67% | $1,613.43 | $1,583.33 | $30.10 |
| QQQ | 3 | 28.61% | 20.89% | 2.02% | 31.67% | $1,565.75 | $1,583.33 | $-17.59 |
| IWM | 4 | 23.06% | 17.72% | -0.93% | 0.00% | $0.00 | $0.00 | $0.00 |
| SPY | 5 | 16.87% | 11.50% | 0.50% | 0.00% | $0.00 | $0.00 | $0.00 |
| EFA | 6 | 11.59% | 11.92% | -0.62% | 0.00% | $0.00 | $0.00 | $0.00 |
| GLD | 7 | 7.32% | 4.65% | -0.15% | 0.00% | $0.00 | $0.00 | $0.00 |
| TLT | 8 | -0.05% | -1.70% | 0.41% | 0.00% | $0.00 | $0.00 | $0.00 |
Paper Trading (PT) Status
Paper Trading (PT) runner exists for the same strategy family and should use the shared
alpaca_paper_main profile. Latest review snapshot did not find matching recent paper orders. Shared-account isolation must be by strategy_id and client_order_id; Paper Trading (PT) profit and loss is not treated as Real Trading (RT) live-account profit and loss.Code Walkthrough
Review entrypoints
live runner · paper runner · T+1 open backtest · manifest · review note
signal generationtarget sizingExecutionGatewayAlpacaBrokerdaemon guard
Signal components
cross_sectional_momentum: average 63/126/252-day returns, top 3 monthly carry-forward.
regime_adaptive: SPY volatility regime chooses momentum, short-term mean reversion, or defensive inverse-volatility allocation.
Execution path: completed daily closes -> merged_weights(cs, ra, 0.5) -> $5k target values -> threshold drift check -> ExecutionGateway.create_order_intent -> AlpacaBroker. This review report performed read-only status checks only.
Export Provenance
| Metrics producer | /workspace/steve/paw_auto_research/tools/eval_strategy.py using generated CSV inputs under reports/review_20260603/inputs/ |
| Strategy Eval command | python /workspace/steve/paw_auto_research/tools/eval_strategy.py --positions reports/review_20260603/inputs/positions.csv --asset-returns reports/review_20260603/inputs/asset_returns.csv --gross-returns reports/review_20260603/inputs/gross_returns.csv --metadata-json reports/review_20260603/inputs/metadata.json --factor-returns reports/review_20260603/inputs/factor_returns.csv --existing-returns reports/review_20260603/inputs/existing_returns.csv --strategy-label merged_cs_ra_dual_review_20260603 --start-date 2007-01-03 --oos-start-date 2023-01-01 --aum-for-tc 5000000 --out-dir reports/review_20260603/eval-report |
| Data source and cutoff | Algorithm Trading (AT) and signal inputs use yfinance adjusted daily OHLC. Live runner excludes the current America/New_York date to avoid partial daily bars. Current report snapshot checked at 2026-06-05T01:28:34.230049+00:00 and used signal date 2026-06-03. |
| Live status source | Read-only Alpaca live_steve account query for Real Trading (RT) positions and open orders. No paper/live orders were submitted. |
| Rolling Sharpe Ratio chart producer | Local report-generation script in this review session computed 126-trading-day and 30-trading-day rolling values from gross_returns.csv; net visualization subtracts annual Transaction Costs (TC) drag / 252 before rolling calculation. |
| Publisher / deploy path | Cloudflare Pages project merged-cs-ra-live-review-20260605; deployed from reports/live_review_20260605 using npx -p wrangler@4.95.0 -p esbuild wrangler pages deploy ... --branch main. |
Known Risks
- P0 bias/lookahead test currently fails; timestamp-specific validation is required before expanding live size.
- Schedule guard is weekday/time based and should be exchange-calendar aware.
- Duplicate-submit guard is local state keyed by date; process/state loss needs review.
- Marketable DAY limits have simplified spread/latency/partial-fill assumptions; 3 historical expired orders exist.
- Portfolio interaction maximum correlation is high at 0.890; diversification benefit may be weaker than desired.