README.md

metadata

title: Updraft
emoji: 🪁
colorFrom: red
colorTo: yellow
sdk: docker
app_port: 7860
pinned: false
short_description: Rides the strongest quality stocks; 3 names, monthly

🪁 Updraft

The Quality-Trend 3 strategy, running on its own.

Three large US stocks with strong quality and capital discipline and the strongest price trends, held on an Alpaca paper account and rebalanced automatically once a month. The page shows the live portfolio, growth vs the S&P 500 since the first trade, the trade log and this month's ranking.

Rule (backtested in the RS-VCP / Quant Model research, scripts/finalize_qt3_managed.py):

1. Universe — the 1,000 largest US companies by market cap; price ≥ $5; ≥ $5M a day traded; US-dollar statements.
2. Quality pool — top 30% of a capital-discipline composite: basic-diluted gap, quality score (ROE, ROA, gross margin, FCF conversion), −equity financing, FCF yield, ROE, buyback yield, net share reduction.
3. Trend — the pool ranked by 0.5 · pct(MA50/MA200) + 0.5 · pct(6-1-month return).
4. Monthly — keep holdings still ranked in the top 6, sell the rest, buy from the top until 3 are held, re-split equally (1% kept in cash). Sells first, waited for, then buys sized to the cash actually available (no margin).

Backtest 1999-08 → 2026-09, after 10 bps per side, point-in-time lists incl. delisted names: +30.9%/yr (S&P +8.5%), worst fall −34%, any 12 months median +24% (in profit 87% of the time). Backtests flatter; expect less.

How it runs without a PC

• Data — each day the Space builds the ranking itself from the Sharadar API (Nasdaq Data Link, ~40 requests, ~70 s). It reproduces the research snapshot exactly (verified 2026-10-13: same top 10, same 250-name pool).
• Trading — POST /api/rebalance?dry=0 with X-Rebalance-Token. It trades only when all hold: market open; one of the first 5 trading days of the month (or the very first run); not already rebalanced this month; ranking data ≤ 5 days old. Every other call is a harmless no-op. Orders carry client_order_id = qt3-YYYYMMDD-SYM-side, so a retried call cannot duplicate them.
• Schedule (built in) — the Space runs its own timer (qt3/scheduler.py): every 15 minutes it rebuilds the ranking if it is older than 20 hours, and on weekdays 10:00–15:30 New York time it attempts the guarded rebalance above — so it trades on the first trading day of each month and does nothing otherwise. No GitHub needed. Set AUTO_REBALANCE=0 to switch it off. The GitHub workflows in .github/workflows/ remain an optional backup.
• Keep-awake — a free Space sleeps after 48 h without visitors, and a sleeping Space's timer stops. A Hugging Face Scheduled Job (updraft-keepalive, id 6acdd86bb0ad3e9da5ea6936, cron 0 */12 * * *, cpu-basic) visits /api/health every 12 hours — a few seconds of compute, well under $0.10 a month; pause or delete it under the account's Jobs. Alternatives: a free uptime monitor (UptimeRobot / cron-job.org) or paid hardware with sleep off.
• State — all in Alpaca (positions, orders, portfolio history). A restart only rebuilds the ranking cache.

Settings (Space → Settings → Variables and secrets)

Secret	Purpose
NASDAQ_DATA_LINK_API_KEY	Sharadar data for the ranking
ALPACA_API_KEY_ID, ALPACA_API_SECRET_KEY	the paper account this Space manages (use a dedicated account)
REBALANCE_TOKEN	a long random string; the same value goes into the GitHub secret UPDRAFT_REBALANCE_TOKEN

Optional variables: ALPACA_DATA_FEED (default iex), CASH_BUFFER (0.01), QT3_INCEPTION (YYYY-MM-DD for the public stats). Live trading is off unless both ALPACA_PAPER=false and ALPACA_ALLOW_LIVE=yes are set.

API

Public: GET / page · /api/health · /api/status · /api/portfolio · /api/performance · /api/orders · /api/ranking · /api/plan (dry-run of what a rebalance would do now). Private (token): POST /api/rebalance?dry=0 (force=1 overrides only the window / once-a-month guards) · POST /api/refresh (rebuild the ranking now).

Notes

Educational paper-trading demonstration — not investment advice. The page shows derived rankings only, not raw Sharadar data. Each stock is a third of the money: one stock halving costs ~17%. Taxes and slippage beyond 10 bps are not modelled.