A reference for building a Jane-Street-style systematic trading operation at retail scale. Focused on programmatic execution of quantitative strategies, not discretionary trading. Written for the operator who already programs and now wants to convert that into a private wealth engine.
The honest filter: at retail capital, most "edge" strategies real shops run (latency arb, true HFT market making, capacity-heavy stat arb on the full Russell 3000) are unavailable. The remaining surface is narrower but real. Mathematical strength matters more on the research side (signal discovery, risk modeling) than the execution side.
| Strategy | Core Math | Min Capital | Realistic Sharpe (retail) | Capacity | PDT/Reg Issues |
|---|---|---|---|---|---|
| Statistical arbitrage | Cointegration (Engle-Granger, Johansen), PCA, OU mean reversion, Kalman filtering | $25k+ (equities) or $5k+ (crypto) | 0.8-1.5 | Medium | PDT bites under $25k on margin equities |
| Market making | Avellaneda-Stoikov, inventory control, Poisson order flow models | $25k (crypto) / impractical (US equities) | 1.0-2.5 in crypto | Low at retail | None on crypto/futures; impossible on equities without rebates |
| Mean reversion (single-name / ETF) | Z-score, half-life of mean reversion (Ornstein-Uhlenbeck), Hurst exponent | $5k+ | 0.5-1.2 | Medium | PDT if margin equities |
| Momentum / trend following | Time-series momentum (Moskowitz et al.), breakout filters, channel models, Kaufman AMA | $5k+ (futures preferred) | 0.6-1.0 (lumpy) | High | None on futures; PDT on intraday equities |
| Options vol arbitrage | Black-Scholes greeks, vol surface modeling, variance swaps, GARCH | $25k+ (defined risk); $100k+ (undefined) | 0.7-1.5 | Medium | Level 3+ options approval, margin requirements |
| Pairs trading | Same toolkit as stat arb but bivariate; spread half-life, dynamic hedge ratios | $10k+ | 0.6-1.2 | Low | PDT same as equities |
| Factor investing | Cross-sectional regression, Fama-French + custom factors, portfolio optimization (mean-variance, risk-parity) | $25k+ (50+ names for diversification) | 0.4-0.8 | Very high | None (rebalance monthly, no PDT issue) |
| Event-driven | NLP on filings/news, M&A spread arb, earnings drift | $25k+ | 0.5-1.0 | Medium | PDT possible; options approval helpful |
Feasibility by capital tier:
The big trap: at retail capital, the strategies with the highest theoretical Sharpe (HFT market making, latency arb) are also the ones you cannot run. The strategies you can run (medium-frequency mean reversion, trend, factors) are the ones the literature has been chewing on for 25 years and where edge is thin. Plan around capacity-tolerant, low-frequency edge with strong risk management, not against high-Sharpe fantasy.
| Asset | Min Account | Leverage | Tax Treatment | API Coverage |
|---|---|---|---|---|
| US Equities | $0 (cash) / $2k margin / $25k for PDT-margin | 2x intraday, 4x day-trading | Short-term cap gains (≤1yr ordinary income); wash sale applies | Public, Alpaca, IBKR, Schwab, Tradier |
| US Options | $2k+ (Level 2); $25k+ (Level 3/4) | Defined by structure | Same as equities; LEAPS get LTCG; complex spreads tricky | Tradier (best), TastyTrade, IBKR, Public |
| Futures (CME) | $500-2000 per micro contract | 10-50x typical | Section 1256: 60% LTCG / 40% STCG regardless of holding period. No wash sale. | IBKR, NinjaTrader, Tradovate, AMP |
| Crypto spot | $0 | 1x (3x on Kraken Pro) | Property; FIFO; no wash sale (yet, IRS proposed) | Coinbase Advanced, Kraken, Gemini |
| Crypto perps | $0 (offshore) | 5-100x | Murky; treat as ordinary income for self-custody DEX | Hyperliquid, dYdX, Bybit (non-US), Binance.US (limited) |
| FX | $100 (offshore); $25k (Oanda US) | 50x retail US | Section 988 ordinary by default; can elect 1256 | Oanda, IG, IBKR |
| Treasuries / Bonds | Varies | None | Federal income; state-exempt for Treasuries | TreasuryDirect.gov, Public, IBKR |
Two regulatory facts to internalize:
The PDT change (2026): The SEC approved elimination of the $25k pattern day trader rule effective June 4, 2026, replacing it with a risk-based margin framework: no minimum balance for cash-equivalent traders, $2k or maintenance margin (whichever greater) for leveraged. This materially expands what's runnable under $25k. (Britannica explainer, NerdWallet)
PUT to amend orders without cancel/replace (Feb 2026 update).Mercury is banking, not brokerage. For trading purposes it enables:
How this composes with trading: keep operating capital in Mercury Treasury (yielding) → programmatically ACH to broker when buying power needed → ACH back to Mercury when flat. The yield differential between idle broker cash and Mercury Treasury on $100k+ is meaningful (typically 100-200 bps).
| Broker | Free Tier | Assets | Latency | Python SDK | Notes |
|---|---|---|---|---|---|
| Alpaca | Yes (commission-free) | US equities, options, crypto | 50-200ms | alpaca-py (official) |
Best paper/live parity. Most popular retail algo broker. |
| Interactive Brokers | $0 commission tier; data add-ons cost | Global multi-asset (everything) | 10-50ms | ib_async (successor to ib_insync after creator's 2024 death) |
Gold standard. TWS/Gateway must run. |
| Tradier | $10/mo brokerage; sandbox free | Equities, options | 100-300ms | Community | Best-priced options API; commission $0.35/contract or flat $10/mo unlimited. |
| TastyTrade | Free with account | Equities, options, futures, crypto | 100ms+ | Community + official | Designed by options traders for options traders; great complex order types. |
| TradeStation | Free with account | Equities, options, futures, crypto | 100ms+ | REST/WebSocket | Solid for futures + equities mix. |
| Schwab | Free; manual approval 1-3d | Equities, options | Variable | Community (schwab-py) | Inherited TD Ameritrade API. Required register as Individual Developer. |
| Robinhood | Unofficial only | Equities, options, crypto | High | robin_stocks | TOS forbids automation. Avoid for serious work. |
| Exchange | Free Tier | Assets | Latency | SDK | |
|---|---|---|---|---|---|
| Coinbase Advanced Trade | Yes | Spot, some futures | ~100ms | Official Python | |
| Kraken | Yes | Spot, futures, margin (3x) | 50-150ms | krakenex |
|
| Binance.US | Yes | Spot | 100ms | Official + python-binance |
|
| Hyperliquid | Yes | Perps + spot DEX | 20-50ms (custom L1) | Official Python + Rust | |
| dYdX v4 | Yes | Perps DEX | 50-100ms | Official | |
| Bybit / OKX / BitMEX | Yes | Perps | 50-150ms | Official | Non-US only |
For market making in crypto specifically, Hummingbot ships connectors for most of the above. CCXT (ccxt/ccxt, ~33k stars) is the universal abstraction layer if you need to swap exchanges.
Ranked, with current state as of mid-2026:
| # | Framework | Repo | Lang | Last Significant Release | License | Best For |
|---|---|---|---|---|---|---|
| 1 | NautilusTrader | nautechsystems/nautilus_trader | Rust core + Python | Active monthly | LGPL-3.0 | Production-grade event-driven; closest to institutional. |
| 2 | VectorBT (PRO) | polakowo/vectorbt | Python (Numba) | OSS less active; PRO active | Apache 2.0 (OSS) | Vectorized parameter sweeps. Fastest for research grids. |
| 3 | QuantConnect LEAN | QuantConnect/Lean | C# (with Python wrapper) | Daily commits | Apache 2.0 | Cloud or self-host; multi-asset; live integration baked in. |
| 4 | Zipline-Reloaded | stefan-jansen/zipline-reloaded | Python | Active maintenance | Apache 2.0 | Daily-bar equities research; pairs with pyfolio-reloaded and Stefan Jansen's ML for Trading. |
| 5 | Backtrader | mementum/backtrader | Python | Original author inactive since ~2019; community PRs | GPL-3.0 | Most documentation/tutorials online. Easy to learn, slow on big grids. |
| 6 | Freqtrade | freqtrade/freqtrade | Python | Very active (25k+ stars) | GPL-3.0 | Crypto-only. FreqAI for ML strategies. |
| 7 | Hummingbot | hummingbot/hummingbot | Python | Very active | Apache 2.0 | Crypto market making + DEX arb. |
| 8 | Jesse | jesse-ai/jesse | Python | Active | MIT | Crypto; clean API; honest backtests. |
| 9 | bt | pmorissette/bt | Python | Maintenance mode | MIT | Portfolio rebalancing strategies on top of ffn. |
| 10 | PyAlgoTrade | gbeced/pyalgotrade | Python | Inactive (~2020 last) | Apache 2.0 | Historical interest only. |
| 11 | QSTrader | mhallsmoore/qstrader | Python | Light maintenance | MIT | Institutional-flavored from QuantStart blog. |
| 12 | vnpy | vnpy/vnpy | Python + C++ | Active (China-focused but international support) | MIT | Multi-broker live trading platform. |
Recommended workflow: VectorBT for parameter sweeps and signal discovery → NautilusTrader for execution-realistic backtest → live deploy on NautilusTrader or broker-specific code. Avoid the temptation to live-trade directly from a vectorized research notebook.
| Source | URL | Cost | Asset Coverage | Granularity |
|---|---|---|---|---|
| Polygon.io | $29-$199/mo retail | US equities/options/FX/crypto | Trade/quote down to 1s; tick on higher tiers | |
| Databento | ~$100-500/mo metered | US equities, options, futures (60+ venues) | Nanosecond tick + L2 from direct exchange feeds | |
| Tiingo | Free tier; $30/mo | US equities, crypto, FX, news | Daily + intraday | |
| Alpha Vantage | Free; $50/mo+ | Multi-asset | Daily, intraday | |
| Norgate Data | $30-90/mo | US/AU/CA equities + futures continuous contracts | Daily; survivorship-bias-free | |
| QuantConnect data | Bundled with LEAN cloud | Multi-asset | Tick/minute/hour/daily | |
| Kibot | One-time pay | US equities, futures | Tick + minute | |
yfinance |
GitHub | Free (unofficial Yahoo scrape) | Global equities, ETFs | Daily mostly; intraday unreliable, survivorship-biased, prone to break |
yfinance is fine for prototyping and research. It is unfit for production-grade backtests. Use Norgate or Polygon for serious work. Databento if you need L2 / tick.
Curated, with stars and current status (May 2026 snapshot):
| Repo | URL | Stars | Why It Matters |
|---|---|---|---|
| QuantConnect/Lean | https://github.com/QuantConnect/Lean | ~10k | Institution-quality event-driven engine. Read the algorithm framework. |
| nautechsystems/nautilus_trader | https://github.com/nautechsystems/nautilus_trader | ~5k | Rust core; modern architecture you'd actually deploy. |
| stefan-jansen/machine-learning-for-trading | https://github.com/stefan-jansen/machine-learning-for-trading | ~14k | Companion notebooks to the book; best ML-for-trading walkthrough on GH. |
| hudson-and-thames/mlfinlab | https://github.com/hudson-and-thames/mlfinlab | ~4k | Implementations of de Prado's Advances in FinML. (Note: now partially closed-source; older OSS version still useful.) |
| OpenBB-finance/OpenBB | https://github.com/OpenBB-finance/OpenBB | ~37k | Free Bloomberg-alternative for research; broad data integrations. |
| ccxt/ccxt | https://github.com/ccxt/ccxt | ~33k | Universal crypto exchange interface. The standard. |
| freqtrade/freqtrade | https://github.com/freqtrade/freqtrade | ~28k | Battle-tested crypto bot platform with FreqAI. |
| hummingbot/hummingbot | https://github.com/hummingbot/hummingbot | ~8k | Production crypto market-making framework. |
| ranaroussi/quantstats | https://github.com/ranaroussi/quantstats | ~5k | Tearsheet generation. Steal the metric implementations. |
| robcarver17/pysystemtrade | https://github.com/robcarver17/pysystemtrade | ~3k | Companion to Systematic Trading by Rob Carver. Full futures portfolio system. |
| polakowo/vectorbt | https://github.com/polakowo/vectorbt | ~5k | Numba-accelerated vectorized backtester. |
| jesse-ai/jesse | https://github.com/jesse-ai/jesse | ~6k | Clean crypto-focused framework. Good API design study. |
| ranaroussi/yfinance | https://github.com/ranaroussi/yfinance | ~14k | Useful for prototyping. Know its limits. |
| microsoft/qlib | https://github.com/microsoft/qlib | ~16k | Microsoft Research's AI-for-quant platform. Strong reinforcement-learning integration. |
| google/tf-quant-finance | https://github.com/google/tf-quant-finance | ~5k | Derivative pricing in TF. Reference implementations for Heston, SABR, etc. |
| jpmorganchase/python-training | https://github.com/jpmorganchase/python-training | ~4k | JPM's internal Python training materials. Reveals how a bank does it. |
| AI4Finance-Foundation/FinRL | https://github.com/AI4Finance-Foundation/FinRL | ~10k | Deep RL for trading. Academic but well-maintained. |
| wilsonfreitas/awesome-quant | https://github.com/wilsonfreitas/awesome-quant | ~20k | Curated list-of-lists. Use as discovery surface. |
| wangzhe3224/awesome-systematic-trading | https://github.com/wangzhe3224/awesome-systematic-trading | ~3k | Better filtered than awesome-quant; focused on real systems. |
| ib-api-reloaded/ib_async | https://github.com/ib-api-reloaded/ib_async | ~3k | The IBKR Python interface. Read the source for asyncio patterns. |
| crflynn/stochastic | https://github.com/crflynn/stochastic | ~0.5k | Stochastic process simulators (OU, Heston, fBM). Useful for synthetic-data testing. |
| cvxpy/cvxpy | https://github.com/cvxpy/cvxpy | ~6k | Convex optimization. Backbone of portfolio construction. |
| lballabio/QuantLib | https://github.com/lballabio/QuantLib | ~6k | C++ derivatives pricing library. Industry standard. |
| AI4Finance-Foundation/FinRL-Meta | https://github.com/AI4Finance-Foundation/FinRL-Meta | ~1k | Standardized datasets/environments for RL trading research. |
Latency-sensitive (HFT): not realistic for retail. Co-located NY4 servers ($2000-10000/mo), kernel-bypass networking, FPGA NICs. Don't bother.
Mid-frequency (sub-second to minute bars): the current i7-265K + 30GB desktop is overkill for execution. The dGPU is irrelevant for non-ML strategies.
Overnight / swing / daily bars: cheapest VPS imaginable suffices. A $5 Hetzner CX11 will run a daily-bar strategy.
Research and backtesting:
| Resource | Need |
|---|---|
| Cores | More is better. Parameter sweeps embarrassingly parallel. 16+ cores changes life. |
| RAM | 32GB minimum. 64-128GB unlocks full-history L1/L2 tick loaded in memory. |
| Storage | NVMe SSD essential. Tick data for 1 year of US equities = 500GB-2TB depending on schema. Plan for STARGATE's 916GB drive to host. |
| GPU | RX 6800 XT is useful for ML training but not inference. PyTorch ROCm is functional on Arch. For most quant ML (gradient boosting, linear models) GPU is irrelevant; XGBoost/LightGBM on CPU is faster than naive GPU. |
| Network | Wired ethernet, low-jitter. Wifi adds 5-30ms jitter that ruins fill quality on anything sub-minute. Use STARGATE wired for live execution. |
Recommended split for this setup:
| Provider | URL | Datacenters | Entry Price | Use Case |
|---|---|---|---|---|
| Beeks Financial Cloud | https://www.beeksgroup.com/ | NY4, LD4, CME Aurora, TY3, +14 more | $39/mo VPS, $315-$3270/mo dedicated | Best NY4 proximity, broker peering. Latency 0.8ms to LD4/NY4 hubs. |
| QuantVPS | https://www.quantvps.com/ | NY4 (Equinix) | $42/mo (Lite) → $129+ | NY4 colo at retail prices. Solid for futures via Rithmic/CQG. |
| Speedy Trading Servers | https://www.speedytradingservers.com/ | Chicago (CME), NY4, LD4 | $35-150/mo | Futures-trader-focused. |
| TradingFXVPS | https://www.tradingfxvps.com/ | Global (NY, London, Tokyo, Frankfurt) | $25-90/mo | MetaTrader/FX focus. |
| ForexVPS.net | https://www.forexvps.net/ | NY, LD, FR, TY, HK | $35/mo | FX VPS. |
| AWS us-east-1 | https://aws.amazon.com/ec2/ | N. Virginia (close to NY4 but not in it) | $4-200+/mo | When you want full IaaS; ~3-5ms to NY4 from us-east-1d. |
| Hetzner | https://www.hetzner.com/cloud | Falkenstein, Helsinki, Ashburn (US) | EUR 4.59/mo | Cheapest sane option for non-latency-sensitive overnight strategies. Ashburn for US ~10-15ms to NYSE. |
| Vultr | https://www.vultr.com/ | NJ (close to NY4), Chicago | $5-40/mo | Cheap and ubiquitous. |
| Linode (Akamai) | https://www.linode.com/ | Newark (close to NY4) | $5+/mo | Same tier as Vultr. |
Recommendation by stage:
| Stage | Capital | Focus | Structure | Action |
|---|---|---|---|---|
| Seed | $5k | Paper trade everything. Live-trade one small futures or crypto strategy to feel real fill quality. | Sole proprietor; personal account. | Open public.com + Alpaca paper + Coinbase Advanced. |
| Live | $25k | 2-3 uncorrelated strategies, post-PDT-elimination. Begin tracking metrics rigorously. | Sole prop; consider TTS election. | Add IBKR for futures (better margin on /MES). |
| Scaling | $100k | Multi-strategy with formal allocation. Add Trader Tax Status filing. Move execution to STARGATE + Mercury treasury sweep. | TTS election; possibly LLC if filing 475(f). | Move to NY4 VPS if any sub-minute strategy. |
| Pro | $500k+ | Diversified across 5+ strategies and 3+ asset classes. Real options book. Possibly run for outside capital under exemption. | LLC; consider S-corp for self-employment tax. | Consult quant tax accountant (Green Trader Tax). |
Prop firm shortcut: instead of personal capital you can lease buying power from a prop firm. Realistic ones:
| Firm | Asset | Eval Cost | Profit Split | Notes |
|---|---|---|---|---|
| FTMO | FX, indices, metals | $99-$999 | 80-90% | Largest, OANDA-owned (Dec 2025). Strong rep. |
| Topstep | CME futures (ES, NQ, CL, GC, etc.) | $49-$149/mo + $149 activation | First $10k 100%, then 90% | Longest-running (since 2012); paid out >$1.4B. |
| The5ers | FX, indices | $39-$1095 | 50-100% (instant funding tier) | Older firm with instant-funding paths. |
| Apex Trader Funding | Futures | $25-$300/mo | 90% | Futures-only; very large account ceiling. |
| FundedNext | FX | $59-$999 | 80-90% | Successor space to MFFU (shut down by CFTC). Now dominant in FX. |
Prop firms are useful as leverage on a tested edge. They are not useful as a substitute for an edge. Eval rules (max drawdown, daily loss limits, news restrictions) bias toward conservative, low-risk strategies, sometimes punishing real edge.
f of equity per trade. Simple, robust. f = 0.01-0.02 typical.position = target_vol * equity / (px * realized_vol). Rob Carver's bread-and-butter.f* = (bp - q) / b. The textbook answer. In practice, use half-Kelly or less because the inputs are estimated and overestimation is fatal.cvxpy or riskfolio-lib.X% drawdown. Mandatory.N consecutive losers or M% drawdown over K days.P% of equity.For STARGATE deployment:
| Layer | Tool | Purpose |
|---|---|---|
| Tick storage | TimescaleDB (Postgres extension) or DuckDB on Parquet | Historical research data. |
| Live PnL | Grafana + Prometheus | Real-time dashboard. |
| Alerts | Alertmanager → email (your ~/.local/bin/email script) + ntfy push |
Drawdown alarms, position breaches. |
| Logs | Loki or just journald + Promtail | Strategy decision audit trail. |
| Order management | NautilusTrader's built-in OMS, or roll your own with ib_async + Redis |
Reconcile broker fills with internal state. |
| Process supervision | systemd | Auto-restart on crash. |
q-fin.ST (statistical finance) and q-fin.TR (trading and microstructure).Caveat: reference, not advice. At $50k+ engage a quant-specialized CPA — Green Trader Tax is the most-cited.
To qualify, you typically need to demonstrate:
Benefits: deduct trading expenses (data, computers, home office, education) on Schedule C; deduct margin interest as business expense.
Once TTS-qualified, electing 475(f) on securities:
Trap: electing 475(f) on commodities/futures converts Section 1256 60/40 treatment to ordinary income, usually worse. Elect on securities only.
Deadline: for individuals, attach the 475 election statement to the prior year's tax return or extension by April 15. So for tax year 2026 elections, the statement was due April 15, 2026. After making the election, file Form 3115 with that year's return.
(Green Trader Tax: Section 475 MTM, IRS Topic 429)
A single-member LLC defaults to disregarded entity (tax-identical to sole prop), so the gain from an LLC is procedural, not tax-magic:
At ~$50k+ profits, structure a trading LLC + S-corp management company combo. The S-corp pays you a reasonable salary (enables ~$70k/yr solo 401k contribution, HSA, etc.) and the trading LLC retains the rest as pass-through ordinary. This is the Green Trader Tax "Pro Trader Entity" structure.
Beyond the algo book, hold a tail in things that don't depend on any single counterparty, market, or political consensus:
| Asset | Vehicle | Counterparty Risk | Yield |
|---|---|---|---|
| I-Bonds | TreasuryDirect.gov | US Treasury only | CPI-linked; $10k/yr cap per SSN |
| T-Bills / Notes | TreasuryDirect or Public/IBKR ladder | US Treasury | Whatever the curve gives |
| Physical gold / silver | Coins (Eagles, Maples); allocated storage (Goldmoney, BullionVault) | Self-custody = none | None; storage cost |
| Bitcoin (cold storage) | Hardware wallet (Coldcard, Trezor, Ledger); multi-sig (Sparrow, Specter) | None at protocol level | None; protocol scarcity |
| Productive land | Direct ownership | State legitimacy | Crop income, timber, lease |
| Cash USD outside the banking system | Physical paper | Inflation + theft | Negative real |
| Cash in foreign currencies | Swiss or SG accounts in CHF/SGD | Foreign banking | Variable |
Principle: things that survive any one institution failing. Mercury Treasury yields well but is ultimately a sweep into MMFs at partner banks. A small physical-asset tail (~5-10% of net worth) is cheap insurance against the rest of the structure failing for non-market reasons.
TreasuryDirect doesn't have a true public API, but its data feeds and ACH automation let you ladder T-Bills via recurring purchases programmed against your Mercury account.
~/dev/trading/ repo. Git. Direnv. uv or poetry for Python isolation.(R_p - R_f) / σ_p; annualize multi-period by √(periods_per_year).f* = p/a - q/b where p = win prob, q = 1-p, b = win size, a = loss size. Use ≤ half.λ = -ln(2) / β where β is the AR(1) coefficient minus 1.H < 0.5 mean-reverting, H > 0.5 trending. Compute on log-returns via R/S analysis or DFA.√((1-γ)Φ⁻¹(1-1/N) + γΦ⁻¹(1-1/(Ne))) where N is number of trials, γ Euler-Mascheroni.Hardware: moirai (research) + STARGATE (execution)
OS: Arch Linux (both)
Languages: Python 3.12 (research/exec), Rust (perf-critical components later)
Frameworks: VectorBT (sweep) + NautilusTrader (execute)
Storage: TimescaleDB (live PnL, fills) + Parquet on disk (research data)
Broker: public.com (equities/options/bonds), IBKR (futures), Coinbase Advanced (crypto)
Banking: Mercury (operating + treasury sweep)
Monitoring: Grafana + Prometheus + Loki on STARGATE
Alerts: ntfy + ~/.local/bin/email
Supervision: systemd units per strategy
Tax: TraderFyles or TradeLog for wash-sale reconciliation; Green Trader Tax CPA
Backups: git for code; restic snapshots of TimescaleDB to Hetzner S3