Systematic Trading — Core Doctrine

Distilled lessons from the FX trading research + system build (May 2026), generalized to any systematic trading system (FX, equities, crypto, commodities).

These are the durable rules that survive any specific strategy / market / timeframe.

1. The 5 Universal Truths

1.1 The Sharpe Lie

A backtest Sharpe of 1.5 over 2 years has a 95% confidence interval of roughly 0.7–2.3 (Lo, 2002). A single backtest number means almost nothing.

Required mitigation:

• Compute Deflated Sharpe Ratio (Bailey & López de Prado, 2014) — deflates for multiple testing
• Compute Probabilistic Sharpe Ratio (PSR) — probability that true Sharpe > benchmark
• Use walk-forward validation, not in-sample backtest only
• Minimum 3 years of out-of-sample data before believing the number

1.2 Costs Are a Real Number — Don't Pretend Otherwise

Realistic retail trading costs include:

• Spread: 1-3 pips on majors (broker), 5-10 pips on exotics
• Slippage: 0.3-1.0 pips per side
• Commission: $30-60 per million notional
• Swap (rollover): 0.5-1.5% annualized retail markup on top of base rate diff
• Tax: variable; in NL, active trading triggers Box 1 reclassification (up to 49.5%)

A strategy that doesn't survive realistic costs in backtest definitely won't survive live.

1.3 Risk Layer > Alpha Layer

Most retail blow-ups are sizing failures, not signal failures.

Required components:

• Per-trade risk cap: 0.5-1% of equity, never higher
• Position sizing: fractional Kelly (half-Kelly maximum), never raw Kelly
• Correlation budget: cap effective positions at 2-3 (n_eff = n / (1 + (n-1)*ρ̄))
• Daily/weekly/monthly loss limits with auto kill switch
• Consecutive-loss circuit breaker (5 losses → halt)
• Equity-curve filter (stop when equity < N-day SMA)
• Pre-news blackout (avoid trading 30 min before high-impact events)
• Weekend flatten (Friday close all positions if strategy is short-horizon)

1.4 Black Swans Are Uncompensated

The major historical retail-FX blow-ups all share a pattern:

• 2015 SNB unpeg: -29% intraday on EUR/CHF → FXCM took $225M neg balances → Alpari bankrupt
• 2016 GBP flash: -6% in 40 seconds at Tokyo open
• 2020 COVID: -5% USD/JPY single day
• 2024 yen carry unwind: -7.4% USD/JPY in 3 sessions

Required defenses:

• EU-regulated broker (mandatory negative-balance protection since 2018)
• Hard stop-loss orders, not "mental stops"
• Position size such that even a 20% gap loss is recoverable
• Diversification across uncorrelated pairs (max n_eff ≤ 3)

1.5 Literature > Discovery

The DSR will crush any strategy you "discover" by searching parameters. But strategies that have been peer-reviewed and independently replicated have already absorbed the multiple-testing penalty.

Trust hierarchy:

1. Peer-reviewed in a top journal (JF, JFE, RFS) AND independently replicated AND post-publication out-of-sample
2. Peer-reviewed but not yet replicated → cautious implementation
3. Working paper from a credible group (NBER, BIS, Fed staff) → cautious implementation
4. Industry whitepaper (AQR, Bridgewater, Two Sigma) → useful directionally
5. Blog post / YouTube / Reddit → noise

2. The Workflow That Survives

2.1 Research → Backtest → Walk-Forward → Paper → Live (months apart)

• Research (1-2 weeks per strategy): read papers, define the signal
• Backtest (1 week): implement, run on 10+ years of data with realistic costs
• Walk-forward (1 week): rolling/anchored split, OOS performance only
• Paper trade (3-6 months minimum): real broker API, fake money, log every trade
• Live micro (3 months): 0.1× target size, real capital
• Scale up (3 months): 0.25× → 0.5× → 1.0× as DSR confirms edge

Total: 12-18 months before full-size live. Anyone selling you a shortcut is selling you survivorship bias.

2.2 The Backtest Discipline Checklist

Before believing any backtest result:

• No look-ahead bias (signal uses only data available at bar close)
• No survivorship bias (use the universe that existed historically)
• Realistic costs included (spread + slippage + commission + swap)
• Out-of-sample data > 3 years
• Walk-forward validation (not just in-sample)
• Deflated Sharpe Ratio computed
• PSR > 70% AND DSR > 50% before live consideration
• Cross-venue validation (your backtest data ≠ your broker's quotes)
• Crisis windows replayed (2008, 2015, 2020, 2022 minimum)
• Strategy assumptions match strategy reality (win rate, R-multiple, max DD)

2.3 The Journal Discipline

Every trade gets logged with:

• Strategy that generated it
• Regime at time of entry (quiet / normal / elevated / shock)
• Reasoning (why this signal, why this size)
• Pre-trade checklist result (all 20 gates passed?)
• Exit reason (signal flip / stop / target / kill switch / manual)
• PnL outcome + return %

Weekly LLM review of the journal looks for behavioral leaks — not strategy bugs.

3. The "Sell It" Trap

If you build a working trading system, do NOT:

1. Sell it as a "signal service" before you have 12+ months of verified live track record
2. Sell automated execution on customer accounts without MiFID II / equivalent licensing
3. Sign affiliate deals where the broker pays you when buyers lose money
4. Make any claim about future returns

What you CAN do:

1. Open-source the research / tool and monetize via consulting
2. Sell the methodology as a course / book / cohort program (clear disclosure)
3. Build a SaaS for other systematic traders (data, backtest, journal) — B2B PLG
4. Manage your own capital, build a track record, THEN consider a regulated fund structure

See knowledge/plg/ for the PLG monetization paths that don't blow up legally.

4. Specific to Retail FX (as of 2026)

• ESMA leverage cap: 30:1 majors, 20:1 cross majors, 10:1 commodities
• Negative balance protection: mandatory for EU-regulated brokers
• Best EU brokers for algo: OANDA (cleanest API), IBKR Ireland (multi-asset), Pepperstone
• AVOID: SVG / Vanuatu / Belize / Marshall Islands / Seychelles regulated brokers
• Dutch tax: active leveraged trading risks Box 3 → Box 1 reclassification (up to 49.5%)

5. Strategies with Known Edge (May 2026)

Confirmed working on 15-year real FX data after retail costs:

• Dollar Carry (Lustig-Roussanov-Verdelhan 2014) — Sharpe 0.23, PSR 83%
• Global Imbalance (Della Corte-Riddiough-Sarno 2016) — Sharpe 0.23, PSR 83%
• Carry + GI 50/50 blend — Sharpe 0.24, PSR 84%

Don't work at retail (tested + documented):

• Single-pair TSMOM
• Cross-sectional momentum (any lookback)
• FX Value / PPP reversion (on daily frequency)
• Vol-managed momentum
• Pure regime filtering

See fx-trading/06-real-data-findings.md and fx-trading/07-academic-strategies.md for full evidence.

6. Honest Performance Expectations

For a well-built retail systematic FX system:

• Realistic CAGR: 2-8% net
• Realistic Sharpe: 0.2-0.5 net
• Realistic max DD: 20-30%
• Time to break-even: 2-5 years
• Time to "meaningful supplementary income": account size > €250k AND 3 years live

If a vendor promises higher, they are lying, selling fraud, or describing institutional performance you can't replicate.

Sources

• Bailey & López de Prado (2014). The Deflated Sharpe Ratio.
• López de Prado (2018). Advances in Financial Machine Learning.
• Moskowitz, Ooi, Pedersen (2012). Time Series Momentum.
• Lustig, Roussanov, Verdelhan (2014). Countercyclical Currency Risk Premia.
• Della Corte, Riddiough, Sarno (2016). Currency Premia and Global Imbalances.
• Bailey, Borwein, López de Prado, Zhu (2014). Pseudo-Mathematics and Financial Charlatanism.
• ESMA Mandatory Broker Disclosures (2018-present).
• BIS Sterling Flash Event Report (Oct 2016).
• Van Tharp (1999). Trade Your Way to Financial Freedom.
• Ralph Vince (1990). Portfolio Management Formulas.