Table of Contents
Scott Welsh sits down to talk systematic trading—why he puts 99% of his money in algos, why breakouts and trend following still work, and how he avoids tinkering once a system is live. The conversation covers his journey from classic trading books to building simple, durable robots that run end-to-end without human interference, plus what he’s learned competing in trading championships and focusing on high-momentum instruments like GBP/JPY.
In this piece, you’ll learn Scott’s core philosophy—trend following via clean breakouts—along with practical rules on re-optimization (why he rarely does it), the power of time-based exits in FX, how to size risk for both longevity and contests, and the mindset required to actually let algos work. If you’re a beginner wanting a clear, copy-and-paste framework for simple, high-conviction systems, this breakdown gives you a blueprint you can adapt without getting lost in complexity.
Scott Welsh Playbook & Strategy: How He Actually Trades
Core Philosophy: Simple Trend Systems, Fully Automated
Scott builds robots to catch clean momentum and then gets out of the way. The edge isn’t prediction; it’s a ruleset that buys strength, cuts losers quickly, and survives the dull months. The goal is durability first, profits second.
- Trade mechanical systems only; no discretionary overrides once live.
- Favor trend-following breakouts on instruments that actually trend.
- Keep inputs minimal (price + one volatility measure) to avoid curve-fit.
- Judge systems by multi-year robustness, not last month’s equity curve.
- Maintain a written “no-tinkering while live” rule for at least 90 days.
Instruments & Timeframes: Hunt Movers, Not Everything
He concentrates on pairs and markets that trend hard, so breakouts matter. On FX, JPY crosses—especially GBP/JPY—are his workhorses; for indices/futures, he uses the same logic on liquid contracts.
- Primary FX universe: GBP/JPY first, EUR/JPY as a backup; avoid chronic mean-reverters.
- Add equity index futures (e.g., ES/NQ) or metals only if the same breakout rules test out-of-sample.
- Core timeframes: 30–6 minutes for FX; 60-minute or daily for futures.
- One symbol per robot; scale the portfolio by adding robots, not by mixing logic inside one bot.
- Require at least 10 years of bar data (or full available history) before trusting the symbol/timeframe combo.
Entry Logic: Volatility-Adjusted Breakouts
Entries are designed to be obvious to a computer and hard to second-guess. He prefers channel/band breaks with a volatility filter so you don’t buy every wiggle.
- Define a breakout as price closing above the upper band/channel by ≥ 1.0×ATR(20) filter (or Donchian/Keltner/Bollinger boundary + ATR buffer).
- Trade only during liquid sessions (e.g., London and early NY for FX); skip illiquid hours.
- Minimum distance from the 200-period moving average: enter only when price is on the trend side of the MA to avoid counter-trend chops.
- Limit one fresh entry per direction per symbol until a stop/exit occurs.
- If slippage > half the average stop size over 20 trades, auto-disable the entry for that session.
Stop Loss & Exits: Let Winners Breathe, Kill Quickly
He uses fixed risk with time-based and structure-based exits to keep robots moving. Time exists to prevent “forever holds” when trends stall.
- Initial stop = 1.5–2.0×ATR(20) from entry; place immediately.
- Trail only on new channel highs/lows or a 2×ATR chandelier; never on every bar.
- Time-based exit: close after 48–72 hours on intraday systems (or 10 bars on daily) if price fails to extend ≥ 0.5R.
- Hard exit on the opposite channel break or close back inside bands two bars in a row.
- No manual profit-taking; exits are 100% rule-driven.
Position Sizing: Conservative by Default, Aggressive by Design
Sizing is the difference between a cute backtest and a survivable strategy. Scott optimizes for staying power, then optionally layers risk for contests or small accounts.
- Risk per trade: 0.25%–0.5% of account on standard portfolios; up to 1.0% only on isolated, high-quality systems.
- Cap symbol exposure: ≤ 1 open position per symbol per direction.
- Portfolio heat: total open risk ≤ 2% (standard) or 4% (aggressive).
- Reduce size by 50% after a 1.5× average drawdown stretch; restore only after new equity highs.
- Use round-turn cost + conservative slippage in lot size math; never size off theoretical fills.
Portfolio Construction: Stack Independent Robots
He stacks multiple uncorrelated robots instead of one “perfect” system. Each robot is narrow (one instrument + one logic), which makes performance easier to diagnose.
- Build 3–6 robots across different symbols and/or timeframes (e.g., GBP/JPY 60m breakout, EUR/JPY 30m breakout, ES 60m Keltner, NQ daily Keltner).
- Correlation rule: don’t run two robots that trigger together > 60% of the time.
- Replace, don’t repair: if a robot degrades across multiple regimes, retire it and backfill with a fresh, simple variant.
- Rebalance monthly by target % risk per robot, not by cash.
- Keep a “bench” of validated robots ready; promote only after paper-trading for 30 days.
Optimization & Maintenance: Minimal Touch, Scheduled Only
Tweaking kills edges. He schedules checks and only makes changes when clear, multi-metric degradation shows up.
- Review robots monthly for: trade count, win rate, average win/loss, MAR, and slippage vs. baseline.
- If two of those metrics breach 2 standard deviations for 3 consecutive months, move the robot to quarantine (paper-trade) and stop live trading.
- Do parameter reviews quarterly; change at most one parameter per review cycle.
- Keep inputs coarse (e.g., 20/40/60) rather than fine-grained (e.g., 17/19/23).
- Any change requires a full re-validation on unseen data and a 30-day live-sim burn-in.
Execution & Infrastructure: Set It Up So You Can Ignore It
Automation only works if the pipes don’t leak. He standardizes the tech so that trades fire reliably without supervision.
- Run robots on a VPS close to your broker; require 99.9% uptime and auto-restart on failure.
- Log every order event with timestamp, price, spread, and latency; rotate logs weekly to cloud storage.
- If the platform disconnects> 60 seconds during liquid hours, halt entries for that session.
- Use server-side stops when supported; if not, sync hard stops every tick.
- Maintain a sandbox account for updates; never deploy untested code mid-week.
Session Filters & News: Avoid the Chop, Dodge the Landmines
Breakouts fail most when liquidity vanishes or spreads explode. Session rules and basic news awareness protect the edge.
- Enable entries only from 06:00–12:00 London time and 08:00–11:00 New York time for FX intraday robots.
- Block entries 15 minutes before and after tier-1 releases that historically blow out spreads (rate decisions, NFP, CPI).
- Spread guard: cancel any pending entry if live spread > 1.5× its 20-day median.
- Skip Sunday open and the last 2 hours of Friday for new entries.
- Holiday mode: reduce risk by 50% during known thin weeks.
Data Hygiene & Reality Checks: Trust the Boring Stuff
Good data keeps simple systems honest. He treats data work as part of the edge, not an afterthought.
- Use bid-based backtests for FX; add a realistic spread model per session.
- Require at least 300 trades in historical tests for intraday robots or 100 trades for daily.
- Run walk-forward on 3+ distinct regimes (calm, volatile, mixed) and demand positive expectancy in each.
- Never optimize to a profit target; optimize to a stable win/loss distribution and controlled drawdown.
- Keep an “ugly chart” dashboard that shows losers prominently, so you don’t fall in love with winners.
Behavior & Process: Be the Custodian, Not the Hero
The final edge is discipline. Scott’s approach wins because he refuses to intervene once rules are set.
- Daily routine: check health dashboard, not charts; confirm VPS, broker connection, and overnight fills.
- Weekly routine: export P&L by robot and review variance vs. expected; take no action unless thresholds are breached.
- Monthly routine: portfolio heat audit, risk rebalance, and bench review.
- Personal rule: no discretionary trades in the same account as robots.
- If you feel the itch to override, reduce risk next month instead—never mid-trade.
A Starter Robot You Can Replicate
Here’s a compact template that mirrors his breakout style without complexity. It’s designed for trenders and emphasizes survival over perfection.
- Market: GBP/JPY, 60-minute bars.
- Entry long: close > upper Keltner (20, 2.0) AND distance from 200-EMA > 0.25×ATR(20).
- Entry short: close < lower Keltner with the same filters.
- Stop: 1.8×ATR(20); trail on new 20-bar channel highs/lows after trade reaches +1R.
- Exit: time stops at 48 bars if unrealized < +0.5R; otherwise, exit on opposite band cross.
- Risk: 0.4% per trade; max one position per direction; portfolio heat cap 2%.
- Sessions: entries allowed 06:00–12:00 London, 08:00–11:00 New York; flat new entries late Friday.
- Kill-switch: disable after 3 consecutive -1R losses within 48 hours; re-enable next session cycle.
Trade the breakout, not your feelings: rules before opinions.
Scott Welsh keeps it simple: strength begets strength, so his systems buy strength and ignore opinions. He builds entry rules that trigger on clean breakouts and then forbids himself from second-guessing them. That discipline turns nerve-wracking decisions into routine executions, which is exactly how he avoids the “I knew it” trap after the fact. In his world, the edge is following the plan—every time—not forecasting the next candle.
To make that stick, Scott Welsh strips emotion from the workflow: the robot fires, the stop is set, and the outcome is just data. When volatility expands, the rules already account for wider ranges, so he doesn’t have to “feel” scared or brave. If price fails to follow through, the system cuts it; if it runs, the system rides it. The point isn’t to be right today—it’s to be relentlessly consistent over hundreds of trades.
Risk is small per trade, cap portfolio heat to survive
Scott Welsh treats position size as the backbone of his strategy. He risks a small, fixed percentage per trade, so a single loser can’t dent the account. That keeps emotions in check and turns each signal into just another attempt rather than a make-or-break event. He sizes from the stop distance, not a hunch, so volatility automatically translates to smaller positions on wild days.
Beyond single trades, Scott Welsh caps total portfolio heat so multiple open positions can’t gang up during a streak. If drawdown expands beyond his historical norm, he methodically cuts risk until equity stabilizes, then restores size only after new highs. He avoids doubling down to “win it back,” preferring steady compounding that survives long flat patches. When markets are favorable, he lets the number of independent systems do the heavy lifting instead of cranking risk on one setup. The result is a smoother equity curve that keeps him in the game long enough for the edge to matter.
Stack simple robots across symbols, timeframes, and regimes.s
Scott Welsh spreads his edge by running multiple narrow, rule-based robots instead of one “do-everything” monster. Each robot has a single job—one symbol, one timeframe, one breakout logic—so it’s easy to test, monitor, and replace. By mixing GBP/JPY with other trenders and pairing 30–60 minute systems with daily ones, Scott lets different market moods hand off the baton. When one lane chops, another can trend, and the portfolio keeps moving.
Scott Welsh also treats correlation as a risk setting, not an afterthought. If two robots trigger together too often, he parks one or shifts the timeframe to reduce overlap. He adds capacity by adding new, simple robots that pass robustness checks instead of grafting more filters onto existing code. That modularity makes it painless to retire underperformers and promote fresh candidates without reinventing the stack. The payoff is a steadier equity curve powered by many small engines rather than a single fragile one.
Use ATR stops and time exits to avoid churn.
Scott Welsh prefers stops that flex with the market, not arbitrary pip counts. He anchors risk to ATR so the initial stop lives outside normal noise yet still kills losers fast. As volatility expands, the stop naturally breathes wider; when things quiet down, it tightens itself. That keeps him from getting clipped to death in choppy stretches while preserving the ability to bail when the breakout clearly fails. Time exists is Scott Welsh’s antidote to dead money. If a trade hasn’t produced meaningful progress within a set number of bars or hours, he closes it—even if the price hasn’t hit the stop. This rule frees capital, reduces frustration, and keeps the robots focused on fresh momentum. Combined with occasional structure-based trails, the approach captures the meat of the move without marrying stale positions. The result is fewer whipsaw re-entries and a portfolio that turns over with purpose rather than hope.
Focus on mechanics over prediction: test, deploy, don’t tinker
Scott Welsh doesn’t try to outguess the market; he out-processes it. He defines every step—entry, stop, exit, size—so the robot behaves the same on Monday as it does on Friday. That mechanical certainty beats clever forecasts because it compounds small, repeatable edges without drama.
Once a ruleset passes his tests, Scott Welsh ships it and leaves it alone. He evaluates by metrics over many trades, not by how the last signal felt. If performance drifts, he quarantines and replaces rather than “improving” a live system midstream. The goal isn’t perfect trades—it’s perfect execution of a good plan, over and over.
Scott Welsh’s message lands with refreshing clarity: build simple breakout systems, automate everything, and let the math—not your mood—drive the account. He keeps 99%+ of his capital in algorithms that enter on strength and exit by rule, favoring trend-following logic over forecasts and steering clear of any human “management” once the trade is on. That bias toward mechanical execution comes from years of seeing clean breakouts and big trends do the heavy lifting while discretionary tweaks do the damage.
The operational edge is surprisingly plain: volatility-aware stops, time-based exits, and enough patience to let a robust rule set prove itself across regimes. Scott highlights how a 20-day breakout paired with an eight-day time exit and reasonable stops/targets can work on higher timeframes, and how buying “overbought” within a trend often beats the crowd’s instincts—as long as you don’t cut winners short or babysit losers. He reviews systems on a schedule, not on feelings, accepts that trend followers will have losing years, and only makes deliberate, infrequent changes after multi-month evaluation.
Risk is treated like plumbing: sized from stops, capped at the portfolio level, and never allowed to balloon just because a setup “feels right.” He’d rather stack independent, narrow robots—one symbol, one timeframe, one job—than chase a single perfect system, and he’s blunt that most traders struggle because they don’t know their stats or worst-case scenarios before they click. In short, Scott Welsh’s playbook is rules first, predictions last: trade breakouts that can run, kill trades that stall, and let a disciplined portfolio of simple algos compound without your interference.
