Table of Contents
Alejandro Perez sits down for a straight-talking interview on building and running algorithmic systems that survive real markets—not just backtests. From Hong Kong, he lays out why most traders misjudge sideways periods, how to think like a portfolio “owner” instead of a button-clicker, and why expectations for live performance should be lower than what the backtest shows. Perez has helped traders turn proven discretionary ideas into robust algos, and in this conversation, he explains how he manages them like employees—with clear KPIs, risk bands, and performance reviews.
In this piece, you’ll learn the core of Alejandro Perez’s approach: setting realistic return bands from backtests, modeling drawdowns for prop-firm rules, using market-regime filters (trend, volatility, noise) that actually align with your system’s logic, and avoiding over-optimization with clean in-sample/out-of-sample validation. You’ll also see how he times maintenance cycles, when he tightens or changes settings before hitting a max drawdown, and why “earn while you learn” can work—if you borrow a framework, not just signals. By the end, you’ll have a simple, beginner-friendly blueprint to run algos like a business and keep your strategy alive through the boring patches as well as the big months.
Alejandro Perez Playbook & Strategy: How He Actually Trades
Risk & Return Bands
Before anything else, Alejandro frames expectations: live results should ride inside realistic “return bands,” not your best backtest line. That mindset makes it easier to keep trading when the system hits normal drawdowns and prevents panicky tweaks that kill edge.
- Define an expected monthly return band from backtests and walk-forward (e.g., −8% to +12%); only intervene if equity exits that band.
- Size every algo so that a worst historical drawdown × 1.3 (safety multiplier) never breaches account rules.
- Cap daily loss at 1/20th of the worst backtest drawdown to avoid clustering losses.
- If equity hits −0.75× worst backtest drawdown, cut risk by 30% automatically; restore only after a new equity high.
Portfolio of Algos = “Employees”
Alejandro manages algos like a team: each has a job, KPI, and probation period. Underperformers don’t get endless chances; they get reviewed, resized, or replaced.
- Give every strategy a one-page job card: market, setup, entry/exit logic, risk, and KPI.
- KPI = CAR/MDD and win-rate × payoff tracked monthly; demote any algo below the 25th percentile of its history.
- 90-day probation for new algos with half risk; promote only after matching backtest efficiency within your return band.
- Keep overlap < 0.6 (pairwise correlation of P&L); if two algos move together, halve the smaller one.
Market-Regime Filters (Trend, Volatility, Noise)
His “simple first” regime approach: let a few clean filters decide when a strategy is most itself. Don’t add filters to avoid losses—add them to concentrate the edge.
- For trend systems, require price above 200-EMA and ADX(14) > 18; otherwise, cut size by 50%.
- For mean-reversion, require Bollinger width < 1.2× its 6-month median.
- Skip trades when spread/ATR > 4% or 1-minute noise ratio (true range/range) exceeds 0.6.
- If two regime flags disagree, trade but halve position; if three disagree, stand down.
From Idea → Code → Test (Keep It Boring)
Alejandro’s pipeline avoids curve-fitting by locking rules before results, then proving durability on data the system hasn’t “seen.” Keep the loop repeatable and dull.
- Freeze logic in plain English first; then code exactly that logic.
- In-sample: last 2–3 years; out-of-sample: prior 3–5 years; walk-forward: rolling 6-month windows.
- Ship only if out-of-sample CAR/MDD ≥ 70% of in-sample and max DD inflation ≤ 1.5×.
- Run a 10,000-trade Monte Carlo (resample by day) and size to 95th percentile drawdown.
Position Sizing & Leverage Discipline
He keeps sizing stupid-simple so mistakes are obvious and reversible. The goal: never let any single idea wreck the portfolio.
- Base size on ATR(14), so each trade risks 0.35–0.5% of equity at the stop.
- Cap per-asset margin at 25% and portfolio margin at 60%.
- Limit concurrent directional exposure (e.g., total USD-long risk ≤ 1.5%).
- If two losers occur within one hour across correlated algos, auto-reduce all open risk by 20%.
Settings, Not Tweaks (Guardrails for Parameters)
Alejandro prefers “settings ranges” over constant parameter tinkering. Settings are chosen from the backtest distribution, not from last week’s pain.
- Pre-approve 3 setting bundles (conservative/normal/aggressive) per algo.
- Switch bundles only if two monthly KPIs miss the target and equity is outside the return band.
- No parameter not present in the original spec; if you add one, it’s a new strategy.
- Log every change with timestamp → bundle → rationale → forward result.
Prop-Firm & Rule-Bound Accounts
He trades within external constraints by designing to the rules, not around them. The edge survives because the risk model assumes those rules are permanent.
- Design to max daily loss = 0.8× firm limit and max total loss = 0.8× evaluation limit.
- Use time-of-day throttles so the book’s risk footprint is smallest during session opens if rules penalize intraday swings.
- If an account hits −50% of the total loss limit, cut all sizes 50% until a new equity high.
Execution, Spreads & Slippage Reality
Alejandro accepts that fill quality is a strategy feature, not an afterthought. He prices in “live tax” so the system remains profitable after friction.
- In testing, add spread = 1.5× broker median and slippage = 0.5–1.0 pip (or 0.02× ATR).
- No market orders during scheduled high-impact news unless the strategy was built for it.
- Reject trades when the effective spread > 3× 30-day median.
- Latency budget: if round-trip > 150 ms for fast algos, cut size by 25% or skip.
News, Holidays & Session Effects
He keeps trading during many events—but only with pre-defined behavior. The idea is to avoid discretionary overrides while respecting known landmines.
- Maintain a calendar; freeze entries 2 minutes before to 3 minutes after Tier-1 events unless the algo is “news-aware.”
- Holiday mode: halve size and widen stops 1.3× when liquidity is thin (e.g., late sessions, bank holidays).
- For Asia/NY/London session handoffs, close the mean-revert risk 10 minutes before the handoff.
Correlation, Buckets & Net Exposure
Alejandro groups risk by “buckets” (currency, setup, timeframe) so diversification is intentional. That structure prevents hidden concentration.
- Cap any bucket at 35% of total risk; if breached, downsize newest entries first.
- Require at least 3 independent edges (e.g., trend-follow, mean-revert, breakout) active most weeks.
- If rolling 20-day P&L correlation between two algos > 0.6, halve the smaller one until it drops.
Daily & Weekly Workflow
His cadence is consistent: monitor, review, and act only when rules say so. That rhythm keeps emotions out and shipping continuous.
- Daily (15–20 min): check broker status, slippage report, KPI dashboard; no discretionary trade changes.
- Weekly (45–60 min): review per-algo equity vs. band, correlation matrix, bucket limits; rotate setting bundle only if triggers fire.
- Monthly: Monte Carlo refresh, parameter drift check, and a “hire/fire” review for algos on probation.
- Keep a rolling 5-slot roadmap: one slot each for research, code, test, deploy, and observe—never skip a slot to rush live.
Size Risk First: Set Return Bands and Hard Daily Loss Limits
Alejandro Perez starts with risk, not signals. He defines a realistic return band from robust testing, so he knows what “normal” looks like in live trading. If equity stays inside that band, he resists tinkering and lets the system work. Only when performance breaks the band does he adjust size or settings, keeping emotion out of the cockpit.
Perez also hardwires loss containment before the first trade. He caps daily loss at a small, fixed fraction of the worst historical drawdown, so a bad session can’t snowball. He scales positions by volatility, so each trade risks roughly the same account percentage. And when drawdown reaches a pre-set tripwire, he automatically cuts risk and waits for a new equity high before restoring it.
Manage Algos Like Employees: KPIs, Probation, Promote or Replace
Alejandro Perez treats each strategy like a staff member with a clear job description and measurable targets. He assigns every algo a one-page mandate—market, setup, timeframes, risk, and expected efficiency—so there’s no ambiguity about what success looks like. Performance is reviewed on a schedule, not on emotion, with KPIs like CAR/MDD and win-rate times payoff acting as the “quarterly report.” If an algorithm drifts below its historical efficiency or starts duplicating another’s risk, Perez resizes it before it becomes a problem.
New strategies go through a probation period with reduced size until they prove they can match live efficiency within predefined return bands. Promotion brings normal risk and a longer leash; demotion means smaller size or fewer concurrent instances. If the same issues persist after review, Alejandro Perez retires the algorithm and reallocates capital to edges that are actually doing the job.
Trade the Regime: Align Trend or Mean Reversion With Volatility
Alejandro Perez starts by identifying the market’s “mood” before he lets any strategy speak. If the trend is dominant, he prioritizes breakout and momentum logic with wider stops and fewer profit targets; if the tape is choppy, he rotates into mean reversion with tighter stops and quicker exits. He keeps filters simple—trend, volatility, and noise—so decisions stay stable across instruments and don’t overfit to last week’s pain. The payoff is consistency: setups fire when the environment matches their DNA, not when a trader is bored.
When volatility expands, Alejandro Perez scales position size down and lets distance-to-stop breathe; when volatility contracts, he nudges size up but trims profit expectations. He also halves exposure when regime signals conflict, trading light until the market picks a side. Correlated strategies don’t run at full throttle in the same regime; he staggers them to avoid duplicate risk. The result is a portfolio that adapts without constant tinkering—rules decide the throttle, not gut feel.
Keep Testing Honest: Freeze Rules, Walk-Forward, Monte Carlo Drawdowns
Alejandro Perez keeps testing brutally simple: freeze the rules before you ever see a result. He writes the strategy in plain English, codes exactly that logic, and refuses to “upgrade” it after peeking at equity curves. Then he splits data into in-sample and out-of-sample and demands that live-like efficiency holds up, not just the cherry-picked window. If the edge only works on the period it was built on, it’s not an edge—it’s nostalgia.
Next, Alejandro Perez stress-tests the idea with Monte Carlo resampling to estimate a realistic worst-case drawdown and sizes from that, not from hope. He bakes spreads and slippage into tests so the system is profitable after friction, not just on paper. Parameters live inside pre-approved ranges, and switching a range is a documented settings change—not a stealth tweak. Promotions to full size require out-of-sample performance within the expected return band; miss twice, and the system gets demoted or archived. Every change is logged with rationale and forward results, so accountability survives the next bad week.
Diversify Smartly: Limit Correlation, Bucket Exposure, Balance Timeframes and Strategies
Alejandro Perez doesn’t chase variety for its own sake; he diversifies to control damage. He groups positions into buckets—by underlying, setup type, and timeframe—so hidden overlap can’t sneak past him. If a bucket gets too heavy, he cuts the newest or least efficient positions first to keep total risk balanced. Rolling correlation between algos is monitored like a vital sign; when two systems start moving together, one gets resized until independence returns.
Time diversification is just as deliberate for Alejandro Perez. He offsets entries across sessions and mixes holding periods so wins and losses don’t stack in the same hour. Trend, mean reversion, and breakout logic share capital, but they never run full throttle simultaneously if net exposure tilts the same way. The result is a portfolio that can take a punch from one market or regime without going to the canvas—because no single idea is allowed to carry the whole fight.
Alejandro Perez’s core message is simple: build your trading around what the market and your data will realistically give you, not what you wish would happen. His use of return bands and pre-defined loss limits keeps expectations grounded and decisions unemotional, so a normal drawdown doesn’t trigger destructive tweaks. He treats strategies like employees with clear mandates and KPIs, promoting what works and resizing or replacing what doesn’t—no sacred cows, just accountability and capital efficiency.
Equally important, Alejandro Perez only lets a setup trade in the environment where it actually has an edge. Clean regime checks—trend, volatility, and noise—decide when to push, when to coast, and when to stand down. His testing loop is deliberately boring: freeze rules, prove them out-of-sample, and size to Monte Carlo drawdowns with real-world friction layered in. Finally, he diversifies like a risk manager—by bucket, timeframe, and correlation—so no single idea can sink the book. Put together, his playbook is a blueprint for longevity: clear guardrails, honest testing, adaptive execution, and a portfolio that can survive the dull months and still be around for the big ones.
