What Is Model Decay (And Why It Destroys Your Profits)

Your EA was trained on historical data. That data had specific volatility patterns, correlation structures, and spread environments. The model learned those patterns and built its profitability around them.

Then the market changed. Volatility spiked. Correlations inverted. Spreads widened. The patterns your EA memorized no longer apply.

This is called concept drift in machine learning—the statistical properties of the target variable change over time. In trading, it's called regime shift. Same problem, different name.

The cost? Every month your EA runs on stale patterns, it bleeds edge. Some months it's obvious (drawdown appears). Most months it's invisible (you just make a little less). By the time you notice, weeks of profit have vanished.

Your EA Isn't Broken—It's Decaying

Every trading bot decays. It's not a flaw in the code. It's a feature of markets.

Markets aren't static. They evolve. Volatility cycles through regimes. Correlations break down during stress. Institutional order flow changes with leverage caps and regulations. Retail access changes with broker rules. Spreads widen and tighten. Economic data surprises.

Each of these shifts pushes the market away from the historical distribution your EA learned from. The bot's accuracy drops. Its edge evaporates.

The timeline varies:

But all of them decay. None are immune. An EA that's 95% accurate in backtest might be 60% accurate six months after deployment.

Three Months Without Monitoring: Here's What Happens

Let's say you deployed an EA last month and it made $1,200 with a 65% win rate. Fast forward three months without checking it. Here's the silent degradation:

  1. Month 1: Win rate stays at 65%, but average profit per trade drops to $85. Total profit: $980. You didn't notice because it was still positive.
  2. Month 2: Win rate dips to 62%. Average profit: $78. Total profit: $670. You chalked it up to a "slow month."
  3. Month 3: Win rate hits 58%. Average profit: $55. A losing month at $220. Suddenly you panic and shut it down.

What actually happened? The market regime shifted. Your model's predictive patterns stopped working. The decay was gradual and invisible until the losses became undeniable.

Now you're faced with a choice:

Most traders pick option 1. The cost? You lose a strategy that probably still has edge—it just needs retraining.

How to Spot Decay Before Your Account Blows Up

Decay isn't random. If you know what to measure, you'll see it coming.

  1. Sharpe ratio trends down. If your EA's monthly Sharpe drops from 2.0 to 1.2 to 0.8, decay is happening. The absolute profit might be the same, but consistency is falling apart.
  2. Win rate creeps lower. A healthy EA with a 60% win rate that drops to 55%, then 50%, then 45% is losing statistical edge. Plot this monthly. A downward trend = decay in progress.
  3. Average profit per trade shrinks. Even if trade count stays the same, smaller profits per trade indicate the signal is weaker. Market conditions have shifted away from the model's assumptions.
  4. Largest loss increases. If your typical drawdown was 2-3% and suddenly hits 5-6%, the model is making worse decisions. Regime change often punishes old strategies hard before blowing them up.
  5. Backtest diverges from live performance. If your last backtest showed 40% annual return but you're running at 15%, you have a decay problem.

The fix is simple: pull recent data (last 6-12 months), retrain the model, backtest on fresh data, redeploy. But most traders don't do this because it requires expertise in data science and access to MT4/MT5 development.

The Difference Between Backtest and Live Performance

Here's the painful truth: your backtest is a lie.

Not intentionally. But historical backtests are biased. They're tested on data the model "saw" during training, which means they're overfitted to patterns that may not repeat. In live trading, those patterns are gone.

Even worse, backtests don't account for market regime changes. You run a 5-year backtest and assume the future will look like the past. It won't. The market never looks the same twice.

This is why profitable backtests fail live. And why live performance decays—the backtest assumptions don't hold in real market conditions.

A proper EA development process handles this:

  1. Train the model on historical data
  2. Test on out-of-sample data (recent periods the model never saw)
  3. Validate on walk-forward testing (rolling windows to catch regime drift)
  4. Deploy with continuous monitoring
  5. Retrain monthly on recent data

Most DIY traders skip steps 4 and 5. They build once, deploy, and hope. When decay hits, they assume the model is broken instead of recognizing it needs retraining.

Why Automated Monitoring Beats Manual Checking

You could manually check your EA's metrics every month. Most traders don't. It requires discipline, spreadsheet tracking, and honest evaluation.

Automated monitoring removes that friction. A dashboard shows you:

If decay risk hits 60+, the system alerts you. You know it's time to retrain before profits evaporate.

The alternative is hoping. Hoping the market stays favorable. Hoping you remember to check your EA. Hoping you catch decay before three months of losses pile up.

The real question isn't whether monitoring costs $200/month. It's how much you're losing monthly to decay.

Why Hiring Beats DIY Maintenance

You could learn to retrain your EA yourself. You could master MQL5, understand walk-forward optimization, and commit 5-10 hours per month to maintenance per strategy.

Or you could focus on what you actually do well: identifying which strategies make sense to trade.

A monitoring and maintenance service handles:

This is what professional trading operations do. Not because they're paranoid. Because they've watched decay destroy accounts. They know the cost of ignoring it.

The gap between professional maintenance and DIY is massive. A professional catches decay at month 1. A DIYer catches it at month 3 (if they're lucky). That's two months of preventable losses.

We've completed 660+ projects on MQL5. Speed isn't luck—it's discipline. We deliver a working monitoring dashboard in 45 minutes, full system in hours. Every EA includes a full backtest report so you'll never wonder if decay is happening.

Key Takeaways

• All EAs decay as market conditions shift
• Decay is invisible until it causes losses
• Monthly monitoring catches the problem early
• Retraining is faster and cheaper than rebuilding
• The cost of maintenance is far less than the cost of inaction

Your EA didn't fail. The market changed. The question is whether you'll adapt faster than the decay spreads.