Autopsy Report
Patient: Your EA
Date of death: Usually 3–6 months post-purchase
Cause of death: Unknown to the buyer at time of purchase. Predictable in hindsight with the right pre-purchase checks. Six causes account for nearly all cases. All six are documented below with diagnostic tests you can run before buying.
Why Do Profitable EAs Stop
Working After You Buy Them?
Published 15 June 2026 · 12 min read
Six causes explain nearly all EA failures after purchase: curve-fitting on historical data, no live forward test before sale, market regime change, spread regime change at your broker, developer abandonment, and wrong broker from day one. The key insight is that all six are diagnosable before purchase if you ask the right questions. This page gives you the diagnostic for each.
The Psychology of EA Purchase Euphoria
The pattern is so consistent it has become a cliché in EA trading communities: the backtest looks extraordinary, the first week of live trading is promising, and then performance quietly deteriorates over weeks until the account is flat or negative. By this point, the buyer has usually convinced themselves it is "temporary," endures another month of losses, and eventually turns the EA off in frustration.
The emotional sequence — excitement, hope, doubt, frustration, acceptance — is predictable because the underlying causes are predictable. The EA was sold on evidence that was not representative of future conditions. This is not always deliberate fraud. Sometimes the developer genuinely believed their backtest. Sometimes they were using a market environment that no longer exists. The result is the same.
Understanding why EAs fail is more useful than feeling deceived. The six causes below give you a framework for evaluating any EA before purchase and for extending the life of one you already own.
6 Autopsied Causes — Click to Expand
The EA's parameters were fine-tuned on historical data until the backtest looked perfect. This is not evidence the strategy works — it is evidence the parameters were calibrated to the past. Applied to new data, the edge disappears because real future prices don't follow the same statistical patterns as the optimised historical window.
How to test for this before buying
Ask the developer for walk-forward analysis results. This runs the EA on out-of-sample data it was not optimised on. If only standard backtests exist — and especially if they show suspiciously smooth equity curves — curve-fitting is likely.
An EA that was never run on live conditions before being sold has only been tested in the controlled environment of the backtester. Backtests use perfect execution — no slippage, no requotes, instant fills. Real trading has all of these. An EA that looked profitable at zero-spread perfect execution often fails at real broker conditions.
How to test for this before buying
Request a live forward test account link (Myfxbook or similar) showing at minimum 3 months of real trading with documented broker and lot size. If none exists, the EA has not been forward-tested.
A strategy optimised for ranging conditions fails during trending periods, and vice versa. If XAUUSD was in a low-volatility range during the backtest period but transitions to a high-volatility trend during the period after you buy, a range-based EA produces losses on every single trade — correctly following its logic in conditions that no longer suit it.
How to test for this before buying
Ask whether the strategy has a regime filter. If the answer is "it works in all conditions," be sceptical — most strategies have a preferred regime and the developer should know what it is.
If your broker widens its average spread after you start trading — due to policy change, risk management decisions, or removing a liquidity provider — an EA calibrated for tighter spreads suddenly has a smaller edge or no edge at all. A 1-pip spread increase on a 10-pip target scalping strategy represents a 10% reduction in maximum profit per trade.
How to test for this before buying
Monitor your broker's spreads over time. If you notice consistent widening outside news hours, compare to a second broker. Test whether the EA performs differently on the alternative broker. The difference isolates spread impact.
The developer sold enough copies to consider the product complete and moved on. No updates for 18+ months means no broker compatibility checks, no adaptation for changing market conditions, no bug fixes. Older code built for a different gold volatility environment runs unmodified in conditions it was never tested against.
How to test for this before buying
Check the EA's changelog or update history. Ask the developer directly when the last update was and what triggered it. A developer who cannot answer this is not actively maintaining the product.
The developer tested the EA on a low-spread ECN broker with excellent execution. You run it on a standard account market maker with higher spreads and frequent requotes. The exact same EA produces different results. The developer's impressive results are real — just not replicable in your configuration.
How to test for this before buying
Ask the developer specifically which broker they use and have tested on. If they cannot name one, or if they push a specific broker without explaining why, investigate independently. Then compare your broker's spread and execution to their tested environment.
Red Flags Visible Before Buying
Each of the six causes leaves traces that are visible before purchase if you know where to look. A suspiciously smooth backtest equity curve (no real drawdowns, consistent month after month) is a curve-fitting signal. No Myfxbook forward test link is a forward-test-absent signal. No update history in the product page is an abandonment signal. Pushing a specific broker you must use is a "only works on one broker" signal.
The developer who is honest about losing periods, who shows drawdown statistics prominently rather than hiding them, and who can name the specific broker their results were achieved on is demonstrating the opposite pattern. These are positive signals — not guarantees, but meaningful evidence.
Structural durability looks like: a named strategy (M15 range breakout + H4 EMA filter), hard stops on every trade (no martingale), a developer who maintains and updates, and documentation that includes honest discussion of losing periods. Goldie Razor V2.8.4 is an example of what this looks like applied — each of these elements is present and verifiable rather than implied by marketing language.
Will Your EA Last? — 7-Question Durability Test
Apply this to an EA you are evaluating or already running. Each "No" is a specific structural weakness.
1. Was the EA forward-tested for 3+ months before going on sale?
2. Is the strategy logic documented (not a black box)?
3. Has it received an update within the last 12 months?
4. Does it work on multiple brokers (not just one specific broker)?
5. Does it have a defined drawdown limit or auto-pause mechanism?
6. Is the strategy logic time-invariant (not dependent on specific 2018–2020 data quirks)?
7. Does the developer actively communicate with buyers?
How to Extend an EA's Working Life
For EAs you already own, several steps extend working life and reduce the cost of underperforming periods. First: strict risk management. Running at 0.5–1% risk per trade means a losing month produces a 5–10% drawdown rather than an account destruction event. This gives you time to assess whether the problem is temporary or structural.
Second: broker diversification. If your EA is underperforming, compare its results on a demo account at a second broker using identical settings. If the second broker produces better results, the problem is broker-specific — switch, and you may recover the edge. If both produce identical underperformance, the issue is more likely market-regime-related.
Third: stay in contact with the developer. Developers who are still active can tell you whether the underperformance is expected, whether an update is coming, or whether the EA needs a settings adjustment for current conditions. Silence from a developer after you report problems is a signal that the product is effectively abandoned.
Related Reading
Due diligence before buying
A systematic checklist for evaluating a gold EA before committing any money.
Questions that reveal EA durability
The specific questions that separate long-lasting EAs from ones that fail fast.
Is automation profitable in the first place?
The foundational question: what percentage of gold EAs actually produce returns.
Free EAs and the abandonment problem
Why free gold EAs face a particular version of the developer abandonment problem.
Docs that prove an EA is maintained
What documentation you should require before trusting an EA with live capital.
Frequently Asked Questions
Community data suggests the most common failure window is 3–6 months post-purchase. This aligns with the time it takes for the market to present conditions outside the historical window the EA was optimised on. Some over-fitted EAs fail within weeks. A small number perform consistently for 12+ months — these are invariably the ones that were forward-tested before sale and are actively maintained.
To a degree. Signs of curve-fitting in a backtest include: near-perfect smooth equity curve with very low drawdown, performance that improves dramatically when parameters are slightly adjusted, and a backtest that only covers 1–2 years (insufficient data to average across multiple market regimes). The definitive test is walk-forward analysis — running the EA on out-of-sample data it was not optimised on. Developers who provide walk-forward reports are demonstrating significantly more rigour than those who only provide backtests.
A market regime is the dominant character of price behaviour over a period — ranging, trending, high-volatility, low-volatility, risk-on, risk-off. EAs optimised for a ranging market produce steady returns during ranging and lose money during trending periods. The flip is also true. EAs that incorporate regime detection (e.g. an EMA trend filter that changes behaviour based on whether the market is trending) are more robust across different regimes than pure single-mode strategies.
Only if you have MQL5 programming skills and access to the EA's source code, and you understand the original strategy logic well enough to know what to change. Most retail buyers receive a compiled .ex5 file with no source access. In that case, the only options are: wait for the developer to release an update, contact the developer to report the issue, or accept the EA has reached end-of-life and move on. Trying to fix a compiled EA without source access is not possible.
Several steps: First, run it on a strict risk-managed lot size (max 1% per trade) so that underperforming periods do not cause major damage while you assess. Second, monitor live performance weekly and set a maximum monthly drawdown threshold where you pause the EA (e.g. 5–8% in a month). Third, stay in contact with the developer — ask about upcoming updates or known issues. Fourth, keep the broker you use updated: if your broker changes their execution or spread regime, test whether a different broker restores performance.
Goldie Razor V2.8.4
M15 breakout + H4 EMA filter — built for XAUUSD on MT5