What Is an Expert Advisor? EA Trading Explained for MT4 and MT5

By Jakub Rož · Founder & CEO, For Traders

An expert advisor (EA) is a program written in MQL4 or MQL5 that runs inside MetaTrader and automatically opens, manages and closes trades according to coded rules, without you clicking anything. It is not a signal service and not an indicator — it executes.

Key takeaways

• An expert advisor is executable trading logic attached to one chart in MT4 or MT5; it places and manages orders on its own using MQL4/MQL5 code.
• The core of every EA is the OnTick event — on each incoming price tick the code re-checks entry, exit and risk conditions, then acts or does nothing.
• An indicator draws, a script runs once, an EA runs continuously and trades — the distinction matters when you install files into the right MetaTrader folder.
• Most 'profitable' EAs die for four reasons: curve-fitted backtests, unrealistic spread and slippage assumptions, no news filter, and no drawdown governor.
• One EA per chart is a hard architecture limit; magic numbers are how multiple EAs avoid managing each other's positions.
• Automated strategies are permitted on the For Traders Challenge within the rules — the same daily loss limit and max drawdown apply to a bot as to your hands.

Watch: related video

What is an expert advisor in trading?

An expert advisor (EA) is a compiled MQL program attached to a MetaTrader chart that opens, modifies and closes positions on its own, based entirely on coded rules — no clicking, no hesitation, no emotion. That's the expert advisor meaning in one sentence: it's an executable object with order-sending permission, sitting on your chart, waiting for its conditions to trigger.

Expert advisor meaning in plain terms

Think of an EA as a trader who never sleeps, never second-guesses a stop loss, and never widens a daily loss limit "just this once." You write (or buy) the logic — entry conditions, position sizing, stop and target placement, maybe a trailing routine — and the EA executes it tick by tick. The EA trading meaning that matters for funded traders is simple: it's automation with real order permissions, not a suggestion engine. When the smiley face is green in the top-right corner of your MT4 or MT5 chart and Auto Trading is enabled, the EA can place trades with real fills, real slippage, and real drawdown consequences on your equity curve — simulated capital during a Challenge, live capital on a funded account.

What an EA is not: signals, indicators and copy trading

This is where most retail traders get confused, so let's be precise. A signal service sends you a message — "buy XAUUSD at 2410" — and you decide whether to click. An indicator overlays or plots data on your chart (think RSI, ATR bands, a moving average cross) but never touches your order book. Copy trading mirrors another trader's fills into your account through a broker-side or third-party bridge. None of these three has order-sending permission baked into the object itself. An EA does. That's the dividing line: if the program can independently send, modify or close an order without a human clicking a button, it's an EA (or its equivalent on another platform). If it only displays information or relays a suggestion, it's not.

Where the term comes from (MetaTrader, MQL4, MQL5)

"Expert advisor" is MetaTrader-native vocabulary — it doesn't exist as a formal term on other platforms, which is why the ecosystem around it looks fragmented from the outside. MetaTrader 4 (MT4) runs EAs written in MQL4; MetaTrader 5 (MT5) runs EAs written in MQL5, a more modern, object-oriented language with better backtesting and multi-asset support. Switch platforms and the vocabulary switches too: cTrader calls the same concept a "cBot," written in C#; NinjaTrader calls it a "strategy," written in NinjaScript. In retail forums, you'll see forex EA bot, "algo," and trading robot thrown around interchangeably — and functionally, they're describing the same thing: a compiled program with permission to trade your account without you touching the mouse.

How do expert advisors work? The tick-by-tick logic

An expert advisor works as an event-driven program: it wakes up when the platform fires specific events — attach, price update, removal — and runs code in response. Understanding how do expert advisors work at this level is the difference between debugging a broken EA in five minutes and staring at logs for an hour.

OnInit: what happens the moment you drag the EA onto a chart

The instant you drop an MQL5 expert advisor onto a chart, OnInit() fires once. This is the setup pass — it validates your input parameters, checks the symbol and timeframe actually make sense for the strategy, sets up any indicator handles the EA will read from, and confirms magic number and lot sizing logic are ready. If OnInit throws an error — wrong symbol, bad lot step, missing indicator buffer — the EA won't trade at all, and you'll see it in the Experts log before a single order gets sent.

OnTick: the decision loop that runs on every price update

OnTick is the engine room. Every incoming quote — every tick — triggers this function, and the EA runs its full check: session filter, spread filter, news filter, then entry conditions, then position sizing. Here's the part every "EA explained" article skips: on the overwhelming majority of ticks, the correct output is do nothing. A well-built EA is idle on roughly 99% of ticks — it's waiting for its specific condition, not force-fitting a trade into every price move. That idle-by-default behavior is a sign of good code, not a bug.

Tick frequency itself isn't constant across instruments. XAUUSD generates ticks far more often than a quiet cross like EUR/USD during Asian session — gold's volatility and volume mean OnTick fires constantly during London and New York overlap, which pushes more CPU cycles and demands tighter execution logic than a slow-moving pair would need.

Order execution, SL/TP placement and trade management

Once entry conditions pass, the EA calls an order-send function that packages the request — symbol, volume, price, stop loss, take profit — and fires it to the broker's server. Fill confirmation, slippage tolerance, and requote handling all happen here. After the trade's live, the same OnTick loop keeps managing it every tick: trailing stops, partial closes, breakeven shifts — all logic re-evaluated on each new price, not on a timer.

OnDeinit, magic numbers and the one-EA-per-chart rule

When you remove the EA, switch timeframes, or the terminal shuts down, OnDeinit() runs — releasing indicator handles and logging the reason for the shutdown. Two hard constraints matter for automated trading MetaTrader setups: first, the magic number is the ID tag stamped on every order an EA opens, so it can recognize and manage its own trades without touching another EA's positions on the same account. Second, only one EA can run per chart — so multi-pair automation means either multiple charts open simultaneously, or a single EA looping through a symbol array inside one OnTick call.

Expert advisor vs indicator vs script vs bot

An indicator draws, a script executes once and exits, and an expert advisor runs continuously and trades. That's the whole distinction — everything else in this section is just the mechanics behind it.

People throw the word "bot" around like it means something specific in MetaTrader. It doesn't. "Bot" is a colloquial umbrella term — traders use it to mean EA, sometimes to mean a signal-copying script, sometimes to mean a Telegram alert feeding a manual clicker. Inside the MQL4/MQL5 ecosystem there's no file type called a "bot." If someone's selling you a "bot," ask which of the three actual categories it falls into.

The attribute-by-attribute comparison

Which MetaTrader folder each file goes into

All three are compiled in MetaEditor, but they don't share a home. The folder they land in is the tell:

• MQL5/Indicators — indicator files only
• MQL5/Scripts — one-shot scripts
• MQL5/Experts folder — expert advisors, full stop

Drop a file in the wrong folder and MetaTrader won't complain — it just won't show up where you expect it in the Navigator window. That silence is the most common source of "why isn't this working" support tickets on every prop and retail forum.

Why the distinction matters when you buy something

Two beginner errors show up constantly. First: buying an "indicator" off a marketplace expecting it to trade for you — it won't, it can only draw signals for you to act on manually. Second: downloading an EA, dragging it onto the chart, then wondering why nothing happens, only to find it was dropped into the Indicators folder instead of MQL5/Experts and never appeared in the Navigator at all.

Before you pay for anything labeled "automated," ask the seller directly: does this place trades itself, or does it just draw levels? The answer tells you which folder it belongs in — and whether it needs Auto Trading switched on to do anything at all.

Types of expert advisors you'll actually encounter

Not all EAs are built the same, and the category tells you almost everything about how it'll behave once real spread and slippage enter the picture. Broadly, you'll run into four archetypes: trend/breakout systems, scalping and news EAs, grid/martingale systems, and execution assistants that never place an entry at all.

Trend-following and breakout EAs

These trade rarely — sometimes a handful of times a month — and their edge lives or dies on the fill quality of that first leg. A breakout EA waiting for price to clear a session high doesn't care about your average spread; it cares about the spread and slippage at the exact moment volatility spikes through the level. Miss the fill by a few points on XAUUSD during a fast breakout and the whole trade's R:R shifts. These are the most "honest" EAs in the sense that losses show up as losses — no hidden risk, just a string of small stops and the occasional trend that pays for all of them.

Scalping and news EAs (the latency-sensitive ones)

A scalping EA or news trading EA lives and dies by execution speed and spread. These systems are designed around tiny targets — sometimes a handful of pips — so a widened spread during rollover or a 200ms latency lag on a slow VPS eats the entire expected edge before the trade even starts working. This is why the same news EA that backtested beautifully on tick data can bleed live: the backtest assumed fills that a retail VPS three hops from the server can't deliver during NFP or FOMC.

Grid and martingale EAs — why the equity curve looks perfect until it doesn't

A grid EA or martingale EA produces, statistically, the smoothest backtest equity curve you will ever see — months of tiny, steady gains with barely a dip. The mechanism is the problem: instead of realizing a loss, the system adds to the position at a worse price, hiding risk in an open, unrealized drawdown rather than a closed trade. The equity curve stays flat right up until a sustained move refuses to revert, and the "smooth" system produces one drawdown large enough to erase a year of gains — often past a challenge's max daily loss limit in a single session.

Execution assistants: trade managers, trailing stops and risk governors

The underrated category: EAs that never place a single entry. A trailing stop EA, breakeven manager, or ATR-based partial-close tool just manages positions you opened manually — moving your stop to breakeven at 1R, trailing by a multiple of ATR, scaling out at fixed R multiples. For a discretionary trader running a challenge, this is often the smartest first EA to run: it removes the emotional part (moving a stop back "just in case") without handing entry decisions to code you haven't stress-tested.

Are expert advisors profitable? The four failure modes

Some EAs are genuinely profitable over time; most vendor EAs sold on forums and marketplaces are not. The gap between a gorgeous backtest and a live account almost always traces back to one of four things: curve fitting, unrealistic execution assumptions, no news filter, or no drawdown governor. Understand these four and you can tell a real system from a curve-fit toy in about ten minutes.

Curve fitting: 300 optimisation passes and one lucky parameter set

Run a genetic optimiser across 6 input parameters over 5 years of EURUSD data and it will happily test hundreds of combinations until it finds one that produces a beautiful, straight-line equity curve. That's not a strategy — that's the optimiser finding the exact noise pattern of the past. It has no reason to repeat. The honest test is out-of-sample: optimise on 2019-2023, then run the locked parameters untouched on 2024-2026 data you never showed the optimiser. If performance collapses, you've found curve fitting, not edge. Forward testing on a demo account for 60-90 days before risking a challenge account catches what optimisation alone never will.

Spread and slippage assumptions the Strategy Tester won't model

MetaTrader's Strategy Tester defaults to fixed spread and can generate fills that never happened, especially on older tick data with gaps. A scalper showing a 96% win rate in the tester on XAUUSD is measuring against a spread that doesn't exist at 3am server time or during a spike. Live, gold spreads widen from 20 cents to $2+ in seconds, and slippage on a market order during a fast move can eat an entire scalp's target in one fill. Backtest vs live gap is smallest for trend/breakout EAs holding trades for hours, and largest for high-frequency scalpers stacking dozens of thin-margin trades a day — the exact type most likely to be marketed with the flashiest equity curve.

No news filter: NFP, FOMC and the 40-pip gap through your stop

A stop-loss is a request to the broker, not a guarantee of price. Around NFP or an FOMC rate decision, liquidity thins and price can gap 40 pips through a stop before any fill occurs — you get filled at the next available price, not your stop level. An EA without a news filter treats a scheduled 2pm Fed announcement exactly like a quiet Tuesday afternoon. Professional systems either flatten positions before high-impact releases or widen stops and cut size going in. Vendor EAs almost never do this because a news filter reduces the backtest's trade count and makes the curve look less impressive.

No drawdown governor: the missing daily loss limit inside the code

Serious EAs carry a hard-coded daily loss limit and an equity floor check — hit either one and the EA flattens every open position and refuses to open a new one until the next session. This is a governor, not a suggestion buried in the strategy description. Retail EAs sold on marketplaces almost never include this logic, because the vendor's backtest never needed to survive a challenge's max drawdown rule. That's precisely what blows funded accounts: the EA doesn't know it just breached your daily loss limit, so it opens trade number six on the same losing day.

How to install and run an expert advisor step by step

Installing an EA takes five minutes; trusting it with real risk should take a month of testing first. The mechanics are near-identical whether you're asking what is an expert advisor in MT4 or what is an expert advisor in MT5 — the file extension changes (.ex4 vs .ex5), the workflow doesn't.

Installing and enabling the EA in MT4 or MT5

This is the exact sequence for how to install an expert advisor without hunting through forums:

1. Open the platform, click File → Open Data Folder.
2. Navigate to MQL4/Experts (MT4) or MQL5/Experts (MT5).
3. Paste the .ex4 or .ex5 file into that folder. If you only have source code, open it in MetaEditor and compile it first — the compiled file is what lands in Experts.
4. Right-click Expert Advisors in the Navigator panel and hit Refresh.
5. Drag the EA from the Navigator onto the chart you want it running on.
6. In the input dialog that pops up, tick Allow Algo Trading (MT5) or Allow live trading (MT4) on the Common tab, then set your inputs on the Inputs tab.
7. Click OK, then confirm the Auto Trading button in the toolbar is toggled on (green).
8. Look for the small smiley face in the top-right corner of the chart — that's your confirmation the EA is attached and live. A sad face means algo trading is blocked somewhere in the chain.

Backtesting properly in the Strategy Tester

Skipping this step is how traders find out their EA was curve-fit only after it's blown a challenge account. Run it properly:

1. Open Strategy Tester (Ctrl+R), select your EA and the correct symbol.
2. Set modelling to Every tick based on real ticks — anything less understates slippage and skips fills that matter on fast candles.
3. Add a realistic spread, not the tightest number you've ever seen quoted — use the average spread your broker or prop platform actually gives you.
4. Split your data: optimize on an in-sample period, then run the untouched out-of-sample period to see if the edge survives.
5. Once the backtest holds up, forward test on a demo account for at least a month before risking a challenge attempt on it.

Best expert advisor settings: what to actually tune

Traders obsess over indicator periods and ignore the settings that actually keep an account alive. Prioritize in this order: risk per trade (fixed % of balance, not fixed lot size), a max spread filter that blocks entries during widened spreads around news, defined session hours so the EA isn't taking trades in dead Asian-session chop it was never tested on, and a hard news blackout window around high-impact releases. An RSI period tweak from 14 to 12 will not save an EA that's still trading through FOMC.

Do you need a VPS to run an EA?

Yes, if your strategy is latency-sensitive or needs to run 24/5 — a scalping or news-reaction EA loses its edge with the extra milliseconds of round-trip time from a home connection, and a laptop that sleeps or reboots mid-week will simply stop trading. No, if it's a daily-timeframe EA and your machine reliably stays on and connected. A VPS located close to your broker's trade server cuts execution lag meaningfully in 2026's tighter latency environment, where sub-20ms round trips are common for co-located setups — worth the monthly cost once your position sizing gets serious.

Where to get an EA: MQL5 Market, custom builds, or write your own

The honest hierarchy is simple: an EA you understand beats an EA you rent. You can buy one off the MQL5 Market, hire a coder for a custom build, or write your own in MetaEditor — but only the last two force you to actually know what your money is doing while you sleep.

What MQL5 Market vetting does and doesn't verify

Every MQL5 expert advisor listed on the marketplace passes a technical review before it goes live: the code has to compile cleanly, it can't contain obvious malware or terminal-breaking bugs, and it has to run without crashing MetaTrader. That's it. Vetting does not certify profitability, does not check the strategy logic for hidden risk, and does not stress-test the equity curve. A "verified" signal history attached to a listing can still be a small account running at absurd leverage, or a martingale system that's simply not been hit by the losing streak yet. You're looking at a track record, not a guarantee — the same distinction that separates a demo-only backtest from live simulated performance in any prop evaluation.

Red flags in vendor EA marketing

• No maximum drawdown figure anywhere in the listing — only "average monthly return"
• Performance shown is backtest-only, with no forward or live signal linked
• Examples use fixed lot sizes rather than percentage risk, which hides true exposure as the account grows
• Copy that says "no stop loss needed" — this almost always means a martingale or grid recovery mechanism instead
• Results pulled from a single symbol and a single, favorable period (say, one low-volatility year on EURUSD) with no multi-asset or multi-regime data

Should a beginner buy an EA or build one in MetaEditor?

Build one, even a bad one. MetaEditor ships free with every MetaTrader install, and the MQL5 wizard will generate a working skeleton EA from a menu of built-in signals and money-management modules in minutes — you don't need to be a programmer to get something running on a chart. The real value isn't the EA you end up with; it's that coding your own manual rules forces you to define entries, exits, and risk in exact numbers instead of vague chart feel. That discipline pays off even if the EA never touches a live or funded account.

If MetaTrader isn't your platform, the same logic exists elsewhere: cTrader supports cBots written in C#, TradingView lets you wire Pine Script alerts into execution via webhooks, and NinjaTrader is the go-to for futures automation on CME products. Different ecosystem, same principle — automation only works if you understand the rules underneath it.

Do prop firms allow expert advisors on funded challenges?

Yes — most prop firms that allow expert advisors, including For Traders, permit automated strategies on the Challenge, as long as the EA plays by the same rules a human trader has to follow. What gets an account flagged isn't the fact that a robot is clicking the buttons — it's latency arbitrage, tick-scalping exploits, copy trading across multiple funded accounts, or HFT-style abuse of simulated pricing gaps. Automation itself is not the violation.

The rules that usually apply to automated trading

Automated trading rules across the industry tend to target the same handful of abuses: exploiting broker-feed latency between servers, arbitraging stale quotes during low-liquidity windows, running the same EA logic across multiple accounts to multiply payouts from one edge, and tick-scalping strategies that rely on execution quirks rather than genuine price movement. If your EA just trades a coded strategy — trend-following, mean-reversion, breakout — on normal spreads and normal timing, you're inside the lines.

Why the daily loss limit and max DD are the real constraint

An EA doesn't get its own risk rules. The daily loss limit and max drawdown apply identically whether a human or a script pulled the trigger. This is where martingale-style EAs die fast: a strategy that doubles size after every loss can blow through both the daily loss limit and max drawdown in a single session while technically "still in a trade" and never having hit a hard stop. The evaluation doesn't care that the EA had a plan — it cares that equity breached the line.

Running an EA on a For Traders Challenge

On a For Traders Challenge, the EA has to respect the same daily loss limit and max drawdown as manual trading — no exceptions for automation. Tactically:

• Hard-code an equity floor into the EA itself, not just a mental stop — the code should refuse new entries once daily loss limit is close.
• Add a hard daily loss cut-off inside the logic, separate from the platform's own limit, so the EA stops itself before the account does.
• Build in a news blackout for NFP and FOMC — spread widens and slippage spikes exactly when a naive EA is most likely to get filled badly.
• Cap the spread the EA will trade through; refuse entries above a threshold.
• Forward test on the free demo first — prove the equity curve before you pay for an evaluation attempt.

If you've already forward-tested for weeks and the equity curve is clean, Instant Funding lets you skip the evaluation phase entirely. If the strategy still needs seasoning, the Two-Step Challenge gives the EA more time and more sessions to prove consistency before performance rewards are on the line.

What tends to get an automated account disqualified

XAUUSD and US100 are the two most-traded instruments on the platform — and the two where a naive EA gets hurt worst, because ATR expansion during news or session opens can blow past backtested assumptions in seconds.

Building an EA workflow that survives live conditions

An EA that passes backtest doesn't automatically survive live spread, latency, and news volatility — you need a launch checklist and daily monitoring routine, plus kill criteria decided before you're emotionally attached to the equity curve. Most EAs that blow accounts weren't bad code; they were undermonitored code.

The pre-launch checklist

Before an EA touches a funded account or challenge, run it through this list — every box, no shortcuts:

• Out-of-sample results: performance on data the EA never saw during optimization, not just the curve-fit backtest window
• Forward testing length: minimum 4-8 weeks live-demo before real capital, longer for strategies with wide drawdown swings
• Worst historical drawdown: know the exact number, not "should be fine" — this becomes your kill threshold later
• Spread sensitivity test: rerun the backtest with 2-3x the broker's typical spread on XAUUSD or US100 to see if the edge survives real execution costs
• News filter: hard-coded pause around NFP, FOMC, CPI — non-negotiable for gold and index EAs
• Hard equity floor: a coded stop that disables the EA if account equity hits a defined line, independent of the daily loss limit
• Magic number set: unique identifiers per EA instance so you can audit which strategy did what when things go wrong
• VPS uptime: a virtual private server with verified uptime history — a dropped connection during an open position is how set-and-forget becomes set-and-lose

Monitoring: what to check daily even on a set-and-forget EA

Set-and-forget describes the entry method, not your obligation to watch the account. Check daily:

• Open exposure — total lots and correlated positions across magic numbers
• Journal errors — failed orders, rejected modifications, terminal disconnects
• Requote and slippage logs — if execution is consistently worse than backtest fills assumed, your realised R:R will drift below the modeled distribution
• Realised R:R versus backtest — a widening gap here is the earliest honest signal something's off, before drawdown even shows it

When to switch the EA off

Define kill criteria before launch, not during a losing streak, when judgment is compromised:

• Consecutive losses beyond the worst historical streak in backtest
• Drawdown exceeding backtest max by a defined multiple — 1.5x is a common line
• A regime shift the EA has never traded through — a volatility expansion well outside the historical ATR range it was built on

This is the same discipline that separates traders who pass a Trading Challenge from those who don't — not smarter code, but the willingness to shut a losing system off before it proves you wrong twice.

Pros and cons of trading with an expert advisor

Pros

• Removes hesitation and revenge trading — the code takes every valid signal, including the ones you'd talk yourself out of
• Trades sessions you can't watch, which matters for Asian-session or overnight US index strategies
• Forces you to define entries, exits and position size as hard numbers instead of feel
• Reacts in milliseconds and never mis-clicks a lot size or forgets a stop
• Backtestable and forward-testable, so a strategy can be measured before real risk

Cons / risks

• Backtests flatter almost every EA — fixed spread modelling and zero slippage hide real-world degradation
• No market context: an EA won't stand aside for FOMC unless you code the filter yourself
• Grid and martingale logic can breach a daily loss limit or max DD in one session
• Technical dependencies — VPS downtime, terminal updates, broker symbol suffixes and disconnects all break automation
• Buying a black-box EA means you can't diagnose or adapt it when the regime changes

Frequently Asked Questions