What RSI settings should an EA use?

The Builder's RSI block defaults to a period of 14 on the close price of the chart timeframe, and the shipped reversal template buys below 30 and sells above 70. Those are the conventional values, not measured ones — and it matters that the recorded parameter search on this template moved the stop and the target only, across 12 combinations and 3 windows, and never varied the period or the levels. So there is no evidence here that 80/20 beats 70/30. Treat any level you pick as a hypothesis you still owe a test, and change one value at a time.

Does the RSI 30/70 reversal strategy actually work?

Not as it ships here. Measured on USDJPY M5 over a year, the Builder's rsi_reversal template closes at a profit factor of 0.94 across 682 trades — a small, steady loss. Its recorded search across 12 stop and target combinations returned no_robust_edge, and the best result it found anywhere was a profit factor of 0.91, still below break-even. It is one of only two searched templates in the gallery that never won even in-sample. That does not make RSI useless; it makes an unfiltered 30/70 reversal a starting point rather than a strategy.

Should I add a trend filter to an RSI EA?

The measured evidence points that way, though it is indirect. In the mean-reversion neighbourhood of the gallery — eight templates all measured under one identical setup — the only one that closed above a profit factor of 1.0 was ma_distance_pullback, a six-node design that buys a pullback while the trend is still intact, at 1.10 across 159 trades. Every pure reversal in that group, RSI included, closed below 1.0. The catch is that it also took the fewest trades by a wide margin, so it bought its result with selectivity, and a smaller sample is a weaker claim.

Can I run the same RSI EA on more than one currency pair?

You can, and the Builder ships a template that does exactly that — but its record is the strongest warning on this page. dualpair_rsi_basket fires the same reversion logic on two pairs from one design, and under identical test conditions its drawdown was 36.38% against 0.97% for the single-pair template. That is roughly 37 times deeper for a profit factor that got worse, not better, because mean-reversion entries on correlated pairs tend to fire together and stack while the move keeps going against them. Add symbols only with a cap on total open exposure.

How to Build an RSI EA in MT5 Without Coding (and What the Shipped Template Actually Scores)

You build an RSI Expert Advisor in MT5 without coding by placing four kinds of block on a canvas. A Start node, an RSI indicator, a condition that compares the RSI value to a level, and an order block. The server then generates and compiles the MQL5 for you. It takes about fifteen minutes, and no MQL5 syntax at any point.

The harder question is what you have when you are done. This guide builds the EA and then reads the Builder’s own record of the same design, which is less flattering than any tutorial will tell you.

What you’ll build

A single-position EA on one pair that buys when RSI drops below an oversold level and sells when it rises above an overbought one, with a fixed stop and target, compiled to a real .ex5 you can drop into MetaTrader 5. Eight blocks. No code.

That is the mechanical half, and it is genuinely easy. The rest of this page is the half that decides whether the file is worth running.

Before you start: what RSI measures, and what it does not

You need a free account with a verified email address (designing is free and needs no sign-in; compiling to .ex5 does), about fifteen minutes, and a MetaTrader 5 terminal to test the result in.

RSI reports where the last n closes sit between their own recent gains and losses, on a 0–100 scale. In this recipe it is doing exactly one job: saying that price has moved far and fast in one direction relative to its own recent behaviour. What it is not saying is that the move is over. That distinction is the whole risk of the strategy — an oversold reading in a downtrend is a description of the downtrend, not a signal against it.

The Builder’s RSI block exposes these settings, and the defaults are the conventional ones:

Block setting	Default	What it controls
`period`	14	How many bars the calculation looks back. Range 1–999
`applied_price`	`PRICE_CLOSE`	Which price of each bar feeds the calculation
`shift`	0	Which bar to read — 0 is the forming bar, 1 the last closed one
`timeframe`	current chart	Any of M1 through MN1, read independently of the chart
`symbol`	current chart	Blank means the chart’s own symbol

The timeframe row is the one worth noticing early. The block can read RSI from a higher timeframe than the chart it trades on. That is how a multi-timeframe filter gets built without writing code.

Step-by-step: build the RSI EA in the Builder

Open the EA Builder and either load the rsi_reversal template from the gallery or start from a blank canvas. Loading the template is faster; building it yourself is how you learn which wire does what.
Place a Start node from the Utility palette. A flow without one fails validation with EAB-E202 before it will generate anything.
Add the RSI indicator block and set period. Leave it at 14 for the first build so your result is comparable to the baseline below, and set shift to 1 if you want the rule evaluated on closed bars only.
Add a Compare condition and wire the RSI block into it. This step is not optional plumbing: wiring an indicator straight into an order block is rejected with EAB-E205 — “insert a condition node (Compare/Cross) in between” — precisely because a raw indicator value is not a decision.
Set the comparison. Buy side: RSI less than 30. Sell side: RSI greater than 70. The Compare block’s own default operator is >, so the buy branch is the one you have to remember to flip.
Wire each condition into its own Open Buy / Open Sell action block.
Set the risk in the Settings tab — Money Management and Exit Strategy. The shared defaults are a 50-pip stop-loss, a 100-pip target and a fixed 0.01 lot. A flow with no valid stop raises a CRITICAL preflight diagnosis and makes you confirm before it will build.
Press Measure to run the built-in approximate backtest, and read it against the bar the panel prints: profit factor at least 1.2, at least 60 trades, drawdown no worse than 20%.
Press Compile .ex5, move the downloaded file into MQL5/Experts via File > Open Data Folder, and refresh the Navigator. Then test it properly in the Strategy Tester before it goes anywhere near an account.

Start node
RSI indicator, period 14
Compare — RSI below the oversold level
Compare — RSI above the overbought level
Open Buy
Open Sell
Stop-loss and take-profit
Compile to .ex5

The eight-block RSI reversal chain, from Start node to compiled file

Two blocks that beginners reach for, and should not, yet.

Cross asks whether one line crossed another on this bar. That is a stricter and different question from “is below”. Used here, it would fire only on the bar RSI dips through 30 — not on every bar it stays there.

The palette also carries an RSI Divergence condition block. It detects price making a new extreme while RSI does not, and defaults to level 30, pivot_bars 70 and near_band 10. It needs an RSI block with shift set to 1 wired into it. That is a genuinely different strategy, not a tuning of this one.

RSI entry, exit and parameter settings

Here is every value this recipe exposes, and what moving it actually does.

Setting	Template value	What moving it does
RSI `period`	14	Shorter reacts faster and fires far more often; longer smooths and fires rarely. It changes the trade count more than the win rate
Oversold level	30	Lower is stricter — fewer, deeper entries. The template’s identity lives here
Overbought level	70	Mirror of the above on the sell side
`shift`	0	Set to 1 to decide on closed bars only. On M5 this is the difference between a handful of decisions an hour and hundreds
Stop-loss	50 pips (shared default)	The searched range was 20–50
Take-profit	100 pips (shared default)	The searched range was 20–100
Lot size	0.01 fixed	Not a strategy setting; it scales whatever the strategy does, including the losses

The sensitivity question everyone asks, answered honestly. “Should I widen 30/70 to 20/80?” is the first thing every RSI tutorial suggests, and this page cannot tell you, because the recorded search on this template never varied the levels. What it varied was the stop and the target — SL20-50 × TP20-100, 12 combinations across three windows on H1 — and the verdict was no_robust_edge with a best in-sample profit factor of 0.91. So the levels remain an untested hypothesis here, and the settings that were tested could not lift the design to break-even from any direction. Anyone quoting a “best” RSI level to you is quoting a belief.

Backtest the RSI EA

This is the Builder’s own measured record for this template, not a re-run. Every template in the comparison below shares the single set of conditions in the table that follows it. How we test states the standard those records are held to.

`rsi_reversal` baseline	Value
Profit factor	0.94
Trades	682
Max drawdown	-0.97%
Net	-51.37
Flagged weak by the manifest	yes
Listed as recommended in the gallery	yes
Clears the Measure panel’s own pass bar	no

Test conditions
Experiment ID	EXP-RSI-RECIPE-BASELINE-001
Parent experiment	EXP-BUILDER-TEMPLATE-BASELINE-001
Symbol / timeframe	USDJPYm M5
Period	2025-06-01 – 2026-06-09
Model	M1 OHLC (Model=1)
Deposit	10,000
Environment	Exness MT5
Baseline measured	2026-06-18
Last verified	2026-08-25

0.94Profit factor of the shipped RSI reversal template at stock settings

0.91Best profit factor its recorded parameter search found anywhere — still a loss

37xHow much deeper the drawdown got when the same RSI rule was run on two pairs

Read that table with the last two rows together. The gallery marks this template recommended and the manifest marks it weak, and both flags are correct about different things: it is a clean, well-formed example of the pattern, and it lost money. Across the 53 templates carrying a measured baseline, 40 are marked recommended and all 53 are flagged weak. “Recommended” is a statement about the wiring, never about the edge.

Where RSI sits among its neighbours. These eight designs were all measured under the one conditions block above, which is the only reason it is fair to line them up:

Template	Blocks	Profit factor	Drawdown	Trades
`ma_distance_pullback`	6	1.10	-0.26%	159
`envelope_reversal`	10	0.95	-1.16%	989
`cci_level`	7	0.95	-0.94%	789
`rsi_reversal`	8	0.94	-0.97%	682
`bb_bounce`	9	0.92	-0.98%	616
`stoch_bb`	14	0.90	-1.41%	888
`hl_range_reversal`	8	0.90	-1.33%	573
`mfi_stddev_revert`	12	0.80	-1.66%	362

Exactly one cleared 1.0, and it is the one that does not fight the trend. ma_distance_pullback buys a pullback while the trend is still intact. That is mean reversion inside a direction rather than against it. It is also the smallest design in the group, at six blocks. Every pure “price went too far, so reverse” template, RSI included, closed below break-even.

That is the strongest argument on this page for putting a trend filter in front of an RSI entry. Its limit is worth stating plainly, though. It took 159 trades against RSI’s 682, so it bought that result with selectivity — and a quarter of the sample is a weaker claim, not a stronger one.

Combining a losing rule with more rules did not fix it — it multiplied it

The gallery contains two compositions that include this exact RSI template, measured the same way. Neither rescued it, and both made the ride much worse:

portfolio_mean_reversion (RSI + Bollinger bounce + envelope reversal): profit factor 0.92, drawdown -3.51% — 3.62× the single rule’s — across 2,287 trades.
portfolio_trend_revert_breakout (MA crossover + RSI + breakout): profit factor 0.93, drawdown -2.48%, 2.56× deeper, across 1,343 trades.
dualpair_rsi_basket, the same reversion logic fired on two pairs from one design: profit factor 0.87, drawdown -36.38% against 0.97%, and a net of -3,433.95 across 12,953 trades.

The profit factor barely moved in any of them — 0.94 to 0.92, 0.93, 0.87 — while the drawdown went up by multiples. Diversification spreads a positive expectancy; it does not create one. Applied to a rule that loses slowly, it just makes it lose faster, and the two-pair version is a demonstration of what correlated mean-reversion entries do when they all fire into the same move. Every figure here is a backtest on one year of one symbol, and none of it forecasts anything.

Common mistakes to avoid

Reading “oversold” as “about to turn.” This is the failure mode specific to this indicator, and it is why the pure reversal templates cluster below 1.0. In a sustained move, RSI sits pinned at an extreme for a long time and the rule fires again and again into a trend that is not finished. The stop is what ends each of those trades, which is why widening it makes this particular design worse rather than safer.

Compiling the template and calling it done. It carries a recommended badge and a losing record simultaneously. The badge is about the wiring.

Tuning until the backtest looks good. The recorded search here already did that honestly — 12 combinations, three windows — and reported no_robust_edge. If your own tuning produces a winner where a systematic search did not, suspect the tuning before you believe the winner. That is what overfitting looks like from the inside.

Adding pairs to smooth the equity curve. The two-pair basket is in the table above: 37 times the drawdown of the single-pair version.

Trusting an M5 result built on interpolated bars. The baseline uses the M1 OHLC model, so intra-bar movement is simulated rather than real. For a rule that trades inside the day, that is the coarse end of trustworthy — re-run anything promising on real tick data in MT5’s own tester.

Running it beside another EA without separating them. Each EA on one account needs its own magic number, or one EA’s exit logic closes the other’s positions.

Next steps and variations

When not to use RSI at all. In a strongly trending market, an unfiltered RSI reversal is a machine for selling strength and buying weakness. If you cannot state a condition that switches the rule off during trends, the honest options are to use a trend filter or to use a different idea — mean reversion explains which market conditions the family needs, and what it costs when they are absent.

The variations worth trying, in order of evidence. Add a higher-timeframe direction filter first, since the one design in the neighbourhood that cleared 1.0 is the one that requires the trend intact — the RSI block’s own timeframe setting does this without any code. Then try the RSI Divergence condition block, which asks a different and stricter question than a level cross. Only after that touch the levels, and change one at a time.

Then test it properly. The Measure panel is a filter, not a verdict: it is bar-level and approximate. How to run a backtest in MT5 walks a real run end to end, including which of the report’s two drawdown numbers to believe, and the rest of the backtesting guides cover the optimisation and walk-forward steps that come after. How to build a MACD EA is the sibling recipe with the opposite failure mode — its search did find a winning setting out of sample, which is a more dangerous result than never finding one. How to create an MT5 EA without coding is the wider tour of the block vocabulary if this recipe was your first flow.

And if you want to see what a documented record looks like, the published EAs each carry a complete closed-trade list and a run manifest. Comparing one of those against the table above is the fastest way to calibrate what “measured” should mean before you trust any strategy, including your own.