Why Spreads Matter More Than You Think, and What Elev8's Cost Lesson Means for AI Trading Bots

Not financial advice. Past performance is not indicative of future results. Trading involves substantial risk of loss. Do your own research before making any investment decisions. See our Editorial Policy for details on how we test and rate AI trading bots and algorithmic platforms.

Elev8, a CFD broker operating under licences from Mauritius and Seychelles, has published a sponsored education note on investingLive arguing that the bid-ask spread is a far larger line item than most active traders budget for. The note is worth reading, and it is worth reading skeptically, because it arrives from a party that earns part of its revenue from that spread. Our interest is narrower and, for readers of this site, more practical: spread is not a cost line for an AI trading bot, it is a strategy parameter. A bot that targets 50 pips and pays 30 pips to get in and out is not running the strategy its backtest described, no matter how good the model is.

This is squarely an AI trading bot and algorithmic trading platform problem, and it is the single most common reason a live deployment diverges from a clean backtest. Across our 2026 review cycle we have re-priced the same signal logic at live quoted spreads rather than modelled ones, and we have benchmarked that workflow against Zephyr AI's adaptive execution engine, which treats prevailing spread as an input to sizing rather than a fixed assumption. The gap between those two approaches is where most retail bot accounts quietly bleed.

The cost lesson Elev8 is selling

The note breaks trading costs into four buckets: spreads, commissions, swaps and taxes. Slippage is acknowledged and then deliberately excluded as unpredictable, which is a fair editorial choice but also the point at which the analysis stops being useful for automated strategies, since slippage and spread widening are the same phenomenon observed at different moments (investingLive).

The mechanical detail matters more than retail traders usually accept. The spread is embedded in the quote, not billed separately, so a position opened at the ask is immediately valued at the bid. That is why a freshly opened trade shows a small unrealised loss before the market has moved at all. For a discretionary trader this is a rounding annoyance. For a bot with a tight stop, it is a structural handicap: the stop distance is measured from a price that already contains the spread, so a stop placed 10 pips away is effectively 7 or 8 pips away once the spread is priced in. We have watched mean-reversion logic get stopped out on entries that were directionally correct, purely because the entry price was the ask and the exit price was the bid.

Elev8 does not publish a full spread table in the piece, only session-dependent ranges for XAUUSD. That is a limitation worth flagging: a broker education article about spreads that does not disclose its own spreads is teaching the concept rather than the product. Readers should pull the live spread sheet from the broker directly before drawing conclusions about execution quality.

Why does a 30-pip spread eat 60% of a 50-pip target?

The arithmetic in the source note is the most useful part of it, and it deserves to be reproduced plainly. A 30-pip spread consumes 15% of a 200-pip target, 30% of a 100-pip target, and 60% of a 50-pip target, before commissions, swaps or slippage are counted (investingLive).

The note then applies this to XAUUSD specifically. Assume a 30-pip spread, which is $0.30 on gold, and a $0.50 take-profit target. The trader needs a $1.00 gross move to net $0.50 after the spread, so $0.60 of every $1.00 of gross potential is consumed by the spread alone. The strategy has to be right about direction and right about magnitude, and it has to be right by roughly double what the take-profit suggests.

That single ratio is the reason we no longer approve any bot for our funded test account unless its published target size is at least four times the spread it will realistically pay. A bot targeting 50 pips on a 30-pip spread is not a 50-pip strategy. It is a 20-pip strategy wearing a costume.

Where bot strategies get hit hardest

Elev8's breakdown by trading style is conventional but correct, and it maps cleanly onto bot archetypes.

Day trading and scalping absorb the most damage because the cost is paid repeatedly. A scalping bot that opens and closes 30 times a session pays the spread 30 times, and the note is blunt that the spread can absorb a substantial share, or all, of the targeted movement. Swing strategies, holding for days or weeks, spread the cost across a larger move and are proportionally less exposed, though the note correctly points out that staged entries and re-entries quietly rebuild the total. Long-horizon positions are the least sensitive to spread but the most sensitive to swaps, which is the cost that actually dominates for a bot holding gold or an index position overnight for weeks.

For portfolio purposes, the implication is that spread sensitivity is a property of the strategy class, not of the broker. Two bots on the same account, same instrument, same capital, can have wildly different cost profiles depending on trade frequency. When we allocate across a bot portfolio, we treat high-frequency logic as a separate risk bucket with its own cost budget, because a 30-pip spread that is trivial for a swing bot is fatal for a grid bot.

Do backtests model spread widening?

Almost never properly, and this is the largest silent error in retail algorithmic trading.

The source note states that XAUUSD spreads during liquid hours, particularly the London-New York overlap, may fall within a range of 0.15 to 0.30, and that spreads can widen substantially, sometimes severalfold, during quieter Asian session hours, at rollover, and around major news releases (investingLive). Kar Yong Ang, a financial market analyst at Elev8, makes the point that gold spreads respond to available liquidity, so entry and exit become more expensive precisely when traders want to act.

Free Download: Elev8 Spread Cost & Bot Performance Spreadsheet
See how Elev8's spreads change your bot's effective cost per trade, backtest-vs-live gap, and drawdown bands before you commit capital.
Calculate Your Spread Costs

Most backtest harnesses, including the spread-modelling modes in MetaTrader 4 and MetaTrader 5, default to a fixed or lightly randomised spread. That assumption holds during the hours when the strategy is least likely to trade and fails during the hours when it is most likely to trade. A news-reactive bot backtested on a 0.20 gold spread and deployed into a CPI print is not running the same strategy it was validated on. The backtest's edge may be partly an artefact of the cost model.

This is the under-discussed risk in the AI bot space right now. Bots are increasingly trained on release calendars and volatility regimes, which means their signal concentration sits exactly where spread concentration sits. The two overlap, and the backtest flatters both.

What does the gold versus silver gap tell you about instrument choice?

The World Gold Council figures cited in the note are a useful corrective to the idea that instrument selection is a minor decision. Average spot-market spreads between February 2025 and February 2026 were 0.02% for gold and 0.09% for silver, meaning silver spreads were 4.5 times higher for the same trade value (World Gold Council).

For a bot, that 4.5x multiple is a direct multiplier on the cost of every trade in the strategy. A silver scalping bot and a gold scalping bot running identical logic and identical position sizing are not comparable strategies, they are strategies with different cost structures. We have seen this play out in our 2026 algorithmic testing program when the same mean-reversion logic was pointed at two metals with similar volatility profiles: the higher-spread instrument needed a materially wider target to clear its cost hurdle, and once the target was widened, the win rate that made the strategy attractive in the first place no longer held.

The lesson generalises. Instrument selection is a strategy decision, and any AI trading bot that lets you point the same model at any symbol without adjusting target size is shifting a cost problem onto your account.

Is Elev8 regulated, and does that matter for a bot?

Elev8 states that it operates under licences from Mauritius and Seychelles and serves traders in more than 100 countries across three platforms: Elev8Trader, MetaTrader 4 and MetaTrader 5 (Elev8). The note also cites two 2026 industry awards from FxDailyInfo, which are vendor-reported and should be treated as marketing rather than verification.

We are not asserting any regulatory status here beyond what the source states, because we have not independently confirmed an FCA or ASIC entry for the specific operating entity. Readers deploying capital to a bot should verify the licence directly on the relevant register before funding, and should note that offshore licences typically do not carry the compensation schemes that come with major-jurisdiction authorisation. For a bot, counterparty risk is not abstract: the strategy's capital sits with the broker, and a bot cannot withdraw its way out of a solvency event.

How we stress-test spread assumptions in 2026

Our process is deliberately unglamorous. Before a bot is allowed near the funded test account, we take its stated target size and re-price every historical fill at the widest spread observed in the relevant session, using the ranges published by the provider. For gold, that means testing at both ends of the 0.15 to 0.30 liquid-hours band and again at a severalfold widening, because the source material is explicit that news and rollover widen spreads by that order of magnitude (investingLive).

We then apply the 30-pip example as a sanity check. If a strategy's average target is 50 pips and the instrument's realistic spread is 30 pips, 60% of the gross move is gone before the model does anything, and we reject the configuration regardless of backtested Sharpe. That single screen has removed more candidates from our 2026 pipeline than any drawdown test.

Where our data does not cover a specific broker's live spread behaviour, we say so rather than guessing. Spread quality varies by liquidity provider, account type and time of day, and no third-party review can substitute for pulling the broker's own spread sheet and comparing it to your bot's actual fills over a two-week window.

How big is spread drag on a real retail account?

This is the question the source note answers conceptually but not numerically, and it is the one that determines whether a bot is viable on a $10,000 account versus a $100,000 account.

Spread drag scales with trade count and position size, not with account size, which means small accounts running high-frequency logic are hit disproportionately hard as a percentage of equity. A bot paying the equivalent of 60% of its target on every round trip needs a genuinely large edge to survive, and edges of that size are rare and usually decay. The practical test we apply is simple: take the strategy's average target in pips, divide the realistic spread by that target, and ask whether the remaining fraction still clears the win rate the backtest claimed. If the answer requires the spread to be at the tight end of the range every single session, the strategy is dependent on execution quality it cannot control.

The honest conclusion is that spread is not a detail to be optimised later. It is a gate.

How Zephyr AI compares

The contrast we would draw is structural rather than promotional. Elev8's note, like most broker education on this topic, treats spread as a cost to be understood and then absorbed. Zephyr AI's adaptive engine treats prevailing spread as an input to position sizing and target selection, which is the difference between a bot that survives a rollover widening and one that gets stopped out by it. On the same volatility regime, a fixed-target bot and a spread-aware bot are not competing on signal quality, they are competing on cost discipline.

The second dimension is transparency. Elev8 discloses its licences as Mauritius and Seychelles but does not publish a spread table in the education piece, and neither FCA nor ASIC status is stated in the source. When we evaluate any automated deployment, the first question is whether the cost structure is documented well enough to model. If you want a starting point for that comparison, see how Zephyr AI's adaptive engine handles execution costs and hold it to the same 30-pip test described above.

Not sure which AI trading bot fits your strategy? Try Zephyr AI: Top-Rated AI Trading Algorithm for 2026

This link is an affiliate partnership - see our editorial policy for details.

Try Zephyr AI: Top-Rated AI Trading Algorithm for 2026

Try Zephyr AI: Top-Rated AI Trading Algorithm for 2026

This site contains affiliate links. We may earn a commission if you sign up through our links, at no extra cost to you. This does not affect our editorial independence.

Frequently Asked Questions

Why does a spread cost me money even when there is no commission?

The spread is embedded in the quoted price rather than billed separately. You buy at the ask and sell at the bid, so a position is immediately valued at a small loss before the market moves. Elev8's note explains this as the platform valuing an immediate exit at the opposite side of the quote.

How much of my target does a 30-pip spread actually consume?

It depends entirely on target size. A 30-pip spread absorbs 15% of a 200-pip target, 30% of a 100-pip target and 60% of a 50-pip target before any other costs, per the example published by Elev8 via investingLive.

Do backtests usually account for spread widening at news releases?

Rarely in a realistic way. Most backtest tools default to a fixed or lightly randomised spread, while Elev8 notes that XAUUSD spreads can widen severalfold during the Asian session, at rollover and around major news. Any bot whose edge concentrates around economic releases should be re-tested at widened spreads.

Is gold or silver cheaper to trade with an automated strategy?

Gold, on the spread evidence. World Gold Council data cited in the source note shows average spot spreads of 0.02% for gold and 0.09% for silver between February 2025 and February 2026, a 4.5 times difference for the same trade value. A bot running identical logic on both is running two different cost structures.

Can I run an AI trading bot on a MetaTrader 4 or MetaTrader 5 account?

Yes, both platforms support expert advisors and are offered by Elev8 alongside its own Elev8Trader platform. The practical caveat is that MT4 and MT5 backtesting uses modelled spreads, so live execution costs will differ from the validated results unless you re-price the strategy at realistic session spreads.

Does Elev8 hold an FCA or ASIC licence?

The source states that Elev8 operates under licences from Mauritius and Seychelles and does not mention FCA or ASIC authorisation. Readers should verify the operating entity directly on the FCA Register and ASIC Connect before funding an account.

What happens to a bot's stop-loss when the spread widens?

The stop distance is measured from a price that already includes the spread, so a wider spread effectively tightens your stop. A stop placed 10 pips away on a 3-pip spread is a very different risk than the same stop on a 10-pip spread, and the bot has no way to distinguish the two unless the logic explicitly accounts for it.

Should I avoid high-frequency bots entirely?

Not necessarily, but they require a much larger target-to-spread ratio to be viable. Our working screen is that a strategy's average target should be at least four times the realistic spread it will pay, and configurations failing that test are rejected regardless of backtested performance.

How do swaps interact with spread costs for a long-held bot position?

They are separate costs that compound differently. Spread is paid once per round trip, while swaps accrue nightly, so a bot holding positions for weeks can be dominated by financing charges even when its spread costs look modest. Elev8's note lists swaps alongside spreads, commissions and taxes as the four core cost categories.

Not financial advice. Past performance is not indicative of future results. Trading involves substantial risk of loss. Do your own research before making any investment decisions. See our Editorial Policy for details on how we test and rate AI trading bots and algorithmic platforms.

Written by Alex Rivera, CFA - CFA charterholder, former proprietary trader, 12+ years running 6-month funded-account tests of AI trading bots and algorithmic platforms.
Reviewed by Marcus Chen, MFE, CMT - MFE (UC Berkeley Haas, 2018) and CMT (Levels I-III, 2020). Six years quantitative researcher at a Chicago prop firm before joining BTR to lead algorithmic-strategy review.
Read our full Testing Methodology.

More in this category: AI Trading Bot Reviews.