Petko AleksandrovRisk warning. Leveraged products carry substantial risk to your capital. Selecting a suitable symbol reduces avoidable costs; it does not make any strategy profitable.
The Direct Answer
Picking what your algorithm trades involves more than finding a market that moves a lot.
Three trading tests decide it. Are the spreads small relative to what your rules expect to make per trade? Do the liquid hours overlap with when your system actually operates? Are the overnight financing charges tolerable given how long positions stay open?
All three need modelling inside the historical test using your own broker’s trading conditions, not generic assumptions. Trading strategies clearing a small edge on paper frequently disappear once real execution costs arrive, and the symbol you chose determines how large those costs are.
The 2022 version of this page catalogued instruments by category: shares, ETFs, futures, options, metals, and contracts for difference. Useful background, wrong question. What follows is the selection framework instead, with the taxonomy condensed into one comparison section near the end.
Why Spreads Decide More Than Volatility
The spread, meaning the difference between bid and ask, is an immediate cost, paid on every entry, before price has done anything at all.
New algorithmic trading enthusiasts fixate on movement. Volatile symbols produce large candles and look full of opportunity, which is true and incomplete, because the instruments producing the biggest moves frequently charge the widest spreads to participate in them.
What matters is the ratio between the two. Here is the calculation that should govern the decision:
| Metric | Example |
| Expected gross gain per trade | 8 pips |
| Median spreads paid | 1.2 pips |
| Commission expressed in pips | 0.6 pips |
| Expected slippage | 0.3 pips |
| Total estimated trading cost | 2.1 pips |
| Cost as a share of gross gain | 26.25% |
Twenty-six percent of the edge consumed before anything goes wrong. That’s survivable on a strategy with a genuine advantage and fatal on a marginal one, which describes most generated trading strategies I have seen.
What to record before approving any symbol:
- Typical spreads during normal conditions, measured yourself rather than taken from marketing
- The median rather than the advertised minimum, since minimums appear rarely
- Maximum widening during volatile periods
- Behaviour at daily rollover, which is frequently the worst moment of the day
- Weekend reopening, particularly for anything trading Sunday evening
- Widening around scheduled announcements
- Whether the quote is static or variable
- The units used, since points and pips differ by a factor of ten on five-digit quotes
- Whether your historical trading test models any of this realistically
That last point catches almost everyone. Testing with a fixed one-pip assumption on a symbol whose real costs triple during the hours your system operates produces results describing a market that doesn’t exist.
Spread Sensitivity by Approach
Not every strategy suffers equally.
| Approach | Sensitivity | Why |
| Scalping | Very high | Costs can exceed the entire expected move |
| Intraday mean reversion | High | Small targets, frequent entries |
| Breakout systems | High during volatility | Costs widen precisely when signals fire |
| Swing trading | Moderate | Larger targets absorb costs more comfortably |
| Long-horizon trend following | Lower per trade | Though financing charges grow instead |
A useful rule of thumb: if your average expected gain is under twenty times the typical cost, the symbol probably isn’t suitable for that approach. Under ten times, it almost certainly isn’t. Those thresholds are my own working preferences rather than researched standards, and I’d welcome being shown better ones.
Sessions Determine When Your System Can Work
Trading conditions are not uniform across the day, and systems running continuously experience every part of that variation.
| Instrument type | Session behaviour |
| Major currency pairs | Trading runs through the business week, though liquidity swings enormously by hour |
| Index CFDs | Usually follow exchange hours plus broker-specific extended periods |
| Futures | Exchange-defined sessions with maintenance breaks |
| Shares | Concentrated in official exchange hours |
| Cryptocurrency | Continuous, but liquidity and volatility still vary by hour and weekday |
| Metals | Extended hours with session-dependent conditions |
Verify exact trading hours with your own broker or exchange. Published schedules vary, and extended sessions differ between firms offering apparently identical symbols.
Why this affects automated systems specifically:
- Breakout logic clusters around opens, so a system running only during quiet hours may barely trigger.
- Mean-reversion approaches often struggle during the London and New York overlap, where directional moves persist.
- Thin sessions produce wider quotes and more slippage on exactly the same rules.
- Orders near maintenance windows may be rejected, delayed, or repriced, halting trading briefly.
- Historical bars may carry a different timezone from your live server, shifting every session filter you wrote.
That final point deserves emphasis. Testing with data stamped in one timezone while running live on a server using another means your carefully constructed hour filters operate at the wrong hours. I have watched this happen to several people, and the symptom is a system that tested beautifully and behaves nothing like it should.
Session settings to establish:
- Broker server time, and its offset from UTC.
- Whether the server observes daylight saving, and on which schedule.
- Rollover time and how long conditions deteriorate around it.
- Friday close and Sunday reopening times.
- Exchange holidays affecting the underlying.
- Any daily maintenance break.
- Which trading hours your rules should be permitted to operate.
Swaps: The Cost Nobody Models
Overnight financing gets applied when a leveraged position stays open past rollover. Charges differ between long and short sides, and one weekday typically carries a triple charge to cover the weekend.
Most retail traders find this out from a statement rather than from a trading plan.
| Item | Example |
| Average holding period | 4 nights |
| Average financing per night | −$3.50 |
| Total per trade | −$14.00 |
| Average gross gain per trade | $55.00 |
| Financing as a share of gross gain | 25.45% |
Another quarter of the edge, gone, and this one never appears in a simple historical test because most testing tools either ignore financing entirely or apply a static figure that bears no relationship to current rates.
What to verify per symbol:
- Long financing rate
- Short financing rate
- Calculation method, since some firms charge in points and others in account currency
- Which weekday carries the triple charge
- Exact rollover time
- Contract size used in the calculation
- Currency conversion where your balance differs
- Whether positive rates are stable, because they change without notice
Positive financing deserves particular scepticism. Strategies built specifically to collect it, holding one side of a rate differential, work until the differential narrows or the broker adjusts its rates, at which point the entire premise evaporates. I’ve seen people describe such approaches as low risk. They aren’t; the risk simply sits somewhere less visible than price movement.
Trading approaches where this matters most: anything holding positions across multiple nights, position-sizing rules that scale up over time, and systems trading currency pairs with wide interest rate differentials.
Where it barely matters: intraday trading systems closing before rollover, which sidestep the question entirely. That alone is an argument for intraday designs on symbols with punishing financing rates.
Contract Specifications Are Not Optional
A symbol name tells you almost nothing. Before any automated strategy touches a market, record the following from the specification window of the account you intend to use:
| Specification | Why it matters |
| Tick size | The minimum price increment |
| Tick value | What that increment is worth in money |
| Contract size | Determines your actual exposure |
| Minimum volume | The smallest position permitted |
| Volume step | Precision available for sizing |
| Maximum volume | Upper exposure limit |
| Stop level | Minimum distance for protective orders |
| Typical spreads | Immediate transaction cost |
| Commissions | Additional execution charge |
| Long and short financing | Overnight cost, both directions |
| Trading hours | When the system may operate |
| Margin requirement | Capital tied up per position |
| Symbol suffix | Required for correct operation |
That last row causes more failures than it should. A strategy coded for EURUSD attached to EURUSD.m may behave unpredictably, and one written for a broker using no suffix will not find its symbol at a firm that appends one.
Test with the same broker you intend to trade with. Specifications differ enough between firms that results from one rarely transfer cleanly to another.
Comparing Realistic Candidates
| Symbol | Spreads | Sessions | Financing | Typical problem for algorithmic trading |
| EURUSD | Tightest available during liquid hours | Weakest conditions near rollover | Relevant for multi-night holds | Small edges vanish after costs |
| GBPJPY | Consistently wider than majors | Strongly session-dependent movement | Can materially affect held positions | More movement, more execution risk |
| Gold | Widens sharply on news | Behaviour changes across regional sessions | Broker-dependent, sometimes heavy | Rollover and announcement volatility |
| Index CFD | Broker-specific schedule | Exchange open and close dominate | Frequently substantial | Gaps and session breaks |
| Cryptocurrency CFD | Variable, often wide | Continuous, with weekend quirks | Can be very high | Persistent volatility, broker differences |
| Futures | Tick-based cost structure | Exchange sessions and expiry dates | No retail-style financing, though roll effects apply | Contract rollover and liquidity migration |
Treat those columns as prompts for your own trading measurements rather than as findings. Actual figures come from your broker, and they change.
A Selection Scorecard
Score trading candidates before building anything on them.
| Criterion | Weight | What to measure |
| Cost efficiency | 30% | Median spreads relative to average expected gain |
| Session fit | 25% | Liquidity during your system’s operating hours |
| Financing burden | 20% | Overnight cost relative to expected return |
| Slippage | 10% | Gap between intended and executed prices |
| Data quality | 10% | Historical depth, missing bars, timestamp consistency |
| Contract stability | 5% | How often specifications change |
Anything scoring poorly on the first three probably cannot be rescued by strategy quality. A brilliant trading system on an unsuitable instrument still loses to costs.
Note what the scorecard deliberately excludes: popularity and volatility. Neither predicts whether a rule-based approach will work there, and both attract beginners for reasons unrelated to expected outcomes.
Testing With Realistic Assumptions
Every candidate under consideration needs its own historical trading run, configured with:
- Data sourced from your intended broker rather than a generic feed
- Variable spreads matching observed behaviour, not a flat figure
- Commissions applied per lot
- Slippage assumptions taken from demo observation
- Overnight financing where positions cross rollover
- Session filters using the correct server timezone
- Daylight saving transitions handled
- Weekend gaps preserved
- Exchange closures reflected
- Contract rollover effects for futures
Run identical trading strategies across several candidates using the same settings. The comparison tells you more than any individual result, because a strategy performing acceptably on three symbols and catastrophically on a fourth has told you something about the fourth.
How This Fits Your Wider Trading Process
Instrument selection sits early in an automated workflow, and getting it wrong invalidates everything downstream.
The usual sequence runs: define the rules, pick the market, gather trading data, run historical tests, stress the results, forward test on a demo account, then deploy small. Most people reverse the first two steps, choosing a market because it interests them and then hunting for rules that suit it. Either order can work, though starting from the rules and then asking which markets they suit produces fewer wasted weeks in my experience.
What changes when you approach it cost-first is the shape of the trading strategies you end up running. Tight-cost symbols make short-horizon approaches viable. Expensive symbols push you toward longer holds where each trade has room to absorb the charges, which in turn raises the financing question again. The constraints interact, and there’s no configuration that dodges all of them.
One further consideration for anyone running several trading systems at once. Each additional market adds monitoring burden, data storage, and another set of specifications to track when your broker changes something. Two well-understood markets frequently beat six half-understood ones, particularly early on, and the temptation to broaden coverage usually arrives before the ability to manage it does.
There’s also the question of what your trading platform supports. Not every market is available through every terminal, symbol naming differs, and some brokers restrict automated execution on specific instruments entirely. Check the account agreement rather than assuming, because discovering a restriction after building around a market wastes considerable effort.
Automation Changes the Cost Calculation
Manual traders absorb costs differently from software-driven systems, which is worth stating plainly.
A discretionary trader takes perhaps two or three positions weekly and pays those trading costs a handful of times. Automated trading changes that arithmetic entirely. Software frequently multiplies that by ten or twenty, since software has no reluctance about entering, no fatigue, and no opinion about whether today feels right. Higher frequency amplifies every cost factor described above.
Software also removes one advantage manual trading retains: the ability to notice that conditions look wrong and simply not participate. Software keeps trading through widening quotes, thin holiday sessions, and the fifteen minutes around a central bank announcement, unless somebody wrote a filter telling it not to. Those filters are where much of the practical work lives.
Common Mistakes
- Selecting by volatility alone: Movement without a favourable cost ratio is expensive movement.
- Trusting advertised minimums: Marketing quotes the best case; you pay the median.
- Ignoring financing on multi-night systems: A quarter of the edge, invisible until the statement arrives.
- Testing on one broker, deploying on another: Specifications differ enough to change outcomes.
- Forgetting the timezone: Session filters built on the wrong offset run at the wrong hours.
- Assuming crypto’s continuous market means uniform conditions: Weekend liquidity is thinner and costs reflect it.
- Copying a symbol because somebody else’s system uses it. Their costs are not your costs.
- Never remeasuring. Trading conditions change, and a symbol suitable two years ago may not be now.
Where the Asset Overview Fits
The original article’s taxonomy still has value as background, condensed here.
Shares allow trading during exchange hours, carry commissions rather than spreads at many brokers, and suit longer-horizon approaches. The earlier version described them as the safest instruments, which was unsupportable. Risk depends on the individual company, diversification, holding period, leverage, and concentration rather than on the category itself.
Contracts for difference offer access to many underlying markets with flexible sizing, which suits algorithmic execution, at the cost of financing charges and counterparty exposure to your broker. The earlier claim that this market is “not fully regulated” was too broad: treatment varies by jurisdiction, product, and legal entity, with some regions imposing strict rules and others very few.
Futures offer centralized pricing, transparent volume, and standardized specifications, with contract expiry and rollover to manage. Larger contract sizes raise the capital needed.
Options introduce time decay and volatility exposure that most retail trading software handles poorly. Possible, considerably harder.
Metals behave differently from currency pairs despite similar quoting, with sharper reactions to announcements. The original article described gold’s role in hedging currency devaluation more categorically than the evidence supports; for this purpose, what matters is measurable behaviour rather than macroeconomic narrative.
Cryptocurrency allows continuous trading with variable costs and, at many brokers, substantial financing charges.
One further correction: the earlier version ranked markets by popularity and implied that popularity drives volume in ways that affect you directly. Without current volume evidence, those comparisons weren’t supportable and have been removed.
Worked Example: Approving a Symbol
Suppose a trading strategy generating roughly forty trades monthly, holding an average of two nights, with an expected gross gain of 30 points per trade.
Step 1: Measure median spreads across your operating hours for a week. Say 1.8 points.
Step 2: Add commissions converted to points, perhaps 0.5, and observed slippage, say 0.4. Total cost 2.7 points, which is 9% of the expected gain. Acceptable.
Step 3: Check financing. Two nights at the quoted long rate, converted to points, comes to 1.9 per trade. Combined cost now 4.6 points, or roughly 15%.
Step 4: Confirm your operating hours sit inside the liquid session and avoid rollover. If they don’t, either adjust the hours or reconsider the symbol.
Step 5: Run the historical test with all of the above applied, then compare against the same rules on two alternative symbols.
Step 6: Score all three against the framework above and pick the best fit, not the one with the prettiest equity curve.
The whole process takes an afternoon per candidate. Skipping it costs considerably more later.
Frequently Asked Questions
Can one trading strategy work across several symbols?
Sometimes, and portability is a useful robustness signal, since rules working on three related markets are less likely to be fitted to one specific history. Parameters usually need adjusting, because typical ranges and cost structures differ. Be cautious about applying settings developed on a major currency pair to a metal or index, where movement characteristics differ substantially. Test each market independently before trading it, rather than assuming the original configuration carries across unchanged.
How do I measure spreads properly?
Record them yourself across at least a full week of trading, sampling every hour rather than glancing occasionally. Most platforms let you log this automatically, or you can attach a simple indicator that writes values to a file. What you want is a distribution: median, worst case, and behaviour by hour. Advertised figures describe optimal conditions during liquid hours, which is rarely when your system is placing orders.
Are fixed spreads better for automated systems?
They make trading cost modelling simpler and historical testing more reliable, which has real value. The trade-off is that guaranteed pricing usually sits higher than the variable average, so you pay a premium for predictability. Providers offering such quotes may also widen them or reject orders during volatile periods, which undermines the guarantee precisely when it matters. Neither structure is universally preferable; it depends on whether your rules need cost certainty or the lowest average.
What about symbols with no overnight charges?
Some accounts advertise swap-free structures, typically for religious compliance, and they usually replace financing with an administration fee after a holding period. Read the trading terms carefully, since the charge structure may be less favourable for longer holds despite appearing simpler. Also verify that the offering applies to every symbol you intend to trade, because coverage is frequently partial and the excluded instruments tend to be the expensive ones.
Does higher volatility mean better opportunity?
Not by itself. Volatility widens the potential gain and the potential loss simultaneously, while typically widening costs as well, so the net effect depends entirely on your rules. Some approaches genuinely need movement and underperform in quiet conditions. Others do better in ranges. Judge a market on the relationship between expected gain and total trading cost rather than on how dramatic its chart looks, which is where most beginners go wrong.
How often should I re-check these figures?
Quarterly at minimum, and after any broker announcement affecting trading costs. Costs drift, financing rates change with interest rates, and session schedules shift with daylight saving. A symbol approved eighteen months ago may no longer clear your threshold. Set a calendar reminder, since nothing prompts you otherwise and the deterioration happens gradually enough that live results decline before anyone works out why.
Should I trade the same symbol across multiple strategies?
Doing so concentrates exposure, which is the main trading concern. Several systems trading one market can all open positions simultaneously, producing far more risk than the individual sizing suggests. Correlated symbols create the same problem less visibly: three dollar-denominated positions represent one bet. Cap total exposure per underlying rather than per strategy, and check correlation between systems before running them together on anything.
Which market suits beginners starting with automation?
Major currency pairs, generally, because costs are lowest, liquidity is deepest, historical data is most available, and behaviour has been studied more thoroughly than anywhere else. Starting there removes several variables while you learn how your trading system behaves. Exotic pairs, thinly traded shares, and cryptocurrency all introduce cost and execution complications that obscure whether your rules are working, which is exactly what you need to observe first.
Final Checklist
Before any instrument enters live trading:
- Median spreads measured across your actual operating hours.
- Cost ratio calculated against expected gain per trade.
- Session schedule confirmed with your broker, including rollover and maintenance.
- Server timezone identified and filters adjusted accordingly.
- Financing rates recorded for both directions, plus the triple-charge weekday.
- Full contract specification saved, including suffix and stop level.
- Historical test run with realistic costs applied.
- Results compared against at least two alternative candidates.
- Scorecard completed.
- Review date set for three months out.
Disclosure: Educational content only, not a recommendation to trade any market. Broker links elsewhere on this site may carry commercial arrangements. Every market can produce gains or losses; suitability depends on your rules, costs, risk controls, and whether the market’s structure matches what your system needs.
