How to Compare Trading Performance Across Multiple Accounts
Compare trading performance across multiple accounts using R, costs, adherence, and drawdown, without letting a consolidated P&L hide a weak account.
To compare trading performance across multiple accounts, review each account on its own first and pool the results only afterward. Give every account its own record of purpose, rule set, cost structure, review period, and planned risk. Then express results in a unit that doesn’t depend on account size, such as each trade’s result after its direct transaction costs divided by the planned risk of that decision (R). Add accounts together only when the question is about the total: total money made, total costs paid, or total exposure open at the same time. A consolidated P&L answers “what did I make?” It cannot tell you which account, rule set, or decision process is working, and a large profitable account can hide a smaller account that is steadily losing.
This article covers how to compare accounts for one trader: a personal cash account, a second broker, an evaluation or funded prop account, or an account kept for a different style. It doesn’t cover choosing sync software, tracking a long-term investment portfolio, or how to calculate cash-flow-adjusted returns when deposits, withdrawals, or payouts occur. The general three-layer review (results, risk, and execution) belongs to the trading performance framework. This page covers what changes when that review runs across several accounts at once.
How do you compare trading accounts, step by step?
- Keep each account’s record separate.
- Define each account’s purpose, rules, risk unit, costs, and review period.
- Normalize trade-level results by planned risk (R) where planned risk was recorded before entry.
- Compare results and rule adherence within each account first.
- Consolidate only the figures where a total answers the question: total financial result, total costs, or exposure open at the same time.
Why can’t you just add the accounts together?
Summing accounts is fine for bookkeeping. It stops being useful once you start asking about performance, for three reasons.
- The biggest account decides the total. Dollar P&L grows with position size. When one account trades at two or three times the risk of another, the combined figure mostly reflects that one account. A small account can lose on most of its decisions without changing the total’s sign.
- The accounts often aren’t running the same process. A personal account, a prop evaluation with a provider’s loss limits, and a longer-horizon account can differ in setups, risk rules, sessions, and pressure. Pooling them averages across different processes and produces a number that describes none of them.
- A pooled result can reverse the account-level results. When accounts differ in size and in their mix of trades, a comparison that holds within every account can flip once the accounts are combined. This is Simpson’s paradox, and a worked example appears below.
The fix isn’t to avoid consolidation. It’s to keep the account-level record as the primary evidence and to treat any consolidated figure as a derived view built for a stated purpose.
What should you record for each account before comparing?
A comparison is only as good as the labels that let you tell the accounts apart. Record these fields once per account, and add a new version whenever they change:
| Field | What to record | Why the comparison needs it |
|---|---|---|
| Identity | Broker or provider, a stable internal label or masked account identifier (not the full account number), and simulated or live status | Stops executions from different accounts merging under one label |
| Purpose | Personal, evaluation, funded, style-specific, or test | Accounts with different jobs aren’t expected to behave the same way |
| Rule set | Your personal rules and, where one applies, the provider’s rules for that account, with the date each version took effect | The same result means different things under different constraints |
| Risk unit | Planned risk per decision in currency, and how it was set | Needed to convert results into a size-independent unit |
| Loss allowance | The loss the account can actually absorb: your own limit, or the provider’s limit where one applies | An account-size label may not match what you can actually lose |
| Cost structure | Commissions, exchange and data fees, platform charges, and evaluation or reset fees where relevant | Gross results aren’t comparable when costs differ |
| Review period | Start and end dates, sessions traded, and number of decisions | Accounts opened at different times cover different market conditions |
Two account-identity problems are worth checking before any analysis. Imported executions should keep the account they came from, and timestamps from brokers that report in different time zones need one normalized session convention. Trading journal data quality covers both checks. It also covers how to keep account-level charges separate from per-execution costs.
How do you normalize results across accounts of different sizes?
No single measure makes every account comparable. Each one answers a narrower question and has a known weak point.
| Measure | What it answers | Where it misleads |
|---|---|---|
| Account result in currency | How much each account made or lost after transaction costs and its account or program costs | Dominated by position size. Not a measure of process quality |
| R per decision (trade P&L after directly attributable transaction costs ÷ planned risk) | How much each decision earned relative to the risk you meant to take | Only as good as your planned-risk records. Breaks if planned risk wasn’t written down before the entry |
| Drawdown as a share of the loss allowance | How close each account came to the loss it can absorb | Depends on how the loss allowance is defined. A trailing provider limit and a fixed personal limit aren’t the same unit |
| Percentage return on account value | Growth of the account’s own capital | A simple balance change is distorted by deposits, withdrawals, and payouts unless a cash-flow-adjusted method is used. Misleading when the “account size” is a label rather than capital |
| Adherence rate | The share of decisions that matched the rules in force for that account | Only comparable when the rules are written precisely enough to classify each decision |
For most discretionary comparisons, R per decision and adherence rate are the two that travel best. R removes the size difference, and adherence separates “the rules didn’t work here” from “the rules weren’t followed here.” Keep the currency result beside them so the financial result stays visible. Just don’t use it to rank the accounts.
Keep two layers apart. Trade-level R is a trade’s result after the costs directly attributable to it, such as commissions and exchange fees, divided by its planned risk. Account economic result is trading P&L minus the account’s or program’s own costs, such as platform and market-data subscriptions, evaluation, reset, or activation fees, and shared operating costs. Don’t subtract account-level costs from individual trades and still call the result R, unless you’ve written down how those costs were allocated.
Percentage return needs extra care. A simple beginning-to-ending balance change mixes trading results with deposits, withdrawals, and payouts, so it’s unreliable for comparing accounts whenever any of those happened. A cash-flow-adjusted method can still produce a valid percentage return. As defined in the GIPS Standards, a time-weighted return negates the effects of external cash flows, and a money-weighted return reflects their timing and size. Which one fits depends on whether you’re judging the trading or what the account’s owner actually earned. Choosing and calculating either is a separate task, and until it’s done, don’t compare the accounts using simple percentage return. Even a correctly calculated return doesn’t fix the prop-account denominator problem. A provider’s advertised account size may be a notional buying-power label, not money you deposited, and the loss that ends the account may be a small fraction of that figure.
Worked example: a consolidated profit that hides a weak account
The numbers below are hypothetical and chosen to show the arithmetic. They don’t describe any real account or typical result.
A trader runs two accounts over the same eight-week period with the same setup list:
| Account A: personal | Account B: evaluation | Consolidated | |
|---|---|---|---|
| Decisions | 40 | 40 | 80 |
| Planned risk per decision | $250 | $100 | Mixed |
| Trading P&L after directly attributable transaction costs | +$3,000 | −$1,200 | +$1,800 |
| Net R | +12.0R | −12.0R | 0.0R |
| Average R per decision | +0.30R | −0.30R | 0.00R |
| Account and program costs (such as platform, data, or evaluation fees) | −$150 | −$250 | −$400 |
| Net account economic result | +$2,850 | −$1,450 | +$1,400 |
| Decisions matching the account’s rules | 36 of 40 | 26 of 40 | 62 of 80 |
R uses trading P&L only. The account and program costs affect each account’s economic result, not its R. The consolidated line says the trader made $1,800 from trading, or $1,400 after account costs. The account-level lines say something different. In risk units, account B gave back exactly what account A earned, and B’s decisions matched their rules far less often. The consolidated dollars look positive only because A trades at 2.5 times B’s risk per decision.
That finding doesn’t prove the evaluation’s rules caused the drift or that B’s setups have no edge. Forty decisions is a small sample; trading data statistical reliability covers how far a sample that size can be trusted. What it does justify is a specific next question: what is different about the decisions in account B? The answer might be the provider’s loss limit, a different session, pressure to hit a target, or a difference in execution. The consolidated number would never have raised that question.
When can a pooled number reverse the account-level result?
A second, separate hypothetical with two different accounts shows the stronger failure. The trader wants to know whether decisions that followed their rules perform better than decisions that didn’t:
| Aligned decisions | Deviated decisions | |
|---|---|---|
| Account C | 10 decisions, average +0.8R | 30 decisions, average +0.5R |
| Account D | 30 decisions, average −0.2R | 10 decisions, average −0.5R |
| Pooled | 40 decisions, average +0.05R | 40 decisions, average +0.25R |
Within each account, aligned decisions did better than deviated ones. Pooled, the deviated decisions appear to do better, because most of them happened in account C, where everything performed better. The pooled comparison mostly measures which account a decision came from, not whether it followed the rules.
This is Simpson’s paradox: an association that holds inside every group can weaken or reverse when the groups are combined. Simpson (1951) described it for contingency tables. The best-known case is Bickel, Hammel, and O’Connell (1975). In their data, Berkeley’s overall graduate admissions appeared to favor men, but that pattern did not hold department by department. The overall figure reflected which departments women and men applied to. Multi-account trading records are exposed to the same problem whenever accounts differ in size, market conditions, or trade mix. The safeguard is to make each comparison within an account first, and to pool only when the accounts are similar enough on the grouping factor to make the pooled number meaningful.
Are copied trades separate evidence?
No. When one decision is sent to several accounts, through a trade copier or by placing the same order in each account, that is one decision, not several. Counting it once per account inflates the number of decisions and makes a pattern look better supported than it is. For the evidence count, record the decision once and list the accounts it was executed in.
The risk works the other way. A decision copied to three accounts puts roughly three times the planned risk on the same outcome, even if each account looks well within its own limits. The consolidated view is the right place to see that total. Correlated event exposure explains how to measure exposure that looks spread out but depends on one driver. Copied positions are the simplest case of it.
Execution evidence is the one exception. If a copied order filled at different prices in different accounts, those fills are separate observations of execution quality, even though they share one decision. Keep the decision count and the fill count as separate fields.
Which costs belong to which account?
Compare accounts after each account’s own costs, not on a gross basis. Accounts at different brokers or providers can differ in commissions, exchange and data fees, platform subscriptions, and, for prop accounts, evaluation fees, reset fees, and payout splits. A gross comparison can rank accounts differently from the net result.
Keep costs at the level where they occur. Transaction costs belong to the account and execution that generated them, and they are the only costs that go into trade-level R. A platform subscription or data fee is an account-level cost, and one shared across several accounts is a cost of the whole operation. When you decide to spread a shared cost across accounts for analysis, write down how you allocated it and keep the allocated figure separate from what the source reported. Evaluation and reset fees count toward the result of the attempt that incurred them. Leaving them out makes a series of failed evaluations look cheaper than it was.
Don’t let closed or failed accounts disappear
The accounts that are easiest to forget are the ones that no longer exist: a failed evaluation, a closed broker account, an account abandoned after a bad month. A comparison that includes only open accounts compares survivors. Its average will look better than the trader’s actual record for the same reason that Brown, Goetzmann, Ibbotson, and Ross (1992) found survivorship can bias performance studies that keep only the managers still in operation.
Professional performance reporting builds rules against this. The GIPS Standards for Firms (2020) require firms to include every actual, fee-paying, discretionary segregated account in at least one composite. They require terminated portfolios to stay in the composite’s history up to their last full measurement period, and they don’t allow portfolios to be moved between composites without a documented reason. The standards state that these requirements exist to stop firms from cherry-picking their best performance. The GIPS standards govern investment firms reporting to clients. They don’t govern an individual trader’s journal, but they provide a useful principle against selecting which results to keep. In practice, keep a closed or failed account’s record in the comparison for the periods it was active. Document why it was closed or reclassified. And don’t remove an account, or move it to a different group, after you’ve seen how it performed.
What is a consolidated view actually good for?
A consolidated view is the right tool for three questions:
- Total financial result: how much the whole trading operation made or lost after all costs, including shared and account-level costs.
- Total concurrent exposure: how much risk is open at the same moment across every account, which no single account’s record shows.
- Total behavior under one rule: whether a personal rule that is meant to apply everywhere, such as a daily loss limit across all your accounts, was followed across all of them.
For process questions, stay at the account level or group only accounts that share the same rules, setups, and conditions. Examples are which rules work, where drift appears, and whether an evaluation changes behavior. For prop accounts in particular, keep the stage and attempt IDs so a comparison doesn’t mix an evaluation attempt with a funded period. The prop firm trading journal covers that stage-by-stage record.
A multi-account comparison scorecard
One row per account, plus a consolidated row that only includes totals, is usually enough:
| Field | Per account | Consolidated row |
|---|---|---|
| Period and decisions | Dates, sessions, independent decisions, and executions | Total decisions, with copied decisions counted once |
| Trading P&L and account result | Trading P&L after transaction costs, then the account result after that account’s own account and program costs | Sums of both, plus shared costs not assigned to any account |
| Net R and average R | Sum and mean of that account’s trade-level R, each calculated against that decision’s planned risk | Sum and mean of decision-level R across independent decisions, with a copied decision counted once. Different dollar risk per account is fine when R is defined consistently. Never divide pooled dollar P&L by a blended risk figure |
| Worst drawdown | As a share of that account’s loss allowance | Peak combined exposure open at the same time |
| Adherence | Aligned, deviated, and unclassified counts against that account’s rules | Adherence to any rule meant to apply across all accounts |
| Status | Open, closed, or failed, with the reason | All accounts, including closed ones |
| Review decision | Keep, investigate, or revise for that account | Any change to the overall operation |
The final row matters most. A comparison should end in a small decision about a specific account, such as “investigate account B’s after-loss entries” or “keep collecting samples in both.” A new rule based on a blended number is usually the wrong output.
Where Costante fits
Costante doesn’t connect to brokers, prop firms, or exchanges. It doesn’t import or sync account data, and it doesn’t calculate consolidated P&L, account returns, or combined exposure. Account balances, fills, costs, and provider rule status stay in your broker and provider records and in your own performance journal.
Costante covers the behavioral side of each decision. It supports session planning, self-defined behavioral guardrails, pre-trade and in-session checks, low-friction logging of trades and behavioral decisions, and structured review that keeps a good outcome separate from a good process. That is the record behind the adherence column above. When an account comparison raises a question like “why do decisions in this account drift more often?”, that record is where the answer is likely to be. Costante doesn’t guarantee discipline, account results, or evaluation outcomes, and the trader remains responsible for the method, the risk, and each order.
Frequently asked questions
How do I compare two trading accounts of different sizes?
Divide each decision’s P&L, after its direct transaction costs, by the risk you planned for it, then compare the average R per decision and the adherence rate for each account. Keep net currency P&L next to them for context, but don’t use it to rank the accounts, because it grows with position size.
Should I combine my trading accounts into one performance report?
Keep a report for each account first. Add a consolidated row for total P&L after all costs, total exposure open at the same moment, and any rule meant to apply across all accounts. Don’t use the consolidated figure to judge which account or process is working.
Do copied trades count as separate trades?
For the evidence count, no: one decision copied into several accounts is one decision. For risk, the copies add up, because each account carries its own exposure to the same outcome. For execution quality, each fill is a separate observation.
Should a failed prop evaluation stay in my performance comparison?
Yes. Keep it for the period it was active, including its fees, and mark it as closed with the reason. Removing failed or closed accounts leaves only the survivors in the comparison and makes the overall record look better than it was.
Can I compare percentage returns across my accounts?
Not with a simple balance change if deposits, withdrawals, or payouts happened in the period. A cash-flow-adjusted return, either time-weighted or money-weighted, can handle those flows, but until you’ve calculated one, don’t compare the accounts on percentage return. Even then, a prop account’s advertised size may be notional rather than money at risk. For comparing decisions, R per decision and drawdown as a share of each account’s loss allowance are usually more reliable.
Sources
- Simpson, E. H. (1951). The Interpretation of Interaction in Contingency Tables. Journal of the Royal Statistical Society, Series B, 13(2), 238–241.
- Bickel, P. J., Hammel, E. A., & O’Connell, J. W. (1975). Sex Bias in Graduate Admissions: Data from Berkeley. Science, 187(4175), 398–404.
- Brown, S. J., Goetzmann, W., Ibbotson, R. G., & Ross, S. A. (1992). Survivorship Bias in Performance Studies. Review of Financial Studies, 5(4), 553–580.
- CFA Institute. Global Investment Performance Standards (GIPS®) for Firms, 2020. Accessed September 23, 2026. Supports the composite-inclusion, terminated-portfolio, and portfolio-movement provisions (Section 3.A), the statement that these requirements prevent cherry-picking, and the glossary definitions of time-weighted and money-weighted return.
Costante provides educational workflow tools, not financial advice. Trading involves risk.