Decentralised News Research · AI trading performance
“The bot made money” answers an accounting question. It does not tell you whether paying for the bot improved the outcome available from a simpler investment. A rising market can reward activity that adds little value.
Begin with a reconciled result using DN’s AI trading bot P&L audit. Then use the DN Benchmark Comparison Worksheet below to test what that result means. It separates profitability, opportunity cost and the strategy’s stated objective.
1. Decide what “added value” means
There are three separate questions. Profitability: did the strategy finish ahead after costs? Opportunity cost: did it improve on an alternative you could reasonably have held? Objective: did it deliver the return, exposure and loss profile you selected before the evaluation?
A bot designed to maximise participation in a rising asset should face a demanding comparison with holding that asset. A strategy designed to limit market exposure needs an additional comparison with a simpler reduced-exposure portfolio. These questions can produce different verdicts without contradiction.
Write the objective down first. “Lower drawdown” is weak justification when introduced only after disappointing returns. Likewise, beating cash is not enough to demonstrate skill if the strategy accepted substantial directional or liquidation risk.
2. Choose the benchmark before seeing the outcome
CFA Institute’s benchmark guidance emphasises a relevant, measurable comparison chosen in advance, with matching reporting periods and an explanation of its construction. The underlying principle is useful here: a benchmark should answer the investment question, not rescue the result.
| Strategy question | Useful comparison | What to disclose |
|---|---|---|
| Did timing a single spot asset help? | Holding the same asset over the same interval. | Entry cost, valuation prices and remaining cash. |
| Did rotating among several assets help? | A basket fixed before the evaluation. | Weights, constituent changes and rebalancing costs. |
| Did a defensive strategy improve the trade-off? | A stated asset/cash blend, alongside full exposure. | Cash return, custody and allocation policy. |
| Did automation help with regular contributions? | Passive purchases on the same funding dates. | Contribution timing and execution assumptions. |
| Did leverage or a long/short strategy add value? | A documented comparison reflecting the objective and exposures. | Gross/net exposure, financing and liquidation differences; unlevered holding is context alone. |
For a real comparison, describe the alternative precisely. “Hold crypto” leaves the asset, currency, custody and entry time undefined. “Hold a fixed quantity of BTC bought at the stated executable price” is reproducible. A custom cash blend also needs a cash-return assumption; a stablecoin balance or lending position must not be described as risk-free.
If no defensible strategy benchmark exists, say so. You can still show a simple opportunity-cost comparison, with its limitations, without calling the difference risk-adjusted outperformance.
3. Put both sides on the same accounting basis
Use identical start and end timestamps, timezone, reporting currency and starting budget. Mark open positions at both ends. Count all capital reserved for the strategy, including unspent averaging-order cash and collateral, rather than dividing profit only by the first small order.
Include relevant trading charges, funding, borrowing, subscriptions and platform costs once. A subscription paid from a separate bank account is still an economic cost: show that expense and its funding explicitly. Do not give the bot additional money while leaving the benchmark at the original budget.
The CFTC’s AI-bot advisory specifically identifies fees, spreads and subscription costs as considerations when evaluating returns. For the benchmark, distinguish a frictionless reference from a costed implementation a reader could actually replicate.
Actual execution prices already incorporate their execution differences. Do not subtract estimated slippage again from actual-fill P&L. Apply comparable endpoint treatment: mark both portfolios, or estimate the cost of closing both. Charging an exit cost to only one side tilts the comparison.
4. Example A: a profitable bot trails holding
Fictional DN exercise. Each strategy starts with an all-in $10,000 budget. There are no additional deposits, withdrawals, taxes or asset distributions. Both results end at marked values, with no hypothetical final liquidation charge. All amounts and fees are synthetic, not current exchange quotes.
The bot produces $1,800 of investment P&L before the separately listed charges. This includes the change in marked open-position P&L. Its $250 trading costs, $100 funding payment and $100 subscription are all funded within its starting budget.
Bot net return = ($11,350 ÷ $10,000) − 1 = 13.50%
The passive alternative buys a fictional asset at $100. Its entry fee is 0.10% of the $10,000 budget, or $10, leaving $9,990 to invest. That buys 99.9 units. At the same endpoint, the price is $120.
Costed holding return = ($11,988 ÷ $10,000) − 1 = 19.88%
| Measure | Bot | Costed holding |
|---|---|---|
| Starting budget | $10,000 | $10,000 |
| Ending marked value after costs | $11,350 | $11,988 |
| Net monetary gain | $1,350 | $1,988 |
| Net return | 13.50% | 19.88% |
| Drawdown or recovery | Unknown from endpoints. | Unknown from endpoints. |
The bot made money but finished $638 behind the costed alternative. Its net return gap is −6.38 percentage points: 13.50% minus 19.88%. That is different from a relative percentage difference in ending wealth; $638 divided by $11,988 is approximately 5.32%.
The frictionless asset price rose 20%. Keeping that reference visible explains the passive entry-cost effect, but the investable comparison here is 19.88%. This example establishes a return shortfall. It cannot establish which portfolio experienced larger interim losses because no path was supplied.
5. Example B: lower return, smaller observed drawdown
A separate fictional exercise. The values below are complete net equity observations at six monthly observation points, with costs already included and no external flows. They are not continuations of Example A.
We also define, before this exercise, a simple alternative: 50% of the starting budget follows the holding series and 50% remains in zero-yield cash, without rebalancing. Its value at each observation is $5,000 plus half the full-holding value. This is a synthetic exposure comparison, not a claim that cash is free of custody or currency risk.
| Observation | Bot | Full holding | 50% holding / 50% cash |
|---|---|---|---|
| Month 0 | 10,000 | 10,000 | 10,000 |
| Month 1 | 10,300 | 11,000 | 10,500 |
| Month 2 | 10,094 | 8,800 | 9,400 |
| Month 3 | 10,400 | 10,500 | 10,250 |
| Month 4 | 10,600 | 11,500 | 10,750 |
| Month 5 | 10,800 | 12,000 | 11,000 |
Sampled maximum drawdown = the largest such decline in the supplied observations
The bot’s largest observed decline is $206 from its $10,300 peak: 2%. Full holding falls $2,200 from $11,000: 20%. The cash blend falls $1,100 from $10,500: approximately 10.48%.
| Measure | Bot | Full holding | 50/50 blend |
|---|---|---|---|
| Net period return | 8% | 20% | 10% |
| Sampled maximum drawdown | 2% | 20% | 10.48% |
| Peak / trough observations | Month 1 / Month 2 | Month 1 / Month 2 | Month 1 / Month 2 |
| First observation recovering that peak | Month 3 | Month 4 | Month 4 |
| Peak to first recovery | 2 monthly intervals | 3 monthly intervals | 3 monthly intervals |
The return objective favours full holding. A predeclared defensive objective may make the bot’s smaller observed decline relevant, but it does not erase the 12-percentage-point return gap. The simpler cash blend narrows the exposure comparison and still earns two percentage points more than the bot, with a larger sampled drawdown.
Neither alternative dominates the bot on both displayed measures. Whether the trade-off is worthwhile depends on the original objective, additional risks, implementation burden and stronger evidence than six observations. There is no general rule that sacrificing return for a smaller historical drawdown is a good decision.
CME’s discussion of drawdown explains why monthly reports can conceal larger losses within the month. These calculations establish sampled declines and recovery observations, not the exact intramonth path or a future loss limit.
6. Match cash flows and exposure without rewriting history
If $1,000 becomes available halfway through a real test, the comparison must not assume the passive portfolio invested it at the beginning. For an investor-outcome comparison, give both alternatives the same contributions and withdrawals on the same dates, then compare resulting wealth and costs.
For a strategy-return comparison, GIPS return methodology distinguishes time-weighted returns, which seek to neutralise external funding effects, from money-weighted returns that reflect their timing and size. Do not compare one method with the other as if they were interchangeable. A simple percentage based on initial capital is sufficient for our no-flow examples; it is not a universal calculation.
Keep leverage separate. A bot with three times the directional exposure has not demonstrated three times the skill. Financing, liquidation thresholds, collateral fluctuations and changing position sizes can alter the outcome. Multiplying a passive return by three does not recreate a path-dependent leveraged strategy.
Likewise, a lower average market allocation can explain some loss reduction. Use an allocation rule fixed in advance for a defensive comparator. A benchmark rebuilt after the event to copy the bot’s exact successful trades would stop being an independent alternative.
DN Benchmark Comparison Worksheet
Complete these fields in your own notes before accepting an outperformance claim. This worksheet has no account connection or executable calculator.
| Record | What to enter | Decision check |
|---|---|---|
| Objective | The intended return/exposure/loss objective and when it was set. | Was success defined before the outcome? |
| Comparison policy | Asset or basket, weights, cash treatment and rebalancing rule. | Does it answer the selected question? |
| Capital and time | Identical budget, currency, timestamps and external-flow schedule. | Did either side receive an advantage? |
| Net result | Marked starting/ending equity; itemised cost and funding treatment. | Are open losses and subscription costs included once? |
| Return gap | Bot return minus benchmark return, in percentage points. | Use comparable return methods. |
| Wealth gap | Bot ending wealth minus benchmark ending wealth. | Capital flows and outside costs must match. |
| Path and exposure | Observed drawdown, peak/trough/recovery dates, frequency, leverage and cash allocation. | What risks remain unmeasured? |
| Evidence and verdict | Live/simulated status, source coverage, strategy versions and qualified conclusion. | Separate arithmetic, authentication and AI attribution. |
Check the price and income convention
Document valuation venue, timestamp, currency conversion and applicable distributions. Where an investment pays income, comparing a strategy that includes that income with a price-only reference can distort the result. Do not assume two similarly named price series use identical inputs.
Check cost fairness
Distinguish actual charges from modelled costs and a frictionless reference. Show whether the endpoint is marked or closed. Verify that a subscription, funding payment, entry fee or slippage adjustment appears only once and is funded consistently.
Check the sample and revisions
Preserve losing periods, strategy changes and interruptions. Record which rules were selected using earlier data and which were evaluated later. A short favourable window or many correlated trades does not establish persistent skill; disclose missing coverage instead of assigning a confidence score.
7. State a conclusion the evidence can support
Use conclusions such as “profitable but behind the stated passive alternative,” “lower return with smaller sampled drawdown,” or “not comparable because cash-flow and valuation records are missing.” Avoid turning a single positive return difference into “AI alpha.” That requires stronger evidence about risk exposures, repeatability and what actually generated the decisions.
The SEC’s investment-adviser marketing guide describes restrictions on misleading performance presentation, including gross and hypothetical results. Its legal requirements concern advisers within its stated scope; they are not a universal regulatory classification for crypto software.
Similarly, NFA’s guidance on hypothetical performance explains hindsight and simulation limitations. Clearly separate a backtest, a forward simulation and an actual funded record. DN’s paper-trading guide and claim-verification checklist address those earlier evidence questions.
DN Alpha Thesis
The strongest reason to use a trading bot is a demonstrable improvement against an alternative that matches your objective, after the full cost and risk comparison. Activity, complexity and an AI label are not substitutes for that evidence.
Read the win-rate red flags before trusting a headline statistic. If you are evaluating the workflow in our Binance and 3Commas setup guide, record the comparison policy before the first test. A working automation setup still needs an independently evaluated strategy.
Frequently asked questions
Is positive profit enough to show that a trading bot added value?
No. Profitability is one question. Compare the net result with a relevant alternative using matching dates, capital and costs, then examine whether the risk and operational trade-off met the objective set in advance.
Should every bot be compared with buy-and-hold?
Buy-and-hold can show opportunity cost, but it may not match a defensive, leveraged or long/short strategy’s objective. Add an appropriate comparison and disclose material exposure differences rather than treating one baseline as a universal verdict.
What does a 6.38-percentage-point return gap mean?
In Example A, the bot returns 13.50% and the costed holding alternative returns 19.88%. The bot-minus-holding gap is −6.38 percentage points. This is distinct from the relative percentage difference in ending wealth.
Does lower drawdown make a lower-return bot better?
Not automatically. A smaller observed drawdown may suit a predeclared defensive objective, but it does not establish lower overall risk or superior value. Compare simpler alternatives and consider the sample, costs, leverage and unobserved losses.
How should deposits and withdrawals be treated?
For an investor-outcome comparison, apply the same external flows on the same dates to both alternatives. For a strategy-return comparison, use consistent cash-flow-adjusted methods. Do not treat a new deposit as trading profit.
Does this worksheet prove an AI trading advantage?
No. It helps structure a fair comparison. It does not authenticate account records, reproduce a backtest, establish AI involvement or predict future performance. All numerical examples in this article are fictional.
Methodology, limitations and disclosure
DN reviewed six primary sources and built the synthetic calculations independently. Example A supplies endpoint arithmetic only. Example B supplies all six observations used for its sampled drawdown and recovery calculations. Neither exercise represents a real asset, live account, platform result or market forecast.
The worksheet is educational and makes no claim of GIPS compliance. Taxes and individual suitability are outside its scope. Regulatory sources retain their own legal scope. Documentation references are neutral; this article contains no affiliate signup links or platform ranking. DN may have commercial relationships elsewhere, which do not determine the comparison method.
Change log: 19 October 2026 — initial research, worksheet and synthetic examples prepared. Corrections: press@decentralised.news; identify the affected section and evidence without sending credentials or private account records.
Primary sources
- CFA Institute: Guidance Statement on Benchmarks for Firms — effective 1 April 2021; benchmark selection and presentation.
- CFA Institute: GIPS Standards Handbook for Asset Owners — valuation, external flows and return methods.
- CME Group: evaluating trading managers — drawdown and observation limits.
- CFTC: customer advisory on AI trading bots — performance claims and costs.
- SEC: Investment Adviser Marketing — guide dated 28 April 2021; performance-presentation context.
- NFA: Interpretive Notice 9025 — hypothetical-performance limitations.
