Profit alone is not enough to judge an automated trading strategy. This reference guide explains the trading bot performance metrics that reveal return quality, risk, consistency, capacity, and resilience, then provides a reusable review template for comparing backtests, paper trading, and live results.
Overview
A robust performance review should answer three questions: how much did the strategy make, how much risk did it take, and how dependable was the result? A stock trading bot can show an attractive net return while relying on excessive leverage, a small number of lucky trades, favorable market conditions, or unrealistic execution assumptions.
Review metrics over the same period, using the same starting capital, fees, slippage assumptions, benchmark, and data frequency. Separate backtest, paper, and live results rather than combining them. Backtests estimate what might have happened, paper trading tests the operational workflow without real fills, and live trading includes actual execution, liquidity, latency, and behavioral risks.
No single measure proves that an algorithmic strategy is reliable. Use a group of related metrics and inspect the equity curve, individual trades, losing periods, exposure, and execution records. For additional due diligence, compare this analysis with a structured AI trading bot evaluation checklist and review how assumptions were handled in the backtesting process.
Template structure
Use the following fields for each trading bot performance review.
1. Return and benchmark
- Starting and ending equity: Record account value before and after the review period.
- Net return:
(Ending equity - Starting equity) / Starting equity. State whether fees, commissions, borrow costs, and slippage are included. - Annualized return: Convert the result to a yearly rate only when the observation period is long enough for that comparison to be meaningful. Clearly label the annualization method.
- Benchmark return: Compare with a relevant passive or strategy-specific benchmark, while noting differences in risk and exposure.
2. Drawdown and loss behavior
- Maximum drawdown: The largest peak-to-trough decline in the equity curve. A common calculation is
(Trough equity - Prior peak equity) / Prior peak equity. - Average drawdown: The typical decline from a local equity peak, which can show whether the maximum loss was an isolated event.
- Recovery time: The number of days or trades from a drawdown start until the account reaches its previous peak. A strategy with a moderate drawdown may still be difficult to operate if recovery takes a long time.
- Worst day, week, or month: These figures help translate risk into a time period a trader can plan around.
3. Risk-adjusted performance
- Volatility: The variation in periodic returns, usually calculated from daily, weekly, or trade-level returns. State the sampling interval because different intervals produce different results.
- Sharpe ratio: A common form is
(Average portfolio return - Risk-free return) / Standard deviation of returns. It measures return relative to total variability. The ratio is useful for comparisons, but it can be distorted by non-normal returns, short samples, and the chosen time interval. - Sortino ratio: This replaces total standard deviation with downside deviation. It focuses more directly on returns below a chosen target or minimum acceptable return.
- Calmar-style comparison: Some reviews compare annualized return with maximum drawdown. This can be a useful recovery and capital-efficiency lens, but the convention should be labeled because calculation periods vary.
4. Trade quality and consistency
- Win rate:
Winning trades / Total closed trades. It must be read alongside average win and average loss. - Average trade:
Total net profit / Number of closed trades. Compare it with estimated costs per trade; a small average trade can be highly sensitive to fees and slippage. - Profit factor:
Gross profit / Gross loss, where gross loss is expressed as a positive number. It describes the relationship between winning and losing dollars, not the sequence of outcomes. - Expectancy: A simple form is
(Win rate x Average win) - (Loss rate x Average loss). Expectancy estimates the average result per trade before considering changes in future market conditions. - Largest win and largest loss: These reveal whether the total result depends heavily on one or two trades.
5. Capital use and implementation
- Exposure: The percentage of time or capital invested. A low-exposure bot and a fully invested bot should not be compared using return alone.
- Turnover: The amount of buying and selling relative to account size. High turnover can increase costs and execution sensitivity.
- Position concentration: Record the largest position, sector concentration, and number of simultaneous positions.
- Fill quality: Compare intended entry and exit prices with actual or modeled fills. Include rejected orders, partial fills, delays, and missed trades when available.
- Operational incidents: Log outages, data gaps, duplicate orders, stale signals, and manual interventions separately from market performance.
How to customize
Start with a consistent reporting sheet containing the strategy name and version, market universe, timeframe, data source, signal rules, position-sizing method, leverage, order types, fees, slippage, and review dates. Without these fields, a later result may look different even though the code appears unchanged.
Choose metrics that match the strategy. A day trading bot should emphasize intraday drawdown, average trade after costs, fill quality, turnover, and the effect of spread and latency. A swing trading strategy should also track overnight exposure, gap risk, holding-period distribution, and performance around market catalysts. A diversified portfolio bot may need concentration, correlation, and portfolio-level drawdown measures rather than trade-level statistics alone.
Set review thresholds before examining the result. For example, you might require a strategy to stay below a predefined drawdown limit, maintain positive after-cost expectancy, and remain within an acceptable exposure range. Thresholds are decision rules, not guarantees. If a limit is breached, pause, investigate, and determine whether the cause is market regime change, implementation error, or ordinary variation.
Keep live and simulated results in separate columns. Differences are informative: live performance may deteriorate because of spreads, partial fills, unavailable liquidity, delayed data, or a mismatch between the coded and intended rules. A paper trading bot can help identify operational problems, but it cannot reproduce every feature of real execution.
Examples
Consider two hypothetical strategies with the same 20% backtest return. Strategy A has a 10% maximum drawdown, a 52% win rate, a 1.35 profit factor, moderate turnover, and results spread across many trades. Strategy B has a 30% drawdown, a 78% win rate, a 1.05 profit factor, and most of its profit comes from one unusually large trade.
Strategy B may look better if a reader focuses on win rate, but its drawdown and weak profit factor suggest a less favorable risk profile. The high win rate may also conceal occasional large losses. Strategy A is not automatically superior, yet its results provide a more balanced starting point for investigation.
Another misleading combination is a high Sharpe ratio with a short backtest and very smooth modeled returns. Smoothing, infrequent pricing, missing costs, or limited market conditions can reduce measured volatility and make the ratio appear stronger than the underlying strategy. Conversely, a strategy with uneven but positively skewed returns may have a modest Sharpe ratio while still being useful if its drawdown, expectancy, and risk limits are acceptable.
Use this compact review table for each test or monthly update:
| Field | Backtest | Paper | Live |
|---|---|---|---|
| Period and strategy version | Record dates and code version | Record dates and configuration | Record dates and configuration |
| Net return after costs | State assumptions | Use simulated fills | Use account records |
| Maximum drawdown | Equity curve | Paper equity curve | Real equity curve |
| Sharpe and Sortino | State return interval | State return interval | State return interval |
| Profit factor and expectancy | Trade log | Trade log | Trade log |
| Exposure and turnover | Portfolio history | Order history | Broker records |
| Incidents and differences | Model limitations | Operational notes | Execution notes |
Before increasing position size, compare the live record with the expected range from testing. Position sizing should reflect the drawdown a trader can tolerate, not merely the bot's strongest historical period. See the position-sizing guide for a separate framework.
When to update
Revisit the performance review monthly, after a meaningful batch of trades, and whenever the bot's code, data, broker, fees, position sizing, or execution settings change. Update it after a prolonged drawdown, a material change in market behavior, or an operational incident. A new news filter, indicator, universe, or order type can change the strategy's risk even if the headline return remains similar.
At each update, save a dated snapshot rather than overwriting the previous report. Recalculate results with the current fee and slippage assumptions, compare backtest, paper, and live periods, and inspect trades that differ from expectations. Confirm that API permissions, alerts, and order controls remain appropriate using the bot monitoring checklist.
For a practical next step, create the review sheet today, define your acceptable drawdown and exposure limits, and record the next reporting period before changing the bot. Evaluate net return, maximum drawdown, volatility, Sharpe, Sortino, profit factor, expectancy, win rate, average trade, exposure, turnover, and recovery time together. That disciplined record will make it easier to distinguish a genuinely robust automated trading strategy from one that was merely profitable under a narrow set of assumptions.