What is backtesting?
Backtesting applies a trading rule to historical prices and records the simulated outcomes.
Read the guide →
How TradingPal builds its historical trading records
A setup’s trade history and an account’s return answer different questions. This guide follows the calculation from a historical signal to a limited-capital portfolio, including allocation, costs and the gaps between simulation and live trading.
From an eligible trade to an account result
Imagine ten setups reach their entry prices on the same day, but an account has room for only two. Counting all ten tells you how the setups performed. The account’s return depends on the two positions it actually took, their size and when their capital became available again.
TradingPal uses separate calculations for those questions. The historical engines below supply the trade records and account simulations. Paper and live measurement then compare those assumptions with actual fills.
- Support and resistance: tests the named Support trendlines and Resistance trendlines rules on weekly and monthly charts.
- Chart patterns: tests wedges, pennants and triangles with defined entries, safety exits and final exits.
- Autopilot: replays eligible pattern trades through one account with position and cash limits.
- Paper and live accounts: record actual positions, fills, costs and outcomes for comparison with the simulations.
Illustrative
Checks on the historical decisions
Each stage needs a rule that can be applied consistently. The introductory backtesting guide explains how hindsight and repeated tuning can distort a result; these are the checks used here.
- Information timing: simulated decisions use information available at that point. Later pattern confirmation can still affect which trades enter the published sample, as explained in the pattern section.
- Possible fills: an entry must fall within the candle’s traded range. A gap past the trigger uses the worse opening price.
- Fixed trade rules: the entry, safety exit, target and trailing rule stay fixed within a test. Research variants are evaluated as separate rules.
- Recorded exclusions: impossible fills, duplicate positions and capacity skips have explicit reasons.
- Separate samples: trade statistics describe qualifying resolved setups; account statistics describe the positions the limited account could hold.
- Reserved-data evaluation: research under a fixed protocol checks rules on stocks or periods held apart from development. That procedure is distinct from chronological replay and does not apply to every published statistic.
Support and resistance setup records
The support-and-resistance record belongs to the complete named setups. It includes the line, surrounding price context, entry, safety exit and target. Results for these rules cannot be assigned to every nearby line on a chart.
At each historical decision date, the calculation reconstructs the automatic lines for that window and checks whether the full setup qualifies. A weekly setup and a monthly setup use their respective chart intervals.
For filters evaluated on reserved data, candidate rules are specified before testing them on separate stocks or periods. The support and resistance guide introduces the chart concepts used in the trade rules.
- Chart reconstruction: use the historical window available for the decision.
- Setup qualification: check the complete support or resistance rule.
- Entry and exit: use the defined entry region and safety-exit conditions for that interval. Stops are close-confirmed; the weekly book requires two qualifying closes.
- Completed record: retain entry, safety exit, target, final exit, holding time and result in R.
- Account simulation: replay those trades with the same capital limits used for setup-card accounts.
Chart-pattern trade records
A pattern trade begins only after the shape could be established on an earlier daily candle. The simulation then checks whether a later candle reaches the entry trigger beyond the boundary. This prevents a shape first identified after a move from receiving an entry at the start of that move.
Later confirmation is a separate limitation. The detector can use prices after the initial crossing to reject immediate failures, so the published sample contains patterns that ultimately passed those checks. It does not establish that each confirmed pattern was already known at the first crossing.
- Entry: use the first eligible reach of the trigger, with a worse opening fill if price gaps through it.
- Range check: reject a planned fill outside the candle’s range or an entry already beyond the target.
- Two exits in one candle: follow the engine’s specified ordering when a daily candle could contain both target and stop events. Daily prices cannot reveal the full intraday sequence.
- Exit policy: fix the structural safety exit at entry. Some bullish patterns switch to a 10-day moving-average exit after reaching the measured-move target.
- Duplicates: remove overlapping same-symbol, same-direction trades from ranking statistics; the account holds at most one position per symbol.
The account behind a setup-card return
Setup-card accounts use the most recent 20-year window. They report the completed setup statistics separately from the return of positions the account could take. In the ten-signal example, all eligible completed signals contribute to the relevant trade statistics, while available cash and positions determine the account’s holdings.
Planned risk is measured from entry to the safety exit. The R-multiple guide shows how that distance determines position size.
- Plan about 1% of current equity at risk per trade before caps reduce the size.
- Hold at most 16 positions, with one per symbol.
- Commit at most 20% of equity to one name.
- Limit total entry notional to 95% of equity, leaving roughly 5% uncommitted and using no leverage.
- Record signals skipped because of the account limits.
The Autopilot portfolio simulation
Autopilot starts with stored historical pattern trades and moves through the calendar. It must choose among eligible signals while respecting available cash and a 16-position limit. Capital becomes available again when a position closes.
Candidate rankings use trades completed before the decision date, blending a stock’s own history with its pattern-family history. The displayed ranking does not establish the order in which prices reach their triggers. Daily data cannot identify the exact order of same-day crossings, so the account tests multiple allocation orders as described below.
The live selector also limits a position to 1% of the stock’s median dollar volume. The current public simulation does not model that additional liquidity cap, which can create differences in accepted trades and position size.
- Position risk: 0.65% of current equity, sized from the breakout price before the modeled fill adjustment. Setup cards use about 1%.
- Position limits: up to 16 open positions and one per symbol, with no daily or per-family entry cap.
- Capital: at most 20% in one name and 95% gross exposure, with sufficient cash and no leverage.
- Tradability: minimum $5 share price, $2 million in 20-day median dollar volume and a stop width of at least 0.3%.
- Eligibility: only the pattern families and market-side conditions allowed by the configured policy.
Why the record keeps one path across 25 runs
Return to the ten signals competing for two positions. Daily prices tell us which triggers were reached, but may not tell us which was reached first. Taking different pairs can change later cash availability and years of compounded returns.
TradingPal runs 25 reproducible allocation paths using different same-day orderings. Each uses the same historical market prices. The spread shows how sensitive the result is to those allocation choices.
The headline comes from one representative run. These repeated allocations test sensitivity within the existing sample; testing on reserved stocks or periods is a separate procedure.
- Setup cards: select the actual run with the median ending account value.
- Autopilot: keep the previously selected path on routine refreshes. Do not switch paths when another run becomes the median.
- Associated statistics: take return, drawdown, Sharpe, capacity measures and the equity curve from that same run.
- Path spread: retain card return quartiles and Autopilot’s 10th-to-90th-percentile CAGR range.
Illustrative
Which costs each record includes
Execution costs differ across the published records. Check the cost basis before comparing their returns.
Limits on price, liquidity and position size restrict the simulated trades, but they do not account for every expense or fill difference. Paper and live records provide a separate check on actual execution.
- Setup cards: pattern and support-and-resistance records are before brokerage friction.
- Public Autopilot record: model entry and exit costs that increase with position size and decrease with available liquidity. The reference budget is 25 basis points for a round trip, not a fixed charge on every entry. Open holdings are valued at daily closes; new position sizes use the prior closing account value.
- Remaining assumptions: commissions, short-borrow fees and taxes are not separately modeled. Actual fills and market impact can differ. The former synthetic failed-entry losses have been removed; the current published path does not assume a random missed-fill rate.
Illustrative
Read the outputs in groups
Start with trade results, then account growth and losses, and finally the amount of supporting data. The metrics guide provides worked examples. The available fields vary by surface.
- Trades: win rate, profit factor, average and median R, average gain and loss, payoff ratio and dollar expectancy.
- Growth: cumulative return and CAGR, the annual rate connecting the starting and ending values.
- Losses: maximum drawdown on the reported path, resampled drawdown percentiles and the proportion of resampled paths falling below half the starting account.
- Risk-adjusted return: monthly-return Sharpe using a 0% risk-free rate and Probabilistic Sharpe; Autopilot also reports Deflated Sharpe, Sortino and Calmar.
- Activity and sample: trade and symbol counts, holding time, time in market, peak positions, skipped signals, test dates and allocation-path ranges.
- Contributions: monthly results and the contributions of different years and pattern families.
- Execution: actual account returns, exposure, entry slippage and differences between live and modeled signals and outcomes.
Illustrative
Compare simulated trades with actual execution
The paper and live measurement layer matches real signals to the production pattern model. It compares the prices and results an account received with the simulated trade for the same setup.
A persistent difference in fill costs or trade outcomes can change the expected result. Small samples remain uncertain, so the reports include sample warnings as well as measures of divergence.
- Entry slippage: the difference between actual and modeled entry, measured in basis points. One basis point is one-hundredth of a percentage point.
- Signal matching: identify real signals without matching backtest trades, called “ghosts,” and new backtest signals the live process missed.
- Outcome difference: compare actual and modeled R for the same setup.
- Recent performance: compare rolling live R with modeled expected R, including sample-size flags.
- Risk response: drawdown and divergence checks contribute to reported risk states and can reduce the trading mode.
When the published records update
Pattern and support-and-resistance records update through the nightly process. The table below shows the latest completed published pattern result.
The public Autopilot record is rebuilt after strategy-affecting changes. Opening a page reads the published record; it does not run a new simulation.
| Pattern | Usual break | Win rate | Avg return | Backtested trades | Fresh (45d) |
|---|---|---|---|---|---|
| Bullish pennant | Up | 54.5% | +3.6% | 6,817 | 86 |
| Bearish pennant | Down | — | — | — | 0 |
| Ascending triangle | Up | 48.9% | +2.6% | 1,753 | 19 |
| Descending triangle | Down | 48.9% | +2.6% | 1,753 | 6 |
| Symmetrical triangle | Either way | 48.9% | +2.6% | 1,753 | 80 |
| Falling wedge | Up | 44.3% | +2.9% | 1,149 | 60 |
| Rising wedge | Down | 52.0% | +1.6% | 390 | 9 |
Historical results of a simulated strategy, refreshed nightly. Triangle rows show their usual break direction's family; each guide breaks out both directions.
Where the simulation can differ from a real account
The ten-signal example shows why a collection of attractive trades may produce a different account result. Limited capital and uncertain entry ordering explain part of that gap. The assumptions below account for other differences that remain.
Use these limits when deciding what the record can support. A historical gain describes the stated rules and data; changes in markets, execution or the available stock universe can produce a different result.
- Surviving stocks: the store does not include every failed or delisted company, so the historical universe may be optimistic.
- Later confirmation: the sample includes patterns confirmed after the initial crossing. That limits claims about alerts available at the first crossing.
- Open positions: Autopilot’s historical replay holds open positions at entry value until settlement, so its drawdown can omit deeper losses inside a trade. Live account metrics mark the account daily.
- Compressed prices: daily, weekly and monthly candles hide the sequence of prices inside the interval.
- Provisional inventory: some source histories and older timeout exclusions remain under review. Known open trades stay in the portfolio until an explicit exit, but previously omitted entries may still affect the historical sample.
- Changing markets: a rule’s past behavior may weaken under different conditions.
- Different periods: the default Autopilot record begins in 2006; the ten-year view begins in 2016. Compare the same dates and accounting basis.
How TradingPal backtests: FAQ
Why does TradingPal use three backtest engines?
Because they answer different questions. The support-and-resistance engine tests named setups, the pattern engine tests wedges, pennants, and triangles, and the Autopilot backtest measures how one constrained account chooses among eligible pattern trades.
Why keep one path across 25 runs?
Breakouts that cross on the same day compete for scarce portfolio slots, and daily data does not say which crossed first. TradingPal runs 25 reproducible allocation paths and publishes one actual representative path: median ending value for setup cards and a fixed, previously selected path for Autopilot. It is a path-sensitivity check, not 25 independent market samples.
Does the 95% cap mean 95% of the account is at risk?
No. It is a 95% gross-exposure cap: the total entry notional cannot exceed 95% of equity, so there is no leverage and roughly 5% remains uncommitted. Planned loss risk is much smaller: about 1% per trade on setup cards and 0.65% in the current Autopilot policy, before caps can size it down.
Are commissions and slippage included?
Pattern and support-and-resistance cards remain before brokerage friction. The public Autopilot record models liquidity- and size-dependent costs at entry and exit, using a 25-basis-point round-trip reference. It does not separately charge commissions, short borrow or taxes. Actual execution may differ; no synthetic failed-entry losses are included.
Why do setup cards use 1% risk while Autopilot uses 0.65%?
A single setup-family card uses about 1% risk to show that edge in a constrained account. The current Autopilot policy spreads risk across a wider 16-position book; its validated setting is 0.65% per position. The two simulations share core caps but are not identical portfolios.
Are the 25 runs out-of-sample validation?
No. They test sensitivity to allocation order inside the same historical sample. Held-out validation is a separate research discipline used where a rule or gate is tuned under a frozen protocol.
Why can two drawdown numbers differ?
Realized max drawdown is the worst dip on the representative historical path. Bootstrap p95 drawdown samples observed trade blocks with replacement to build many synthetic paths, then reports a stress threshold that about 95% stayed below. Autopilot's settled-equity drawdown can still miss deeper moves inside open trades.
How often do the records update?
Pattern and support-and-resistance results update through the nightly process. The public Autopilot record is rebuilt after strategy-affecting updates and served from storage; opening a page never starts a backtest.
Keep learning
Track record metrics
A track record describes trade outcomes, account growth and the losses along the way.
Read the guide →
Risk/reward & R-multiples
A trade’s entry, safety exit and target let you compare its planned loss with its potential gain.
Read the guide →
Support trendlines
The Support trendlines strategy uses a planned entry above a qualifying support line on weekly or monthly charts.
Read the guide →
Educational content, not investment advice. Backtest statistics are historical results of a simulated strategy. Publication timing varies by record; the numbers describe the past, not the next trade.