Use this paper-trading template to describe how an AI agent should evaluate stretched prices, decide whether a simulated mean reversion setup qualifies, record invalidation, and produce reviewable journal notes. Trading Boy does not execute live trades, hold funds, or provide financial advice.
A mean reversion agent is only useful when the rules define what normal means, what stretched means, and what evidence shows the move is returning toward its reference point. Use this page as a reusable scaffold for a paper trading agent that studies deviations from VWAP, moving averages, volatility bands, z-scores, range extremes, or other written references.
The goal is not to predict every bounce. The goal is to make a paper workflow that can be audited later. A good mean reversion template says which market regime is eligible, which extreme is valid, which signal is required before entry, where the idea is wrong, when the agent must skip, and what the journal should contain after the simulated trade closes.
Before filling in the template, write the source rules in the AI trading agent rules workflow, connect the paper risk boundaries in AI agent risk controls, and compare the expected output with the pre-trade checklist. Those pages keep this template from becoming a loose market opinion generator.
| Block | What to define | Paper review question |
|---|---|---|
| Universe | Assets, sessions, liquidity floor, spread limit, and timeframe. | Was the agent evaluating only the market it was assigned? |
| Reference mean | VWAP, moving average, prior range midpoint, fair value band, or z-score baseline. | Was the reference observable before the simulated trade? |
| Extreme condition | Distance from the reference, volatility expansion, RSI threshold, band breach, or range extension. | Was the move actually stretched by the written definition? |
| Confirmation | Stall, reclaim, failed breakdown, volume shift, reduced momentum, or other required trigger. | Did the agent wait for confirmation instead of buying or fading blindly? |
| Invalidation | Price level, time stop, volatility break, thesis failure, or news condition that cancels the setup. | Could the reviewer tell exactly when the idea was wrong? |
| Exit logic | Reference retest, partial mean return, trailing rule, max hold time, or stop condition. | Did the exit follow the template rather than a hindsight explanation? |
Role: You are a mean reversion paper-trading agent inside Trading Boy. You help review simulated setups only. You do not execute live trades, hold funds, place orders, provide financial advice, or describe paper alerts as recommendations.
Objective: Evaluate whether the watched asset has moved far enough away from its written reference mean to qualify for a paper mean reversion test. Your job is to produce disciplined simulated decisions and journal evidence, not confident predictions.
Universe: Review only [asset list], [session], [timeframe], and [liquidity requirement]. Skip any asset with spreads wider than [spread limit], abnormal data, missing candles, unresolved exchange issues, or conditions outside the paper test.
Reference mean: Use [VWAP / moving average / range midpoint / z-score baseline / volatility band] as the mean. Record the current value, the distance from price, and whether the reference was available before the decision.
Eligible extreme: A setup can qualify only when price is at least [distance threshold] from the reference, volatility is within [allowed range], and the move is not caused by excluded news, halted liquidity, or a structural break.
Confirmation: Do not enter on distance alone. Require [confirmation trigger], such as a reclaim of the band, failed continuation, lower momentum, range re-entry, or another observable signal from the written rules.
Risk: Use a maximum simulated position size of [paper size], a maximum paper loss of [loss limit], a maximum daily drawdown of [drawdown limit], and a correlated exposure cap of [correlation cap]. Skip if the setup requires breaking any risk limit.
Invalidation: The paper idea is wrong if [price level], [time condition], [volatility expansion], [news event], or [trend continuation condition] occurs. State the invalidation before any simulated entry.
Exit plan: Plan the exit around [mean retest], [partial return], [time stop], [stop condition], and [journal review point]. Do not move the target or stop in the journal unless a written rule allows that adjustment.
Output: For every simulated entry, exit, or skip, write the setup name, reference mean, distance, confirmation, invalidation, risk check, reason for action or skip, expected review question, and final paper result.
Mean reversion can look simple because stretched prices often move back toward a reference. In practice, some stretched moves are early trend moves, forced liquidations, news repricing, or liquidity breaks. The template should protect the paper workflow from treating every deviation as a trade.
These constraints also make the page compatible with a broader paper trading workflow. The agent can generate journal evidence, but the human reviewer still evaluates whether the rule version is worth changing, retiring, or testing again.
A good output says the asset is 2.1 standard deviations below the chosen reference, names the range-bound regime rule, confirms the required reclaim, states the simulated stop, explains why risk is inside the cap, and records whether the paper exit happened at the partial mean return, full reference retest, stop, or time limit.
A weak output says price is oversold and likely to bounce without naming the reference mean, regime, threshold, confirmation, invalidation, risk cap, or skip criteria. That kind of output is hard to evaluate and should be rejected by the paper workflow.
After a paper session, evaluate the agent like an operations system. Ask whether each simulated decision followed the written template, not whether one outcome happened to be profitable. The right comparison is rule quality, consistency, and review value.
Start with the journal. The entries should identify the reference mean, distance, confirmation trigger, invalidation, risk check, and exit plan. If those fields are missing, improve the template before drawing conclusions from the result. Use AI paper trading agent evaluation to compare behavior across versions and avoid mixing data from different rule sets.
Then inspect skipped opportunities. A mean reversion agent that skips because the market is trending, spreads are wide, the stop is too far, or the move is news-driven may be doing exactly what the rules require. Skips help distinguish a disciplined agent from an agent that simply looks for action.
Finally, connect the findings to the broader AI paper trading agent workflow. If the agent obeyed the template but the idea failed repeatedly, the issue may be the strategy definition. If the idea worked but the journal was incomplete, the issue may be output structure. If the agent kept violating boundaries, return to the AI trading agent prompt template and make the instructions stricter.
Replace the bracketed fields in the template with observable values. Keep every field specific enough that another reviewer can read the journal later and understand whether the agent followed the rule.
Define asset class, watchlist, session, minimum volume, maximum spread, excluded events, and the timeframe the agent is allowed to inspect.
Define the exact stretch condition, such as z-score, percentage distance from VWAP, band deviation, or range extension.
Define the journal fields required for entries, exits, skips, prompt version, rule version, and post-trade review notes.
For limitations, read paper trading limitations. For first setup design, start with the rules workflow. For ongoing guardrails, keep risk controls close to the template instead of treating risk as a later review step.
It is a reusable paper-trading structure that tells an agent how to define an extreme move, when to skip, where the idea is invalid, how simulated risk is capped, and what journal evidence must be produced.
No. The template should require regime, liquidity, volatility, news, correlation, and risk checks before any simulated entry is allowed.
No. Trading Boy is for paper trading, evaluation, and workflow review. Trading Boy does not execute live trades, hold funds, or provide financial advice.