F 072 Trade Sizing When AI Signals Disagree

F 072 sizes trades by blending multiple AI signals into a single confidence score, then scaling position size proportionally to that score when signals disagree. You enter the signals’ individual scores and the maximum position size you are willing to take, and the calculator returns a recommended trade size that accounts for disagreement by reducing exposure.

Formula

The calculator uses a straightforward averaging method: it converts each signal score from a 0–100 scale to a 0–1 fraction, averages those fractions, and multiplies by the maximum position size you specify. The formula is:

Recommended Size = Max Size × ( (S1/100 + S2/100 + S3/100) / 3 )

Where:

• S1, S2, S3 are the individual AI signal scores (0–100), representing each model’s confidence that the trade should be taken.
• Max Size is the largest position you would take if all signals fully agreed (in any unit: shares, contracts, lots, or dollars).

The output is in the same unit as your maximum size. For example, if you set max size to 1,000 shares, the result is in shares.

Worked Example

Using the default values from the calculator: Signal 1 = 70, Signal 2 = 55, Signal 3 = 40, and Max Size = 1,000 units.

1. Convert each score to a fraction: 70/100 = 0.70, 55/100 = 0.55, 40/100 = 0.40.
2. Average the fractions: (0.70 + 0.55 + 0.40) / 3 = 1.65 / 3 = 0.55.
3. Multiply by max size: 1,000 × 0.55 = 550 units.

The calculator displays 550 as the recommended trade size, with an average score of 55.00 and a disagreement level of 0.30 (the range from 40 to 70 divided by 100).

Why Disagreement Matters

When multiple AI models or signals are used to decide on a trade, they rarely all agree. One model might see a strong bullish pattern, another might be neutral, and a third might flag risk. If you size the trade based on the most optimistic signal, you take on more risk than the collective wisdom suggests. Conversely, if you only trade when all signals agree, you miss many opportunities. The averaging method sits in between: it treats each signal equally and scales your exposure to the overall level of conviction.

This approach is common in ensemble trading systems, where combining several models reduces the impact of any single model’s error. The trade size reflects the average confidence, so when signals are split (e.g., 80, 50, 20), the size is moderate, not full. When they all agree at 90, the size is close to the maximum. This is a simple, transparent way to handle disagreement without complex weighting schemes.

How to Choose Signal Scores

Your signal scores should come from the AI models or strategies you are using. Many trading platforms output a confidence score between 0 and 100, but if yours does not, you can normalize any score range to 0–100. For example, if a model returns a probability between 0 and 1, multiply by 100. If it returns a z-score, you might map it to a percentile.

• Use consistent scales: All signals must be on the same 0–100 scale for the average to be meaningful. If one signal is a probability and another is a 1–5 star rating, convert them first.
• Consider weighting: The calculator treats all signals equally. If you have reason to trust one model more than others, you can adjust its score upward or use a weighted average outside the calculator.
• Set max size realistically: This should be the largest position you are comfortable with when all signals are at 100. It is not the amount you want to risk, but the full position size before scaling.

Common Mistakes and Misconceptions

A frequent error is using the average score as a percentage of your account instead of as a fraction of your maximum size. For example, if your max size is $10,000 and the average score is 55, the calculator gives $5,500. Some traders mistakenly use 55% of their entire account, which may be far larger than intended. Always define max size as the full position you would take at 100% confidence, not your total capital.

Another misconception is that disagreement alone should reduce size further. The calculator already accounts for disagreement through the average: a wide spread (e.g., 90, 50, 10) produces an average of 50, which is lower than a tight cluster (e.g., 60, 55, 50) at 55. If you want to penalize disagreement more heavily, you can subtract a penalty from the average before multiplying, but the basic formula does not do that.

Also, note that the calculator does not consider volatility, market conditions, or your risk tolerance. It only translates signal confidence into a position size. For a complete risk management system, you should combine this with stop-loss levels and position sizing based on account risk.

Limitations of the Averaging Approach

The simple average assumes all signals are equally informative and independent. In practice, AI models may be correlated—if they are trained on similar data, they may make similar errors. When signals are correlated, averaging does not reduce risk as much as it would with independent models. If your signals are highly correlated, consider using a more conservative method, such as taking the minimum score or applying a higher penalty for disagreement.

The formula also ignores the order or recency of signals. If one signal is based on stale data, it should perhaps have less weight. A weighted average or a decay factor could address that, but the calculator uses equal weights for simplicity. Finally, the 0–100 scale is linear, meaning a score of 80 is assumed to be twice as confident as 40. If your models’ confidence is not linear, the output may not reflect true conviction.

Practical Tips for Using the Result

When you get the recommended size, treat it as a starting point, not a final order. Consider the following before executing:

• Check liquidity: If the calculated size is too large for the market’s typical volume, you may move the price. Reduce the size accordingly.
• Set a stop-loss: The calculator does not tell you where to place a stop. Use your usual risk rules to determine the stop distance, and ensure the potential loss fits your risk tolerance.
• Re-evaluate when signals change: If new data arrives and a signal score changes, re-run the calculation. Do not hold a position sized for old confidence levels.
• Backtest the method: Before using this sizing in live trading, test it on historical data to see how it would have affected your returns and drawdowns.

Frequently Asked Questions

What if I have only two AI signals? You can still use the calculator by entering the third signal as the same as one of the others, or you can simply average the two scores manually. The calculator requires three inputs, but you can set the third equal to the average of the first two to get the same result as a two-signal average.

Can I use this calculator for crypto, stocks, or forex? Yes, the calculator is unit-agnostic. It works for any tradeable asset as long as you define max size in the appropriate units (shares, contracts, lots, or currency). The logic of scaling by confidence applies universally.

How do I convert my AI signal from a different range to 0–100? If your signal ranges from A to B, use the formula: normalized = (signal – A) / (B – A) × 100. For example, a probability of 0.75 (range 0–1) becomes 75. A star rating of 4 out of 5 becomes 80.

What is a good disagreement level? There is no universal threshold. A disagreement level above 0.5 indicates that the signals span more than 50 points on the 0–100 scale, which is quite high. In such cases, the average score will be moderate, and the trade size will be reduced accordingly. You may decide to skip the trade if disagreement exceeds a certain level, but that is a personal risk preference.

Does this calculator replace professional risk management? No. It is a simple tool for position sizing based on signal confidence. It does not account for market risk, portfolio correlation, or your overall risk tolerance. Always consult a qualified financial advisor before making trading decisions.

Related

• AI-Generated Propaganda in Your Music: What to Do
• Playwright MCP Clear: What It Resets and What It Can’t
• Claude Cowork Alternatives: Tools for AI Workflows

in sig1 | Signal 1 score (0-100) | - | 70 | 0 | 100
in sig2 | Signal 2 score (0-100) | - | 55 | 0 | 100
in sig3 | Signal 3 score (0-100) | - | 40 | 0 | 100
in max_size | Maximum position size | units | 1000 | 0 | 1000000
out size | Recommended trade size | units | max_size * ((sig1/100 + sig2/100 + sig3/100) / 3) | 0
out avg_score | Average signal score | - | (sig1 + sig2 + sig3) / 3 | 2
out disagreement | Disagreement level | - | (max(sig1,sig2,sig3) - min(sig1,sig2,sig3)) / 100 | 2