Probabilistic Thinking: Judging Decisions, Not Outcomes

Resulting: the error that prevents learning

Resulting means inferring the quality of a decision from the quality of its outcome. In domains with little randomness this works well. In trading, where a sound decision loses perhaps half the time, it is actively misleading.

Expectancy as the unit of judgement

Expectancy = (Win% x AvgWin) - (Loss% x AvgLoss)
             measured in R, where R = risk per trade

Strategy A: 70% win rate, 0.5R wins, 1.5R losses
   (0.70 x 0.5) - (0.30 x 1.5) = 0.35 - 0.45 = -0.10R

Strategy B: 35% win rate, 3.0R wins, 1.0R losses
   (0.35 x 3.0) - (0.65 x 1.0) = 1.05 - 0.65 = +0.40R

Strategy A feels good and loses money.
Strategy B feels terrible and makes money.

Annual result depends on expectancy x frequency:
   B at 80 trades/year = +32R
   At 0.5% risk per R, roughly +16% before costs

Expectancy is what makes it possible to follow Strategy B. Without the number, the experience of losing 65 percent of trades is unbearable. With it, each loss is an expected component of a positive-expectancy process rather than evidence of failure.

Thinking in distributions

Every trade is a draw from a distribution rather than a prediction. Once that is internalised, questions change: not "will this work?" but "what is the distribution of outcomes, and is the average favourable?"

• Any single trade is uninformative. With a 40 percent win rate, losing four in a row happens constantly and says nothing.
• Streaks are expected, not anomalous. The expected longest losing streak in 500 trades at a 60 percent loss rate is around twelve.
• The sequence matters as much as the set. The same trades in a different order produce very different drawdowns, which is what Monte Carlo simulation reveals.
• Your realised result is one sample. Both good and bad years contain substantial luck, which is why process metrics are more stable than outcome metrics.
• Sample size determines what you can conclude. Fifty trades supports almost no conclusion about a modest edge; five hundred supports several.
• Tails matter disproportionately. In skewed strategies, the average is dominated by a small number of outcomes, so the median experience differs greatly from the mean.

Calibrating your own confidence

1. 1

   Record a probability with every prediction

   Not just the direction, but how confident you are as a percentage. This is what makes calibration measurable.

2. 2

   Group predictions by confidence level

   All the trades where you said 70 percent, all where you said 55 percent, and so on.

3. 3

   Compare stated confidence with realised frequency

   If your 70 percent predictions occur 52 percent of the time, you are systematically overconfident by a measurable amount.

4. 4

   Adjust rather than resolve

   Calibration improves through feedback, not through intention. Knowing you are overconfident by 15 points lets you discount accordingly.

5. 5

   Notice where calibration is worst

   Most traders are best calibrated on routine setups and worst on trades they find exciting, which is directly actionable.

6. 6

   Use calibration to inform size, not to override rules

   Better-calibrated confidence is useful for allocation between strategies, not as a reason to deviate within one.

Putting it into practice

Each reframing replaces an unanswerable question with a measurable one. That substitution is the practical content of probabilistic thinking, and it is what allows a trader to continue executing a sound process through the periods when the outcomes are unfavourable.