Skip to main content

the-virtuous-cycle

The Two Directions of Analysis​

When studying conditional behavior in markets, there are two fundamentally different ways to analyze the same event space. Both use the same dataset of events, but they move in opposite conceptual directions.

1. Forward Analysis — Adding Conditions​

In this direction, you start from a global set of events and apply an additional condition to see how performance changes. For example:

“What happens to persistence rate if I only include events where ATR > 1?”

By applying a filter, you’re restricting the sample space. You’re effectively asking:

P(Outcome∣Condition)P(\text{Outcome} \mid \text{Condition})

This is a forward-looking or experimental perspective — you use your knowledge about how the world behaves to test what happens when you operate only within a certain context.

  • The sample composition changes.
  • The success rate (e.g., 70%) may rise or fall.
  • You are learning how the rule performs under controlled subcontexts.

It’s an active mode of analysis — you’re simulating what happens if you act under new constraints. This is how filters, volatility caps, or regime restrictions are discovered.

2. Backward Analysis — Cohort Differentiation​

In this direction, you hold the outcome labels fixed and study what the world looked like in each case. For example:

“Out of all slope-persistence events, what did ATR look like for the ones that succeeded versus failed?”

Here you aren’t filtering; you’re diagnosing. You’re asking:

P(Condition∣Outcome)P(\text{Condition} \mid \text{Outcome})

The success/failure ratio (such as 70/30) remains constant — what changes is your understanding of the context distribution inside each cohort.

  • The contextual features (ATR, VWAP distance, Bollinger position, etc.) vary between cohorts.
  • You learn what conditions were associated with success or failure.
  • It’s a passive and diagnostic mode of analysis — a way to uncover structure in historical outcomes.

This is how you identify hidden dependencies and context-sensitive behavior.

3. How the Two Approaches Work Together​

Forward and backward analysis form a continuous feedback loop:

StepQuestionMethod
1“When this condition holds, how does performance change?”Forward (apply filters)
2“When the rule worked vs. failed, what conditions differed?”Backward (compare cohorts)
3“Can I refine the rule to exclude failure conditions?”Forward again (test refined filter)

Each informs the other:

  • Forward analysis tests interventions.
  • Backward analysis reveals causes and structure.

Together, they create a cycle of hypothesis → observation → refinement that turns a raw conditional statement into a context-aware trading rule.

4. The Virtuous Cycle of Analysis​

Forward and backward analysis aren’t competing methods — they reinforce each other. Together they create a feedback system that steadily improves both understanding and performance.

  1. Observe → Build an initial conditional rule from data. “Slopes above 0.15 tend to persist 70% of the time.”

  2. Diagnose (Backward) → Study where it failed. “Failures mostly occurred when ATR was above 1.5.”

  3. Refine (Forward) → Add that observation as a new filter. “Test slope > 0.15 only when ATR < 1.5.”

  4. Re-evaluate → The success rate shifts; a new structure emerges. “Now persistence rises to 83%, but only in aligned HTF regimes.”

  5. Repeat → Each iteration sharpens both the rule and your model of the market.

This cycle turns raw statistics into structured understanding:

  • Backward analysis discovers why behavior differs.
  • Forward analysis tests what happens if those insights are enforced.

Over time, the alternation builds resilience into your logic — you’re not just fitting filters to past data, you’re sculpting a context-aware framework that evolves with new evidence.