# Observation, model and inference

**DL-15 · Dealer Logic · Core**

**Lesson question:** What can dealer analytics actually establish?

Original fictional examples; educational writing, not live market data, orders or production approval.

## By the end

- Explain the mechanism in the lesson question.
- Reproduce the worked comparison and recognize its failure case.

**Prerequisites:** DL-01

A trade print is an observation. A signed dealer inventory inferred from public data is a model. A predicted hedge response is conditional inference. A future price target goes further still.

Keeping these layers separate is more useful than giving every line an authoritative label.

## Understand the mechanism

Open interest lacks a universal dealer-direction tag, and trade classification can be uncertain. Offsetting structures matter. A study of one historical product window is not today's positioning feed for every instrument.

The hedge identities are valid under their assumptions; the challenge is establishing the inputs and the effect's scale relative to other flows.

## Walk through the process

Ask for provenance, timestamp, coverage, ownership rules, units and sensitivity analysis. Prefer an honest range over false precision. Record missing information.

'The data do not identify net dealer sign' is a valid research result. It can prevent an unjustified position.

## Evidence boundary

The arithmetic is original conditional teaching material, not a measurement of current dealer positions. The cited Cboe research is historical product-specific evidence, not a live inventory feed. Model outputs should remain distinguishable from observed market facts.

## A worked example

Two allocations of identical OI produce +4,000 and −4,000 shares per dollar of Gamma. The observed contracts did not change; the allocation assumption did. The chart displays that sign uncertainty.

Without stronger ownership evidence, a claim that dealers must buy the next decline is not established. A later coincidental matching move would not prove the hidden input.

## Graph: Observation, model and inference—worked comparison

**Horizontal:** Inventory allocation  
**Vertical:** Modeled Gamma (shares/USD)

Fictional data from the written example. The chart organizes the comparison; its shape is not a forecast.

Complete numerical data are embedded in `scripts/dealer-logic/DL-15.js` and displayed as an accessible table in the HTML lesson.

## What would invalidate the reading?

A story can be internally consistent but built on an invented sign. Price agreement after the fact is not causal verification.

## Check your understanding

### What should be reported when available data cannot identify net dealer sign?

**Answer:** Unknown or model-dependent, with assumptions shown.

**Why:** Uncertainty is part of the useful output.

### What is the critical failure to avoid?

**Answer:** A story can be internally consistent but built on an invented sign. Price agreement after the fact is not causal verification.

**Why:** Explain this risk before treating the example as a decision rule. A correct calculation does not guarantee a favorable market outcome. **Critical risk check.**

## Takeaway

Return to the question: What can dealer analytics actually establish? State the observed facts, the assumptions and the failure condition separately.

## References and provenance

The cited references support definitions and conventions. The teaching prose, tickets and numerical scenarios are original; they are not attributed as historical market observations.

- [OIC Greeks](https://www.optionseducation.org/advancedconcepts/understanding-options-greeks)

- [Cboe 0DTE positioning study, May 2025](https://www.cboe.com/insights/posts/0-dt-es-decoded-positioning-trends-and-market-impact)