Desk instincts
for gas markets.

A working library of drills for the trader who wants to turn storage, weather, basis, and curve information into a position.

TODAY'S LENSThe spread is an option to wait.

A Cal+1 / Cal+2 trade is not just a z-score. Put a price on patience, then ask what the boundary is really doing.

LOCAL FIELD NOTE

Choose a drill

Desk prompts

page 1 / 1
01CURVE / SPREADS

The free boundary of a calendar spread

◆◆◇

“Buy it at two standard deviations” is a rule. Can you derive the rule that survives transaction costs?

DESK READ

PHYSICAL MECHANISM

What makes this relationship exist at all.
actor → asset → constraint
WHO MUST ACT

WHAT CARRIES IT

WHAT BINDS IT

dS = κ(θ − S) dt + σ dW

THE SETUP

Let S(t) = Cal+1 − Cal+2. Assume the spread mean-reverts around θ. You can enter long when the spread is cheap, pay round-trip cost c, and exit at the target.

DESK EXPRESSION

Conditional long Cal+1 / short Cal+2 when the spread clears the cost-adjusted entry boundary; size for liquidity and exit at fair value or regime break.

WHAT TO VERIFY

    WHERE IT BREAKS

      LIVE SKETCH

      Entry boundary, not a z-score

      A simulated spread path meets an entry boundary.

      drag the controls
      spread / valuetime / delivery month →

      TRADE FRAME

      Turn the model into a testable position.
      model → trade → falsify
      DATA

      SIGNAL / POLICY

      VALIDATION

      FAILURE MODE

      IMPLEMENTATION SPEC

      What you would actually code.
      data → state → decision → validation
      DATA CONTRACT
        STATE VARIABLES
          DECISION RULE

          VALIDATION

          FALSIFICATION

          PM MEMO / ALLOCATION DECISION

          WHO PAYS

          WHY IT PERSISTS

          CAPACITY

          PORTFOLIO FACTOR

          EDGE HALF-LIFE

          WHY THIS TRADE FAILS

            POSITION SIZING

            HIDDEN CONVEXITY / FRAGILITY / MODEL RISK

              BAYESIAN TRACKER

              Update a regime view in odds form. Correlated observations are discounted, so ten runs of one model are not ten independent updates.