Evaluating contrarian sentiment thresholds for bond futures

What the evaluation asked

The evaluation asked which numeric levels of an after-close nonprofessional sentiment survey could serve as a complete long or short procedure in US Treasury bond futures. A complete procedure here means a numeric long or short cue, a numeric unwind, a stretch that stays flat, and a fill rule that matches when the reading actually arrives.

The daily sentiment reading

The survey is a daily sentiment reading: a one-to-one-hundred after-close survey of nonprofessional futures traders' bullishness. It is published after the close. On that scale, 80 to 100 marks extreme optimism, 1 to 20 marks extreme pessimism, and 50 marks relative neutrality.

Contrarian cues and rule-based entry

A common contrarian mapping treats a reading of 25 or below as a long cue and a reading of 75 or above as a short cue, on the premise that a one-sided crowd is already reflected in price. That mapping is a contrarian strategy: a signal procedure that fades crowded bullish or bearish survey readings.

Rule-based entry, in this setting, is a signal that fires only when the published sentiment reading crosses a pre-set numeric threshold and that stays flat until a specified opposite threshold is crossed.

When a reading may change a position

Because the reading is available only after the session ends, a historically faithful test may enter or exit only at the next day's opening price. That next-open execution constraint means a reading released after the close may change a position only at the following session's opening price.

The historical test assumed one contract, a one-tick spread of 1/32 of a point, a 10% annual financing charge, and at most one open position. The single-position constraint required that the prior trade be closed before a new one could start.

The long and short grids

Threshold-grid evaluation tested many paired entry and exit crossings on the same historical sample to see which complete procedures remain acceptable after costs and a drawdown screen.

Long-only procedures were specified as buy crossings below 10, 15, 20, 25, or 30 paired with sell crossings above 45, 50, 60, 70, 75, or 80 on the January 1, 1987 through December 7, 1994 sample.

Short-only procedures were specified as sell crossings above 90, 85, 80, 75, or 70 paired with cover crossings below 55, 50, 40, 30, 25, or 20 on that same sample.

The drawdown screen and rejected shorts

Maximum drawdown was defined as the lowest equity balance after each completed round-trip, and combinations with an adverse drawdown were to be discarded unless total equity return was at least ten times that drawdown. That round-trip-drawdown screen rejects a completed-trade equity path unless total return is large relative to the worst post-round-trip equity dip, illustrated here as a 10-to-1 bar.

Four short combinations were identified as ones not to use because they produced a net loss per trade: sell above 90 and cover below 20 or 25, and sell above 85 and cover below 20 or 25.

A same-length passive hold

Over the sample the contract rose by an average of 0.00851 point per day, so each rule's average holding period could be compared with a passive hold lasting the same number of days.

Editorial: why a grid cell is unfinished

Editorial: picking a cell from an in-sample grid is incomplete until that same bundled procedure is walked forward. The archive describes the historical workflow on one sample. It does not treat an in-sample grid winner as a finished rule.

Editorial: walk-forward analysis is a way to keep a multi-threshold entry and exit search honest by re-validating the whole procedure, including inputs, state, and next-session execution, on later samples.