新西兰储备银行分析

Step-Function Bootstrap from the Quarterly Bank Bill Strip

Instrument. ASX 90-Day New Zealand Bank Bill futures settle on the IMM quarterly cycle (March, June, September, December) against BKBM, the NZ Financial Markets Association's 90-day bank bill benchmark. Each contract prices 100 − the average BKBM over its roughly three-month reference window.

The identification problem. The monthly PyFedWatch/CME method used for the Fed, ECB and BoE assigns one meeting per contract window. The RBNZ holds about seven OCR decisions a year — roughly two inside each quarterly window — so one contract is one equation in two unknown decision steps, under-determined by about a factor of two. This is the same problem the Bank of Japan page describes for TONA, and we solve it the same way.

Model. The overnight path is treated as piecewise-constant, changing only on decision dates. The level after decision k is R_k = R₀ + Σ_j≤k δ_j, where each δ_j is the step at decision j. Each contract contributes one linear equation whose coefficients are the day-weighted fraction of that contract's window falling after each decision — so a step early in a window loads that window's average, while a step late in the window barely moves it but shows up in full in the next window. The short-and-wide system is resolved with a sparsity (ℓ₁) prior via iteratively reweighted least squares, which localises each move to a single decision rather than smearing a half-move across two. The recovered steps are quantised to 25bp and fed into the same expanding-tree methodology used for the Fed and ECB, so the decision tabs, granular basis-point outcomes and aggregated cut/hold/hike figures are produced by identical downstream code.

The BKBM anchor and the constant-spread assumption

Unlike TONA (BoJ) or SARON (SNB), which sit within a few basis points of their policy rates, BKBM is a genuine bank credit rate. We anchor R₀ on the latest observed 90-day BKBM and carry the current BKBM − OCR spread forward unchanged, exactly as the €STR/SONIA calculations do for the ECB and BoE. The RBNZ's own published research flags bank bill futures as a noisier proxy for OCR expectations than overnight-indexed-swap (OIS) pricing, which this site does not currently license. The spread is persisted as an explicit, auditable number for every run. Its known failure mode: it widens in a tightening cycle and in credit stress — precisely when the probabilities matter most — so a constant spread biases the implied path toward hikes in exactly those episodes.

Honesty guards (no invented data)

• A stale strip, fewer contracts than the decision horizon, or a degenerate fit (large residual) each yield no per-decision probabilities rather than a forced decomposition.
• Decisions beyond the coverage of the available contracts receive no figure.
• The BKBM anchor and the OCR shown above come from live data; if neither can be sourced, no probabilities are published.

Contrast with the ECB. The €STR calculation reads one meeting per monthly contract directly. The RBNZ calculation adds two stages in front — the quarterly-strip bootstrap that recovers each decision's step, and the BKBM→OCR spread adjustment — and is otherwise identical.